Green Belt MTB 2.5

© Lean Sigma Corporation
Lean Six Sigma Training - MTB

Lean Sigma Corporation
Lean Six Sigma Green Belt Training
Featuring Examples from Minitab 16
1
www.LeanSigmaCorporation.com
© Lean Sigma Corporation Lean Six Sigma Training - MTB
2
1.0 Define Phase
1.1 Overview of Six Sigma
3
Green Belt Training: Define Phase
1.1 Six Sigma Overview 1.4 Lean Fundamentals
1.1.1 What is Six Sigma 1.4.1 Lean and Six Sigma
1.1.2 Six Sigma History 1.4.2 History of Lean
1.1.3 Six Sigma Approach Y = f(x) 1.4.3 The Seven Deadly Muda
1.1.4 Six Sigma Methodology 1.4.4 Five-S (5S)
1.1.5 Roles and Responsibilities
1.2 Six Sigma Fundamentals

1.2.1 Defining a Process
1.2.2 VOC and CTQs
1.2.3 QFD
1.2.4 Cost of Poor Quality (COPQ)
1.2.5 Pareto Analysis (80 : 20 rule)
1.3 Lean Six Sigma Projects
1.3.1 Six Sigma Metrics
1.3.2 Business Case and Charter
1.3.3 Project Team Selection
1.3.4 Project Risk Management
1.3.5 Project Planning
4
5
1.1.1 What is Six Sigma
What is Six Sigma?
• What is “sigma”?
• In statistics, sigma (s) refers to “standard deviation,” which is a
measure of variation.
• You will come to learn that variation is the enemy of any quality
process. We need to understand, manage, and minimize process
variation.
• What is “Six Sigma”?

• Six Sigma is an aspiration or goal of process performance.
• A Six Sigma “goal” is for a process average to operate approximately
6s away from customer’s high and low specification limits.
6
What is Six Sigma?
• A process whose average is about 6s away from the
customer’s high and low specification limits has
abundant room to “float” before approaching the
customer’s specification limits.
• A Six Sigma process only yields 3.4 defects for every
million opportunities! In other words, 99.9997% of the

products are defect-free!
7
What is Six Sigma: Sigma Level
• Sigma level measures how many “sigma” there are

between your process average and the nearest
customer specification.
• Let us assume that your customers upper and lower
specifications limits (USL & LSL) were narrower

than the width of your process spread.
• The USL & LSL below stay about 1 standard
deviation away from the process average.
Therefore, this process operates at 1 sigma.
8
• A process operating at 1 sigma has a defect rate of

approximately 70%.
• This means that the process will generate

defect-free products only 30% of the time.

• What about processes with more than 1 sigma
level?
• A higher sigma level means a lower defect rate.
• Let us take a look at the defect rates of processes at

different sigma levels.
9
• This table shows each sigma level’s corresponding
defect rate and DPMO (defects per million opportunities).
• The higher the sigma level, the lower the defective rate
and DPMO.

• How does this translate into things you might easily
relate to?
10
• Let us take a look at processes operating at 3 sigma.

• 3 sigma processes have a defect rate of approximately
7%. What would happen if processes operated at 3
sigma?
• Virtually no modern computer would function*.

• 10,800,000 health care claims would be mishandled each year.
• 18,900 US savings bonds would be lost every month.
• 54,000 checks would be lost each night by a single large bank.
• 4,050 invoices would be sent out incorrectly each month by a
modest-sized telecommunications company.
• 540,000 erroneous call details would be recorded each day
from a regional telecommunications company.
• 270 million erroneous credit card transactions would be
recorded each year in the United States.
(*http://www.qualityamerica.com)
11
• What if processes operated with 1% defect rate?
• 20,000 lost articles of mail per hour*.
• Unsafe drinking water almost 15 minutes per day.
• 5,000 incorrect surgical operations per week.
• Short or long landings at most major airports each day.
• 200,000 wrong drug prescriptions each year.

• No electricity for almost 7 hours per month.
• Even at 1% defect rate, some processes would be

unacceptable to you and many others.
• So what is Six Sigma?

• Sigma level is the measure!
• Six is the goal!
12
(* Implementing Six Sigma – Forest W. Breyfogle III) www.LeanSigmaCorporation.com
What is Six Sigma: The Methodology
• Six Sigma itself is the goal, not the method.
• In order to achieve Six Sigma, you need to improve your

process performance by:
• Minimizing the process variation so that your process has
enough room to fluctuate within customer’s spec limits

• Shifting your process average so that it is centered between
your customer’s spec limits.
• Accomplishing these two process improvements (along

with stabilization and control), you can achieve Six
Sigma.
• DMAIC is the systematic methodology prescribed to

achieve Six Sigma. 13
What is Six Sigma: The Methodology
• DMAIC is a systematic and rigorous methodology that
can be applied to any process in order to achieve Six
Sigma.
• It consists of 5 phases of a project:
• Define
• Measure

• Analyze
• Improve
• Control.
• You will be heavily exposed to many concepts, tools, and
examples of the DMAIC methodology through this
training.
• You will be capable of applying the DMAIC methodology
to improve the performance of any process at the
completion of the curriculum.
14
1.1.2 Six Sigma History
15
Six Sigma History
Polaroid

16
Six Sigma History
• The “Six Sigma” terminology was originally adopted by Bill
Smith at Motorola in the late 1980s as a quality management
methodology.
• As the “Father of Six Sigma,” Bill forged the path for Six Sigma
through Motorola’s CEO Bob Galvin who strongly supported

Bill’s passion and efforts.
• Starting from the late 1980s, Motorola extensively applied Six

Sigma as a process management discipline throughout the
company, leveraging Motorola University.
• In 1988, Motorola was recognized with the prestigious Malcolm

Baldrige National Quality Award for its achievements in quality
improvement.
17
Six Sigma History
• Six Sigma has been widely adopted by companies as an
efficient way of improving the business performance since
General Electric implemented the methodology under the
leadership of Jack Welch in the 1990s.
• As GE connected Six Sigma results to its executive

compensation and published the financial benefits of Six
Sigma implementation in their annual report, Six Sigma
became a highly sought-after discipline of quality.
18
Six Sigma History
• Most Six Sigma programs cover the aspects, tools, and topics
of Lean or Lean Manufacturing.
• The two work hand in hand, benefitting each other.

• Six Sigma focuses on minimizing process variability, shifting
the process average, and delivering within customer’s

specification limits.
• Lean focuses on eliminating waste and increasing efficiency.
• Lean and its popularity began to form and gain significant

traction in the mid 1960s with the Toyota initiative “TPS” or
Toyota Production System.
• The concepts and methodology of Lean, however, were

fundamentally applied much earlier by both Ford and Boeing in
the early 1900s.
19
Six Sigma History
• Despite the criticism and immaturity of Six Sigma in many
aspects, its history continues to be written with every company
and organization striving to improve its business performance.

Polaroid
20
1.1.3 Six Sigma Approach
21
Six Sigma Approach: Y = f(x)
• The Six Sigma approach to problem solving uses a

transfer function.
• A transfer function is a mathematical expression of

the relationship between the inputs and outputs of a

system.
• Y = f(x) is the relational transfer function that is used

by all Six Sigma practitioners.
• It is absolutely critical that you understand and

embrace this concept.
22
• “Y” refers to the measure or output of a process.
• Y is usually your primary metric
• Y is the measure of process performance that you are trying
to improve.
• f(x) means “function of x.”
• x’s are factors or inputs that affect the Y

• Combined, the Y = f(x) statement reads “Y is a
function of x.”
• In simple terms: “My process performance is
dependent on certain x’s.”
• The objective in a Six Sigma project is to identify the
critical x’s that have the most influence on the output
(Y) and adjust them so that the Y improves.
23
• Let us look at a simple example of a pizza delivery

company that desires to meet customer expectations
of on-time delivery.
• Measure = on-time pizza deliveries
• Y = percent of on-time deliveries
• f(x) would be the x’s or factors that heavily influence timely

deliveries
• x1: might be traffic
• x2: might be the number of deliveries per driver dispatch
• x3: might be the accuracy of directions provided to the driver
• x4: might be the reliability of the delivery vehicle
• etc.
• The statement Y = f(x) in this example will refer to the
proven x’s determined through the steps of a Six Sigma
project.
24
• With this approach, all potential

X2
X3 x’s are evaluated throughout the
X7 X4 X5 DMAIC methodology.
X8

X1
• The x’s should be narrowed
X6 down until the vital few x’s that
significantly influence “on-time
pizza deliveries” are identified!
Y=f(x)
25
• This approach to problem solving will take you through
the process of determining all potential x’s that might
influence on-time deliveries and then determining
through measurements and analysis which x’s do
influence on-time deliveries.
• Those significant x’s become the ones used in the Y =

f(x) equation.
• The Y = f(x) equation is a very powerful concept and

requires the ability to measure your output and quantify
your inputs.
• Measuring process inputs and outputs is crucial to
effectively determining the significant influences to any
process.
26
1.1.4 Six Sigma Methodology
27
Six Sigma Methodology
• Six Sigma follows a methodology that is

conceptually rooted in the principles of a five-phase
project.
• Each phase has a specific purpose and specific
tools and techniques that aid in achieving the phase

objectives.
• The 5 phases of DMAIC:

1. Define
2. Measure
3. Analyze
4. Improve
5. Control
28
• Define and quantify the

Define problem and objective
• Baseline process, validate measures,

Measure and identify all possible x’s and
dependent Y’s

• Analyze and validate causes
Analyze identifying critical factors
Improve • Develop solutions
• Implement
and
Control sustain
solutions
29
Six Sigma Methodology: Define Phase
• The goal of the Define phase is to establish a solid foundation
and business case for a Six Sigma project.
• Define is arguably the most important aspect of any Six Sigma

project.

• All successful projects start with a current state challenge or
problem that can be articulated in a quantifiable manner.
• It is not enough to just know the problem, you must quantify it
and also determine the goal.
• Once problems and goals are identified and quantified, the rest
of the define phase will be about valuation, team, scope, project
planning, timeline, stakeholders, Voice Of the Customer (VOC),
and Voice Of the Business (VOB).
30
Six Sigma Methodology: Define Phase
• Define Phase Tools and Deliverables

• Project Charter – Establish the:
• Business Case
• Problem Statement
• Project Objective

• Project Scope
• Project Timeline
• Project Team.
• Stakeholder Assessment
• High-Level Pareto Chart Analysis
• High-Level Process Map
• VOC/VOB and CTQs Identified and Defined
• Financial Assessment
31
Six Sigma Methodology: Measure Phase
• The goal of the Measure phase is to gather baseline

information about the process (process performance,
inputs, measurements, customer expectations etc.).
• Throughout the Measure phase you will seek to
achieve a few important objectives:

• Gather All Possible x's
• Assess Measurement System and Data Collection
Requirements
• Validate Assumptions
• Validate Improvement Goals
• Determine COPQ (Cost of Poor Quality)
• Refine Process Understanding
• Determine Process Stability
• Determine Process Capability.
32
Six Sigma Methodology: Measure Phase
• Measure Phase Tools and Deliverables

• Process Maps, SIPOC, Value Stream Maps
• Failure Modes and Effects Analysis (FMEA)
• Cause-and-Effect Diagram
• XY Matrix
•

Six Sigma Statistics
• Basic Statistics
• Descriptive Statistics
• Measurement Systems Analysis
• Variable and/or Attribute Gage R&R
• Gage Linearity and Accuracy or Stability
• Basic Control Charts
• Process Capability (Cpk, Ppk) and Sigma Levels
• Data Collection Plan
33
Six Sigma Methodology: Analyze Phase
• The Analyze phase is all about establishing verified
drivers.
• In the DMAIC methodology, the Analyze phase uses

statistics and higher-order analytics to discover
relationships between process performance and process

inputs (in other words, what are the root causes or
drivers of the improvement effort).
• Ultimately, the Analyze phase establishes a reliable

hypothesis for improvement solutions.
• Establish the Transfer Function Y = f(x)
• Validate the List of Critical x's and Impacts
• Create a Beta Improvement Plan (e.g., pilot plan).
34
Six Sigma Methodology: Analyze Phase
• Analyze Phase Tools and Deliverables

• The Analyze phase is about proving and
validating critical x’s using the appropriate and
necessary analysis techniques. Examples include:
• Hypothesis Testing

• Parametric and Non-Parametric
• Regression
• Simple Linear Regression
• Multiple Linear Regression
• The Analyze phase is also about establishing a

set of solution hypotheses to be tested and further
validated in the Improve phase.
35
Six Sigma Methodology: Improve Phase
• The goal of the Improve phase is. . .you guessed it!

"make the improvement." Improve is about
designing, testing, and implementing your solution.
• To this point you have defined the problem and
objective of the project, brainstormed possible x's,

analyzed and verified critical x's. Now it's time to
make it real!
• Statistically Proven Results from Active Study/Pilot
• Improvement/Implementation Plan
• Updated Stakeholder Assessment
• Revised Business Case with Return on Investment (ROI)
• Risk Assessment/Updated FMEA
• New Process Capability and Sigma.
36
Six Sigma Methodology: Improve Phase
• Improve Phase Tools and Deliverables

• Any Appropriate Tool from Previous Phases
• Design of Experiment (DOE)

• Full Factorial

• Fractional Factorial
• Pilot or Planned Study Using:

• Hypothesis Testing
• Valid Measurement Systems
• Implementation Plan
37
Six Sigma Methodology: Control Phase
• The last of the 5 core phases of the DMAIC

methodology is the Control phase.
• The goal of the Control phase is to establish

automated and managed mechanisms to maintain

and sustain your improvement.
• A successful control plan also establishes a reaction

and mitigation plan as well as an accountability
structure.
38
Six Sigma Methodology: Control Phase
• Control Phase Tools and Deliverables

• Statistical Process Control (SPC/Control Charts)
• IMR, XbarS, XbarR, P, NP, U, C etc.
• Control Plan Documents

• Control Plan
• Training Plan
• Communication Plan
• Audit Checklist
• Lean Control Methods

• Poka-Yoke
• Five-S
• Kanban
39
Six Sigma DMAIC Roadmap
• Goal: Problem Statement, Objective, Business Case, Project Scope, Team

Define • Main Tools: Project Charter, Pareto, Process Maps
• Goal: Brainstorm/Prioritize Possible x’s, Validate measurement, Capability

• Tools: Basic Statistics, C & E, XY Matrix, Capability Analysis, MSA, Process

Measure Maps, Control Charts
• Goal: Identify critical x’s

Analyze • Tools: Hypotheses Tests (Normal/Non Normal), Regression and Correlation
• Goal: Design, Test, and Implement Improvement

Improve • Tools: DOE, Implementation/Change/Communication Plan
• Goal: Lock-in the Improvement

Control • Tools: Control Plan, Poka-Yoke, SPC, SOPs, Training Plans etc.
40
41
Roles and Responsibilities
• The various roles in a Six Sigma program are

commonly referred to as “Belts.”
• In addition to Belts, there are also other key roles

with specific responsibilities.

• Let us explore the different roles and their
corresponding responsibilities in a Six Sigma
program.
42
• Each of the four Six Sigma belts represents a

different level of expertise in the field of Six Sigma.
• Six Sigma Master Black Belt (MBB) Expert
• Six Sigma Black Belt (BB)
• Six Sigma Green Belt (GB)
•

Six Sigma Yellow Belt (YB)
Novice
• In addition to Belts, there are other critical and

complementary roles:
• Champions
• Sponsors
• Stakeholders
• Subject Matter Experts (SMEs).
43
Roles and Responsibilities: MBB
• The Master Black Belt (MBB) is the most experienced,
educated, and capable Six Sigma expert.
• A typical MBB has managed dozens of Black Belt level

projects.

• The MBB can simultaneously lead multiple Six Sigma
Belt projects while mentoring and certifying Black Belt
and Green Belt candidates.
• The MBB typically works with high-level operations

directors, senior executives, and business managers to
help with assessing and planning business strategies
and tactics.
44
Roles and Responsibilities: MBB
• MBB commonly advises management team on the cost
of poor quality of an operation and consults on methods
to improve business performance.
• Typical Responsibilities of a MBB

• Identifies and defines the portfolio of projects required to

support a business strategy
• Establishes scope, goals, timelines, and milestones
• Assigns and marshals resources
• Trains and mentors Green Belts and Black Belts
• Facilitates tollgates or checkpoints for Belt candidates
• Reports-out/updates stakeholders and executives
• Establishes organization’s Six Sigma strategy/roadmap
• Leads the implementation of Six Sigma.
45
Roles and Responsibilities: BB
• The Black Belt (BB) is the most active and valuable
experienced Six Sigma professional among all the Six
Sigma Belts.
• A typical BB has
• led multiple projects

• trained and mentored various Green Belts candidates
• understood how to define a problem and drive effective
solution.
• The BB is well rounded in terms of project management,

statistical analysis, financial analysis, meeting facilitation,
prioritization, and a range of other value-added
capabilities, which makes a BB highly valuable asset in
the business world.
46
• BBs commonly serve as the dedicated resource

continuing their line management role while
simultaneously achieving a BB certification.
• Typical Responsibilities of a BB

• Project Management
• Defines projects, scope, teams etc.
• Marshals resources
• Establishes goals, timelines, and milestones
• Provides reports and/or updates to stakeholders and
executives.
47
• Typical Responsibilities of a BB (continued)

• Task Management
• Establishes the team’s Lean Sigma roadmap
• Plans and implements the use of Lean Sigma tools
• Facilitates project meetings

• Does project management of the team’s work
• Manages progress toward objectives.
• Team Management
• Chooses or recommend team members
• Defines ground rules for the project team
• Coaches, mentors, and directs project team
• Coaches other Six Sigma Belts
• Manages the team’s organizational interfaces.
48
Roles and Responsibilities: GB
• The Green Belt (GB) is considered as a less intense
version of Six Sigma professional than the Black Belt
(BB).
• A GB is exposed to all the comprehensive aspects of Six

Sigma with less focus on the statistical theories and

some other advanced analytical methodologies such as
Design of Experiment (DOE).
• When it comes to project management, a GB has almost

the same responsibilities as a BB.
• In general, the GB works on less complicated and

challenging business problems than a BB.
49
Roles and Responsibilities: GB
• Typical Responsibilities of a Green Belt
• Project Management
• Defines the project, scope, team etc.
• Marshals resources
• Sets goals, timelines, and milestones
• Reports-out/updates stakeholders and executives.
• Task Management

• Establishes the team’s Lean Sigma Roadmap
• Plans and implements the use of Lean Sigma tools
• Facilitates project meetings
• Does Project Management of the team’s work
• Manages progress toward objectives.
• Team Management
• Chooses or recommends team members
• Defines ground rules for the project team
• Coaches, mentors, and directs project team
• Coaches other Six Sigma Belts
• Manages the team’s organizational interfaces.
50
Roles and Responsibilities: YB
• The Yellow Belt (YB) understands the basic objectives
and methods of a Six Sigma project.
• YB has an elementary understanding about what other

Six Sigma Belts (GB, BB, MBB) are doing to help them
succeed.

• In a Six Sigma project, YB usually serves as a subject
matter expert regarding some aspects of the process or
project.
• Supervisors, managers, directors, and sometimes

executives are usually trained at the YB level.
51
Roles and Responsibilities: YB
• Typical Responsibilities of a Yellow Belt
• Helps define process scope and parameters
• Contributes to team selection process

• Assists in information and data collection
• Participates in experiential analysis sessions (FMEA,

Process Mapping, Cause and Effect etc.)
• Assists in assessing and developing solutions
• Delivers solution implementations.

52
Roles and Responsibilities: Champions & Sponsors
• Champions and sponsors are those individuals

(directors, executives, managers etc.) chartering,
funding, or driving the Six Sigma projects that BBs
and GBs are conducting.

• Champions and sponsors need to have a basic
understanding of the concepts, tools, and
techniques involved in the DMAIC methodology so
that they can provide proper support and direction.
53
Roles and Responsibilities: Champions & Sponsors
• Champions and sponsors play critical roles in the

successful deployment of Six Sigma.
• Strong endorsement of Six Sigma from the leadership
team is critical for success.

• Typical Responsibilities of a Champion/Sponsor
• Maintains a strategic oversight
• Establishes strategy and direction for a portfolio of projects
• Clearly defines success
• Provides resolution for issues such as resources or politics
• Establishes routine tollgates or project reviews
• Clears the path for solution implementation
• Assists in project team formation.
54
Roles and Responsibilities: Stakeholders
• Stakeholders are usually the recipients or

beneficiaries of the success of a Six Sigma project.
• Stakeholders are individuals owning the process,

function, or production/service line that a Six Sigma

Belt focuses on improving the performance of.
• BBs and GBs need to keep strong working

relationships with stakeholders because without their
support, it would be extremely difficult to make the
Six Sigma project a success.
55
Roles and Responsibilities: SMEs
• Subject Matter Experts (SMEs) are commonly known
as the experts of the process or subject matter.
• Six Sigma Belts should proactively look to key SMEs to

round out their working project team.

• SMEs play critical roles to the success of a project.
• Based on SMEs’ extensive knowledge about the process,
they have the experience to identify which solutions can
work and which cannot work.
• SMEs who simply do not speak up can hurt the chances of
the process’ success.
• SMEs are also the same people who prefer to keep the
status quo. Six Sigma Belts may find many of them unwilling
to help implement the changes.
56
• Throughout this module we have reviewed the

various common roles and corresponding
responsibilities in any Six Sigma program:
• Six Sigma Master Black Belt
• Six Sigma Black Belt
•

Six Sigma Green Belt
• Six Sigma Yellow Belt
• Champion and Sponsors
• Stakeholders
• Subject Matter Experts (SMEs)
• These Six Sigma belts and other roles are designed

to deliver value to the business effectively and
successfully.
57
1.2. Six Sigma Fundamentals
58

1.2.3 QFD
1.2.5 Pareto Analysis (80:20 rule)
59
60
Defining a Process
• The basic method of defining and understanding a
process is the process map.
• Process maps help determine where and how a process
begins as well as all the steps and decisions in between.
• By learning the various types and methods of process
maps, you can become adept at setting project scopes,

identifying value-added and non-value-added steps,
identifying problems in a process, etc.
• This module covers:
• High-level process maps
• Detailed process maps
• Functional maps.
• In the Measure section we will touch on several other
types and methods of process mapping.
61
What is a Process Map?
• A process map is a graphical representation of a
process flow.
• It illustrates how the business process is accomplished
step by step.
• It describes how the materials or information
sequentially flow from one business entity to the next.

• It illustrates who is responsible for what between the
process boundaries.
• It depicts the inputs and outputs of each individual
process step.
• Always encourage your project team to map the current

state of the process instead of the ideal state. Be
honest with each other!
62
Process Map Basic Symbols
• The following four symbols are the most commonly
used symbols in a process map.
Terminator (Oval):
Shows the start and end points in the process.

Process (Rectangle):
Indicates a single process step.
Decision (Diamond):
Indicates a question with two choices (e.g.
Yes/No)
Flow Line (Arrow):

Shows the direction of the process
flow.
63
Additional Process Symbols
• Additional Process Symbols:
Alternative Process:
Indicates a process step as an alternate of a normal step.
Predefined Process:
Indicates a formally-defined process step. Other

documentation or instruction is needed to support further
details of the step.
Manual Operation:
Indicates a process step conducted manually.
Preparation:
Indicates a preparation step.
Delay:
Indicates a waiting period in the process.
64
• Additional File and Information Related Symbols:
Data (I/O):
Shows the inputs and outputs of a process.
Document:
Indicates a process step that results in a document.

Multi-Document:
Indicates a process step that results in multiple documents.
Stored Data:
Indicates a process step that stores data.
Magnetic Disk:
Indicates a database.
65
• Additional Control of Flow Symbols:
Off-Page Connector:
Indicates the process flow continues onto another page.
Merge:

Indicates multiple processes merge into one.
Extract:
Indicates a process splits into multiple parallel processes.
Or:
Indicates a single data processing flow diverges to multiple
branches with different criteria requirements.
Summing Junction:
Indicates multiple data processing flows converge into one.
66
How to Plot a Process Map
• Step 1: Define the boundaries of the process you

want to map.
• A process map can depict the flow of an entire process or a

segment of it.

• You need to identify and define the beginning and ending
points of the process before starting to plot.
• Use operational definitions where possible.
67
• Step 2: Define and sort the process steps with the

flow.
• Consult with process owners and SMEs or observe the

process in action to understand how the process is actually
performed.

• Record the process steps and sort them according to the
order of their occurrence.
68
• Step 3: Fill the step information into the appropriate

process symbols and plot the diagram.
• In the team meeting of process mapping, place the sticky

notes with different colors on a white board so you can
move them around while the map is under-construction.

• The flow lines can be plotted directly on the white board.
• Decision steps. Rotate the sticky note 45 degrees.
• When the map is completed on the white board, record the

map using Excel, PowerPoint, Visio, Quality Companion, or
other preferred software.
69
• Step 3:
• To illustrate the responsibility of different organizations
involved in the process, use a Swim Lane Process Map.

70
• Step 4: Identify and record inputs/outputs and their
corresponding specifications for each process step.
• The process map helps in understanding and documenting

Y = f(x) of a process, where Y represents the outputs and x
represents the inputs.

• The inputs of each process step can be controllable or non-
controllable, standardized operational procedures, or noise.
• Inputs are the source of variation in the process and need to

be analyzed qualitatively and quantitatively.
• The outputs of each process step can be products,

information, services, etc. They are the little Y’s within the
process.
71
• Step 5: Evaluate the process map and adjust if

needed.
• If the process is too complicated to be covered in one single

process map, you may create additional detailed sub-
process maps for further information.

• Number the process steps in the order of their occurrence
for clarity.
72
High Level Process Map
• Most high-level process maps are also referred to as
flow charts.
• The key to a high-level process map is to over-simplify
the process being depicted so that it can be understood
in its most generic form.
• As a general rule, high-level process maps should be no

more than 4–6 steps.
• Below is an oversimplified version of a high-level process
map for cooking a 10lb prime rib for a dozen holiday
guests.
73
Detailed Process Map
• Detailed process maps or multi-level maps take the high-
level map much further.
• Detailed maps can be 2–4 levels deeper than your high-
level process map.
• A good guideline used to help create the second level is
to take each step in the high level map and break it down

into 2–4 steps (no more).
• Repeat this process (level 3, level 4 etc.) until reaching

the desired level of detail.
• Some detailed maps are 2 or 3 levels deep, others can
be 5–6 levels deep. Obviously, the deeper the levels, the
more complex and the more burdensome.
74
• At left is our prime

rib cooking
example at level 2
detail.
• This process map

has a few more
decision points and
process steps.
• You can see that

going only one
more level deep
adds a fair amount
of information to
the process map.
75
Functional Process Map
• The functional map adds dimension to the high-level or
detailed map.
• The dimension added is identifying which function or job
performs the step or makes the decision.
• Below is a generic example of a functional map. Note that
functions are identified in horizontal lanes and each process

step is placed in the appropriate lane based on which function
performs the step.
76
77

1.2.2 VOC and CTQs
Voice of the Customer
• VOC stands for “Voice of the Customer.”

• Voice of the customer is a term used for a data-
driven plan to discover customer wants and needs.
• VOC is an important component to a successful Six
Sigma project.

• There are also other “Voices” that need to be heard
when conducting projects. The 3 primary forms are:
• VOC: Voice of the Customer

• VOB: Voice of the Business
• VOA: Voice of the Associate.
78
Gathering VOC
• Gathering VOC should be performed methodically.

• The two most popular methods of collecting VOC
are
1. Indirect
2. Direct.

• 1. Indirect data collection for VOC involves passive
information exchange:
• Warranty claims
• Customer complaints/compliments
• Service calls
• Sales reports.
• 2. Indirect methods are less effective, sometimes
dated, require heavy interpretation, and are also
more difficult to confirm.
79
Gathering VOC
• Direct data collection methods for VOC are active

and planned customer engagements:
• Conducting interviews
• Conducting customer surveys
• Conducting market research
•

Hosting focus groups.
• Direct methods are typically more effective for
several reasons:
• Less need to interpret meaning
• Researchers can go a little deeper when interacting with
customers
• Customers are aware of their participation and will respond
better upon follow-up
• Researchers can properly plan engagements (questions,
sample size, information collection techniques etc.).
80
Gathering VOC
• Gathering VOC requires consideration of many factors
such as product or services types, customer segments,
manufacturing methods or facilities etc.
• All this information will influence the sampling strategy.
• Consider which factors are important and build a sample

size plan around them.
• Also, consider response rates and adjust the initial sample
strategy to ensure adequate input is received.
• Once a sampling plan is in place, collect data via the

direct and indirect methods discussed earlier.
• After gathering VOC it will be necessary to translate it into
something meaningful: CTQs.
81
Critical to Quality: CTQ
• CTQ stands for Critical to Quality.
• CTQs are translated from VOC or
“voice of the customer” feedback.
• VOC is often vague, emotional, or simply
a generalization about products or services.
• CTQs are the quantifiable, measureable, and meaningful

translations of VOC.
• Organizing VOC helps to identify CTQs.
• One effective way to organize VOC is to group or

bucket it using an affinity diagram.
• Affinity diagrams are ideal for large amounts of soft
data resulting from brainstorming sessions or
surveys.
82
Affinity Diagram: Building a CTQ Tree
• Steps for conducting an Affinity Diagram exercise:
• Step 1: Clearly define the question or focus of the exercise

(“Why are associates late for work?”).
• Step 2: Record all participant responses on note cards or sticky

notes (this is the sloppy part, record everything!).

• Step 3: Lay out all note cards or post the sticky notes onto a
wall.
• Step 4: Look for and identify common themes.
• Step 5: Begin moving the note cards or sticky notes into the
themes until all responses are allocated.
• Step 6: Re-evaluate and make adjustments.

83
• Define the question or focus
• Record responses on note cards or sticky notes
• Display all note cards or sticky notes on a wall if
necessary.

84
• Look for and identify common themes within the

responses.

85
• Group note cards or sticky notes into themes until all
responses are allocated.
• Re-evaluate and make final adjustments.

86
CTQ Tree
• Example of a generic CTQ tree transposed from a

white board to a software package.

87
Kano
• Another VOC categorization technique is the Kano.
• The Kano model was developed by Noriaki Kano in the
1980s.
• The Kano model is a graphic tool that further categorizes
VOC and CTQs into 3
distinct groups:

• Must Haves
• Performance Attributes
• Delighters.
• The Kano helps to

identify CTQs
that add incremental
value vs. those that are
simply requirements and having
more is not necessarily better.
88
Validating VOC and CTQs
• After determining all CTQs,

confirm them with the customer.
• Confirming can be accomplished
by conducting surveys through one
or more of the following methods:

• Group sessions
• One-on-one meetings
• Phone interviews
• Electronic means (chat, email, social media etc.)
• Physical mail.
• Consider your confirming audience and try to avoid
factors that may influence or bias responses such
as inconvenience or overly burdensome time
commitments.
89
Translating CTQs to Requirements
• Lastly, CTQs must be transformed into specifics that
can be built upon in a process.
• A requirements tree translates CTQs to meaningful

and measureable requirements for production

processes and products.
90
1.2.3 Quality Function Deployment
91
History of QFD
• Developed by Shigeru Mizuno (1910–1989) and Yoji

Akao (b. 1928) in Japan. Quality Function
Deployment (QFD) aims to design products that
assure customer satisfaction and value – the first
time and every time.

• The QFD framework can be used for translating
actual customer statements and needs (“The voice
of the customer”) into actions and designs to build
and deliver a quality product.
92
What is QFD?
• Quality Function Deployment (QFD) is a

construction methodology and quantification tool
used to identify and measure customer’s
requirements and transform them into meaningful
and measureable parameters.

• QFD helps to prioritize actions to advance process
or product to meet customer’s anticipations.
• QFD is an excellent tool for contact between cross-

functional groups.
93
Purpose of QFD
The quality function deployment has many purposes.

Among the most important are:
• Market analysis to establish needs and expectations
• Examination of competitors’ abilities

• Identification of key factors for success
• Translation of key factors into product and process

characteristics.
94
Phases of QFD
Four Key Phases of QFD
• Phase I: Product Planning Including the “House of

Quality” (Requirements Engineering Life Cycle)

• Phase II: Product Design (Design Life Cycles)
• Phase III: Process Planning (Implementation Life

Cycle)
• Phase IV: Process Control (Testing Life Cycle)
95
How to build a House of Quality
• Determine Customer Requirements (What’s from VOC/CTQ)
• Technical Specifications/Design Requirements (How’s)
• Develop Relationship Matrix (What’s and How’s)
• Prioritize Customer Requirements

• Conduct Competitive Assessments
• Develop Interrelationship (How’s)
• Prioritize Design Requirements
96
House of Quality
Correlation Matrix
[How’s vs. How’s]
Technical Specifications
Why’s

[How’s]
Customer
Customer
Relationship Matrix Customer Market
Requirements
Importance Evaluation
[What’s vs. How’s] Rating [What’s vs.
[What’s]
Why’s]
Degree of Technical Difficulty

Overall Importance Ratings
How Much’s Technical Competitive Evaluations
Target Goals
97
Step 1: Determine Customer Requirements
• Identify the important customer requirements.

These are the “What’s” and are typically
determined through the VOC/CTQ process.
• Use the results from your requirements tree

diagram as inputs for the customer requirements
in your HOQ.

On-Time Delivery
Customer Requirements
Increase Product Quality
Alternative Solutions
Reduce Turn-Around Time
Optimize Resource
Reduce COPQ
High Productivity
98
Step 2: Technical Specification
• Potential choices for product features

• Voice of Designers or Engineers
• Each “What” item must be refined to “How’s”

Technical Specification
Design and Development
Supply and Distribution
Material Management
Asset Management
Customer Support
Research and
Development
Development
Resource
99
Step 3: Develop Relationship Matrix (What’s & How’s)
• This is the center portion of the house. Each cell

represents how each technical specification
relates to each customer requirement.

100
Step 4: Prioritize Customer Requirements
• This is the right portion of the house. Each cell
represents customer requirements based on
relative importance to customers and
perceptions of competitive performance.

101
Step 5: Competitive Assessments
• This is the extreme right portion of the

house. Comparison of the organization’s
product to competitors’ products.

102
Step 6: Correlation Matrix
• This is the top portion of the house. It identifies

the way “how” items either support (positive) or
conflict (negative) with one another.

+9 Strong Positive
+3 Positive
-3 Negative
-9 Strong Negative
103
Step 7: Prioritize Design Requirements
• Overall Importance Ratings
Function of relationship ratings and customer
prioritization ratings
• Technical Difficulty Assessment

Similar to customer market competitive evaluations

but conducted by the technical team
104
Step 7: Prioritize Design Requirements
• Technical Specification Competitive Evaluation
Helps to establish the feasibility and realization of
each “how” item
• Target Goals
How much is good enough to satisfy the customer

105
House of Quality

1 3 4 5
7
106
Pros of QFD
• Focuses the design of the product or process on

satisfying customer’s needs and wants.
• Improves the contact channels between customers,

advertising, research and improvement, quality and

production departments, which sustains better
decision making.
• Reduces the new product development project

period and cost.
107
Cons of QFD
• The relationship matrix can be too obscure with

many process inputs and/or many customer
constraints.
• It can be very complicated and difficult to implement

without experience.
• If throughout the process new ideas, specifications,

or requirements are not discovered, you run the risk
of losing team members’ trust in the process.
108
QFD Summary
When used properly, the quality function deployment is

an extremely valuable approach to product/process
design.
There are many benefits of QFD that can only be
realized when each step of the process is completed

thoroughly:
• Logical way of obtaining information and presenting it
• Smallest product development cycle
• Considerably condensed start-up costs
• Fewer engineering alterations
• Reduced chance of supervision during design process
• Collaborating environment
• Preserving everything in characters.
109
1.2.4 Cost of Poor Quality
110
Cost of Poor Quality
• Cost of Poor Quality (COPQ) is the

expense incurred due to waste,
inefficiencies, and defects.
• The COPQ has been proven to range from 5% to

30% of gross sales for most companies.
• The COPQ can be staggering when considering
process inefficiencies, hidden factories, defective
products, rework, scrap, etc.
• Understanding COPQ and where to look for it will
help uncover process inefficiencies, defects, and
hidden factories within your business.
111
• There are 7 common forms of waste that are often

referred to as the “7 deadly muda.”
• Technically, there are more than 7 forms of waste
but if you can remember these you will capture over
90% of your waste.

1. Defects
2. Overproduction
3. Over-Processing
4. Inventory
5. Motion
6. Transportation
7. Waiting
112
• The “7 deadly muda” are very important to understand.
They are the best way to identify the COPQ.
• The presence of any muda causes many other forms of
inefficiencies and hidden factories to manifest themselves.
• There are four key

categories of costs related
to muda:
1. Costs Related to Production
2. Costs Related to Prevention
3. Costs Related to Detection
4. Costs Related to Obligation
113
COPQ: Costs Related to Production
• Costs related to production are the direct costs of

the presence of muda. These forms of COPQ are
usually understood and easily observable. They are
in fact the “7 deadly muda” themselves.
1. Defects

2. Overproduction
3. Over-Processing
4. Inventory
5. Motion
6. Transportation
7. Waiting
114
COPQ: Costs Related to Prevention
• Costs related to the prevention of muda are those

associated with trying to reduce or eliminate any of
the “7 deadly muda.”
• Costs for error proofing methods or devices
• Costs for process improvement and quality programs

• Costs for training and certifications
• etc.
• Any costs directly associated with the prevention of

waste and defects should be included in the COPQ
calculation.
115
COPQ: Costs Related to Detection
• Costs related to the detection of muda are those

associated with trying to find or observe any of the
“7 deadly muda.”
• Costs for sampling
• Costs for quality control check points

• Costs for inspection costs
• Costs for cycle counts or inventory accuracy inspections
• etc.
• Any costs directly associated with the detection of
waste and defects should be included in the COPQ
calculation.
116
COPQ: Costs Related to Obligation
• Costs related to obligation are those associated

with addressing the muda that reaches a customer.
• Repair costs
• Warranty costs
• Replacement costs

• Customer returns and customer service overhead
• etc.
• Any costs directly associated with customer

obligations should be included in the COPQ
calculation.
117
COPQ: Types of Cost
• There are two types of costs to be considered when

determining COPQ
1. Hard Costs
• Tangible costs that can be traced to the income statement
2. Soft Costs
• Intangible costs: avoidance, opportunity costs, lost revenue etc.

• Calculating the COPQ
1. Determine the types of waste that are present in your
process
2. Estimate the frequency of waste that occurs
3. Estimate the cost per event, item, or time frame
4. Do the math.
118
1.2.5 Pareto Charts and Analysis
119
Pareto Principle
• The Pareto principle is commonly known as the

“law of the vital few” or “80:20 rule.”
• It means that the majority (approximately 80%) of
effects come from a few (approximately 20%) of the
causes.

