Dmaic - Book of Knowledge - Green Belt11ea60e023f

D M A I C
Classroom Session Agenda–Day 1 1
Total Topic Step

Min
60 Introductions and Expectations N/A
20 Meeting Skills N/A
20 Define A,B and C Review and CAP A,B,C
10 Break
10 Measure Overview 1,2,3
90 Measure 1 Review CTQ Tools 1
60 Measure 1–CTQ Tools continued 1

10 Break
45 Measure 2–Performance Standards 2
5 Break
120 Minitab Tutorial and Graphical
Analysis various
© GE Capital, Inc., 2000
DMAIC GB A TX PG V4.2.0
D M A I C
Classroom Session Agenda–Day 2* 2
Total Topic Step

Min
30 Day 1–General Review and Homework A,B,C,1,2
30 Minitab–continued various
15 Measure 3–Overview 3
15 Break
45 Measure 3–Data Collection Plan 3
75 Measure 3–Sampling 3
60 Measurement System Analysis (MSA) 3

15 Break
15 Analyze Overview 4,5,6
150 Analyze 4–Establish Process Capability 4
* Elevator Speeches will be shared throughout the

day
D M A I C
Classroom Session Agenda–Day 3* 3
Total Topic Step

Min
30 Day 2–General Review various
15 Analyze 5–Define Performance Objective 5
15 Break
150 Analyze 6–Identify Variation Sources 6
30 Analyze 6–continued 6
15 Improve Preview 7,8,9
30 Improve 7 and 8–Design Of Experiment 7,8
15 Break
30 Improve 9–Statistical Tolerancing 9
15 Control Preview 10,11,12
60 Control 12–Control Charts 12
* Elevator Speeches will be shared throughout the

day
D M A I C
D M A I C
Assessment Strategy
Introductions
1
The Learning Objectives are divided in 3 categories:

1. Classroom
2. Project, and
3. Test
It is expected that a Green Belt be proficient in all learning

objectives, however, only those designated Test represent
the content of the certification exam.
Each phase of the DMAIC cycle is listed below and the
learning objectives have classified into the 3 categories
within each phase.
DMAIC GB B TX PG V 4.2.0
D M A I C
Overview–Classroom Learning Objectives 2
1 Explain the benefits of Six Sigma to GE’s business.

Compare and contrast Six Sigma’s process improvement approach to quality with traditional
defect prevention strategies (i.e., inspection and testing).
Identify the “vital few” CTQs that apply to all GE customers: responsiveness; marketplace
competitiveness; on time, accurate and complete deliverables; and product/service technical
performance.
Explain the relationship between increasing levels of process complexity and quality
improvement results.
2 Describe the Six Sigma Methodology for quality improvement.

Define the term “sigma” (standard deviation) as it relates to the sigma capability (z value) of a
business or manufacturing process.
Recognize a Six Sigma level of quality (i.e., 99.99966% probability that defects will not be
passed on to the customer).
Define key Six Sigma terms and acronyms, including CTQ, opportunity, defect, DPMO, and Six
Sigma capability (Z value).
Explain the Master Back Belt (MBB), Black Belt (BB) and Green Belt (GB) roles in Six Sigma.
Describe Six Sigma’s focus on repeatable processes.
Describe Six Sigma’s focus on inputs (X’s) over outputs (Y’s) using the formula Y=f (X).
Describe the statistical objective of Six Sigma (i.e., reduce process variation).
Describe the relationship between DPMO and process capability (i.e., as DPMO goes down,
process capability goes up.
Describe the financial benefits of Six Sigma to GE.
D M A I C
Overview–Project Learning Objectives 3
1. Determine if DMAIC is the right strategy by identifying the

conditions under which the DFSS methodology would be more
appropriate.
Compare and contrast the DFSS design methodologies to
DMAIC.
D M A I C
Overview–Test Learning Objectives 4
1 Describe the 5 phases of DMAIC, including the purpose, tools,

and outputs for each phase.
D M A I C
Define–Project Learning Objectives 5
1 Identify project CTQs.

Define “CTQ” (Critical to Quality Characteristic).
Identify customer(s) in a quantifiable way.
– Recognize the components of a process (i.e., supplier, input(s), sub-
process, output(s), customer(s).
– Distinguish between internal and external customers.
Compile and evaluate customer CTQ data.
– Distinguish between customer driven CTQs and process driven CTQs.
– Recognize sources of existing customer data.
– Assess customer requirements and expectations.
– Recall the vital few customer CTQs.
– Analyze the voice of the customer and it’s impact on CTQ data.
– Translate customer needs into requirements (CTQs).
D M A I C
Define–Project Learning Objectives (continued) 6
2 Use a process/product drill-down tree to: define the limits of a

project (project bounding); clarify what the project is and is not;
identify other areas for improvement.
Create a process/product drill-down tree.
Integrate measurements to clarify areas needing improvement.
Given an example of a process/product drill down tree, identify
viable Six Sigma projects.
D M A I C
3 Develop a team charter.

Describe the purpose of a charter.
Identify the five major elements of a charter.
Define the business case for a project in terms of its potential benefits, the
consequences of not doing it, its relationship to other activities, and its fit with
business initiatives/target.
Develop a problem statement.
Describe the customer’s pain.
Identify key considerations and potential pitfalls to consider when developing a
problem statement.
Develop a SMART goal statement (specific, measurable, attainable, relevant, time
bound).
Assess the scope of the project.
Identify the 8 steps for bounding a project.
Define project milestones.
Select a project team and define team roles.
Identify team roles and responsibilities.
Evaluate a proposed Six Sigma project.
Recognize characteristics of a “good” project.
Recognize characteristics of a “bad” project.
D M A I C
4 Map a business process.

Describe the goal of process mapping.
Identify the components of a process map (COPIS).
Describe the steps involved in creating a process map.
Define and name a process.
Given a business process, use brainstorming and
storyboarding techniques to: identify its outputs, customers,
suppliers, and inputs; identify customer requirements for
primary outputs; and identify process steps
D M A I C
5 Obtain approval for a Six Sigma project.

Identify the steps in the project approval process.
Enter a project into QPT.
D M A I C
Define–Test Learning Objectives 10
1 Recognize the components of the 12 Step Process and how they

may be applied to a Six Sigma project.
2 Describe the purpose of the define phase and it’s key
deliverables: CTQs, team charter, and process map.
Identify the five key objectives of the Define Phase.
3 Describe the CAP tools and their connection to Six Sigma.
D M A I C
Measure1–Classroom Learning Objectives 11
1 Describe and define the deliverables of Step 1.
D M A I C
Measure 1–Project Learning Objectives 12
1 Identify the project Y.

Identify the tools that may be used to select the relevant CTQ
or Y on which to focus.
Explain the purpose Quality Function Deployment, Process
Map, and FMEA tools have.
Define performance standards for Y including specification
limits as well as defect and opportunity definitions.
D M A I C
Measure 1–Project Learning Objectives (continued) 13
2 Select CTQ characteristics.

Select the Critical to Quality (CTQ) characteristic to be
improved in a project.
Narrow the focus of a project to an actionable level.
Establish the project team and gained consensus on the project
definition.
3 Select and apply appropriate tools to narrow the focus of a Six

Sigma project by identifying key areas for improvement.
Identify tools that may be used to narrow the focus of a project,
including Process Map, and FMEA.
Recognize the purpose and benefits of each tool.
D M A I C
4 Use Process Mapping to identify potential breakdowns, rework loops, and

sources of variation in a process.
Use the C.O.P.I.S. model to illustrate a customer focused process.
Identify the elements of a process (input, mechanism, control, output,
process boundary).
Identify and distinguish between internal and external process controls.
Recognize the purpose and benefits of process mapping.
Recognize the three types of process maps.
Describe the process mapping process, including the following steps:
Determine the scope.
Determine the steps in the process.
Arrange the steps in order.
Recognize ISO 9000 standard symbols for process mapping.
Validate a process map.
Evaluate a process map.
D M A I C
5 Use a Failure Modes and Effects Analysis to identify the potential

failure modes of a process or product.
Recognize the purpose and benefits of FMEA.
Describe how FEMA works.
Describe FMEA, including preparation, process, and
improvement steps.
Define the terms “failure mode,” “cause,” and “effect,” as they
relate to FMEA, and recognize examples of each.
Assign degree of severity, likelihood of occurrence, and ability
to detect ratings, and calculate a risk priority number (RPN).
Complete an FMEA form.
– Recognize when and by whom an FMEA is prepared, updated, and
completed.
D M A I C
6 Conduct a test-retest study and analyze the results.

Describe the purpose and procedure for conducting a test-
retest study.
Plot and test-retest study data.
Use descriptive statistics to evaluate test-retest study data.
7 Establish a Data Collection Plan for a Six Sigma project.

Describe the purpose and benefits of a Data Collection Plan.
Write a data collection strategy.
Define a clear strategy for collecting reliable data efficiently.
D M A I C
Measure–Test Learning Objectives 17
1 Recall the DMAIC 12 Step process, and distinguish between characterization

phases (DMA) and optimization phases (IC).
Define product characterization.
Define process optimization.
2 Recognize how statistics can be applied to the problem solving process.

Define the terms “precision” and “accuracy” as they relate to a Six Sigma
process.
Relate precision/variation and accuracy/mean to quality and customer
satisfaction.
State the goal of Six Sigma in statistical terms.
Define the term “Upper Specification Limit (USL).”
Define the term “Lower Specification Limit (LSL).”
Define the term “target (T).”
Define “σ.”
D M A I C
3 Explain how statistics can be used to solve problems.

Identify project variables using the formula Y = f (X1,…..,Xn ).
Describe the relationship between any dependent variable (Y)
and independent variables (X).
Explain how the shape, mean, and standard deviation
characterize a process.
Express the capability of a process in terms of a standard
measure (z-value).
Define hidden factories and how capability impacts cycle time.
4 Identify the key deliverables of the Measure phase of DMAIC.
D M A I C
5 Relate and apply the Quality Function Deployment (QFD) process to Six
Sigma.
Explain the purpose of QFD.
Describe the phases of QFD.
Explain QFD flowdown for product and service applications.
Generate/build a House of Quality (Product Planning Chart).
Identify what the customer wants (the “what’s”).
Identify the functions or processes that impact customer wants (the how’s).
Evaluate the impact of each function/process on customer wants.
Calculate the overall magnitude of the impact each function/process has on
customer wants (prioritize actions).
Analyze and diagnose a completed House of Quality.
Describe other QFD applications.
Determine when QFD is appropriate to use.
Recognize QFD pitfalls.
Describe an example of QFD from GE Medical Systems.
D M A I C
7 Define Performance Standards for a Six Sigma project.

Describe the purpose and characteristics of a performance standard.
Describe the purpose and characteristics of an operational definition.
Define the term “defect.”
Given an example of a problem or process, write an operational definition.
Describe and distinguish between continuous and discrete data.
Recognize the components of a performance standard, including
product/process characteristic, measure, target value, specification limits,
and defect.
Given a CTQ type, identify performance standard sources and
discrete/continuous data measurement methods.
Given an example, define the measurable characteristic, determine whether
it is continuous or discrete, determine the specification limit if applicable,
identify a defect.
D M A I C
9 Establish the accuracy of the measurement system and the data

(Analyze the measurement system).
Describe measurement as a process that includes
Measurement, Analysis, Improvement, and Control phases.
Describe measurement as a system that includes operators,
gages, and environment.
Define the terms “(gage) resolution,” “precision,” “accuracy,”
and “bias” as used in Measurements System Analysis (MSA).
Using the MSA checklist, document the existing measurement
system.
Recognize the sources of variation in a measurement system.
D M A I C
10 Conduct a Gage R&R study and analyze the results.

Develop and implement a Data Collection Plan to collect Gage
R&R study data.
Describe equipment and appraiser sources of variation.
Describe the total R&R variation in terms of Reproducibility
(AV) and Repeatability (EV).
Set up, collect, and enter data into a Minitab data sheet.
Calculate both the appraiser variation (reproducibility) and
equipment variation (repeatability).
Describe the concepts of stability and linearity in gage studies.
Compare R&R variation to the tolerance (specification window).
Create graphs and charts (ANOVA method) to analyze study
results.
Recall and apply rules of thumb (guidelines) for analyzing R&R
study results.
D M A I C
11 Describe and define the use of gage R& R for discrete data.
Describe the use of the attribute R & R spreadsheet for discrete
data.
12 Recall and list the deliverables of the Measure phase of

DMAIC.
D M A I C
Analyze–Classroom Learning Objectives 24
1 Perform hypothesis testing for a continuous Y and discrete X.

Determine process stability with run charts and other tools
Determine the data shape with histograms, normal probability
plots, and Anderson-Darling tests.
Select and use the appropriate tool to determine the p-value.
Determine whether or not to accept the null hypothesis or the
alternative hypothesis.
2 Perform hypothesis testing for a discrete Y and discrete X.

Use chi-square testing to determine the goodness-of-fit and as
a test of independence.
Based on the chi-square test, determine whether or not to
accept the null hypothesis or the alternative hypothesis.
D M A I C
Analyze–Classroom Learning Objective (continued) 25
3 Perform hypothesis testing for a continuous Y and continuous X.

Use a scatterplot to determine correlations between variables.
Use a linear regression analysis to quantify correlations and
predict values.
Determine process stability with run charts and other tools.
Determine whether or not to accept the null hypothesis or the
alternative hypothesis.
Describe the use of multiple regression for this type of data.
Describe the implications of multiple regression in statistical
analysis.
D M A I C
Analyze–Project Learning Objectives 26
1 Use benchmarking to assist in developing project goals.

Describe the purpose of benchmarking.
Describe the uses of five different types of benchmarking, including
competitive benchmarking, product benchmarking, process benchmarking,
best practices benchmarking, strategic benchmarking, and parameter
benchmarking.
Apply benchmarking methodology to a variety of situations.
List potential sources of benchmarking data and how to access such
sources.
Describe the advantages and disadvantages of Internal, Competitive, and
Functional benchmarking and their relationship to Best Practices.
2 Set realistic and achievable defect reduction goals based on current baseline,
GE guidelines, benchmarking results, and the process entitlement.
Describe the methods used to set project goals.
Use GE Standards for defect reduction in combination with benchmark
results to determine project goals.
Use the process entitlement to validate the achievability of the project goals.
D M A I C
Analyze–Test Learning Objectives 27
3 Develop consensus within a project team on the acceptability of

the project goals.
Use a Cause & Effect (Fishbone) diagram to identify Xs that
may impact the Y that is important in a project and provide a
visual display of all possible causes of a specific problem.
Recognize the purpose and benefits of a Cause & Effect
diagram.
Write a problem statement.
Brainstorm categories appropriate to a problem.
Recognize the 4 Ps: policies, procedures, people and plant.
Brainstorm and analyze causes for each category to identify the
most likely cause(s) of a problem.
Determine which causes need to be verified with data.
D M A I C
4 Use a Pareto Chart to separate the vital few from the trivial many
in a process to determine where to focus improvement efforts.
Recognize the purpose and benefits of a Pareto Chart.
Describe the Pareto Principle.
Describe the steps involved in building a Pareto Chart:
Collect data.
Total results and arrange data in descending order.
Draw and label a Pareto Chart.
Analyze results.
Compare before and after Pareto Chart to evaluate
improvement effectiveness.
5 Describe process map analysis.

Describe value added/ non-value added analysis.
D M A I C
6 Identify variation sources.

Brainstorm a list of potential vital X’s.
Use a histogram to aid in determining variation, center, and shape of a
process.
Use a dot plot to aid in determining variation, center, and shape of a
process.
Use a box plot to aid in determining variation, center, and shape of a
process.
Use a run chart to determine process stability.
Define the terms population and sample and relate the two to each other.
Use statistical tests to validate sampling techniques.
Define the theoretical framework for hypothesis testing.
Define and follow the hypothesis testing protocol.
Define the terms null hypothesis and alternative hypothesis.
Develop the null hypothesis for your project.
Develop the alternative hypothesis for your project.
Define type I and type II errors in relation to hypothesis testing.
Define the relationship between the confidence interval and the p-value.
D M A I C
2 Apply statistical principles of the Standard Normal Probability Distribution to

predict the probability of a defect and process capability.
Use continuous data to describe a process by its average, standard deviation, and
normal curve.
Define the term “random variable.”
Interpret uniform, triangular, normal, and exponential distributions.
Relate probability to distribution curves.
Define the terms “mean” and “standard deviation” as they relates to a normal distribution
curve.
Recognize and distinguish between population and sample computational equations.
Use Minitab to calculate a mean and standard deviation.
Recognize the Descriptive Statistics tool as a method for validating calculations.
Calculate statistical measures of variation, including range, deviation, sum-of-square,
standard deviation, and coefficient of variation.
Calculate capability (Z value).
Perform basic statistic calculations using Minitab.
Describe the purpose and characteristics of Descriptive Statistics tools, including
Histogram, Dot Plot, Box and Whisker Plot, Run Chart.
Be able to distinguish a normal distribution from other common non-normal distributions.
D M A I C
Analyze–Test Learning Objectives (continued) 31
3 Characterize a process using discrete data.

Define the terms unit (U), opportunity (OP), and defect (D).
Recognize formulas for DPU, TOP, DPO, and DPMO.
Use Z tables to convert DPMO to “Z”.
Run and interpret a Minitab Product Report.
Compare and contrast Classical Yield (Yc), Throughput Yield
(YTP), and Rolled Yield (YRT).
Calculation the distribution of defects for a given DPU.
Calculate submitted, observed, and escaping defect levels.
Recall DPU application rules.
Determine how DPU controls Throughput Yield (YTP).
Explain how complexity impacts quality.
D M A I C
4 Use Process Centering strategies to perform a capabilities analysis.

Explain the concept of Process Centering.
Distinguish between special (assignable) and common (random) cause
variation.
Choose rational subgroups for proper sampling and analysis.
Differentiate between entitlement, short term process capability, and long
term process capability.
Interpret Minitab “hand calculations,” histogram, and box plots.
Calculate the long and short term standard deviation and Z-values.
Explain the general long term 1.5 Z shift.
Use Minitab Six Sigma Process Report to obtain short and long term
process capability measures - ZST, ZLT, ZbenchLT, Zshift, DPMOST,
DPMOLT.
Using capability measures and a 2x2 matrix, determine if there is a control
problem or a technology problem.
D M A I C
5 Determine process capability.

Define the term “process entitlement” as it relates to process capability.
Define and provide examples of common cause and special cause variation.
Define and use rational subgrouping of data.
Define shift and drift of a process.
Describe the components of variation.
Calculate variation for a given process.
Calculate the standard deviation for a process.
Calculate process capability.
Define the difference between long term and short term capability and their uses in a six sigma
project.
Define the use of the sum of squares and the standard deviation.
Use the universal equation for Z to calculate Z scores.
Define, derive, and use the Z-Bench for a process.
Relate and convert between the Z score and Defects per Million Opportunities.
Use data collected in the Measure phase to generate a process capability chart for a process.
Interpret the results generated by the process capability report to determine the short term and
long term process capability of a process.
Use process capabilities to compare your process with a benchmark process.
Determine whether the deficiencies in a process are due to control problems or technology
problems.
D M A I C
8 Describe the statistical analysis tools and process for

normal/non-normal data.
Describe the use of the normality test.
Describe the use of Mood’s median for non-normal data.
D M A I C
Improve–Project Learning Objectives 35
1 Characterize X’s as either operating parameters or critical

elements.
2 Develop a strategy for those X’s identified as operating

parameters.
Develop a mathematical model of a proposed solution.
Determine the best configuration or combination of X’s
3 Develop a strategy for those X’s identified as critical elements.

Optimize process flow issues.
Standardize the process.
Develop a practical solution.
Explain the needs and process to do screening experiments.
The implications for this in DOE is described.
D M A I C
4 Perform optimizing experiments in order to develop a proposed

solution.
Identify factors for optimization experiments.
Identify factor levels for optimizing experiments.
Design optimizing experiment to include randomization and
replication.
Perform experiments and collect data.
Analyze data with various tools including regression analysis.
D M A I C
5 Develop a proposed solution.

Interpret the outputs of various tools to determine the optimum
solution.
Determine if the optimum solution will meet project goals.
Present the proposed solution to management.
Use statistical tolerancing to define the control mechanisms for
implementation.
D M A I C
6 Establish operating tolerances.

Describe the concept of tolerances and describe an example of
this concept.
Describe the use of simulation and the use of Crystal Ball.
Describe and show an example of Crystal Ball.
7 Pilot the proposed solution.
D M A I C
Improve–Test Learning Objectives 39
1 Describe and define the deliverables of Steps 7 and 8 in Six

Sigma.
2 Describe and define the deliverables of Step 9 in Six Sigma.
D M A I C
Control–Project Learning Objectives 40
1 Develop/modify and implement Quality Plans.

Describe the purpose and characteristics of a Quality Plan.
Recognize the components of a Quality Plan.
Plan ongoing process controls, including monitoring and auditing strategies.
Explain the benefits of monitoring as compared to First Article Inspection
(FAI) and Information Management methods.
Determine what to monitor for a given process.
Determine the appropriate amount of monitoring data to collect, and how
frequently the monitoring should occur.
Recognize methods for detecting changes in a process.
Recognize the steps that should be taken if a process change is detected.
Explain the purpose of auditing.
Describe guidelines for effective auditing.
Compare and contrast manufacturing control methods.
Explain the purpose and process of variable data charting (SPC).
Explain the purpose and process of process management charting.
D M A I C
2 Develop and implement risk management strategies.

Explain the value of a risk management process.
Define the terms “risk” and “risk management” as they relate to DMAIC/Six
Sigma.
Determine when to use risk management.
Recognize different types of risks.
Recognize the steps involved in risk management, including identifying
risks, rating risks, abating risks, and executing risk management plans.
Recognize methods for identifying risks.
Describe methods and tools for rating risks.
Using the Probability of Occurrence Rating guide and Consequence of
Occurrence/Risk Impact chart, prioritize risks according to risk factor score.
Determine when and how to implement a Risk Abatement plan.
Integrate lessons learned from prior risk management efforts.
Describe the formal risk review process.
Explain the criticality of tracking and executing risk abatement plans.
D M A I C
3 Develop and implement mistake proofing strategies.

Recognize examples of mistake proofing.
Describe principles underlying the process of mistake proofing.
Recognize the difference between errors and defects.
Explain how defects originate.
Identify ten types of human error.
Recognize human error-provoking conditions.
Identify the three key mistake proofing techniques: shutdown, control, and
warning.
Distinguish between prediction/prevention and detection methods of mistake
proofing.
Recognize typical mistake proofing tools.
Describe the 5 steps involved in mistake proofing, including identifying
problems, prioritizing problems, finding the root cause, creating solutions,
and measuring results.
Recognize the advantages of mistake proofing as a proactive tool.
Explain how mistake proofing fits into the Six Sigma process.
D M A I C
4 Select and apply the appropriate Control Chart.

Recall and explain SPC Concepts, including controlled
variation, uncontrolled variation, common causes, and special
causes.
Recognize the five main uses of control charts.
Distinguish between variable, attribute, and process focused
control charts.
Determine control limits.
Distinguish between control limits and specification limits.
Recognize the four states of a process.
D M A I C
Control–Test Learning Objectives 44
1 Describe and define the deliverables of Steps 10 and 11 in Six Sigma.
2 Describe and define the deliverables of Step 12 in Six Sigma.
3 Develop and implement Variable Control Charts.

Describe the purpose of Statistical Process Control Charts.
Given a control chart, recognize when a special cause is acting on a process.
Recognize other types of variable control charts, including X Bar Chart, R Chart,
Individuals Chart, and Moving Range Chart.
State the five main uses of control charts.
Describe data collection and sampling techniques.
Establish and maintain control limits.
Select the appropriate Variable Control Chart.
Distinguish between control limits and specification limits.
Determine whether a process is “in control” or “out of control.”
Recognize Western Electric rules for identifying an out of control process.
Recognize and apply Minitab rules.
Build an Individuals and Moving Range chart.
Use knowledge of the process to eliminate or reduce assignable/special causes.
D M A I C
Control–Test Learning Objectives 45
4 Develop and implement Attribute Control Charts.

Describe the purpose of Attribute Control Charts.
Define and relate the terms “a defect” and “a defective.”
Recognize types of Attribute Control Charts, including C-
Charts, U-Charts, P-Charts, NP Charts.
Select the appropriate Attribute Control Chart.
Use Minitab to generate each type of Attribute Control Chart.
Determine the appropriate Attribute Chart subgroup size.
D M A I C
Glossary And Minitab Primer 46
1 Define the typical terms used in Six Sigma methodologies.
2 Describe the use of the Minitab software program.
D M A I C
Introduction Exercise (60 Minutes) 1

Introductions
Desired Outcome: Introduction of Classmates and CD Rom Learnings
What How Who Timing
Team Go around your tables and introduce All 2 mins.

Preparation yourself to your table-team members
(name and business only, at this
time.)
Exercise Discuss the importance of the All 7 mins.

DMAIC Step that your team has
been assigned.
Create a list of items that describe
the deliverables for the step and why
this step is important in the DMAIC
cycle. Use your knowledge from your
CD Rom Learnings.
Close Choose a spokesperson to 1 min.

Exercise report out
Report When asked by the instructor, 1 min.

Out each team member will introduce
themselves:
– Name
– How long you’ve been
with GE
– Your Title and what you do
– Previous Quality
Experience
– What Phase your project
is in
After last team member has
introduced himself/herself, the
spokesperson will discuss the
deliverables of the above-
mentioned exercise.
DMAIC GB C TX PG V 4.2.0
D M A I C
Classroom Session–Objectives 2
Share and get feedback on project deliverables for Define and

Measure phases
Review Define phase
Review Measure phase
Learn how to apply CTQ tools in the Measure phase of a
project
Learn how to use Minitab for statistical and graphical analysis
during a project
Learn knowledge and skills in Analyze phase and apply to
Capital Logistics case
Preview Improve phase
Learn how to apply Design of Experiment (DOE) in the Improve
phase of a project
Preview Control phase
Learn how to apply control charts in the Control phase of a
project
D M A I C
Iterative Process 3
Steps A,B,C
Steps
10,11,12 Steps 1,2,3
Steps 7,8,9 Steps 4,5,6
The Phases of the 12-Step Process Phase 4 (Improvement). This phase is usually initiated by
selecting those product performance characteristics which
Phase 1 (Define). This phase defines the project. It must be improved to achieve the goal. Once this is done, the
identifies customer CTQ’s and ties them to business characteristics are diagnosed to reveal the major sources of
needs. Further, it defines a project charter and the variation. Next, the key process variables are identified by
business process bounded by the project. way of statistically designed experiments. For each process
variable which proves to be significant, performance
Phase 2 (Measure). This phase is concerned with specifications are established.
selecting one or more product characteristics; i.e.,
dependent variables, mapping the respective process, Phase 5 (Control). This phase is related to ensuring that the
making sure the measurement system is valid, making
new process conditions are documented and monitored via
the necessary measurements, and recording the results.
statistical process control methods. After a “settling in”
period, the process capability would be reassessed.
Phase 3 (Analyze). This phase entails estimating the Depending upon the outcomes of such a follow-up analysis,
short and long-term process capabilities and it may be necessary to revisit one or more of the preceding
benchmarking the key product performance metrics. phases.
Following this, a gap analysis is often undertaken to
identify the common factors of successful performance;
i.e., what factors explain best-in-class performance.
D M A I C
Statistical Thinking 4
D M A I C
Practical Statistical Statistical Practical
Problem Problem Solution Solution
Problem Characterize the Root cause Verify critical
statement process analysis X’s and ƒ(x)
– Project Y – Stability – Critical X’s Change
– Magnitude – Shape Measure the process
– Impact – Center influence of the Control the
– Variation critical X’s on gains
Data Integrity the mean and – Risk
– MSA Capability variability analysis
– Brainstorm – ZBench ST & LT – Test – Control
potential X’s – Model plans
– Sampling plan – Estimate
Collect data
The Practical-To-Statistical-To-Practical
Transformation Process
D M A I C
The 12-Step Process 5
Step Description Focus Tools Deliverables

Define
A Identify Project CTQ’s Project CTQ’s
B Develop Team Charter Approved Charter
C Define Process Map High Level Process Map
Measure
1 Select CTQ Characteristics Y Customer, QFD, FMEA Project Y
2 Define Performance Y Customer, Blueprints Performance Standard for
Standards Project Y
3 Measurement System Y Continuous Gage R&R, Data Collection Plan & MSA
Analysis test/Retest, Attribute Data for Project Y
R&R
Analyze
4 Establish Process Y Capability Indices Process Capability for
Capabilities Project Y
5 Define Performance Y Team, Benchmarking Improvement Goal for
Objectives Project Y
6 Identify Variation Sources X Process Analysis, Prioritized List of all X’s
Graphical Analysis,
Hypothesis Tests
Improve
7 Screen Potential Causes X DOE-Screening List of Vital Few X’s
8 Discover Variable X Factorial Designs Proposed Solution
Relationships
9 Establish Operating Y, X Simulation Piloted Solution
Tolerances
Control
10 Define & Validate Y, X Continuous Gage R&R, MSA
Measurement System on Test/Retest, Attribute
X’s in Actual Application R&R
11 Determine Process Y, X Capability Indices Process Capability Y, X
Capability
12 Implement Process Control X Control Charts, Mistake Sustained Solution,
Proofing, FMEA Documentation
D M A I C
The “Basics” For Effective Meetings 6
DMAIC is a powerful methodology which will allow teams to

significantly improve the processes on which they are working.
However, in order for teams to efficiently use the tools and
techniques associated with DMAIC, they must be able to work
together effectively. Although not the focus of this class, we will
briefly focus on the minimum requirements for effective
meetings.
Further training in effective meeting skills is provided in
“Facilitating Teams through Change Projects” Training
D M A I C
Meeting Skills 7
The “Basics” For Effective Meetings
Roles:
Leader or Facilitator
Timekeeper
Scribe
Note Taker
Tools:
Agenda
Desired Outcomes
Ground Rules
Decision-Making Process
Brainstorming
Define These Roles For A Meeting Or Any Team Rules Of Brainstorming:

No judgment of ideas
During this training, you will practice these roles so that
you can use them in your project team meetings to Record all ideas
improve meeting effectiveness. No discussion of ideas during brainstorming
The effectiveness of any meeting can be improved simply

by having an agenda. That effectiveness can be improved
even further by including the desired outcomes. Desired
outcomes, or the goals you hope to achieve during the
meeting, provide additional focus and give clear purpose
for your meeting.
Ground rules, or the “code of conduct,” further increase

the effectiveness of your meeting by specifying the
behavior that is expected from all participants. For
ongoing teams, such as DMAIC teams, ground rules
should be established and agreed upon by all team
members early in the project.
D M A I C
Practice Meeting (5 minutes) 7
Desired Outcome: Name your team
What How Who Timing
Team Choose a facilitator, timekeeper All 30 secs.