• This principle was first introduced in early 1900s and
has been applied as a rule of thumb in various
areas.
• Example of applying the Pareto principle:
• 80% of the defects of a process come from 20% of the
causes.
• 80% of sales come from 20% of customers.
120
Pareto Principle
• The Pareto principle helps us to focus on the vital

few items that have the most significant impact.
• In concept, it also helps us to prioritize potential
improvement efforts.
• Since this 80:20 rule was originally based upon the

works of Wilfried Fritz Pareto (or Vilfredo Pareto),
the Pareto principle and references to it should be
capitalized because Pareto refers to a person
(proper noun).
• Mr. Pareto is also credited for many works associated with
the 80:20, some more loosely than others:
• Pareto’s Law
• Pareto efficiency
• Pareto distribution etc.
121
Pareto Charts
• A Pareto chart is a chart of descending bars with

an ascending cumulative line on the top.
• Sum or Count:
The descending bars on a Pareto chart may be set on a
scale that represents the total of all bars or relative to the
biggest bucket, depending on the software you are using.

• Percent to Total: A Pareto chart shows the percentage to
the total for individual bars.
• Cumulative Percentage: A Pareto chart also shows the

cumulative percentage of each additional bar. The data
points of all cumulative percentages are connected into an
ascending line on the top of all bars.
122
Pareto Charts
• Case study time!

• Next we will use Minitab, SigmaXL, and JMP to run Pareto
charts on exactly the same data set.
• The following table shows the count of defective products by
team.
• Input the tabled data below into your software program and

follow the instructions over the next few pages to run Pareto
charts in the appropriate software.
123
Create a Pareto Chart in Minitab
• Steps to generate a Pareto chart using Minitab:
1. Open the spreadsheet with the count data for individual

categories.
2. Click on Stat → Quality → Pareto Chart.

3. A new window with the title “Pareto Chart” pops up.
4. Select “Category” into the box “Defects or attribute data

in” and “Count” into the box “Frequency in.”
5. Click “OK.”
6. The Pareto chart is created in a newly-generated window.
124

125

• The above Pareto chart generated in Minitab presents the count of defective
products by team.
• The bars are descending on a scale with the peak at 50, which is
approximately the total count of all defective products for all teams.
• The table below the chart contains counts, individual percentages, and
cumulative percentages.
• The cumulative percentages are the red data points driving the red line that
spans across the graphic.
126
Create Pareto Chart in SigmaXL
• Steps to generate a Pareto chart using SigmaXL:
1. Open the spreadsheet with the count data for individual
categories.
2. Select both columns of “Count” and “Category.”
3. Click SigmaXL → Graphical Tools → Basic Pareto Chart.

4. A new window named “Pareto Chart” pops up.
5. Click “Next>>.”
6. A new window named “Basic Pareto Chart” pops up.
7. Select “Category” as the “Pareto Category (X)” and “Count”
as the “Optional Numeric Count (Y).
8. Click “Finish.”
9. The Pareto chart is created in a new tab.
127

128

• The above Pareto chart generated in SigmaXL presents the count of
defective products by team.
• The bars are descending on a scale with the peak at 25, which is
approximately the size of the largest bar.
• Compared with Minitab, it is a bit more difficult to ascertain the total
number of defective items in the Pareto chart created in SigmaXL.
129
Create Pareto Chart in JMP
• Steps to generate a Pareto chart using JMP:
1. Open the JMP data table of the count data for individual
categories.
2. Click Graph → Pareto Plot.

3. In the new window, select “Count” as “Frequency” and
“Category” as “Y, cause.”
4. Click “OK.”
5. The Pareto chart is created in a new window.
130

131

• The above Pareto chart generated in JMP presents the count of
defective products by team.
• The bars are descending on a scale with the peak at 45, approximately
the total count of all defective products for all teams.
• The cumulative percentages make the black line spanning across the
graphic on top of the bars.
132
Pareto Analysis
• The Pareto analysis is used to identify the root

causes by using multiple Pareto charts.
• In Pareto analysis, we drill down into the bigger
buckets of defects and identify the root causes of
defects that contribute heavily to total defects.

• This "drill down" approach effectively solves a
significant portion of the problem.
• Next you will see an example of three-level Pareto
analysis.
• The second-level Pareto is a Pareto chart that is a subset of
the tallest bar on the first Pareto.
• The third-level Pareto is a subset of the tallest bar of the
second-level Pareto.
133
Pareto Analysis: First Level

• First-level Pareto
• Shows the count of defective items by team
• Next level will only show the defective items of team 4
134
Pareto Analysis: Second Level

• Second-level Pareto
• Shows the count of the defective items by section for only team 4
• Next level will only show the defective items of section 3
135
Pareto Analysis: Third Level

• Third-level Pareto
• Shows the count of defective items by associate for only section 3 of
team 4
• Next level will only show the defective items of Dave
136
Pareto Analysis: Conclusion
• After drilling down three levels we find that most of the defective
products are from Dave who is in Section 3 of Team 4.
• Determining what Dave might be doing differently and solving that

problem can potentially fix about 30% of the entire defective
products (13/44).

137
1.3 Six Sigma Projects
138

1.2.3 QFD
139
140
Six Sigma Metrics
• There are many Six Sigma metrics and/or measures

of performance used by Six Sigma practitioners.
• In addition to the ones we will cover here, several
others (Sigma level, Cp, Cpk, Pp, Ppk, takt time, cycle time,
utilization etc.) will be covered in other modules

throughout this training.
• The Six Sigma metrics of interest here in the define
phase are:
• Defects per Unit (DPU)
• Defects per Million Opportunities (DPMO)
• Yield (Y)
• Rolled Throughput Yield (RTY).
141
Defects per Unit: DPU
• DPU stands for “Defects per Unit“

• DPU is the basis for calculating DPMO and RTY,
which we will cover in the next few pages.
• DPU is found by dividing total defects by total units.

• DPU = D/U
• For example, if you have a process step that

produces an average of 65 defects for every 598
units, then your DPU = 65/598 = 0.109.
142
DPMO: Defects per Million Opportunities
• DPMO is one of the few important Six Sigma

metrics that you should get comfortable with if you
are associated with Six Sigma.
• In order to understand DPMO it is best if you first

understand both the nomenclature and the nuances
such as the difference between defect and
defective.
• Nomenclature
• Defects = D
• Unit = U
• Opportunity to have a defect = O
143
• In order to properly discuss DPMO, we must first explore
the differences between "defects" and "defective."
• Defective
• Defective suggests that the value or function of the entire unit or
product has been compromised.
• Defective items will always have at least one defect. Typically,
however, it takes multiple defects and/or critical defects to cause

an item to be defective.
• Defect
• A defect is an error, mistake, flaw, fault, or some type of
imperfection that reduces the value of a product or unit.
• A single defect may or may not render the product or unit
"defective" depending on the specifications of the customer.
• Summary
• Defect means that part of a unit is bad.
• Defective means that the whole unit is bad.
144
• Now let us turn our attention to defining

"opportunities" so that we can fully understand
Defects per Million Opportunities (DPMO).
• Opportunities

• Opportunities are the total number of possible defects.
• Therefore, if a unit has 6 possible defects, then each unit
produced is equal to 6 defect opportunities.
• If we produce 100 units, then there are 600 defect
opportunities.
145
• Calculating Defects per Million Opportunities
• The equation is DPMO = (D/(U × O)) × 1,000,000

• Example: Let us assume:
• There are 6 defect opportunities per unit

• There are an average of 4 defects every 100 units.
• Opportunities = 6 × 100 = 600
• Defect rate = 4/600
• DPMO = 4/600 × 1,000,000 = 6,667
146
• What is the reason or significance of 1,000,000?

• Converting defect rates to a per million value
becomes necessary when the performance of your
process approaches Six Sigma.

• When this happens, the number of defects shrinks
to virtually nothing. In fact, if you recall from the
“What is Six Sigma” module, sigma is 3.4 defects
per million opportunities.
• By using 1,000,000 opportunities as the barometer

we have the resolution in the measurement to count
defects all the way up to Six Sigma.
147
RTY: Rolled Throughput Yield
• Rolled Throughput Yield (RTY) is a process
performance measure that provides insight into the
cumulative effects of an entire process.
• RTY measures the yield for each of several process

steps and provides the probability that a unit will come

through that process defect-free.
• RTY allows us to expose the "hidden factory" by

providing visibility into the yield of each process step.
• This helps us identify the poorest performing process

steps and gives us clues into where to look to find the
most impactful process improvement opportunities.
148
• Calculating RTY:
• RTY is found by multiplying the yields of each
process step.
• Let us take the 5-step process below and calculate the

RTY using the multiplication method mentioned above.
• The calculation is: RTY = 0.90 × 0.91 × 0.99 × 0.98 × 0.97 =

0.77
• Therefore, RTY = 77%.

149
• You may have noticed that in order to calculate RTY

we must determine the yield for each process step.
• Before we get into calculating yield, there are a few

abbreviations that need to be declared.

• Abbreviations
• Defects = D
• Unit = U
• Defects per Unit = DPU
• Yield = Y
• e = 2.71828 (mathematical constant)
150
• Calculating Yield
• The yield of a process step is the success rate of
that step or the probability that the process step
produces no defects.
• In order to calculate the yield, we need to know the DPU

and then we can apply it to the yield equation below.
• Example
• Let us assume a process step has a DPU of 0.109

(65/598)
• Y = 2.718 ^ -0.109 = 0.8967. Rounded, Y = 90%.

151
• Below is a table using the above process yield data

that we used in the earlier RTY calculation.

• This table allows us to see the DPU and yield of
each step as well as the RTY for the whole process.
152
RTY: Using an Estimate of Yield
• Calculating RTY using yield estimation

• It is possible to “estimate” yield by taking the inverse of DPU

or simply subtracting DPU from 1.
• Yield Estimation = 1 − DPU
• Yield Estimate for process step 1: 1 − 0.10870 = 0.90
• RTY using the Yield Estimation Method
• RTY = 0.90 × 0.91 × 0.99 × 0.98 × 0.97 = 0.77 = 77%
153
154
Business Case and Project Charter
• Earlier we stated that DMAIC is a structured and

rigorous methodology designed to be repeatedly
applied to any process in order to achieve Six
Sigma.

• We also stated that DMAIC was a methodology that
refers to 5 phases of a project.
• Define, Measure, Analyze, Improve, and Control
• Given that the premise of the DMAIC methodology is

project-based, we must take the necessary steps to
define and initiate a project, hence the need for. . .
• Project Charters
155
Project Charter
• The purpose of a project charter is to provide vital

information about a project in a quick and easy-to-
comprehend manner.
• Project charters are used to get approval and buy-in

for projects and initiatives as well as declaring:
• The scope of work
• Project teams
• Decision authorities
• Project lead
• Success measures
156
Project Charter
• Key Elements of Project Charters

• Title
• Project Lead
• Business Case
• Problem Statement
• Project Objective

• Primary and Secondary Metrics
• Project Scope
• Project Timeline
• Project Constraints
• Project Team
• Stakeholders
• Approvers
• Constraints
• Dependencies
• Risks
157
Project Charter: Key Elements
• Title
• Projects should have a name, title, or some reference that
identifies them.
• Branding can be an important ingredient in the success of a
project so be sure your project has a reference name or title.

• Leader
• Any projects needs a declared leader or someone who is
responsible for project’s execution and success.
• You may hear references to RACI throughout in your Six
Sigma journey.
• RACI stands for Responsible, Accountable, Consulted,
Informed and identifies the people that play those roles.
• Every project must have declared leaders indicating who is
responsible and who is accountable.
158
• Business Case
• A business case is the quantifiable reason why
the project is important.
• Business cases help shed light on problems. They

explain why a business should care.
• Business cases must be quantified and stated

succinctly.
• COPQ is a key method of quantification for any

business case.
159
• Problem Statement and Objective

• A properly written problem statement has an objective
statement woven into it.
• There should be no question as to the current state or the
goal.

• A gap should be declared, the gap being the difference
between the present state and the goal state.
• The project objective should be to close the gap or reduce
the gap by some reasonable amount.
• Valuation or COPQ is the monetary value assigned to the

gap.
• Lastly, a well-written problem statement refers to a timeline
expected to be met.
160
Project Charter: Problem Statement Examples
• Currently, process defect rates are 17% with a goal of
2%. This represents a gap of 15%, costing the business
$7.4 million dollars. The goal of this project is to reduce
this gap by 50% before Nov 2010 putting process defect
rates at 9.5% and saving $3.7MM.

• Process cycle time has averaged 64 minutes since Q1
2009. However, production requirements put the cycle
time goals at 48 min. This 16-min gap is estimated to
cost the business $296,000. The goal of this project is
reduce cycle time by 16 min. by Q4 2010 and capture all
$296,000 cost savings.
161
• Metrics
• A measure of success is an absolute for any project.
• Metrics give clarity to the purpose of the work.
• Metrics establish how the initiative will be judged.

• Metrics establish a baseline or “starting point.”
• For Six Sigma projects…metrics are mandatory!
162
• Primary Metric
• The primary metric is a generic term for a Six Sigma
project's most important measure of success. The primary
metric is defined by the Black Belt, GB, MBB, or Champion.
• A primary metric is an absolute MUST for any project and it
should not be taken lightly. Here are a few characteristics of
good primary metrics.

• Primary metrics should be:
• tied to the problem statement
• measureable
• expressed with an equation
• aligned to business objectives
• tracked at the proper frequency (hourly, daily, weekly,
monthly etc.)
• expressed pictorially over time with a run chart, time
series, or control chart
• validated with an MSA.
163
• The primary metric is the reason for your work.
• It is the success indicator.
• It is your beacon.

• The primary metric is of utmost importance and
should be improved, but not at the expense of your
secondary metric.
164
• Secondary Metric
• The secondary metric is the thing you do not want
sacrificed on behalf of a primary improvement.
• A secondary metric is one that makes sure problems are not
just "changing forms" or "moving around."
• The secondary metric keeps us honest and ensures we are

not sacrificing too much for our primary metric.
• If your primary metric is a cost or speed metric, then your
secondary metric should probably be some quality
measure.
• Example: If you were accountable for saving energy in an
office building and your primary metric was energy
consumption then you could shut off all the lights and the
HVAC system and save tons of energy. . .except that your
secondary metrics are probably comfort and functionality of
the work environment.
165
• Elements of a Good Project Charters (continued)

• Scope Statement – defined by high-level process map
• Stakeholders Identified – who is affected by the project
• Approval Authorities Identified – who makes the final call

• Review Committees Defined – who is on the review team
• Risks and Dependencies Highlighted – identify risks and

critical path items
• Project Team Declared – declare team members
• Project Timeline Estimated – set high-level timeline

expectations.
166
167
Project Team Selection
• Six Sigma project team selection is the cornerstone

of a successful Six Sigma project.
• Teams and Team Success

• A team is a group of people who share complementary skills
and experience.
• A team will be dedicated to consistent objectives.
• Winning teams share similar and coordinated goals.
• Teams often execute common methods or approaches.
• Team members hold each other accountable for achieving

shared goals.
168
• What makes a team successful?

• Shared goals
• Commitment
• Leadership

• Respect
• Effective communication
• Autonomy
• Diversity (capabilities, knowledge, skills,

experience etc.)
• Adequate resources.
169
• Keys to Team Success
• Agreed focus on the goal or the problem at hand
• Focus on problems that have meaning to the business
• Focus on solvable problems within the scope of influence; a
successful team does not seek unattainable solutions.
• Team Selection
• Selected teammates have proper skills and knowledge

• Adequately engaged management
• Appropriate support and guidance from their direct leader
• Successful teams use reliable methods
• Follow the prescribed DMAIC methodology
• Manage data, information, and statistical evidence
• Successful teams always have exceeds players
• Winning teams typically
• Have unusually high standards.
• Have greater expectations of themselves and each other.
• Do not settle for average or even above average results.
170
• Principles of Team Selection:

• Select team members based on
• Skills required to achieve the objective
• Experience (Subject Matter Expertise)
• Availability and willingness to participate

• Team size (usually 4–8 members)
• Don’t go at it alone!
• Don’t get too many cooks in the kitchen!
• Members’ ability to navigate
• The process
• The company
• The political landscape
• Be sure to consider the inputs of others

• Heed advice
• Seek guidance
171
Project Team Development
• All teams experience the following four stages of
development. It is helpful to understand these phases so
that you can anticipate what your team is going to
experience.
• The four stages of team development process:

• Forming
• Storming
• Norming
• Performing
• Teammates seek something different at each stage:

• In the forming stage they seek inclusion
• In the storming stage they seek direction and guidance
• In the norming stage they seek agreement
• In the performing stage they seek results.
172
• Patterns of a team in the Forming stage:

• Roles and responsibilities are unclear
• Process and procedures are ignored
• Scope and parameter setting is loosely attempted
• Discussions are vague and frustrating
•

There is a high dependence on leadership for guidance
• Patterns of a team in the Storming stage:
• Attempts to skip the research and jump to solutions
• Impatience for some team members regarding lack of
progress
• Arguments about decisions and actions of the team
• Team members establish their position
• Subgroups or small teams form
• Power struggles exist and resistance is present
173
• Patterns of a team in the Norming stage:

• Agreement and consensus start to form
• Roles and responsibilities are accepted
• Team members’ engagement increases
• Social relationships begin to form
•

The leader becomes more enabling and shares authority
• Patterns of a team in the Performing stage:

• Team is directionally aware and agrees on objectives
• Team is autonomous
• Disagreements are resolved within the team
• Team forms above average expectations of performance
174
• Well-structured and energized project teams are the

essential components of any successful Six Sigma
project.
• To have better chances of executing the project

successfully, you will need to understand and
effectively manage the team development process.
175
176
Risk
Risk is defined as a future event that can impact the

task/project if it occurs.

177
What is Project Risk Management?
• The main purpose of risk management is to
foresee potential risks that may inhibit the project
deliverables from being delivered on time, within
budget, and at the appropriate level of quality, and
then to mitigate these risks by creating,
implementing, and monitoring contingency plans.

• Risk management is concerned with identifying,
assessing, and monitoring project risks before they
develop into issues and impact the project.
• Risk analysis helps to identify and manage

potential problems that could impact key business
initiatives or project goals.
178
Three Basic Parameters of Risk Analysis
• Risk Assessment:
The process of identifying and evaluating risks,
whether in absolute or relative terms.
• Risk Management:

Project risk management is the effort of responding
to risks throughout the life of a project and in the
interest of meeting project goals and objectives.
• Risk Communication:
Communication plays a vital role in the risk analysis
process because it leads to a good understanding of
risk assessment and management decisions.
179
Why is Risk Analysis Necessary?
What can happen if you omit the risk analysis?
• Vulnerabilities cannot be detected

• Mitigation plans are introduced without proper
justification

• Customer dissatisfaction
• Not meeting project goals
• Remake the whole system
• Huge cost and time loss
180
Project Risk Analysis Steps
The project risk analysis process consists of the following
steps that evolve through the life cycle of a project.
• Risk Identification:
• Identify risks and risk categories, group risks, and define ownership.
• Risk Assessment:
• Evaluate and estimate the possible impacts and interactions of risks.

• Response Planning:
• Define mitigation and reaction plans.
• Mitigation Actions:
• Implement action plans and integrate them into the project.
• Tracking and Reporting:
• Provide visibility to all risks.
• Closing:
• Close the identified risk.
181
Risk Identification
The first action of risk management is the identification
of individual events that the project may encounter
during its lifecycle.
The identification step comprises:

• Identify the risks
• Categorize the risks
• Match the identified risks to categories
• Define ownership for managing the risks.
182
Risk Identification
Source of Risk:
Identification of risk sources provides a basis for
systematically examining changing situations over
time to uncover circumstances that impact the ability
of the project to meet its objectives.

183
Risk Identification
Source of
Risk Description
Human
The risks originated from human resources (e.g., availability, skill etc.)
Resources
Physical The risks originated from physical resources (e.g., hardware or software,
Resources availability of the required number at the right time etc.)
The risks originated from technology (e.g., development environment, new or

Technology
complex technologies, performance requirements, tools etc.)
The risks are associated with a supplier (e.g., delays in supplies, capability of
Suppliers
suppliers etc.)
The risks derived from the customer (e.g., unclear requirements, requirement
Customer
volatility, change in project scope, delays in response etc.)
The risks are associated with information security, security of personnel, security
Security
of assets, and security of intellectual property
Legal The risks are associated with legal issues that may impact the project
Project The risks are associated with project management processes, organizational
management maturity, and ability
184
Risk Identification
Risk Parameters:
Parameters for evaluating, categorizing, and prioritizing
risks include the following:
• Risk likelihood (i.e., probability of risk occurrence)

• Risk consequence (i.e., impact and severity of risk
occurrence)
• Thresholds to trigger management activities.
185
Risk Assessment
The risk assessment consists of evaluating the range

of possible impacts should the risk occur.
Follow these steps when assessing risks:

1) Define the various impacts of each risk

2) Rate each impact based on a logical severity level
3) Sort and evaluate risks by severity level
4) Determine if any controls already exist
5) Define potential mitigation actions.
186
Risk Mitigation Planning
The risk owners are responsible for planning and
implementing mitigation actions with support from the
project team.
• All team members, inclusive of partners and suppliers,
may be requested to identify and develop mitigation
measures for identified risks.

• The project core team members are responsible for
identifying an appropriate action owner for each identified
risk.
• After mitigation actions are defined, the project core team
will review the actions.
• The risk owner must track all mitigation actions and
expected completion dates.
• The risk owner and the project core team members must
hold all action owners accountable for the risk mitigation
planning.
187
Risk Mitigation Action Implementation
• The action implementation is the responsibility of

the risk owner.
• The action owners are responsible for the execution

of the tasks or activities necessary to complete the

mitigation action and eliminate or minimize the risk.
• The risk owner or the project manager will monitor

completion dates of the mitigation action
implementation.
188
Risk Occurrence and Contingency Plans
• Whenever any risk occurs, the project team should

implement contingency plans to ensure that
project deliverables can be met.
• The details of each occurrence should be recorded

in the risk register or other tracking tool.
• The risk register or risk management plan (see next

slide) will be maintained by the project manager and
reviewed on a regular basis.
189
Risk Tracking and Reporting
• Risk tracking and reporting provides critical visibility to all risks.
• Risk owners must report on the status of their mitigation actions.
• Depending on the risk severity, project managers need to report
the risk status of each category of risk to senior management.
This template is available in the “Lean Sigma Corporation Templates.xls” file

190
Risk Closure
• The risk owners are responsible for recommending

the risk closure to the project manager.
• A risk is closed only when the item is not considered

a risk to the project anymore.

• When a risk is closed, the project manager needs to
update the risk status in the lessons learned
document.
191
Risk Analysis Features
The risk analysis should be:

• Systematic
• Comprehensive
• Data driven
• Adherent to evidence

• Logically sound
• Practically acceptable
• Open to critique
• Easy to understand.
192
Project Risk Analysis Advantages
• Helps strategic and business planning

• Meets customer requirements
• Reduces schedule slips and cost overruns
• Promotes an effective usage of resources
• Promotes continuous improvement

• Helps to achieve project goals
• Minimizes surprises from customers and stakeholders
• Allows a quick grasp of new opportunities
• Enhances communication
• Reassures stakeholders that the project stays on track.
193
194
What is Project Management?
• Project management is the process of defining,

planning, organizing, managing, leading, securing,
and optimizing the resources to achieve a set of
planned goals and objectives.

• It is the application of knowledge, skills, tools, and
techniques to project activities in order to meet
project requirements.
195
What is a Project Plan?
• A project plan is a crucial step in project

management for achieving a project’s goals.
• A project plan is a formal approved document used to

guide and execute project tasks.

• It provides an overall framework for managing project
tasks, schedules, and costs.
• A project plan is a coordinating tool and

communication device that helps teams, contractors,
customers, and organizations define the crucial
aspects of a project or program.
196
Project Planning Stages
1. Determine project scope and objectives: Explore
opportunities, identify and prioritize needs, consider
project solutions.
2. Plan the project: Identify input and resources
requirements such as human resources, materials,
software, hardware, and budgets.

3. Prepare the project proposal: Based on stakeholder
feedback, plan the necessary resources, timeline,
budget etc.
4. Implement the project: Implement the project by
engaging responsible resources and parties. Ensure
execution and compliance of the defined plans.
5. Evaluate the project: Regularly review progress and
results. Measure the project’s effectiveness against
quantifiable requirements.
197
Planning and Scheduling Objectives
• To optimize the use of resources (both human and other
resources).
• To increase productivity.
• To achieve desired schedules and deliverables.

• To establish an approach to minimize long-term
maintenance costs.
• To minimize the chaos and productivity losses resulting

from planned production schedules, priority changes,
and non-availability of resources.
• To assess current needs and future challenges.

198
Project Planning Activities
• Statement of work (SoW)

• Work breakdown structure (WBS)
• Resource estimation plan
• Project schedule
• Budget or financial plan

• Communication plan
• Risk management plan
199
Project Planning Activities: SOW
Statement of Work (SOW) is a formal document often
accompanying a contract that outlines specific expectations,
limitations, resources and work guidelines intended to
define the work or project.
SOW’s:

• Define the scope of the project.
• Establish expectations and parameters of the project.
• Identify technical requirements for the project.
• Provide guidance on materials to be used.
• Establish timeline expectations.
• Etc.
200
Project Planning Activities: WBS
Work Breakdown Structure (WBS) is a decomposition of
project components into small and logical bodies of work or
tasks.
Work Breakdown Structures:
• Identify all required components of a project.
• Cascades components into sub-components and tasks.

• WBS’s are not by themselves project plans or schedules
but they are a necessary step to help establish project
plans and timelines.
• WBS’s also enable logical reporting and summarizations of
project progress.
201
Project Planning Activities: WBS
Work Breakdown Structure Example: below is an

example of a WBS for the assembly of a bicycle.

202
Project Planning Activities: Resource Planning
Resource Estimation Plan

• Estimate Resource Requirements (Use Your WBS)
• Parts, Hardware, Software, Human Resources etc.
• Plan resources
• Establish who is responsible for what and when.
• Determine quantity requirements and delivery dates etc.

203
Project Planning Activities: Project Scheduling
Project Scheduling:
• Assign beginning times to each activity in the WBS
(days are used for start and durations times in example below).
• Assign duration times to each activity in the WBS.

• Identify People Responsible and set completion dates.
• Represent schedules as Gantt charts or network diagrams.
• Identify critical dependencies between tasks.

204
Project Planning Activities: Project Scheduling
Project Schedule – Gantt Chart:
• The advantage of a Gantt chart is its ability to display the status
of each task/activity at a glance.
• Because it is a graphic representation, it is easy to demonstrate
the schedule of tasks and timelines to all the stakeholders.

205
Project Scheduling: Preparing a Gantt Chart in Excel
There are many tools available to create Gantt charts. Some of
the most common and compatible for businesses are: Microsoft
Project, Visio and Excel. In Microsoft Excel we use the bicycle
example:
1. Select rows & columns of your tasks, start times and durations.

206
2. On the Insert tab select Bar > Stacked Bar.

3. On the Design tab choose a design style from the style dropdown.

207
4. Activate the “start tasks” on your graph.
5. From the format tab select “format selection”.
5 4

208
6. Select Fill > No Fill.
7. Select Shadow > Presets > No Shadow > Close.
7
6

209
8. Format your new Gantt chart to your own preferences.

210
Project Scheduling: Critical Path Method
Critical Path Method (CPM) is a project modeling

technique used to identify the set of activities that
are most influential to a projects completion timeline.
• Critical Path tasks are those that others are dependent upon.
• Project timelines cannot be shortened without shortening the tasks
or activities that are identified as critical path items.

Steps To Using The Critical Path Method
1. List all activities necessary to complete the project.
2. Determine the time or duration of each activity.
3. Identify the dependencies between the activities.
Use Your Work Breakdown Structure (WBS).
211
Critical Path Method Example: Let’s continue with our
bicycle example. Note that the tasks in the WBS are
already set to start based on their dependencies.

• For example, Installing the wheels in step 1.1.5 requires the whole set of
steps in 1.2 to be completed so 1.1.5 starts on day 14 when the wheels
are ready.
212
Continuing this example; the last task, 1.3.4 Adjusting the Brake System,
completes on day 23 and can’t begin until day 21 because the brake
system must be attached to the bike frame before being calibrated.

213
Earlier if you noticed when we conducted our resource plan that we were
out of tire tubes then you might have managed your resources differently.
Let’s assume the tubes take 21 days to receive from the supplier. Below is
the adjustment to our Gantt chart. Notice that the project timeline is
pushed out by 8 days, finishing on day 31 instead of day 23.

214
Project Scheduling: PERT
Program Evaluation & Review Technique (PERT): is
a method of evaluation that can be applied to time or
cost.
• PERT provides a weighted assessment of time or cost
• PERT uses 3 parameters for estimation:
• Optimistic or Best Case Scenario represented by ‘O’.

• Pessimistic or Worst Case Scenario represented by ‘P’.
• Most Likely Scenario represented by ‘ML’.
• PERT Equation:
(O + (4*ML) + P) / 6
• The PERT equation provides for heavier weighting of the most
likely scenario but also considers the best and worst cases. In
the event that a best or worst case scenario is an extreme
situation, the PERT will account for it in a weighted manner.
215
Let’s apply the PERT formula to our estimates of

receiving our needed tire tubes from the supplier and
then change our schedule based on the result.
• Your experience with this supplier tells you that they typically over
estimate the time required to deliver. Therefore, their 21 day lead
time on the tubes should be considered a “worst case scenario”.

• Procurement has indicated that tubes have arrived in as few as 6
days from this supplier. So, your best case scenario is 6 days.
• The most likely scenario you decide will be the median of this
suppliers delivery time which is 10 days.
• Therefore:
(O + (4*ML) + P) / 6
(6 + (4*10) + 21) / 6
(6 + 40 + 21) / 6
67 / 6 = 11
216
Using 11 days in your project plan instead of 21 gives you a more
confident estimate of time and actually removes tire tubes as the
primary detracting critical path item.

217
Budget or Financial Plan

• Planned expenses
• Planned revenues
• Budget forecast

218
Communication Plan
• Establish communication procedures among
management, team members, and relevant
stakeholders.
• Determine the communication schedule.
• Define the acceptable modes of communication.

219
Risk Management Plan
• Identify the sources of project
risks and estimate the effects of
those risks.
• Risks might arise from new
technology, availability of
resources, lack of inputs from
customers, business risks etc.

• Assess the impact of risk to the
customers/stakeholders.
• Calculate the probability of risk
occurrence based on previous
similar projects or industry
benchmarks
• Initiate mitigation and contingency plans
• Review risks on a periodic basis.
220
Project Planning Tools Advantages
• Project planning tools are very useful to organize and
communicate project plans, status, and projections.
• They help link tasks and sub-tasks or other work elements to get
a whole view of what needs to be accomplished.
• They allow a more objective comparison of alternative solutions

and provide consistent coverage of responsibilities.

• They allow for effective scope control and change management.
• They facilitate effective communication with all project

participants and stakeholders.
• They help define management reviews.
• They act as an effective monitoring mechanism for the project.
• They establish project baselines for progress reviews and control

points.
221
222

1.4 Lean Fundamentals

1.2.3 QFD
223
1.4.1 Lean and Six Sigma
224
What is Lean?
• A lean enterprise intends to eliminate waste and

allow only value to be pulled through its system.
• Lean manufacturing is characterized by:
• Identifying and driving value
• Establishing flow and pull systems

• Creating production availability and flexibility
• Zero waste
• Waste Elimination
• Waste identification and elimination
is critical to any successful lean
enterprise.
• Elimination of waste enables flow,
drives value, cuts cost, and provides
flexible and available production.
225
The 5 Lean Principles
• The following 5 principles of lean are taken from the

book Lean Thinking (1996) by James P. Womack
and Daniel T. Jones.
1. Specify value desired by customers.
2. Identify the value stream.

3. Make the product flow continuous.
4. Introduce pull systems where continuous flow is possible.
5. Manage toward perfection so that the number of steps

and the amount of time and information needed to serve
the customer continually falls.
226
Lean & Six Sigma
Six • Lean and Six Sigma both have

Lean the objectives of producing high
Sigma
value (quality) at lower costs
Eliminate Eliminate (efficiency).
Waste Defects
• They approach these objectives in

Process Process
Flow Yield somewhat different manners but
in the end, both Lean and Six
Pull Capable Sigma drive out waste, reduce
Systems Systems defects, improve processes, and
stabilize the production
Easy Rigorous environment.
Visual Analytical
Approach Approach
• Lean and Six Sigma are a perfect

combination of tools for improving
Quality & Value for the Customer
Efficiency for the Business quality and efficiency.
227
228

1.4.2 History of Lean
History of Lean
Polaroid

229
History of Lean
• Lean thinking originated, as far as is known, the 1400s.
• Henry Ford established the first mass production system

in 1913 by combining standard parts, conveyors, and
work flow.

• Decades later, Kiichiro Toyoda and Taiichi Ohno at
Toyota improved and implemented various new concepts
and tools (e.g., value stream, takt time, kanban etc.)
based on Ford’s effort.
• Toyota developed what is known today as the Toyota

Production System (TPS) based on lean principles.
230
History of Lean
• Starting in the mid 1990s, Lean became extensively

recognized and implemented when more and more
Fortune 100 companies began to adopt Lean and
Six Sigma.

• The term “Lean manufacturing” was introduced by
James Womack in the 1990s.
• Lean and Six Sigma share similar objectives, work

hand in hand, and have benefited from one another
in the past 30 years.
231
1.4.3 Seven Deadly Muda
232
The 7 Deadly Muda
• The Japanese word for waste is “muda.”
• There are 7 commonly recognized forms of waste,

often referred to as the “7 deadly muda.”
1. Defects

2. Overproduction
3. Over-Processing
4. Inventory
5. Motion
6. Transportation
7. Waiting
233
The 7 Deadly Muda: Defects
• Defects or defectives are an

obvious waste for any working
environment or production
system.

• Defects require rework during production and/or after
the product is returned from an unhappy customer.
• Some defects are difficult to solve and they create

“workarounds” and hidden factories.
• Eliminating defects is a sure way to improve product

quality, customer satisfaction, and production costs.
234
The 7 Deadly Muda: Overproduction
• Overproduction is
wasteful because your
system expends energy
and resources to
produce more materials
than the customer or

next function requires.
• Overproduction is one of the most detrimental of the

seven deadly muda because it leads to many others:
• Inventory
• Transportation
• Waiting etc.
235
The 7 Deadly Muda: Over-processing
• Over-processing
occurs any time more
work is done than is
required by the next
process step,
operation, or

consumer.
• Over-processing also includes being over capacity

(scheduling more workers than required or having
more machines than necessary).
• Another form of over processing can be buying tools

or software that are overkill (more precise, complex,
or expensive than required).
236
The 7 Deadly Muda: Inventory

• Inventory is an often overlooked waste. Look at the
picture above and imagine all the time, materials,
and logistics that went into establishing such an
abundance of inventory.
• If this were your personal business, and inventory

velocity was not matched with production, how
upset would you be?
237
The 7 Deadly Muda: Motion
• Motion is another form of waste
often occurring as a result of poor
setup, configuration, or operating
procedures.
• Wasted motion can be experienced

by machines or humans.
• Wasted motion is very common with

workers who are unaware of the
impact of small unnecessary
movements in repetitive tasks.
• Wasted motion is exaggerated by

repetition or recurring tasks.
238
The 7 Deadly Muda: Transportation
• Transportation is
considered wasteful
because it does
nothing to add value or
transform the product.
• Imagine for a moment

driving to and from
work twice before
getting out of your car
to go into work. . .
• That is waste in the form of transportation.
• The less driving you have to do, the better.
• In a similar way, the less transportation a product has to
endure, the better. There would be fewer opportunities for
delay, destruction, loss, damage etc.
239
The 7 Deadly Muda: Waiting

• Waiting is an obvious form of waste and is typically a
symptom of an upstream problem.
• Waiting is usually caused by inefficiency, bottlenecks, or
poorly-designed work flows within the value stream.
• Waiting can also be caused by inefficient administration.
• Reduction in waiting time will require thoughtful
applications of lean and process improvement.
240
241

1.4.4 Five-S (5S)
What is 5S?
• 5S is systematic method to organize, order, clean,
and standardize a workplace…and to keep it that
way!
• 5S is a methodology of organizing and improving the work
environment.
• 5S is summarized in five Japanese words all starting

with the letter S:
• Seiri (sorting)
• Seiton (straightening)
• Seiso (shining)
• Seiketsu (standardizing)
• Shisuke (sustaining)
• 5S was originally developed in Japan and is widely
used to optimize the workplace to increase
productivity and efficiency.
242
Five-S (5S)
Shine

Sort Set in Order
Standardize Sustain
243
Goals of 5S
• Reduced waste
• Reduced cost
• Establish a work environment that is:
• self-explaining
• self-ordering
• self-regulating

• self-improving.
• Where there is/are no more:
• Wandering and/or searching
• Waiting or delaying
• Secret hiding spots for tools
• Obstacles or detours
• Extra pieces, parts, materials etc.
• Injuries
• Waste.
244
Benefits of 5S Systems
• Reduced changeovers
• Reduced defects
• Reduced waste
• Reduced delays
• Reduced injuries

• Reduced breakdowns
• Reduced complaints
• Reduced red ink
• Higher quality
• Lower costs
• Safer work environment
• Greater associate and equipment capacity.
245
Reported Results of 5S Systems
• Cut in floor space: 60%

• Cut in flow distance: 80%
• Cut in accidents: 70%
• Cut in rack storage: 68%
• Cut in number of forklifts: 45%

• Cut in machine changeover time: 62%
• Cut in annual physical inventory time: 50%
• Cut in classroom training requirements: 55%
• Cut in nonconformance in assembly: 96%
• Increase in test yields: 50%
• Late deliveries: 0%
• Increase in throughput: 15%
246
Sorting (Seiri)
• Go through all the tools, parts,
equipment, supply, and material in
the workplace.
• Categorize them into two major

groups: needed and unneeded.

• Eliminate the unneeded items from
the workplace. Dispose of or
recycle those items.
• Keep the needed items and sort

them in the order of priority. When
in doubt…throw it out!
247
Straightening (Seiton)
• Straightening in 5S is also called

setting in order.
• Label each needed item.
• Store items at their best locations
so that the workers can find them

easily whenever they needed any
item.
• Reduce the motion and time
required to locate and obtain any
item whenever it is needed.
• Promote an efficient work flow
path.
• Use visual aids like the tool board
image on this page. 248
Shining (Seiso)
• Shining in 5S is also called sweeping.
• Clean the workplace thoroughly.
• Maintain the tidiness of the workplace.

• Make sure every item is located at the
specific location where it should be.
• Create the ownership in the team to

keep the work area clean and
organized.
249
Standardizing (Seiketsu)
• Standardize the workstation and the layout of tools,

equipment and parts.
• Create identical workstations with a consistent way

of storing the items at their specific locations so that

workers can be moved around to any workstation
any time and perform the same task.
250
Sustaining (Shisuke)
• Sustaining in 5S is also called self-discipline.
• Create the culture in the team to follow the first four
S’s consistently.
• Avoid falling back to the old ways of cluttered and
unorganized work environment.
• Keep the momentum of optimizing the workplace.