Preparation and scribe
Determine Brainstorm a list of possible team Facilitator 4 mins.

Team Name names
Using N/3*, prioritize your
brainstormed list
Identify your top name
Close Choose a spokesperson to All 30 secs.

Exercise report out
* N/3 is a prioritization technique used to narrow a large list down to the top
priorities. The number of ideas (N) is divided by 3 (N/3) and team members get
that number of votes to identify their top choices. All votes should not be put on
one choice.
D M A I C
Define Review Module Objectives 1

Define Review Objectives
Review CAP Tools
Review steps in Define phase
Review and get feedback on Green Belt project deliverables for
Define phase
Review Steps in Define phase
Review and get feedback on Greenbelt project

deliverables for Define phase
DMAIC GB D TX PG V 4.2.0
D M A I C
2
First Things First (10 minutes)
Desired Outcome: Learning the components for successful change
What How Who Timing
Team Think of a change initiative that you Each 5 mins.

Preparation have experienced in the past in Team
which the change was unsuccessful. Member
Why did the change initiative fail?
Prepare List On a flip chart, list your group’s Facilitator 5 mins.

ideas for why change initiatives
fail. What are the common themes?
Be prepared to report findings.
D M A I C
Change Acceleration Process (CAP) 3
Techn
ical St
rate g y
rat egy
iza tional St
an
al/Org
Cultur
Change Initiative
(Target)
CAP Complements Technical Strategy With Cultural

Tools To Achieve The Change Initiative
Change can be depicted via the equation:
Q x A = E
Q = Quality of solution to be implemented
A = Acceptance of solution to be implemented
E = Effectiveness of the implemented solution
It is as important to have a strategy for developing

acceptance as it is to have a plan for implementing the
solution.
A Technical Strategy and a Cultural/Organizational

Strategy are both necessary in order to effectively
achieve your change target. CAP gives us tools to
develop the Cultural/Organizational or Influence
Strategy.
D M A I C
Cap Model 4
Leading Change
Creating
Creating A
A Shared
Shared Need
Need
Shaping A Vision
Mobilizing
Mobilizing Commitment
Commitment
Current Transition Improved
State State State
Making
Making Change
Change Last
Last
Monitoring Progress
Changing Systems And Structures
Leading Change Making Change Last

All implementation projects require a Champion who Once change is started, it endures and flourishes, and
sponsors the change if they are to be successful. learnings are transferred throughout the organization.
Creating A Shared Need Monitoring Progress

The reason to change, whether driven by threat or Progress is real; benchmarks are set and realized; indicators
opportunity, is instilled within the organization and widely are established to guarantee accountability.
shared through data, demonstration, demand, or diagnosis.
The need for change must exceed the resistance to change. Changing Systems And Structures
Management practices are used to complement and reinforce
Shaping A Vision
change.
The desired outcome of change is clear, legitimate, widely Did the themes that you’ve developed in your last exercise align
understood, and shared. with our CAP Model? This CAP Model helps us to focus on the
organizational changes aspect of projects.
Mobilizing Commitment
There is a strong commitment from key constituents to invest
in the change, make it work, and demand and receive
management attention.
D M A I C
Cap Tools And Six Sigma 5
Creating A Shaping A Mobilizing Making Change Monitoring

Shared Need Vision Commitment Last Progress
Define SIPOC In Frame / Out Threat vs.

Threat vs. Of Frame Opportunity Matrix
Opportunity Matrix GRPI
Project Scope ARMI
Contract
Measure
Analyze
Improve Making Change Making Change

Last Checklist Last Checklist
Control Pans Control Pans
Measures And Measures And
Rewards Rewards
Control
The above-mentioned tools are covered in the Tools

Section of the Self-Paced Workbook. For further training,
you can attend the week-long course sponsored by
Learning Services.
D M A I C
Homework 6
Desired Outcome: Identify Cap Tools which you will use in your project
What How Who Timing
Prepare a Identify at least one CAP Tool that Each As

plan for you have used in your project Student Homework
CAP Tools thus far
If you have not used any CAP
Tools, then identify a CAP Tool that
you could, in retrospect, have
used
For tomorrow, be prepared to
share with the class which tool
you used, how you used it, and
how it helped you with your
project
D M A I C
Define Phase Flowchart 7
D M A I C
DEFINE
PHASE
OVERVIEW
Define A: Define B: Define C:

Identify Project Develop Team Define
CTQ’s Charter Process Map
D M A I C
Define Phase Overview 8
What is the Define phase?

The Define phase is when your team identifies:
Who your customers are and what their requirements are for your products and
services
The reason for doing the project and project boundaries
The project team members and how they will work together
What process you are trying to improve and what the process map looks like
Why is the Define phase important?

This phase is important because it clearly and precisely describes the goals of
the project, aligns the project with organizational priorities and lays the
groundwork that will allow the team to remain focused.
Steps involved in the Define phase:

Define A: Identify Project CTQ’s
Define B: Develop Team Charter
Define C: Define Process Map
D M A I C
Define A–Identify Project CTQ’s 9
What does it mean to Identify Project CTQ’s?

Critical to Quality Characteristics, CTQ’s, are the key measurable characteristics of a
project or process whose performance standards must be met in order to satisfy the
customer. Green Belt improvement projects typically focus on one or two CTQ’s of a
process or product.
Why is it important to Identify Project CTQ’s?

Project CTQ’s are important because they ensure that the improvement team is solving
problems that are both critical to your customer and aligned with your business
strategy. If project CTQ’s are not identified and validated in this manner, valuable
resources may be wasted on counterproductive projects that neither increase customer
satisfaction nor add value to the business.
What are the project tasks for completing Define A?

A.1: Identify Customer
A.2: Collect Voice of the Customer data to identify customer CTQ
A.3: Build a process/product drill-down tree to identify project CTQ’s
DEFINE STEP
OVERVIEW

Identify Project Develop Team Define Process
CTQ’s Charter Map
A.1 Identify Customer

A.2 Collect Voice of the Customer data to Identify Project CTQ’s
A.3 Build a process/product drill-down tree to Identify Project CTQ’s
D M A I C
10
Define B–Define Team Charter
What does it mean to Develop a Team Charter?

A charter is a document that establishes a purpose and plan for the project. It contains a
statement of the problem, the scope of the project (including the process to be improved),
and an improvement goal, a plan and schedule for the project, estimated financial
benefits, and a list of team members and their roles. The charter becomes the blueprint
for the project when key stakeholders approve it.
Why is it important to Develop a Team Charter?

The charter documents the expectations, boundaries, and business case for your project
and can help you identify necessary resources. It is also a key communication tool that
can help your team stay focused and help you share information about your project with
each other and with the business. An approved charter ensures that your team, sponsor,
and stakeholders agree up front on what is being done and that your project will have a
beneficial impact on the business.
What are the project tasks for completing Define B?

B.1 Define the business case
B.2 Develop problem statement
B.3 Develop goal statement
B.4 Assess project scope
B.5 Select project team and define roles
B.6 Develop charter
B.7 Get sign-off for team charter
DEFINE STEP
OVERVIEW

CTQ’s Charter Map
B.1 Define the business case B.5 Select project team and define roles
B.2 Develop problem statement B.6 Develop charter
B.3 Develop goal statement B.7 Get sign-off for team charter
B.4 Assess project scope
D M A I C
Define C–Define Process Map 11
What does it mean to Define a Process Map?

A high-level process map is a chronological display of the most significant four to five
steps, events, and operations in a process. It provides a structure for defining a process
in a simplified, visual manner. The high-level map gives you an overall view of an entire
process and lays the foundation for thinking about the process in more detail.
Why is it important to Define a Process Map?

A high-level (COPIS) map will help you understand the process and validate your
project scope. It is a bridge between the problem and scope statements in your charter
and the more detailed maps you will develop to help you improve the process. A high-
level map provides focus for the team and helps you identify areas that are within (as
well as beyond) your control. In addition, process mapping serves as a communication
tool that helps you to clarify the process to others, both internally and externally to the
business.
What are the project tasks for completing Define C?

C.1 Develop high-level Process Map
DEFINE STEP
OVERVIEW

CTQ’s Charter Map
C.1 Develop high-level Process Map
D M A I C
Elevator Speech–Homework Activity* 12
Desired Outcome: Share elevator speech for the project with members of your
team
What How Who
Preparation Develop an elevator speech for your project. All
Elevator Each team member shares the elevator Each

Speech speech for their project. (See the description Team
of an elevator speech on the following page.) Member
Feedback Make a note of any feedback you receive All

from your class members. Did your elevator
speech provide a clear understanding of your
project?
Elevator speeches will be shared with the entire class throughout the 3-day workshop.
* For additional information on elevator speeches, see the Self-Paced Workbook,

pp 197-198
D M A I C
Elevator Speech 13
Here is what our project is about (describe the problem or

issue)
Here is why it’s important to you (describe the benefit of doing
the project)
Here is what success will look like (describe the goal of the
project and where we currently are)
What is it? Why use it?
This is a tool that helps the team members put together a It helps the team link the need for change with the vision
short “sell” pitch for the project. Being able to clearly and of the future. All the team members will be using a
simply state the need for change and describe the future common “sell” pitch. For teams that have thoroughly
state is essential for rallying the support and commitment debated and documented both the need and vision, this is
of key constituents. the synthesis event whereby they distill the essence of the
project. For teams struggling to get started in terms of
The metaphor of the elevator is useful in challenging the need and vision, it can be the place to begin to bring
team to be clear, precise and simple. some focus to the team’s more rambling discussion of
need and/or vision.
Imagine a chance meeting of a team member and a key
stakeholder in an empty elevator with a 90-second ride. The elevator speech should be able to deal with the
questions that will arise once the project is announced to
the broader constituent base.
Describe the need for change and the vision of the new
state, as one might respond to the question, “Why are we
doing this project?”
D M A I C
DMAIC Checklist–Define A 14
A.1 Identify customer

Have you considered both internal and external customers?
Have you focused on the most important customer segment(s)?
A.2 Collect voice of the customer data to identify customer CTQ’s

Have you collected data to understand customer requirements?
Have you coordinated your data collection with others who are collecting similar
data, so as not to overwhelm the customer?
Have you validated the customer CTQ’s (Big Y) with the customer?
Have you verified a VOC Research method that you’ve selected that reflects the
true VOC?
A.3 Build a process/product drill-down tree to identify project CTQ’s

Is your project CTQ focused on what’s most important to the customer?
If you have multiple CTQ’s, have you used the appropriate tool to prioritize them?
D M A I C
DMAIC Checklist–Define B 15
B.1 Define the business case

Is the business case compelling to the team?
Does this business case illustrate why this project needs to be done now ?
Have I considered intellectual infringements?
B.2 Develop problem statement

Does your problem statement identify a problem and not prejudge a root cause or
attempt to solve a problem?
Is the problem statement linked to the project CTQ?
Is the problem statement based on facts not assumptions?
Has another Improvement Team tried to solve this or a similar problem? What can
you learn from their effort?
B.3 Develop goal statement

Is your goal statement SMART (Specific, Measurable, Attainable, Relevant, Time-
bound)?
Does your goal statement focus on what needs to be improved, and without
predetermining a solution?
How will you know the team is successful?
D M A I C
DMAIC Checklist–Define B (continued) 16
B.4 Project scope

Does your team agree on the project scope?
Is your project scope within the team’s control?
B.5 Roles
Are all functions represented?
Does each team member fully represent their manager’s input on the project?
D M A I C
DMAIC Checklist–Define C 17
C.1 Develop high-level process map

Does the COPIS scope (start and stop) match the scope definition in your team
charter?
Have you mapped the process from your customer's perspective?
Is this the “As-Is” process? Was the map validated with the people who actually
do the process
D M A I C
Forms Of Intellectual Property 18
Patent
Best Protection against independent development by others
Gives right to exclude others from making, using selling, the invention of 20
years from filing
Can wait 2-3 years for patent to issue
Right granted by the government for new, useful, non-obvious inventions
Trade Secret
Rights exist only so long as actually kept secret
Protects against theft or misappropriation, not independent development
Does not require inventiveness in the patent sense
Copyright
Protects original works of authorship fixed in a tangible medium
Does not protect against independent development
Trademark
Defines sources of goods or services
Arbitrary, fanciful marks get stronger protection
D M A I C
Intellectual Property Assessment in Chartering Phase 19
Throughout the project consider implications to Intellectual

Property Policies and Procedures through the following:
Commercial benefit for licensing–consider this in the
cost/benefit
Infringement Avoidance–GE has a policy covering avoiding
infringing on other’s IP-Assure no infringement
Identify and Capture IP–Throughout this project consider
whether IP is developed and take actions to capture and
protect
Note: When writing your Business Case be certain to consider and include IP
issues.
Consult the business’ IP Business Champion and IP

Designer with questions, and for guidance
D M A I C
D M A I C
Measure Module Objectives 1
Review steps in Measure phase

Start work on Green Belt project deliverables for Measure phase
DMAIC GB E TX PG V 4.2.0
D M A I C
Measure Phase Flowchart 2
D M A I C
MEASURE
PHASE
OVERVIEW
Measure 3:
Measure 2: Establish Data
Measure 1:
Define Collection Plan,
Select CTQ
Performance Validate
Characteristic
Standards Measurement
System, & Collect
Data
D M A I C
3
The12-Step Process

Define
Measure
Standards Project Y
R&R
Analyze
Graphical Analysis,
Hypothesis Tests
Improve
Relationships
Tolerances
Control
Capability
D M A I C
Measure Phase Overview 4
What is the Measure phase?

This phase is concerned with selecting one or more product characteristics to
measure, defining how the characteristics will be measured, planning data
collection, and collecting data.
Why is the Measure phase important?

This phase is important because it ensures that accurate and reliable data is
collected to measure current process performance related to the customer CTQ.
Steps involved in the Measure phase:

Measure 1: Select CTQ Characteristic
Measure 2: Define Performance Standards
Measure 3: Establish Data Collection Plan, Validate Measurement
System & Collect Data
D M A I C
5
Mapping Tools To Generation Of CTQ Components
Define A
Output
Output Unit
Unit
C-O-P-I-S
C-O-P-I-S
Define A
Output
Output
Characteristic
Characteristic C-O-P-I-S
C-O-P-I-S
High Level Need Project

Project YY Measure 1
(VOC) Operational
Operational QFD
QFD
C&E
Definition
Definition
C&E
C-O-P-I-S
C-O-P-I-S
Measure 1
QFD
QFD
CTQ
CTQ Project
Project YY
Measure
Process
ProcessMap
C&E
C&E
Map
Measure Pareto
Pareto
Measure 2
Specification
Specification VOC
Limits
Limits VOC
QFD
QFD
QFD Measure 2
Target
Target QFD
Pareto
Pareto
VOC
VOC
Measure 2
Defect
Defect FMEA
FMEA
C&E
C&E
##of
ofDefect
Defect Measure 2
FMEA
Opportunities
Opportunities FMEA
Process
ProcessMap
Map
Per
Per Unit
Unit
D M A I C
D M A I C
Problem Characterize the Verify critical
statement process X’s and ƒ(x)
– Project Y – Stability Change
– Magnitude – Shape process
– Impact – Center Control the
– Variation gains
Data Integrity – Risk
– MSA Capability analysis
– Brainstorm – ZBench ST & LT – Control
potential X’s plans
– Sampling plan
Collect data
D M A I C
1
Measure 1–Select CTQ Characteristics
What does it mean to Select a CTQ Characteristic?

The goal for this step is to drill-down from what you learned in Define to identify the
specific sub-process or system characteristic that will be the subject of your Green Belt
project. In the Measure phase, you need to further narrow the scope by focusing on one
particular factor that impacts the CTQ. Remember that every project is different, and the
level that you need to drill-down to depends on how broad in scope you want your
project to be. By the end of this step, you should have identified exactly what aspect of
the product/service you will measure for your project.
Why is it important to Select a CTQ Characteristic?

It is important to manage the scope of your project. By drilling-down to a sub-process or
sub-system, if necessary, you keep the connection to the high-level customer CTQ
while, at the same time, keeping the project scope manageable.
What are the project tasks for completing Measure 1?

1.1 Identify the measurable CTQ characteristic that will be improved (Project Y).
MEASURE STEP
OVERVIEW
Measure 1: Measure 2: Measure 3:

Select CTQ Define Establish Data
Characteristic Performance Collection, Validate
Standards MSA
1.1 Identify measurable CTQ characteristics that will be improved (Project Y).
DMAIC GB F TX PG V 4.2.0
D M A I C
2
Define A
Output
Output Unit
Unit
C-O-P-I-S
C-O-P-I-S
Define A
Output
Output
Characteristic
Characteristic
C-O-P-I-S
C-O-P-I-S

(VOC) Operational
Operational QFD
QFD
C&E
Definition
Definition C&E
C-O-P-I-S
C-O-P-I-S
Measure 1
QFD
Project
Project YY QFD
CTQ
CTQ Measure
Measure
Process
ProcessMap
C&E
C&E
Map
Pareto
Pareto
Specification
Specification Measure 2
Limits
Limits VOC
VOC
QFD
QFD
QFD Measure 2
Target
Target QFD
Pareto
Pareto
VOC
VOC
Measure 2
Defect
Defect FMEA
FMEA
C&E
C&E
##of
ofDefect
Defect Measure 2
Opportunities
Opportunities FMEA
FMEA
Process
ProcessMap
Per
Per Unit
Unit Map
DMAIC GB F TX PG V 4.2.0
D M A I C
1
CTQ Tools Module Objectives
Learn how to apply a variety of CTQ tools to help prioritize &

define CTQ elements to be improved
If you don’t already know your Project Y, then one or more of
these tools can help you:
– Process Mapping
– Cause & Effect (Fishbone)
– Failure Modes and Effects Analysis (FMEA)
– Pareto Charts
– Quality Function Deployment (QFD)
This group of CTQ tools can be used to do some

preliminary analysis of your process in order to identify
the sub-processes that contribute most to satisfying the
customer CTQ requirement. Most likely, you will use 1
or 2 of these tools to gather your Project Y (Measure,
Step 1 deliverable).
Note: FMEA’s and Pareto Charts can also be used in

other DMAIC phases and will be covered in detail in later
sections of the program.
DMAIC GB G TX PG V 4.2.0
D M A I C
Process Mapping
D M A I C
3
Process Mapping–CTQ Tool #1
A Graphical Representation of
Steps, Events, Operations, and Relationships
of Resources Within a Process
Applications Enables group to see the entire process as a team

Techniques for examining a process to determine Magnifies normally overlooked areas and displays
where and why major breakdowns occur their relevancy
Graphically displays steps, events, operations and
relationships of resources
Used to design an improved process
Benefits
Provides a structure for breaking down a complex
process
Uncovers problem spots
Determines what data to collect.Targets selected
improvements.
D M A I C
4
Levels Of A Process
Acquire New
Core Business
Obtain Request to Conduct Prepare Initialize

Buy Service Underwriting Contract Customer Service
COPIS/SIPOC S Establish Prepare Negotiate Close

C
Terms Docs Contract Deal
Underwriters Customer/
Customer Service
Detailed
Subprocess
Map
Tasks Procedures
Earlier in Define, you developed a high-level or COPIS If the Project Y is a cost measure, which of the blocks
process map. By looking at a process from a “big picture” adds the most cost?
perspective, you evaluated customer needs and supplier
inputs, and determined initial measurement objectives. If the Project Y is a function measure, which block has
the most errors or problems?
Now, you will look in more detail at the subprocesses
defined in the COPIS map. Subprocess maps provide Like working with a puzzle, you begin to assemble the
specifics on the process flow that you can then analyze pieces of an area on which it makes sense to focus our
using several useful techniques. Choose what to efforts.
subprocess map by determining which of the major steps
in the COPIS have the biggest impact on the output (Y’s).
The block (or blocks) selected is the one on which you
create a subprocess map–using it to understand how and
why it impacts the output.
If the Project Y is a time measure, which of the blocks

consumes the largest portion of total time, or which one
has the most variation or delays?
D M A I C
5
Versions Of A Process
What You Think It Is... What It Really Is.. What It Should Be... What It Could Be...
Subprocess mapping is simply drawing a picture of the Reconciling the map requires your team to observe the
process–documenting the flow of the process. But, there process–making sure you have recorded all the existing
are at least four major versions of a process map that you steps.
can draw.
As the team moves forward and does process analysis
First, you can document what individuals who touch the and problem-solving, eventually you will move toward
process think it is. Certainly with the daily contact, you the third version of the process map–the “should be”
would expect people to know how the process works. But map created as part of their Improvement plan. At this
many people can easily explain how things work when point, a check must be made as to whether customer
things go right. You need to know how the process works CTQ’s have been met or exceeded.
in all conditions, so you need to go further.
Reconciling what the process map is, into what it really

is, is a second version of the process. These first two
versions of the process constitute what is referred to as
the “as is” process map. A thorough “as is” process map
is one of the short-term goals of good process mapping
and a deliverable for the Analyze phase.
D M A I C
6
Outside-In Focus
Does your process overlap with the customer’s?

Should the customer’s process be mapped?
Should you partner with the customer to
assess the processes?
Continue to keep focused on the customer’s perspective.
D M A I C
7
Subprocess Mapping Techniques
Process flowchart
Alternate path method
Deployment or cross-functional map/flowchart
There are a number of ways to draw a subprocess map.

Your team can choose the approach or approaches most
appropriate (or comfortable) for your use. Whatever type
of subprocess map is chosen, make sure that it accurately
reflects your process as it currently exists.
D M A I C
8
Process Mapping Symbols
These symbols represent the standard symbols for

process mapping.
Most projects use the simplified symbol sets shown on

the following page.
D M A I C
9
Process Flowchart
Customer Call
Automated Touch- Rep
Customer Yes Chooses Placed In
Criteria Customer
Prepare Yes
System Tone Answers
Calls Routing Que, On
Met? LoanWaits?
Offer
Answers Phone Phone
Option Hold
No
Call Gets No
Routed To
Call
Voice-Activated
Ends
System
Key:
Start Review Or
Task Decision Direction
End
The business process map most people are familiar with

uses four simple symbols: oval, rectangle, diamond, and
arrow.
It is generally used when the process is fairly small and

simple, or when documenting the work done by a single
person or group.
D M A I C
10
Alternate Path Flowchart
85 Customer Call 80
Automated Touch- Rep
Customer Yes Chooses Placed In
Criteria Customer
Prepare Yes
System Tone Answers
Calls Routing Que, On
Met? LoanWaits?
Offer
Answers Phone Phone
Option Hold
15
No
20
Call Gets No
Routed To
Call
Voice-Activated
Ends
System
Key:
60
40
Direction
Task Percentages Decision
The “alternate path” process map method arose from

reengineering efforts, where the mapping of very large
processes made “decision diamonds” more of a hindrance
than a help.
In this technique, diverging or alternate “paths” are noted

by split arrows. Teams can then note relative percentages
of times/incidences the process follows each path.
Process mapping software tools also make it easier to add

“icons” showing the tools or methods employed at
different points in the process.
Showing the alternate path allows us to identify sources

of variation which maybe adding complexity to our
process.
D M A I C
11
Deployment Or Cross-Functional Flowchart
Sign File
Customer Negotiate
Contract Contract
Contract Initiate Revise Revise File

Negotiate
Call Contract Contract Contract
Admin.
Review Review
Attorney
Contract Contract
Write File
Dealer Contract Contract
Who PROCESS FLOW
This approach to process mapping emphasizes who

performs which tasks and which steps are done
concurrently.
It provides a clear visual perspective of the hand-offs and

relationships between groups involved in the process.
(Note: Either of the mapping techniques described on the
previous pages can be placed on this type of map.)
Always include the process customer and key suppliers as

“bands” on this type of map. (Because customers are
visible at the top, some people refer to this as a “Service
Blueprint.”)
D M A I C
12
Steps In Deployment Mapping
1. Identify the participants of the process

from start-to-top point. Place the
customer at the upper left corner.
2. Identify the “trigger” or initial step and
identify the “final” step
3. Identify who receives the output of the
initial step and what activity they
perform
4. Repeat by identifying who receives the
output from the second step and what
activity they perform
5. Continue identifying steps and align
vertically with participants
For Your Process/Subprocess: 3. Next, identify who receives the output of the second
step and what activity they perform. Write this activity
1. Identify the participants involved in the process, on a card and place it to the right of the second step on
write their names on cards, and pin or tape the cards the same row as the participant performing the activity.
in a vertical column on the wall with the customer at
4. Identify who receives the output of this activity and what
the top of the column.
activities that participant performs.
2. Identify the action that initiates the process, write it 5. Continue to create cards and position them in the same
on a card, and place it next to the appropriate manner until the process is complete (i.e., the customer
participant card. In many instances,the process is receives the final process output).
initiated by the customer but it is not always the case.
Also, identify the final step in the process and place in
the appropriate row at the end of the map. Here again,
the last step may be the customer’s action.
D M A I C
13
Tips In Subprocess Mapping
Involve people who know (focus on) the “as is”

Clarify process boundaries
Brainstorm steps–write on Post-its®
– Use verb-noun format (e.g., Prepare contract, not contracting )
– Don’t include “Who” in step description
Combine, eliminate duplicates, clarify steps
Organize steps into proper “flow” and add arrows
Respect the boundaries

Don’t start “problem-solving”
Validate and refine before analyzing
Here are some guidelines on building a subprocess map. Combine and Clarify–Make sure brainstormed steps
These are not absolute–but they should help you avoid are clear and don’t overlap.
some of the pitfalls of process mapping.
Organize in “Flow”–Creating the map is last. With all
Focus on “As Is”–To find out why problems are steps visible, it’s typically much easier to create a
occurring in a process, you need to concentrate on meaningful map without getting stuck on one or two
how it’s working now. minor issues.
Clarify Boundaries–If you’re working from a well-
Is the customer involved in your subprocess step?
done high-level map, this should be easy. If not, you’ll
If so, how?
need to clarify start and stop points.
Brainstorm Steps–It’s usually much easier to identify
the steps before you try to build the map. Starting each
step description with a verb (e.g., “Collate Orders”;
“Review Credit Data”) helps you focus on action in the
process. Who does the step is best left in parentheses
(or left out) –you want to avoid equating a person with
the process step.
D M A I C
14
Verifying CTQ Elements For Process Mapping
Using Process Mapping in the Measure phase allows us to:

Select an upstream step that is generating defects
Determine where to collect the data
Define the output measure (a good place to begin looking is a “No Exit” out
of Decision Diamonds)
D M A I C
15
Subprocess Mapping–Breakout Activity (20 minutes)
Desired Outcome: Practice creating a subprocess map
What How Who Timing
Team Choose a facilitator, All 30 secs.

Preparation timekeeper, scribe and/or note
taker
Prepare Prepare the subprocess map for All 19 mins.

Subprocess your own project
Map
Brainstorm the challenges of
subprocess mapping on your
own projects
Close Choose a subprocess to report All 30 secs.

Exercise your results
D M A I C
16
Cause & Effect Diagrams–CTQ Tool #2
A visual tool used by an improvement team to brainstorm and

logically organize possible causes for a specific problem or effect.
Machines Methods Materials
Potential High-Level Causes Problem

Statement
Measurement Mother Nature People
Procedure: 7. As a team, determine the three to five most

likely causes.
1. Draw a blank diagram on a flip chart.
8. Determine which likely causes you will need to
2. Define your problem statement. verify with data.
3. Label branches with categories appropriate to your Alternative procedure:

problem. The categories shown are often used, but any
categories can be used. Brainstorm causes independent of categories.
4. Brainstorm possible causes and attach them to Use the Affinity Diagram and brainstorm items into
appropriate branch. categories.
5. For each cause ask, “Why does this happen?”
6. Analyze results; do any causes repeat?
D M A I C
17
Cause & Effect Diagrams–Example
Cause & Effect for Capital Logistics

Example Of A CTQ
Measurements People
Capital Truck
Dock Clock
Capital Driver
or
Watches
3rd Party Carrier
or
Computer Tracking
Capital Truck
Lease Driver
New employees
Why are the
Weight of freight 52' trailers
deliveries not
Temperature timely?
Distance 48' trailers
Overnight Delivery Volvo Trucks
Unload Delivery Mack Trucks

Precipitation
Drop off & Hook up Freightliner Trucks
Environment Methods Machines
The C&E answers the question of “What am I going to

measure?” I can now focus on weight of freight or
distance or temperature, etc. (This is the output
characteristic.)
D M A I C
18
FMEA–CTQ Tool #3
An FMEA is a structured approach to:

Identifying the ways in which a process can fail to meet critical customer
requirements
Estimating the risk of specific causes with regard to these failures
Evaluating the current control plan for preventing these failures from
occurring
Prioritizing the actions that should be taken to improve the process
Identify Ways The Product Or Process Can Fail, Then

Plan To Prevent Those Failures
Purpose & Benefits of FMEA Types of FMEA

Improves the quality, reliability, and safety of products System FMEA: is used to analyze systems and
Helps to increase customer satisfaction subsystems in the early concept and design stages.
Reduces product development timing and cost Focuses potential failure modes associated with the
functions of a system caused by design.
Documents and tracks actions taken to reduce risk.
Design FMEA: is used to analyze products before they
are released to production
Process FMEA: is used to analyze manufacturing,
assembly and transactional processes
FMEA’s will be covered in detail during the Improve

phase of the CD Rom. This section is a brief introduction.
D M A I C
19
Potential Failure Modes and Effects Analysis
Worksheet
Process/Product: Invoicing FMEA Date: (Original) 10/15/96
FMEA Team: Invoice Process Mgnt. Team (Revised) 5/25/97
Black Belt: E. Jones Page: 1 of 1
FMEA Process Action Results
Occurrence
Occurrence
Responsibility
Detection
Detection
Severity
Severity
Potential Potential
Recommended And Target
RPN
RPN
Item/Process Potential Current Action
Effect(s) Of Cause(s) Of Completion
Step Failure Mode Controls Action Taken
Failure Failure Date
Enter
Amt Inaccurate Overbill 8 Wrong Ct 8 Rare 9 576 DMAIC Team E. Jones All DMAIC 8 4 2 64
Owed Wrong Price 5 “ 9 360 to Investigate 5/15/97 Tasks 8 3 2 48
Root Causes Complete
Underbill 5 Wrong Ct 8 Rare 9 360 of Count & 5 1 2 10
Wrong Price 5 “ 9 225 Price 5 1 3 15
Accessories
Missing No Payment 6 Sale Error 5 Reviewed 3 90 6 5 3 90
De Error 7 “ 3 126 6 7 3 126
De Error
Delay 3 Sale Error 5 Reviewed 3 45 3 5 3 45
De Error 7 “ 3 63 3 7 3 63
De Error
Delayed Late Bill 3 Sale “Too 7 Measured 4 84 3 7 4 84
Busy”
System Down 3 “ 4 36 3 3 4 36
No Bill 6 Sales Busy 7 Measured 4 168 6 7 4 168

System Down 3 “ 4 72 6 3 4 72
Total Risk Priority Number 2,205 Resulting Risk Priority Number 821
In this example,the “Wrong Ct” component of billing

accuracy becomes the highest priority defect type.
D M A I C
20
Failure Mode And Effects Analysis
How it Works Function

Part/Process
Failure Mode
Effects Controls
Severity Detectability
(1-10) (1-10)
Causes
Occurrence
(1-10)
RPN
RPN
Risk
Risk Priority Number
Priority Number
RPN
RPN = S x O x D == 11 to
= S x O x D to 1000
1000
Review the product, service, or process Assign Severity, Occurrence, and Detection Factors
Determine failure modes Calculate RPN.Prioritize RPNs from high to low.
List one or more potential effects for each failure Identify the top issues, those with high RPN’s.
mode. Answer the question: “If the failure occurs, Determine preventive or remedial actions, especially
what are the consequences?” for high-priority issues.
Identify potential causes Recalculate RPN after actions have been implemented
List current controls
D M A I C
21
FMEA: Calculating Risk Priority Number (RPN)
RPN = Severity x Occurrence x Detection

Severity Scale
Rating Criteria – A Failure Could:
Bad
10 Injure a customer or employee
9 Be illegal
8 Render product or service unfit for use
7 Cause extreme customer dissatisfaction
6 Result in partial malfunction
5 Cause a loss of performance which
4
is likely to result in a complaint
Cause minor performance loss
Occurrence Scale
3 Cause a minor nuisance, but be Rating Time Period Probability
overcome with no performance loss
10 More than once per day > 30%
Good
2 Be unnoticed and have only minor
9 Once every 3-4 days ≤ 30%
Detection Scale
effect on performance
1 Be unnoticed and not affect the 8 Once per week ≤ 5% Rating Definition
performance 7 Once per month ≤ 1% 10 Defect caused by failure is not detectable
6 Once every 3 months ≤ .03% 9 Occasional units are checked for defect
8 Units are systematically sampled and inspected
5 Once every 6 months ≤ 1 Per 10,000
7 All units are manually inspected
4 Once per year ≤ 6 Per 100,000
6 Units are manually inspected with mistake-proofing
3 Once every 1-3 years ≤ 6 Per Million modifications
2 Once every 3-6 years ≤ 3 Per 10 Million 5 Process is monitored (SPC) and manually inspected
1 Once every 6-100 years ≤ 2 Per Billion 4 SPC is used with an immediate reaction to out-of-
control conditions
3 SPC as above with 100% inspection surrounding out -
of-control conditions
2 All units are automatically inspected
1 Defect is obvious and can be kept from affecting the
customer
Total RPN is determined by multiplying all of the

individual scores.
Note: Occurrence, detection and severity are independent

estimates of risk.
D M A I C
22
Using A FMEA In The Measure Phase
Helps to prioritize Project Y’s

Helps to identify where and how a process may fail to meet a
CTQ (defects)
In this process, you may uncover an issue where you

need to contain a problem (e.g., regulatory, non
compliance, legal). If this occurs, it raises the priority.
D M A I C
23
Pareto Charts–CTQ Tool #4
Is there a defect that occurs frequently?
Segment data to look for a Frequency

significant factor that
influences the process
C A E D B
Category of Defect
The Pareto chart is a bar chart with the bars

(segmentation levels) arranged in descending order. It is
an essential tool to help prioritize improvement targets by
identifying the 20% of the problems that cause 80% of
the poor performance (Pareto principle).
D M A I C
24
Pareto Charts (continued)
Example April 1 – June 30

200 100
Number Of Units Investigated: 5,000
180 *f = frequency Cumulative
90
Summation Line
160 (Cum Sum line)
80
f* of D+B+F
Cumulative Percentage
140 70
f* of D+B
120 60
Frequency
f* of D
100 50
LEGEND
80 40
A: Illegible
B: Bank Info Incomplete
60 30 C: Missing Signature
D: Personal Information
Incomplete
40 20 E: Employment History
Incomplete
20 10 F: Address Incomplete
0
Defect Type D B F A C E Other* Total
# Defects 105 40 20 10 5 3 25 208
% Defects 50.5 19.2 9.6 4.8 2.4 1.5 12.0 100
Cum % 50.5 69.7 79,3 84.1 86.5 88.0 100.0
Approximately 80% of Defects from Defects D + B + F
A fully documented Pareto chart will include the

“Cumulative Summation line”, or Cum Sum line, which
depicts the running total of the frequency of each
subsequent bar (segmentation factor). The right-hand axis
on the graph will show the cumulative percent of defects.
By reading the Cum Sum line against the cumulative
percentage, your team can determine which of the
segmentation levels comprise 80% of the total for the
problem, and direct their attention to those levels.
In this example, I would likely focus on the ‘D” segment,

and this would become my project Y.
You should also create a Pareto showing $’s. This may

cause your focus to change.
* Other- Try to keep this category as small as possible.

Find out what is in “other”.
D M A I C
25
Using Pareto Charts In The Measure Phase
Quantitative tool use to determine segmented areas of focus

Graphically displays most frequent occurrence of outcomes
(little y’s)
Helps you get from Big Y to little y
D M A I C
26
QFD–CTQ Tool #5
Definition of QFD (Quality Function Deployment)
Structured Methodology to Identify

and Translate Customer Needs
and Wants Into Measurable
Features and Characteristics of a Product or Service
D M A I C
27
QFD Flowdown–Product Application
Functional
Requirements Part Characteristics
(HOW’s)
GE
Requirements
(HOW’s) Processes
House
Customer
(WHAT’s)
of Process
Requirements
House (HOW’s) Variables

Functional
Quality
(WHAT’s)
of
Characteristics
#1 (HOW’s)
Quality
(WHAT’s)
#2 Part House
Processes
of
(WHAT’s)
House
Quality
GE
of
Y #3
Quality
Y
#4
Key Functional
Requirements
Key Part
Characteristics X
Key
Manufacturing Key
Processes Process
Variables
Product application of QFD links functional requirements

to key process variables.
D M A I C
28
QFD Flowdown–Services Application
Service
Requirements
(HOW’s) Service
Functions/Processes
(HOW’s)
Customer Wants
Process Controls
House (HOW’s)
(WHAT’s)
of
Requirements
Quality
House
(WHAT’s)
#1
Functions/Proceses
Service
of
Quality
(WHAT’s)
#2 House
Service
of
Quality
Y #3
Critical-to-Quality
Characteristics
(CTQ’s) Key X
Service
Processes Key
Process
Variables
Services application of QFD links project deliverables to

key process tasks.
D M A I C
29
House Of Quality Summary
Roof
Roof
1b. Correlation
Correlation Target
1b. Customer
Customer Target
Importance
Importance Direction
Conflict Direction
Conflict
How
How important
important Resolution
are Resolution
arethe
thecustomer
customer
wants
wants to
to the
the
Current Rating
Competitor #1
Competitor #2
customer?
customer?
3.Characteristics/measures
3.Characteristics/measures
HOW'S
HOW'S
How
How do
do you
you satisfy
satisfythe
thewants
wants??
1a.
1a. Customer
Customer H L L M
Needs
Needs H
WHAT'S
WHAT'S M M L 2.
2. Competitive
Competitive
H Assessment
Assessment
What
Whatdoes
doesthe
the
customer
customer
want?
L M Where
Whereare
arewe
competitors
weand
and
want? competitorsrelative
relativeto
to
M L H customer importance?
customer importance?
Voice
Voiceof
of the
the
Customer
Customer L M
4.4.Relationships Importance
Relationships ImportanceRatings
Ratings
HH Strong
Strong 99
MM Medium
Medium 3
3 6.
6. Target
Target values
values of
of HOW's
HOW's (units)
(units)
LL Weak
Weak 11
5.
5. Competitive
Competitive
Benchmarks
Benchmarks
How
How do
do competitors
competitors
perform
performrelative
relative to
to each
each
HOW?
HOW?
It is not necessary to develop every room for every house of To what extent will a HOW impact a WHAT? Using a 9, 3, and
quality you build. 1 scale forces a wider spread between the most important and
Room 1a least important items.
Customer Needs (in their language) Room 5
High-level wants identified by the customer Competitive Benchmarks
Room 1b How the competition performs relative to the
Customer Importance measures/characteristics.
Customer ranking of the wants. A typical Room 6
score is 1-5. Target and Specifications
1–least important Target values for the HOW’S. Setting measurable targets allows
5–most important the team to define what is required to achieve customer
Room 2 satisfaction.
Competitive Components Roof
Customer’s view of how GE compares with the competition Relationship
Room 3 Impact of the HOW’S on each other.
Characteristics/Measures ++ Strong positive
Measurable attributes that can be used to indicate how well the + Positive
customer’s needs are met. - Negative
Room 4 -- Strong Negative
Relationships
Strengths of interrelation between the WHAT’S and the
Importance Ratings (IR)
HOW’s. H–Strong (9)
IR = Σ (Customer Importance X Strength Of Relationship)
M–Medium (3)
L–Weak (1)
D M A I C
House Of Quality Example–Lunch Selection 30
Relationship O = HOLD CONSTANT

High Positive
Positive
Lunch QFD High Negative
Negative
Number of measured ingredients

Percent carbohydrate provided
Percent nutrition provided
Product Requirement
Number of dishes used
Cost of ingredients
Customer Needs Weight of portion
PB&J Sandwich
Time to prepare
Instant Soup
Importance
Fast Food
Direction for improvement
Fills us up 5 H M M 2 3 4
Is nutritious 4 M H L L 3 3 2
Tastes good 3 H L 2 3 4
Is easy to make 4 H H 2 1 4
Is easy to clean up 2 M H 2 3 5
Sticks with us 4 H M 2 2 4
Is inexpensive 1 L L L M 4 3 1
Is clean 2 M M H 2 3 4
Importance Rating 94* 52 59 48 60 10 18

Target (Obtained from VOC) 16 33 25 3 1 0.50 0
PB&J 2 33 25 3 1 0.25 0 Key
Instant Soup 8 15 10 5 3 0.50 1 H=9
Fast Food 16 40 73 10 0 3.00 0 M=3
Units Of Measure (Obtained from VOC) oz % % min $ L=1
*Importance Rating For “Weight Of Portion” = 5(9) + 4(3) + 4(9) + 1(1) = 94
Roof Relationship Symbols indicate if the quality goals of the CTQ’s are conflicting. For example, if I chose to focus on
“Weight of Portion” I need to be concerned about the negative impact on the “Number of Measured Ingredients”
D M A I C
31
Analyzing A House Of Quality
Empty
Empty columns
columns in in Strong
Strong negative
negative relationship
relationship
Room
Room 33 -- Perhaps
Perhaps an an in
unnecessary
unnecessary in roof –Identify
roof –Identify trade-offs
trade-offs
measure
measure
characteristic
characteristic
signaling
signaling itit does
does not
not
affect
affect customer
customer
Current Rating
Competitor #1
Competitor #2
wants
wants
L L M
H Highest
Highest score
score on
on
competitive
competitive
M M L comparison
comparison -- AbleAble to
to
lead
lead the
the market
market with
with
existing
existing
L M product/service
product/service
L H
L M
Empty Low
Low score
score on
on competitive
competitive comparison,
comparison, but
but
Emptyrows
rowsininRoom
Room33- - high
Unaddressed customer high score
score on
on competitive
competitive benchmarks
benchmarks --
Unaddressed customer Market
Market technical
technical advantages
advantages to
to improve
improve
want
wantwould
wouldbebeaamajor
major
problem
customer
customer perception
perception
problem
Low
Low competitive
competitive benchmarks
benchmarks -- Poor
Poor long-term
long-term market
market performance
performance
D M A I C
Measure 1–Select CTQ Characteristics
CTQ And Performance Standard Worksheet 32
Define A
Here’s our next step: Output
Output Unit
Unit AA
Lunch
LunchPortion
Portion
Output
Output Define A
Characteristic
Characteristic Healthy
HealthyLunch
Lunch
Project
Operational Total
TotalWeight
Weight
Operational In
Definition InOunces
OuncesPer
PerServing
Serving
High-Level Definition
Need (VOC)
Measure 1
Project
Project YY
Measure
Measure Weight
Weight
Prepared
Prepared
Lunch
CTQ
Lunch
Measure 2
Specification LSL=14
LSL=14oz.
oz.
Specification USL=18
Limits* USL=18oz.
oz.
Limits*
Measure 2
Target*
Target* 16
16oz.
oz.
You may not know Spec, Target Measure 2

or Defect at this time (in Step 1). Defect
Defect**
If you know it now, write it down. <<14
14oz.
oz.Or
Or>>18
18oz.
oz.
This information will come from
the VOC.
##of
ofDefect
Defect Measure 2
Opportunities
Opportunities
Per
Per Unit
Unit 11
D M A I C
Methodology Step: Measure 1–Select CTQ Characteristics
33
CTQ And Performance Standard Worksheet
Define A
Output
Output Unit
Unit AA
Lunch
LunchPortion
Portion
Define A
Output
Output
Characteristic Healthy
Healthy
Characteristic
Lunch
Lunch
Project
Operational
Operational %
%Carbohydrate
Carbohydrate
Definition Per
PerServing
ServingPer
PerUSDA
USDA
Need (VOC)
Measure 1
Project
Project YY
Measure
Measure %
%Carbohydrate
CarbohydrateContent
Prepared
Prepared Content
Lunch
CTQ
Lunch
Measure 2
Specification
Specification
Limits*
Limits* >>20%
20%
Measure 2
Target*
Target*
25%
25%
Measure 2
Defect
Defect**
You may not know Spec, Target
or Defect at this time (in Step 1). <<20%
20%
If you know it now, write it down.
the VOC. ##of
ofDefect
Defect Measure 2
Opportunities
Opportunities
Per
Per Unit
Unit 11
D M A I C
34
Another Example
What are your needs for this class?
# Tools Used
# Completed
Assessment
Customer
Exam Pass
Importance
Knowledge
Project Per
In Project
Needs
Quarter
Score
Rate
Learn The
Material
5 9 3 3 1
Pass The
Exam
4 3 9 1 1
Apply This
Material To Project
2 1 1 9 3
Importance
Rating 59 53 37 15
Key
High = 9
Medium = 3
Low = 1
D M A I C
35
Points To Remember
The process may look simple, but requires effort

Many entries look obvious–after they’re written down
If there are NO “tough spots” the first time
– It probably isn’t being done right!
Focus on the end-user customer
Charts are not the objective. Charts are the means for
achieving the objective.
QFD is a perishable document. It has a shelf life and must be
updated.
QFD is a valuable decision support tool, not a decision-maker
QFD is a valuable decision support tool (assigning

numbers to H-M-L does not make the numbers and cut-
off’s absolute decision points).
D M A I C
Methodology Step: Measure 1–Select CTQ Characteristics
36
CTQ And Performance Standard Worksheet
Define A
Output
Output Unit
Unit Knowledge
Knowledge
Assessment
Assessment
Define A
Output
Output GB
GBKnowledge
Knowledge
Characteristic
Characteristic Level
Level
Project
Project YY %
%Correct
Correct== Measure 1
Operational
Operational Total
TotalCorrect
Correct
Definition
Questions
QuestionsDivided
DividedBy
By100
100
Need (VOC)
Knowledge
Knowledge Measure 1
Project
Project YY Assessment
AssessmentScore
Score
Measure
Measure (Percentage
(PercentageCorrect)
Correct)
Learn
Learn
Six
CTQ
Six Sigma
Sigma
Measure 2
Specification
Specification ≥≥80%
80%
Limits*
Limits*
Measure 2
Target*
Target* 85%
85%
Measure 2
You may not know Spec, Target Defect
Defect**
or Defect at this time (in Step 1). <<80%
80%
If you know it now, write it down.
the VOC. ##of
ofDefect
Defect Measure 2
Opportunities
Opportunities 11
Per
Per Unit
Unit
D M A I C
37
Common QFD Pitfalls
QFD on everything
– Set the “right” granularity
– Don’t apply to every last project
Lack of teamwork
– Wrong participants
– Lack of team skills
– Lack of support or commitment
Too much “chart focus”
“Hurry up and get done”
Failure to integrate and implement QFD
D M A I C
38
QFD Activity
Background:
Capital Logistics has established a customer help desk to handle
customer inquiries regarding deliverables. However, customers are
dissatisfied with the help desk.
A team consisting of Capital Logistics customer service

representatives and Capital Logistics customers has been
assembled. The team’s objective is to use QFD to identify the sub-
processes that will be the focus of DMAIC improvement projects.
D M A I C
QFD–Breakout Activity (20 minutes) 39
Desired Outcome: Practice constructing a House Of Quality
What How Who Timing
Preparation Choose a facilitator, scribe All 5 mins.

and timekeeper
Complete Review the partially completed All 10 mins.

Relationships House Of Quality on the next
page
Complete the relationship room
of the house
Calculate Calculate the “How” importance All 5 mins.

“How” ratings
Importance
Which “How” is most important
Ratings
to satisfy the customer?
D M A I C
40
QFD Activity
HOWS
Service Characteristics
or good on customer
Rep (CSR) excellent
resolved on first call
the day helpdesk is
% customers rating
Number of hours in
Customer Service
resolve problems
survey of (CSR).
courteousness
Cycle time to
% problems
Importance
available
Quick problem resolution 5
WHATS
Customer Wants
Available when needed 4
Courteous Service 3
Importance rating
Units % Hrs.
Hrs. %
days
Target 100 <3 24x7 100
To what extent will the “How’s” impact the “What”?
Use L=1
M=3
H=9
Add one more service characteristic (CTQ) and place

rating on the CTQ.
D M A I C
41
CTQ Tools: Summary
There are a variety of tools available to identify the CTQ

characteristic for your project (Project Y)
Select the tool that is most appropriate for your project
D M A I C
D M A I C
Measure 2–Define Performance Standards 1
What does it mean to Define Performance Standards?

A performance standard defines the process to be measured, how it will be measured,
and how much variance will be tolerated. In step 1 of Measure, you determined the
specific subprocess that is now the subject of your Green Belt project. In step 2, you’ll
create a standard for the performance of that process. This is the point when you
determine the best way to turn what the customer wants into a numeric measurement.
This measurement will later be compared to the measurement of your current process in
order to see how well you’re meeting the customer’s need.
Why is it important to Define Performance Standards?

The performance standard translates the customer need into a clearly defined measure
for which performance data can be collected. Once you know what constitutes acceptable
and unacceptable performance for your process, you can define a defect. What you
define as a defect becomes the basis for determining performance capability and
improvement goals for the project.

2.1 Develop operational definition for process to be measured
2.2 Identify target performance
2.3 Set specification limit(s)
2.4 Define unit, defect and defect opportunity
MEASURE STEP
OVERVIEW

Standards MSA
2.1 Develop operational definition for process to be measured

2.2 Identify target performance
2.3 Set specification limit(s)
2.4 Define unit, defect and defect opportunity
DMAIC GB H TX PG V 4.2.0
D M A I C
Operational Definitions 2
Defining The Measure

Definition An operational definition is a clear, concise
description of a measurement
and the process by which it is to
be collected
Purpose
To remove ambiguity
Everyone has a consistent understanding
To provide a clear way to measure the

characteristic
Identifies what to measure
Identifies how to measure it
Makes sure that no matter who does
Always Pilot Your Operational
the measuring, the resultsDefinitions
are consistent
Operational definitions guide what properties will be

measured and how they will be measured.
There is no single right way to write an operational

definition. There is only what people agree to for a
specific purpose. The critical factor is that any two
people using the operational definition will be measuring
the same thing.
D M A I C
Operational Definitions For Output Measures 3
Features
What: Must have specific and concrete criteria
How: Must have a method to measure criteria
Must be useful to both you and the customer
(the “wing-to-wing” concept)
Example: Loan Application Cycle Time

What: Loan application cycle time is the number of hours from
receipt of a loan application, to successful notification of
decision for the loan application
How: The clock starts when the computer attaches the time of
application receipt at data entry
The clock stops when the phone caller notes time of completed
application decision notification in desk log
Must Have Specific And Concrete Criteria Example:

Everyone has the same understanding of the definition. If a team were measuring “length of lines” at a bank,
Different people can use the definition and know that they would need to know whom to include in a count
their data will be measured in the same way. of people in line–do children count, or a spouse or
friends waiting with someone?
Example:
If a team were studying long wait lines at a bank, they Must Be Useful To Both You And
would agree on what to measure (e.g., length of the lines, The Customer
how long people wait) and agree on a way to measure it.
The operational definition makes it possible to make
process improvements that are important to the
Must Have A Method To Measure Criteria customer and measurable for you. The definition
The definition tells you how to get a value (either should relate to issues that are important to both you
a number or yes/no). The definition clearly states how to and your customer.
get a measurement.
Example:
Are customers really concerned with the length of a
line, or just the length of time they must spend in line?
It is important to include the customer’s perspective
when developing operational definitions. © GE Capital, Inc., 2000
D M A I C
4
Why Start Our Measurement Work With The Project Y Metric?
The Project Y is the key metric for the project and is defined
from the customer’s perspective
A customer-focused Project Y measure allows us to assess the
process from the customer’s perspective
Changes in the values of a good Project Y measure will predict
changes in the degree to which the process meets the customer
CTQ’s
A Project Y measure expresses the “voice of the process”–
changes in input or process conditions drive changes in true
output measures
Project Y data is the first key to understanding process

performance and the starting point for understanding how
to control process output.
D M A I C
Define Performance Standards 5
CTQ Elements (Performance Standards)–5 Major Elements ONLY

Timely Call Pick-up
(Output characteristic) Established in Define, Step A
Hold Time
(Project Y Measure) Established in Measure, Step 1
Customer Need
Customer waits 10 seconds or less

on hold too long CTQ (Target) Established in Measure, Step 2
Not greater than 20 seconds

(specification/ tolerance limit) Established in Measure, Step 2
A CTQ applies
To each individual
output unit Any response taking More
than 20 seconds (Defect) Established in Measure, Step 2
CTQ’s Are The Bridge Between Our Process Output

And Customer Satisfaction
In Measure, Step 2, we are completing our CTQ Element Specification: the variation around the target the
Tree. Use the template on the next page or the one in customer will accept.
your GB Workbook.
Defect: failure to meet specification. Anything
that results in customer dissatisfaction.
Definitions:
Characteristic: a word or phrase that describe some
aspect of the product or service.
Project Y: a definition of how the product/service’s

characteristic is to be quantified. The key metric for the
project.
Target: where we “aim” our product/service. If there

were no variation in the product/service, this is the value
we would always achieve.
D M A I C
6
Define A
Output
OutputUnit
Unit C-O-P-I-S
C-O-P-I-S
AACall
Call
Define A
Output
Output Timely
TimelyCall
CallPick-up
Pick-up
Characteristic
Characteristic C-O-P-I-S
C-O-P-I-S