• Promote innovations of workplace improvement.
• Sustain the first fours S’s using:
• 5S Maps
• 5S Schedules
• 5S Job cycle charts
• Integration of regular work duties
• 5S Blitz schedules
• Daily workplace scans.
251
Simplified Summary of 5S
1. Sort – “when in doubt, move it out.”
2. Set in Order – Organize all necessary tools, parts, and

components of production. Use visual ordering
techniques wherever possible.

3. Shine – Clean machines and/or work areas. Set regular
cleaning schedules and responsibilities.
4. Standardize – Solidify previous three steps, make 5S a

regular part of the work environment and everyday life.
5. Sustain – Audit, manage, and comply with established

5S guidelines for your business or facility.
252
Five-S (5S)
• A few words about 5S and the Lean Enterprise
• As a method, 5S generates immediate improvements.
• 5S is one of many effective lean methods that create

observable results.
• It is tempting to implement 5S alone without

considering the entire value stream.
• However, it is advisable to consider a well-planned

lean manufacturing approach to the entire production
system.
253
254

2.0 Measure Phase
Green Belt Training: Measure Phase
2.1 Process Definition
2.1.1 Cause and Effect Diagrams
2.1.2 Cause and Effects Matrix
2.1.3 Process Mapping
2.1.4 FMEA: Failure Modes and Effects Analysis
2.1.5 Theory of Constraints
2.2 Six Sigma Statistics
2.2.1 Basic Statistics

2.2.2 Descriptive Statistics
2.2.3 Distributions and Normality
2.2.4 Graphical Analysis
2.3 Measurement System Analysis
2.3.1 Precision and Accuracy
2.3.2 Bias, Linearity, and Stability
2.3.3 Gage R&R
2.3.4 Variable and Attribute MSA
2.4 Process Capability
2.4.1 Capability Analysis
2.4.2 Concept of Stability
2.4.3 Attribute and Discrete Capability
2.4.4 Monitoring Techniques
255
256

2.3.3 Gage R&R
257
2.1.1 Cause and Effect Diagram
258
What is a Cause and Effect Diagram?
• A cause and effect diagram is also called a Fishbone
Diagram or Ishikawa Diagram. It was created by Kaoru
Ishikawa and is used to identify, organize, and display
the potential causes of a specific effect or event in a
graphical way similar to a fishbone.
• It illustrates the relationship between one specified event

(output) and its categorized potential causes (inputs) in a
visual and systematic way.
Affinity Affinity Affinity
Cause Cause Cause
Effect
Cause Cause Cause
Affinity Affinity Affinity

259
Major Categories of Potential Causes
• P4ME
• People: People who are involved in the process
• Methods: How the process is completed (e.g., procedures,
policies, regulations, laws)
• Machines: Equipment or tools needed to perform the
process

• Materials: Raw materials or information needed to do the
job
• Measurements: Data collected from the process for
inspection or evaluation
• Environment: Surroundings of the process (e.g., location,
time, culture).
260
How to Plot a Cause and Effect Diagram
• Step 1: Identify and define the effect/event being

analyzed.
• Clearly state the operational definition of the

effect/event of interest.

• The event can be the positive outcome desired or
negative problem targeted to solve.
• Enter the effect/event in the end box of the Fishbone

diagram and draw a spine pointed to it.
261
• Step 1

Effect or Event
Being Analyzed
262
• Step 2: Brainstorm the potential causes or factors of

the effect/event occurring.
• Identify any factors with a potential impact on the

effect/event and include them in this step.

• Put all the identified potential causes aside for use
later.
263
• Step 2
Cause Cause
Cause Cause

Cause Cause
Cause
Cause
Cause Cause
Cause
264
• Step 3: Identify the main categories of causes and
group the potential causes accordingly.
• Besides P4ME (i.e., people, methods, machines,

materials, measurements, and environment), you can
group potential causes into other customized

categories.
• Below each major category, you can define sub-

categories and then classify them to help you visualize
the potential causes.
• Enter each cause category in a box and connect the

box to the spine. Link each potential cause to its
corresponding cause category.
265
• Step 3
People Methods Machines
Cause Cause Cause

Effect or Event
Being Analyzed
Cause Cause Cause
Materials Measurements Environment
266
• Step 4: Analyze the cause and effect diagram.
• A cause and effect diagram includes all the possible

factors of the effect/event being analyzed.

• Use a Pareto chart to filter causes the project team
needs to focus on.
• Identify causes with high impact that the team can take
action upon.
• Determine how to measure causes and effects

quantitatively. Prepare for further statistical analysis.
267
Benefits to Using Cause and Effect Diagram
• Helps identify and sort the potential causes of an

effect occurring quickly.
• Provides a systematic way to brainstorm potential

causes effectively and efficiently.

• Identifies areas requiring data collection for further
quantitative analysis.
• Locates “low-hanging fruits.”
268
Limitation of Cause and Effect Diagrams
• A cause and effect diagram only provides qualitative

analysis of correlation between each cause and the
effect.
• One cause and effect diagram can only focus on

one effect or event at a time.
• Further statistical analysis is required to quantify the

relationship between various factors and the effect
and identify the root causes.
269
Cause and Effect Diagram Example
• Case study:
• A real estate company is interested to find the root causes of

high energy costs of its properties.
• The cause and effect diagram is used to identify, organize,

and analyze the potential root causes.
270
• Step 1: Identify and define the effect/event being

analyzed: high energy costs of buildings.

High Energy
Cost of Buildings
271
• Step 2: Brainstorm the potential causes or factors of

the high energy costs.
Bad energy Aged HVAC

consuming habits units
Poor

maintenance of Most lights are still on
HVAC units during non-business
Inaccurate hours
Building metering
materials Building air
Temperature set leakage
point is too high in
Temperature set winter
Fuel prices
point is too low in
summer increasing
272
• Step 3: Identify the main categories of causes and

group the potential causes accordingly.
People Methods Machines
Poor Maintenance Lights On After

of HVAC Units Hours

Temp Set point too Aged HVAC Units
low in summer
Bad Energy Temp Set Point
Consuming Habits too high in winter
High Energy
Cost of Buildings
Fuel Princes
Inefficient Building Inaccurate Building Air Increasing
Materials Metering Leakage
Materials Measurements Environment
273
• Step 4: Analyze the cause and effect (C&E) diagram.

• After completing the C&E diagram, the real estate company conducts
further research on each potential root cause.
• It is discovered that:
• The utility metering is accurate
• The building materials are fine and there is not significant amount of air
leakage from the building

• The fuel prices increased recently but were negligible
• Most lights are off during the non-business hours except that some lights
have to be on for security purposes
• The temperature set points in the summer and winter are both adequate
and reasonable
• The high energy costs are probably caused by the poor HVAC
maintenance on aged units and the wasteful energy consuming habits.
• Next, the real estate company needs to collect and analyze the data
to check whether root causes identified in the C&E diagram are
statistically the causes of the high energy costs.
274
275
What is a Cause and Effect Matrix?
• The cause and effect matrix (XY Matrix) is a tool to

help subjectively quantify the relationship of several
X's to several Y's.
• Among the Y's under consideration, two important

ones should be the primary and secondary metrics
of your Six Sigma project.
• The X's should be derived from your cause and

effect diagram. Let us take a peek as what it looks
like on the next page.
276
Cause and Effects Matrix

277
How to Use a Cause and Effect Matrix
1. Across the top enter your output measures. These are the
Y’s that are important to your project.
2. Next, give each Y a weight. Use a 1–10 scale, 1 being least

important and 10 most important.

3. Below, in the leftmost column, enter all the variables you
identified with your cause and effect diagram.
4. Within the matrix itself, rate the strength of the relationship

between the X in the row and the corresponding Y in that
column. Use a scale of 0, 3, 5, and 7.
5. Lastly, sort the “Score” column to order the most important

X’s first.
278
Cause and Effect Matrix Notes

279
After You Have Completed the C&E Matrix
After you have completed your cause and effects
matrix, build a strategy for validating and/or
eliminating the x’s as significant variables to the Y=f(x)
equation.
• Build a data collection plan

• Prepare and execute planned studies
• Perform analytics
• Review results with SMEs
• etc.
280
281

What is a Process Map?
• A process map is a graphical representation of a process
flow.
• It visualizes how the business process is accomplished

step by step.
• It describes how the information or materials sequentially

flow from one business entity to the next.
• It illustrates who is responsible for what between the
process boundaries.
• It depicts the input and output of each individual process
step.
• In the Measure phase, the project team should map the

current state of the process instead of the ideal state.
282
Process Map Basic Symbols
• The following four symbols are the most commonly
used symbols in a process map.
Terminator (Oval):
Shows the start/end points in the process.

Process (Rectangle):
Indicates a single process step.
Decision (Diamond):
Indicates a question with two choices (e.g., Yes/No).
Flow Line (Arrow):

Shows the direction of the process flow.
283
Alternative Process:
Indicates a process step as the alternate of a normal
process step.
Predefined Process:
Indicates a formally defined process step. Other
documentation or instruction is needed to support further
details of the step.

Manual Operation:
Indicates a process step conducted manually.
Preparation:
Indicates a preparation step.
Delay:
Indicates a waiting period in the process.
284
• Additional file- and information-related symbols:
Data (I/O):
Shows the inputs and outputs of a process.
Document:
Indicates a process step that results in a document.

Multi-Document:
Indicates a process step that results in multiple
documents.
Stored Data:
Indicates a process step that stores data.
Magnetic Disk:
Indicates a database.
285
• Additional control of flow symbols:
Off-Page Connector:
Indicates the process flow continues onto another page.
Merge:
Indicates multiple processes merge into one.

Extract:
Indicates a process splits into multiple parallel processes.
Or:
Indicates a single data processing flow diverges to multiple
branches with different criteria requirements.
Summing Junction:
Indicates multiple data processing flows converge into one.
286
• Step 1: Define boundaries of the process you want

to map.
• A process map can depict the flow of an entire process

or a segment of it.

• You need to identify and define the beginning and
ending points of the process before starting to plot.
• Use operational definition.
287
• Step 2: Define and sort the process steps with the

flow.
• Consult with process owners and subject matter experts

or observe the process in action to understand how the
process is actually performed.

• Record the process steps and sort them according to
the order of their occurrence.
288
• Step 3: Fill the step information into the appropriate

process symbols and plot the diagram.
• In the team meeting of process mapping, place the

sticky notes with different colors on a white board to
flexibly adjust the under-construction process map.

• The flow lines are plotted directly on the white board.
For the decision step, rotate the sticky note by 45°.
• When the map is completed on the white board,

record the map using Excel, PowerPoint, or Visio.
289
• Step 3:
• To illustrate the responsibility of different organizations
involved in the process, use a Swim Lane Process Map.

290
• Step 4: Identify and record the inputs/outputs and their
corresponding specifications for each process step.
• The process map helps in understanding and documenting

Y=f(x) of a process where Y represents the outputs and x
represents the inputs.

• The inputs of each process step can be controllable or non-
controllable, standardized operational procedure, or noise.
They are the source of variation in the process and need to be
analyzed qualitatively and quantitatively in order to identify the
vital few inputs that have significant effect on the outcome of
the process.
• The outputs of each process step can be products, information,

services, etc. They are the little Y’s within the process.
291
• Step 5: Evaluate the process map and adjust it if

needed.
• If the process is too complicated to be covered in one single

process map, you may create additional detailed sub-
process maps for further information.

• Number the process steps in the order of their occurrence
for clarity.
292
High Level Process Map
• Most high-level business process maps are also referred
to as flow charts.
• The key to a high-level process map is to over-simplify
the process being depicted so that it can be understood
in its most generic form.
• As a general rule, high-level process maps should be 4–

6 steps and no more.
• Below is an oversimplified version of a high-level process

map for cooking a 10lb prime rib for a dozen holiday
guests.
293
• Detailed process maps or multi-level maps take the high-
level map much further.
• Detailed maps can be two, three, or more levels deeper
than your high-level process map.
• A good guideline used to help create the second level is

to take each step in the high-level map and break it down
into another two to four steps each (no more).
• Repeat this process (level 3, level 4 etc.) until reaching
the desired level of detail.
• Some detailed maps are two or three levels deep, others

can be five or six levels deep. Obviously, the deeper the
levels, the more complex and the more burdensome.
294
Functional Process Map
• The functional map adds dimension to the high-level or
detailed map.
• The dimension added is identifying which function or job
performs the step or makes the decision.
• Below is a generic example of a functional map. Note that
functions are identified in horizontal "lanes" and each process

step is placed in the appropriate lane based on which function
performs the step.
295
What is SIPOC?
• A SIPOC (Suppliers-Input-Process-Output-
Customers) is a high-level visualization tool to help
identify and link the different components in a
process.

• It is usually applied in the Measure phase in order to
better understand the current state of the process
and define the scope of the project.
296
Key Components of a SIPOC
• Suppliers: vendors who provide the raw material,

services, and information. Customers can also be
suppliers sometimes.
• Input: the raw materials, information, equipment,
services, people, environment involved in the

process.
• Process: the high-level sequence of actions and
decisions that results in the services or products
delivered to the customers.
• Output: the services or products delivered to the
customers and any other outcomes of the process.
• Customers: the end users or recipients of the
services or products.
297
How to Plot a SIPOC Diagram
• The first method:
• Step 1: Create a template that can contain the information of

the five key components in a clear way.
• Step 2: Plot a high-level process map that covers five steps

at maximum.
• Step 3: Identify the outputs of the process.
• Step 4: Identify the receipt of the process.
• Step 5: Brainstorm the inputs required to run each process

step.
• Step 6: Identify the suppliers who provide the inputs.

298
How to Plot a SIPOC Diagram
• The second method:
• Step 1: Create a template that can contain the information of

the five key components in a clear way.
• Step 2: Identify the receipt of the process.

• Step 3: Identify the outputs of the process.
• Step 4: Plot a high-level process map that covers five steps

at maximum.
• Step 5: Brainstorm the inputs required to run each process

step.
• Step 6: Identify the suppliers who provide the inputs.

299
Benefits of SIPOC Diagrams
• A SIPOC diagram provides more detailed

information than process maps and it demonstrates
how each component gets involved in the process.
• It helps visualize and narrow the project scope.

• It serves as a great communication tool to help
different process owners understand the entire
process, their specific roles and responsibilities.
300
SIPOC Diagram Example
• Example of plotting a SIPOC diagram for Mom cooking
scrambled eggs for two kids

301
Creating a SIPOC
• Step 1: Vertically List High-Level Process

• If you followed the general rules for a high-level process
map, then you should have no more than 4–6 steps for your
process.
• List those steps in a vertical manner as depicted below.

302
Creating a SIPOC
• Step 2: List Process Outputs

303
Creating a SIPOC
• Step 3: List Output Customers

304
Creating a SIPOC
• Step 4: List Process Inputs

305
Creating a SIPOC
• Step 5: List Suppliers of Inputs

306
SIPOC Benefits
• Visually communicate project scope

• Identify key inputs and outputs of a process
• Identify key suppliers and customers of a process
• Verify:
• Inputs match outputs for upstream processes

• Outputs match inputs for downstream processes.
• This type of mapping is effective for identifying
opportunities for improvement of your process.
• If you have completed your high-level process map,

follow the outlined steps to create a process map
of Suppliers, Inputs, Process, Outputs,
and Customer.
307
What is Value Stream Mapping?
• Value stream mapping is a method to visualize and

analyze the path of how information and raw
materials are transformed into products or services
customers receive.

• It is used to identify, measure, and decrease the
non-value-adding steps in the current process.
308
Non-Value-Added Activities
• Non-value-adding activities are activities in a

process that do not add any other value to the
products or services customers demand.
• Example of non-value-adding activities:

• Rework
• Overproduction
• Excess transportation
• Excess stock
• Waiting
• Unnecessary motion.
• Not all non-value-adding activities are unnecessary.

309
How to Plot a Value Stream Map
• Plot the entire high-level process flow from when the
customer places the order to when the customer receives
the products or services in the end.
• A value stream map requires more detailed information for
each step than the standard process map.
• Cycle time

• Preparation type
• Actual working time
• Available time
• Scrap rate
• Rework rate
• Number of operators
• Assess the value stream map of current process, identity
and eliminate the waste.
310
Basic Value Stream Map Prototype
Customer' Customer
Master
s Order Service Customer
Scheduler
Representative

Sub Process 1 Sub Process 2 Sub Process 3 Sub Process 4
Number of Number of Number of Number of

Operators=... Operators=... Operators=... Operators=...
Cycle Time=... Cycle Time=... Cycle Time=... Cycle Time=...
Preparation Time=... Preparation Time=... Preparation Time=... Preparation Time=...
Actual Working Actual Working Actual Working Actual Working
Time=... Time=... Time=... Time=...
Available Time=... Available Time=... Available Time=... Available Time=...
Scrap Rate=... Scrap Rate=... Scrap Rate=... Scrap Rate=...
Rework Rate=... Rework Rate=... Rework Rate=... Rework Rate=...
311
Additional Mapping Techniques
• Spaghetti Chart
• Thought Process Mapping

312
Spaghetti Chart
• A spaghetti chart is a graphical tool to map out the

physical flow of materials, information, and people
involved in a process. It can also reflect the
distances between multiple workstations the
physical flow has been through.

• A process that has not been streamlined has messy
and wasteful movements of materials, information,
and people, resembling a bowl of cooked spaghetti.
313
How to Plot a Spaghetti Chart
• Step 1: Create a map of the work area layout.
• Step 2: Observe the current work flow and draw the

actual work path from the very beginning of work to
the end when products exit the work area.

• Step 3: Analyze the spaghetti chart and identify
improvement opportunities.
314
Spaghetti Chart Example
Workstation Workstation Workstation

1 2 3
Workstation
4
Workstation

5
Workstation
6
Work
Workstation
station
7
8
315
Thought Process Mapping
• A thought process map is a graphical tool to help

brainstorm, organize, and visualize the information,
ideas, questions, or thoughts regarding reaching the
project goal.

• It is a popular tool generally used at the beginning of
a project in order to:
• identify knowns and unknowns
• communicate assumptions and risks
• discover potential problems and solutions
• identify resources, information, and actions required to meet
the goal
• present relationship of thoughts.
316
How to Plot a Thought Process Map
• Step 1: Define the project goal.
• Step 2: Brainstorm knowns and unknowns about the

project.
• Step 3: Brainstorm questions and group the unknowns

and questions into five phases (Define, Measure,

Analyze, Improve, and Control).
• Step 4: Sequence the questions below the project goal

and link the related questions.
• Step 5: Identify tools or methods that would be used to

answer the questions.
• Step 6: Repeat steps 3 to 5 as the project continues.
317
Thought Process Map Example
Scheduling
Car Battery
Process Map
How often?
Routing Process
Traffic

Which area?
What time?
Deliver the Transportation
Staff pizza to clients
Too few? in time
Online ordering system

Training
System
Phone ordering system
318
319

2.1.4 FMEA
What is FMEA?
• The FMEA (Failure Modes and Effects Analysis)

is an analysis technique to identify, evaluate, and
prioritize a potential deficiency in a process so that
the project team can design action plans to reduce
the probability of the failure/deficiency occurring.

• FMEA activity follows the process mapping and is
followed by a Pareto analysis.
• FMEA is completed in cross-functional

brainstorming sessions in which attendees have a
good understanding of the entire process or of a
segment of it.
320
Basic FMEA Terms
• Process Functions
• Process steps depicted in the process map. FMEA is based on a
process map and one step/function is analyzed at a time.
• Failure Modes
• Potential and actual failure in the process function/step. It usually
describes the way in which failure occurs. There might be more
than one failure mode for one process function.

• Failure Effects
• Impact of failure modes on the process or product. One failure
mode might trigger multiple failure effects.
• Failure Causes
• Potential defect of the design that might result in the failure modes
occurring. One failure mode might have multiple potential failure
causes.
• Current Controls
• Procedures currently conducted to prevent failure modes from
happening or to detect the failure mode occurring.
321
Basic FMEA Terms
• Severity Score
• The seriousness of the consequences of a failure mode
occurring.
• Ranges from 1 to 10, with 10 indicating the most severe
consequence.

• Occurrence Score
• The frequency of the failure mode occurring.
• Ranges from 1 to 10, with 10 indicating the highest frequency.
• Detection Score
• How easily failure modes can be detected.
• Ranges from 1 to 10, with 10 indicating the most difficult
detection.
322
Basic FMEA Terms
• RPN (Risk Prioritization Number)

• The product of the severity, occurrence, and detection
scores.
• Ranges from 1 to 1000.
• The higher RPN is, the more focus the particular
step/function needs.

• Recommended Actions
• The action plan recommended to reduce the probability of
failure modes occurring.
323
How to Conduct an FMEA
• Step 1: List the critical functions of the process based on the
process map created.
• Step 2: List all potential failure modes that might occur in each
function. One function may have multiple potential failures.

• Step 3: List all potential failure effects that might affect the
process or product.
• Step 4: List all possible causes that may lead to the failure mode
happening.
• Step 5: List the current control procedures for each failure mode.
324
• Step 6: Determine the severity rating for each potential failure
mode.
• Step 7: Determine the occurrence rating for each potential

failure cause.

• Step 8: Determine the detection rating for each current control
procedure.
• Step 9: Calculate RPN (Risk Prioritization Number).
• Step 10: Rank the failures using RPN and determine the
precedence of problems or critical inputs of the process. A
Pareto chart might help to focus on the failure modes with high
RPNs. The higher the RPN, the higher the priority the
correction action plan.
325
• Step 11: Brainstorm and create recommended action plans for
each failure mode.
• Step 12: Determine and assign the task owner and projected
completion date to take actions.
• Step 13: Determine the new severity rating if the actions are

taken.
• Step 14: Determine the new occurrence rating if the actions

are taken.
• Step 15: Determine the new detection rating if the actions are
taken.
• Step 16: Update the RPN based on new severity, occurrence,

and detection ratings.
326
FMEA Example
Case study:
• Joe is trying to identify, analyze, and eliminate the failure
modes he experienced in the past when preparing his
work bag before heading to the office every morning. He
decides to run an FMEA for his process of work bag
preparation.

• There are only two steps involved in the process.
• Putting the work files in the bag
• Putting a water bottle in the bag.
327
FMEA Example
• Step 1: List the critical functions of the process

based on the process map created.

328
FMEA Example
• Step 2: List all the potential failure modes that might

occur in each function.

329
FMEA Example
• Step 3: List all the potential failure effects that might

affect the process or product.

330
FMEA Example
• Step 4: List all the possible causes that may lead to

the failure mode happening.

331
FMEA Example
• Step 5: List the current control procedures for each

failure mode.

332
FMEA Example
• Step 6: Determine the severity rating for each

potential failure mode.

333
FMEA Example
• Step 7: Determine the occurrence rating for each

potential failure cause.

334
FMEA Example
• Step 8: Determine the detection rating for each

current control procedure.

335
FMEA Example
• Step 9: Calculate the RPN (Risk Prioritization

Number).

336
FMEA Example
• Step 10: Rank the failures using the RPN and

determine the precedence of problems or critical
inputs of process.

337
FMEA Example
• Step 11: Brainstorm and create recommended

action plans for each failure mode.

338
FMEA Example
• Step 12: Determine and assign the task owner and

projected completion date to take actions.

339
FMEA Example
• Step 13: Determine the new severity rating if the actions are
taken.
• Step 14: Determine the new occurrence rating if the actions
are taken.
• Step 15: Determine the new detection rating if the actions are
taken.

• Step 16: Update the RPN based on new severity, occurrence,
and detection ratings.
340
341
What is the Theory of Constraints?
• Processes, systems, and organizations are

vulnerable to their weakest part.
• Any manageable system is limited by constraints in
its ability to produce more (and there is always at
least one constraint).

342
Performance Measures
Making sound financial decisions based on these three
measures is a critical requirement.
• Throughput – rate at which a system generates money
through sales.
• Operational Expense – money spent by the system to

turn inventory into throughput.
• Inventory – money the system has invested in
purchasing things it intends to sell.
343
Five Focusing Steps
Objective: To ensure ongoing improvement

efforts are focused on the constraints of
a system.
1. Identify the system’s constraint(s).

2. Decide how to exploit the constraint(s).

3. Subordinate everything else to the decision in step 2.
4. Elevate the constraint(s).
5. If in previous steps a constraint has been broken, return to
step 1, but do not allow inertia to cause a system’s constraint.
344
Logical Thinking Processes
Focusing Step Thinking Process Tools
Identify the system’s • Identify the problems

1 constraint(s) • Find the root causes
Cause and effect diagram
Decide how to exploit

2 • Develop a solution Future reality tree
the constraint(s)

Subordinate everything • Identify the conflict
3 else to the decision in preventing the solution Evaporating cloud
step 2 • Remove the conflict
• Construct and execute an Prerequisite tree

4 Elevate the constraint
implementation plan Transition tree
If in previous steps a constraint has been broken, return to step 1, but do not allow inertia
5 to cause a system’s constraint
345
Simulation Exercise
Resources needed:
• 3 “production line” participants
• 1 timer per each “production line” participant
• 5 small boxes of 15 widgets each (paperclips,
pens/pencils, candy, etc.)

Widget Value Chain:
Get box of
Count widgets Return box to
widgets and
and fill box instructor
empty it
Participant 1 Participant 2 Participant 3
346
Simulation Exercise
Widget Value Chain:
Get box of
Count widgets Return box to
widgets and
and fill box instructor
empty it
Participant 1 Participant 2 Participant 3

Five focusing steps:
1. Identify the system’s constraint(s).
2. Decide how to exploit the constraint(s).
3. Subordinate everything else to the decision in step 2.
4. Elevate the constraint(s).
5. If in previous steps a constraint has been broken, return to step
1, but do not allow inertia to cause a system’s constraint.
347
348

2.3.3 Gage R&R
349
350
What is Statistics?
• Statistics is the science of collection, analysis,

interpretation, and presentation of data.
• In Six Sigma, we apply statistical methods and

principles to quantitatively measure and analyze the

process performance to reach statistical conclusions
and help solve business problems.
351
Types of Statistics
• Describing what was going on
• Inferential Statistics
• Making inferences from the data at hand to more general

conditions
352
Descriptive Statistics
• Descriptive statistics is applied to describe the

main characteristics of a collection of data.
• Descriptive statistics summarizes the features of the

data quantitatively.

• Descriptive statistics is descriptive only and it does
not make any generalizations beyond the data at
hand.
• The data used for descriptive statistics are for the

purpose of representing or reporting.
353
Inferential Statistics
• Inferential statistics is applied to infer the

characteristics or relationships of the populations
from which the data are collected.
• Inferential statistics draws statistical conclusions

about the population by analyzing the sample data
subject to random variation.
• A complete data analysis includes both descriptive

statistics and inferential statistics.
354
Statistics vs. Parameters
• The word statistic refers to a numeric measurement

calculated using a sample data set, for example,
sample mean or sample standard deviation. Its
plural is statistics (the same spelling as “statistics”
which refers to the scientific discipline).

• The parameter refers to a numeric metric describing
the population, for example, population mean and
population standard deviation. Unless you have the
full data set of the population, you will not be able to
know the population parameters.
355
Continuous Variable vs. Discrete Variable
• Continuous Variable
• Measured
• There is an infinite number of values possible
• Examples: temperature, height, weight, money, time

• Discrete Variable
• Counted
• There is a finite number of values available
• Examples: count of people, count of countries, count of
defects, count of defectives
356
Types of Data
• Nominal
• Categorical data
• Examples: a set of colors, the social security number
• Ordinal
• Rank-ordering data

• Examples: the first, second place in a race, scores of exams
• Interval
• Equidistant data
• Examples: temperature with Fahrenheit or Celsius scale
• Ratio
• The ratio between the magnitude of a continuous value and
the unit value of the same category
• Examples: weight, length, time
357
358
Basics of Descriptive Statistics
• Descriptive statistics provides a quantitative

summary for the data collected.
• It summarizes the main features of the collection of

data.

• Shape
• Location
• Spread
• It is a presentation of data collected and it does not

provide any inferences about a more general
condition.
359
Shape of the Data
• Distribution is used to describe the shape of the

data.
• Distribution (also called frequency distribution)

summarizes the frequency of an individual value or a

range of values of a variable (either continuous or
discrete).
• Distribution is depicted as a table or graph.
360
Shape of the Data
• Simple example of distribution
• We are tossing a fair die. The possible value we obtain from

each tossing is a value between 1 and 6.
• Each value between 1 and 6 has a 1/6 chance to be hit for

each tossing.
• The distribution of this game describes the relationship

between every possible value and the percentage of times
the value is being hit (or count of times the value is being
hit).
361
Shape of the Data
• Examples of continuous distribution

• Normal Distribution
• T distribution
• Chi-square distribution
• F distribution

• Examples of discrete distribution
• Binomial distribution
• Poisson distribution
362
Location of the Data
• The location (i.e., central tendency) of the data

describes the value where the data tend to cluster
around.
• There are multiple measurements to capture the

location of the data:
• Mean
• Median
• Mode.
363
Mean
• The mean is the arithmetic average of a data set.
1 n
x   xi
n i 1

n is the number of values in the data set
• For example, we have a set of data: 2, 3, 5, 8, 5,

and 9. The arithmetic mean of the data set is
235859
 5.33
6
364
Median
• The median is the middle value of the data set in

numeric order.
• It separates the finite set of data into two parts: one

with values higher that the median and the other

with values lower than the median.
• For example, we have a set of data: 45, 32, 67, 12,

37, 54, and 28. The median is 37 since it is the
middle value of the sorted list of values (i.e., 12, 28,
32, 37, 45, 54, and 67).
365
Mode
• The mode is the value that occurs most often in the

data set.
• If no number is repeated, there is no mode for the

data set.

• For example, we have a data set: 55, 23, 45, 45, 68,
34, 45, 55. The mode is 45 since it occurs most
frequently.
366
Spread of the Data
• The spread (i.e., variation) of the data describes the

degree of data dispersing around the center value.
• There are multiple measurements to capture the

spread of the data:

• Range
• Variance
• Standard Deviation.
367
Range
• The range is the numeric difference between the

greatest and smallest values in a data set.
• Only two data values (i.e., the greatest and the

smallest values) are accounted for calculating the

range.
• For example, we have a set of data: 34, 45, 23, 12,

32, 78, and 23. The range of the data is 78−12 = 66.
368
Variance
• The variance measures how far on average the

data points spread out from the mean.
• It is the average squared deviation of each value
from its mean.
• All the data points are accounted for calculating the

variance.
n
1
s 2   ( xi  x) 2
n i 1
where
1 n
x   xi
n i 1
369
Standard Deviation
• Standard deviation describes how far the data

points spread away from the mean.
• It is simply the square root of the variance.
• All the data points are accounted for calculating the
standard deviation.

1 n
s 
n i 1
( xi  x ) 2
where
1 n
x   xi
n i 1
370
2.2.3 Normal Distribution & Normality
371
What is Normal Distribution?
• The normal distribution is a probability distribution

of a continuous random variable whose values
spread symmetrically around the mean.
• A normal distribution can be completely described

by using its mean (μ) and variance (σ2).
• When a variable x is normally distributed, we denote

x ~ N(μ, σ2).
372
Z Distribution
• The Z distribution is the simplest normal
distribution with the mean equal to zero and the
variance equal to one.
• Any normal distribution can be transferred to a Z
distribution by applying

x
z
s
where

x ~ N ,s 2
 s 0
373
Shape of Normal Distribution
• The probability density function curve of normal

distribution is bell-shaped.
• Probability density function of normal distribution
( x )2
1 
f ( x)  2s 2

e
2s 2
μ
374
Location of Normal Distribution
• If a variable is normally distributed, the mean, the

median, and the mode have the same value.
• The probability density curve of normal distribution

is symmetric around a center value which is the

mean, the median, and the mode at the same time.
375
Spread of Normal Distribution
• The spread or variation of the normally-distributed

data can be described using the variance or the
standard deviation.
• The smaller the variance or the standard deviation,

the less variability in the data set.
376
68-95-99.7 Rule
• The 68-95-99.7 rule or the empirical rule in statistics

states that for a normal distribution:
• About 68% of the data fall within one standard deviation of

the mean

• About 95% of the data fall within two standard deviations of
the mean
• About 99.7% of the data fall within three standard deviations

of the mean.
377
378

68-95-99.7 Rule
Normality
• Not all the distributions with a bell shape are normal

distributions.
• To check whether a group of data points are

normally distributed, we need to run a normality test.

• There are different normality tests available:
• Anderson-Darling test
• Sharpiro-Wilk test
• Jarque-Bera test.
• More details of normality test will be introduced in

the Analyze module.
379
Normality Testing
• To check whether the population of our interest is

normally distributed, we need to run normality test.
• Null Hypothesis (H0): The data points are normally

distributed.
• Alternative Hypothesis (Ha): The data points are not

normally distributed.
• There are many normality tests available. For

example, Anderson-Darling test, Sharpiro-Wilk test,
Jarque-Bera test, and so on.
380
Use Minitab to Run a Normality Test
• Case study: we are interested to know

whether the height of basketball players is
• Data File: “One Sample T-Test” tab in “Sample
Data.xlsx”

• Null Hypothesis (H0): the height of basketball
players is normally distributed.
• Alternative Hypothesis (Ha): the height of basketball

players is not normally distributed.
381
• Steps to run a normality test in Minitab
1) Click Stat → Basic Statistics → Normality Test.
2) A new window named “Normality Test” pops up.

3) Select “HtBk” as the “Variable.”
4) Click “OK.”
5) The normality test results appear in the new window.
382

383
• Null Hypothesis (H0): The data are normally distributed.
• Alternative Hypothesis (Ha): The data are not normally distributed.
• Since the p-value of the normality is 0.275 greater than alpha level
(0.05), we fail to reject the null and claim that the data are normally
distributed.

384
385
What is Graphical Analysis?
• In statistics, graphical analysis is a method to

visualize the quantitative data.
• Graphical analysis is used to discover the structure

and patterns in the data, explaining and presenting

the statistical conclusions.
• A complete statistical analysis includes both

quantitative analysis and graphical analysis.
386
Graphical Analysis Example
• There are various graphical analysis tools available.

Here are four most commonly used examples:
• Box Plot
• Histogram
• Scatter Plot
•

Run Chart.
387
Box Plot
• A box plot is a graphical method to summarize a

data set by visualizing the minimum value, 25th
percentile, median, 75th percentile, the maximum
value, and potential outliers.

• A percentile is the value below which a certain
percentage of data fall. For example, if 75% of the
observations have values lower than 685 in a data
set, then 685 is the 75th percentile of the data.
388
Box Plot
Median
(Q2)
Minimum Value in Maximum Value in
the Data 25th Percentile (Q1) 75th Percentile the Data
(Q3)

Potential Potential
Interquartile Range
Outliers Outliers
(IQR)
Maximum (Minimum Value in the Data, Q1 – I.5*IQR)
Minimum (Maximum Value in the Data, Q3 + I.5*IQR)
Interquartile Range = 75th Percentile – 25th Percentile

389
How to Use Minitab to Generate a Box Plot
• Data File: “Box Plot” tab in “Sample Data.xlsx”

• Steps to render a Box Plot in Minitab
1) Click Graph → Boxplot.
2) A new window named “Boxplots” pops up.
3) Click “OK” in the window “Boxplots.”

4) Another new window named “Boxplot – One Y, Simple”
pops up.
5) Select “HtBk” as the “Graph Variables.”
6) Click the “Data View” button and a new window named
“Boxplot – Data View” pops up.
7) Check the boxes “Median symbol” and “Mean symbol.”
8) Click “OK” in the window “Boxplot – Data View.”
9) Click “OK” in the window “Boxplot – One Y, Simple.”
10) The box plot appears automatically in the new window.
390

391
Median (Q2)
75th Percentile (Q3)

25th Percentile (Q1)
Mean
392
Histogram
• A histogram is a graphical tool to present the

distribution of the data.
• The X axis represents the possible values of the

variable and the Y axis represents the frequency of

the value occurring.
• A histogram consists of adjacent rectangles erected

over intervals with heights equal to the frequency
density of the interval.
• The total area of all the rectangles in a histogram is

the number of data values.
393
Histogram
• A histogram can also be normalized. In this case,

the X axis still represents the possible values of the
variable, but the Y axis represents the percentage of
observations that fall into each interval on the X
axis.

• The total area of all the rectangles in a normalized
histogram is 1.
• With the histogram, we have a better understanding

of the shape, location, and spread of the data.
394
Histogram
12 Histogram with
Count
8 frequency (count) as
the Y axis
4

5.5 6 6.5 7
Normalized 0.30
Probability
histogram with 0.20
proportion 0.10
(probability) as the Y
axis 5.5 6 6.5 7
395
How to Use Minitab to Generate a Histogram
• Data File: “Histogram” tab in “Sample Data.xlsx”
• Steps to render a histogram in Minitab

1) Click Stat → Basic Statistics → Graphical Summary.

2) A new window named “Graphical Summary” pops up.
3) Select “HtBk” as the “Variables.”
4) Click “OK.”
5) The histogram appears in the new window.
396

397

Histogram
398
Scatter Plot
• A scatter plot is a diagram to present the

relationship between two variables of a data set.
• A scatter plot consists of a set of data points.

• On the scatter plot, a single observation is
presented by a data point with its horizontal position
equal to the value of one variable and its vertical
position equal to the value of the other variable.
399
Scatter Plot
• A scatter plot helps to understand:

• Whether the two variables are related to each other or not
• How is the strength of their relationship

• What is the shape of their relationship
• What is the direction of their relationship
• Whether outliers are present.
400
How to Use Minitab to Generate a Scatter Plot
• Data File: “Scatter Plot” tab in “Sample Data.xlsx”

• Steps to render a histogram in Minitab:
1) Click Graph → Scatterplot.
2) A new window named “Scatterplots” pops up.

3) Click “OK” in the window “Scatterplots” and another window
named “Scatterplot – Simple” pops up.
4) Select “MPG” as the “Y variables.”
5) Select “weight” as the “X variables.”
6) Click “OK.”
7) The scatter plot appears in the new window.

401

402

403
Run Chart
• A run chart is a chart used to present the data in

time order. It captures the process performance over
time.
• The X axis of the run chart indicates the time and

the Y axis indicates the observed values.
• Run chart looks similar to control charts except that

a run chart does not have control limits plotted. It is
easier to produce a run chart than a control chart.
• It is often used to identify the anomalies in the data

and discover the pattern of data changing over time.
404
How to Plot a Run Chart in Minitab
• Steps to plot a run chart in Minitab:

1) Data File: “Run Chart” tab in “Sample Data.xlsx.”
2) Click Stat → Quality Tools → Run Chart.
3) A new window named “Run Chart” pops up.