ProjectYY
Process
ProcessMap
QFD
QFD
Map Measure 1
C&E Time
Time From
FromCall
(VOC) Operational
Operational
C&E
C-O-P-I-S
C-O-P-I-S Placement
Placementtoto
Call
Definition Pareto
Pareto Customer
CustomerService
Service
Definition Pick-Up
Pick-UpInInSecs.
Secs.
Customer
Customer Measure 1
waits
waits QFD
QFD
on CTQ
CTQ Project
ProjectYY
Process
ProcessMap
Map HOLD
HOLDTIME
TIME
on Hold
Hold Measure
Measure
C&E
C&E
Pareto
Pareto
too
too Long
Long
Measure 2
Specification
Specification VOC
VOC USL=20
USL=20Seconds
Seconds
Limits
Limits QFD
QFD
QFD
QFD Measure 2
Target
Target Pareto
Pareto 10
10Seconds
Seconds
VOC
VOC
Measure 2
Defect
Defect FMEA
FMEA >>20
20Seconds
Seconds
C&E
C&E
##of
ofDefect
Defect Measure 2
Opportunities
Opportunities FMEA
FMEA 11
Process
ProcessMap
Per
PerUnit
Unit Map
D M A I C
Establishing A Performance Standard 7
A performance standard
translates customer needs
into quantified requirements
for our product or process Output
Output Timely Resolution
Characteristic
Characteristic of a Call
Specific Needs Total time to resolve

Statement Project
Project YY inquiry minus initiation
time of inquiry
Quick
Quick
Answer CTQ
CTQ Target
Target
5 Minutes or
Answer less
Specification/
Specification/ Not Greater
tolerance
tolerance Than 60
limit(s)
limit(s) (USL/USL)
(USL/USL) Minutes (USL)
Any Response
Defect
Defect Taking More Than
60 Minutes
Output Characteristic: A word or phrase that describes

some aspect of the product or service. Comes from the
Product/Process Drill-down tree.
Project Y: A definition of how the product/process’s
characteristic is to be quantified. State as a formula if
possible (Operational Definition) and state the metric
(Project Y Measure)
Target Value: Where we will “aim” our product/process.
If there were no variations in the product/process, this is
the value we would always achieve.
Specification Limits: How much variation is the
customer willing to “tolerate” in the delivery of our
product or process?
Defect: An output which is unacceptable to the customer.
Anything that results in customer dissatisfaction.
D M A I C
Identify Project CTQ’s 8
Translating Needs into CTQ’s: Example 1–Pizza

Characteristic: temperature of pizza
Project Y metric: temperature (degrees F) of Pizza at customer door step
Need
Need == Hot
Hot
Pizza Target: 160 degrees
Pizza
Specification: LSL=140 degrees and USL=180 degrees
Defect: any pizza temperature falling outside of the specifications
Translating Needs into CTQ’s: Example 2–Customer Service

Characteristic: CSA Knowledge level
Need
Need == Project Y: training test score
Problem
Problem Target: 100 points
Solved on First
Solved on First Specification: LSL= 90 points
Call
Call Defect: any score less than 90 points
Translating Needs into CTQ’s: Example 3–Information

Characteristic: Time to find information
Project Y: maximum time to find any piece of data
Need
Need == II want
want Target: 1 minute
to
to find
find all
all info
info Specification: USL=1.5 minutes
without
without effort
effort Defect: any time greater than 1.5 minutes
LSL=Lower Specification Limit

USL= Upper Specification Limit
D M A I C
Specify Target And Specification Limits For Project Y 9
Once the Project Y metric has been established, the target and
specification limits can be established.
Output
Output
Characteristic Timely response
Characteristic
Specific Needs Variance between

Statement Project
Project YY customer request and
actual delivery
Customer
Customer
receives
receives approval
approval
on CTQ
CTQ Target
Customer
on customer
customer Target request date
request
request date
date
Specification/
Specification/ Variance not
tolerance
tolerance greater than
limit(s)
limit(s) one day
Any response
Defect
Defect greater than
1 day
Target/Nominal Value
Optimal value for Y
Specification/Tolerance Limit(s)
Window or range of acceptable output values for Y
Also called “performance standards”
Information about where to set target and specification
limits comes from the customer. This information can
be collected not only from VOC but also from:
– Contractual agreements
– Industry standards
– Customer observations
– Process data
D M A I C
Project Y 10
Six Sigma Performance Definitions

Unit: The item produced or processed relative to the Project Y
as reviewed by the customer
Defect: Any Project Y measurement value that does not meet
the Y performance standards
Defect Opportunity: Any Y measurement event which provides a
chance of not meeting the performance standard
Defective: A unit with one or more defects
The definitions listed above are based on the Project Y

and performance standards and will be used later in the
Measure phase to calculate Process Sigma or
capability.
The “defective” terminology is used when there is
more than one defect opportunity per unit. A single
unit could have more than one defect-the total number
of defects is important because it represents the overall
magnitude of the problem. It only takes one defect per
unit to represent a problem from a customer’s
perspective so looking at the percent of units with at
least one defect gives us a perspective of how the
customer sees the overall process performance. Notice
that if units contain a number of defects, it is possible
to reduce the overall total number of defects without
having much impact on the percent of defective units.
We need to be aware of both the total number of
defects and the percent of units containing defects.
D M A I C
Project Y Alignment 11
Key output metrics that are

Business Big Y’s aligned with strategic
goals/objectives of the
business. Big Y’s provide a
direct measure of business
performance.
Process Y’s
Y
PROCESS
PROCESS Y
Key output metrics that
Y summarize process
Management
Management Y performance
Project Y
Key project metric defined
from the customer’s
perspective
X1 X2 X3
Any parameters that

influence the Y
Does Your Project Y Pass The Alignment Test?

All quality projects should be aligned to a Big Y and
CTQ for the business. This ensures that project activity
is organized and targeted to impact specific CTQ’s that
the customer will feel.
Not all Y’s are useful measures of business
performance or customer impact. Your Project Y
should be correlated to a higher level Business Big Y
or CTQ.
You should be able to describe the link between your
Project Y and the related Business Big Y or CTQ in
specific terms.
D M A I C
6σ Performance Standards Definitions–Breakout Activity (5 minutes) 12
Desired Outcome
Practice defining units and defect for your CTQ
What How Who Timing

Team Choose a facilitator All
Preparation
Define units For your process output, agree on the Facilitator 5 mins.
and defect unit and defect for your Project Y. Use
the worksheet on page 17 of your GB
Workbook.
Close Choose a spokesperson to report your All

exercise results
D M A I C
1
Minitab and Graphical Analysis Module Objectives
Understand the structure of Minitab*

Understand data entry and correct data structure for analysis in
Minitab
Review variation
Be able to create and interpret basic graphs in Minitab
In this module, you will focus on learning how to use

Minitab and how to perform basic graphical analysis of
data. We are stepping away from the DMAIC Flow for a
moment.
If you’ve been given some historical data, you can start to

view the data graphically. We will use Minitab to help us
view this data.
In order to be fully effective in the Analyze phase, we

need to expose you to Minitab’s capabilities now. This
will help you think about the best way to collect data in
Measure
Step 3.
*Note: For another tutorial, the student can go to:

www.minitab.com
DMAIC GB I TX PG V 4.2.0
D M A I C
2
Minitab Windows
Menu
Menu Bar
Bar
Session
Session Window:
Window:
•• Analytical
Analytical Output
Output
Data
Data Window:
Window:
•• A
A Worksheet,
Worksheet, not
not aa
Spreadsheet
Spreadsheet
Graph
Graph Window
Window
Data Window–Stores data to be used for analysis. You can move through the windows using Ctrl-Tab, the
Discussed in more detail on next page. windows menu, or the following shortcuts.
Data Window Crtl+A

Session Window–Stores numeric output and character
Session Window Crtl+M
graphs that result from executing a statistics command.
Info Window Crtl+I
Graph Window–Generated whenever you run a command History Window Crtl+H
that results in a graphical output.
Info Window–Stores synopsis of your data

(not shown) window.
History Window–Stores all commands initiated (not

shown) from either the menus or the session window (not
shown).
D M A I C
3
Data Window
Minimizes the Window Closes the Window
Data Entry Arrow

Maximizes
the Window
Column Names
Are Entered Here
Data is Entered
Here
Scroll Bars
Data window is a worksheet, not a spreadsheet. If the value you enter into the first row is something
Variables are usually entered in columns, observations other than a number or date, Minitab assumes that the
in rows. entire column will be text (T).
Each column has a title area which is 31-characters or
less and this title must be unique.
Everything in a column is considered to be the same
variable.
Column names are above the column, not in the first
row.
Minitab cells contain values only, not formulas, like
Excel.
To perform mathematical functions, use
Calc > Calculator.
The orientation of the data arrow determines whether
the enter key will take you to the next row or the next
column.
D M A I C
4
Minitab Menus–Summary
File Menu Print and save the window that is currently active
File menu changes depending on the window that is currently active
Allows open, close, and save
Edit Menu Similar to the edit menu in most standard Windows applications
Manip. Menu Sort, code or manipulate data
Calc Menu Calculate or generate data
Stat Menu Basic statistics and quality tools

Most often used by Green Belts
Graph Menu Contains the commands that you will use to do graphical analysis during
your project
Help Menu Minitab has a comprehensive Help system with detailed documentation of
all features, complete with examples of how all the menu commands are
used, and how to interpret graphical and statistical output which result from
the use of the commands
Window Menu Allows you to manage multiple graphs on the screen
Six Sigma* Product & process report can be used for determining performance levels
Menu (Calculating Sigma)
Notes on “Saving”: Notes on “Opening”:
Save Project As–When you save a project, you save all If you use the icon, only projects will be shown. You
the worksheet, session, info, history and graph windows must use the file > open worksheet to view all Minitab
that are currently open. Minitab will use the file extension worksheets.
of .mpj.
*Six Sigma add-on in Minitab can be obtained via a link
Save Current Worksheet As–When you save as a in support central.
worksheet, you save only the worksheet portion (you will
lose session window and graphs). Minitab will use the
file extension of .mtw.
D M A I C
5
Using Minitab: A Typical Session
1. Enter data
2. Select menu command (for desired statistical/graphical
function)
3. Enter command parameters in the dialog window
4. View results in session window or graph window
5. Copy output to another application
6. Print output
7. Save file
Command Parameters–Minitab sometimes calls these

parameters:
X Y
Factor Response
Independent Dependent
Input variable
By variable variables
graph variables
D M A I C
6
Step 1: Minitab allows you to enter data in four different ways:
1. Open an existing Minitab worksheet

2. Type data into the worksheet
3. Import data files from other compatible software packages (such
as Excel)
4. Paste data from other applications
D M A I C
7
Enter Data Manually
Only one row for

heading
C1 C2 C3 C4 C5 C6 C7 C8 C9 C10 C11 C12 C13 C14
Type data
directly into
worksheet
One way to enter data in Minitab is to type the data

directly into the worksheet. For your Green Belt project,
you will probably import or copy the data from Excel.
D M A I C
8
Importing Data From Excel
1.3 How to import an Excel data file:

1. File > Open Worksheet
2. Select Files of Type: Excel
3. Highlight the file to be imported
4. Double-click or click Open
Importing Data: When importing from Excel, an Excel file can contain
only one worksheet and the information contained in
To determine whether a file can be opened by Minitab, Row 1 will become the titles in Minitab.
choose Open Worksheet from the File menu. Notice that
the area labeled Files of Type contains a drop-down Note: During import, Minitab will convert percentages,
menu. This is indicated by the arrow (triangle) at the right dollars, etc.
of the Files of Type selection box.
When you choose a file type from the drop-down list,

only those files containing the chosen file extension are
displayed in the file list. When you are importing files, it
is important to choose the specific file type you are
importing.
D M A I C
9
Paste data from Excel

1.In Excel:
Highlight Data (and Column Names) to be copied
Using Your Mouse
2.Copy the Data to the Windows Clipboard

Edit > Copy (or CTRL-C on Your Keyboard)
3.Go to Minitab:
ALT > Tab
4. Position the Cursor where you want the data to fill

See example below.
5. Go to the Edit Menu:

Edit > Paste/Insert Cells (or Ctrl-V on the keyboard)
Insertion point
Note: In order for this to work, you

must make all data non-format specific
in Excel before you cut-paste.
D M A I C
10
Tips for moving data back and forth:
Structure the data so that each variable is in a single column

Each column must have a title
The column title must have fewer than 31 characters and be
on a single line
All data must immediately follow the column names
Do not put empty rows between rows of data
Columns containing dollar signs or commas cannot be transferred to Minitab
using Copy or Paste, but can be imported using the import command.
Reformat these numbers to include only decimal points.
After movement into Minitab, check column heading type
(D vs. T.)
If you want to use a file in both Minitab and Excel, keep

these points in mind.
D M A I C
11
Tables vs. Variable Columns

Format Needed For Minitab:
Typical Excel Format: Sales Office Revenue Month
Central 387,980 January
Central 45,700 February
Sales Office January February March April
387,980 45,700 456,789 349,050 Central 456,789 March
Central
578,990 600,987 456,789 456,798 Central 349,050 April
Southwest
435,800 542,700 345,988 564,050 Southwest 578,990 January
Northeast
497,050 827,900 456,789 687,050 Southwest 600,987 February
Southeast
613,242 61,689 456,789 434,567 Southwest 456,789 March
Northwest
Southwest 456,798 April
Northeast 435,800 January
Northeast 542,700 February
Northeast 345,988 March
Northeast 564,050 April
Southeast 497,050 January
Southeast 827,900 February
The best format for Southeast 456,789 March
Southeast 687,050 April
analysis of data in Northwest 613,242 January
Minitab is variable Northwest
Northwest
61,689
456,789
February
March
columns. Northwest 434,567 April
The format of data typically used in Excel is table format.

The best format for analysis of data in Minitab is variable
columns.
There are multiple ways to manipulate data. The major

point is that Minitab likes to have the data in columns.
D M A I C
12
Tips For Importing Data Into Minitab–Let’s practice

File name: sales_data.xls
Sales 1999
Sales Office January February March April
Central $387,980.00 $45,700.00 $456,789.00 $349,050.00
Southwest $578,990.00 $600,987.00 $456,789.00 $456,798.00
Northwest $435,800.00 $542,700.00 $345,988.00 $564,050.00
Southeast $497,050.00 $827,900.00 $456,789.00 $687,050.00
Northwest $613,242.00 $61,689.00 $456,789.00 $434,567.00
Total $2,513,062.00 $2,078,976.00 $2,173,144.00 $2,318,658.00

What is wrong with the format of this data in terms of its suitability for use in
Minitab?
Above is a set of data that has been captured and entered Note: For additional practice, see the Green Belt Training
into an Excel spreadsheet. Support Central Site.
To make the data suitable for use in Minitab:
1. Reformat the data in Excel (hide columns, change

dollar format to numeric) before copying and pasting
the data into Minitab. Arrange the data in 3 columns
as seen on the previous page.
2. Import the data directly into Minitab using Options
and Preview in the Open Worksheet dialog box to
customize the data structure.
Minitab also has several menu options that allow you to

reorganize and restructure the data once you have
imported it into Minitab. These options are available in the
Manip and Calc menus. You may want to first manipulate
in Excel prior to exporting the data!
D M A I C
13
Step 2: Select menu command
Most of the commands you use in your Green Belt

project are in the Stat and Graph menus.
D M A I C
14
Step 3: Enter parameters in dialog box
D M A I C
15
Step 4: View results in session or graph window
Note: Sometimes there is only a session window.

Therefore, check both the session window & the graph (if
one exists).
D M A I C
16
Step 5: Copy output to another application
Copying graphs:
1. Make sure the graph window is active in Minitab
2. Click right mouse button and select “Copy Graph”
3. Open the application into which you want to copy the graph or table,
e.g., Microsoft Word
4. Paste graph using Paste from the toolbar or Ctrl-V
Copying session window output:

1. Highlight the text lines you want to copy
2. Use the right mouse button to copy the text
3. Open the application into which you want to copy the text
4. Paste the text using Paste from the toolbar or Ctrl-V
Note: You can also copy graphs from the window >
manage graphs.
D M A I C
17
Steps 6 and 7: Print and Save
When you save a project, you can save the worksheet or

session window individually, or save an entire project.
D M A I C
18
Recruitment Cycle Time Case Study
We will use the following case study to practice basic graphical analysis in Minitab.
You are working on a project to reduce cycle time for filling temporary positions in the
Information Technology department within the company. Currently, you work with one
agency who finds temporary contractors to fill positions on request by the IT department.
The performance standard for cycle time to fill these positions is 20 (working) days
maximum. This is measured from the time the agency receives the request to the time
the position is filled.
You have completed the Define and Measure phases. In the Measure phase, you collected
cycle time data for a sample of positions filled within the last 6 months.
The data is contained in file fill_time.mtw. The column labels are as follows:
C1: yrs exp–years experience of the person who filled the position
C2: agent–specifies the agent who processed the request and found the resource
C3: site–specifies the company site from which the request originated (there are three sites in
the area who request temporary resources)
C4: type of resource–specifies the type of resource requested (DBA, Programmer,
Systems Analyst)
C5: cycle time–time to fill position (working) days
D M A I C
19
Graphical Analysis Of Data
Key Questions:
How is my data distributed (variation)?
What relationships exist between the
Y variable and X variables?
D M A I C
20
Review–Variation
All repetitive activities have variation (fluctuation)

Variation is a primary source of customer dissatisfaction
In order for our customers to “feel the quality” at GE, we must
reduce variation
A fundamental principle of process improvement is the

measurement, reduction, and control of variation.
Variation is present in all processes, whether they are
personal processes (getting to work, fixing dinner) or
business processes (approval cycle time, order to
delivery). The output of a process will vary as it is
repeatedly performed. Although our customers may
accept some variation, when variation is too extreme, our
customers will be dissatisfied.
D M A I C
21
Using Data To Understand Variation
Plot The Data Using Variation Tools

Study Variation For A Period Of Time Study Variation Over Time
Histogram Run Chart
Measurement
Frequency
Measurement Time
For Continuous Data For Discrete Data For Continuous Data For Discrete Data
– Histogram – Bar Chart – Run Chart – Control Charts

– Box Plot – Pie Chart – Control Chart – Run Chart
– Histogram
Variation is a fact of life. There will always be

fluctuations in our processes.
Graphically displaying measures provides a basic

description of the variation and its sources.
Following any data collection effort, the first step toward

understanding variation is to plot the data.
D M A I C
22
Review–Continuous vs. Discrete Data
Reminder: Data Type Is Critical!

Data type dictates how much variation we will see:
Continuous data–the most information about variation in the process
Discrete data–less information about variation in the process
Application Cycle Time

Upper Specification Limit = 30 Days
Continuous
Y = days to process
Discrete
Y = late/on-time USL
Actual Times
No. Rec’d No. Late

28 23 13 34 24 29
21 16 24 11 49 21
30 2
21 25 26 27 27 29
30 29 30 20 10 30
12 11 27 23 24 28
17 9 30 29 29 28
Less variation information The most variation information
Continuous data holds several advantages over discrete

data:
– It gives us more information about our process.
– Smaller sample sizes are required when we use
continuous data.
If you were the customer:

– Which type of data would you want?
– Which type of data shows what you see?
Consider this: If you were the process owner which

would you prefer?
– Simply how many applications were processed
within the customer specification?
OR
–The exact time it took to process each application?
D M A I C
23
5 M’s & 1 P
Sources Of Variation
Machines P
Methods R
Materials O
Measurements
C
E
Mother Nature
S
People
S
Machines–The various appliances used in the transformation In a training session, failure to regulate the thermostat
from inputs into outputs. For example, a PC can turn various can result in a non-conducive learning environment.
sources of information into an organized manual that then
relates to a training service. People–The staffing that influences customer needs
and requirements. While often dominant in the
Methods–The procedures, formal or otherwise, that transform service industry, this is an area still too often blamed
inputs into outputs. For example, there is a standard procedure for failure to meet or exceed customer requirements.
for billing collections in the GE Capital businesses.
Sometimes in a service environment, the following
Materials–The components, tangible or otherwise, that are categories are used:
transformed from inputs into outputs. For example, paper stock
and ink quality will affect a product brochure’s quality. — Policies–Higher level decision rules or
management practices.
Measurements–The tools that monitor a process’sperformance. — Procedures–The way in which tasks are performed.
For example, a doctor’s blood pressure reading of a patient — Plant–The building, equipment, work space, and
would determine subsequent activity related to treatment. environmental factors that affect performance.
— People–The human element.
Mother Nature–The environmental elements within a process
that influence a customer need or requirement.
— Parts–Systems, documents, and other supplies that
are needed to perform the service.
D M A I C
24
Two Types Of Variation
Common Versus Special Causes
Type of Variation Characteristics

Characteristics
Always Present
Common Cause Expected
Predictable
Normal
Not Always Present

Special Cause Unexpected
Unpredictable
Not Normal
To distinguish between common and special causes variation, use display

tools that study variation over time such as Run Charts and Control Charts.
There are two types of variation: The distinction between common and special causes is
important to determine the basic strategy for process
Common causes
improvement and control.
Special causes
Common causes are characteristics of the process and

exist as a result of the presence and interaction of
different process variables. Common causes affect
everyone working in the process, and affect all of the
outcomes. Common causes are always present and thus
are predictable within bounds.
Special causes are those causes that occur due to

occasional extraordinary circumstances. Special causes
are not always present, do not affect everyone working in
the process, do not affect all of the outcomes, and are not
predictable.
D M A I C
25
Describing Variation For A Period Of Time: Data Distributions
Key Questions:
What is the shape of the distribution–symmetrical, skewed,
twin peaks, flat?
What is the central tendency (“center”)
of the distribution?
What is the variation (“spread”) of the distribution–wide or
narrow?
When describing variation, we usually focus on two

things:
– Center
– Spread
D M A I C
26
Statistics
Statistics is concerned with making inferences about general

populations and about characteristics of general populations
We study outcomes of random experiments
If a particular outcome is not known in advance, then we do not
know the exact value assigned to the variable of that outcome:
– The number of invoices received weekly
– The cost in dollars of reworking each part
– The number of surfaces that are rough on a cast part
– The number of calls received every Monday between the hours of 8-9
a.m.
We call such a value a random variable
D M A I C
27
Some Distributions
A random variable can be expressed in terms of a distribution
Uniform Distribution
P(X) Single roll of a die
X All permissible values P(X)

are equally likely
Triangular Distribution
Sums of a pair of dice
P(X)
Rapidly descending
P(X), no tails
X
D M A I C
28
Distributions
Normal Distribution
P(X)
Process/repair times
Error fluctuations about an

operating point
X
Exponential Distribution
Time between arrivals
P(X)
Time between random
(unrelated) failures
X Events with no memory from one

to the next
Processes may have many differently shaped population

distributions. The shapes may be uniform, symmetric,
skewed, “ramped,” “camel-backed,” exponential, normal,
and non-normal. The shapes may be mixtures of normal
and non-normal distributions–or unmixed. Stable
distributions can have many different shapes.
Why does the process, or product, result in a particular

shape (distribution) and not some other shape? The
answers to these questions may provide a better
understanding of the process and how to improve it. The
particular shape is not as important as why this shape and
why not some other shape? Why this distribution, why
not some other distribution?
D M A I C
29
Shape
Shape is the distribution pattern exhibited by the data

Assess shape using a histogram, or more precisely with a Normal
Probability Plot
Roughly Normal Distribution Center Skewed Distribution
7 Center 20
6
Frequency
5
Frequency
4
10
3
2
1
0 0
12 13 14 15 16 17 18 19 0 10 20 30 40 50 60 70 80 90
Bimodal Distribution
Spread Spread
6 Center Center
5
Frequency
4
3
2
1
0
7 9 11 13 15 17 19 21 23
The shape of a data set can be determined by examining a

histogram, or, if testing the distribution for normality,
using a “Normal Probability Plot.” There are many
different shapes a data set may assume. The plots above
illustrate three common shapes. The top left is a normal
distribution, the middle is a bimodal distribution, and the
top right plot is a skewed distribution.
Because of its predictive qualities, the normal distribution

is of special interest.
D M A I C
30
The Normal Curve
Is The Data Distribution Normal?
Definition:
A probability distribution is where the most frequently occurring value is in the
middle and other probabilities tail off symmetrically in both directions.
Characteristics:
The curve does not reach zero
The curve can be divided in half with equal pieces falling either side of the most
frequently occurring value
The peak of the curve represents the center of the process
The area under the curve represents 100% of the product the process is capable of
producing
The normal curve is a probability distribution that forms

the basis for many decisions we will make about our
processes. The curve is noted by its “bell-shaped” nature,
where most of the values fall in the middle and fewer
values fall in either direction. The curve has several
important characteristics:
Infinity–The curve theoretically does not reach zero. It
goes out to infinity in either direction.
Symmetry–Roughly half of the values will be above
the average, and half below.
Centering–The peak of the curve tells us where the
process is centered.
Process Totality–The area under the curve represents
virtually 100% of whatever we are measuring.
D M A I C
31
The Normal Curve (continued)
Specific Characteristics
34.13% 34.13%
13.60% 13.60%
2.14% 2.14%
0.13% 0.13%
-3s -2s -1s X +1s +2s +3s

68.26%
95.46%
99.73%
68.26% Fall Within +\- 1 Standard Deviation

The normal curve can be divided into a series of Most common statistical terms and analysis tools are
segments. Each segment is mathematically called a based on a normal data distribution. If we use these tools
standard deviation from the mean. It is also noted by the with a data set that is not normal, the accuracy of the
small s. tools may be compromised.
As you can see, the curve is first segmented into one

standard deviation which represents approximately 34%
of whatever you are measuring.
Because the curve is symmetrical, going one standard

deviation in the other direction represents approximately
68% of whatever it is you are measuring. Going out +/– 2
standard deviations is equal to approximately 95% of
whatever you are measuring and +/– 3 standard
deviations is equal to 99.73% of whatever you are
measuring.
D M A I C
32
Normal Probability Plot
We use a normal Probability Plot to determine if our data is

normal Normal Probability Plot
.999
.99
.95
Probability
.80
.50
.20
.05
.01
.001
2 12 22 32
Cycle Time
Average: 16.3921 Anderson-Darling Normality Test
StDev: 5.61675 A-Squared: 0.208
N: 240 P-Value: 0.864
Use this method to determine if If p>0.05, then data is normal

distribution is normal
What Is A Normality Test? When Should I Use A Normality Test?
A normality test is a statistical process used to determine There are two occasions when you should use a
if a sample, or any group of data, fits a standard normal normality test:
distribution. A normality test can be done mathematically — When you are first trying to characterize raw data,
or graphically. normality testing is used in conjunction with
graphical tools such as histograms and box plots.
A normality test can be thought of as a “litmus test” for — When you are analyzing your data, and you need to
determining if a distribution is a normal distribution. calculate basic statistics such as Z values or to
employ statistical tests that assume normality, such
The Y axis is the cumulative % of data points which fall as t-Test and ANOVA.
below the value on the X axis.
Interpreting A Normal Probability Plot
Why Is A Normality Test Useful?
— When plotted data follows a straight line, the
Many statistical tests (tests of means and tests of Anderson-Darling p-value will exceed 0.05 and will
variances) assume that the data being tested is normally increase, I have normally distributed data.
distributed. A normality test is used to determine if that
assumption is valid. — Rule: p > .05 indicates data is normal.
D M A I C
33
Normal Probability Plot (continued)
Distribution Type:
Normal Bimodal curve Skewed Long-Tailed
Roughly Normal Distribution Bimodal Distribution Skewed Distribution Long-Tailed Distribution

7 6 20 6
6 5 5
Frequency
5
Frequency
Frequency
4
Frequency
4
3 10
3 3
2 2
2
1 1
1
0
0
12 13 14 15 16 17 18 19 0 0
7 9 11 13 15 17 19 21 23
0 10 20 30 40 50 60 70 80 90 7 9 11 13 15 17 19 21 23
Normal Probability Plot for a Normal Probability Plot for a Normal Probability Plot for a Normal Probability Plot for
Normal Distribution Bimodal Distribution Skewed Distribution Long-Tailed Distribution
99 99
ML Estimates ML Estimates
ML Estimates
Mean: 14.6382
Mean: Mean: 15.0790
95 95
StDev: 5.47084
StDev: 90 StDev: 12.6232 90
80 80
70
Percent
Percent
70
Percent
60 60
Percent
50 50
40 40
30 30
20 20
10 10
5 5
1 1
0 10 20 30 0 10 20 30
How Distribution Looks On Normality Curve:
Straight line Zig-zag Two lines “S” curve

(Stable Operations)
The Normal Probability Plot is another way besides the

histogram to plot data and look for normality. Normal
data, when plotted with the data value on the X-axis and
specially spaced percentiles of the normal distribution on
the Y-axis, will fall on a straight line. 95% confidence
limits are shown around the line.
In the top diagram on the left, we see that all of the data
points roughly form a straight line and fall within the
limits. We would conclude that there is no serious
departure from normality in this data. In the other
diagrams, we see that many of the data points fall outside
the limits and do not form a straight line. We would
therefore conclude that these data sets significantly depart
from the normal distribution.
What types of process might result in data that follows a

bimodal, skewed, and long-tailed distribution?
D M A I C
34
What If Your Data Is Not Normal?
If you conclude that Y is non-normally distributed, there are

two general approaches:
Approach 1: Variance-based Thinking (VBT) Methodology
– possibly multiple processes embedded
– segmentation and stratification
– span reduction
Approach 2: Non-parametric techniques (beyond the scope of this course)
Expectation: Green Belts Should

Be Able To Do Approach 1
If your data is non-normal, you may be able to proceed. It

is important that you consult with your MBB before
proceeding.
D M A I C
35
Minitab–Histogram

Datafile: Fill_time.mtw
Question: What does the distribution of cycle 30
time to fill positions look like?
Frequency
20
Tool: Histogram
10
Data Type: Continuous. The X axis is cycle time,
the Y axis is the number of times
cycle time fell within the range of each 0
bar or interval in the histogram. The Y 0 10 20 30
axis is calculated for you.
Cycle Time
Data A single column of data for each
Structure: histogram
To Make A Graph > Histogram

Histogram: Variable: Cycle Time
Click OK
D M A I C
36
Minitab–Normal Probability Plot

.999
.99
Question: Are the cycle time data .95
normally distributed?
Probability
.80
.50
.20
Tool: Normality Test
.05
.01
Data Type: Continuous Y .001
2 12 22 32
Data Structure: A single column of data Cycle Time

Average: 16.3921 Anderson-Darling Normality Test
N: 240 P-Value: 0.864
To Make A Normal Stat > Basic

Probability Plot: Statistics > Normality Test
Variable: Cycle Time
Test For Normality:
Choose Anderson-Darling
Click OK
Since p > 0.05, the data is normal.
There are a few ways to determine normality. We have

standardized on this Anderson-Darling normality test. We
will discuss p-values later in the Analyze phase.
D M A I C
37
“Center” Or Central Tendency
Descriptive Statistics: Normal Distribution

Represents the nominal value
of the process.
Mean (X)
x
Median (“middle” data point)
Long-tailed Distribution
Quartile Values (Q1, Q3)
Skewed Distributions
Q1 Q3
A good guess of central tendency can be made by

visually examining a histogram.
A more precise estimate of central tendency can

be made using descriptive statistics such as the mean,
median, or quartile values. The distribution shape dictates
which metrics should be used.
D M A I C
38
“Center” Or Central Tendency (continued)
The Mean, sometimes called the average, is

the most likely or expected value. The formula
for the mean is:
X=
∑X i
1. The sum of all data
values
n 2. Divide by number of data

values
The Median is literally the middle of the data

set where 50% of the data is greater than the
median, and 50% of the data is less than the
median. The most commonly used symbol
~
for the median is X. The procedure for
calculating the median is :
Order the numbers from smallest to

largest
If the number of values (N) is odd, the
median is the middle value. For example,
if the ordered values are 3, 4, 6, 9, 20, the
median is 6.
If the number of values (N) is even, the
median is the average of the two middle
values. For example, if the ordered values
are 1,5,8,9,12,18, the median is 8.5.
For very skewed data, we can describe the
central tendency in terms of the quartile
values, Q1 or Q3.
Q1 is the data point that divides the lowest
25% of the data set from the remaining 75%
and is used to describe performance when the
data is skewed toward the right.
Q3 is the data point that divides the highest
25% of the data set from the remaining 75%
and is used to describe performance when the
data is skewed toward the left.
D M A I C
39
Variation
Descriptive Statistics Normal Distribution
Represents the variation of s

the process
Standard Deviation (s)

Span Long-tailed Distribution
Stability Factor (SF) = Q1/Q3
p=.01 p=.99
Skewed Distributions
Q1 Q3 Q1 Q3
A question to be answered in describing variation is

“What is the spread of the data?” That is, how much
variation exists in the data? By examining a histogram, a
good guess of variation can be made.
A precise measure of variation is accomplished using

basic descriptive statistics such as standard deviation,
span, or stability factor.
Standard Deviation–measures average distance from the

mean. Smaller is better.
Span–measures unusually low & high distance from the

median. Smaller is better.
Stability Factor–ratio of first quartile and third quartile,

1.0 is best.
D M A I C
Variation (continued) 40
Normal Distributions: Skewed Or Stable Ops Distributions:

The standard deviation is the average distance, or The stability factor (SF) is the ratio of the first
quartile and third quartile values. The formula for
deviation, that a given point is away from the
Stability Factor is:
mean. The formula for the standard deviation is:
Xi-X
SF = Q1/Q3
As Q1 and Q3 get closer together, their ratio
decreases and the variation of the middle 50%
A- subtract decreases.
each data
value from
the mean Xi X
Q Q
1 3
E- take the
∑ (X − X )
square root of
2 B- square each
the result
i difference Q Q
1 3
n −1 C- sum the
squared
differences
D- divide by 1 less than the
number of data values
Q More Q
1 1 Less
0 Variatio 1 Variatio
Q n Q
3 3 n
Long-tailed Distributions: 0 More Variation Less Variation 1
Span is used to indicate the amount of variation in a

long-tailed distribution. It is the distance between the
two extremes of the data set. For example, in customer
delivery span, it is the number of days between the
earliest and latest delivery. Because we don’t want span
to be determined by only one or two data points, we
typically use P95 and P5 (although this changes
depending upon sample size). The span is the
difference between P95 and P5.
The guidelines of what span to use varies with the
sample size:
Sample Size Span P

100-500 90%-10% P90, P10
501-2000 95%-5% P95, P5
2001-5000 98%-2% P98, P2
5001+ 99%-1% P99, P1
D M A I C
41
The Computational Equations
Population N
Mean ∑ Xi
i =1
µ=
N
Population N
∑ (Xi − µ)
2
Variance 2 i =1
σ =
N
Population
Standard N
∑ (Xi − µ)
2
Deviation σ= i =1
N
n
Sample
∧
∑X i
Mean µ=X= i =1
Sample ∑ (X
n
− X)
2
Variance ∧ i =1
i
σ 2 = s2 =
n−1
Sample
∑ (X − X)
n
Standard 2
∧ i =1
i
Deviation σ = s=
n−1
The difference between population and sample:

population has all the data points, N
sample only has a portion of the total data points,
n<N
The divisor for the population variance is the population
size N, whereas the divisor for the sample variance is the
sample size minus one (n-1). The divisor (n-1) is used
rather than N because this leads to an unbiased estimate
for the population variance.
D M A I C
42
Deviation
Deviation==Measures
Measures Of Variation
Of Variation
Random data (“1” to “99”, n = 10 samples):
n Xi Xbar Dev =(Xi-Xbar) (Xi-Xbar)^2
1 20 35.7 -15.70 246.49
2 1 35.7 -34.70 1204.09
3 6 35.7 -29.70 882.09
4 35 35.7 -0.70 0.49
5 54 35.7 18.30 334.89
6 67 35.7 31.30 979.69
7 43 35.7 7.30 53.29
8 99 35.7 63.30 4006.89
9 5 35.7 -30.70 942.49
10 27 35.7 -8.70 75.69
----- ------ ---------- --------------
357 0.0 8726.10
Results & Take-Aways:

Deviation = measure of Variation, measure of Spread
SST = 8726.1 … large Deviations = large SST
Variance = “Average” SST = SST/df [df = deg. of freedom = n-1]

s2 = Sample Variance = SST/9 = 969.6
σ2 = Population Variance = SST/10 = 872.6
Standard Deviation = Square Root of Variance:

It’s All Variation… Deviation, SST, Variance,
Sample StDev = s = 969.6 = 31.1
Population SD = σ = 872.6 = 29.5
And Standard Deviation = Measures Of Spread
A key concept is that the variance is the average SST for

a population and a near average for a sample = SST/(n-1)
= SST/df.
As the degrees of freedom (df) increase (which is the

same as larger sample size (n)), the near average and
average SST become essentially the same.
Finally, the standard deviation is the square root of the

variance.
D M A I C
43
Summary–Measures
Summary–Measures Of Variation
Of Variation:
Range = (Max - Min)

Deviation = (Xi - Xbar) [Xbar = mean or mu]
Sum-of-Squares (of the squared Deviations) = SST
Variance = Average SST = SST/df (df = degrees of freedom)
Std Deviation = SqrRoot (Variance) and
Coeff. of Variation = CV = (StDev/Xbar)* 100
i.e., ratio of StDev to Mean–expressed as %
SST is key concept in measuring variation.
The greater SST, the greater the variation.
SST is always conserved: given a set of numbers (data), it is fixed

and unchanging.
SST can be allocated to different COMPONENTS OF VARIATION–

i.e., within and between group.
Components Of Variation: The Key For Process

Centering And ANOVA
Note that for any given set of numbers (Xis), the value
for SST is fixed, unchanging. The only way to change
SST is to change one or more of the numbers–which
means that the spread around the mean will likely be
changed.
D M A I C
44
Variation For A Period Of Time
Descriptive Statistics Summary
Shape Normality Plot Center Spread

(central tendency) (variation)
Normal Probability Plot for a Normal Distribution
ML Estimates
Mean:15.7224
StDev:1.74183
Standard
Percent
Mean X( ) Deviation (s)

normal
Normal Probability Plot for an Exponential Distribution
ML Estimates
Mean: 15.0790
StDev:12.6232
Quartile Stability Factor

Percent
Q1 or Q3 (SF)
skewed
Normal Probability Plot for a Long-Tailed Distribution
99 ML Estimates
Mean: 14.6382
( )
95
StDev:5.47084
~
90
Percent
80
70
60
50 Median X Span
40
30
20
10
5
long-tailed 1
0 10 20 30
Normal Probability Plot for a Bimodal Distribution
99 ML Estimates
The different processes must be stratified

Mean: 14.6382
95
StDev:5.47084
90
Percent
80
70
60
50
40
before descriptive statistics
can be calculated.
30
20
10
5
bimodal 1
0 10 20 30
Q1= 25th Percentile

Q2 = Median,, 50th Percentile
Q3 = 75th Percentile
Range = Maximum Value–Minimum Value
Span = 99th Percentile–1st Percentile
SF = Q1/Q3
IQR = Inner Quartile Range = Q3-Q1 (middle 50th of
data)
D M A I C
45
Displaying Variation For A Period Of Time
Histogram
Measurements Graphical Display
Time Time Histogram of Time Estimates
Estimates Estimates
Round 1 Round 2
16.48 13.84 # of Occurrences 10
18.89 13.50
13.18 15.41
11.11 14.35
14.67 14.37
16.53 14.63 5
14.79 13.58
18.06 14.75
14.48 11.95
14.89 14.36 0
15.63 16.17
10 11 12 13 14 15 16 17 18 19
13.95 15.15
13.74 12.48 Time Estimates (in seconds)
17.67 14.12
10.23 19.00
13.67 13.81
11.35 12.97
15.03 14.19 Illustrates
Shape (pattern) of the data
Central tendency (center) of the data
Variation (spread) of the data
After data is collected, it should be displayed graphically

and described using basic statistical measures. The first
graphical display used for describing variation is the
histogram.
Histograms combined with basic descriptive statistics

provide a clear picture of the current performance of a
process.
D M A I C
46
Displaying Variation For A Period Of Time (continued)
Box Plots
* Outlier
Highest Value
Third Quartile (75%) value Q3

Each segment
represents
Median Q2
25% of the
data points First Quartile (25%) value Q1
Lowest Value
Box Plots are graphical summaries of the patterns of

variation in sets of data.
The horizontal lines at the top and bottom of the plot

represent the highest and lowest values. The horizontal
line in the middle of the solid box represents the median,
and the solid box represents the middle 50% of the data,
with 25% of the data on the top side of the box and 25%
of the data on the bottom side of the box.
Points noted with an asterisk(*) represent extreme values

or outliers.
Outlier calculations
High: Q3 + 1.5 (Q3-Q1)
Low: Q1 - 1.5 (Q3-Q1)
D M A I C
47
Summary–Variation For A Period Of Time
10
Data
28 23 13 34 24 29
21 16 24 11 49 21 Histogram
21 25 26 27 27 29
30 29 30 20 10 30 5
12 11 27 23 24 28
17 9 30 29 29 28
10 11 12 13 14 15 16 17 18 19
Box Plots and histograms provide similar information but

in different graphical formats.
D M A I C
48
Minitab–Graphical Summary

Question: How can I examine the shape,

central tendency, and variation Descriptive Statistics
(spread) of my data using one Variable: Cycle Time
Minitab command?
Anderson-Darling Normality Test
A-Squared: 0.208
P-Value: 0.864
Tool: Descriptive Statistics Mean

StDev
Variance
16.3921
5.6167
31.5478
Skewness 6.39E-02
Kurtosis -2.4E-01
N 240
4 10 16 22 28
Data Type: Continuous Y Minimum

1st Quartile
3.0800
12.4067
Median 16.6400
3rd Quartile 20.3067
95% Confidence Interval for Mu Maximum 32.2667
95% Confidence Interval for Mu
To Make A Stat > Basic Statistics > Display 16.0 16.5 17.0 17.5
15.6779 17.1063
95% Confidence Interval for Sigma
Histogram: Descriptive Statistics: 5.1552 6.1698
95% Confidence Interval for Median
Variable: Cycle Time 95% Confidence Interval for Median

15.7274 17.3441
Click Graphs button

Select Graphical Summary
Click OK twice–once in each
dialog box
This graphical summary is one of the most widely used Skewness–A measure of asymmetry. A value more
charts in Minitab. than or less than zero indicates skewness in the data.
Useful Items: But, a zero does not necessarily indicate symmetry.
Anderson-Darling normality TEST–We use the p- Kurtosis–A measure of how different a distribution is
value to determine whether our distribution is normal from the normal distribution. A negative value
or non-normal. If p > 0.05, the distribution is normal. typically indicates a distribution more peaked than the
(we will discuss p-value & hypothesis testing in normal. A positive value typically indicates a
Analyze, Step 6.) distribution flatter than normal.
Mean–Average of the samples. N–sample size.
StDev–Standard deviation of the samples. Confidence Interval for Mu–Tells us the interval in
which the true population mean lies (with 95%
Variance–(Standard Deviation Squared.) A measure of confidence.)
how far the data are spread about the mean.
D M A I C
49
Variation Over Time
Run Chart
A graphical tool to monitor the “stability of Project Y
Allows observation of time order properties such as trend
Should be used before any detailed data analysis
Example of a Run
Chart
Median
Is the process stable over time?
An important basic tool for understanding variation is the

Run Chart. Run Charts are simple time-ordered plots of
data. On these plots one can perform tests for certain
patterns in the data. Presence of these patterns indicate
special causes.
It is important to note that the data should be in time

order for run charts to be valid.
D M A I C
50
Run Charts–Special Cause Patterns
If p < 0.05, then there is significant statistical evidence to show that one of the
trends below exists.
Mixture Cluster
Oscillating Trend
Minitab Run Chart Oscillating Pattern
The Minitab Run Chart tests for two kinds of Runs: Oscillation represents more than expected number of
(A) Consecutive points on the same side of the median. runs of type (B).
(Thus a new run starts when the median is crossed).
(B) Consecutive points in the same direction up or down. Trend Pattern
(Thus, a new run starts when the direction is changed).
A trend represents less than expected number of runs
of type (B).
Mixture Pattern
At this point in time, our biggest concern is whether
A mixture represents more than expected number of runs the process is stable or not. We use the above-
of type A (above). mentioned checks to indicate stability. Minitab will do
this for us.
Cluster Pattern
A cluster represents fewer than expected number of runs

of type (A).
D M A I C
51
Minitab–Run Chart

Datafile: Fill_time.mtw Run Chart for Cycle Time
Question: Does cycle time show any 32
patterned variation over
time?
Cycle Time
22
Tool: Run Chart 12
Data Type: Continuous Y 2
40 140 240
Observation
Data Structure: A single column of data is all that Number of runs about median: 124.000 Number of runs up or dow n: 151.000
is necessary. It is assumed that Expected number of runs:
Longest run about median:
121.000
9.000
Expected number of runs:
Longest run up or dow n:
159.667
4.000
the data is entered in the order in Approx P-Value for Clustering:
Approx P-Value for Mixtures:
0.651
0.349
Approx P-Value for Trends:
Approx P-Value for Oscillation:
0.091
0.909
which it is collected (time
order).
To Make A Stat > Quality Tools > Run Chart

Run Chart: Single Column: cycle time
Subgroup size: 1
The 4 p-values circled above indicate stability. Since

p>0.05, there are no issues with Mixtures, Clusters,
Oscillations or Trends. Therefore, our process appears
stable (no special causes).
D M A I C
52
Two Types Of Variation
Investigating Common vs. Special Causes

For new process data, use a Run Chart to look for special causes
Investigate special cause points for positive quick-fixes
Common cause variation requires systematic improvement
effort
For new process data, a Run Chart can be used to detect

special cause situations. For processes that are already in
statistical control, Control Charts are the preferred
method. Control Charts are discussed in the Control
phase of DMAIC training. But, here in the Measure and
Analyze phase we suggest using the Run Chart to test for
stability.
D M A I C
53
Two Types Of Variation (continued)
Reacting To Common vs. Special Causes

How you interpret variation . . .
Common Causes Special Causes
Focus on systematic Mistake 1

Common Tampering
process change
Causes (increases variation)
True
variation
type... Investigate
Mistake 2
Special Under-reacting special causes
for possible
Causes (missed prevention)
quick-fixes
Why is it important to know the source of variation and treat it Treating special causes as common causes is essentially under-
according to the appropriate strategy? Because not reacting reacting to individual data points. The problem with making
appropriately to the type of variation present in a process can this type of mistake is missing an opportunity to prevent defects
seriously impact customer satisfaction and the amount of and reduce process variation. An example of this type of
variation and defects, and it can increase costs. mistake would be failing to correct the steering on a car
heading for a ditch, and ending up in the ditch.
When interpreting variation, you can make two types of Appropriately reacting to the source of variation in a process
mistakes: provides the correct economic balance between over-reacting
Treating common causes as special causes and under-reacting to variation from a process.
Treating special causes as common causes
Treating common causes as special causes is essentially over-

reacting to individual data points and is often referred to as
“tampering.” The problem with making this type of mistake is
that it typically leads to increased variation, costs, and defects.
It’s like a novice driver who over-reacts to “slop” in the steering
of a car. The result is more variation in the path of the car.
D M A I C
54
Graphical Analysis Tools
Looking For Patterns In Data
Continuous Y Discrete Y
Boxplot
Pareto Chart
Scatterplot
D M A I C
55
Box Plots
What differences do you see between the output from the

different shifts?
60
Measure
30
10
Shift 1 Shift 3 Shift 4 Shift 6

Shift 2 Shift 5
Box plots can be used to identify differences in variation

between subgroups that exist in your data.
Look for 2 major items here:
Center–are the medians similar?
Spread–do the boxes and the “Whiskers”have similar
heights?
D M A I C
56
Minitab–Box Plots

Question: Is there a difference in variation

in cycle time between agents?
30
Tool: Box Plot

Cycle Time 20
Data Type: Continuous Y, Discrete X

10
To Make: Graph > Box plot

A Boxplot
Graph variables: Y Cycle time 0
1 2 3 4
X Agent
Agent
Click OK
Additional Questions: Is there a difference in
variation in cycle time between sites? Between
types of resource?
D M A I C
57
Scatter Diagrams–Analyzing Relationships
Use Scatter Diagrams To Study The Relationship Between

Two Variables.
40
Cycle Time (Days) (Y)
35
30
25
20
15
10
5
1K 2K 3K 4K 5K 6K 7K 8K 9K 10K
Size Of Loan (X)
A Scatter Diagram is an important graphical tool for

exploring the relationship between two continuous
variables.
Here, it appears that size of Loan and Cycle Time have a

relationship (Larger Loans take longer to process–
positive relationship).
D M A I C
58
Warning! Correlation Does Not Imply Causation
Correlation Between Number Of Storks And Human

Population
100 200 300
80 80
70 70
Population
(In Thousands)
60 60
50 50
100 200 300
Number Of Storks
Source: Box, Hunter, Hunter. Statistics For Experimenters. New York, NY: John Wiley & Sons. 1978
This is a plot of the population of Oldenburg at the end of

each year against the number of storks observed in that
year, 1930-1936.
Even strong correlations do not imply causation. (For

example, there will likely be a positive correlation–but
not causation–between the occurrence of vaporlocks in
automobiles and the use of public swimming pools.)
D M A I C
59
Common Construction Mistakes
Examples
Problem Variable Axis Variable Axis
# Of Errors In
Bank
Size Of Loan Closing Cost
Errors
Estimate
Length Of Time To
Loan Delays # Of Changes
Approve Loan
Computers Are Computer’s Computer’s System

Too Slow Response Time Load
Delays Unloading # Of People On Time To Unload

Airline Baggage Plane Baggage
One of the most common mistakes that occurs with Scatter

Diagrams is mixing up the X and Y variables.
The X variable is the potential cause and is plotted on

the horizontal axis.
The Y variable is the effect and is plotted on the vertical

axis.
D M A I C
60
Interpretation Of A Scatter Diagram
Look For:
Common patterns in the data
Range of the predictor variable (X)
Irregularities in the data pattern
Common Patterns In The Data

See whether the potential cause variable and the effect
variable are related to one another.
Range Of The Predictor Variable (X’s)

The range is the difference between the largest and
smallest values.
Check that the range of the potential cause variable is
wide enough to show possible relationships with the
effect variable.
Irregularities In The Data Pattern

Check whether the data pattern indicates possible
problems in the data.
D M A I C
61
Interpreting A Scatter Diagram
Look For Patterns

1 3 5
Strong Positive Correlation Strong Negative Correlation No Correlation
2 4 6
Positive Correlation Negative Correlation Other Pattern
For all charts: Y = Participant satisfaction (scale: 1 – worst to 100 – best)

X = Trainer experience (# of hours)
Interpretation Of Patterns: As trainer experience increases, participant satisfaction decreases

Positive Relationship: X Increases/Y Increases No Relationship: For any value of X, there are many values of Y
Graphs 1 and 2–as X increases, Y increases Graph 5–there is no relationship between X and Y
Graph 1–Strong, positive relationship between No one line best describes the data
X and Y There is no relationship between trainer experience and
– Not much scatter participant satisfaction
– Points form a line with a positive slope Non-Linear Relationship–The relationship between X and Y is
Graph 2–Weaker, positive relationship between X and Y complex and cannot be summarized with a straight line.
– Scatter is wider Graph 6–Inverted u
– Points still form a line with a positive As trainer experience increases, participant satisfaction also
slope increases until a critical threshold is reached, at which point
participant satisfaction decreases. This is not a linear relationship.
As trainer experience increases, participant satisfaction also
increases.
Negative Relationship: X Increases/Y Decreases
Graphs 2 and 3–as X increases, Y decreases
Graph 3–strong negative relationship between X and Y
– Not much scatter
– Points form a line with a negative slope
Graph 4–weaker negative relationship between X and Y
– Scatter is wider
D M A I C
62
Minitab–Scatter Plot

Question: Is there an association

between cycle time and
years experience? 30
Tool: Plot
Cycle Time 20
Data Type: Continuous Y, Continuous X 10
To Make A Graph > Plot

0
Scatter Plot: Y Cycle time
0 5 10 15 20 25
X Years
Yrs Exp
Experience Click OK
D M A I C
63
Pairs Exercise–Analyze Patterns
With A Partner:
Review scatter diagrams on following
pages as assigned
Determine the type of relationship you
observe (e.g., negative, positive, strong, weak, etc.)
Describe an example from your business
of this type of relationship
D M A I C
Common Scatter Diagram Patterns (continued) 64
Strong,
Plot +/- Other Weak, Other
Example
1
Effect
Potential Cause
2
Effect
Potential Cause
3
Effect
Potential Cause
D M A I C
Strong, Weak,
Plot +/- Other Other
Example
4
Effect
Potential Cause
5
Effect
Potential Cause
6
Effect
Potential Cause
D M A I C
Plot +/- Other Strong, Weak, Example

Other
7
Effect
Potential Cause
8
Effect
Potential Cause
9
Effect
Potential Cause
D M A I C
Plot +/- Other Strong, Example

Weak, Other
10
Effect
Potential Cause
11
Effect
Potential Cause
12
Effect
Potential Cause
D M A I C
68
Pareto Charts
Is There A Defect That Occurs Frequently?
Frequency
C A E D B
Category of Defect
We were introduced to Pareto charts in the CTQ tools

module.
D M A I C
69
Minitab–Pareto Chart

Question: Which agent has the most cycle
time defects? Pareto Chart for Agent
Y
Tool: Pareto Chart
180
Data Structure: A single column where the 160
defect categories are recorded 140
OR 120
Percent
Count
100
A tally table in two columns. 80
One column contains the Type 60 100
80
of Defect, the second column 40 60
contains the frequency of the defect. 20 40
20
0 0
To Make A Stat>Quality Tools>Pareto Chart Defect
Pareto Chart: Chart Defects Data In: Agent
4 1 2 3
By Variable: Cycle _Time_Defect Count

Percent
35
56.5
12
19.4
8
12.9
7
11.3
Choose: One Chart Per Page, Cum % 56.5 75.8 88.7 100.0
Independent Ordering of Bars

Click: OK
Additional Which site has the most cycle

Questions: time defects? Which type of
resource has the most
cycle time defects?
To find out which site has the most defects: Follow steps Alternate method (subsetting a worksheet)
above except, place “Site” in “Chart Defects Data In” Box. MANIP > SUBSET WORKSHEET
Name: No Transactions
Alternate Method (Splitting A Worksheet)
Include or Exclude: CLICK ON “Specify which rows
to include”
MANIP > SPLIT WORKSHEET
Specify which rows to include:
By variable: Cycle_Time_Defect CLICK ON “Rows that Match”
Click: OK
CLICK ON “Condition” Button
Now you have 2 worksheets-one with “N” transactions Condition: Double Click on C6 then = “N”
and one with “Y” transactions (‘cycle_time_defect’) = “N”
Activate the correct worksheet and then follow the steps Click “OK” twice
in the slide above
Now you have a worksheet with “N” transactions
D M A I C
D M A I C
1
Measure
Define 3–Plan
ReviewData Collection And Collect Data
Objectives 1
What does it mean to Establish A Data Collection Plan, Validate the

Measurement System, and Collect Data?
A project data collection plan is a written strategy for collecting the data you will use
in your project. A validated measurement system is one that has been shown to
provide reliable data that represents the process output.
Why is it important to Establish A Data Collection Plan, Validate the

Measurement System, and Collect Data?
A project data collection plan and a validated measurement system are important
because they define a clear strategy for collecting reliable data efficiently.The data
collection plan helps ensure that resources are used effectively to collect only data
that is critical to the success of the project. A validated measurement system is
important because it ensures that the collected data accurately represents the true
nature of your process. Data collection is costly; therefore, you need to make sure
that both the data collection plan and the measurement system are sound.

3.1 Develop a plan to collect data
3.2 Validate the measurement system
3.3 Collect data per plan
MEASURE STEP
OVERVIEW

Standards MSA
3.1 Develop a plan to collect data

3.2 Validate the measurement system
3.3 Collect data per plan

DMAIC GB J TX PG V 4.2.0
D M A I C
Measure 3–Plan data collection, validate measurement
system, and collect data
Data Collection
Define Review Plan Worksheet
Objectives
2
Data collection purpose:
What to measure (output measure and segmentation factors)

Data Type Operational
Measure Measure Range Of
(Discrete)/ Definition
(Name) Type (Y, X) Values
(Continuous)
How to measure:
Sampling Plan (Scheme, Frequency, Size)
Measurement Procedure
Refer to more detailed documentation, if necessary.

D M A I C
Measurement System
Define Review Analysis (MSA) Worksheet
Objectives
3
Type of Gage R&R Conducted (Check One): Date

Gage R&R (Continuous Data) Conducted:
Gage R&R Results
1) Two-Way ANOVA Significant?
Part p-value Y N
Oper p-value Y N
Oper & Part p-value Y N
Pass?
2) % Tolerance [ <30% ] Y N
3) % Contribution [ <8% ] Y N
4) % Study [ <30% ] Y N
5) # Distinct Categories [ >4 ] Y N
Graphical Output OK?