4) Select “Measurement” as the “Single Column.”
5) Enter “1” as the “Subgroup Size.”
6) Click “OK.”
7) The run chart appears automatically in the new window.
405

406

407
Run Chart Example
• Run charts are used to identify the trend, cycle,

seasonal pattern, or abnormality in the data.

• The time series in this chart appears stable.
• There are no extreme outliers, trending or seasonal
patterns.
408
Run Chart Example
• Create another run chart using the data listed in

column “Cycle” in the “Run Chart” tab of “Sample
Data.xlsx.”

Seasonal Pattern
409
Run Chart Example
• Create another run chart using the data listed in

column “Trend” in the “Run Chart” tab of “Sample
Data.xlsx.”

Trending Pattern
410
2.3 MSA (Measurement System Analysis)
411

2.3.3 Gage R&R
412
413
What is Measurement System Analysis
• Measurement System Analysis (MSA) is a

systematic method to identify and analyze the
variation components in the measurement.
• It is a mandatory step in any Six Sigma project to

ensure the data are reliable before making any data-
based decisions.
• The MSA is the check point of data quality before

we start any further analysis and draw any
conclusions from the data.
414
Data-Based Analysis
• Here are some examples of data-based analysis

where MSA is the prerequisite:
• Correlation analysis
• Regression analysis
• Hypothesis testing

• Analysis of variance
• Design of experiments
• Multivariate analysis
• Statistical process control.
415
Measurement System
• A measurement system is a process to obtain data.
Measurement
Inputs (Xs) Output (Y)
System
• Y (output of the measurement system)

• Observed values
• X’s (inputs of the measurement system)

• True values
• Measurement errors
416
Measurement Errors
Observed Value = True Value + Measurement Error
• True Value
• The actual value we are interested to measure
• It reflects the true performance of the process we are

measuring
• Measurement Error
• The errors brought in by measurement system
• Observed Value
• The observed/measured value obtained by the measurement
system
417
Measurement Errors
• Types of Observed Values:

• Continuous measurements
• Weight
• Height
• Money
• Discrete measurements

• Red/Yellow/Green
• Yes/No
• Ratings of 1–10
• A variable MSA is designed for continuous

measurements and an attribute MSA is for discrete
measurements.
418
Measurement Errors
• Sources of measurement errors:

• Human
• Environment
• Equipment
• Sample
•

Process
• Material
• Method.
• Fishbone diagrams can help to brainstorm the

potential factors affecting the measurement system.
419
Measurement Errors
• The more errors the measurement system brings in,

the less reliable the observed values are.
• A valid measurement system brings in minimum

amount of measurement errors.

• The goal of MSA is to qualify the measurement
system by quantitatively analyzing its
characteristics.
420
Characteristics of a Measurement System
• Any measurement systems can be characterized by

two aspects:
• Accuracy (location related)
• Precision (variation related).

• A valid measurement system is both accurate and
precise.
• Being accurate does not guarantee the measurement
system is precise.
• Being precise does not guarantee the measurement system
is accurate.
421
Accuracy vs. Precision
• Accuracy:
• The level of closeness between the average observed value
and the true value
• How well the observed value reflects the true value.
• Precision:

• The spread of measurement values
• How consistent the repeated measurements deliver the
same values under the same circumstances.
422
• Accurate and precise

• (high accuracy and high precision)

423
• Accurate and not precise

• (high accuracy and low precision)

424
• Precise and not accurate

• (high precision and low accuracy)

425
• Not accurate and not precise

• (low accuracy and low precision)

426
MSA Conclusions
• If the measurement system is considered both

accurate and precise, we can start the data-based
analysis or decision making.
• If the measurement system is either not accurate or

not precise, we need to identify the factor(s)
affecting it and calibrate the measurement system
until it is both accurate and precise.
427
Stratification of Accuracy and Precision
• Accuracy
• Bias
• Linearity
• Stability

• Precision
• Repeatability
• Reproducibility
428
429
Bias
• Bias is the difference between the observed value

and the true value of a parameter or metric being
measured.
• It is calculated by subtracting the reference value

from the average value of the measurements.
Bias = Grand Mean – Reference Value
where the reference value is a standard agreed upon
430
Bias
Bias
Reference value Average of the measurements

Distribution of measurements
• The closer the average of all measurements is to the

reference level, the smaller the bias.
• The reference level is the average of measurements of the

same items using the master or standard instrument.
431
Bias
• To determine whether the difference between the

average of observed measurement and the
reference value is statistically significant (we will
explain more details about statistical significance in
the Analyze module), we can either conduct a
hypothesis testing or compare the reference value

against the confidence intervals of the average
measurements.
• If the reference value falls into the confidence

intervals, the bias is not statistically significant and
can be ignored. Otherwise, the bias is statistically
significant and must be fixed.
432
Bias
• Potential causes of bias:

• Errors in measuring the reference value
• Lack of proper training for appraisers
• Damaged equipment or instrument
• Measurement instrument not calibrated precisely
•

Appraisers read the data incorrectly.
433
Linearity
• Linearity is the degree of the consistency of bias

over the entire expected measurement range.
• It quantifies how the bias changes over the range of

measurement.

• For example, a scale is off by 0.01 pounds when
measuring an object of 10 pounds. However, it is off
by 10 pounds when measuring an object of 100
pounds. The scale’s bias is not linear.
434
435

Linearity
Linearity
• Create a scatter plot for bias (Y-axis) and reference

level (X-axis).
• Find a best fit linear regression line and compute the
slope of the line.
• The closer the slope is to zero, the better the

measurement system performs.
436
Linearity
• Formula of the linearity of a measurement system:
Linearity = |Slope| × Process Variation
where
n
1 n n
 xi yi   xi  yi

n i 1 i 1
Slope  i 1
2
1 
 
n n
 x
i 1
 2
i
n
  x 
 i 1
i

xi is the reference value;

yi is the bias at each reference level;
n is the sample size.
437
Linearity
• Potential causes of linearity:

• Errors in measuring the lower end or higher end of the
reference value
• Lack of proper training for appraisers

• Measurement instrument not calibrated correctly at the

lower or higher end of the measurement scale
• Innate nature of the instrument.
438
Stability
• Stability is the consistency level needed to obtain

the same values when measuring the same objects
over an extended period of time.
• A measurement system that has low bias and

linearity close to zero but cannot consistently
perform well would not deliver reliable data.
• Stability is evaluated using control charts.
439
440

Bias
Bias
Stability
Time
Stability
• Control charts are used to evaluate the stability of a

measurement system.
• When there are no data points out of control, the

measurement system is considered stable.

441
Stability
• Potential causes of instability:

• Inconsistent training for appraisers
• Worn equipment or instrument

• Measurement instrument not calibrated
• Appraisers do not follow the procedure consistently.
442
443

2.3.3 Gage R&R
Repeatability
• Repeatability evaluates whether the same

appraiser can obtain the same value multiple times
when measuring the same object using the same
equipment under the same environment.

• It refers to the level of agreement between the
repeated measurements of the same appraiser
under the same condition.
• Repeatability measures the inherent variation of the

measurement instrument.
444
Reproducibility
• Reproducibility evaluates whether different

appraisers can obtain the same value when
measuring the same object independently.
• It refers to the level of agreement between different

appraisers.
• It is not caused by the inherent variation of the

measurement instrument. It reflects the variability
caused by different appraisers, locations, gauges,
environments etc.
445
Gauge R&R
• Gauge R&R (i.e., Gauge Repeatability &

Reproducibility) is a method to analyze the
variability of a measurement system by partitioning
the variation of the measurements using ANOVA
(Analysis of Variance).

• Gauge R&R only addresses the precision of a
measurement system.
446
Gauge R&R
• Data collection of a gauge R&R study:

• let k appraisers measure n random samples independently
and repeat the process p times.
• Different appraisers perform the measurement

independently.
• The order of measurement (e.g., sequence of

samples and sequence of appraisers) is
randomized.
447
Gauge R&R
• The potential sources of variance in the

measurement:
• Appraisers: s appraisers
2
• s 2
Parts: parts
Appraisers × Parts: s appraisers parts
2
•
Repeatability: s repeatability
2
•

• Variance Components
s total
2
 s appraisers
2
 s parts
2
 s appraisers
2
 parts  s 2
repeatability
448
Gauge R&R
• A valid measurement system has low variability in

both repeatability and reproducibility so that the
total variability observed can reflect the true
variability in the objects (parts) being measured.
s total
2
 s reproducib
2
 s 2
 s 2

ility repeatability parts
where
s reproducib
2
ility  s 2
appraisers  s 2
appraisers parts
• Gauge R&R variance reflects the precision level of

the measurement system.
s R2&R  s repeatabil
2
ity  s 2
reproducibility
449
Gauge R&R
• Variation Components
Variation total  Z 0  s total

Variation repeatability  Z 0  s repeatability
Variation reproducibility  Z 0  s reproducibility

Variation parts  Z 0  s parts
where
s 2
total s 2
reproducibility s 2
repeatability s 2
parts
Z0 is a sigma multiplier that assumes a specific confidence level in

the spread of the data.
450
Gauge R&R
• The percentage of variation R&R contributes to the

total variation in the measurement:
Variation R& R
Contributi on% R& R  100%
Variation total

where Variation R&R  Z 0  s repeatabil
2
ity  s reproducibility
2
451
452
Variable Gage R&R
• Whenever something is measured repeatedly or by different

people or processes, the results of the measurements will
vary. Variation comes from two primary sources:
1. Differences between the parts being measured
2. The measurement system.

• We can use a gage R&R to conduct a measurement system
analysis to determine what portion of the variability comes
from the parts and what portion comes from the
measurement system.
• There are key study results that help us determine the

components of variation within our measurement system.
453
Key Measures of a Variable Gage R&R
• %Contribution: The percent of contribution for a source is 100

times the variance component for that source divided by the total
variation.
•%Study Var (6*SD): The percent of study variation for a source
is 100 times the study variation for that source divided by the total
variation.

• %Tolerance (SV/Tolerance): The percent of spec range
taken up by the total width of the distribution of the data based on
variation from that source.
• Distinct Categories: The number of distinct categories of parts
that the measurement system is able to distinguish. If a
measurement system is not capable of distinguishing at least five
types of parts, it is probably not adequate.
454
Variable Gage R&R Guidelines (AIAG)
• Percent Tolerance and Percent Study Variation

• 10% or less – Acceptable
• 10% to 30% – Marginal
• 30% or greater – Unacceptable

• Percent Contribution
• 1% or less – Acceptable
• 1% to 9% – Marginal
• 9% or greater – Unacceptable
• Distinct Categories
• Look for five or more distinct categories to
indicate that your measurement system is
acceptable.
455
Guidelines for Distinct Categories
Distinct categories is the number of categories of parts that
your measurement system can distinguish. If it is below five,
it is likely not able to distinguish between parts.
Number of Categories Conclusion

Measurement system cannot

Distinct Categories = 1
discriminate between parts
Measurement system can only
Distinct Categories = 2 distinguish between high/low or
big/small
Measurement system is of little or
Distinct Categories = 3 or 4 no value
According to AIAG, the
Distinct Categories = 5+ measurement system can
acceptably discriminate parts
456
Use Minitab to Implement a Variable MSA
• Data File: “Variable MSA” tab in “Sample Data.xlsx” (an
example in the AIAG MSA Reference Manual, 3rd Edition)
• Step 1: Initiate the MSA study
1) Click on Stat → Quality Tools → Gage R&R → Create Gage R&R
Study Worksheet.
2) A new window named “Create Gage R&R Study Worksheet” pops
up.

3) Select 10 as the “Number of Parts.”
4) Select 3 as the “Number of Operators.”
5) Select 3 as the “Number of Replicates.”
6) Enter the part name (e.g., Part 01, Part 02, Part 03, …).
7) Enter the operator name (e.g., Operator A, Operator B, Operator C).
8) Click on the “Options” button and another window named “Create
Gage R&R Study Worksheet – Options” pops up.
9) Select the radio button “Do not randomize.”
10) Click “OK” in the window “Create Gage R&R Study Worksheet –
Options.”
11) Click “OK’ in the window “Create Gage R&R Study Worksheet.”
12) A new data table is generated.
457

458
• Step 2: Data collection

• In the newly-generated data table, Minitab has provided the
template where we organize the data.
• In the “Variable MSA” tab in “Sample Data.xlsx,” there are all
the measurement data collected by three operators (i.e.,
operator A, B, and C). The data are listed in the

standardized order.
459
• Step 3: Enter the data into the MSA template generated
in Minitab
• Transfer the data from the “Measurement” column in “Variable
MSA” tab of “Sample Data.xlsx” to the last column in the MSA
template Minitab generates.
• Although the run order in the “Variable MSA” tab of “Sample
Data.xlsx” is different from the run order in the Minitab MSA

template, we can directly use the raw data listed in the “Variable
MSA” tab of “Sample Data.xlsx” for our MSA purpose.
460
• Step 4: Implement Gauge R&R
1) Click Stat → Quality Tools → Gage Study → Gage R&R
Study (Crossed).
2) A new window named “Gage R&R Study (Crossed)” appears.
3) Select “Part” as “Part numbers.”
4) Select “Operator” as “Operators.”
5) Select “Measurement” as “Measurement data.”

6) Click on the “Options” button and another new window
named “Gage R&R Study (Crossed) – ANOVA Options” pops
up.
7) Enter 5.15 as the “Study variation (number of standard
deviations)”.
8) Click “OK” in the window “Gage R&R Study (Crossed) –
ANOVA Options.”
9) Click “OK” in the window “Gage R&R Study (Crossed).”
10) The MSA analysis results appear in the new window and the
session window.
461

462
• 5.15 is the recommended standard deviation multiplier by
the Automotive Industry Action Group (AIAG). It corresponds
to 99% of data in the normal distribution. If we use 6 as the
standard deviation multiplier, it corresponds to 99.73% of
the data in the normal distribution.

463
• Step 5: Analyze the MSA results

The percentage of variation R&R contribution to the total variation is 27.86% and
the precision level of this measurement system is not good. Actions are required to
calibrate the measurement system.
464

465
Use Minitab to Implement an Attribute MSA
• Data File: “Attribute MSA” tab in “Sample Data.xlsx”

(an example in the AIAG MSA Reference Manual,
3rd Edition).
• Steps in Minitab to run an attribute MSA

466
• Step 1: Reorganize the original data into four new
columns (i.e., Appraiser, Assessed Result, Part, and
Reference).
1) Click Data → Stack → Blocks of Columns.
2) A new window named “Stack Blocks of Columns” pops up.
3) Select “Appraiser A,” “Part,” and “Reference” as block one.

4) Select “Appraiser B,” “Part,” and “Reference” as block two.
5) Select “Appraiser C,” “Part,” and “Reference” as block three.
6) Select the radio button of “New worksheet” and name the
sheet “Data.”
7) Check the box “Use variable names in subscript column.”
8) Click “OK.”
9) The stacked columns are created in the new worksheet
named “Data.”
10) Name the four columns from left to right in worksheet “Data”:
Appraiser, Assessed Result, Part, and Reference.
467

468

469
• Step 2: Run MSA using Minitab
1) Click Stat → Quality Tools → Attribute Agreement Analysis.
2) A new window named “Attribute Agreement Analysis” pops
up.
3) Click in the blank box next to “Attribute column” and the
variables appear in the list box on the left.
4) Select “Assessed Result” as “Attribute column.”
5) Select “Part” as “Sample.”

6) Select “Appraiser” as “Appraisers.”
7) Select “Reference” as “Known standard/attribute.”
8) Click the “Options” button and another window named
“Attribute Agreement Analysis – Options” pops up.
9) Check the boxes of both “Calculate Cohen’s kappa if
appropriate” and “Display disagreement table.”
10) Click “OK” in the window “Attribute Agreement Analysis –
Options.”
11) Click “OK” in the window “Attribute Agreement Analysis.”
12) The MSA results appear in the newly-generated window and
the session window.
470

471
Within Appraiser
Agreement Percent:
the agreement
percentage within each
individual appraiser.

472
Each Appraiser vs.

Standard Agreement
Percent: the
agreement percentage

between each
appraiser and the
standard. It reflects
the accuracy of the
measurement system.
473
Between Appraiser
Agreement Percent: the
agreement percentage
between different
appraisers.

All Appraisers vs.
Standard Agreement
Percent: overall
agreement percentage of
both within and between
appraisers. It reflects how
precise the measurement
system performs.
474
• Kappa statistic is a coefficient indicating the agreement
percentage above the expected agreement by chance.
• Kappa ranges from -1 (perfect disagreement) to 1 (perfect

agreement).

• When the observed agreement is less than the chance
agreement, Kappa is negative.
• When the observed agreement is greater than the chance

agreement, Kappa is positive.
• Rule of thumb: If Kappa is greater than 0.7, the

measurement system is acceptable. If Kappa is greater
than 0.9, the measurement system is excellent.
475
Kappa statistic of the

agreement within each
appraiser

Kappa statistic of the agreement
between individual appraiser and
the standard
476
Kappa statistic of
the agreement
between appraisers

Kappa statistic of the
overall agreement
between appraisers
and the standard
477
478

2.3.2 Bias, Linearity and Stability
2.3.3 Gage R&R
479
480
What is Process Capability?
• The process capability measures how well the

process performs to meet given specified outcome.
• It indicates the conformance of a process to meet

given requirements or specifications.

• Capability analysis helps to better understand the
performance of the process with respect to meeting
customer’s specifications and identify the process
improvement opportunities.
481
Process Capability Analysis Steps
• Step 1: Determine the metric or parameter to measure
and analyze.
• Step 2: Collect the historical data for the parameter of

interest.

• Step 3: Prove the process is statistically stable (i.e., in
control).
• Step 4: Calculate the process capability indices.
• Step 5: Monitor the process and ensure it remains in

control over time. Update the process capability indices if
needed.
482
Process Capability Indices
• Process capability can be presented using various

indices depending on the nature of the process and
the goal of the analysis.
• Popular process capability indices:

• Cp
• Pp
• Cpk
• Ppk
• Cpm
483
Cp
• Cp stands for capability of the process.
USL  LSL
Cp 
6  s within
where

s within 
sp
sp 
  (x  x )
i j ij i
c4 (d  1)  (n  1)
i i
4(n  1)
d  i (ni  1) c4 
(4n  3)
USL and LSL are the upper and lower specification limits.
484
Cp
• Cp measures the process’ potential capability to meet the
two-sided specifications.
• It does not take the process average into consideration.

• High Cp indicates the small spread of the process with
respect to the spread of the customer specifications.
• Cp is recommended when the process is centered

between the specification limits.
• Cp works when there are both upper and lower

485
Pp
• Pp stands for performance of the process.
USL  LSL
Pp 
6  s overall

where
s ( xij  x) 2
s overall  s 
c4 (n) i j
n 1
4(n  1)
c4 
(4n  3)
486
Pp
• Similar to Cp, Pp measures the capability of the process
to meet the two-sided specifications.
• It only focuses on the spread and does not take the

process centralization into consideration.

• It is recommended when the process is centered
between the specification limits.
• Cp considers the within-subgroup standard deviation and

Pp considers the total standard deviation from the sample
data.
• Pp works when there are both upper and lower

487
Cpk
• Cpk stands for the capability of the process with a k

factor adjustment.
C pk  (1  k )  C p

where
m USL  LSL
k m
USL  LSL
2
2
μ is the process mean; n is the sample size.
488
Cpk
• The formulas to calculate Cpk can also be expressed

as follows:
 USL     LSL 
C pk  min  , 
 3  s within 3  s within 

where
s within 
sp
sp 
  (x  x )
i j ij i
c4 (d  1)  (n  1)
i i
4(n  1)
d  i (ni  1) c4 
(4n  3)
489
Cpk
• Cpk measures the process’ actual capability by

taking both the variation and average of the process
into consideration.
• The process does not need to be centered between

the specification limits to make the index
meaningful.
• Cpk is recommended when the process is not in the

center between the specification limits.
• When there is only a one-sided limit, Cpk is

calculated using Cpu or Cpl.
490
Cpk
• Cpk for upper specification limit:

USL  
C pu 
3  s within
• Cpk for lower specification limit:

  LSL
C pl 
3  s within
μ is the process mean.
491
Ppk
• Ppk stands for the performance of the process with a

k factor adjustment.
Ppk  (1  k )  Pp

where
m
k
USL  LSL
2
USL  LSL
m
2
492
Ppk
• The formulas to calculate Ppk can also be expressed

as follows:
 USL     LSL 
Ppk  min  , 
 3  s overall 3  s overall 

s
s overall  ( xij  x) 2
c4 (n) s  i j
n 1
4(n  1)
c4 
(4n  3)
μ is the process mean. n is the sample size.
493
Ppk
• Similar to Cpk, Ppk measures the process capability

by taking both the variation and the average of the
process into consideration.
• Ppk solves the decentralization problem Pp cannot

overcome.
• Cpk considers the within-subgroup standard

deviation, while Ppk considers the total standard
deviation from the sample data.
• When there is only a one-sided specification limit,

Ppk is calculated using Ppu or Ppl.
494
Ppk
• Ppk for upper specification limit:

USL  
Ppu 
3  s overall
• Ppk for lower specification limit:

  LSL
Ppl 
3  s overall
495
Cpm
• Cp, Pp, Cpk, and Ppk all consider the variation of the
process. Cpk and Ppk take both the variation and the
average of the process into consideration when
measuring the process capability.

• It is possible that the process average fails to meet
the target customers require while the process still
remains between the specification limits. Cpm
(Taguchi’s capability index) helps to capture the
variation from the specified target.
496
Cpm
• Formula to calculate Cpm
min(T  LSL,USL  T )
C pm 
3  s 2  (  T )2

T is the specified target.
Note: Cpm can work only if there is a target value specified.
497
Use Minitab to Run a Process Capability Analysis
• Data File: “Capability Analysis” tab in “Sample Data.xlsx”

• Steps in Minitab to run a process capability analysis:
3) Select “HtBk” as the variable.
4) Click “OK.”
5) The histogram and the normality test results appear in the new window.

6) In this example, the p-value is 0.275, greater than the alpha level (0.05). We
fail to reject the hypothesis and conclude that the data are normally
distributed.
7) Click Stat → Quality Tools → Capability Analysis → Normal.
8) A new window named “Capability Analysis (Normal Distribution)” pops up.
9) Select “HtBk” as the single column and enter “1” as the subgroup size.
10) Enter “6” as the “Lower spec” and “7” as the “Upper spec”
11) Click “Options” button and another new window named “Capability Analysis
(Normal Distribution) – Options” pops up.
12) Enter “6.5” as the target and click “OK.”
13) Click “OK” in the “Capability Analysis (Normal Distribution)” window.
14) The capability analysis results appear in the new window. 498

499

500

Process capability
indices 501
1) If the p-value of the normality test is smaller than the alpha level (0.05), we
reject the null hypothesis and conclude that the data are not normally
distributed.
2) Click Stat → Quality Tools → Capability Analysis → Nonnormal
3) A new window named “Capability Analysis (Nonnormal Distribution)” pops up.

4) Select “HtBk” as the single column.
5) Enter “6” as the “Lower spec” and “7” as the “Upper spec.”
6) Click “Options” button and another new window named “Capability Analysis
(Nonnormal Distribution) – Options” pops up.
7) Enter “6.5” as the target and click “OK.”
8) Click “OK” in the “Capability Analysis (Nonnormal Distribution)” window.
9) The capability analysis results appear in the new window.
502

503

Process capability
indices 504
505
What is Process Stability?
• A process is said to be stable when:

• the process is in control
• the future behavior of the process is predictable at least
between some limits
• there is only random variation involved in the process.
• the causes of variation in the process are only due to

chance or common causes
• there are not any trends, patterns, or outliers in the control
chart of the process.
506
Root Causes of Variation in the Process
• Common Cause:
• Chance
• Random and anticipated
• Natural noise
• Inherent in the process
•

Unable to be eliminated from the process.
• Special Cause:
• Assignable cause
• Unanticipated
• Unnatural pattern
• Signal of changes in the process
• Able to be eliminated from the process.
507
Control Charts
• Control charts are the graphical tools to analyze

the stability of a process.
• A control chart is used to identify the presence of

potential special causes in the process and to

determine whether the process is statistically in
control.
• If the samples or calculations of samples are all in

control, the process is stable and the data from the
process can be used to predict the future
performance of the process.
508
Popular Control Charts
• I-MR Chart
• Xbar-R Chart
• Xbar-S Chart
• C Chart
• U Chart

• P Chart
• NP Chart
• EWMA Chart
• CUSUM Chart
Note: More details of the control charts will be introduced in the Control
module.
509
Process Stability vs. Process Capability
• Process stability indicates how stable a process
performed in the past.
• When the process is stable, we can use the data from

the process to predict its future behavior.

• Process capability indicates how well a process performs
with respect to meeting the customer’s specifications.
• The process capability analysis is valid only if the

process is statistically stable (i.e., in control, predictable).
• Being stable does not guarantee that the process is also

capable. However, being stable is the prerequisite to
determine whether a process is capable.
510
2.4.3 Attribute & Discrete Capability
511
Process Capability Analysis for Binomial Data
• If we are measuring the count of defectives in each

sample set to assess the process performance of
meeting the customer specifications, we use
“%Defective” (percentage of items in the samples
that are defective) as the process capability index.

N defectives
%Defective 
N overall
where Ndefectives is the total count of defectives in the samples
and Noverall is the sum of all the sample sizes.
512
Process Capability Analysis for Poisson Data
• If we are measuring the count of defects in each

sample set to assess the process performance of
meeting the customer specifications, we use Mean
DPU (defects per unit of measurement) as the
process capability index.

N defects
DPU 
N overall
where Ndefects is the total count of defects in the samples and

Noverall is the sum of all the units in the samples.
513
514
Capability and Monitoring
• In the Measure phase of the project, process

stability analysis and process capability analysis are
used to baseline the performance of current
process.

• In the Control phase of the project, process stability
analysis and process capability analysis are
combined to monitor whether the improved process
is maintained consistently as expected.
515
516

3.0 Analyze Phase
Green Belt Training: Analyze Phase
3.1 Inferential Statistics 3.4 Hyp Testing: Non-Normal Data
3.1.1 Understanding Inference 3.4.1 Mann-Whitney
3.1.2 Sampling Techniques and Uses 3.4.2 Kruskal-Wallis
3.1.3 Sample Size 3.4.3 Moods Median
3.1.4 Central Limit Theorem 3.4.4 Friedman
3.2 Hypothesis Testing 3.4.5 One Sample Sign
3.2.1 Goals of Hypothesis Testing 3.4.6 One Sample Wilcoxon
Lean Six Sigma Training - SXL

3.2.2 Statistical Significance 3.4.7 One and Two Sample Proportion
3.2.3 Risk; Alpha and Beta 3.4.8 Chi-Squared (Contingency Tables)
3.2.4 Types of Hypothesis Tests 3.4.9 Test of Equal Variances
3.3 Hypothesis Testing: Normal Data
3.3.1 One and Two Sample T-Tests
3.3.2 One sample variance
3.3.3 One Way ANOVA
517
3.1 Inferential Statistics
518

3.3.3 One Way ANOVA
519
3.1.1 Understanding Inference
520
What is Statistical Inference?
• Statistical inference is the process of making

inferences regarding the characteristics of an
unobservable population based on the
characteristics of an observed sample.

• We rely on sample data to draw conclusions about
the population from which the sample is drawn.
• Statistical inference is widely used since it is difficult

or sometimes impossible to collect the entire
population data.
521
Outcome of Statistical Inference
• The outcome or conclusion of statistical inference is

a statistical proposition about the population.
• Examples of statistical propositions:

• Estimating a population parameter

• Identifying an interval or a region where the true population
parameter would fall with some certainty
• Deciding whether to reject a hypothesis made on
characteristics of the population of interest
• Making predictions
• Clustering or partitioning data into different groups.
522
Population and Sample
• A statistical population is an entire set of
objects or observations about which
statistical inferences are to be drawn
based on its sample.
• It is usually impractical or impossible to obtain the data for the entire
population. For example, if we are interested in analyzing the

population of all the trees, it is extremely difficult to collect the data for
all the trees that existed in the past, exist now, and will exist in the
future.
• A sample is a subset of the population (like a piece of the

pie above). It is necessary for samples to be
representative of the population.
• The process of selecting a subset of observations within a

population is referred to as sampling.
523
Population and Sample
• Population Parameters • Sample Statistics

(Greek letters) (Roman letters)
_
• Mean: μ • Sample Mean: X
• Standard deviation: σ • Standard deviation: S
• Variance: σ2 • Variance: S2

~
• Median: η • Median: X
• The population parameter is the numerical summary

of a population.
• The sample statistic is the numerical measurement
calculated based on a sample of that population. It is
used to estimate the true population parameter.
524
Descriptive Statistics vs. Statistical Inference
• Descriptive statistics summarize the characteristics of a
collection of data.
• Descriptive statistics are descriptive only and they do not
make any generalizations beyond the data at hand.
• Data used for descriptive statistics are for the purpose of

representing or reporting.
• Statistical Inference
• Statistical inference makes generalizations from a sample at
hand of a population.
• Data used for statistical inference are for the purpose of
making inferences on the entire population of interest.
• A complete statistical analysis includes both
descriptive statistics and statistical inference.
525
Error Sources of Statistical Inference
• Statistical inference uses sample data to best

approximate the true features of the population.
• A valid sample must be unbiased and representative

of the population.

• Two sources of error in statistical inference:
• Random sampling error
• Selection bias.
526
Error Sources of Statistical Inference
• Random Sampling Error

• Random variation due to observations being selected
randomly
• It is inherent to the sampling process and beyond one’s
control.

• Selection Bias
• Non-random variation due to inadequate design of sampling
• It can be improved by adjusting the sampling size and
sampling strategy.
527
3.1.2 Sampling Techniques
528
What is Sampling?
• Sampling is the process of selecting objects or

observations from a population of interest about
which we wish to make a statistical inference.
• It is extensively used to collect information about a

population, which can represent physical or
intangible objects.
Population Sampling Sample
529
Advantages of Sampling
• It is usually impractical or impossible to collect the

data of an entire population:
• High cost
• Time consuming
• Unavailability of historical records
•

Dynamic nature of the population.
• Advantages of sampling a representative subset of

the population:
• Lower cost
• Faster data collection
• Easier to manipulate.
530
Uses of Sample
• With valid samples collected, we can draw statistical

conclusions about the population of our interest.
• There are two major uses of samples in making

statistical inference:

• Estimation: estimating the population parameters using the
sample statistics.
• Hypothesis testing: testing a statement about the population
characteristics using sample data.
• This module covers sample size calculation for

estimation purposes only. Sample size calculations
for hypothesis testing purposes will be covered in
the Hypothesis Testing module.
531
Basic Sampling Steps
1) Determine the population of interest
2) Determine the sampling frame
3) Determine the sampling strategy

4) Calculate the sample size
5) Conduct sampling
532
Population
• Population in statistics is the entire set of objects or

observations about which we are interested to draw
conclusions or make generalizations based on some
representative sample data.
• A population can be either physical or intangible.

• Physical: trees, customers, monitors etc.
• Intangible: credit score, pass/fail decisions etc.
• A population can be static or dynamic
• Characteristics of individuals are relatively static over time
• Items making up the population continue to change or be
generated over time
• The population covers all the items with
characteristics we are interested to analyze.
533
Sampling Frame
• In the ideal situation, the scope of a population

could be defined.
• Example: Mary is interested to know how the soup she is
cooking tastes. The population is simply the pot of soup.
• However, in some other situations, the population
cannot be identified or defined precisely.

• Example: to collect the information for an opinion poll, we do
not have a list of all the people in the world at hand.
• A sampling frame is a list of items of the population

(preferably the entire population; if not, approximate
population).
• Example: The telephone directory would be a sampling
frame for opinion poll data collection.
534
Basic Sampling Strategies
• Simple Random Sampling
• Matched Random Sampling
• Stratified Sampling

• Systematic Sampling
• Proportional to Size Sampling
• Cluster Sampling
535
Simple Random Sampling
• Simple random samples are selected in such a

way that each item in the population has an equal
chance of being selected.
• There is no bias involved in the sample selection.

Such selection minimizes the variation between the
characteristics of samples and the population.
• It is the basic sampling strategy.
536
Matched Random Sampling
• Matched random samples are samples randomly

selected in pairs, each of which has the same
attribute.
• Example:

• Researchers are interested in understanding the weight of
twins.
• Researchers are interested in understanding the patients’
blood pressure before and after taking some medicine.
537
Stratified Sampling
• A population can be grouped or “stratified” into distinct and
independent categories. An individual category can be
considered as a sub-population. Stratified samples are
randomly selected in each category of the population.
• The categories can be gender, region, income level etc.

• Stratified sampling requires advanced knowledge of the
population characteristics.
• Example: A fruit store wants to measure the quality of all

their oranges. They decide to use stratified sampling by
region to collect sample data. Since about 40% of their
oranges are from California, 40% of the sample is selected
from the California oranges sub-population. 538
Systematic Sampling
• Systematic samples are selected at regular

intervals based on an ordered list where items in the
population are arranged according to a certain
criterion.

• Systematic sampling has a random start and then
every ith item is selected going forward.
• For example, we are sampling the every 5th unit

produced on the production line.
539
Cluster Sampling
• Cluster sampling is a sampling method in which

samples are only selected from certain clusters or
groups of the population.
• It reduces the cost and time spent on the sampling

but bears the risk that the selected clusters are
biased.
• For example, selecting samples from the region

where researchers are located so that the cost and
time spent on travelling is reduced.
540
Sampling Strategy Decision Factors
• When determining the sampling strategy, we need
to consider the following factors:
• Cost and time constraints
• Nature of the population of interest

• Availability of advanced knowledge of the population
• Accuracy requirement.
541
Sample Size
• The sample size is a critical element that can

influence the results of statistical inference.
• The smaller the sample size, the higher the risk that
the sample statistic will not reflect the true

population parameter.
• The greater the sample size, the more time and

money we will spend on collecting the samples.
542
Sample Size Factors
• Is the variable of interest continuous or discrete?
• How large is the population size?
• How much risk do you want to take regarding

missing the true population parameters?
• What is the acceptable margin of error you want to

detect?
• How much is the variation in the population?
543
Sample Size Calculation for Continuous Data
• Sample size equation for continuous data
 Z  s 
2
n0   2 
 d 
 

where
n0 is the number of samples.
Z is the Z score when risk level is α/2. When α is 0.05, it is
2
1.96. When α is 0.10, it is 1.65.
s is the estimation of standard deviation in the population
d is the acceptable margin of error.
544
Sample Size Calculation for Continuous Data
• When the sample size calculated using the formula

 Z  s 
2
n0   2 
 d 
 
exceeds 5% of the population size, we use a correction

formula to calculate the final sample size.
n0
n
 n0 
1  
 N
where  Z  s 
2
n0 is the sample size calculated using equation n0   2 

d
N is the population size.  
545
Sample Size Calculation for Discrete Data
• Sample size equation for continuous data

2
 Z 
n0   2   p  (1  p)
 d 
 
where

n0 is the number of samples.
Z is the Z score when risk level is α/2. When α is 0.05, it
2
is 1.96. When α is 0.05, it is 1.65.
s is the estimation of standard deviation in the population.
d is the acceptable margin of error.
p is the proportion of one type of event occurring (e.g.,
proportion of passes).
p × (1 – p) is the estimate of variance.
546
Sample Size Calculation for Discrete Data
• When the sample size calculated using the formula

2
 Z 
n0   2   p  (1  p )
 d 
 
exceeds 5% of the population size, we use a

correction formula to calculate the final sample size.
n0
n
 n0 
1  
 N
where  Z 
2
n0 is the sample size calculated using equation n0   2   p  (1  p)

d
N is the population size.  
547
Sampling Errors
• Random Sampling Error

• Random variation due to observations being selected
randomly
• It is inherent in the sampling process and beyond
one’s control.

• Selection Bias
• Non-random variation due to inadequate design of
sampling
• It can be improved by adjusting the sampling size and
sampling strategy.
548
549

3.1.3 Sample Size
Sample Size
• The sample size is a critical element that can

influence the results of hypothesis testing.
• The smaller the sample size, the higher the risk that
the statistical conclusions will not reflect the

population relationship.
• The greater the sample size, the more time and

money we will spend on collecting the samples.
550
Sample Size Calculation
• General sample size formula for continuous data

2
  Z  Z  s 
    
n  2  
 d 
 

• General sample size formula for discrete data
 Z  Z 
2

n 2   p  (1  p)
 d 
 
551
Sample Size Calculation
• n is the number of observations in the sample.
• α is the risk of committing a false positive error.
• β is the risk of committing a false negative error.

• s is the estimation of standard deviation in the
population
• d is the size of effect you want to be able to detect.
• p is the proportion of one type of event occurring

(e.g., proportion of passes).
552
Use Minitab to Calculate the Sample Size
• Case study:
• We are interested in comparing the average retail price of a
product between two states.
• We will run a hypothesis test on the two sample means to

determine whether there is a statistically significant
difference between the retail price in the two states.
• The average retail price of the product is $23 based on our

estimation and the standard deviation is 3. We want to
detect at least a $2 difference with a 90% chance when it is
true and we can tolerate the alpha risk at 5%
• What should the sample size be?
553
• Steps to calculate the sample size in Minitab
1) Click Stat → Power and Sample Size → 2-Sample t.
2) A window named “Power and Sample Size for 2-Sample t”

appears.

3) Enter “2” as “Differences.”
4) Enter “0.9” as “Power values.”
5) Enter “3” as “Standard deviation.”
6) Click “OK.”
7) The sample size calculation results appear in a new

window.
554

555
• The sample size for each group is 49 based on the sample
size calculator.
• When the absolute value of the difference to detect decreases,
the power decreases accordingly if the sample size is kept the
same.

556
• 0.05 is the significance level by default.
• To customize the significance level, click on the “Options”
button in the window “Power and Sample Size for 2-Sample t,”
specify the significance value in the box, and click “OK.”

557
3.1.4 Central Limit Theorem
558
What is Central Limit Theorem?
• The Central Limit Theorem is one of the

fundamental theorems of probability theory.
• It states a condition under which the mean of a large

number of independent and identically-distributed

random variables, each of which has a finite mean
and variance, would be approximately normally
distributed.
559
What is Central Limit Theorem?
• Let us assume Y1, Y2, …, Yn is a sequence of n i.i.d.

random variables, each of which has finite mean μ
and variance σ2, where σ2 > 0.
• When n increases, the sample average of the n

random variables is approximately normally
distributed, with the mean equal to μ and variance
equal to σ2/n, regardless of the common distribution
Yi follows where i = 1, 2, …, n.
560
Independent and Identically Distributed
• A sequence of random variables is independent

and identically distributed (i.i.d.) if each random
variable is independent of others and has the same
probability distribution as others.

• It is one of the basic assumptions in Central Limit
Theorem.
561
Central Limit Theorem Example
• Let us assume we have10 fair die at

hand.
• Each time we roll all 10 die together

we record the average of the 10 die.