1) Effective Resolution [ >50% ] Y N
2) Stability [R Chart] Y N
3) Consistency Between Xbar consistency Y N
Between Oper
4) Systematic Shift [Oper/Part Inter. Plot] Y N
Attribute Gage R&R

1) Repeatability [ >90% ]
2) % Reproducibility [ >90% ]
3) % Accuracy [ >90% ]
_______________________________________________________________
1) Gage R&R Pass? Y N, If NO:
Plan for improvement:
_______________________________________________________________
_______________________________________________________________
D M A I C

D M A I C
1
Plan For Data

Define Collection
Review Objectives
1
Develop
Establish Ensure Data Collect Data
1 Data Collection
Goals
2 Operational
Definitions And
Procedures
3 Consistency
And Stability 4 And Monitor
Consistency
Clarify purpose Write and pilot Test and validate Train data
of data collection operational definitions measurement collectors
Identify what systems
Develop and pilot Pilot process
data to collect
data collection forms and make
and procedures adjustments
Establish a sampling Collect data

plan
Monitor data
accuracy and
consistency
Data Collection Is The First Step To Understanding The

Variation The Customer Feels
Data collection occurs multiple times throughout

DMAIC. The data collection plan described here
can be used as the guide for data collection
regardless of where you are in DMAIC or what
type of data you are collecting. Using this model
will help ensure that you collect useful, accurate
data that is needed to answer your process
questions.
DMAIC GB K TX PG V 4.2.0
D M A I C
Step 1: Establish
Define Review Data Collection Goals
Objectives
2
In order to establish your data collection goals you must:

State the purpose of the data collection
Identify what data is required
Asking these questions may help you clarify your goals:

What do I need to know about my process?
What data do I need?
What is the plan for analysis once the data is collected?
What data is already available?
D M A I C
Preparing For TheObjectives

Define Review Analyze Phase 3
Segmentation:
– An analysis technique that involves temporarily dividing a large group of
data into smaller logical categories to look for areas of very good vs. very
poor performance
– Can be used to understand which X’s drive variation of the Project Y
Collect Project Y data to identify patterns and performance
trends and establish current baseline defect rate
Collect segmentation factor data to be used for later analysis
Segmentation Helps Us Understand

Variation In Project Y
Think ahead about how you plan to analyze the For example, in deal businesses where the Project Y
data. Collect additional information is the cycle time of closing the deal, a potential
corresponding to the Project Y that may be segmentation factor is client size (large, medium, or
helpful in subsequent analysis. small).
– Include possible external factors that may .
be useful for segmenting the total Y data
set.
– Consider how to collect continuous data
instead of discrete data.
Thinking about segmentation factors now allows
you to gather these conditions along with data
on your Project Y so they don’t have to be
reconstructed after the measurement is
completed.
Segmentation factors include external variables
outside our control that may explain patterns in
the Project Y measures.
D M A I C
Common Segmentation
Define Review Factors
Objectives
4
Segmenting by external factors will help us identify the

drivers of variation in the process
Possible categories:
Customer
Product
Market
Time
Geography
Size of account
Degree of dissatisfaction
What Data Is Needed? 2. Y = On-time delivery of services

Project Y data, collected over X time period Objective:
Examples: Measure on-time delivery performance for
1. Y = Application cycle time customer service center
Objective: Identify possible sources of poor or irregular
performance.
Understand cycle time performance of credit
approvals process. Data Required:
Identify potential sources of poor or irregular Deviations from target date/time for individual
performances. orders (minutes/hours, early or late)
Data Required: Data for segmenting factors [e.g., customer,
market, geography, time (day, month, hour),
Physical cycle time data (collect several request type]
business cycles) for individual credit
applications.
Data on possible segmentation factors external
to the process (e.g., customer information,
time/date applications received with product
type, processing center).
D M A I C
Common Segmentation
Objectives
5
Other Categories
Factor Example
What type Complaints, defects, problems
When Year, month, week, day
Where Country, region, city, work site
Who Business, department, individual,
customer type, market segment
Tip: Begin with factors “outside” the process box–often these are factors that were not considered
when the process was first designed
In service-related processes, “When” elements to

pay special attention to include: relationship to
deadline (e.g., end of quarter), time/number in
queue, or timing versus systems changes.
Note: Your team will need to segment the data in

several different ways in order to uncover where
the most significant differences occur.
D M A I C
How To Collect
Define Review Data For Segmentation
Objectives
6
Identify the factors for segmentation before you start collecting

data
Make sure the segmentation factors can be measured reliably
Record the segmentation factors for each Y data point
collected
Segmentation factors are typically easy to collect, so collect
more segmentation factors rather than fewer
D M A I C
Step 1: Breakout
Define Review Activity (5 Minutes)
Objectives
7
Objective Determine Segmentation Method
For your Project Y:

Instructions
– Brainstorm a list of segmentation factors
– Remember to also segment on “unlikely” parameters
Time 5 Minutes
D M A I C
Step 2: Develop
Define ReviewOperation Definitions and Procedures
Objectives
8
Clearly specify variables to be collected:

Operational definitions for all metrics
Specific descriptions of how to take the measurement
Specify the details of the data collection process:
How to collect the data
How to record the data
The period of time for data collection
The sampling plan to be followed
To help ensure consistency, clearly define each

metric and the process for collecting it. Attention to
these details will help ensure that the data you
collect will give you an accurate picture of the
variation in your process.
D M A I C
Operational Definitions
9
Defining The Measure

Definition An operational definition is a clear, concise description of a
measurement and the process by which it is to be collected
To remove ambiguity
– Everyone has a consistent understanding
Purpose
To provide a clear way to measure the characteristic
– Identifies what to measure
– Identifies how to measure it
– Makes sure that no matter who does
the measuring, the results are consistent
Always Pilot Your Operational Definitions
Operational definitions guide what properties

will be measured and how they will be
measured.
There is no single right way to write an
operational definition. There is only what
people agree to for a specific purpose. The
critical factor is that any two people using the
operational definition will be measuring the
same thing.
D M A I C
Operational Definitions–Scale
Define Review Objectives Of Scrutiny
10
Choosing The Level Of Measurement

Measure one scale or level smaller than what your customer
measures
For Example:
If your customer measures cycle time in days, your scale of
scrutiny would be hours
If your customer measures cycle time in hours, your scale of
scrutiny would be in minutes
Scale of scrutiny may expose larger true variation
The scale of scrutiny is how finely you measure

your process. Measuring one level smaller than
your customer allows you to more fully understand
and capture the variation in the process.
D M A I C
1
Sampling
1
Sampling is the process of:

Collecting only a portion of the data that is available or could be
available, and drawing conclusions about the total population
(statistical inference)
Population Sample
x x x
x x
x x
x x x x x x
x x x x x
x
x x x
N = 5,000 n = 100
What is the From the sample,

average we infer that the
resolution time? average
resolution time
( ) is 1.2 days
Sampling is the process of collecting a portion or

subset of the total data that may be available. All of
the data available is often referred to as a
population (N). The purpose of sampling is to draw
conclusions about the population using the sample
(n). This is know as statistical inference.
In the example above, the population is all of the

written inquiries received at a processing center last
month (5,000). The manager of the process wants
to know the average resolution time for the
inquiries received last month. Measuring the
resolution time for each inquiry is too expensive;
therefore, a decision is made to take a sample.
A random sample of 100 inquiries is made and the

average resolution is estimated.
DMAIC GB L TX PG V 4.2.0
D M A I C
When To Review
Define Sample Objectives 2
When to sample
Collecting all the data is impractical or too costly
Data collection can be a destructive process
When measuring a high-volume process
When not to sample
A subset of data may not accurately depict the process, leading
to a wrong conclusion (every unit is unique – e.g., structured
deals)
Statistically Sound Conclusions Can Often Be Drawn

From A Subset Of The Total Available Data
One of the first questions to ask is ‘Do I need to

sample?” The major reason sampling is done is for
efficiency reasons-it is often too costly or time
consuming to measure all of the data. Sampling
provides a good alternative to collect data in an
effective and efficient manner.
If the circumstances surrounding the data collection

plan do not justify sampling, then sampling should
not be done. This is often the case in low volume
processes (e.g., deal processes).
D M A I C
Goal Of AReview
Define Useful Sample
Objectives
3
Representative Samples
Representative sample:
All parts of the target population are represented
(i.e., selected for measurement) equally
The customer’s view is captured
How to guarantee a representative sample:
Design a sampling strategy
Understand special characteristics of the population before
sampling
Regardless of the situation, a sample must be

“representative” of the population.
For practical purposes a sample is representative if it

accurately represents the target population.
Consideration that may hinder collection of a
representative sample include:
The cost and ease of obtaining samples

Time constraints
Unknown characteristics of the population
Samples that are not representative of a target

population are called biased samples. Often, the
biases are not recognized until the collected data has
been analyzed.
D M A I C
Sample
DefineSize For Continuous
Review ObjectivesData
4
How Do I Determine Sample Size?

Sample size (n) depends on three things
Level of confidence required for the result, “How confident I am that the result
represents the true population”
– Level of confidence increases as sample size increases
Precision or accuracy (∆) required in the result, “The error bars or uncertainty
in my result”
– Precision increases as sample size increases
Standard deviation of the population (σ), “How much variation is in the
total data population”
– An estimate of standard deviation is needed to start
As standard deviation increases, a larger sample size is needed to obtain
reliable results 2
 1.96 × σ 
n= 
 ∆ 
In this equation, “1.96” represents a 95% confidence level
Consider the following example: Therefore, from the statistical theory we can answer
We want to estimate average call length in handling according to the formula:
customer inquiries, and we want our estimate to be
accurate to within + 1 minute.  1.96 × σ 
2
n= 
 ∆ 
Based on a small random sample of 30 inquiries we
know that the variation in call length, as measured
by a statistic called the standard deviation, is 5 Where n = sample size, u = standard deviation and
minutes. ∆ = degree of precision required. In our example,
the required sample size is:
We want to have 95% confidence that the estimate n = [(1.96x5)/1] 2 = 96.04
will be in the range of specified accuracy – i.e., + 1 or 96 samples
minute.
D M A I C
Common Segmentation
Objectives
5
Other Categories
Factor Example
What type Complaints, defects, problems
When Year, month, week, day
Where Country, region, city, work site
Who Business, department, individual,
customer type, market segment
Tip: Begin with factors “outside” the process box–often these are factors that were not considered
when the process was first designed
In service-related processes, “When” elements to

pay special attention to include: relationship to
deadline (e.g., end of quarter), time/number in
queue, or timing versus systems changes.
Note: Your team will need to segment the data in

several different ways in order to uncover where
the most significant differences occur.
D M A I C
Calculating A Sample
Define Review Size
Objectives
6
Continuous Data To Derive Sample Size Formula
– σ Standard error
CI –X X ± Zα/2 * n
–
Given X ± ∆
σ
Solve for n ∆ = Zα /2 n
Z α /2 *σ
n =
∆ 2
n = (∆ )
Zα /2 *σ
n = sample size
∆ = precision of the estimate. How much error is
ok? The smaller ∆, larger the sample size. May be a
business decision.
CI = Confidence Interval
Zα/2 = Z score we usually set at 1.96 for 96%
confidence.
σ = estimated standard deviation
D M A I C
How To Estimate
Define Review Standard Deviation When It Is Unknown
Objectives
7
3 Ways To Estimate σ
Use an existing Xbar or R chart
σˆ = R /d2 where d2 is control chart factor
(
σˆ = UCL − X /3 )
Collect a small pre-sample & calculate s (n = 30)
Ask subject matter experts to take an educated guess at the
plausible range of data
σˆ = (Highest known value - Lowest known value )/6
Control Chart Factor d2
n d2
2 1.128
3 1.693
4 2.059 σˆ = R /d 2
5 2.326
6 2.534
Range or moving range chart should be in

statistical control (stable) to use this estimate of σ
D M A I C
Sample
DefineSize For Continuous
Review ObjectivesData (continued)
8
What Is The Performance For Delivery Time?

Population
Y = Delivery 2 Calculate sample size (n) based on:
Time  1.96 × σ  Precision (∆)
n= 
(Days)  ∆  95% confidence Level (1.96)
Standard deviation (σ)
Sample
n
Values
∆ ∆
⌧-∆ ⌧ ⌧+∆
Calculate average (⌧)
Conclusion:
I know with 95% confidence that the population mean is ⌧ + ∆
Whenever samples are taken to estimate a Confidence intervals are important when precise
population there will be differences between the estimates of populations are required, and the
“true” population values and the sample values. degree of precision in the estimate needs to be
known.
Through statistical theory we can determine the
amount of variation we can expect in our estimates. For those statistically inclined, an important
This is known as a confidence interval. Confidence statistical theory supporting confidence intervals is
intervals are stated in terms of an interval and a the “central limits theorem”. The central limit
confidence level. theorem states that regardless the shape of the
population, sample averages will always be
For example, when estimating average inquiry “normally” distributed and inversely proportioned
resolution time from samples, we may obtain a 95% to the square root of the sample size.
confidence interval for the average 1.2 - 2.3 days.
This means that if repeated samples were taken
from the same population, 95 times out of 100 we
would expect the sample average to be between 1.2
and 2.3 days.
D M A I C
Finite Population
Define Review Correction
Objectives
9
If You Have A Finite Population

1. Calculate sample size (n)
2. If n/N > .05

OR Where n = sample size;
If n > N… N = population size
3. Calculate n finite
nfinite = n/(1+n/N)
Operational definitions guide what properties

will be measured and how they will be
measured.
There is no single right way to write an
operational definition. There is only what
people agree to for a specific purpose. The
critical factor is that any two people using the
operational definition will be measuring the
same thing.
D M A I C
Sample
DefineSize For Discrete
Review Data
Objectives
10
How Do I Determine Sample Size?

Sample size (n) depends on three things:
Level of confidence required for the result, “How confident I am
that the result represents the true population”
– Level of confidence increases as sample size increases
Precision or accuracy (∆) required in the result, “The error bars
or uncertainty in
my result”
– Precision increases as sample size increases
Estimated proportion defective of the population (P)
– An estimate at P is needed to start
– Sample size is maximized at P = 0.5
2
 1.96 
n=  P(1 − P )
 ∆ 
In this equation, “1.96” represents a 95% confidence interval
You need to know the approximate proportion

defective for the population to be sampled to calculate
the sample size. For example:
P ∆ n
.05 .02 456
.50 .02 2400
.95 .02 456
This calculation procedure is based on the binomial

model and should only be used when nP > 5.
As for continuous measurements, if the sample size (n)

is more than 5% of the population size (N), the finite
population correction should be used n (finite) = n/(1 +
n/N) and n (finite) should be used as the sample size.
D M A I C
Sample
DefineSize For Discrete
Review Data (continued)
Objectives
11
What Is The Defect Rate (P) Of A Process?

Population
Y = Proportion 2 Calculate sample size (n) based on:

 1.96 
Defective n=  P(1 − P ) Precision (∆)
 ∆  95% confidence Level (2)
Estimated proportion defective (P)
Sample
n
Values
Conclusion:
I know with 95% confidence that the population
proportion defective is P + ∆
Calculate Proportion
Defective (P)
Recalculate n* based on the calculated P. If the new required sample size (n*) is more than the number
of samples taken, take (n*-n) samples and recalculate P based on the full sample size. If it is not
practical to take more samples, then use the actual n and P to recalculate the actual precision (∆).
Minimum Sample Size
2
 1.96 
n=  P(1 − P )
 ∆ 
Example
We want to estimate the defect rate (P) within +
0.02 (I.e., ∆ = 0.02). We expect P to be
approximately 0.05.
2
 1.96 
n=  0.05(1 − 0.05 ) = 456
 0. 2 
D M A I C
How To Estimate
Define Review P and ∆
Objectives
12
Estimate P
Take a small pre-sample of data (n = 100) and calculate P
Use X from an existing control chart
Set P = .5 as a worst case (largest n)
Estimate ∆
See estimation of ∆ for continuous data
To determine ∆ for a given sample size
Px (1− P )
∆ = Ζ α/2
n
Sample Size & P
0 .5 1
P
D M A I C
Sample
DefineSize Considerations
Review Objectives
13
Beyond The Formulas…

The formulas give an approximate sample size
Don’t forget these important factors:
– Is the population homogeneous?
If not, you will need to segment before sampling
– What is the opportunity for bias?
Plan ahead to make sure your data is representative of the true
population
What Is The Impact On The Customer If Your Sample Size

Is Not Representative Of The Process?
D M A I C
Sampling ExerciseObjectives
Define Review
14
Objective Apply Sample-Size Formulas
1. In your table-team, answer the assigned questions and be

Instructions
prepared to report your answers. Assume confidence level of
95%.
2. Determine the sample size needed if the following is known.
P = 0.20, ∆ = 0.0784
3. Give an estimated proportion defective guessed to be 5%, how
many observations should we take to estimate the proportion
defective within 2%?
4. We want to estimate the average cycle time within 2 days. A
preliminary estimate of the population standard deviation is 8
days. How many observations should we take?
5. We want to estimate average hold time at one of our cost
centers within ± 2 seconds. We will assume hold time standard
deviation is 10 seconds. How many calls do we need to
sample?
Time 10 Minutes
D M A I C
Measurement System Analysis–Objectives 1
Recognize that observed variation of a product/process

includes the true variation of the product/process & the
variation due to the measurement system
Identify & describe possible sources of variation in a
measurement process
Describe the importance of a validated measurement system
Describe the terms precision, accuracy & resolution in relation
to MSA
Use appropriate tools to validate measurement system,
analyze, and interpret results
– Gage R&R for continuous data
– Attribute R&R for discrete data
One of the objectives of the Measure Phase is to validate

your measurement system. A Gage R&R Study will help
us do this!
The focus of this module is to review the methodology

and tools to validate your measurement system.
MSA = Measurement System Analysis
Gage R&R = Gage Repeatability & Reproducubility
DMAIC GB M TX PG V 4.2.0
D M A I C
Possible Sources Of Variation 2
Observations
Inputs Process Outputs Inputs Measureme
nt Outputs Measuremen
ts
Process
Data
Observed Process Variation
Actual Process Variation Measurement Process

Measurement Process
Variation
Variation
Long-term Short- Variation Variation “Other”

Variation Variation “Other”
Process term due to due to sources
due to due to sources
gage operator (Environment
Variation Process gage operator (Environment
, etc.)
Variation , etc.)
Accuracy
Accuracy Precision Discriminatio
Precision Discriminatio
(Bias)
(Bias) (Measureme nn
(Measureme (Resolution)
nt Error) (Resolution)
nt Error)
To Address Actual Process Variability; The Variation Due To

The Measurement System Must First Be Identified And
Separated From That Of The Process
Measurement System Variation Why is MSA important?
We will be evaluating the variation in our measurement Data is only as good as the process that
systems. We will include the variation due to the operator measures it
and gage (Accuracy, Precision and Discrimination). MSA identifies how much variation is present in
the measurement process
When collecting data, we are seeing the process through Understanding measurement variation is
the lens of our measurement system. We never really see necessary for identifying “true” process variation
the actual process variation-instead, we see the actual and maximizing true Y improvements
process variation and the measurement process variation. Without MSA, you run the risk of making
Therefore, the total observed variation can be broken decisions based on an inaccurate picture of your
down into 2 parts: the actual process variation and the process
variation (or measurement error) created by the
MSA helps direct efforts aimed at decreasing
measurement process.
measurement variation
Excessive measurement variation distorts our
understanding of what the customer feels
D M A I C
Measurement Error vs. Process Variation 3
Call Taking Y= Time to Respond (Response Time)

– Process Variation: Time to answer a question varies for different
operators, locations, etc.
– Measurement Error: Error in capturing the time measurement due to
vague definition of when to stop.
Application Processing Y= Time to Decision
– Process Variation: Different types of applications vary in time to
complete. Associates vary in their productivity.
– Measurement Error: Error in measurement time. Time stamp
inconsistencies or application arrival times not recorded accurately.
Deal Approval Y = Time to Complete
– Process Variation: Differing cycle times. Clients are treated differently.
– Measurement Error: Inconsistencies in start-up or set-up definitions.
Deal Approval Y = Agreeable Terms and Conditions
– Process Variation: Deal value variation (net income generated).
– Measurement Error: Errors in recording net income.
Measurement Error vs Process Variation make process improvements. The Minitab results
obtained from your Gage R&R will indicate where the
When conducting a Gage R&R, you will be looking at the largest sources of measurement variation are hiding.
Measurement Error and determining if it is at an Your goal will be to fix or reduce this variation and then
acceptable level. We want to minimize the measurement to re-run the Gage R&R. You cannot collect and/or use
error (or measurement process variation). By conducting your data until the Gage R&R passes.
a Gage R&R, you will know how much of the total
variation is due to the measurement process itself. In
other words, by knowing the Measurement Error, we can
assume that the variation we are observing in the process
is mainly due to the process variation and not due to the
measurement error. If the measurement error is found to
be unacceptable, the measurement process must be fixed
prior to collecting and/or analyzing data and trying to
D M A I C
Planning Your MSA 4
Questions to ask:
What are you measuring?
Who is measuring?
What do you use to measure (gage)?
How are you measuring? Do you always use the same
procedure?
Under what conditions are you measuring (6 m’s)?
What is the resolution of your measurement system? Does the
customer measure the same way?
Which analysis will you use (continuous, Attribute and/or
Destructive)?
D M A I C
Performing An MSA 5
An MSA can be conducted with Continuous or Discrete Data

(we will review each method).
When conducting the MSA, the following guidelines are
suggested:
– Choose 2-3 “operators” who normally perform the measurements.
– For continuous data, choose 10 parts/samples (e.g., applications, calls,
deals, trucks, etc) to measure. The parts should be as dissimilar as
possible (within the normal measurement range). These same, ten parts
will be used throughout the entire MSA study.
– For discrete data, choose 30-40 parts/samples (higher is better).
– The parts must be labeled with a number. The part numbers will remain
constant throughout the MSA.
– Each operator will measure the parts 2-3 times. A blank data collection
form should be used for each trial.
– The Gage should be “calibrated” as per the calibration procedure prior to
conducting the MSA.
Choosing Parts for your Study: The “Parts” are NOT

randomly chosen. They should be chosen specifically for
the study. For example, you need to choose parts that
vary from one another (in size, weight, length of time,
etc.) because one output in the MSA tells you if your
measurement system can distinguish between parts.
Measuring the Parts: The operators should measure

each of the 10 parts and record them on a data collection
form. On their second round of measurement, they
should have a new blank form. Never allow the operator
to see previous measurements or measurements from
other operators.
Analyzing the Results: After all the data is collected,

you will run the MSA in Minitab (for Continuous data) or
in Excel (for Discrete data). The remaining pages in this
section explain the MSA output.
D M A I C
Measurement System Analysis 6
Five Components Of An MSA:
Accuracy–the differences between observed average measurement and a

standard
Repeatability–variation when one person repeatedly measures the same unit

with the same measuring equipment
Reproducibility–variation when two or more people measure the same unit with
the same measuring equipment
Stability–variation obtained when the same person measures the same unit
with the same equipment over an extended period of time
Linearity–the consistency of the accuracy across the entire range of the

measurement system
Think in terms of shooting at a target.
Accuracy: How well do you hit the target? Do you hit

where you are aiming?
Repeatability: Can you keep hitting the target in the

same place?
Reproducibility: Can another person with the same gun

hit the target in the same place?
Stability: If you come back to shoot at other times are

you still able to shoot as accurately (consistency over
time)?
Linearity: Are you as accurate at 50 meters as you are at

250 meters?
D M A I C
Process Flow For MSA (Measurement System Analysis) 7
Conduct MSA
Take Preventive And

OK?
Corrective Actions
No
Yes
Continue Process
Improvement
For Attribute R&R:
90% matches for each Repeatability, Reproducibility,

and Accuracy analysis.
Examples of Corrective Actions: Note: Attribute = Discrete

Review data and decisions
Institute stopgap measures
Examples of Preventive Actions:

Correct mechanical or definitional errors
Institute or upgrade training
Examples of Process Improvement:

Measure gage over time to address stability
Study linearity of gage
D M A I C
Attribute Data AR&R’s 8
With Discrete Data, we will look at these things:

– Repeatability: Do repeated measures match within an operator?
– Reproducibility: Do repeated measures match between operators?
– Accuracy: Do the measures match the standard value?
An MSA helps us determine if our measurement system must
be improved, and if so, gives guidance as to how
Although it has been emphasized that it is best to find a Y

that is associated with continuous data, this is not always
possible. In some cases, the response will be discrete. The
measurement system must still be validated.
Attribute = Discrete
AR&R = Attribute Repeatability and Reproducibility

is the method we’ll use to analyze Discrete
Data.
D M A I C
AR&R Guidelines For Evaluation 9
Repeatability: 90% of repeated measures within an operator

match
Reproducibility: 90% of the repeated measures across
operators match
Accuracy: 90% of the individual measures match the standard
Guidelines for determining how to design an AR&R study: The examples that follow balance all
1. For accuracy: The sample size is the total number three characteristics of an AR&R study.
of measurements. (40 parts x 3 repeated measures
x 3 operators = 360 measures.)
2. For repeatability: The more repeated measures on

the same unit, the better the sensitivity. Increasing
the number of units has a minimal effect on
sensitivity. (Rather than measure 40 parts 3 times,
measure 10 parts 12 times.)
3. For reproducibility: The more operators, the better

the sensitivity. More units have a minimal effect on
sensitivity. (Rather than measure 40 parts with
3 operators, measure 10 parts with 12 operators.)
D M A I C
How To Determine Sample Size In The AR&R 10
How many samples should I take for an AR&R Study?

Typically, 30-40 samples should give you a good indication of
your measurement system
Ten samples may be enough if running 30-40 samples would
be too costly or time consuming. Use your “Business Sense”
and learn from the study. Always ensure you choose samples
that represent typical measures in your process.
For example, if your process requires a Customer Service

Representative to code customer complaints as “A-K”
categories, you would need to ensure your samples
represented each of those codes. You also should choose
samples that clearly fit that category and also ones that
may fall within a “gray” area. (if such samples exist.)
Therefore, the AR&R study will show whether your
current measurement system is adequate.
D M A I C
AR&R Data Sheet 11
'True' Operator 1 Operator 2 Operator 3

Sample Answer Trial1 Trial2 Trial3 Trial1 Trial2 Trial3 Trial1 Trial2 Trial3
1 N N N N N N N N N N
3 N N N N N N N D N N
4 D D D D D D D D D D
5 D D D D D D D D N D
7 D N D N D D D D D D
8 N N N N D N D N N N
10 N N N N N N N D N D
12 D N N N D D D D D D
16 N D D D N N N N N N
19 N N N N D D D N N N
20 N N N N N N N D D N
21 D D D D D D D D D N
24 N N N N N N N D D D
30 D N D N D D D D D D
33 N D N N D D N N N N
35 N N N N N N N N D N
38 N N N N N N N N N D
This data sheet uses 3 operators, 3 trials, and 40 units to An extra column is added to define the true state of the
measure repeatability, reproducibility and accuracy. standard according to a subject matter expert. In this
Repeatability of the measurement system is assessed example, N = Non-defective and D = Defective.
by determining the proportion of times each operator
matches on (e.g., three) repeated measures of one unit.
For this example, there are 120 opportunities for a
repeatability match.
Reproducibility of the measurement system is assessed
by determining the proportion of times all operators
match on (e.g., nine) repeated measures of one unit.
For this example, there are 40 opportunities for a
reproducibility match.
Accuracy of the measurement system is assessed by
determining the number of times each individual
measure matches a standard. For this example, there
are 360 opportunities for an accuracy match.
D M A I C
AR&R Example: Repeatability 12
Operator 1 Operator 2 Operator 3

Sample Tr1 Tr2 TR3 Match? Tr1 Tr2 Tr3 Match? Tr1 Tr2 Tr3 Match?
1 N N N Y N N N Y N N N Y
3 N N N Y N N N Y D N N N
4 D D D Y D D D Y D D D Y
5 D D D Y D D D Y D N D N
7 N D N N D D D Y D D D Y
8 N N N Y D N D N N N N Y
10 N N N Y N N N Y D N D N
12 N N N Y D D D Y D D D Y
16 D D D Y N N N Y N N N Y
19 N N N Y D D D Y N N N Y
20 N N N Y N N N Y D D N N
21 D D D Y D D D Y D D N N
24 N N N Y N N N Y D D D Y
30 N D N N D D D Y D D D Y
33 D N N N D D N N N N N Y
35 N N N Y N N N Y N D N N
37 N N N Y N N N Y D N N N
38 N N N Y N N N Y N N D N
Operator 1 0.925 Operator 2 0.950 Operator 3 0.800
Repeatability of the measurement system is assessed

by determining the number of times each operator
matches on (e.g., three) repeated measures of one unit.
In this example:
– A Y in the “Match?” column indicates a match, and
an N indicates a non-match. On Unit 2, operator 2,
for example, all three measures matched, while on
Unit 3, Operator 3, only trials 2 and 3 matched.
– There are 40 samples *3 operators which yield 120
opportunities for a defect. Subtract (number of non-
matches/120) from 1 and multiply by 100% to get
the percent match for repeatability. 13/120 = .1083
did not match.
– The percent matched = 100 (1- .1083) = 89.17%
– Since 89.17% matched is very close to 90%
matched, gage repeatability technically fails but is
practically acceptable.
D M A I C
AR&R Example: Reproducibility 13
Operator 1 Operator 2 Operator 3

Sample Tr1 Tr2 TR3 Tr1 Tr2 Tr3 Tr1 Tr2 Tr3 Match?
1 N N N N N N N N N Y
3 N N N N N N D N N N
4 D D D D D D D D D Y
5 D D D D D D D N D N
7 N D N D D D D D D N
8 N N N D N D N N N N
10 N N N N N N D N D N
12 N N N D D D D D D N
16 D D D N N N N N N N
19 N N N D D D N N N N
20 N N N N N N D D N N
21 D D D D D D D D N N
24 N N N N N N D D D N
30 N D N D D D D D D N
33 D N N D D N N N N N
37 N N N N N N D N N N
38 N N N N N N N N D N
Operator 1 Operator 2 Operator 3 0.525
Reproducibility (plus repeatability) of the measurement

system is assessed by determining the number of times
operators match on the same unit.
In this example:
– A Y in the “Match?” column indicates a match, and
an N indicates a non-match. On Unit 1, for example,
all three operators matched, while on Unit 3, only
Operators 1 and 2 matched.
– There are 40 samples or 40 opportunities for a defect
in reproducibility. 19/40 = .475 did not match.
– The percent matched = 100 (1-.475) = 52.50%
matched.
– Since 52.50% matched is lower than 90% matched,
gage reproducibility is not acceptable.
D M A I C
AR&R Example: Accuracy 14
Std Operator 1 Operator 2 Operator 3 Not

Sample Value Tr1 Tr2 Tr3 Tr1 Tr2 Tr3 Tr1 Tr2 Tr3 Match?
1 N N N N N N N N N N 0
3 N N N N N N N D N N 1
4 D D D D D D D D D D 0
5 D D D D D D D D N D 1
7 D N D N D D D D D D 2
8 N N N N D N D N N N 2
10 N N N N N N N D N D 2
12 D N N N D D D D D D 3
16 N D D D N N N N N N 3
19 N N N N D D D N N N 3
20 N N N N N N N D D N 2
21 D D D D D D D D D N 1
24 N N N N N N N D D D 3
30 D N D N D D D D D D 2
33 N D N N D D N N N N 3
35 N N N N N N N N D N 1
37 N N N N N N N D N N 1
38 N N N N N N N N N D 1
# of N 5 2 4 6 5 5 5 4 4 40
Accuracy of the measurement system is assessed by

determining the number of non-matches to the standard
for each individual measure
In this example,
– The “Non-Match?” column contains a count of non-
matches to the standard value (std value) for the unit
(row). For example, all operators matched the
standard for unit 2, so the number of non-matches is
0. For unit 3, operator 3 failed to match the standard
on trial 1, so the number of non-matches is 1.
– There are the 40 units *9 repeated measures, or 360
individual measurements. 40/360 = .1111 did not
match.
– The percent matched = 100 (1-.1111) = 88.89%
matched.
– Since 88.89% matched is very close to 90%
matched, gage accuracy technically fails but is
practically acceptable.
D M A I C
AR&R Analysis Example: Accuracy (continued) 15
To summarize the AR&R MSA

Repeatability: 13/120 non-matched = 100 (1-.1083) = 89.17%
(Fail–But Close!)
Reproducibility: 19/40 non-matched = 100 (1-.475) = 52.50%
matched (Fail)
Accuracy: 40/360 non-matched = 100 (1-.1111) = 88.89%
matched (Fail–But Close!)
Note: For Repeatability, 89.17% (and Accuracy

88.89%) we considered both of these as
“Passing” due to the closeness to 90%. Use your
judgment. If an improvement can be made, feel
free to improve the measurement system. The
90% rate is an approximation not a “hard” limit.
D M A I C
Measurement System Analysis (MSA) Worksheet 16

Gage R&R (Continuous Data) Conducted: 1/2/01
Gage R&R Results

Part p-value Y N
Oper p-value Y N
Pass?
2) % Tolerance [ <30% ] Y N
3) % Contribution [ <8% ] Y N
4) % Study [ <30% ] Y N
5) # Distinct Categories [ >4 ] Y N

Between Oper
√
Attribute Gage R&R
1) % Repeatability [ >90% ] 89.17% OK
2) % Reproducibility [ >90% ] 52.5% Need to improve. See plan below
3) % Accuracy [ >90% ] 88.89% OK
Gage R&R Pass? Y √ N, If NO:
Plan for improvement: Will work on Reproducibility problem by conducting a team meeting
with the 3 operators and discussing the differences in their measurements. Suspect that additional
training is needed for less experienced operators because the data showed less experienced
operators answering differently. Will also investigate if there is anything unique with each of the
samples where reproducibility was an issue.
D M A I C
Causes Of Measurement System Variation 17
If you fail the Gage R&R, here are some factors that could
cause measurement process variation (measurement error).
Repeatability
Operational definitions
Maintain stability
Reproducibility
Consistent use of gage
Training
Varying work environment
AUDIT-follow-up on training
Human/physical characteristics
Performance measures
Unclear requirements
If the Gage R&R fails, you need to improve the

measurement system before moving on. Use these factors
above as places to begin looking for areas of variation in
your measurement system. By improving some of these
factors, your measurement system variation should be
reduced (or brought to an acceptable level).
D M A I C
Causes Of Measurement System Variation (continued) 18
Accuracy
Error in master
Instrument used improperly by appraiser
Operational definition
Standard not understood
D M A I C
Other Considerations 19
Temporal Effects
Gage R&R depends upon our ability to measure things multiple
times
If the item is an event, it may not happen the same way twice.
For this reason, Gage R&R may be impossible.
If the event is recorded, it may be possible to conduct a Gage
R&R study
If the event cannot be recorded, but multiple judges can
observe at once, reproducibility can be estimated, but
repeatability cannot
Video cameras or audio voice recordings can allow a

Gage R&R to be conducted. Use today’s technology to
help you in validating your measurement system.
D M A I C
Temporal Effects Example 1

Many Olympic sports are judged based on a numerical score
sheet filled out by a judge. The difference among different
judge’s readings of one live event is a measure of
reproducibility.
For this same case, repeatability can be estimated only if the

competition is taped. A random sample of 10 performances
shown to five judges, two times each, would allow estimates of
both repeatability and reproducibility.
D M A I C
Temporal Effects Example 2

Call centers field service calls from around the world at
centralized locations. Call duration and call quality are both
recorded and tracked very closely.
Call quality is calculated based on a scoring sheet filled out by

a quality monitor. Calls are scored on a continuous scale from
1-100 and can be assumed to be continuous for the purposes
of a Gage R&R calculation.
D M A I C
Temporal Effects Example 2 (continued)

In one monitoring scenario, monitors listen in live and score the
calls. For a “live listen” it is impossible to calculate repeatability,
but if two monitors listen at the same time, it is possible to
calculate reproducibility.
In another monitoring scenario, calls are recorded at random

and scored at a later date. For this recorded scenario, both
repeatability and reproducibility may be calculated.
D M A I C
Continuous Data MSA, First Step 23
Test-Retest Study
D M A I C
Test–Retest Study 24
Best Practice Hint: Do Test-Retest before Gage R&R–a quick look at the
situation.
Why: Determine the precision of the system, instrument, device, or gage–where

Precision = Measurement Error = Repeatability
How:
Repeatedly measure the same item
Same conditions, operator, device, and location on item–same, same,
same
Completely mount and dismount item for each measurement–exercise gage
through full range of normal use
Data: Twenty (20) or more measurements is an adequate sample size. If

measurements are difficult or expensive, then 10-15 may be OK. More is better.
Calculate the sample mean ( Χ) and standard deviation (s) of the repeated
measurements.
Test–Retest studies are sometimes called Calibration

Tests. This study is done prior to a Gage R&R.
A Test–Retest Study will indicate if you have either a

precision or accuracy problem.
A Test–Retest Study is typically run using a certified or

working “Standard.” The same item is measured
repeatedly.
D M A I C
Test–Retest Study Guidelines 25
Device precision should be less than 1/10 of the tolerance:
s < 1/10 x tolerance*
If s exceeds 1/10 x tolerance*, then the measurement system is

unacceptable because it lacks precision. Action is required to find
and remove the sources of this error, including the replacement of
the device.
Device accuracy may be estimated if you know the true value

of the test unit:
Inaccuracy = Bias = X - True Value
Precision is a term used to describe the amount of

measurement error (repeatability error) in a measurement
system.
Precision is also used to describe the term ∆ (delta) in

confidence intervals. There it indicates ½ width of a
confidence interval.
If your gage lacks precision or accuracy, you may need to

remove or replace the measurement device. Perhaps a
better or updated device can be used.
*For a one-sided specification you can: (1) Use the mean

minus the tolerance. Note: the new formula would be:
(½) S ≤ 1/10 (USL – X) or (½) S ≤ 1/10 ( X-LSL). (2) use
the natural boundary (for example, use 0 for a cycle time
measurement) as the other spec. (Tolerance = USL-0).
D M A I C
Process Flow For MSA (Measurement System Analysis) 26
Conduct MSA
Take Preventive And

OK?
Corrective Actions
No
Yes
Continue Process
Improvement
For continuous data we’ll look at these major items: %Tolerance,

% Contribution, % Study and # Distinct Categories.
MSA is a set of methods for estimating the current

amount of variation in the Measurement System.
Examples of Corrective Actions:

Review data and decisions
Institute stopgap measures
Examples of Preventive Actions:

Correct mechanical or definitional errors
Institute or upgrade training
Examples of Process Improvement:

Measure gage over time to address stability
Study linearity of gage
D M A I C
Precision & Accuracy 27
Target Analogy
== Xbar
True Value = Bull's Eye
I. Precise, not accurate

Xbar
==
True Value
II. Accurate, not precise

Xbar
==
True Value
III. Precise and accurate
Precision: What kind of repeatability does the whole

system have? What is the scatter? (standard deviation)
Accuracy: Does the average reading agree with the actual

size of the part? (mean)
D M A I C
Measurement System 28
Observations
Inputs Outputs Inputs Measurement Outputs Measurements
Process Process Data
Product Measurement Total

Variation Variation Variation
(Actual (Observed
Example variation) variation)
#1
Example σ 2Actual (Part) + σ 2Meas. System = σ 2Observed (Total)

#2
Measurement
System Variability
- Investigated
through “R&R
Study”
The measurement system will also contribute some σ2 = variance

variation to the overall total observed variation.
Here are two examples of the effect of measurement

variation. This variation comes from both the gage and
the operator.
In a Gage R&R Study, we will determine how large the

measurement variation contributes to the overall variation
we’re observing in our process. The key will be to
minimize the measurement variation so that we are
mainly seeing process variation.
D M A I C
Types Of Variation Estimated By The Gage R&R 29
Equipment Variation (EV)

(Repeatability)
(Sources of variation from within the process)
Within Gage–Within Operator–Within Part/Process–Etc.
The variation introduced into the measurement process
from within one or more elements of the measurement
process–such as: within operator variation–within gage
variation–within part variation–within method variation.
Appraiser Variation (AV)

(Reproducibility)
(Source of variation from across the process)
Across Gages–Across Operators–
Across Parts/Process–Etc.
The variation introduced into the measurement process by
effects going across the measurement process–such as
different appraisers–different part configurations–different
checking methods.
Equipment variation–variation within a sample group.
Appraiser variation–variation between sample groups.
D M A I C
Relationship Between EV, AV And R&R 30
R&R is the Reproducibility (AV) and Repeatability (EV) of the

Measurement System. It represents the total variation in the
Measurement System.
R&R
EV AV
σ 2Equipment + σ 2Appraiser = σ 2Total (R&R)
EV = Equipment Variation Repeatability: variation when one person

repeatedly measures the same unit with the same
AV = Appraiser Variation measuring equipment.
The variation due to the equipment and appraiser do not Reproducibility: variation when two or more
directly add up to determine the total (R&R) variation. people measure the same unit with the same
There is an overlap between EV and AV. You can use the measuring equipment.
Pythagorean theorem (right angles) to add to the
influence of each EV and AV to calculate the total
variation.
σ 2= variance
D M A I C
Relating R&R To Specification Window 31
How much of the tolerance is used up by the Measurement

System variation?
99%
R&R
(5.15σ)
Lower Spec. Limit Upper Spec. Limit

Specification Window (tolerance)
About 50% Of The Tolerance In This Example Is Used Up By The

Measurement System Variation This Leaves Only 50%
For The Process Variation
We’re interested in how this variation compares to the

tolerance with which we are working.
If the measurement system takes up lots of variation, we

don’t have any room for process variation. Minitab calls
this variation % Tolerance. This is one or our 4 key
metrics in the Gage R&R Study.
D M A I C
Gage R&R For Continuous Data: Example 32
GB Case Study: On Time Delivery of Shipments to Customers
Background: The dispatchers for “Capital Logistics” keep a

record of the time when truck drivers radio in to report delivery of
the shipment was made to the customer (this is a requirement from
the customers). Since there are 3 dispatchers recording the
delivery time and 10 truck drivers calling in, the Green Belt needs
to validate these delivery times for repeatability, reproducibility &
accuracy.
First, we will walk through the process to conduct a Gage

R&R Study in Minitab, analyze the results and draw
conclusions. We will use Capital Logistics’ study of the
delivery of shipments to customers.
Then, you will do an activity to conduct a Gage R&R

Study in Minitab with a new set of data from the same
Capital Logistics’ study.
D M A I C
Gage Reproducibility & Repeatability (GR&R) Study: Steps 33
1. Set up a Minitab Worksheet

2. Perform study–collect & enter Data. (use file: “Gage
R&R-Continuous Data.mtw)
3. Perform calculations & prepare Charts.
– Perform Gage R&R Study–ANOVA method (ANOVA =
Analysis of Variance)
4. Analyze–interpret & draw conclusions.
5. Investigate variation in measurement system, take action, make
recommendations–keep, improve, or replace the
measurement system.
If measurement system is changed, repeat above steps to
validate accuracy, repeatability & reproducibility of new
measurement.
We have standardized on using the “ANOVA” method in

Minitab. ANOVA simply stands for Analysis Of
Variance. Minitab will partition the total variation and
allocate this variation to gage variation or process
variation. In the Gage R&R Study, we analyze the gage
variation and want to understand how much gage
variation is present. If we have too much measurement
variation, we’ll need to fix our measurement system prior
to collecting any data. We want to ensure the variation
we are observing is mainly process variation.
D M A I C
Case Study Setup 34
Does my measurement system provide me a clear view of my

process...
To help answer this question the GB
established the following analysis:
1. The GB developed a master matrix which listed the times

each truck was supposed to arrive at the customer. The GB
gave a copy of the matrix to each of the three dispatchers.
2. The GB had each of the truck drivers radio a message

from a radio truck, identifying himself as one of the 10 units
listed in the matrix..
3. So that the GB could get repeatability & reproducibility

data, the GB had the message recorded and then sent to the
dispatchers throughout a 3 day period. The GB was able to
program the phone system to deliver the messages precisely
at the same time both days. The dispatchers were able to
hear the call-in simultaneously via speakerphone.
4. Therefore, both days, the dispatchers, all in the same

room at the same time, would receive the radio message and
record the difference in minutes, from the target time listed on
the matrix. At the end of the day, the GB would collect the
data collection sheets from each dispatcher and give them a
blank form for the following day. This procedure was then
repeated the second day.
5. The GB then consolidated the data into a single Minitab

file for analysis. (Gage R&R–continuous data.mtw).
…or does it cloud what I see?
D M A I C
Gage R&R For Continuous Data: Example 35
Step 1: Set up Minitab data sheet.

GB asked 3 dispatchers to record the truck driver call-in time
for the 10 different truck drivers.
Data sheet for recording the data was set up in Minitab to keep
track of the 10 trucks (Part/Truck), 3 dispatchers (Oper), 2 runs
(Trial) and the “Y” of Difference from Target (Meas).
Open the File: Gage R&R–Continuous Data.mtw
D M A I C
Gage R&R For Continuous Data: Example (continued) 36
Step 2: Perform Study–Collect & Enter Data.

Truckers called in to report delivery and dispatchers recorded
the difference from target time.
The data was collected and recorded on a data sheet and was
input into Minitab (see Data column).
Minitab File: Gage R&R – Continuous Data.mtw
Use Minitab file: Gage R&R–Continuous Data.mtw Columns 1-3 in the Minitab worksheet represent
the data collected by the dispatchers as each
truck driver called on their delivery times.
1. Part/Truck–indicates which of the 10 trucks
2. Oper–indicates which one of the 3 dispatchers
3. Meas–indicates the difference, in minutes,

from the target time
4. Trial–indicates the trial number when the

operator took the measurement.
D M A I C
Step 3: Perform Calculations & Prepare Charts

Perform Gage R&R Study–ANOVA method
Now let’s perform the Gage R&R study on this In Minitab:

dispatching process example.
Click on STAT > Quality Tools > Gage R&R Study
ANOVA is the best method–it breaks down the overall In the Gage R&R Study dialog box:
variation into three categories:
1. Select Columns for part, operator & measurement.
part-to-part
repeatability 2. Select the ANOVA method
reproducibility
3. Select Options: Enter Tolerance width (note:
Tolerance was set to 20 minutes here). The
and….it breaks down reproducibility further into its
customer wanted the delivery within +/- 10 minutes
components:
of the target time.
operator
operator by part 4. Click “OK”, twice.
The standard value for a 2-sided specification is 5.15.
This is the number of standard deviations needed to
capture 99% of your process measurements. If you
only have a 1-sided spec, you would use 2.575 (half of
5.15).
D M A I C
Minitab Output will generate the following items which we will

use to determine if our measurement system is acceptable:
1. Graphical Summary
2. Two-Way ANOVA Table with Interaction (p-values)
3. % Comparisons Table (% Contribution)
4. % Comparisons Table (% Tolerance)
5. % Comparisons Table (% Study)
6. Discrimination Index
We will cover each of these 6 outputs in this section.
D M A I C
The Central Question In A Gage R&R Study 39
There are 2 ways to (1) FUTURE STATUS: (2) PRESENT STATUS:

answer the question Will my measurement Is my measurement
“Is my measurement variation be too large variation too large right
variation too big?” when I reach the 6-Sigma now, compared to a
goal (i.e., with small process realistic estimate of current
spread, good capability)? process spread (which
may be narrow next year)?
Number
Numberof of R&R
R&R%%Study
Study
R&R
R&R Distinct
Distinct
R&R
R&R Variation
Variation
%Contribution
%Contribution Categories
Categories
%Tolerance
%Tolerance (Ratio
(Ratioof
ofStdDev’s
StdDev’s--
(Ratio
(Ratioof
ofVariances)
Variances) (Discrimination
(Discrimination --AIAG)
AIAG)
Index)
Index)
Red
30%
30% 8%
8% 44 30%
30%
Yellow
10%
10% 2%
2% 10
10 15%
15%
Green
CAUTION:
The magnitude of these
%’s are exaggerations
If the Gage R&R (as a percent of tolerance) is less than If the discrimination index (number of distinct
10%, the MS is green, or acceptable; if Gage R&R is categories) is greater than 10, the measurement system
10-30% , the MS is yellow, or marginal; if Gage R&R (MS) is green, or acceptable; if the discrimination
is greater than 30%, the MS is red, or unacceptable. index is 4-10, the MS is yellow, or marginal; if the
This estimate may be appropriate for evaluating how discrimination index is less than 4, the MS is red, or
well the measurement system can perform with respect unacceptable. The discrimination index is just a
to specifications. We can use it to look into the future transformation of %R&R contribution (for ease of
and ask “Will my measurement system variation be too interpretation.) It addresses the same issues and
large when I reach the 6 Sigma goal (with small contains the same information.
process spread, good capability)?” If Gage R&R (as a percent study) is less than 15%, the
If Gage R&R (as a percent contribution) is less than MS is green, or acceptable; if Gage R&R is 15-30%,
2%, the MS is green, or acceptable; if the Gage R&R is the MS is yellow, or marginal; if Gage R&R is greater
2-8%, the MS is yellow, or marginal; if Gage R&R is than 30%, the MS is red, or unacceptable.
greater than 8%, the MS is red, or unacceptable.
Note: the limits shown above can be derived from one
another. Some rounding was done for ease of
interpretation.
D M A I C
Rules Of Thumb–Acceptable Ranges 40
Step 4: Analyzing Gage R&R Results
A. R&R% of Tolerance
1. R&R less than 10%–Measurement System “acceptable”
2. R&R 10% to 30%–May be acceptable–make decision

based on classification of Characteristic, Application,
Customer Input, etc.
3. R&R over 30%–Not acceptable. Find problem, re-visit the

Fishbone Diagram, remove Root Causes. Is there a better gage on the
market, is it worth the additional cost?
RULES OF THUMB:
1. Less than 10% = Good
2. 10% to 30% = Fair–improvements possible; look to

find opportunities in process and training.
3. Over 30% = Problem–measurement system not

acceptable; fix problem with training, improved
instruments, correct measuring process, and/or
improved operational definitions for data collection.
The gage is not adequate for Product/Process
Acceptance decisions.
D M A I C
Rules Of Thumb (continued) 41
B. % Contribution (or Gage R&R StdDev):

GR&R Variance should be “small” compared to Part-to-Part Variance–
applies in cases where Tolerance Width is not meaningful, and %Tolerance
is unavailable–such as one sided specs.
1. % Contribution < 2%–Measurement System “acceptable”
2. % Contribution 2%-8%–Measurement System “marginal”
3. % Contribution > 8%–Measurement System “unacceptable”
Questions to ask for validating the measurement system:

Are we capturing the correct data? Does the data reflect
what is happening in the process?
How big is the measurement error?
Can we detect process improvement if and when it
happens?
What are the sources of measurement error?
Are the measurements being made with measurement
units which are small enough to properly reflect the
variation present?
Is the Measurement System stable over time
Is the Measurement System “capable” for this study?
How much uncertainty should be attached to a
measurement when interpreting it?
How do we improve the measurement system?
D M A I C
Rules Of Thumb (continued) 42
C. Number of Distinct Categories

A “Signal-to-Noise” Ratio = (StdDevparts/process/StdDevGR&R) x 1.41 and
rounded
Guidelines:
< 2 =>no value for process control, parts all “look” the same
= 2 =>can see two groups–high/low, good/bad
= 3 =>can see three groups–high/mid/low
≥ 4 =>acceptable measurement system
(higher is better)
D. R&R % Study
1. % Study less than 15%–measurement system acceptable.
2. % Study 15% to 30%–may be acceptable (marginal)
3. % Study over 30%–not acceptable
D M A I C
Minitab Gage R&R–Session Window Analysis 43
Here are the Results:

% Comparison Table–Found in our Minitab Session Window
Source %Contribution %Study Var %Tolerance
Total Gage R&R 10.67 Fails! 32.66 Fails! 171.53 Fails!

Repeatability 3.10 17.62 92.54
Reproducibility 7.56 27.50 144.43
Oper 2.19 14.81 77.76
Oper*Part/Truck 5.37 23.17 121.70
Part-To-Part 89.33 94.52 496.41
Total Variation 100.00 100.00 525.21
Number of distinct categories = 4 pass!
We fail the Gage R&R!
% Tolerance is 171.53% which fails, based on our % Study, probably the least important, (because %
criteria.This shows us the % of the Tolerance which is contribution tell us roughly the same thing) also fails
being taken up by the gage. In this case, 171.53% of based on our criteria. This gives us a measure of gage
our tolerance is taken up by the variation of the gage. variation compared to 99% of a normal distribution.
Note: You must put in the tolerance in Minitab in
order to get this data. If you do not enter a tolerance,
these figures will not appear. If you do not have a Number of distinct categories passes! This tells us the
tolerance, then % Contribution would be more number of divisions the measurement system can
important in your analysis accurately measure across the seen process variation.
.
% Contribution is 10.67% which fails our criteria. % For a more detailed explanation of output, refer to the
Contribution is based on overall measure of gage end of this section.
variation (regardless of the specification).
D M A I C
Minitab Gage R&R–Graphical Analysis 44
Now Let’s Review the Graphical Summary:
Gage name:
Date of study :
Gage R&R (ANOVA) for Meas Reported by :
Tolerance:
Misc:
Xbar Chart by Oper Oper*Part/Truck Interaction

110 1 2 3 110 Oper
100 100 1
Sample Mean
90 3.0SL=87.96 90 2
Average
80 X=80.75 80
1 70 -3.0SL=73.54
70
3
4
60
60
50
40 50
30 40
0 Part/Truck 1 2 3 4 5 6 7 8 9 10
R Chart by Oper By Oper

15 1 2 3 110
Sample Range
3.0SL=12.52 100
10 90
80
70
5
R=3.833 60
2 50 5
0 -3.0SL=0.00E+00 40
0 Oper 1 2 3
Components of Variation By Part/Truck

500 110
%Total Var 100
400 %Study Var 90
Percent
300 %Toler 80
3 200
70
60 6
100 50
0 40
Gage R&R Repeat Reprod Part-to-Part Part/Truck 1 2 3 4 5 6 7 8 9 10
Chart 4: Chart 5:
The chart shows the 10 parts and the average This chart shows the average reading for all parts by
measurement of those parts broken-out by operator. operator. The red circle is the average. This could show
Ideally, all 3 lines should be on top of each other. We can us if one operator is measuring higher/lower (on average)
see that there is some measurement variation in parts 4, 8 than the others.
and 10. We should investigate why.
Note: For more detailed explanation, refer to the end of

this section
D M A I C
Next Steps For The Green Belt 45
Sit down with your BB and/or Mentor to ensure proper analysis

After an initial look at the Gage R&R Results for this example,
the Green Belt will investigate the following:
– Look at the feasibility of changing the gage to a digital clock
– Talk with the operators to find out why they are measuring differently
– Can a spreadsheet be developed to calculate time automatically?
– Can we implement a driver call-in procedure to reduce variation?
– Why was it harder to measure parts 4, 8 and 10?
Continue on to see the GB’s Next Steps…
D M A I C
If you fail the Gage R&R, here are some factors that could
cause measurement process variation (measurement error).
Repeatability
Calibrating too often
Granularity of the measure (e.g., nearest hour, minute, second)
Maintain stability
Reproducibility
Consistent use of gage
Training
Varying work environment
AUDIT–follow-up on training
Human/physical characteristics
Performance measures
Unclear requirements
If the Gage R&R fails, you need to improve the

measurement system before moving on. Use these factors
above as places to begin looking for areas of variation in
your measurement system. By improving some of these
factors, your measurement system variation should be
reduced, (or brought to an acceptable level).
D M A I C
Causes Of Measurement System Variation (continued) 47
Accuracy
Error in master
Instrument not calibrated
Worn components/trend or drift in master
Calibrating too often
Operational definition
Range of the master
D M A I C
Linearity
Instrument calibrated incorrectly
Error in master
Worn instrument
Stability
Error in master
Worn instrument
Instrument measuring wrong characteristic
Instrument not calibrated properly
D M A I C
Gage R&R Action Plan 49
Step 5: Investigate sources of variation in the measurement

system and make recommendations–keep, improve or replace the
measurement system.
After completing the analysis of the measurement system, the

Green Belt investigated the sources of variation. By observing the
process, the GB found:
The % Tolerance is the largest issue (171.53%). With the current gage and
measurement process in place, the gage is not acceptable. To address this
issue the GB has done the following:
– Changed the Gage
– Implemented a standardized call-in procedure
– Implemented a new spreadsheet
Explanation of Bullets: The GB also noticed that all the data was positive,
The Dispatchers are using a clock on the wall that is (indicating that the deliveries were always late). Upon
not digital. Therefore, the GB implemented a new investigation, the GB found that the dispatchers did not
procedure to use the computer clock. All computers know that they should enter both positive and negative
were synchronized and a procedure was implemented numbers. A new spreadsheet was created to fix this
to have the system automatically synchronize clocks issue
each day with the Atomic Clock.
The GB also found that the Truck Drivers were
inconsistent in calling in their arrival times. After
going into the field and interviewing some truck
drivers, the GB found that a new call-in process needed
to be implemented. Part of the problem was not with
the gage but with “when” the truck driver called in the
delivery time. In many cases, the truck driver was
within the 10 minute window but had forgotten to call
in the arrival. A standardized process was
implemented to address this issue.
D M A I C
Gage R&R Action Plan 50
Step 5 (continued)
The % Contribution also failed (10.67%). From the Minitab data, the GB
found Reproducibility to be the larger issue and the interaction between
certain trucks and certain operators was the issue. To address these issues,
the GB has done the following:
– The data indicates that the measuring of the arrival time for trucks 4 & 10 seem to
vary from operator to operator. By going back and reviewing the data and talking
to the operators, the following was found:
There was a typo on the recording of time from Operator 1 on truck 4. He accidentally
typed in the wrong number. To address this problem, a new spreadsheet was
developed and now the Dispatcher is asked to verify his data prior to saving it in the
spreadsheet.
For Truck 10, the Operators had a very difficult time hearing the recording because the
Truck Driver was using a cell phone with a poor connection. The GB changed the
procedure so that each Dispatcher was now required to repeat-back the time given by
the truck driver. The Truck Driver then confirms that the correct time is being recorded.
– The data also indicated that Operator 2 seems to measure slightly lower than
Operators 1 & 3. The GB found that Operator 2 was sitting on an angle and when
he read the clock his time was distorted. By converting to the computer clock,
this issue should be resolved.
D M A I C
Gage R&R For Continuous Data–Activity (15 minutes) 51
Desired Outcome: Determine if the improved measurement system

utilized by the Capital Logistics dispatchers is “acceptable” or not
What How Who Timing

Preparation Choose a facilitator, scribe
Conduct Complete 5 steps for a Gage Team 10 mins.