• We repeat rolling the die 50 times until
we will have 50 data points.
• Upon doing so, we will discover that

the probability distribution of the
sample average approximates the
normal distribution even though a
single roll of a fair die follows a
discrete uniform distribution. 562
Central Limit Theorem Explained in Formulas
• Let us assume Y1, Y2, …, Yn are i.i.d. random

variables with
E Yi   Y where    Y  
var(Yi )  s Y2 where 0  s Y2  

As n   , the distribution of Y becomes
approximately normally distributed with

E Y  Y
s Y2
var(Y )  s Y2 
n
563
Central Limit Theorem Application
• Use the sample mean to estimate the population

mean.
• If the assumptions of Central Limit Theorem are

met,

E (Yi )  E (Y ) where i  1,2,..., n
564
• Use standard error of the mean to measure the

standard deviation of the sample mean estimate of a
population mean.
s
SEY 
n

where s is the standard deviation of the sample and n is the
sample size.
• Standard deviation of the population mean

s
SDY 
n
where σ is the standard deviation of the population and n is
the sample size.
565
• Use a larger sample size, if economically feasible, to
decrease the variance of the sampling distribution.
• The larger the sample size, the more precise the

estimation of the population parameter.
• Use a confidence interval to describe the region which the

population parameter would fall in.
• The sample distribution approximates the normal

distribution in which 95% of the data stays within two
standard deviations from the center.
• Population mean would fall in the interval of two standard

errors of the mean away from the sample mean, 95% of
the time.
566
Confidence Interval
• The confidence interval is an interval where the

true population parameter would fall within a certain
confidence level.
• A 95% confidence interval indicates that the

population parameter would fall in that region 95%
of the time or we are 95% confident that the
population parameter would fall in that region.
• 95% is the most commonly used confidence level.
• Confidence interval is used to describe the reliability

of a statistical estimate of a population parameter.
567
Confidence Interval
• The width of a confidence interval depends on:

• Confidence level
• Sample size
• Variability in the data.

• The higher the confidence level, the wider the
confidence interval.
• The smaller the sample size, the wider the

confidence interval.
• The more variability, the wider the confidence

interval.
568
Confidence Interval of the Mean
• Confidence interval of the population mean μY of a

continuous variable Y is
  s   s 
Y  Z 2   , Y  Z   
  n 2  n 
where

Y is the sample mean
s is the standard deviation of the population
n is the sample size
Z is the Z score when risk level is α/2. When α is 0.05, it
2
is 1.96. When α is 0.05, it is 1.65.
 is (1 – confidence level). When confidence level is 95%,
α is 5%. When the confidence level is 90%, α is 10%.
Note: Since 95% is the most commonly used confidence level, 0.05 is the
most commonly used α (also called alpha level).
569
Minitab: Calculate the Confidence Interval of the Mean
• Data File:
• “Central Limit Theorem” tab in “Sample Data.xlsx”

3) Select “Cycle Time (Minutes)” as the variable.
4) The confidence level is 0.95 by default.
5) Click “OK.”
6) A new window named “Summary for Cycle Time

(Minutes)” pops up.
570

571
The 95% confidence interval of the mean is shown in the

newly-generated “Summary for Cycle Time (Minutes).”

572
• The confidence level is 95% by default.

• In order to see the confidence interval of “Cycle Time
(Minutes)” at other confidence level, we need to enter the
confidence level of our interest in the window “Graphical Summary”
and click “OK.”
• The following example shows how to generate 90% confidence
interval of the mean.

573
3.2 Hypothesis Testing
574

3.3.3 One Way ANOVA
575
3.2.1 Goals of Hypothesis Testing
576
What is Hypothesis Testing?
• A hypothesis test is a statistical method in which a

specific hypothesis is formulated about a population,
and the decision of whether to reject the hypothesis
is made based on sample data.

• Hypothesis tests help to determine whether a
hypothesis about a population or multiple
populations is true with certain confidence level
based on sample data.
577
Hypothesis Testing Examples
• Hypothesis testing tries to answer whether there is a

difference between different groups or there is some
change occurring.
• Are the average SAT scores of graduates from high school A

and B the same?

• Is the error rate of one group of operators higher than that of
another group?
• Are there any non-random causes influencing the height of

kids in one state?
578
What is Statistical Hypothesis?
• A statistical hypothesis is an assumption about one

or multiple population.
• It is a statement about whether there is any

difference between different groups.

• It can be a conjecture about the population
parameters or the nature of the population
distributions.
• A statistical hypothesis is formulated in pairs:

• Null Hypothesis
• Alternative Hypothesis.
579
Null and Alternative Hypotheses
• Null Hypothesis (H0) states that:

• there is no difference in the measurement of different groups
• no changes occurred
• sample observations result from random chance.
• Alternative Hypothesis (H1 or Ha) states that:

• there is a difference in the measurement of different groups
• some changes occurred
• sample observations are affected by non-random causes.
580
• A statistical hypothesis can be expressed in

mathematical language by using population
parameters (Greek letters) and mathematical
symbols.
• Population Parameters • Mathematical Symbols

(Greek letters)
• Equal: =
• Mean: μ • Not equal: ≠
• Standard deviation: σ • Greater than: >
• Variance: σ2 • Smaller than: <
• Median: η
581
• Examples of null and alternative hypotheses written

in mathematical language.
H0: μ1 = μ2
H1: μ1 ≠ μ2
H0: σ1 = 0

H1: σ1 ≠ 0
H0: η1 = 10
H1: η1 > 10
H0: μ1 = 10
H1: μ1 < 10
582
Hypothesis Testing Conclusion
• There are two possible conclusions of hypothesis

testing:
• Reject the null
• Fail to reject the null.

• When there is enough evidence based on the
sample information to prove the alternative
hypothesis, we reject the null.
• When there is not enough evidence or the sample

information is not sufficiently persuasive, we fail to
reject the null.
583
Decision Rules in Hypothesis Testing
The critical region is the
shaded area under the
sampling distribution and
Sampling distribution based
beyond the critical value.
on the null hypothesis
The critical value defines
the boundary of critical
region.
The critical value is derived from

statistical tables. Different hypotheses

use different statistical tables. It is the
threshold value to determine whether
to reject the null.
The acceptance region is the non-shaded The test statistic is calculated based on a
area under the sampling distribution and function of sample data. Different hypotheses
not beyond the critical value. It is the use different formulas to calculate test statistic. It
opposite area of critical region. can be any value along the x axis.
584
• The test statistic in hypothesis testing is a value
calculated using a function of the sample.
• Test statistics are considered the sample data’s
numerical summary that can be used in hypothesis
testing.
• Different hypothesis tests have different formulas to

calculate the test statistic.
• The critical value in hypothesis testing is a
threshold value to which the test statistic is
compared in order to determine whether the null
hypothesis is rejected.
• The critical value is obtained from statistical tables.
• Different hypothesis tests need different statistical
tables for critical values.
585
• When the test statistic falls into the acceptance

region, we fail to reject the null and claim that there
is no statistically significant difference between the
groups.

• When the test statistic falls into the critical region,
we reject the null and claim that there is a
statistically significant difference between the
groups.
586
• The proportion of the area under the sampling

distribution and beyond the critical value indicates α
risk (also called α level). The most commonly
selected α level is 5%.

• The proportion of the area under the sampling
distribution and beyond the test statistic is the p-
value. It is the probability of getting a test statistic at
least as extreme as the observed one, given the null
is true.
587
• When the p-value is smaller than the α level, we

reject the null and claim that there is a statistically
significant difference between different groups.
• When the p-value is higher than the α level, we fail

to reject the null and claim that there is no
statistically significant difference between different
groups.
588
Steps in Hypothesis Testing
• Step 1: State the null and alternative hypothesis.
• Step 2: Determine α level.
• Step 3: Collect sample data.

• Step 4: Select a proper hypothesis test.
• Step 5: Run the hypothesis test.
• Step 6: Determine whether to reject the null.
589
3.2.2 Statistical Significance
590
Statistical Significance
• In statistics, an observed difference is statistically

significant if it is unlikely that the difference occurred
by pure chance, given a predetermined probability
threshold.

• Statistical significance indicates that there are some
non-random factors causing the result to take place.
• The statistical significance level in hypothesis

testing indicates the amount of evidence which is
sufficiently persuasive to prove that a difference
between groups exists not due to random chance
alone.
591
Practical Significance
• An observed difference is practically significant

when it is large enough to make a practical
difference.
• A difference between groups that is statistically

significant might not be large enough to be
practically significant.
• In some business situations, statistical differences

can have little to no meaning because the difference
is not large enough to be practical for a business to
act upon.
592
Example
• You started to use the premium gas recently, which
was supposed to make your car run better.
• After running a controlled experiment to measure the

performance of the car before and after using the
premium gas, you performed a statistical hypothesis

test and found that the difference before and after
was statistically significant.
• Using premium gas did improve the performance.
• However, due to the high cost of the premium gas,

you decided that the difference was not large enough
to make you pay extra money for it. In other words,
the difference is not practically significant. 593
3.2.3 Risk; Alpha & Beta
594
Errors in Hypothesis Testing
• In statistical hypothesis testing, there are two types

of errors:
• Type I Error
• a null hypothesis is rejected when it is true in fact.

• Type II Error
• a null hypothesis is not rejected when it is not true in fact.
595
Type I Error
• Type I error is also called false positive, false alarm,

or alpha (α) error.
• Type I error is associated with the risk of accepting

false positives.

• It occurs when we think there is a difference
between groups but in fact there is none.
• Example: telling a patient he is sick and in fact he is

not.
596
Alpha (α)
• α indicates the probability of making a type I error. It ranges
from 0 to 1.
• α risk is the risk of making a type I error.
• 5% is the most commonly used α.

• (100% – α) is the confidence level which is used to calculate
the confidence intervals.
• When making a decision on whether to reject the null, we

compare the p-value against α:
• If p-value is smaller than α, we reject the null
• If p-value is greater than α, we fail to reject the null.
• To reduce the α risk, we decrease the α value to which the p-

value is compared. 597
Type II Error
• Type II error is also called false negative, oversight,

or beta (β) error.
• Type II error is associated with the risk of accepting

false negatives.

• It occurs when we think there is not any difference
between groups but in fact there is.
• Example: telling a patient he is not sick and in fact

he is.
598
Beta (β)
• β indicates the probability of making a type II error. It
ranges from 0 to 1.
• β risk is the risk of making a type II error.
• 10% is the most commonly used β.

• (100% – β) is called power, which denotes the probability
of detecting a difference between groups when in fact the
difference truly exists.
• To reduce the β risk, we increase the sample size.
• When holding other factors constant, β is inversely

related to α.
599
3.2.4 Types of Hypothesis Tests
600
Two Types of Hypothesis Tests
• Two-tailed hypothesis test

• It is also called two-sided hypothesis test
• It is a statistical hypothesis test in which the critical region is
split into two equal areas, each of which stays on one side
of the sampling distribution.

• One-tailed hypothesis test
• It is also called one-sided hypothesis test
• It is a statistical hypothesis test in which the critical region is
only on one side of the sampling distribution.
601
Two-Tailed Hypothesis Test
• A two-tailed hypothesis test is used when we care about
whether there is a difference between groups and we do not
care about the direction of the difference.
• Examples of two-tailed hypothesis tests:
H0: μ1 = 10
Ha: μ1 ≠ 10

The null hypothesis (H0) is rejected when:
• the test statistic falls into either half of the critical region
• the test statistic is either sufficiently small or sufficiently large
• the absolute value of the test statistic is greater than the absolute
value of the critical value.
• A two-tailed hypothesis test is the most commonly used
hypothesis test. In the next modules we will cover more details
about it.
602
Two-Tailed Hypothesis Test
Critical

Acceptance Region
Region
α/2 α/2
603
One-Tailed Hypothesis Test
• A one-tailed hypothesis test is used when we care

about one direction of the difference between
groups.
• Examples of one-tailed hypothesis tests:
H0: μ1 = 10

Ha: μ1 > 10
• The null hypothesis is rejected when the test

statistic:
• falls into the critical region which only exits on the right side
of the sampling distribution
• is sufficiently large
• is greater than the critical value.
604

Critical
Acceptance Region
Region
605
• Examples of one-tailed hypothesis test

H0: μ1 = 10
Ha: μ1 < 10

• The null hypothesis is rejected when the test
statistic:
• falls into the critical region which only exits on the left side of
the sampling distribution
• is sufficiently small
• is smaller than the critical value.
606

Critical Acceptance
Region Region
607
3.3 Hypothesis Tests: Normal Data
608

3.3.3 One Way ANOVA
609
3.3.1 1 & 2 Sample T-Tests
610
What is a T-Test?
• In statistics, a t-test is a hypothesis test in which the

test statistic follows a Student t distribution if the null
hypothesis is true.
• We apply a t-test when the population variance (σ)

is unknown and we use the sample standard
deviation (s) instead.
611
What is One Sample T-Test?
• One sample t-test is a hypothesis test to study

whether there is a statistically significant difference
between a population mean and a specified value.
• Null Hypothesis (H0): μ = μ0

• Alternative Hypothesis (Ha): μ ≠ μ0

where μ is the mean of a population of our interest
and μ0 is the specific value we want to compare
against.
612
Assumptions of One Sample T-Test
• The sample data drawn from the population of

interest are unbiased and representative.
• The data of the population are continuous.

• The data of the population are normally distributed.
• The variance of the population of our interest is

unknown.
• One sample t-test is more robust than the z-test

when the sample size is small (< 30).
613
Normality Test
• To check whether the population of our interest is

normally distributed, we need to run normality test.

• Alternative Hypothesis (Ha): The data are not normally
distributed.

• There are a lot of normality tests available:
• Anderson-Darling
• Sharpiro-Wilk
• Jarque-Bera etc.
614
Test Statistic and Critical Value of One Sample T-Test
• Test Statistic
Y
tcalc  , where
s
n
Y is the sample mean, n is the sample size, and s is the sample standard
deviation

n
 (Y  Y )
i
2
s i 1
n 1
• Critical Value
• tcrit is the t-value in a Student t distribution with the
predetermined significance level α and degrees of freedom (n
–1).
• tcrit values for a two-sided and a one-sided hypothesis test
with the same significance level α and degrees of freedom (n
– 1) are different.
615
Decision Rules of One Sample T-Test
• Based on the sample data, we calculated the test

statistic tcalc, which is compared against tcrit to make
a decision of whether to reject the null.
• Null Hypothesis (H0): μ = μ0
• Alternative Hypothesis (Ha): μ ≠ μ0

• If |tcalc| > tcrit, we reject the null and claim there is a
population mean μ and the specified value μ0.
• If |tcalc| < tcrit, we fail to reject the null and claim there
is not any statistically significant difference between
the population mean μ and the specified value μ0.
616
Use Minitab to Run a One-Sample T-Test
• Case study: We want to compare the average height of
basketball players against 7 feet.
• Data File: “One Sample T-Test” tab in “Sample Data.xlsx”
• Null Hypothesis (H0): μ = 7

• Alternative Hypothesis (Ha): μ ≠ 7
617
• Step 1: Test whether the data are normally

distributed

3) Select “HtBk” as the variable.
4) Click “OK.”
5) A new window named “Probability Plot of HtBk” appears,

which covers the results of the normality test.
618

619
• Since the p-value of the normality is 0.275, which is greater than
alpha level (0.05), we fail to reject the null and claim that the data are
• If the data are not normally distributed, you need to use hypothesis
tests other than the one sample t-test.

620
• Step 2: Run the one-sample t-test
1) Click Stat → Basic Statistics → 1 Sample t.

2) A new window named “1 Sample t (Test and Confidence
Interval)” pops up.
3) Click in the blank box under “Samples in columns” and

“HtBk” appears in the list box on the left.
4) Select “HtBk” as the “Samples in columns.”
5) Check the box of “Perform hypothesis test.”
6) Enter the hypothesized value “7” into the box next to
“Perform hypothesis test.”
7) Click “OK.”
8) The one-sample t-test result appears automatically in the
session window.
621

622

Null Hypothesis (H0): μ = 7
Alternative Hypothesis (Ha): μ ≠ 7
Since the p-value is smaller than alpha level (0.05), we reject the
null hypothesis and claim that average of basketball players is
statistically different from 7 feet.
623
What is Two Sample T-Test?
• Two sample t-test is a hypothesis test to study

whether there is a statistically significant difference
between the means of two populations.
• Null Hypothesis (H0): μ1 = μ2

• Alternative Hypothesis (Ha): μ1 ≠ μ2

where μ1 is the mean of one population and μ2 is the mean of
the other population of our interest.
624
Assumptions of Two Sample T-Tests
• The sample data drawn from both populations are

unbiased and representative.
• The data of both populations are continuous.

• The data of both populations are normally
distributed.
• The variances of both populations are unknown.
• Two sample t-test is more robust than a z-test when

the sample size is small (< 30).
625
Three Types of Two Sample T-Tests
1. Two sample t-test when the variances of two

populations are unknown but equal
• Two sample t-test (when σ21 = σ22)
2. Two sample t-test when the variances of the two

population are unknown and unequal
• Two sample t-test (when σ21 ≠ σ22)
3. Paired t-test when the two populations are

dependent of each other
626
Test of Equal Variance
• To check whether the variances of two populations

of interest are statistically significant different, we
use the test of equal variance.
• Null Hypothesis (H0): s 1  s 2

2 2
• Alternative Hypothesis (H1): s 1  s2

2 2

• An F-test is used to test the equality of variances
between two normally distributed populations.
627
• An F-test is a statistic hypothesis test in which the

test statistic follows an F-distribution when the null
hypothesis is true.
• The most known F-test is the test of equal variance

for two normally distributed populations.
• The F-test is very sensitive to non-normality. When

any one of the two populations is not normal, we
use the Brown-Forsythe test for checking the
equality of variances.
628
• Test Statistic
2
s1
Fcalc  2
s2
where
s1 and s2 are the sample standard deviations.

• Critical Value
• Fcrit is the F value in a F distribution with the predetermined
significance level α and degrees of freedom (n1 – 1) and (n1 –
1).
• Fcrit values for a two-sided and a one-sided F-test with the
same significance level α and degrees of freedom (n1 – 1) and
(n1 – 1) are different.
629

statistic Fcalc, which is compared against Fcrit to
make a decision of whether to reject the null.
2 2
• Alternative Hypothesis (Ha): s 1  s 2

2 2

• If Fcalc > Fcrit, we reject the null and claim there is a
variances of the two populations.
• If Fcalc < Fcrit, we fail to reject the null and claim there
the variances of the two populations.
630
Test Statistic & Critical Value of a Two Sample T-Test

when σ21 = σ22
• Test Statistic
Y1 Y 2 (n1  1) s1  (n2  1) s2
2 2
tcalc  s pooled 
1 1 (n1  n2  2)
s pooled 
n1 n2
where

Y 1 and Y 2 are the sample means of the two populations of our interest.
n1 and n2 are the sample sizes. n1 is not necessarily equal to n2.
spooled is a pooled estimate of variance. s1 and s2 are the sample
standard deviations.
• Critical Value
tcrit is the t value in a Student t distribution with the predetermined
significance level α and degrees of freedom (n1 + n2 – 2).
tcrit values for a two-sided and a one-sided t-test with the same
significance level α and degrees of freedom (n1 + n2 – 2) are different.
631
Test Statistic & Critical Value of a Two Sample T-Test

when σ21 ≠ σ22
2
• Test Statistic  s12 s2 2 
 
Y1 Y 2 n  n 
 1 2 
tcalc  2 2
df  2 2
s1 s2  s1   s2 
  n  n 
n1 n2  1   2
n1  1 n2  1
where

Y 1 and Y 2 are the sample means of the two populations of our interest.
n1 and n2 are the sample sizes. n1 is not necessarily equal to n2.
s1 and s2 are the sample standard deviations.
• Critical Value
significance level α and degrees of freedom df calculated using the
formula above.
significance level α and degrees of freedom df are different.
632
Test Statistic & Critical Value of a Paired T-Test
• Test Statistic
d
tcalc 
sd
n
where

d is the difference between each pair of data.
d is the average of d.
n is the sample size of either population of interest.
sd is standard deviation of d.
• Critical Value
significance level α and degrees of freedom (n – 1).
significance level α and degrees of freedom (n – 1) are different.
633
Decision Rules of a Two Sample T-Test

statistic tcalc, which is compared against tcrit to make
a decision of whether to reject the null.

• If |tcalc| > tcrit, we reject the null and claim there is a
statistically significant difference between the means
of the two populations.
• If |tcalc| < tcrit, we fail to reject the null and claim there
the means of the two populations.
634
Use Minitab to Run a Two-Sample T-Test
• Case study: We are trying to compare the average

retail price of a product in state A and state B.
• Data File: “Two-Sample T-Test” tab in “Sample Data.xlsx”

635
• Step 1: Test the normality of the retail price for both state
A and B.
2) A window named “Graphical Summary” pops up.
Select “Retail Price” as the “Variables.”

3)
4) Click in the blank box right below “By variables (optional)”

and “State” appears in the list box on the left.
5) Select the “State” as the “By variables (optional).”
6) Click “OK.”
7) The normality test results would appear in the new windows.
636

637

Both retail price data of state A and B are normally distributed since the
p-values are both greater than alpha level (0.05).
If any of the data series is not normally distributed, we need to use other

hypothesis testing methods other than the two sample t-test.
638
• Step 2: Test whether the variances of the two data

sets are equal.
2 2
• Alternative Hypothesis (Ha): 1 s 2

 s 2
2
1) Click Basic Statistics → 2 Variances.

2) A new window named “2 Variances (Test and Confidence

Interval” pops up.
3) Click in the blank box of “Samples” and “Retail Price”
appears in the list box on the left.
4) Select “Retail Price” as the “Samples.”
5) Click in the blank box of “Subscripts” and “State” appears
in the list box on the left.
6) Select “State” as the “Subscripts.”
7) Click “OK.”
8) The results show up in the session window.
639

640
Because the retail prices at state A
and state B are both normally
distributed, an F test is used to test
their variance equality.
The p-value of F test is 0.870,

greater than the alpha level (0.05),

so we fail to reject the null
hypothesis and we claim that the
variances of the two data sets are
equal. We will use the two sample
t-test (when σ21 = σ22) to compare
the means of the two groups.
If σ21 ≠ σ22, we will use the two

sample t-test (when σ21 ≠ σ22) to
compare the means of the two
groups.
641
• Step 3: Run two-sample t-test to compare the

means of two groups.
1) Click Stat → Basic Statistics → 2-Sample t.
2) A new window named “2-Sample t (Test and Confidence Interval)”
pops up.
3) Click in the blank box right next to “Samples” and the “Retail

Price” appears in the list box on the left.
4) Select “Retail Price” as the “Samples.”
5) Click in the blank box right next to “Subscripts” and the “State”
6) Select “State” as the “Subscripts.”
7) Check the box “Assume equal variances.”
8) Click “OK.”
9) The results of the two-sample t-test (when σ1 = σ2) appear
automatically in the session window.
642

643

Since the p-value of the t-test (assuming equal variance) is
0.665, greater than the alpha level (0.05), we fail to reject
the null hypothesis and we claim that the means of the two
data sets are equal.
644
• If the variances of the two groups do not equal, we will
need to use the two-sample t-test (when σ1 ≠ σ2) to
compare the means of the two groups.
• In the window of “2-Sample (Test and Confidence Interval),”
uncheck the box next to “Assume equal variances” and run the 2-
sample t-test again.

Since the p-value of the t-test (assuming
unequal variance) is 0.666, greater than the
alpha level (0.05), we fail to reject the null
hypothesis and we claim that the means of
two groups are equal.
645
Use Minitab to Run a Paired T-Test
Case study:
• We are interested to know whether the average salaries
($1000/yr.) of male and female professors at the same
university are the same.
• Data File: “Paired T-Test” tab in “Sample Data.xlsx”
• The data were randomly collected from 22 universities. For

each university, the salaries of a male and female
professors were randomly selected.
• Null Hypothesis (H0): μmale – μfemale = 0

• Alternative Hypothesis (Ha): μmale – μfemale ≠ 0
646
• Step 1: Create a new column of the difference

between of two data sets of interest and name it
“Difference.”
1) Right click on the newly-generated column, “Difference.”

2) Select “Formulas” and then “Assign Formula to Column.”
3) A new window named “Assign Formula to C3 (Difference)”
pops up.
4) Enter “ ‘MALES’-‘ FEMALES’ ” into the box right below
“Expression.”
5) Click “OK.”
6) The values would appear in the column “Difference.”
647

648
• Step 2: Test whether the difference is normally

distributed.
• Null Hypothesis (H0): The difference between two data sets
is normally distributed.
• Alternative Hypothesis (Ha): The difference between two
data sets is not normally distributed.

3) Select “Difference” as the “Variable.”
4) Click “OK.”
5) The normality test results would appear in the new
window.
649

650
• The p-value of the normality test is 0.287, greater than
the alpha level (0.05), so we fail to reject the null
hypothesis and we claim that the difference is normally
distributed.
• When the difference is not normally distributed, we need
other hypothesis testing methods.

651
• Step 3: Run the paired t-test to compare the means

of two dependent data sets.
1) Click Stat → Basic Statistics → Paired t.

2) A new window named “Paired t (Test and Confidence
Interval)” pops up.

3) Click in the blank box right next to “First Sample” and the
three columns names appear in the list box on the left.
4) Select “MALES” as “First Sample.”
5) Select “FEMALES” as “Second Sample.”
6) Click “OK.”
7) The paired t-test results appears in the session window.
652

653
• The p-value of the paired t-test is 0.213, greater

than the alpha level (0.05), so we fail to reject the
null hypothesis and we claim that there is no
salaries of male and those of female professors.

654
3.3.2 One Sample Variance
655
What is One Sample Variance Test?
• One sample variance test is a hypothesis testing

method to study whether there is a statistically
significant difference between a population variance
and a specified value.

2

• Alternative Hypothesis (Ha): s 2  s0
2
where σ2 is the variance of a population of our interest and s 0

2
is the specific value we want to compare against.
656
Chi-Square Test
• A chi-square test is a statistical hypothesis test in

which the test statistic follows a chi-square
distribution when the null hypothesis is true.
• A chi-square test can be used to test the equality

between the variance of a normally distributed
population and a specified value.
657
Chi-Square Test
• Test Statistic
(n  1) s 2
 calc 
2
s 02
where
s2 is the observed variance and n is the sample size.

s 0 2 is the specified value we compare against.
• Critical Value
•  crit 2 is the χ2 value in a χ2 distribution with the
predetermined significance level α and degrees of
freedom (n – 1).
•  crit 2 values for a two-sided and a one-sided χ2-
test with the same significance level α and
degrees of freedom (n –1) are different.
658
Chi-Square Test
 2
 2
statistic calc , which is compared against crit to
make a decision of whether to reject the null.
02
• Alternative Hypothesis (Ha): s  s 0
2

• If  calc
2
  2
crit , we reject the null and claim there is
a statistically significant difference between the
population variance and the specified value.
• If  calc
2
  2
crit , we fail to reject the null and claim
there is not any statistically significant difference
between the population variance and specified
value.
659
Use Minitab to Run a One Sample Variance Test
• Case study: We are interested in comparing the variance

of the height of basketball players with zero.
• Data File: “One Sample T-Test” tab in “Sample Data.xlsx”
• Null Hypothesis (H0): σ2 = 0

• Alternative Hypothesis (Ha): σ2 ≠ 0
660
• Step 1: Test whether the data are normally

distributed

3) Select “HtBk” as the “Variable.”
4) Click “OK.”
5) The normality test result appears in the new window.
661

662

• Alternative Hypothesis (Ha): The data are not normally
distributed.
• Since the p-value of the normality is 0.275, greater than alpha
level (0.05), we fail to reject the null and claim that the data are

• If the data are not normally distributed, you need to use other
hypothesis tests.
663
• Step 2: Check whether the population variance is

equal to the specified value (zero).
1) Click Stat → Basic Statistics → 1 Variance.
2) A new window named “1 Variance” appears.
3) Click in the blank box right below “Columns” and the “HtBk”
4) Check the box of “Perform hypothesis test.”

5) Select “Hypothesized variance” from the dropdown menu.
6) Enter “0” as the “Hypothesized variance.”
7) Click “OK.”
8) An alert window pops up saying “Invalid hypothesized value. The
following value is too small: “0”. Please specify: A single numeric
constant. Value > 0.”
9) Click “OK” in the alert message window.
10)Enter “0.0000000001” as the “Hypothesized variance” instead.
11)Click “OK.”
12)The one sample variance test appears in new window.
664

665

666
Since the p-value of the one

sample variance test is smaller
than alpha level (0.05), we reject
the null and claim that the
population variance is
statistically different from zero.

Another way to see whether the
population variance is
statistically different from zero is
to check whether zero stays
between the upper and lower
confidence interval boundaries of
the variance. If yes, we fail to
reject the null hypothesis and
claim that the population
variance is not statistically
different from zero.
667
668

3.3.3 One Way ANOVA
What is One-Way ANOVA?
• One-way ANOVA (one-way analysis of variance) is

a statistical method to compare means of two or
more populations.
• Null Hypothesis (H0): 1   2  ...   k

• Alternative Hypothesis (Ha): at least one μi is different,

where i is any value from 1 to k.
• It is a generalized form of the two sample t-test

since a two sample t-test compares two population
means and one-way ANOVA compares k population
means where k ≥ 2.
669
Assumptions of One-Way ANOVA
• The sample data drawn from k populations are

unbiased and representative.
• The data of k populations are continuous.

• The data of k populations are normally distributed.
• The variances of k populations are equal.
670
How ANOVA Works
• ANOVA compares the means of different groups by

analyzing the variances between and within groups.
• Let us say we are interested in comparing the

means of three normally distributed populations. We

randomly collected one sample for each population
of our interest.
• Null Hypothesis (H0): μ1 = μ2 = μ3

• Alternative Hypothesis (Ha): one of the μ is different from the
others.
671
How ANOVA Works
• Based on the sample data, the means of the three
populations might look different because of two variation
sources.
1) Variation between groups
There are non-random factors leading to the variation between groups.
2) Variation within groups

There are random errors resulting in the variation within each
individual group.
• What we care about the most is the variation between

groups since we are interested in whether the groups are
statistically different from each other.
• Variation between groups is the signal we want to detect

and variation within groups is the noise which corrupts
the signal. www.LeanSigmaCorporation.com
672
How ANOVA Works
Individual observation: Yij
Grand mean of all

observations: Y

Group means of each
individual sample:Y j
k nj
Total Variation = SS(Total) =  (Yij Y )

2
j 1 i 1
k
Between Variation = SS(Between) =  n j (Y j Y ) 2
j 1
k nj
Within Variation = SS(Within) =  (Y
j 1 i 1
ij Y j ) 2
673
How ANOVA Works
• Variation Components
• Total Variation = Variation Between Groups + Variation

Within Groups
k nj k k nj
 (Y ij Y ) 2
=  n (Y
j 1
j j Y ) 2
+  (Y ij Y j ) 2
j 1 i 1 j 1 i 1

• Total Variation = sums of squares of the vertical difference
between the individual observation and the grand mean
• Variation Between Groups = sums of squares of the vertical

difference between the group mean and the grand mean
• Variation Within Groups = sum of squares of the vertical

difference between the individual observation and the group
mean
674
How ANOVA Works
• Degrees of Freedom (DF)

• In statistics, the degrees of freedom is the number of
unrestricted values in the calculation of a statistic.
• Degrees of Freedom Components

• DFtotal= DFbetween + DFwithin
• DFtotal = n – 1
• DFbetween= k – 1
• DFwithin= n – k
where
n is the total number of observations

k is the number of groups.
675
How ANOVA Works
• Signal-to-Noise Ratio (SNR)
• SNR denotes the ratio of a signal to the noise corrupting the
signal.
• It measures how much a signal has been corrupted by the noise.
• When it is higher than 1, there is more signal than noise.
• The higher the SNR, the less the signal has been corrupted by the
noise.
• F-ratio is the SNR in ANOVAk

MS
SSbetween
DF
 (Y j  Y ) 2
j 1
(k  1)
F between
 between

 Y 
nj
MS within SS within k
2
DFwithin ij Y j (n  k )
j 1 i 1
• In ANOVA, we use the F-test to compare the means of different

groups. The F-ratio calculated as above is the test statistic Fcalc.
• The critical value (Fcri) in an F-test can be derived from the F table
with predetermined significance level (α) and with (k –1) degrees
of freedom in the numerator and (n – k) degrees of freedom in the
denominator.
676
How ANOVA Works
• Null Hypothesis (H0): 1  2  ...  k
• Alternative Hypothesis (Ha): at least one μi is

different, where i is any value from 1 to k.

• If |Fcalc| < Fcrit, we fail to reject the null and claim that
the means of all the populations of our interest are
the same.
• If |Fcalc| > Fcrit, we reject the null and claim that there
is at least one mean different from the others.
677
Model Validation
• ANOVA is a modeling procedure. To make sure the

conclusions made in ANOVA are reliable, we need
to perform residuals analysis.
• Good residuals:

• Have a mean of zero
• Are normally distributed
• Are independent of each other
• Have equal variance.
678
Use Minitab to Run an ANOVA
• Case study: We are interested in comparing the average
startup costs of five kinds of business.
• Data File: “One-Way ANOVA” tab in “Sample Data.xlsx”
• Null Hypothesis (H0): μ1 = μ2 = μ3 = μ4 = μ5

• Alternative Hypothesis (Ha): at least one of the five

means is different from others.
679
• Step 1: Test whether the data for each level are

3) Select the “Cost” as the variable.
4) Click in the blank box right next to “By variables (optional)”

and the “Business” appears in the list box on the left.
5) Select the “Business” as the “By variables (optional).”
6) Click “OK.”
7) The normality results appear in the new window.

680

681
Business = X1 Business = X3
Business = X5

Business = X2 Business = X4
682
Null Hypothesis (H0): The data are normally distributed.
Alternative Hypothesis (Ha): The data are not normally
distributed.
• Since the p-values of normality tests for the five data sets
are higher than alpha level (0.05), we fail to reject the

null hypothesis and claim that the startup costs for any of
the five businesses are normally distributed.
• If any of the five data sets are not normally distributed,

we need to use other hypothesis testing methods other
than one-way ANOVA.
683
• Step 2: Test whether the variance of the data for

each level is equal to the variance of other levels.
Null Hypothesis (H0): s 1  s 2  s 3  s 4  s 5

2 2 2 2 2
Alternative Hypothesis (Ha): at least one of the variances is

different from others.

1) Click Stat → ANOVA → Test for equal variances.
2) A new window named “Test for Equal Variances” pops up.
3) Select the “Cost” as the “Response.”
4) Select the “Business” as the “Factors.”
5) Click “OK.”
6) The results shows up in a new window.
7) Use the Bartlett’s test for testing the equal variances between five
levels in this case since there are more than two levels in the data
and the data of each level are normally distributed.
684

685

The p-value of Barlett’s test is 0.777, greater than the alpha level (0.05),
so we fail to reject the null hypothesis and we claim that the variances of
five groups are equal.
If the variances are not all equal, we need to use other hypothesis testing
methods other than one-way ANOVA.
686
• Step 3: Test whether the mean of the data for each level
is equal to the means of other levels.
• Null Hypothesis (H0): μ1 = μ2 = μ3 = μ4 = μ5

• Alternative Hypothesis (Ha): at lease one of the means is

1) Click Stat → ANOVA → One-way.
2) A new window named “One-Way Analysis of Variance” pops
up.
3) Select “Cost” as “Response.”
4) Select “Business” as “Factor.”
5) Check the boxes “Store residuals” and “Store fits.”
6) Click “OK.”
7) The ANOVA results appear in the session window. The fitted
response and the residuals are stored in the data table.
687

688

Since the p-value of the F test is 0.018, lower than the alpha level
(0.05), the null hypothesis is rejected and we conclude that the at
least one of the means of the five groups is different from others.
689
• Step 4: Test whether the residuals are normally

distributed with mean equal zero
1) The residuals have been stored in the data table in step 4.

3) A new window named “Graphical Summary” appears.
4) Select “RESI1” as the “Variables.”
5) Click “OK.”
6) The normality test results show up in a new window.

690

691
• The p-value of the normality test is 0.255, greater than
the alpha level (0.05), and we conclude that the residuals
are normally distributed.
• The mean of the residuals are 0.0000.

692
• Step 5: Check whether the residuals are

independent of each other.
• If the residuals are in time order, we can plot IR charts to
check the independence. When no data points on the IR
charts fail any tests, the residuals are independent of each
other.

• If the residuals are not in time order, the IR charts cannot
deliver reliable conclusion on the independence.
1) Click Stat → Control Charts → Variables Charts for

Individuals → I-MR.
2) A new window named “Individuals – Moving Range Chart”
pops up.
4) Click “OK.”
693

694
• Step 6: Plot residuals versus fitted values and check

whether there is any systematic pattern.
1) Click Graph → Scatter Plot.

Click “OK.”

3)
4) A new window named “Scatterplot – Simple” pops up.
5) Select the “RESI1” as the “Y variables.”
6) Select the “FITS1” as “X variables.”
7) Click “OK.”
8) The charts appears in a new window.
9) If the data points spread out evenly at any of the five
levels, we claim that the residuals have equal variances
across the five levels.
695

696
3.4 Hypothesis Testing Non-Normal Data
697

3.3.3 One Way ANOVA
698
699

3.4.1 Mann-Whitney
What is the Mann-Whitney Test?
• The Mann-Whitney test (also called Mann-Whitney

U test or Wilcoxon rank-sum test) is a statistical
hypothesis test to compare the medians of two
populations that are not normally distributed.
• Null Hypothesis (H0): η1 = η2

• Alternative Hypothesis (Ha): η1 ≠ η2
where η1 is the median of one population and η1 is

the median of the other population.
700
Mann-Whitney Test Assumptions
• The sample data drawn from the populations of
• The data of both populations are continuous or

ordinal when the spacing between adjacent values
is not constant.

• The two populations are independent to each other.
• The Mann-Whitney test is robust for the non-

normally distributed population.
• The Mann-Whitney test can be used when shapes

of the two populations’ distributions are different.
701
How Mann-Whitney Test Works
• Step 1: Group the two samples from two populations

(sample 1 is from population 1 and sample 2 is from
population 2) into a single data set and then sort the
data in ascending order ranked from 1 to n, where n
is the total number of observations.

• Step 2: Add up the ranks for all the observations
from sample 1 and call it R1. Add up the ranks for all
the observations from sample 2 and call it R2.
702
• Step 3: Calculate the test statistics
U  min(U1 ,U 2 )
n1 (n1  1)
where U1  n1n2   R1
2

n2 (n2  1)
U 2  n1n2   R2
2
n1 and n2 are the sample sizes.
R1 and R2 are the sum of ranks for observations from

sample 1 and 2 respectively.
703
• Step 4: Make a decision on whether to reject the null

hypothesis.

• If both of the sample sizes are smaller than 10, the
distribution of U under the null hypothesis is
tabulated.
• The test statistic is U and, by using the Mann-Whitney table,
we would find the p-value.
• If the p-value is smaller than alpha level (0.05), we reject the
null hypothesis.
• If the p-value is greater than alpha level (0.05), we fail to
reject the null hypothesis.
704
• If both sample sizes are greater than 10, the
distribution of U can be approximated by a normal
distribution. In other words, U   follows a
standard normal distribution. s
U 

Z calc 
s
where
n1n2 n1n2 (n1  n2  1)

 s
2 12
When |Zcalc| is greater than Z value at α/2 level (e.g., when α = 5%,
the z value we compare |Zcalc| to is 1.96), we reject the null
hypothesis.
705
Use Minitab to Run a Mann-Whitney Test
• Case study: We are interested in comparing

customer satisfaction between two types of
customers using a nonparametric (i.e., distribution-
free) hypothesis test: Mann-Whitney test.
• Data File: “Mann-Whitney” tab in “Sample Data.xlsx”

706
• Steps to run a Mann-Whitney Test in Minitab:
1) Un-stack the data into two separate columns:

“Overall Satisfaction 1” for customer type = 1
“Overall Satisfaction 2” for customer type = 2
2) Click Stat → Nonparametrics → Mann-Whitney.

3) A new window named “Mann-Whitney” pops up.
4) Select “Overall Satisfaction 1” as the “First Sample.”
5) Select “Overall Satisfaction 2” as the “Second Sample.”
6) Click “OK.”
7) The Mann-Whitney test results appear in the session
window.
707

708
• The p-value of the test is lower than alpha level (0.05);
so we reject the null hypothesis and conclude that there
is a statistically significant difference between the overall
satisfaction medians of the two customer types.

709
710

3.4.2 Kruskal-Wallis
Kruskal-Wallis One-Way Analysis of Variance
• The Kruskal-Wallis one-way analysis of variance

is a statistical hypothesis test to compare the
medians among more than two groups.
• Null Hypothesis (H0): η1 = η2 = … = ηk
• Alternative Hypothesis (Ha): at least one of the medians is

ηi is the median of population i, and k is the number of
groups of our interest.
• It is an extension of Mann-Whitney test.

• If the distributions of k populations are not identically
shaped or there are outliers in the distribution,
Mood’s median test is a more robust than Kruskal-
Wallis.
711
Kruskal-Wallis One-Way Analysis of Variance: Assumptions

• The data of k populations are continuous or ordinal

when the spacing between adjacent values is not

constant.
• The k populations are independent to each other.
• The Kruskal-Wallis test is robust for the non-

712
How Kruskal-Wallis One-Way ANOVA Works
• Step 1: Group the k samples from k populations

(sample i is from population i) into one single data
set and then sort the data in ascending order ranked
from 1 to N, where N is the total number of
observations across k groups.

• Step 2: Add up the ranks for all the observations
from sample i and call it ri, where i can be any
integer between 1 and k.
713
• Step 3: Calculate the test statistic

k
 i
n ( r i  r ) 2
T  ( N  1) i 1
k ni
 ij
( r  r
i 1 j 1
) 2

where
ni is the sample size of sample i.
N is the total number of all the observations across k groups.
rij is the rank (among all the observations) of observation j from
group i.
ni
r
j 1
ij
1
ri  r  ( N  1)
ni 2
714
• Step 4: Make a decision of whether to reject the null

hypothesis.
• Null Hypothesis (H0): η1 = η2 =…= ηk

• The test statistic follows chi-square distribution
when the null hypothesis is true.
 2
• If T is greater than k 1 , we reject the null and claim there is
at least one median statistically different from other
medians.
• If T is smaller than  k 1, we fail to reject the null and claim
2
the medians of k groups are equal.
715
Use Minitab to Run a Kruskal-Wallis One-Way ANOVA

customer satisfaction among three types of
free) hypothesis test: Kruskal-Wallis one-way
ANOVA.
• Data File: “Kruskal-Wallis” tab in “Sample Data.xlsx”

• Null Hypothesis (H0): η1 = η2 = η3
• Alternative Hypothesis (Ha): at least one of the customer
types has different overall satisfaction levels from the others.
716
• Steps to run a Kruskal-Wallis One-Way ANOVA in

Minitab
1) Click Stat → Nonparametrics → Kruskal-Wallis.
2) A new window named “Kruskal-Wallis” pops up.

3) Select “Overall Satisfaction” as the “Response.”
4) Select “Customer Type” as the “Factor.”
5) Click “OK.”
6) The Kruskal-Wallis test results appear in the session

window.
717

718
• The p-value of the test is lower than alpha level (0.05), and we
reject the null hypothesis and conclude that at least the overall
satisfaction median of one customer type is statistically
different from the others.

719
720

3.4.3 Mood’s Median
What is Mood’s Median Test?

• Mood’s median test is a statistical test to compare
the medians of two or more populations.
• Null Hypothesis (H0): η1 = … = ηk
• k is the number of groups of our interest and is equal to or

greater than two.
• Mood’s median is an alternative to Kruskal-Wallis.
• It is the extension of one sample sign test.
• For the data with outliers, Mood’s median test is

more robust than the Kruskal-Wallis test.
721
Mood’s Median Test Assumptions

• The sample data drawn from the populations of interest
are unbiased and representative.
• The data of k populations are continuous or ordinal when

the spacing between adjacent values is not constant.

• The k populations are independent to each other.
• The distributions of k populations have the same shape.
• Mood’s median test is robust for non-normally distributed

populations.
• Mood’s median test is robust for data with outliers.

722
How Mood’s Median Test Works

• Step 1: Group the k samples from k populations
(sample i is from population i) into one single data
set and get the median of this combined data set.
• Step 2: Separate the data in each sample into two

groups. One consists of all the observations with
values higher than the grand median and the other
consists of all the observations with values lower
than the grand median.
723
How Mood’s Median Test Works

• Step 3: Run a Pearson’s chi-square test to
determine whether to reject the null hypothesis.
• Null Hypothesis (H0): η1 = … = ηk

 2
 2
• If calc is greater than crit , we reject the null
hypothesis and claim that at least one median is
 2
 2
• If calc is smaller than crit , we fail to reject the null
hypothesis and claim that the medians of k
populations are not statistically different.
724
Use Minitab to Run a Mood’s Median Test

customer satisfaction among three types of
free) hypothesis test: Mood’s median test.
• Data File: “Median Test” tab in “Sample Data.xlsx”

• Alternative Hypothesis (Ha): at least one customer
type has different overall satisfaction from the
others.
725

• Steps to run a Mood’s median test in Minitab
1) Click Stat → Nonparametrics → Mood’s Median Test.
2) A new window named “Mood’s Median Test” pops up.

3) Select “Overall Satisfaction” as the “Response.”
4) Select “Customer Type” as the “Factor.”
5) Click “OK.”
6) The Mood’s median test results appear in the session

window.
726

727

• The p-value of the test is lower than alpha level (0.05), and we
reject the null hypothesis and conclude that at least the overall
satisfaction median of one customer type is statistically

728
729

3.4.4 Friedman
What is the Friedman Test?
• The Friedman test is a hypothesis test used to

detect the differences in various groups across
multiple attempts.
• Null Hypothesis (H0): The treatments have identical effects
(i.e., η1 = … = ηk).
• Alternative Hypothesis (Ha): At least one of the treatments

has different effects from the others (i.e., at least one of the
medians is statistically different from others).
• It is used as an alternative of the parametric

repeated measures ANOVA when the assumption of
normality or variance equality is not met.
730
Friedman Test Assumptions
• The data are continuous or ordinal when the

spacing between adjacent values is not constant.

• Results in one block are independent of the results
in another.
• The Friedman test is robust for the non-normally

distributed population.
• The Friedman test is robust for populations with

unequal variances.
731
How Friedman Test Works
• Step 1: Organize the data into a tabular view with n

rows indicating the blocks and k columns indicating
the treatments. Each observation xij is filled into the
intersection of a specific block i and a specific
treatment j.

• Step 2: Calculate the ranks of each observation
within each block.
• Step 3: Replace the values in the table created in

step 1 with the order rij (within a block) of the
corresponding observation xij.
732
• Step 4: Calculate the test statistic

k
n (r j  r ) 2
j 1
Q n k
1
 ij
n(k  1) i 1 j 1
( r  r ) 2

where
1 n
r j   rij
n i 1
1 n k
r 
nk i 1 j 1
rij
733
• Step 5: Make a decision of whether to reject the null

hypothesis.
• When n > 15 or k > 4, the test statistic Q follows a chi-

square distribution if the null hypothesis is true and the p-

value is P  k21  Q

• When n < 15 and k < 4, the test statistic Q does not
approximate a chi-square distribution and the p-value can
be obtained from tables of Q for the Friedman test.
• If p-value > alpha level (0.05), we fail to reject the null.
734
Friedman Test Examples
• A number of n water testers judge the quality of k
different water samples, each of which is from a
distinct water source. We will apply a Friedman Test
to determine whether the water qualities of the k
sources are the same.

• There are n blocks and k treatments in this
experiment.
• One tester’s decision would not have any influence

on other testers.
• When running the experiment, each tester judges

the water in a random sequence.
735
Use Minitab to Run a Friedman Test
• Case study: We are interested in comparing the

effect of a drug treatment on enzyme activity. Three
different drug therapies were given to four animals
and each animal belongs to a different litter.
• Data File: “EXH_STAT_MTW,” which is one of the sample

data files that comes with the Minitab installation package

• Alternative Hypothesis (Ha): at least one treatment
effect is statistically different from the others.
736
• Steps to run a Friedman test in Minitab
1) Click Stat → Nonparametrics → Friedman.
2) A new window named “Friedman” pops up.

3) Select “EnzymeActivity” as the “Response.”
4) Select “Therapy” as the “Treatment.”
5) Select “Litter” as the “Blocks.”
6) Click “OK.”
7) The Friedman test results appear in the session window.

737

738
• The p-value of the Friedman test is greater than alpha
level (0.05). We fail to reject the null hypothesis and
conclude that there is not any statistically significant
difference between the treatment effects.

739
740

3.4.5 One Sample Sign
What is the One Sample Sign Test?
• The one sample sign test is a hypothesis test to

compare the median of a population with a specified
value.
• Null Hypothesis (H0): η = η0
• Alternative Hypothesis (Ha): η ≠ η0

• It is an alternative test of one sample t-test when the
distribution of the data is non-normal. It is robust for
the data with non-symmetric distribution.
741
One Sample Sign Test Assumptions


• The one sample sign test is robust for the non-
• The one sample sign test does not have any

assumptions on the distribution. It is a distribution-
free test.
742
How the One Sample Sign Test Works
• Step 1: Separate the sample set of data into two

groups: one with values greater than and the other
with values less than the hypothesized median η0.
Count the number of observations in each group.

• Step 2: Calculate the test statistic.
• If the null hypothesis is true, the number of
observations in each group should not be significantly
different from half of the total sample size.
• The test statistic follows a binomial distribution when
the null is true. When n is large, we use the normal
distribution to approximate the binomial distribution.
743
How the One Sample Sign Test Works
Test Statistic:
n  np
Z calc 
np(1  p)
where
n+ is the number of observations with values greater than the hypothesized median.

n is the total number of observations .
p is 0.5.

hypothesis. If the |Zcalc| is smaller than Zcrit, we fail to
reject the null hypothesis and claim that there is no
significant difference between the population
median and the hypothesized median.
744
Use Minitab to Run a One Sample Sign Test

overall satisfaction of customer type 1 against a
specified benchmark satisfaction (3.5) using a
nonparametric (i.e., distribution-free) hypothesis
test: one sample Wilcoxon test.
• Data File: “One Sample Wilcoxon” tab in “Sample Data.xlsx”

• Null Hypothesis (H0): η1 = 3.5
• Alternative Hypothesis (Ha): η1 ≠ 3.5
745
• Steps to run a one sample sign test in Minitab
1) Click Stat → Nonparametrics → 1-Sample Sign.
2) A new window named “1-Sample Sign” pops up.

3) Select “Overall Satisfaction” as the “Variables.”
4) Click the box of “Test median.”
5) Enter “3.5” in the box next to “Test median.”
6) Click “OK.”
7) The one sample sign test results appear in session

window.
746

747
• The p-value of the one sample sign test is 1.0000, higher than
the alpha level (0.05), and we fail to reject the null hypothesis.
There is not any statistically significant difference between the
overall satisfaction of customer type 1 and the benchmark
satisfaction level.

748
3.4.6 One Sample Wilcoxon
749
What is the One Sample Wilcoxon Test?
• The one sample Wilcoxon test is a hypothesis test

to compare the median of one population with a
specified value.
• Null Hypothesis (H0): η = η0
• Alternative Hypothesis (Ha): η ≠ η0

• It is an alternative test of one sample t-test when the
distribution of the data is non-normal.
• It is more powerful than one sample sign test but it

assumes the distribution of the data is symmetric.
750
One Sample Wilcoxon Test Assumptions



• The distribution of the data is symmetric about a
median.
• The one sample Wilcoxon test is robust for the non-

751
How the One Sample Wilcoxon Test Works
• Step 1: Create the following columns one by one:
• Column 1: all the raw observations (X)
• Column 2: the differences between each observation value

and the hypothesized median (X – η0)

• Column 3: the signs (+ or –) of column 2
• Column 4: the absolute value of column 2
• Column 5: the ranks of each item in column 4 in ascending

order
• Column 6: the product of column 3 and column 5

752
How the One Sample Wilcoxon Test Works
• Step 2: Calculate the test statistic Wcalc , which is the

sum of all the non-negative values in column 6.

hypothesis. Use the table of critical values for the

Wilcoxon test to get the Wcrit with predetermined
alpha level and number of observations.
• If the Wcalc is smaller than the Wcrit, we fail to reject the null
hypothesis and claim that there is no significant difference
between the population median and the hypothesized
median.
753
Use Minitab to Run a One Sample Wilcoxon Test

overall satisfaction of customer type 1 against a
specified benchmark satisfaction (3.5) using a
nonparametric (i.e., distribution-free) hypothesis
test: one sample Wilcoxon test.
• Data File: “One Sample Wilcoxon” tab in “Sample Data.xlsx”

• Null Hypothesis (H0): η1 = 3.5
• Alternative Hypothesis (Ha): η1 ≠ 3.5
754
• Steps to run a one sample Wilcoxon test in Minitab
1) Click Stat → Nonparametrics → 1-Sample Wilcoxon.
2) A new window named “1-Sample Wilcoxon” pops up.

3) Select “Overall Satisfaction” as the “Variables.”
4) Click the radio button “Test median.”
5) Enter “3.5” in the box next to “Test median.”
6) Click “OK.”
7) The one sample Wilcoxon test results appear in the

session window.
755

756
• The p-value of the one sample Wilcoxon test is

0.557, higher than the alpha level (0.05), and we fail
to reject the null hypothesis. There is not any
statistically significant difference between the overall
satisfaction of customer type 1 and the benchmark
satisfaction level.

757
3.4.7 One & Two Sample Proportion
758
What is the One Sample Proportion Test?
• One sample proportion test is a hypothesis test to

compare the proportion of one certain outcome
occurring in a population following the binomial
distribution with a specified proportion.
• Null Hypothesis (H0): p = p0

• Alternative Hypothesis (Ha): p ≠ p0
759
One Sample Proportion Test Assumptions

• There are only two possible outcomes in each trial:

success/failure, yes/no, and defective/non-defective

etc.
• The underlying distribution of the population is

binomial distribution.
• When np ≥ 5 and np(1 – p) ≥ 5, the binomial

distribution can be approximated by the normal
distribution.
760
How the One Sample Proportion Test Works
When np ≥ 5 and np(1 – p) ≥ 5, we use normal

distribution to approximate the underlying binomial
distribution of the population.
pˆ  p0
Test Statistic: Z calc 
p0 (1  p0 )

n
where
p̂ is the observed probability of one certain outcome occurring.
p0 is the hypothesized probability.
n is number of trials.
When |Zcalc| is smaller than Zcrit, we fail to reject the null hypothesis and claim that
there is no statistically significant difference between the population proportion and
the hypothesized proportion.
761
Use Minitab to Run a One Sample Proportion Test

exam pass rate of a high school this month against a
specified rate (70%) using a nonparametric (i.e.,
distribution-free) hypothesis test: one sample
proportion test.
• Data File: “One Sample Proportion” tab in “Sample Data.xlsx”

• Null Hypothesis (H0): p = 70%
• Alternative Hypothesis (Ha): p ≠ 70%
762
• Steps to run a one sample proportion test in Minitab.
1) Click Stat → Basic Statistics → 1 Proportion.

2) A new window named “1 Proportion (Test and Confidence
Interval)” appears.
3) Click on the radio button of “Summarized data.”

4) Enter “77” in the box of “Number of events.”
5) Enter “105” in the box of “Number of trials.”
6) Check the box “Perform hypothesis test.”
7) Enter “0.70” as the “Hypothesized proportion.”
8) Click “OK.”
9) The one sample proportion test results appear in the
session window.
763

764
• The p-value of the one sample proportion test is 0.460,

greater than the alpha level (0.05), and we fail to reject
the null hypothesis.
• We conclude that the exam pass rate of the high school
this month is not statistically different from 70%.

765
What is the Two Sample Proportion Test?
• The two sample proportion test is a hypothesis

test to compare the proportions of one certain event
occurring in two populations following the binomial
distribution.

• Null Hypothesis (H0): p1 = p2
• Alternative Hypothesis (Ha): p1 ≠ p2
766
Two Sample Proportion Test Assumptions

• There are only two possible outcomes in each trial

for both populations: success/failure, yes/no, and

defective/non-defective etc.
• The underlying distributions of both populations are


distribution.
767
How the Two Sample Proportion Test Works
• When np ≥ 5 and np(1 – p) ≥ 5, we use normal

distribution to approximate the underlying binomial
distributions of the populations.
pˆ1  pˆ 2
Test Statistic: Z 
1 1

pˆ 0 1  pˆ 0   
 n1 n2 
x1  x2
where pˆ 0 
n1  n2
p̂1 and p̂2 are the observed proportions of events in the two samples.
n1 and n2 are the number of trials in the two samples respectively.
x1 and x2 are the number of events in the two samples respectively.
When |Zcalc| is smaller than Zcrit, we fail to reject the null hypothesis.
768
Use Minitab to Run a Two Sample Proportion Test
• Case study: We are interested in comparing the exam

pass rates of a high school in March and April using a
nonparametric (i.e., distribution-free) hypothesis test:
two sample proportion test.
• Data File: “Two Sample Proportion” tab in “Sample Data.xlsx”

• Null Hypothesis (H0): pMarch = pApril
• Alternative Hypothesis (Ha): pMarch ≠ pApril
769
• Steps to run a two sample proportion test in Minitab
1) Click Stat → Basic Statistics → 2 Proportions.

2) A new window named “2 Proportions (Test and
Confidence Interval)” pops up.
3) Click the radio button of “Summarized data.”

4) Enter “89” in the box intersecting “First” and “Events.”
5) Enter “112” in the box intersecting “First” and “Trials.”
6) Enter “102” in the box intersecting “Second” and “Events.”
7) Enter “130” in the box intersecting “Second” and “Trials.”
8) Click “OK.”
9) The two sample proportion test results appear in the
session window.
770

771
• The p-value of the two sample proportion test is 0.849,

greater than the alpha level (0.05), and we fail to reject
the null hypothesis.
• We conclude that the exam pass rates of the high school
in March and April are not statistically different.

772
3.4.8 Chi-Squared (Contingency Tables)
773
What is the Chi-Square Test?
• A chi-square test is a hypothesis test in which the

sampling distribution of the test statistic follows a
chi-square distribution when the null hypothesis is
true.

• There are multiple chi-square tests available and in
this module we will cover the Pearson’s chi-square
test used in contingency analysis.
• Null Hypothesis (H0): p1 = p2 =… = pk

• Alternative Hypothesis (Ha): at least on of the proportions is
• k is the number of populations of our interest. k ≥ 2.
774
What is the Chi-Square Test?
• The chi-square test can also be used to test whether

two factors are independent of each other. In other
words, it can be used to test whether there is any
statistically significant relationship between two
discrete factors.

• Null Hypothesis (H0): Factor 1 is independent of factor 2.
• Alternative Hypothesis (Ha): Factor 1 is not independent of
factor 2.
775
Chi-Square Test Assumptions

• There are only two possible outcomes in each trial

for an individual population: success/failure, yes/no,

and defective/non-defective etc.
• The underlying distribution of each population is


distribution.
776
How Chi-Square Test Works
• Test Statistic
n
(Oi  Ei ) 2
 calc
2

i 1 Ei
where

Oi is an observed frequency.
Ei is an expected frequency.
N is the number of cells in the contingency table.
If  calc is smaller than  crit , we fail to reject the null hypothesis.

2 2
777
Use Minitab to Run a Chi-Square Test
• Case study 1: We are interested in comparing the
product quality exam pass rates of three suppliers A, B,
and C using a nonparametric (i.e., distribution-free)
hypothesis test: chi-square test.
• Data File: “Chi-Square Test1” tab in “Sample Data.xlsx”

• Null Hypothesis (H0): pA = pB = pC
• Alternative Hypothesis (Ha): at least one of the suppliers
has different pass rates from the others.
778
• Steps to run a chi-square test in Minitab
1) Click Stat → Tables → Cross Tabulation and Chi-Square.
2) A new window named “Cross Tabulation and Chi-Square”
pops up.
3) Select “Results” as “For rows.”
4) Select “Supplier” as “For columns.”
5) Select “Count” as “Frequencies are in.”

6) Click the “Chi-Square” button.
7) A new window named “Cross Tabulation – Chi-Square” pops
up.
8) Check the boxes of “Chi-square analysis”, “Expected cell
counts,” and “Each cell’s contribution to the Chi-Square
statistic.”
9) Click “OK” in the window named “Cross Tabulation – Chi-
Square.”
10) Click “OK” in the window named “Cross Tabulation and Chi-
Square.”
11) The Chi-square test results appear in the session window.
779

780
• Counts are based on the

sample observation.
• Expected counts are

based on the assumption

that the null hypothesis is
true.
• Since the p-value is

smaller than alpha level
(0.05), we reject the null
hypothesis and claim that
at least one supplier has
different pass rate from
others.
781
• Case study 2: We are trying to check whether there is a
relationship between the suppliers and the results of the
product quality exam using nonparametric (i.e.,
distribution-free) hypothesis test: chi-square test.
• Data File: “Chi-Square Test2” tab in “Sample Data.xlsx”

• Null Hypothesis (H0): product quality exam results are
independent of the suppliers.
• Alternative Hypothesis (Ha): product quality exam results
depend on the suppliers.
782
• Steps to run a chi-square test in Minitab
1) Click Stat → Tables → Cross Tabulation and Chi-Square.
2) A new window named “Cross Tabulation and Chi-Square”
pops up.
3) Select “Results” as “For rows.”
4) Select “Supplier” as “For columns.”

5) Select “Count” as “Frequencies are in.”
6) Click the “Chi-Square” button.
7) Check the boxes of “Chi-square analysis”, “Expected cell
counts,” and “Each cell’s contribution to the Chi-Square
statistic.”
8) Click “OK” in the window named “Cross Tabulation – Chi-
Square.”
9) Click “OK” in the window named “Cross Tabulation and
Chi-Square.”
10) The Chi-square test results appear in the session window.
783

784
The p-value is smaller

than the alpha level
(0.05) and we reject the

null hypothesis.
The product quality

exam results are not
independent of the
suppliers.
785
3.4.9 Tests of Equal Variance
786
What are Tests of Equal Variance?
• Tests of equal variance are a family of hypothesis

tests used to check whether there is a statistically
significant difference between the variances of two
or more populations.
• Null Hypothesis (H0): s 1  s 2  ...  s k

2 2 2

• Alternative Hypothesis (Ha): at least the variance of one
population is different from others.
• k is the number of populations of interest. k ≥ 2.
• A tests of equal variance can be used alone but

most of the time it is used with other statistical
methods to verify or support the assumption about
the variance equality.
787
F-Test
• The F-test is used to compare the variances

between two normally distributed populations.
• It is extremely sensitive to non-normality and serves

as a preliminary step for two sample t-test.

• Test Statistic:
2
s1
Fcalc  2 , where s1 and s2 are the sample standard deviations.
s2
The sampling distribution of the test statistic follows F distribution
when the null is true.
788
Bartlett’s Test

• Bartlett’s test is used to compare the variances
among two or more normally distributed populations.
• It is sensitive to non-normality and it serves as a
preliminary step for ANOVA.

• Test Statistic:
k
( N  k ) ln( S )   (ni  1) ln( Si2 )
2
p k
2  N   ni and S p2 
i 1 1
1  k  1  1 
, where 
N k i
(ni  1) Si2
1  
3(k  1)  i 1  ni  1  N  k 
i 1
The sampling distribution of test statistic follows chi2 distribution when the null is true.
789
Brown-Forsythe Test
• The Brown-Forsythe test is used to compare the

variances between two or more populations with any
distributions.
• It is not so sensitive to non-normality as Bartlett’s

test.
• The Test statistic is the model F statistic from the

ANOVA on the transformed response zij  yij  ~yi
where ~yi is the median response at ith level.
790
Levene’s Test

• Levene’s test is used to compare the variances
between two or more populations with any
distributions.
• It is not so sensitive to non-normality as Bartlett’s

test.
• The test statistic is the model F statistic from the

ANOVA on the transformed response zij  yij  yi
where yi is the mean response at ith level.
791
Brown-Forsythe Test vs. Levene’s Test

k
(N  k)  i i. ..
N ( Z  Z ) 2
F i 1
(k  1) k Ni
 ij i.
( Z  Z
i 1 j 1
) 2
where

N is the total number of observations.
Ni is the number of observations in the ith group.
Zi. is the group mean of the ith group.
Z.. is the grand mean of all the observations.
~ ~
In Brown-Forsythe Test, Z ij  Yij  Yij , where Yij is the group median of the ith group.
In Levene’s Test, Z ij  Yij  Yij , where Yij is the group mean of the ith group.
792
Use Minitab to Run Tests of Equal Variance
variances of the retail price of a product in state A and
state B.
• Data File: “Two-Sample T-Test” tab in “Sample Data.xlsx”

• Null Hypothesis (H0): s A  s B
2 2
• Alternative Hypothesis (Ha): s A2  s B2
793
• Step 1: Run the normality test to check whether all levels
of data are normally distributed

3) Select “Retail Price” as the “Variables.”
4) Click in the blank box right below “By variables” and the
“State” appears in the list box on the left.
5) Select “State” as “By variables.”
6) Click “OK.”
7) The normality test results appear in the new windows.

794

795
• Null Hypothesis (H0): The

data are normally distributed.
• Alternative Hypothesis (Ha):
The data are not normally
distributed.

Both retail price data of state A
and B are normally distributed
since the p-values are both
greater than alpha level (0.05).
796
• Step 2: Run tests of equal variance in Minitab
1) Click Stat → ANOVA → Test for Equal Variances.
2) A new window named “Test for Equal Variances” pops up.

3) Select “Retail Price” as the “Response.”
4) Select “State” as “Factors.”
5) Click “OK.”
6) The results of variances equality tests appear in the new

window.
797

798
If all the groups of data are

normally distributed, use the F-
test or Bartlett’s test in Minitab
to test the equality of the
variances.
If at least one of the groups is

not normally distributed, use
Levene’s test in Minitab to test
the equality of the variances.
If the p-value of the variances equality test is greater than the alpha
level (0.05), we fail to reject the null hypothesis and conclude that the
variances of different groups are identical.
799
Hypothesis Testing Roadmap: Putting it all together

800
801

4.0 Improve Phase
Green Belt Training: Improve Phase
4.1 Simple Linear Regression
4.1.1 Correlation
4.1.2 X-Y Diagram
4.1.3 Regression Equations
4.1.4 Residuals Analysis
4.2 Multiple Regression Analysis
4.2.1 Non-Linear Regression

4.2.2 Multiple Linear Regression
4.2.3 Confidence Intervals
4.2.5 Data Transformation, Box Cox
4.2.6 Stepwise Regression
4.2.7 Logistic Regression
802
803
4.1.1 Correlation
4.1.2 X-Y Diagram

804
805

4.1.1 Correlation
What is Correlation?
• Correlation is a statistical technique that describes
whether and how strongly two or more variables are
related.
• Correlation analysis helps to understand the direction

and degree of association between variables, and it

suggests whether one variable can be used to predict
another.
• Of the different metrics to measure correlation, Pearson’s

correlation coefficient is the most popular. It measures
the linear relationship between two variables.
806
Pearson’s Correlation Coefficient

• Pearson’s correlation coefficient is also called:
• Pearson’s r or coefficient of correlation
• Pearson’s product moment correlation coefficient (r)
• “r” is a statistic measuring the linear relationship
between two variables.

• Correlation coefficients range from -1 to 1.
• If r = 0, there is no linear relationship between the variables.
• The sign of r indicates the direction of the relationship:
• If r < 0, there is a negative linear correlation.
• If r > 0, there is a positive linear correlation.
• The absolute value of r describes the strength of the
relationship:
• If |r| ≤ 0.5, there is a weak linear correlation.
• If |r| > 0.5, there is a strong linear correlation.
• If |r| = 1, there is a perfect linear correlation.
807

• When the correlation is strong, the data points on a
scatter plot will be close together (tight).
• The closer “r” is to -1 or 1, the stronger the relationship.
• -1 Strong inverse relationship

• +1 Strong direct relationship
• When the correlation is weak, the data points are

spread apart more (loose).
• The closer the correlation is to 0 the weaker the relationship.
808

• Pearson’s correlation coefficient is only sensitive to the
linear dependence between two variables.
• It is possible that two variables have a perfect non-linear

relationship when the correlation coefficient is low.

• Notice the scatter plots below with correlation equal to 0.
There are clearly relationships but they are not linear and
therefore can not be determined with Pearson’s
correlation coefficient.
809
Correlation and Causation
• Correlation does not imply causation.
• If variable A is highly correlated with variable B, it

does not necessarily mean A causes B or vice
versa. It is possible that an unknown third variable C

is causing both A and B to change.
• For example, if ice cream sales at the beach are

highly correlated with the number of shark attacks, it
does not imply that increased ice cream sales
causes increased shark attacks. They are triggered
by a third factor: summer.
810
Correlation and Dependence
• If two variables are independent, the correlation

coefficient is zero.
• WARNING! If the correlation coefficient of two

variables is zero, it does not imply they are

independent.
• The correlation coefficient only indicates the linear

dependence between two variables. When variables
are non-linearly related, they are not independent of
each other but their correlation coefficient could be
zero.
811
Correlation Coefficient and X-Y Diagram
• The correlation coefficient indicates the direction and
strength of the linear dependence between two variables
but it does not cover all the existing relationship patterns.
• With the same correlation coefficient, two variables might

have completely different dependence patterns.

• A scatter plot or X-Y diagram can help to discover and
understand additional characteristics of the relationship
between variables.
• Correlation coefficient is not a replacement for examining

the scatter plot to study the variables’ relationship.
812
Statistical Significance of the Correlation Coefficient
• The correlation coefficient could be high or low by

chance (randomness). It may have been calculated
based on two small samples that do not provide
good inference on the correlation between two
populations.

• In order to test whether there is a statistically
significant relationship between two variables, we
need to run a hypothesis test to determine whether
the correlation coefficient is statistically different
from zero.
• Hypothesis Test Statements
• H0: r = 0: Null Hypothesis: There is no correlation.
• H1: r ≠ 0: Alternate Hypothesis: There is a correlation.
813
• Hypothesis tests will produce p-values as a result of

the statistical significance test on r.
• When the p-value for a test is low (less than 0.05), we can
reject the null hypothesis and conclude that “r” is significant;
there is a correlation.

• When the p-value for a test is > 0.05, then we fail to reject
the null hypothesis; there is no correlation.
• We can also use the t statistic to draw the same

conclusions regarding our test for significance of the
correlation coefficient.
814
r
• Test Statistic: t 
1 r 2
n2
• Critical Statistic: t-value in t-table with (n – 2) degrees of

freedom
• If |t| ≤ tcritical, we fail to reject the null. There is no

statistically significant linear relationship between X and
Y.
• If |t| > tcritical, we reject the null. There is a statistically

significant linear relationship between X and Y.
815
Using Software to Calculate the Correlation Coefficient
• We are interested in understanding whether there is

linear dependence between a car’s MPG and its
weight and if so, how they are related.
• The MPG and weight data are stored in the

“Correlation Coefficient” tab in “Sample Data.xlsx.”
We will discuss three ways to get the results.
816
Use Excel to Calculate the Correlation Coefficient
• The formula CORREL in Excel calculates the sample

correlation coefficient of two data series.
• The correlation coefficient between the two data series is -0.83,
which indicates a strong negative linear relationship between
MPG and weight.

817
Use Minitab to Calculate the Correlation Coefficient
• Step 1: Stat → Basic Statistics → Correlation.

• Step 2: Select the two variables of interest in the pop-up window
“Correlation” and click “OK.”
• The correlation coefficient result (-0.832) appears in the session
window. The p-value (0.000) is lower than the alpha level (0.05),
indicating the linear correlation is statistically significant.
818
Interpreting Results
• How do we interpret results and make decisions based
Pearson’s correlation coefficient (r) and p-values?
• Let us look at a few examples:
• r = -0.832, p = 0.000 (previous example). The two variables are
inversely related and the linear relationship is strong. Also, this
conclusion is significant as supported by p-value of 0.00.
• r = -0.832, p = 0.71. Based on r, you should conclude the linear

relationship between the two variables is strong and inversely
related. However, with a p-value of 0.71, you should then
conclude that r is not significant and that your sample size may
be too small to accurately characterize the relationship.
• r = 0.5, p = 0.00. Moderately positive linear relationship, r is
statistically significant.
• r = 0.92, p = 0.61. Strong positive linear relationship but r is not
statistically significant. Get more data.
• r = 1.0, p = 0.00. The two variables have a perfect linear
relationship and r is significant.
819
Correlation Coefficient Calculation
• Population Correlation Coefficient (ρ)

• Sample Correlation Coefficient (r)
• It is only defined when the standard deviations of both X

and Y are non-zero and finite.
• When covariance of X and Y is zero, the correlation
coefficient is zero.
820
821

4.1.2 X-Y Diagram
What is an X-Y Diagram?
• An X-Y diagram is a scatter plot depicting the
relationship between two variables (i.e., X and Y).
• Each point on the X-Y diagram represents a pair of

X and Y values, with X plotted on the horizontal axis
and Y plotted on the vertical axis.

• With an X-Y diagram, you can qualitatively assess
both the strength and direction of the relationship
between X and Y.
• To quantitatively measure the relationship between

X and Y, you may need to calculate the correlation
coefficient.
822
Example 1: Perfect Linear Correlation
• In the chart below, there are 10 data points depicted (10 pairs of X
and Y values), and they were created using the equation Y = 2X.
• As a result, all the data points fall onto the straight line of Y = 2X.
• The chart demonstrates a perfect positive linear correlation
between X and Y since the relationship between X and Y can be
perfectly described by a linear equation in a format of Y = a × X +
b where a ≠ 0.

X Y
1 2
2 4
3 6
4 8
5 10
6 12
7 14
8 16
9 18
10 20
823
Example 2: Strong Linear Correlation
• In this chart, the data points scatter closely around a straight
line.
• When X increases, Y increases accordingly.
• This chart demonstrates a strong positive linear correlation
between X and Y.
• The straight line is the trend line showing how Y’s trend goes
with changes in X.

X Y
1 2
2 5
3 6
4 7
5 10
6 13
7 15
8 16
9 19
10 22
824
Example 3: Weak Linear Correlation
• In this chart, the data points scatter remotely around a straight
line.
• When X increases, Y increases accordingly.
• This chart demonstrates a weak positive linear correlation
between X and Y since the distance between the data points and
the trend line is relatively far on average.

X Y
1 3
2 11
3 7
4 1
5 10
6 2
7 15
8 3
9 11
10 10
825
Example 4: Non-Linear Correlation
• The X-Y diagram also helps to identify any nonlinear

relationship between X and Y.
X Y
1 2.84
2 2.91
3 2.14
4 1.24
5 1.04

6 1.72
7 2.66
8 2.99
9 2.41
10 1.46
X Y
1 35
2 26
3 19
4 14
5 11
6 10
7 11
8 14
9 19
10 26
826
Example 5: Uncorrelated
• In this chart, the Y value of each data point is a constant
regardless of what the X value is.
• Changes in X do not show any relative impact on Y. As a result,

there is no correlation between X and Y.

X Y
1 5
2 5
3 5
4 5
5 5
6 5
7 5
8 5
9 5
10 5
827
Example 6: Outlier Identification
• Using X-Y diagram, you may identify outliers in the data.

• In this chart, the last data point does not seem to follow the trend
of other data points.
• This should require further investigation on the last data point to
determine whether it is an outlier.

X Y
1 2
2 5
3 6
4 7
5 10
6 13
7 15
8 16
9 19
10 58
828
Benefits of Using an X-Y Diagram
• An X-Y diagram graphically demonstrates the

relationship between two variables.
• It suggests whether two variables are associated

and helps to identify the linear or nonlinear

correlation between X and Y.
• It captures the strength and direction of the

relationship between X and Y.
• It helps identify any outliers in the data.
829
Limitations of the X-Y Diagram
• Although the X-Y diagram helps to “spot” interesting

features in the data, it does not provide any
quantitative conclusions about the data and further
statistical analysis is needed to:
• Assess whether the association between variables is

• Measure the strength of the relationship between

variables.
• Determine whether outliers exist in the data.
• Quantitatively describe the pattern of the data.

830
831
Correlation and Regression Analysis
• The correlation coefficient answers the following questions:
• Are two variables correlated?
• How strong is the relationship between two variables?
• When one variable increases, does the other variable
increase or decrease?
• The correlation coefficient cannot address the following

questions:
• How much does one variable changes when the other
variable changes by one unit?
• How can we set the value of one variable to obtain a
targeted value of the other variable?
• How can we use the relationship between two variables to
make predictions?
• The simple linear regression analysis helps to answer these
questions.
832
What is Simple Linear Regression?
• Simple linear regression is a statistical technique to fit
a straight line through the data points.
• It models the quantitative relationship between two

variables.

• It describes how one variable changes according to the
change of another variable.
• Both variables need to be continuous.
• It is simple because only one predictor variable is

involved.
833
Simple Linear Regression Equation
• The simple linear regression analysis fits the data to a
regression equation in the form
Y = α × X + β+ e
where:
• Y is the dependent variable (the response) and X is the

single independent variable (the predictor).
• α is the slope describing the steepness of the fitting line. β is

the intercept indicating the Y value when X is equal to 0.
• e stands for error (residual). It is the difference between the

actual Y and the fitted Y (i.e., the vertical difference between
the data point and the fitting line).
834
Ordinary Least Squares
• The ordinary least square is a statistical method used
in linear regression analysis to find the best fitting line for
the data points.
• It estimates the unknown parameters of the regression

equation by minimizing the sum of squared residuals

(i.e., the vertical difference between the data point and
the fitting line).
• In mathematical language, we look for α and β that

satisfy the following criteria:
n n
  where Q( ,  )   ei   ( yi   xi )
2 2
min Q
 ,
( , )
i 1 i 1
835
Ordinary Least Squares
• The actual value of the dependent variable:
Yi   * X i    ei where i = 1, 2,…, n
• The fitted value of the dependant variable:

^
Yi   * X i   where i = 1, 2,…, n

• By using calculus, it can be shown the sum of squared
error is minimal when
n _ _
(X i  X )(Yi  Y ) _ _
 i 1
n _  Y X
 i
( X
i 1
 X ) 2
836
ANOVA in Simple Linear Regression
X: the independent
variable that we use to
predict;
Y: the dependent
variable that we want
to predict.
X Y
1 5
2 5

3 8
4 7
5 10
6 15
7 15
8 16
n _
9 20
Total Variation = Total Sums of Squares =  (Y  Y )
i 1
i
2
10 22
n ^ _
Explained Variation = Regression Sums of Squares =  (Y  Y )
i 1
i
2
n ^
Unexplained Variation = Error Sums of Squares =  (Y  Y )
i 1
i
2
837
• Linear regression is also analysis of variance (ANOVA).
• Variation Components:
• Total Variation = Explained Variation + Unexplained
Variation
i.e., Total Sums of Squares = Regression Sums of Squares +

Error Sums of Squares
• Degrees of Freedom Components

• Total Degrees of Freedom = Regression Degrees of
Freedom + Residual Degrees of Freedom
i.e., n – 1 = (k – 1) + (n – k), where n is the number of data
points, k is the number of predictors
838
• Whether the overall model is statistically significant can be
tested by using F-test of ANOVA.
• H0: The model is not statistically significant.
• Ha: The model is statistically
n
significant.
^ _
MSR  (Yi  Y ) 2 (k  1)
• Test Statistic: F   in1 ^
 (Yi  Yi ) 2 (n  k )
MSE

i 1
• Critical Statistic: F value in F table with (k – 1) degrees of

freedom in the numerator and (n – k) degrees of freedom in the
denominator.
• If F ≤ Fcritical, we fail to reject the null. There is no statistically

significant relationship between X and Y.
• If F > Fcritical, we reject the null. There is a statistically
significant relationship between X and Y.
839
Coefficient of Determination
• R2 (also called coefficient of determination) measures the
proportion of variability in the data that can be explained
by the model.
• R2 ranges from 0 to 1. The higher R2 is, the better the

model can fit the actual data.

• How to calculate R2:
n ^
SS regression SSerror  i
(Y  Y ) 2
R 
2
 1  1 i 1
n _
 i
SStotal SStotal
(Y  Y ) 2
i 1
840
Use Minitab to Run a Simple Linear Regression
• Case study:
• We want to see whether the score on exam one has
any statistically significant relationship with the score
on the final exam. If yes, how much impact does exam
one have on the final exam?
• Data File: “Simple Linear Regression” tab in “Sample

Data.xlsx”
• Step 1: Determine the dependent and independent

variables. Both should be continuous variables.
• Y (dependent variable) is the score of final exam.
• X (independent variable) is the score of exam one.
• Both variables are continuous.
841
• Step 2: Create a scatter plot to visualize whether

there seems to be a linear relationship between X
and Y.

3) Click “OK.”
4) A new window named “Scatterplot - Simple” pops up.
5) Select “FINAL” as “Y variables” and “EXAM1” as “X
variables.”
6) Click “OK.”
7) A scatter plot is generated in a new window.
842

843
• Based on the scatter plot, the relationship between exam

one and final seems linear. The higher the score on
exam one, the higher the score on the final.

844
• Step 3: Run the simple linear regression analysis.
1) Click Stat → Regression → Regression.

2) A new window named “Regression” pops up.
3) Select “FINAL” as “Response” and “EXAM1” as
“Predictions.”

4) Click the “Storage” button.
5) Check the box of “Residuals” so that the residuals can be
saved automatically in the last column of the data table.
6) Click “OK.”
7) The regression analysis results appear in the new window.
845

846
• Step 4: Check whether the model is statistically

significant. If not significant, we will need to re-
examine the predictor or look for new predictors
before continuing.
R2 measures the percentage of variation in
the data set that can be explained by the

model. 89.5% of the variability in the data
can be accounted for by this linear
regression model.
“Analysis of Variance” section provides a

ANOVA table covering degrees of
freedom, sum of squares, and mean
square information for total, regression
and error.
The p-value of the F-test is lower than the

α level (0.05), indicating that the model is
847
• Step 5: Understand regression equation
The estimates of slope and

intercept are shown in
“Parameter Estimate”
section.

In this example,
Y = 15.6 + 1.85 × X.
One unit increase in the

score of Exam1 would
increase the final score by
1.85.
848
Interpreting the Results
• Rsquare Adj = 89.0%

• 89% of the variation in FINAL can be explained by EXAM1
• P-value of the F-test = 0.000

• We have a statistically significant model

• Prediction Equation: 15.6 + 1.85 × EXAM1
• 15.6 is the Y intercept, all equations will start with 15.6
• 1.85 is the EXAM1 Coefficient: multiply it by EXAM1 score
• Let us say you are the professor and you want to

use this prediction equation to estimate what two
of your students might get on their final exam.
849
• Let us assume the following:
• Student “A” exam 1 results were: 79
• Student “B” exam 1 results were: 94.
• Remember our prediction equation?

• 15.6 + 1.85 × Exam1

• Now apply the equation to each student
• Student “A” Estimate: 15.6 + (1.85 × 79) = 161.8
• Student “B” Estimate: 15.6 + (1.85 × 94) = 189.5
Now you can use your “magic” as the professor

and allocate your time appropriately to the
student(s) who you predict need the most help
Nice Work!
850
851
What are Residuals?
• Residuals are the vertical differences between actual
values and the predicted values or the “fitted line”
created by the regression model.

Predicted Value
Vertical
Distance
Actual Point
852
Why Perform Residuals Analysis?
• Regression equations are generated on the basis of
certain statistical assumptions.
• Residuals analysis helps to determine the validity of
these assumptions.
• The assumptions are:
• The residuals are normally distributed, mean equal to zero.

• The residuals are independent.
• The residuals have a constant variance.
• The underlying population relationship is linear.
• If residuals performance does not meet the

requirements, we will need to rebuild the model by
replacing the predictor with a new one, adding new
predictors, building non-linear models, and so on.
853
Use Minitab to Perform Residuals Analysis
• The residuals of the model are saved in the last column of the
data table

854
• Step 1: Check whether residuals are normally

distributed around the mean of zero.

3) Select “RESI1” as “Variables.”
4) Click “OK.”
5) The histogram and the normality test of the residuals are

displayed in the new window.
855

856
• The mean of residuals is -0.0000.
• The Anderson-Darling test is used to test the normality. Since
the p-value (0.339) is greater than the alpha level (0.05), we
fail to reject the null hypothesis; the residuals are normally
distributed.
• H0: The residuals are normally distributed.

• H1: The residuals are not normally distributed.
857
• Step 2: If the data are in time order, run the IR chart
to check whether residuals are independent.
1) Click Stat → Control Charts → Variable Charts for


pops up.
3) Select “Residuals” as “Variables” and check the box of

“Test for special causes.”
4) Click “OK.”
5) The control charts are shown automatically in the new

window.
858

859
• If no data points are out of control in both the I-chart and MR
chart, the residuals are independent of each other.
• If the residuals are not independent, it is possible that some
important predictors are not included in the model.
• In this example, since the IR chart is in control, residuals are
independent.

860
• Step 3: Check whether residuals have equal variance
across the predicted responses.
• Create a column of the fitted response using the regression formula
returned in the session window.
1) Right click on the last column in the data table.
2) Select Formulas → Assign Formula to Column.
3) A new window named “Assign Formula to C4” appears.
4) Enter the regression equation “15.6+1.85*EXAM1” into the “Expression”
box.

5) Click “OK.”
6) The column of the fitted values is created in the data table.
• Create a scatter plot with Y being the residuals and the X being the
fitted values.
2) A new window named “Scatterplots” appears.
3) Select “RESI1” as the “Y variables” and “Fitted” as the “X variables.”
4) Click “OK.”
5) The scatter plot appears in a new window.
• We are looking for the pattern in which residuals spread out evenly
around zero from the top to the bottom.
861

862

863
864
4.1.1 Correlation
4.1.2 X-Y Diagram

865
866
Linear and Non-Linear
• The word linear originally comes from Latin word linearis
meaning “created by lines.”
• A linear function in mathematics follows the following

pattern (i.e., the output is proportional to its input):
f ( x)   * x  

f ( x1 , x2 ,..., xn )  1 * x1   2 * x2  ...   n * xn  
• A non-linear function does not follow the above pattern.

There are usually exponents, logarithms, power,
polynomial components, and other non-linear functions
of the independent variables and parameters.
867
Non-Linear Relationships Using Linear Models
• Many non-linear relationships can be transformed into

linear relationships, and from there we can use linear
regression methods to model the relationship.
• Some non-linear relationships cannot be transformed

to linear ones and we need to apply other methods to
build the non-linear models.
• In this section, we will focus on building non-linear

regression models using linear transformation (i.e.,
transforming the independent or dependent variables
or parameters to generate a linear function).
868
Assumptions in Using Non-Linear Regression
• The population relationship is non-linear based on a

reliable underlying theory.
• Across the range of all the possible values of the

independent variables, the non-linear relationship

applies. It is possible that at some extreme values
the relationship between variables changes
dramatically.
869
Non-Linear Functions: Transforming to Linear
• Examples of non-linear functions that can be

transformed to linear functions:
• Exponential Function
• Inverse Function
• Polynomial Function

• Power Function.
870
Exponential Function
• Exponential Function

• Transformation
871
Inverse Function
• Inverse Function

• Transformation
where
872
Polynomial Function
• Polynomial Function

• Transformation
where
873
Power Function
• Power Function

• Transformation
874
875
What is Multiple Linear Regression?
• Multiple linear regression is a statistical technique

to model the relationship between one dependent
variable and two or more independent variables by
fitting the data set into a linear equation.

• The difference between simple linear regression and
multiple linear regression:
• Simple linear regression only has one predictor.
• Multiple linear regression has two or more predictors.
876
Multiple Linear Regression Equation
Y  1 * X1   2 * X 2  ...   p * X p    e
• Y is the dependent variable (response).
• X1, X2, … , Xp are the independent variables (predictors).
There are p predictors in total.

• Both dependent and independent variables are
continuous.
• β is the intercept indicating the Y value when all the
predictors are zeros.
• α1, α2, … , αp are the coefficients of predictors. They
reflect the contribution of each independent variable in
predicting the dependent variable.
• e is the residual term indicating the difference between
the actual and the fitted response value.
877
Use Minitab to Run a Multiple Linear Regression
• Case study:
• We want to see whether the scores in exam one, two, and
three have any statistically significant relationship with the
score in final exam. If so, how are they related to final exam
score? Can we use the scores in exam one, two, and three
to predict the score in final exam?

• Data File: “Multiple Regression Analysis” tab in “Sample
Data.xlsx.”
• Step 1: Determine the dependent and independent

variables. All should be continuous.
• Y (dependent variable) is the score of final exam.
• X1, X2, and X3 (independent variables) are the scores of
exam one, two, and three respectively.
• All the variables are continuous.
878
• Step 2: Start building the multiple linear regression

model
1) Click Stat → Regression → Regression.
2) A new window named “Regression” pops up.
3) Select “FINAL” as “Response” and “EXAM1”, “EXAM2”
and “EXAM3” as “Predictors.”

4) Click the “Graph” button, select the radio button “Four in
one” and click “OK.”
5) Click the “Options” button, check the box of “Variation
inflation factors” and click “OK.”
6) Click the “Storage” button, check the boxes of “Residuals”
and “DFITS” and click “OK.”
7) Click “OK” in the window named “Regression.”
8) The regression analysis results appear in the session
window and the four residual plots appear in another
window named “Residual Plots for FINAL.”
879

880
• Step 3: Check whether the whole model is

statistically significant. If not, we need to re-examine
the predictors or look for new predictors before
continuing.
H0: The model is not statistically
significant (i.e., all the parameters of

predictors are not significantly different
from zeros).
H1: The model is statistically significant

(i.e., at least one predictor parameter
is significantly different from zero).
In this example, p-value is much

smaller than alpha level (0.05), hence
we reject the null hypothesis; the
model is statistically significant.
881
• Step 4: Check whether multicollinearity exists in the

model.
• The VIF information is automatically generated in table of
parameter estimates.

882
Multicollinearity
• Multicollinearity is the situation when two or more
independent variables in a multiple regression model are
correlated with each other.
• Although multicollinearity does not necessarily reduce

the predictability for the model as a whole, it may mislead

the calculation for individual independent variables.
• To detect multicollinearity, we use VIF (Variance Inflation

Factor) to quantify its severity in the model.
883
Variance Inflation Factor (1)
• VIF quantifies the degree of multicollinearity for each individual

independent variable in the model.
• VIF calculation:
• Assume we are building a multiple linear regression model using p
predictors.
Y  1  X1   2  X 2  ...   p  X p  

• Two steps are needed to calculate VIF for X1.
• Step 1: Build a multiple linear regression model for X1 by using X2, X3, …, Xp as
predictors.
X1  a2  X 2  a3  X 3  ...  a p  X p  b
• Step 2: Use the R2 generated by the linear model in step 1 to calculate the VIF for
X1.
1
VIF 
1 R2
• Apply the same methods to obtain the VIFs for other Xs. The VIF value ranges from
one to positive infinity.
884
Variance Inflation Factor (2)
• Rules of thumb to analyze variance inflation factor (VIF):
• If VIF = 1, there is no multicollinearity.
• If 1 < VIF < 5, there is small multicollinearity.

• If VIF ≥ 5, there is medium multicollinearity.
• If VIF ≥ 10, there is large multicollinearity.
885
How to Deal With Multicollinearity
• Increase the sample size.
• Collect samples with a broader range for some

predictors.

• Remove the variable with high multicollinearity and high
p-value.
• Remove variables that are included more than once.
• Combine correlated variables to create a new one.
• In this section, we will focus on removing variables with

high VIF and high p-value.
886
• Step 5: Deal with multicollinearity:
• Step 5.1: Identify a list of independent variables with VIF

higher than 5. If no variable has VIF higher than 5, go to
Step 6 directly.

• Step 5.2: Among variables identified in Step 5.1, remove the
one with the highest p-value.
• Step 5.3: Run the model again, check the VIFs and repeat
Step 5.1.
• Note: we only remove one independent variable at a time.
887
• In this example, all three predictors have VIF higher than 5.

Among them, EXAM1 has the highest p-value.
• We will remove EXAM1 from the equation and run the model
again.

888
• Run the new multiple linear regression with only two

predictors (i.e., EXAM2 and EXAM3).
• Check the VIFs of EXAM2 AND EXAM3. They are both
smaller than 5; hence, there is little multicollinearity
existing in the model.

889
• Step 6: Identify the statistically insignificant predictors.

Remove one insignificant predictor at a time and run the
model again. Repeat this step until all the predictors in
the model are statistically significant.
• Insignificant predictors are the ones with p-value higher than alpha
level (0.05). When p > alpha level, we fail to reject the null

hypothesis; the predictor is not significant.
• H0: The predictor is not statistically significant.
• H1: The predictor is statistically significant.
• As long as the p-value is greater than 0.05, remove the

insignificant variables one at a time in the order of the
highest p-value.
• Once one insignificant variable is eliminated from the
model, we need to run the model again to obtain new p-
values for other predictors left in the new model.
890
• In this example, both predictors’ p-values are smaller

than alpha level (0.05). As a result, we do not need to
eliminate any variables from the model.

891
• Step 7: Interpret the regression equation

The multiple linear regression
equation appears
automatically at the top of the
session window.

“Parameter Estimates” section
provides the estimates of
parameters in the linear
regression equation.
892
• Rsquare Adj = 98.4%

• 98% of the variation in FINAL can be explained by the
predictor variables EXAM2 & EXAM3.
• P-value of the F-test = 0.000
• We have a statistically significant model.
• Variables p-value:

• Both are significant (less than 0.05).
• VIF
• EXAM2 & EXAM3 are both below 5; we’re in good shape!
• Equation: -4.34 + 0.722*EXAM2 + 1.34*EXAM3
• -4.34 is the Y intercept, all equations will start with -4.34.
• 0.722 is the EXAM2 coefficient; multiply it by EXAM2 score.
• 1.34 is the EXAM3 coefficient; multiply it by EXAM3 score.
893
• Let us say you are the professor again, and this
time you want to use your prediction equation to
estimate what one of your students might get on
their final exam.
• Assume the following:
• Exam 2 results were: 84
• Exam 3 results were: 102.

• Use your equation: -4.34 + 0.722*EXAM2 + 1.34*EXAM3
• Predict your student’s final exam score:

• -4.34 + (0.722*84) + (1.34*102) =-4.34 + 60.648 + 136.68 =
192.988
Nice work again! Now you can use your “magic” as the
smart and efficient professor and allocate your time to
other students because this one projects to perform
much better than the average score of 162.
894
4.2.3 Confidence & Prediction Intervals
895
Prediction
• The purpose of building a regression model is not only to
understand what happened in the past but more
importantly to predict the future based on the past.
• By plugging the values of independent variables into the

regression equation, we obtain the estimation/prediction

of the dependent variable.
896
Uncertainty of Prediction
• We build the regression model using the sample data to
describe as close as possible the true population
relationship between dependent and independent
variables.
• Due to noise in the data, the prediction will probably

differ from the true response value.
• However, the true response value might fall in a range

around the prediction with some certainty.
• To measure the uncertainty of the prediction, we need

confidence interval and prediction interval.
897
Confidence Interval
• The confidence interval of the prediction is a range in
which the population mean of the dependent variable
would fall with some certainty, given specified values of
the independent variables.
• The width of confidence interval is related to:
• Sample size

• Confidence level
• Variation in the data.
• We build the model based on a sample set {y1, y2,…, yn}.
The confidence interval is used to estimate the value of
the population mean μ of the underlying population.
• The focus of the confidence interval are the
unobservable population parameters.
• The confidence interval accounts for the uncertainty in
the estimates of regression parameters. 898
Prediction Interval
• The prediction interval is a range in which future values of
the dependent variable would fall with some certainty, given
specified values of the independent variables.
• We build the model based on a sample set {y1, y2,…, yn}.

The prediction interval is used to estimate the value of

future observation yn+1.
• The focus of the prediction interval are the future

observations.
• Prediction interval is wider than confidence interval

because it accounts for the uncertainty in the estimates of
regression parameters and the uncertainty of the new
measurement. 899
Use Minitab to Obtain Prediction and Confidence Interval
Steps in Minitab to obtain prediction, confidence interval,

and prediction interval.
1) In the “Regression” window, click on the “Options” button.
2) A new window named “Regression – Options” appears.
3) Click in the blank box right below “Prediction intervals for new
observations” and the columns existing in the data table appear

in the list box on the left.
4) Select both “EXAM2” and “EXAM3” into the box below
“Prediction intervals for new observations.”
5) Check the boxes of “Fits”, “Confidence limits,” and “Prediction
limits.”
6) Click “OK” in the window named “Regression – Options.”
7) Click “OK” in the window named “Regression.”
8) The fitted response (PFIT), the confidence interval (CLIM), and
the prediction interval (PLIM) are automatically added to the data
table.
900
Use Minitab to Obtain Prediction and Confidence Interval

901
902
Remember what Residuals Are?
• Residuals are the vertical difference between actual
values and the predicted values or the “fitted line”
created by the regression model.

Predicted Value
Vertical
Distance
Actual Point
903
Why Perform Residuals Analysis?
• The regression equation generated based on the sample
data can make accurate statistical inference only if certain
assumptions are met. Residuals analysis can help to
validate these assumptions. The following assumptions
must be met to ensure the reliability of the linear regression
model:

• The errors are normally distributed with mean equal to zero.
• The errors are independent.
• The errors have a constant variance.
• The underlying population relationship is linear.
• If the residuals performance does not meet the

requirement, we will need to rebuild the model by replacing
the predictors with new ones, adding new predictors,
building non-linear models, and so on.
904
• The residuals of the model are automatically saved the data
table as long as we check the box “Residuals” in the
“Regression – Storage” window.

905
• Step 1: Check whether residuals are normally

distributed around the mean of zero.

4) Click “OK.”
5) The histogram and analysis results are shown

automatically in the new window.
906

907
• The mean of residuals is -0.0000.
• The Anderson-Darling test is used to test the normality. Since
the p-value (0.339) is greater than the alpha level (0.05), we
fail to reject the null hypothesis; the residuals are normally
distributed.
• H0: The residuals are normally distributed.

• H1: The residuals are not normally distributed.
908
• Step 2: If the data are in time order, run the IR chart
to check whether residuals are independent.


pops up.
3) Select “RESI1” as “Variables” and check the box of “Test

for special causes.”
4) Click “OK.”
5) The control charts are shown automatically in the new

window.
909

910
• If no data points are out of control in both the I-chart and MR
chart, the residuals are independent of each other.
• If the residuals are not independent, it is possible that some
important predictors are not included in the model.
• In this example, since the IR chart is in control, residuals are
independent.

911
• Step 3: Check whether residuals have equal

variance across the predicted responses.
• Create a scatterplot with Y being the residuals and the X
being the fitted values.

2) A new window named “Scatterplots” appears.
3) Select “RESI1” as the “Y variables” and “Fitted” as the “X
variables.”
4) Click “OK.”
5) The scatterplot appears in a new window.
• We are looking for the pattern in which residuals spread out

evenly around zero from the top to the bottom.
912

913
4.2.5 Data Transformation
914
What is the Box-Cox Transformation?
• When a response does not fit the model well, sometimes
using a transformation of the response can improve the
fit.
• Power transformation is a class of transformation

functions that raise the response to some power. For

example, a square root transformation converts X to X1/2.
• Box-Cox transformation is a popular power

transformation method developed by George E. P. Box
and David Cox.
915
Box-Cox Transformation Formula
• The formula of the Box-Cox transformation is:

x 1
y where 0


y  ln x where  0
y is the transformation result.
x is the variable under transformation.
λ is the transformation parameter.
916
Use Minitab to Perform a Box-Cox Transformation
• Minitab provides the best Box-Cox transformation with an

optimal λ that minimizes the model SSE (sum of squared
error).
• Here is an example of how we transform the non-

normally distributed response to normal data using Box-

Cox method.
• Data File: “Box-Cox” tab in “Sample Data.xlsx”
917
• Step 1: Test the normality of the original data set.

3) Select “Y” as “Variable.”
4) Click “OK.”
5) The normality test results are shown automatically in the

new window.
918

919
• Normality Test:
• H0: The data are normally distributed.
• H1: The data are not normally distributed.
• If p-value > alpha level (0.05), we fail to reject the null
hypothesis. Otherwise, we reject the null.
• In this example, p-value = 0.029 < alpha level (0.05). The data

are not normally distributed.
920
• Step 2: Run the Box-Cox Transformation:
1) Click Stat → Control Charts → Box-Cox Transformation.
2) A new window named “Box-Cox Transformation” pops up.
3) Click into the blank list box below “All observations for a chart

are in one column.”
4) Select “Y” as the variable.
5) Select “Run” into the box next to “Subgroup sizes (enter a

number or ID column).”
6) Click “OK.”
7) The analysis results are shown automatically in the new

window.
921

922
• When λ is 0.12, the transformation is the optimal with minimum

SSE.

923
• The transformed Y can also be saved in another

column.
1) Create a new column named “Y1” in the data table.
2) Click on the “Options” button in the “Box-Cox
Transformation” window.
3) A new window named “Box-Cox Transformation –

Options” appears.
4) Click in the blank box under “Store transformed data in”
and all the columns pop up in the list box on the left.
5) Select “Y1” in “Store transformed data in.”
6) Click “OK” in the window “Box-Cox Transformation –
Options.”
7) Click “OK” in the window “Box-Cox Transformation.”
8) The transformed column is stored in the column “Y1.”
924

925
• Run the normality test to check whether the transformed data are normally
distributed.
• Use the Anderson-Darling test to test the normality of the transformed data
• H0: The data are normally distributed.
• H1: The data are not normally distributed.
• If p-value > alpha level (0.05), we fail to reject the null. Otherwise, we reject the null.
• In this example, p-value = 0.327 > alpha level (0.05). The data are normally distributed.

926
927
What is Stepwise Regression?
• Stepwise regression is a statistical method to

automatically select regression models with the best
sets of predictive variables from a large set of
potential variables.

• There are different statistical methods used in
stepwise regression to evaluate the potential
variables in the model:
• F-test
• T-test
• R-square
• AIC.
928
Three Approaches to Stepwise Regression
• Forward Selection
• Bring in potential predictors one by one and keep them if
they have significant impact on improving the model.
• Backward Selection

• Try out potential predictors one by one and eliminate them if
they are insignificant to improve the fit.
• Mixed Selection
• Is a combination of both forward selection and backward
selection. Add and remove variables based on pre-defined
significance threshold levels.
929
How to Use Minitab to Run a Stepwise Regression
• Case study: We want to build a regression model to

predict the oxygen uptake of a person who runs 1.5
miles. The potential predictors are:
• Age
• Weight
•

Runtime
• Runpulse
• RstPulse
• MaxPulse.
• Data File: “Stepwise Regression” tab in “Sample

Data.xlsx”
930
• Sample Data Glance

931
• Steps to run stepwise regression in Minitab

1) Click Stat → Regression → Stepwise.
2) A new window named “Stepwise Regression” appears.
3) Select “Oxy” as the Y and select all the other variables
into the “Predictors” box.
4) Click the “Methods” button and a new window named

“Stepwise – Methods” pops up.
5) Select the method of stepwise regression and enter the
alphas to enter/remove. In this example, we use the
“Forward selection” method and the alpha to enter is 0.25.
6) Click “OK” in the window “Stepwise – Methods.”
7) Click “OK” in the window “Stepwise Regression.”
8) The results appear in the session window.
932

933
Model summary:
One out of six potential
factors is not statistically
significant since its p-
value is higher than the
alpha to enter.

Step History:
Step-by-step
records on how
to come up with
the final model.
Each column
indicates the
model built in
each step.
934
935
What is Logistic Regression?
• Logistic regression is a statistical method to

predict the probability of an event occurring by fitting
the data to a logistic curve using logistic function.
• The dependent variable in a logistic regression can

be binary (e.g., 1/0, yes/no, pass/fail), nominal
(blue/yellow/green), or ordinal
(satisfied/neutral/dissatisfied).
• The independent variables can be either continuous

or discrete.
936
Logistic Function
1
f ( z)  z
1 e
where z can be any value ranging from negative infinity

to positive infinity.
The value of f(z) ranges from 0 to 1, which matches

exactly the nature of probability (i.e., 0 ≤ P ≤ 1).
937
Logistic Regression Equation
• Based on the logistic function

1
f ( z) 
1  e z
we define f(z) as the probability of an event

occurring and z is the weighted sum of the
significant predictive variables.
z  0  1  x1   2  x2       k  xk
938
Logistic Regression Equation
• Logistic Regression Y = F(x)

1
Y
1  e ( 0  1 x1   2  x2   k  xk )
where Y is the probability of an event occurring and

x’s are the significant predictors.
Note:
 When building the regression model, we use the actual Y,
which is discrete (e.g., binary, nominal, ordinal).
 After completing building the model, the fitted Y calculated
using the logistic regression equation is the probability
ranging from 0 to 1. To transfer the probability back to the
discrete value, we need SMEs’ inputs to select the
probability cut point.
939
Logistic Curve
• Logistic curve for binary logistic regression with one

continuous predictor:

Probability
1
of the event Y
occurring 1  e ( 0  1 x )
0
X
940
Odds
• Odds is the probability of an event occurring divided

by the probability of the event not occurring.
P
Odds 
1 P

• Odds range from 0 to positive infinity.
941
Odds
• Probability can be calculated using odds.
odds eln(odds) 1
P  
1  odds 1  e ln(odds)
1  e ln(odds)
• Since in logistic regression model

1
P (  0  1  x1   2  x2   k  xk )
1 e
therefore
ln(odds)  0  1  x1   2  x2       k  xk
942
Three Types of Logistic Regression
• Binary Logistic Regression

• Binary response variable
• Example: yes/no, pass/fail, female/male
• Nominal Logistic Regression
• Nominal response variable

• Example: set of colors, set of countries
• Ordinal Logistic Regression
• Ordinal response variable
• Example: satisfied/neutral/dissatisfied
• All three logistic regression models can use multiple

continuous or discrete independent variables.
• All three logistic regression models can be developed in
Minitab using the same steps.
943
How to Run a Logistic Regression in Minitab
• Data File: “Logistic Regression” tab in “Sample Data.xlsx”
• Response and potential factors
• Response (Y): Female/Male
• Potential Factors (Xs):
• Age
• Weight

• Oxy
• Runtime
• RunPulse
• RstPulse
• MaxPulse.
• We want to build a logistic regression model using the potential

factors to predict the probability that the person measured is
female or male.
944
• Step 1:
1) Click Stat → Regression → Binary Logistic Regression.
2) A new window named “Binary Logistic Regression”

appears.

3) Click into the blank box next to “Response” and all the
variables pop up in the list box on the left.
4) Select “Sex” as the “Response.”
5) Select “Age”, “Weight”, “Oxy”, “Runtime”, “RunPulse”,

“RstPulse”, “MaxPulse” into the “Model.”
6) Click “OK.”
945

946
• Step 2:
1) Step 2.1: The results of the logistic regression model

appear in session window.

2) Step 2.2: Check the p-values of all the independent
variables in the model.
3) Step 2.3: Remove the insignificant independent variable

one at a time from the model and rerun the model.
4) Step 2.4: Repeat step 2.1 until all of the independent

variables in the model are statistically significant.
947
Since the p-values of all

the independent variables
are higher than the alpha
level (0.05), we need to
remove the insignificant
independent variables one

at a time from the model,
starting from the one with
the highest p-value.
MaxPulse has the highest

p-value (0.9390), so it
would be removed from the
model first.
948
After removing MaxPulse from

the model, the p-values of all
the independent variables are
still higher than the alpha level

(0.05).
We need to continue removing

the insignificant independent
variables one at a time from the
model, starting from the one
with the highest p-value.
RunPulse has the highest p-

value (0.4516), so it would be
removed from the model next.
949
After removing RunPulse from


(0.05).

Runtime has the highest p-

value (0.2684), so it would be
950
After removing Runtime from the

model, the p-values of all the
independent variables are still
higher than the alpha level (0.05).

model, starting from the one with
Age has the highest p-value

(0.3821), so it would be removed
from the model next.
951
After removing Age from the

model, the p-values of all the
independent variables are still
higher than the alpha level (0.05).

model, starting from the one with
RstPulse has the highest p-value

(0.3857), so it would be removed
from the model next.
952
After removing RstPulse from

(0.05).

Weight has the highest p-value

(0.0755), so it would be
953
After removing Weight from the

model, the p-value of the only

independent variable “Oxy” is
lower than the alpha level (0.05).
There is no need to remove
“Oxy” from the model.
954
• Step 3:
• Analyze the binary logistic report in the session window and
check the performance of the logistic regression model.
The p-value here is 0.003,
smaller than alpha level (0.05).
We conclude that at least one of
the slope coefficients is not

equal to zero.
The p-value of goodness of fit

tests are all higher than alpha
level (0.05). We conclude that
the model fits the data.
955
• Step 4: get the predicted probabilities of the event

(i.e., Sex = M) occurring using the logistic regression
model.
1) Click the “Storage” button in the window named “Binary
Logistic Regression” and a new window named “Binary
Logistic Regression – Storage” pops up.

2) Check the box “Event probability.”
3) Click “OK” in the window of “Binary Logistic Regression –

Storage.”
4) Click “OK” in the window of “Binary Logistic Regression.”
5) A column of the predicted event probability is added to the

data table with the heading “EPRO1”.
956

957
958

5.0 Control Phase
Green Belt Training: Control Phase
5.1 Lean Controls 5.2.12 Subgroups, Variation, Sampling
5.1.1 Control Methods for 5S 5.3 Six Sigma Control Plans
5.1.2 Kanban 5.3.1 Cost Benefit Analysis
5.1.3 Poka-Yoke (Mistake Proofing) 5.3.2 Elements of the Control Plan
5.2 Statistical Process Control (SPC) 5.3.3 Elements of the Response Plan
5.2.1 Data Collection for SPC

5.2.2 I-MR Chart
5.2.3 Xbar-R Chart
5.2.4 U Chart
5.2.5 P Chart
5.2.6 NP Chart
5.2.7 X-S chart
5.2.8 CumSum Chart
5.2.9 EWMA Chart
5.2.10 Control Methods
5.2.11 Control Chart Anatomy
959
960

5.1 Lean Controls

5.2.2 I-MR Chart
5.2.3 Xbar-R Chart
5.2.4 U Chart
5.2.5 P Chart
5.2.6 NP Chart
5.2.7 X-S chart
5.2.8 CumSum Chart
5.2.9 EWMA Chart
961
5.1.1 Control Methods for 5S
962
What is 5S?
• 5S is a systematic method to organize, order, clean, and
standardize a workplace…and keep it that way!
• 5S is a methodology of organizing and improving the work
environment.
• 5S is summarized in five Japanese words, all starting

with the letter S:
• Seiri (sorting)
• Seiso (shining)
• Seiketsu (standardizing)
• Shisuke (sustaining).
• 5S was originally developed in Japan, and is widely used to

optimize the workplace to increase productivity and efficiency.
963
5S Goals
• Reduced waste
• Reduced cost
• Establish a work environment that is:
• self-explaining
• self-ordering

• self-regulating
• self improving.
• Where there is/are no more:
• Wandering and/or searching
• Waiting or delays
• Secrets hiding spots for tools
• Obstacles or detours
• Extra pieces, parts, materials, etc.
• Injuries
• Waste.
964
5S Benefits
• Reduced changeovers
• Reduced defects
• Reduced waste
• Reduced delays
• Reduced injuries

• Reduced breakdowns
• Reduced complaints
• Reduced red ink
• Higher quality
• Lower costs
• Safer work environment
• Greater associate and equipment capacity
965
5S Systems Reported Results
• Cut in floor space: 60%

• Cut in flow distance: 80%
• Cut in accidents: 70%
• Cut in rack storage: 68%
• Cut in number of forklifts: 45%

• Cut in machine changeover time: 62%
• Cut in annual physical inventory time: 50%
• Cut in classroom training requirements: 55%
• Cut in nonconformance in assembly: 96%
• Increase in test yields: 50%
• Late deliveries: 0%
• Increase in throughput: 15%
966
Sorting (Seiri)
• Go through all the tools, parts,
equipment, supply, and material in
the workplace.
• Categorize them into two major

groups: needed and unneeded.
• Eliminate the unneeded items from

the workplace. Dispose of or
recycle those items.
• Keep the needed items and sort them in the order

of priority. When in doubt…throw it out!
967
Straightening (Seiton)
• Straightening in 5S is also called
setting in order.
• Label each needed item.

• Store items at their best locations
so that the workers can find them

easily whenever they needed any
item.
• Reduce the motion and time
required to locate and obtain any
item whenever it is needed.
• Promote an efficient work flow path.
• Use visual aids like the tool board
image on this page.
968
Shining (Seiso)
• Shining in 5S is also called sweeping.
• Clean the workplace thoroughly.
• Maintain the tidiness of the workplace.

• Make sure every item is located at the
specific location where it should be.
• Create the ownership in the team to

keep the work area clean and
organized.
969
Standardizing (Seiketsu )
• Standardize the workstation and the layout of tools,

equipment, and parts.
• Create identical workstations with a consistent way

of storing the items at their specific locations so that

workers can be moved around to any workstation
any time and perform the same task.
970
Sustaining (Shisuke)
• Sustaining in 5S is also called self-discipline.

• Create the culture in the team to follow the first four S’s
consistently.
• Avoid falling back to the old ways of cluttered and
unorganized work environment.
• Keep the momentum of optimizing the workplace.

• Promote innovations of workplace improvement.
• Sustain the first fours S’s using:
• 5S Maps
• 5S Schedules
• 5S Job cycle charts
• Integration of regular work duties
• 5S Blitz schedules
• Daily workplace scans.
971
Simplified Summary of 5S
• Sort – “when in doubt, move it out.”
• Set in Order – Organize all necessary tools, parts, and

components of production. Use visual ordering
techniques wherever possible.

• Shine – Clean machines and/or work areas. Set regular
cleaning schedules and responsibilities.
• Standardize – Solidify previous three steps, make 5S a

regular part of the work environment and everyday life.
• Sustain – Audit, manage, and comply with established

five-s guidelines for your business or facility
972
973

5.1.2 Kanban
What is Kanban?
• The Japanese word “Kanban” means “signboard.”
• Kanban system is a “pull” production scheduling

system to determine when to produce, what to
produce, and how much to produce based on the

demand.
• It was originally developed by Taiichi Ohno in order

to reduce the waste in inventory and increase the
speed of responding to the immediate demand.
974
Kanban System
• Kanban system is a demand-driven system.
• The customer demand is the signal to trigger or pull

the production.

• Products are made only to meet the immediate
demand. When there is no demand, there is no
production.
• It is designed to minimize the in-process inventory

and to have the right material with the right amount
at the right location at the right time.
975
Kanban System
• Principles of the Kanban System:

• Only produce products with exactly the same amount that
customers consume.
• Only produce products when customers consume.
• The production is driven by the actual demand from

the customer side instead of the forecasted demand
planned by the staff.
976
Kanban Card
• The Kanban card is the ticket or signal to authorize

the production or movement of materials. It is the
message of asking for more.
• It is sent from the end customer up to the chain of

production.
• Upon receiving of a Kanban card, the production

station would start to produce goods.
• The Kanban card can be a physical card or an

electronic signal.
977
Kanban System Example
• The simplest example of a Kanban system is the supermarket
operation.
• Customers visit the supermarkets and buy what they need.
• The checkout scanners send electronic Kanban cards to the

local warehouse asking for more when the items are sold to
customers.
• When the warehouse receives the Kanban cards, it starts to

replenish the exact goods being sold.
• It the warehouse prepares more than what Kanban cards

require, the goods would become obsolete. If it prepares less,
the supermarket would not have the goods available when
customers need them.
978
Kanban System Benefits
• Minimize in-process inventory
• Free up space occupied by unnecessary inventory
• Prevent overproduction

• Improve responsiveness to dynamic demand
• Avoid the risk of inaccurate demand forecast
• Streamline the production flow
• Visualize the work flow.

979
980

5.1.3 Poka-Yoke
What is Poka-Yoke?
• The Japanese term “poka-yoke” means “mistake-

proofing.”
• It is a mechanism to eliminate defects as early as

possible in the process.

• It was originally developed by Shigeo Shingo and
was initially called “baka-yoke” (fool-proofing).
981
Two Types of Poka-Yoke
• Prevention
• Preventing defects from occurring
• Removing the possibility that an error could occur
• Making the occurrence of an error impossible.

• Detection
• Detecting defects once they occur
• Highlighting defects to draw workers’ attention immediately
• Correcting defects so that they would not reach the next
stage.
982
Three Methods of Poka-Yoke
• Contact Method
• Use of shape, color, size, or any other physical attributes of
the items.
• Constant Number Method

• Use of a fixed number to make sure a certain number of
motions are completed.
• Sequence Method
• Use of a checklist to make sure all the prescribed process
steps are followed in the right order.
983
Poka-Yoke Devices
• We are surrounded by poka-yoke devices daily.

• Prevention Devices
• Example: the dishwasher does not start to run when the door is
open.
• Detection Devices
• Example: the car starts to beep when the passengers do not

buckle their seatbelts.
• Poka-yoke devices can be in any format that can

quickly and effectively prevent or detect mistakes.
• Visual, electrical, mechanical, procedural, human etc.
984
Steps to Apply Poka-Yoke
• Step 1: Identify the process steps in need of mistake proofing.
• Step 2: Use the 5-why’s method to analyze the possible

mistakes or failures for the process step.
• Step 3: Determine the type of poka-yoke: prevention or

detection.

• Step 4: Determine the method of poka-yoke: contact, constant
number, or sequence.
• Step 5: Pilot the poka-yoke approach and make any

adjustments if needed.
• Step 6: Implement poka-yoke in the operating process and

maintain the performance.
985
5.2 Statistical Process Control
986

5.2.2 I-MR Chart
5.2.3 Xbar-R Chart
5.2.4 U Chart
5.2.5 P Chart
5.2.6 NP Chart
5.2.7 X-S chart
5.2.8 CumSum Chart
5.2.9 EWMA Chart
987
988
What is SPC?
• Statistical process control (SPC) is a statistical

method to monitor the performance of a process
using control charts in order to keep the process in
statistical control.

• Statistical process control can be used to distinguish
between special cause variation and common cause
variation in the process.
• It presents the voice of the process.
989
Common Cause Variation
• Common cause variation (also called chance

variation) is the inherent natural variation in any
processes.
• It is the random background noise, which cannot be

controlled or eliminated from the process.
• Its presence in the process is expected and

acceptable due to its relatively small influence on
the process.
990
Special Cause Variation
• Special cause variation (also called assignable

cause variation) is the unnatural variation in the
process.
• It is the cause of process instability and leads to

defects of the products or services.
• It is the signal of unanticipated change (either

positive or negative) in the process.
• It is possible to eliminate the special cause variation

from the process.
991
Process Stability
• A process is stable when:
• There is not any special cause variation involved in the

process
• The process is in statistical control

• The future performance of the process is predictable
within certain limits
• The changes happening in the process are all due to

random inherent variation
• There are not any trends, unnatural patterns, and

outliers in the control chart of the process.
992
SPC Benefits
• Statistical process control can be used in different

phases of Six Sigma projects to:
• Understand the stability of a process

• Detect the special cause variation in the process
• Identify the statistical difference between two phases
• Eliminate or apply the unnatural change in the process
• Improve the quality and productivity.
993
Control Charts
• Control charts are graphical tools to present and

analyze the process performance in statistical
process control.
• Control charts are used to detect special cause

variation and determine whether the process is in
statistical control (stable).
• Variation solutions:
• Minimize the common cause variation
• Eliminate the special cause variation when it leads to unanticipated
negative changes in the outcome
• Implement the special cause variation when it leads to
unanticipated positive changes in the outcome.
994
Control Charts Elements
Data Points: individual observations

or their calculation results Upper Line: upper control limit (UCL)
Vertical Axis: Measurement Values

Center Line: the process mean Lower Line: lower control limit (LCL)
Horizontal axis: Time

995
Control Charts Elements
• Control charts can work for both continuous data

and discrete or count data.
• Control limits are approximately three sigma away

from the process mean.

• A process is in statistical control when all the data
points on the control charts fall within the control
limits and have random patterns only.
• Otherwise, the process is out of control and we

need to investigate the special cause variation in the
process.
996
Possible Errors in SPC
• There are two types of possible errors in interpreting

controls charts.

997
• It is similar to the way of defining the type I and type

II errors in hypothesis testing.
• Control charts can be interpreted as a way of testing

the hypothesis about the process stability.

• Null Hypothesis (H0): The process is stable (i.e., in statistical
control).
• Alternative Hypothesis (HA): The process is unstable (i.e.,
out of statistical control).
998
• Type I Error
• False positive
• False alarm
• Considering true common cause variation as special cause
variation
• Type I errors waste resources spent on investigation.

• Type II Error
• False negative
• Miss
• Considering true special cause variation as common cause
variation
• Type II errors neglect the need to investigate critical
changes in the process.
999
Data Collection Considerations
• To collect data for plotting control charts, we need to

consider:
• What is the measurement of interest?
• Are the data discrete or continuous?
• How many samples do we need?

• How often do we sample?
• Where do we sample?
• What is the sampling strategy?
• Do we use the raw data collected or transfer them to
percentages, proportions, rates, etc.?
1000
Subgroups and Rational Subgrouping
• When sampling, we randomly select a group of items (i.e., a
subgroup) from the population of interest.
• The subgroup size is the count of samples in a subgroup. It

can be constant or variable.
• Depending on the subgroup sizes, we select different control

charts accordingly.
• Rational subgrouping is the basic sampling scheme in SPC.
• The goal of rational subgrouping is to maximize the likelihood

of detecting special cause variation. In other words, the control
limits should only reflect the variation between subgroups.
• The number of subgroups, subgroup size, and frequency of

sampling have great impact on the quality of control charts.
1001
Impact of Variation
• The rational subgrouping strategy is designed to

minimize the opportunity of having special cause
variation within subgroups.
• If there is only random variation (background noise)

within subgroups, all the special cause variation
would be reflected between subgroups. It is easier
to detect an out-of-control situation.
• Random variation is inherent and indelible in the

process. We are more interested in identifying and
taking actions on special cause variation.
1002
Frequency of Sampling
• The frequency of sampling in SPC depends on

whether we have sufficient data to signal the
changes in a process with reasonable time and
costs.

• The more frequently we sample, the higher costs it
may trigger.
• We need the subject matter experts’ knowledge on

the nature and characteristics of the process to
make good decisions on sampling frequency.
1003
1004

5.2.2 I-MR Chart
I-MR Chart
• The I-MR chart (also called individual-moving range chart or
IR chart) is a popular control chart for continuous data with
subgroup size equal to one.
• The I chart plots an individual observation as a data point.
• The MR chart plots the absolute value of the difference

between two consecutive observations in individual charts as a
data point.
• If there are n data points in the I chart, there are n –1 data

points in the MR chart.
• The I chart is valid only if the MR chart is in control.
• The underlying distribution of the I-MR chart is normal

distribution.
1005
I Chart Equations
• I Chart (Individuals Chart)
• Data Point: xi
n
x

• Center Line: i 1
i
n
n
• Control Limits:
x i
i 1
 2.66  MR
n
where n is the number of observations
1006
MR-Chart Equations
• MR Chart (Moving Range Chart)
• Data Point: xi 1  xi
xi 1  xi
• Center Line:

n 1
xi 1  xi
• Upper Control Limit: 3.267 
n 1
• Lower Control Limit: 0
where n is the number of observations

1007
Use Minitab to Plot I-MR Charts
• Data File: “IR” tab in “Sample Data.xlsx”
• Steps to plot IR charts in Minitab
appears.

3) Select “Measurement” as the “Variables.”
4) Click “I-MR Options” button and a new window “Individual
– Moving Range Chart – Options” appears.
5) Click on the tab “Tests.”
6) Select the item “Perform all tests for special causes” in the
dropdown box.
7) Click “OK” in the window “Individuals – Moving Range
Chart Options.”
8) Click “OK.”
9) The IR charts appear in the newly-generated window.
1008

1009

1010
I-MR Charts Diagnosis
I Chart (Individuals’ Chart):

Since the MR chart is out of
control, the I chart is invalid.
MR Chart (Moving Range

Chart):

Two data points fall beyond
the upper control limit. This
indicates the MR chart is out
of control (i.e., the variations
between every two contiguous
individual samples are not
stable over time). We need to
further investigate the
process, identify the root
causes that trigger the
outliers, and correct them to
bring the process back in
control.
1011
1012

5.2.3 Xbar-R Chart
Xbar-R Chart
• The Xbar-R chart is a control chart for continuous data with a
constant subgroup size between two and ten.
• The Xbar chart plots the average of a subgroup as a data

point.
• The R chart plots the difference between the highest and

lowest values within a subgroup as a data point.
• The Xbar chart monitors the process mean and the R chart
monitors the variation within subgroups.
• The Xbar is valid only if the R chart is in control.
• The underlying distribution of the Xbar-R chart is normal

distribution.
1013
Xbar Chart Equations
• Xbar chart m
x
j 1
ij
• Data Point: X i 
m
X

i
• Center Line: X  i 1
k
• Control Limits: X  A2 R
where m is the subgroup size and k is the number of

subgroups. A2 is a constant depending on the subgroup size.
1014
R Chart Equations
• R chart (Rage Chart)
• Data Point: Ri  Max

j1, m 
( xij )  Min ( xij )
j1, m 
R i
• Center Line: R  i 1

k
• Upper Control Limit: D4  R
• Lower Control Limit: D3  R

subgroups. D3 and D4 are constants depending on the
subgroup size.
1015
Use Minitab to Plot Xbar-R Charts
• Data File: “Xbar-R” tab in “Sample Data.xlsx”
• Steps to plot Xbar-R charts in Minitab
1) Click Stat → Control Charts → Variable Charts for Subgroups →
Xbar-R.
2) A new window named “Xbar-R Chart” appears.
3) Click in the blank box right below “All observations for a chart are
in one column” and the variables appear in the list box on the
left.

4) Select the “Measurement” into the box below “All observations
for a chart are in one column.”
5) Select the “Subgroup ID” as the “Subgroup size (enter a number
or ID column).
6) Click “Xbar-R Options” button and a new window “Xbar-R Chart –
Options” appears.
7) Click on the tab “Tests.”
dropdown menu.
9) Click “OK” in the window “Xbar-R Chart – Options.”
10) Click “OK.”
11) The Xbar-R charts appear in the newly-generated window.
1016

1017

1018
Xbar-R Charts Diagnosis
Xbar-R Charts:
Since the R chart is in

control, the Xbar chart is
valid.
In both charts, there are

not any data points
failing any tests for
special causes (i.e., all
the data points fall
between the control
limits and spread around
the center line with a
random pattern).
We conclude that the

process is in control.
1019
1020

5.2.4 U Chart
Defect vs. Defective
• A defect of a unit is the unit’s characteristic that

does not meet the customers’ requirements.
• A defective is a unit that is not acceptable to the

customers.

• One defective might have multiple defects.
• One unit might have multiple defects but be still

usable to the customers.
1021
U Chart
• The U chart is a control chart monitoring the

average defects per unit.
• The U chart plots the count of defects per unit of a

subgroup as a data point.

• It considers the situation when the subgroup size of
inspected units for which the defects would be
counted is not constant.
• The underlying distribution of the U chart is Poisson

distribution.
1022
U Chart Equations
• U chart
xi
• Data Point: ui 
ni
k
u i
• Center Line: u 

i 1
k
• Control Limits: u  3 
u
ni
where ni is the subgroup size for the ith subgroup;

k is the number of subgroups;
xi is the number of defects in the ith subgroup.
1023
Use Minitab to Plot a U Chart
• Data File: “U” tab in “Sample Data.xlsx”

• Steps to plot a U chart in Minitab
1) Click Stat → Control Charts → Attributes Charts → U.
2) A new window named “U Chart” appears.
3) Select “Count of Defects” as the “Variables.”
4) Select “Count of Units Inspected” as the “Subgroup Sizes.”

5) Click the button “U Chart Options” and another window named
“U Chart Options” pops up.
6) Click the tab “Tests.”
dropdown menu.
8) Click “OK” in the window “U Chart Options.”
9) Click “OK” in the window “U Chart.”
10) The U chart appears in the newly-generated window.
1024

1025

1026
U Chart Diagnosis
U Chart:
Since the sample sizes are

not constant over time, the
control limits are adjusted
to different values
accordingly.

The data point circled in
red falls beyond the upper
control limit. We conclude
that the process is out of
control.
Further investigation is
needed to determine the
special causes that
triggered the unnatural
pattern of the process.
1027
1028

5.2.5 P Chart
P Chart
• The P chart is a control chart monitoring the

percentages of defectives.
• The P chart plots the percentage of defectives in

one subgroup as a data point.

• It considers the situation when the subgroup size of
inspected units is not constant.
• The underlying distribution of the P chart is binomial

distribution.
1029
P Chart Equations
• P chart
xi
• Data Point: pi  n
i
k
x i
• Center Line: p 
i 1

k
n
i 1
i
p(1  p)
• Control Limits: p  3
ni
where ni is the subgroup size for the ith subgroup;

xi is the number of defectives in the ith subgroup.
1030
Use Minitab to Plot a P Chart
• Data File: “P” tab in “Sample Data.xlsx”

• Steps to plot a P chart in Minitab
1) Click Stat → Control Charts → Attributes Charts → P.
2) A new window named “P Chart” appears.
3) Select “Fail” as the “Variables.”
4) Select “N” as the “Subgroup Sizes.”

5) Click the button “P Chart Options” and another window named
“P Chart Options” pops up.
dropdown menu.
8) Click “OK” in the window “P Chart Options.”
9) Click “OK.”
10) The P chart appears in the newly-generated window.
1031

1032

1033
P Chart Diagnosis
P Chart:
Since the sample

sizes are not constant
over time, the control
limits are adjusted to

different values
accordingly.
All the data points fall

within the control
limits and spread
randomly around the
mean. We conclude
that the process is in
control.
1034
1035

5.2.6 NP Chart
NP Chart
• The NP chart is a control chart monitoring the count

of defectives.
• The NP chart plots the number of defectives in one


• The subgroup size of the NP chart is constant.
• The underlying distribution of the NP chart is

1036
NP Chart Equations
• NP chart
• Data Point: xi
k
x i
• Center Line: n p  i 1

k
• Control Limits: n p  3  n p(1  p)
where n is the constant subgroup size;

xi is the number of defectives in the ith subgroup.
1037
Use Minitab to Plot an NP Chart
• Data File: “NP” tab in “Sample Data.xlsx”

• Steps to plot a NP chart in Minitab
1) Click Stat → Control Charts → Attributes Charts → NP.
2) A new window named “NP Chart” appears.
3) Select “Fail” as the “Variables.”
4) Enter “50” as the “Subgroup Sizes.”

5) Click the button “NP Chart Options” and another window named
“NP Chart Options” pops up.
dropdown menu.
8) Click “OK” in the window “NP Chart Options.”
9) Click “OK.”
10) The NP chart appears in the newly-generated window.
1038

1039

1040
NP Chart Diagnosis
NP Chart:
Four data points, circled

in red, fall beyond the
upper control limit.

NP chart is out of
control.
needed to determine the
special causes that
triggered the unnatural
pattern of the process.
1041
1042

5.2.7 X-S Chart
X-S Chart
• The X-S chart (also called Xbar-S chart) is a control
chart for continuous data with a constant subgroup size
greater than ten.
• The Xbar chart plots the average of a subgroup as a data

point.

• The S chart plots the standard deviation within a
• The Xbar chart monitors the process mean and the S

chart monitors the variability within subgroups.
• The Xbar is valid only if the S chart is in control.
• The underlying distribution of the Xbar-S chart is normal

distribution.
1043
Xbar Chart Equations
• Xbar Chart m
x
j 1
ij
• Data Point: X i 
m
k
X i
• Center Line: X 
i 1

k
• Control Limits: X  A3 s

subgroups. A3 is a constant depending on the subgroup size.
1044
S Chart Equations
• S chart m
 ij
( x
j 1
 x i ) 2
• Data Point: si 
m 1
k
s i

• Center Line: s  i 1
k
• Upper Control Limit: B4  s
• Lower Control Limit: B3  s
where m is the subgroup size and k is the number of subgroups.

B3 and B4 are constants depending on the subgroup size.
1045
Use Minitab to Plot Xbar-S Charts
• Data File: “Xbar-S” tab in “Sample Data.xlsx”

• Steps to plot Xbar-S charts in Minitab
1) Click Stat → Control Charts → Variables Charts for Subgroups
→ Xbar-S.
2) A new window named “Xbar-S Chart” appears.
3) Select “Measurement” by ticking the box below “All observations

for a chart are in one column.”
4) Select “Subgroup ID” as the “Subgroup Sizes.”
5) Click the button “Xbar-S Options” and another window named
“Xbar-S Chart Options” pops up.
dropdown menu.
8) Click “OK” in the window “Xbar-S Chart Options.”
9) Click “OK.”
10) The Xbar-S charts appear in the newly-generated window.
1046

1047

1048
Xbar-S Charts Diagnosis
Xbar-S Charts:
Since the S chart is in

control, the Xbar chart is
valid.
In both charts, there are not

any data points failing any
tests for special causes
(i.e., all the data points fall
between the control limits
and spread around the
center line with a random
pattern).

process is in control.
1049
1050

5.2.8 CumSum Chart
CumSum Chart
• The CumSum chart (also called cumulative sum
control chart or CUSUM chart) is a control chart of
monitoring the cumulative sum of the subgroup
mean deviations from the process target.
• It detects the shift of the process mean from the

process target over time.
• The underlying distribution of the CumSum chart is

normal distribution.
• There are two types of CumSum charts:

• One two-sided CumSum charts
• Two one-sided CumSum charts.
1051
Two-Sided CumSum
• In two-sided CumSum charts, we use a V-mask to

identify out-of-control data points.
• The V-mask is a transparent overlay shape of a

horizontal “V” applied on the top of the CumSum

chart. Its origin point is placed on the last data point
of the CumSum chart and its center line is
horizontal.
• If all of the data points stay inside the V-mask, we

consider the process is in statistical control.
1052
V-Mask
Origin point on the V-

mask, which is applied
Upper boundary right on the last data point
of V-mask of the CumSum chart.

The center line of the V-
mask, which should be
parallel to the x axis
(time).
Lower boundary
of V-Mask
If any data points in the CumSum chart to the left of the origin point
are outside the V-mask, the process is considered out of statistical
control.
1053
Two-Sided CumSum Equations
• Two-Sided CumSum
ci  ci 1  ( xi  T ) i0
• Data Point:
ci  0 i0
s
• V-Mask Slope: k

m
s
• V-Mask Width at the Origin Point: 2h
m
where x i is the mean of the ith subgroup;

T is the process target;
σ is the estimation of process standard deviation;
m is the subgroup size.
1054
One-Sided CumSum
• We can also use two one-sided CumSum charts to

detect the shift of the process mean from the
process target.
• The upper one-sided CumSum detects the upward

shifts of the process mean.
• The lower one-sided CumSum detects the

downward shifts of the process mean.
1055
One-Sided CumSum Equations
• One-Sided CumSum
ci  ci 1  ( xi  T ) i0
• Data Point:
ci  0 i0
• Center Line: T



• Upper Control Limit: ci  max 0, x i  (T  k )  ci 1


• Lower Control Limit: ci

 max 0, (T  k )  x i  ci1 
T is the process target;
k is the slope of the lower boundary of the V-mask.
1056
Use Minitab to Plot a CumSum Chart
• Data File: “CumSum” tab in “Sample Data.xlsx”
• Steps in Minitab to plot CumSum charts
1) Click Stat → Control Charts → Time Weighted Charts →
CUSUM.
2) A new window named “CUSUM Chart” appears.
3) Select “weight” in the box below “All observations for a chart

4) Select “hour” as the “Subgroup sizes.”
5) Enter “8.1” as “Target.”
6) Click the button “CUSUM Options” and another window
named “CUSUM Chart – Options” pops up.
7) Click the tab “Plan/Type.”
8) Select the radio button “Two-sided (V-mask).”
9) Click “OK” in the window “CUSUM Chart Options.”
10) Click “OK.”
11) The CUSUM chart appears in the newly-generated window.
1057

1058

1059
CumSum Chart:
The process is in
control since all of the
data points fall

between the two arms
of the V-mask.
1060
1061

5.2.9 EWMA Chart
EWMA Chart
• The EWMA chart (Exponentially-Weighted Moving

Average Chart) is a control chart monitoring the
exponentially-weighted average of previous and
present subgroup means.

• The more recent data get more weight than more
historical data.
• It detects the shift of the process mean from the

process target over time.
• The underlying distribution of the EWMA chart is

normal distribution.
1062
EWMA Chart Equations
• EWMA Chart
• Data Point: zi   x i  (1   ) zi 1 where 0    1
• Center Line: X

s   
• Control Limits: X  k    1 (1   ) 2i

n  2 

λ and k are user-defined parameters to calculate the EWMA
data points and the control limits.
1063
Use Minitab to Plot an EWMA Chart
• Data File: “EWMA” tab in “Sample Data.xlsx”

• Steps in Minitab to plot EWMA charts
1) Click Stat → Control Charts → Time Weighted Charts →
EWMA.
2) A new window named “EWMA Chart” appears.
3) Select “Gap” in the box below “All observations for a chart

4) Select “Sample” as the “Subgroup sizes.”
5) Enter “0.5” as “Weight of EWMA.”
6) Click “OK.”
7) The EWMA chart appears in the newly-generated window.
1064

1065

1066
EWMA Chart:
One data point falls

beyond the upper
control limit and we

conclude that the
process is out of
control.
needed to discover the
root cause for the
outlier.
1067
1068
Control Methods
• There are many control methods available to keep

the process stable and minimize the variation.
• Most common control methods:

• SPC (statistical process control)

• 5S method
• Kanban
• Poka-Yoke (mistake proofing).
1069
SPC
• SPC (Statistical Process Control) is a quantitative

control method to monitor the stability of the process
performance by identifying the special cause
variation in the process.
• It uses control charts to detect the unanticipated

changes in the process.
• Which control chart to use depends on:
• Whether the data are continuous or discrete
• How large the subgroup size is
• Whether the subgroup size is constant
• Whether we are interested in measuring defects or
defectives
• Whether we are interested in detecting the shifts in the
process mean.
1070
5S
• 5S is a systematic approach of cleaning and

organizing the workplace.
• Seiri (sorting)
• Seiso (shining)
•

Seiketsu (standardizing)
• Shisuke (sustaining)
1071
Kanban
• A Kanban system is a demand-driven system.
• The customer demand is the signal to trigger or pull

the production.

• Products are made only to meet the immediate
demand. When there is no demand, there is no
production.
• It was designed to reduce the waste in inventory

and increase the speed of responding to immediate
demand.
1072
Poka-Yoke
• Poka-yoke is a mechanism to eliminate the defects

as early as possible in the process.
• Contact Method
• Use of the shape, color, size, or any other physical attributes

of the items
• Constant Number Method

• Use of a fixed number to make sure a certain number of
motions are completed
• Sequence Method
• Use of a checklist to make sure all the prescribed process
steps are followed in the right order.
1073
1074
Control Chart Calculations Summary

1075
Control Chart Constants

1076
Unnatural Patterns
• If there are unnatural patterns in the control chart of
a process, we consider the process out of statistical
control.
• Typical unnatural patterns in control charts:

• Outliers

• Trending
• Cycling
• Auto-correlative
• Mixture.
• A process is in control if all the data points on the

control chart are randomly spread out within the
control limits.
1077
Western Electric Rules
• Western Electric Rules are the most popular

decision rules to detect unnatural patterns in the
control charts. They are a group of tests for special
causes in a process.

• The area between the upper and lower control limits
is separated into six subzones.
• Zone A: between two and three standard deviations from the
center line
• Zone B: between one and two standard deviations from the
center line
• Zone C: within one standard deviation from the center line.
1078
Observation Upper control limit
A
B
C
C
B

A
Lower control limit

• Zone A: three sigma zone
• Zone B: two sigma zone Center line
• Zone C: one sigma zone
If a data point falls onto the dividing line of two consecutive

zones, the point belongs to the outer zone.
1079
• Test 1: 1 point more than 3 standard deviations from

the center line (i.e., 1 point beyond zone A)

1080
• Test 2: 9 points in a row on the same side of the

center line

1081
• Test 3: 6 points in a row steadily increasing or

steadily decreasing

1082
• Test 4: 14 points in a row alternating up and down

1083
• Test 5: 2 out of 3 points in a row at least 2 standard

deviations from the center line (in zone A or beyond)
on the same side of the center line

1084
• Test 6: 4 out of 5 points in a row at least 1 standard

deviation from the center line (in zone B or beyond)
on the same side of the center line

1085
• Test 7: 15 points in a row within 1 standard deviation

from the center line (in zone C) on either side of the
center line

1086
• Test 8: 8 points in a row beyond 1 standard

deviation from the center line (beyond zone C) on
either side of the center line

1087
Next Steps
• If no data points fail any tests for special causes, the

process is in statistical control.
• If any data point fails any tests for special causes,

the process is unstable and we will need to

investigate the observation thoroughly to discover
and take actions on the special causes leading to
the changes.
• Process stability is the prerequisite of process

capability analysis.
1088
5.2.12 Subgroups & Sampling
1089
Subgroups
• Rational subgrouping is the basic sampling scheme in
SPC (Statistical Process Control).
• When sampling, we randomly select a group (i.e., a

subgroup) of items from the population of interest.

• The subgroup size is the count of samples in a subgroup.
It can be constant or variable.
• Depending on the subgroup sizes, we select different

control charts accordingly.
1090
Impact of Variation
• The rational subgrouping strategy is designed to
minimize the opportunity of having special cause
variation within subgroups.
• If there is only random variation (background noise)

within subgroups, all the special cause variation would

be reflected between subgroups. It is easier to detect the
out-of-control situation.
• Random variation is inherent and indelible in the

process. We are more interested in identifying and taking
actions on special cause variation.
1091
Frequency of Sampling
• The frequency of sampling in SPC depends on
whether we have sufficient data to signal the changes in
a process with reasonable time and costs.
• The more frequently we sample, the higher the costs

sampling may trigger.

• We need the subject matter experts’ knowledge on the
nature and characteristics of the process to make good
decisions on sampling frequency.
1092
5.3 Six Sigma Control Plans
1093

5.2.2 I-MR Chart
5.2.3 Xbar-R Chart
5.2.4 U Chart
5.2.5 P Chart
5.2.6 NP Chart
5.2.7 X-S chart
5.2.8 CumSum Chart
5.2.9 EWMA Chart
1094
5.3.1 Cost Benefit Analysis
1095
What is Cost-Benefit Analysis?
• The cost-benefit analysis is a systematic method

to assess and compare the financial costs and
benefits of multiple scenarios in order to make
sound economic decisions.

• A cost-benefit analysis is recommended to be done
at the beginning of the project based on estimations
of the experts from the finance team in order to
determine whether the project is financially feasible.
• It is recommended to update the cost-benefit

analysis at each DMAIC phase of the project.
1096
Why Cost-Benefit Analysis?
• In the Define phase of the project, the cost-benefit

analysis helps us understand the financial feasibility
of the project.
• In the middle phases of the project, updating and

reviewing the cost-benefit analysis helps us
compare potential solutions and make robust data-
driven decisions.
• In the Control phase of the project, the cost-benefit

analysis helps us track the project’s profitability.
1097
Return on Investment
• Return on investment (also called ROI, rate of

return, or ROR) is the ratio of the net financial
benefits (either gain or loss) of a project or
investment to the financial costs.

TotalNetBe nefits
ROI  100%
TotalCosts
where
TotalNetBe nefits  TotalBenef its  TotalCosts
1098
Return on Investment (ROI)
• The return on investment is used to evaluate the

financial feasibility and profitability of a project or
investment.
• If ROI < 0, the investment is not financially viable.

• If ROI = 0, the investment has neither gain nor loss.
• If ROI > 0, the investment has financial gains.
• The higher the ROI, the more profitable the project.
1099
Net Present Value (NPV)
• The net present value (also called NPV, net

present worth, or NPW) is the total present value of
the cash flows calculated using a discount rate.
NetCashFlowt
NPV 

(1  r ) t
Where
NetCashFlowt is the net cash flow happening at time t;

r is the discount rate;
t is the time of the cash flow.
1100
Cost Estimation
• Examples of costs triggered by the project:

• Administration
• Asset
• Equipment
• Material
•

Delivery
• Real estate
• Labor
• Training
• Consulting.
1101
Benefits Estimation
• Examples of benefits generated by the project:

• Direct revenue increase
• Waste reduction
• Operation cost reduction
• Quality and productivity improvement
•

Market share increase
• Cost avoidance
• Customer satisfaction improvement
• Associate satisfaction improvement.
1102
Challenges in Cost and Benefit Estimation
• Different analysts might come up with different cost

and benefit estimations due to their subjectivity in
determining:
• The discount rate
• The time length of the project and its impact
•

Potential costs of the project
• The tangible/intangible benefits of the project
• The specific contribution of the project to the relevant
financial gains/loss.
1103
5.3.2 Elements of Control Plans
1104
Control Plans
• The control plans ensure that the changes
introduced by a Six Sigma project are sustained
over time.
• Benefits of the Control phase:

• Methodical roll-out of changes including standardization of

processes and work procedures
• Ensure compliance with changes through methods like auditing
and corrective actions
• Transfer solutions and learning across the enterprise
• Plan and communicate standardized work procedures
• Coordinate ongoing team and individual involvement
• Standardize data collection and procedures
• Measure process performance, stability, and capability
• Plan actions that mitigate possible out-of-control conditions
• Sustain changes over time.
1105
What is a Control Plan?
• A control plan is a management planning tool to
identify, describe, and monitor the process
performance metrics in order to meet the customer
specifications steadily.
• It proposes the plan of monitoring the stability and

capability of inputs and outputs of critical process
steps in the Control phase of a project.
• It covers the data collection plan of gathering the

process performance measurements.
• Control plans are the most overlooked element of

most projects. It is critical that a good solution be
solidified with a great control plan! 1106
Control Plan Elements
• Control Plan
• The clear and concise summary document that details key
process steps, CTQs metrics, measurements, and corrective
actions.
• Standard Operating Procedures (SOPs)
• Supporting documentation showing the “who does what,
when, and how” in completing the tasks.
• Communication Plan

• Document outlining messages to be delivered and the target
audience.
• Training Plan
• Document outlining the necessary training for employees to
successfully perform new processes and procedures.
• Audit Checklists
• Document that provides auditors with the audit questions they
need to ask.
• Corrective Actions
• Activities that need to be conducted when an audit fails.
1107
Control Plan
• The control plan identifies critical process steps that have
significant impact on the products or services and the
appropriate controls mechanisms.
• The control plan includes measurement systems that

monitor and help manage key process step performance.

• Specified limits and targets of the performance metrics
are clearly defined and communicated.
• Sampling plans to collect the measurements are

declared:
• How many samples are needed?
• How often do we need to sample?
• Where should we sample? www.LeanSigmaCorporation.com
1108
Control Plan
Key processes and Critical information regarding key

process steps are measurements is documented and clarified
identified

d
1109
Control Plan
Other key measurement
Measurements are
information is documented:
clearly defined with
sample size, measurement
equations Customer
frequency, people responsible
specifications
for the measurement, etc.
are declared

1110
Control Plan
Where will this measurement or Control plans identify the
report be found? Good control mitigating action or corrective
plans provide linking actions required in the event the
information or other report measurement falls out of spec or
reference information. control. Responsible parties are
also declared.

1111
1112

Control Plan Example
Standard Operating Procedures (SOPs)
• Standard Operating Procedures (SOPs) are
documents that focus on process steps, activities,
and specific tasks required to complete an operation.
• SOPs should not be much more than two to four pages.

• SOPs should be written to the user’s level of required
detail and information.
• The level of detail is dependent on the position’s required skills and
training
• Good SOPs are auditable, easy to follow, and not difficult

to find.
• Auditable characteristics are: observable actions and countable
frequencies. Results should be evident to a third party (compliance
to the SOP must be measurable).
1113
SOP Elements
• SOPs are intended to impart high value information in
concise and well-documented manner.
• SOP Title and Version Number:
• Provide a title and unique identification number with version
information.
• Date:

• List the original creation date; add all revision dates.
• Purpose:
• State the reason for the SOP and what it intends to
accomplish.
• Scope:
• Identify all functions, jobs, positions, and/or processes
governed or affected by the SOP.
1114
SOP Elements
• Responsibilities:
• Identify job functions and positions (not people) responsible for
carrying out activities listed in the SOP.
• Materials:
• List all material inputs: parts, files, data, information,
instruments, etc.
• Process Map:

• Show high level or level two to three process maps or other
graphical representations of operating steps.
• Process Metrics:
• Declare all process metrics and targets or specifications.
• Procedures:
• List actual steps required to perform the function.
• References:
• List any documents that support the SOP.
1115
1116

SOP Template
Communication Plans
• Communication plans are documents that focus on planning
and preparing for the dissemination of information.
• Communication plans organize messages and ensure that the

proper audiences receive the correct message at the right
time.
• A good communication plan identifies:

• Audience
• Key points/message
• Medium (how the message is to be delivered)
• Delivery schedule
• Messenger
• Dependencies and escalation points
• Follow-up messages and delivery mediums.
• Communication plans help develop and execute strategies for

delivering changes to an organization.
1117
Communication Plan Template

1118
Training Plans
• Training plans are used to manage the delivery of training for
new processes and procedures.
• Most GB or BB projects will require changes to processes and/or

procedures that must be executed or followed by various
employees.
• Training plans should incorporate all SOPs related to performing

new or modified tasks.
• Training plans use and support existing SOPs and do not

supersede them.
• Training plans should include logistics:

• One-on-one or classroom
• Instruction time
• Location of training materials
• Master training reference materials
• Instructors and intended audience
• Trainee names.
1119
Training Plan Template

1120
Audits
• What is an audit?
• ISO 9000 defines an Audit as “a systematic and
independent examination to determine whether quality
activities and related results comply with planned
arrangements and whether these arrangements are
implemented effectively and are suitable to achieve

objectives.”
• Audits are used to ensure actions, processes,

procedures, and other tasks are performed as
expected.
1121
Audit Guidelines
• Audits should be directed by managers, supervisors,
and other accountable positions.
• An audit’s purpose must be well-defined and executed

by independent unbiased personnel.
• Auditors must:

• Be qualified to perform their tasks
• Attend and successfully complete an internal auditing
training session
• Be able to identify whether or not activities are being
followed according to the defined SOP
• Base conclusions on facts and objective evidence
• Use a well documented audit checklist.
• Audits should confirm compliance or declare non-

compliance.
1122
Audit Checklists
• Auditors should review the SOPs before preparing
checklists or ensure that existing checklists properly
reference SOPs.
• Audit checklists:
• Serve as guides for identifying items to be examined

• Are used in conjunction with understanding of the procedure
• Ensure a well-defined audit scope
• Identify needed facts during audits
• Provide places to record gathered facts.
• Checklists should include:
• A review of training records
• A review of maintenance records
• Questions or observations that focus on expected behaviors
• Questions should be open-ended where possible
• Definitive observations yes/no, true/false, present/absent, etc.
1123
Audit Checklist Template

1124
5.3.3 Response Plan Elements
1125
What is a Response Plan?
• A response plan should be a component of as
many control plan elements as possible.
• Response plans are a management planning tool to

describe corrective actions necessary in the event of
out-of-control situations.

• There is never any guarantee that processes will
always perform as designed. Therefore, it is wise to
prepare for occasions when special causes are
present.
• Response plans help us mitigate risks and, as

already mentioned, should be part of several control
plan elements. 1126
Response Plan Elements
• Action triggers
• When do we need to take actions to correct a problem or issue?
• Action recommendation
• What activities are required in order to solve the problem in the
process? The action recommended can be short-term (quick fix) or
long-term (true process improvement).

• Action respondent
• Who is responsible for taking actions?
• Action date
• When did the actions happen?
• Action results
• What actions have been taken?
• When were actions taken?
• What are the outcomes of the actions taken?
1127

Note the response plan element
in this control plan template
1128

Note the response plan element in
this communication plan template
1129

Note the
response plan
element in
this audit
checklist
1130

Green Belt MTB 2.5

Uploaded by

Copyright:

Available Formats

Green Belt MTB 2.5

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Green Belt MTB 2.5

Uploaded by

Copyright:

Available Formats

© Lean Sigma Corporation

Lean Six Sigma Training - MTB

Lean Six Sigma Training - MTB

Lean Six Sigma Training - MTB

Lean Six Sigma Training - MTB

• Sigma level measures how many “sigma” there are

Lean Six Sigma Training - MTB

• A process operating at 1 sigma has a defect rate of

• This means that the process will generate

Lean Six Sigma Training - MTB

• A higher sigma level means a lower defect rate.

• Let us take a look at the defect rates of processes at

Lean Six Sigma Training - MTB

• Let us take a look at processes operating at 3 sigma.

Lean Six Sigma Training - MTB

Lean Six Sigma Training - MTB

• Even at 1% defect rate, some processes would be

• So what is Six Sigma?

• In order to achieve Six Sigma, you need to improve your

Lean Six Sigma Training - MTB

• Accomplishing these two process improvements (along

• DMAIC is the systematic methodology prescribed to

Lean Six Sigma Training - MTB

Lean Six Sigma Training - MTB

Lean Six Sigma Training - MTB

• Starting from the late 1980s, Motorola extensively applied Six

• In 1988, Motorola was recognized with the prestigious Malcolm

• As GE connected Six Sigma results to its executive

Lean Six Sigma Training - MTB

• The two work hand in hand, benefitting each other.

Lean Six Sigma Training - MTB

• Lean and its popularity began to form and gain significant

• The concepts and methodology of Lean, however, were

Lean Six Sigma Training - MTB

• The Six Sigma approach to problem solving uses a

• A transfer function is a mathematical expression of

Lean Six Sigma Training - MTB

• Y = f(x) is the relational transfer function that is used

• It is absolutely critical that you understand and

Lean Six Sigma Training - MTB

• Let us look at a simple example of a pizza delivery

Lean Six Sigma Training - MTB

• With this approach, all potential

Lean Six Sigma Training - MTB

Lean Six Sigma Training - MTB

• The Y = f(x) equation is a very powerful concept and

• Six Sigma follows a methodology that is

Lean Six Sigma Training - MTB

• The 5 phases of DMAIC:

• Define and quantify the

• Baseline process, validate measures,

Lean Six Sigma Training - MTB

Improve • Develop solutions

• Define is arguably the most important aspect of any Six Sigma

Lean Six Sigma Training - MTB

• Define Phase Tools and Deliverables

Lean Six Sigma Training - MTB

• The goal of the Measure phase is to gather baseline

Lean Six Sigma Training - MTB