MSA R&R study and record results
Use File: Gage R&R–Continuous
Data.mtw and use columns
C6-C9
Report Complete Report Summary Team 5 mins.

Out
Present findings to class (see
next page for report summary)
Are there any additional items
the GB should consider to
improve the MSA even further?
D M A I C

Gage R&R (Continuous Data) Conducted:
Gage R&R Results

Part p-value Y N
Oper p-value Y N
Pass?
2) % Tolerance [≤ 30%] Y N
3) % Contribution [≤ 8%] Y N
4) % Study [≤ 30%] Y N
5) # Distinct Categories [≥ 4] Y N

Between Oper
Attribute Gage R&R

1) Repeatability [>90%]
2) % Reproducibility [>90%]
3) % Accuracy [>90%]
Gage R&R Pass? Y N, IF NO:
Plan for improvement:________________________________________
__________________________________________________________
D M A I C
Analyzing The Gage R&R Results 53
Rules of Thumb
1. R&R (% Tolerance)
Less than 10%–Measurement System is acceptable
10% to 30%–maybe acceptable–make decision based on classification of
characteristic, hardware application, customer input, etc.
Over 30%–Measurement System is not acceptable. Find problem, re-visit
the fishbone diagram, remove root causes. Validate Measurement System
again
2. % Contribution (or Gage R&R StdDev) :
GR&R Variance should be “small” compared to Part-to-Part
Variance–applies in cases where Tolerance Width is not
meaningful, and % Tolerance is unavailable–Such as one-sided
specs.
1.% Contribution < 2%-Measurement System “acceptable”
2.% Contribution 2%-8%-Measurement System “marginal”
3.% Contribution > 8%-Measurement System “unacceptable”
D M A I C
Analyzing The Gage R&R Results (continued) 54
3. Number of Distinct Categories = a “Signal-To-Noise”

Ratio = (StdDevparts/StdDevGR&R) X 1.41 and rounded
Guidelines:
< 2–no value for process control, parts all “look” the same
= 2–can see two groups–high/low, good/bad
= 3–can see three groups–high/mid/low
> 4–acceptable measurement system (higher is better)
4. R&R % Study
1. % Study less than 15%–measurement system acceptable
2. % Study 15% to 30%–may be acceptable (marginal)
3. % Study over 30%–not acceptable
D M A I C
Summary–Measurement System Analysis 55
Variation in the measurement system will contribute to the

observed variation in a process
The resolution is the ability of the gage to see the variation in
the process
– The gage should be accurate: mean close to the true mean of the
process, and precise: small variation.
Minimize the measurement process variation
Use MSA (Gage R&R for Continuous Data or Attribute R&R for
Discrete Data) to identify the amount of measurement system
variation and process variation
Understand how measurement error impacts your customer
Measurement error is always a bigger deal than you think!
Make Sure Your MSA Is Examining The Actual

Measurement System Itself
D M A I C
Measure Step 3–Green Belt Project Activity (30 minutes) 56
Desired Outcome: Outline of a Data Collection Plan and MSA for your
GB project
What How Timing

Your Individual 1. Data Collection plan* 15 mins.
GB Project
Review and finalize your Data
Collection Plan for your GB project
Have you answered:
What data will you need to
collect?
How will you collect it? From
where?
Who will collect the data?
(can refer to the data collection
worksheet)
2 . MSA* 15 mins.
Develop an MSA for your project Y
data
List factors that might cause
measurement system variation and
how you would reduce the impact
of those factors
*Refer to the Self-Paced Workbook pp 43-49.
D M A I C
Gage R&R Details–Optional Information 57
Pages 58-69 contain Optional/Additional

information that will not be covered in
the 3-Day Workshop
D M A I C
Full Gage R&R Details–Optional Information 58
Graphical Output
•ANOVA
•ANOVAsimplysimplystands
standsfor
for““Analysis
AnalysisofofVariance”
Variance”
•It•Itisisaatool
tool used to analyzethe
used to analyze thetotal
totalvariability
variabilityamong
amongdifferent
differentsources.
sources.
•It•Itpartitions
partitions the total variation into “buckets”, and then allocateseach
the total variation into “buckets”, and then allocates eachbucket
buckettotoeach
eachsource
source
(part,
(part,operator,
operator,oper*part
oper*partinteraction).
interaction).
Gage name:
Date of study :
Tolerance:
Misc:

110 1 2 3 110 Oper
100 100 1
Sample Mean
90 3.0SL=87.96 90
1 2
4
Average
80 X=80.75 80
-3.0SL=73.54 3
70
70
60
60
50
40 50
30 40
0 Part/Truck 1 2 3 4 5 6 7 8 9 10

15 1 2 3 110
Sample Range
3.0SL=12.52 100
10 90
80
70
5
2 R=3.833 60
50
5
0 -3.0SL=0.00E+00 40
0 Oper 1 2 3

500 110
%Total Var 100
400 %Study Var 90
Percent
300 %Toler 80
70
3 200
60 6
100 50
0 40
What do the graphs tell you about the measurement 5 Scatter of individual measures should be equal by
system? Is it acceptable? operator.
1 > 50% of points should be outside control limits. 6 Scatter of individual measures should be equal by part
(truck.)
2 Range of measures by operator should be in control
(this is showing repeatability.)
3 “D” should be largest, then A, B & C approximately

equal.
4 Parallel lines will show no problems with certain parts

(trucks.)
D M A I C
Minitab Gage R&R ANOVA Graphical Output 59
Average of 2 measurements
taken by operator 1. Gage name:
Date of study :
Tolerance:
Misc:

110 1 2 3 110 Oper
100 100 1
Sample Mean
90 3.0SL=87.96 90 2
Average
80 X=80.75 80
-3.0SL=73.54 3
70
70
60
60
50
40 50
30 40
0 Oper 1 Oper 2 Oper 3 Part/Truck 1 2 3 4 5 6 7 8 9 10

15 1 2 3 110
100
Sample Range
3.0SL=12.52
10 90
80
70
5
R=3.833 60
50
0 -3.0SL=0.00E+00 40
0 Oper 1 2 3

500 110
%Total Var 100
Range of 2 measurements 400 %Study Var 90
Percent
taken by operator 1. 300 %Toler 80

70
200
60
100 50
0 40
Xbar chart by Operator: We want the R chart to be “in control”. If it is not, then
Plots the average for each part measured by each this may be an indication that repeatability is poor. Our
operator (average of 2-3 #’s). In our example, it is the data in our session window will confirm if we have a
average of 2 measurements. Repeatability issue. At this point, make a mental note.
Look for any points outside the Upper Control Limit
We want the parts to be “out of control” If they are not (verify the measurement is not a typo!)
out of control, we can’t tell one part from another!
Remember , we’ve chosen parts that are different from In this case, the R Chart looks OK. The chart is in
each other. Usually, we want 50% of our points to be control indicating that Repeatability is probably not an
outside the control limits. issue.
In this example, the Xbar chart looks OK (more than 50%

of the points are outside the control limits.)
R Chart by Operator:
Plots the range of each average point in the Xbar chart.
This is the difference between the highest and lowest
measurement for all 10 parts for each operator.
D M A I C
Gage name:
Date of study :
Tolerance:
Misc:

110 1 2 3 110 Oper
100 100 1
Sample Mean
90 3.0SL=87.96 90 2
Average
80 X=80.75 80
-3.0SL=73.54 3
70
70
60
60
50
40 50
30 40
0 Part/Truck 1 2 3 4 5 6 7 8 9 10

15 1 2 3 110
100
Sample Range
3.0SL=12.52
10 90
80
70
5
R=3.833 60
50
0 -3.0SL=0.00E+00 40
0 Oper 1 2 3

500 110
%Total Var 100
400 %Study Var 90
Percent
300 %Toler 80
70
200
60
100 50
0 40
A B C D
Graphical Display MSA Indicators by Components of

Variation:
This is a graphical display of the components of
variation. You can view the actual data in the session
window.
% Total Var ==> same as saying % Contribution
% Toler ==> % Tolerance
Remember, we want the variation in the parts (Part-to-

Part) to be the major piece of variation and we want the
variation from the gage itself to be small. Therefore, we
want “A-C” to be small and “D” to be largest.
In this case, the total variation of the parts is contributing

89.33% and the variation in the gage is contributing
10.67% (can be found in the session window). We’ll
learn more later on % Contribution and % Tolerance.
D M A I C
Gage name:
Date of study :
Tolerance:
Misc:

110 1 2 3 110 Oper
100 100 1
Sample Mean
90 3.0SL=87.96 90 2
Average
80 X=80.75 80
-3.0SL=73.54 3
70
70
60
60
50
40 50
30 40
0 Part/Truck 1 2 3 4 5 6 7 8 9 10

15 1 2 3 110
100
Sample Range
3.0SL=12.52
10 90
80
70
5
R=3.833 60
50
0 -3.0SL=0.00E+00 40
0 Oper 1 2 3

500 110
%Total Var 100
400 %Study Var 90
Percent
300 %Toler 80
70
200
60
100 50
0 40
Operator by Part Interaction: Shows interaction By Operator: Main effect plot for Operator
between the operator and the part
We want a flat horizontal line. This would indicate that it
We want the lines to be on top of each other. Crossed is unlikely the operator is having an effect...confirm with
lines could indicate the potential for interaction. Confirm the p-value from ANOVA. This chart shows the
suspected interactions with the p-value in ANOVA Table. measurements, by operator. Some dots may be on top of
We want all operators measuring the parts the very each other. Visually, you can see each operator’s
similarly (ie, getting the same measurement). Use this measurements.
chart to help you look at particular parts which operators
may be measuring differently. By reviewing this chart, By Part: Main effect plot for Part.
we notice that the operators are measuring Part 4 & Part
10 differently. We would want to investigate why. This
We want to see lots of changes in slope. This would
chart would also show us if one operator measures higher
indicate that it is likely that the parts are having an effect.
or lower than the others. It may indicate a
Confirm with the p-value from Anova. (Remember, we
Reproducibility issue. It appears that Operator 2 is
set up our study this way. We wanted differences in the
measuring slightly lower than Operators 1 & 3. The
parts.) More importantly, look for parts that have varying
Minitab Session Window can also tell us more.
measurements. For instance, part 10 shows more
variation in measurement than some of the other parts.
D M A I C
Minitab Gage R&R ANOVA Table 62
ANOVA TABLE
Two-Way ANOVA Table With Interaction
Source DF SS MS F P
Part/Truck 9 20587.1 2287.45 39.7178 0.00000

Sources of Oper 2 480.0 240.00 4.1672 0.03256
Variability Oper*Part/Truck 18 036.7 7.59 4.4588 0.00016
Repeatability 30 387.5 12.92
Total 59 22491.2
P< 0.05 indicates statistically
significant!
Are any of these “p-values” less than 0.05?

If so, their sources of variation can be considered statistically significant (i.e,
Active, Influential).
For this gage... parts, operators and the interaction between parts & operators
are statistically significant sources of variation!
There are 4 major outputs other than the graphical Oper*Part/Truck: p=0.00016; This is the interaction
summary. We will discuss each separately. between the operator and the parts (eg, when operators
measure/record certain travel times). Since this is
Explanation of output for ANOVA Table: statistically significant, it is indicating that there may be
some issues when some operators measure/record certain
Part/Truck: p=0.00000; I would expect the parts to be runs. We will look further into the Minitab output to find
significant because we chose truck travel routes that out which run and which operators.
were really different from each other (short runs and long
runs).
Oper: p=0.03256; This is telling us that the Operators are

statistically significant. I need to look further to find out
if it’s a repeatability or a reproducibility issue (Minitab
does this for us in both the Comparisons Table and in the
charts. We’ll show you later!)
D M A I C
Minitab Gage R&R % Tolerance 63
% Comparisons Table
P/T
P/TRatio
Ratio
(Precision-to-Tolerance Total Gage R&R 10.67 32.66 171.53
(Precision-to-ToleranceRatio)
Ratio) Repeatability 3.10 17.62 92.54
Oper 2.19 14.81 77.76
Oper*Part/Truck 5.37 23.17 121.70
Part-To-Part 89.33 94.52 496.41
Total Variation 100.00 100.00 525.21
“R&R
“R&Rasasaa%
%ofof
Tolerance”
Tolerance”
% Tolerance
Acceptable Ranges
Not Acceptable
30%
Questionable
10%
Acceptable
If you wanted to calculate %Tolerance by hand:

5 .1 5 • σ R & R ( M e a s . S y s t .)
% T o le r a n c e = ∗1 0 0 %
T o le r a n c e
This shows us the % of the Tolerance which is being If % Tolerance is unacceptable now, then once you
taken up by the gage. For instance, in this case, 171.53% reach a Six Sigma Process, the measurement variation
of our tolerance is taken up by the variation of the gage. will be too large compared to the process variation.
Refer to the next page for a visual explanation of % In this case, the gage fails % Tolerance.
Tolerance. Note: You must put in the tolerance in
Minitab in order to get this data. If you do not enter a
tolerance, these figures will not appear. If you do not
have a tolerance, then % Contribution would be more
important in your analysis.
% Tolerance = A good indicator for Future Status

For this Gage: %Tolerance GRR = 171.53%
This Gage Fails the Criteria for %Tolerance for the
Total Gage RR Piece.
D M A I C
Minitab Gage R&R % Tolerance Graphically 64
%Tolerance: Graphically
LSL USL
%Tolerance = 10%
%Tolerance = 30%
%Tolerance = 70%
Graphically, this shows the amount of the engineering

tolerance used up just from the measurement system.
D M A I C
Minitab Gage R&R % Contribution 65
% Comparisons Table Variation due to the gage

+ Variation due to the parts
= 100%

%R&R
%R&RContribution
Contribution(Total
(Total
Variation)
Variation)
Total Gage R&R 10.67 32.66 171.53
Repeatability 3.10 17.62 92.54
Oper 2.19 14.81 77.76
Oper*Part/Truck 5.37 23.17 121.70
Part-To-Part 89.33 94.52 496.41
Total Variation 100.00 100.00 525.21
“R&R
“R&Ras
asaa%%of
ofTotal
TotalProcess
Process
Variation”
Variation” Further Explanation:
Source %Contribution
Total Gage R&R 10.67

Repeatability 3.10 3.10 + 7.56 = 10.67
Reproducibility 7.56
% Contribution
Oper 2.19 Now, we know that
Acceptable Ranges Oper*Part/Truck 5.37 Reproducibility is our bigger
Part-To-Part 89.33 issue (contributing more than
Not Acceptable
8.0% Total Variation 100.00 Repeatability).
Questionable
2.0%
2.19 + 5.37 = 7.56
Acceptable
Now, we know that of the Reproducibility, the Oper to Part/Truck
interaction is our bigger issue than the Reproducibility of
Operators. Check with the Oper*Part/Truck Interaction Graph to
find out which parts may be the issue.
% Contribution = a good indicator for Present Status Note: If you don’t have a tolerance, then, %
Contribution becomes a very critical measurement for
The measurement system (gage) is contributing to
Gage R&R. It gives you a good indication of the
10.67% of the total variation and the parts are
measurement variation, now, compared to a realistic
contributing 89.33%.
estimate of current process spread.
This Gage Fails the Criteria for % Contribution, but it
is close to passing.
If you wanted to calculate %R&R Contribution by hand:
 σ 2

% R&R  R & R

=
( M e a s . S y s t .)
  × 1 0 0 %
Contribution  σ 2
T o t a l 
 
D M A I C
Minitab Gage R&R Discrimination Index 66
Discrimination Index
Number of Distinct Categories = 4
Discrimination
DiscriminationIndex
Index
For this gage: the Discrimination Index = 4
• This says that the gage has reached the minimum acceptable
level.
Discrimination Index:
Discrimination Index:
•Provides the number of divisions that the Measurement System can accurately
•Provides the number of divisions that the Measurement System can accurately
measure
measureacross
acrossthe
theprocess
processvariation.
variation.
“Number
“Numberof
ofDistinct
Distinct •Indicates
•Indicateshow
how well
wellaagage
gagecan
candetect
detectpart-to-part
part-to-partvariation
variation---- process
processshifts
shiftsand
and
Categories”
Categories” improvement.
improvement.
Discrimination
DiscriminationIndex
IndexAcceptable
AcceptableValues
Values
•Less than 2, inadequate
•Equal to 2, equal to a go/nogo gauge.
•Minimum acceptable value = 4.
•Optimal = 10 or more.
If you wanted to calculate
Discrimination Index by hand:
 
 σ2 
Std Dev (Parts) D is c rim = 2 •  2 T o ta l  −1
x 1.41  σ R &R 
Std Dev (GRR)  ( M e a s .S y s t . ) 
Discrimination Index will tell us how many categories or

“buckets” the measurement device is able to see or
discriminate. If you’ve chosen parts that are different
(eg, in size, length, width, etc), the Discrimination Index
will indicate how many distinct categories are present. In
our example, Minitab is able to see 4 distinct categories.
Note: If you’ve designed and ran your study with similar
parts (on purpose), this reading would be meaningless to
you.
D M A I C
Minitab Gage R&R Key Indices 67
33Key
KeyIndices
Indicesfor
forMeasurement
MeasurementSystem
SystemCapability
Capability
11 22 33
P/T Discrimination
DiscriminationIndex
P/TRatio
Ratio %R&R
%R&RContribution
Contribution Index
(Precision-to-Tolerance
(Precision-to-ToleranceRatio)
Ratio)
“R&R
“R&Rasasaa%
%ofof “R&R
“R&Ras
asaa%%of
ofTotal
Total “Number
“Numberof
ofDistinct
Distinct
Tolerance”
Tolerance” Measure Variation”
Measure Variation” Categories”
Categories”
%Tolerance GRR = 171.53% % R&R Contribution = 10.67% Discrimination Index = 4
The next question to be answered: What do I do now We also know that Operators and the
that the % Tolerance & % Contribution has failed? Operator*Part/Truck Interaction is significant
(from p-value). This is an indication that the
Well, this is what we have learned so far about the break- operators are measuring differently. It also tells me
down of % Tolerance and % Contribution: there is some variation when certain operators
For % Tolerance and % Contribution, we know the measure certain trucks. Again, the graphs indicate
bigger issue is reproducibility. And, with which trucks and which operators.
Reproducibility, the bigger issue is the interaction
between parts and operators. By looking at the I would investigate those areas mentioned above.
graphical summary, it appears that trucks 4 and 10 Depending on your findings, you may need to implement
may be the issue. Investigate why there is a a new training program with your operators on how to
difference. Is there something unique about those measure/record the time. You also may need to do
truck routes? Is it harder to measure the time for something like having the computer measure the time
Trucks 4 & 10 due to time length? Should there be an when we measure parts similar to Parts 4 & 10. Once
Operational Definition or Procedure on how to you’ve done those items, go back and re-run the Gage
measure? Was there a problem with calculating the R&R using your implemented changes.
time from target?
D M A I C
Gage R&R Acceptable Ranges 68
Acceptable
Acceptable Ranges–ANOVA
Ranges–ANOVA METHOD
METHOD
This tells us the ratio of each
This tells us how much of the component of variation (the
This tells us how much
tolerance is being taken up by standard deviation for each
variation the gage itself is
the variation the gage. component divided by the total
contributing to the variation.
standard deviation).
%Tolerance % Discrimination Index

%Contribution
Not Acceptable
8.0% 30% 4 30%
Questionable
2.0% 10% 10 15%
Acceptable
• Provides the number of divisions that the Measurement System can accurately
measure across the process variation.
• Indicates how well a gage can detect part-to-part variation--process shifts and
improvement.
D M A I C

Gage R&R (Continuous Data) Conducted: 1/2/02
Gage R&R Results

Part p-value 0.00000 Y N
Oper p-value 0.03256 Y N
Oper & Part p-value 0.00016 Y N
Pass?
2) % Tolerance 171.53% [≤ 30%] Y N
3) % Contribution 10.67% [≤ 8%] Y N
4) % Study 32.66% [≤ 30%] Y N
5) # Distinct Categories 4 [≥ 4 ] Y N

Between Oper
_______________________________________________________________
Gage R&R Pass? Y N, If NO:
Plan for improvement:
Look at the gage and possibly change to a digital clock.
Review/Implement a driver call-in procedure. Suspect there is variation between drivers.
Review/develop a new spreadsheet for calculating the time.
Talk with operators to determine why Oper 2 measures differently than Operators 1 & 3.
Talk with customers about the right specification, if possible.
Train all operators and truck drives on new procedures.
Re-run Gage R&R to verify improvements.
D M A I C
D M A I C
1
Analyze Module Objectives
Learn and apply the tools and concepts

in the Analyze phase
You will learn the concepts and tools for the Analyze
phase during this classroom session. After the classroom
session, you can use the Six Sigma training CD to
review how the Analyze phase is applied.
DMAIC GB N TX PG V 4.2.0
D M A I C
2
Analyze Phase Flowchart
D M A I C
ANALYZE
PHASE
OVERVIEW
Analyze 4: Analyze 5: Analyze 6:

Establish Define Identify
Process Performance Variation
Capability Objectives Sources
D M A I C
3
The12-Step Process

Define
Measure
Standards Project Y
R&R
Analyze
Graphical Analysis,
Hypothesis Tests
Improve
Relationships
Tolerances
Control
Capability
D M A I C
4
Analyze Phase Overview
What is the Analyze phase?

The Analyze phase is when your team:
Calculates baseline process capability for the process
Defines the improvement goal for the project
Analyzes historical data to identify the sources of variation
Why is the Analyze phase important?

This phase is important because it clearly defines how well the process is
currently performing and identifies how much the process will be improved.
Steps involved in the Analyze phase

Analyze 4: Establish process capability
Analyze 5: Define performance objectives
Analyze 6: Identify variation sources
D M A I C
5
Statistical Thinking
D M A I C
critical X’s on gains
– Variation
Collect data
D M A I C
D M A I C
Analyze 4–Establish Process Capability 1
What does it mean to Establish Process Capability?

Process capability refers to the ability of a process to produce a defect-free product or
service. In this step, you will determine how consistently your product or process
meets the performance standard for your project Y calculating the sigma level. The
sigma level is calculated through statistical analysis of the collected data.
Why is it important to Establish Process Capability?

You can’t set a measurable goal without a clear understanding of where you are. It is
important to establish process capability in order to baseline your current process
performance. This will be the starting point from which you will set your improvement
goals.
What are the project tasks for completing Analyze 4?

4.1 Graphically analyze data for project Y (continuous data only)
4.2 Calculate baseline sigma for project Y
ANALYZE STEP
OVERVIEW

Establish Define Identify
Process Performance Variation
Capability Objectives Sources
4.1 Graphically analyze data for project Y (continuous data only)

4.2 Calculate baseline sigma for project Y
DMAIC GB O TX PG V 4.2.0
D M A I C
Step 4.1: Graphically Analyze Data For

Project Y
D M A I C
Review: Describing Variation 3
Prior to Calculating Capability, we need to know:

Key question #1–Stability–Variation over time (Run Chart)
How stable is the data?
Key Question #2–Shape, Spread–Variation for a period of
time: Data Distributions (Graphical Analysis)
What is the shape of the distribution–symmetrical, lopsided,
twin peaks, long-tailed? (determination of normality)
What is the central tendency (“center” or “average”)
of the distribution?
What is the variation (“spread”) of the distribution–wide
or narrow?
These attributes of your data will help determine which

statistical calculations are better descriptors of your data
set.
D M A I C
Graphically Analyze Data 4
We can use the following example:

Data set: Holdtime.MPJ
1. First, look at Stability:
STAT > Quality Tools > Run Chart
- Approx p-value for clustering = 0.1139
- Approx p-value for mixtures = 0.8861
- Approx p-value for trends = 0.2883
- Approx p-value for oscillator = 0.7117
p ≥ 0.05 indicates stability. Therefore, data is stable.
2. Secondly, look at Shape:
STAT > Basic Statistics > Normality Test
Anderson-Darling p-value = 0.885
p-value ≥ 0.05 indicates data is normal. Therefore, data is normal.
3. Third, look at Centering and Spread:
STAT > Basic Statistics > Display Descriptive Statistics
Again, p-value = 0.885. Data is normal.
Mean = 8.5 and S = 0.10
p ≥ 0.05 indicates normal data. Therefore, will use mean and standard deviation to
describe data.
D M A I C
Steps 4.2: Calculate Baseline Sigma
D M A I C
What Is Process Capability? 6
A Measurement Scale Which Compares

the Output of a Process to
the Performance Standard
Process capability allows us to compare how “capable”

the process is of meeting customer CTQ’s by looking at
the probability that the process will produce a defect.
D M A I C
Common Metric For Comparison 7
Process Performance
Purchase Order 98% accuracy
Generation
Accounts Receivable 33 days average aging
Customer Service 82% rated 4 or 5 on responsiveness
Supplier Delivery 95% on-time delivery
Which process is performing best?
It’s difficult to compare processes with very different

kinds of measure. By expressing process capability as a
“sigma” value we can compare different types of
processes.
D M A I C
Data Analysis Roundup 8
Process Capability Tools and Terminology

e.g Cycle Time, Length,
e.g Light off Weight…
e.g Light On
Discrete
DiscreteData
Data Continuous
Continuous
Data
Data
Defects per Six Sigma Process
Opportunity Report:
Defects per Million ZLong term
Opportunities
ZShort term = ZBench =
Six Sigma Product reported yield
Report
ZShift
In this step you will learn how to calculate discrete and

continuous data.
A note on terms to describe capability. All the following

are synonyms:
ZLT = ZBenchLT = Sigma CapabilityLT (or, you may see

capability written as σLT or SigmaLT. This is no longer
standard usage to denote capability).
ZST = ZBenchST = Sigma CapabilityST = ZBench

(σ or SigmaST is no longer recommended).
D M A I C
Process Capability Continuous Data 9
Verify we have a normal distribution

Calculate ZLSL and/or ZUSL
Determine probability of a defective
Determine ZBench
We will look at how to do each of these steps in detail!
D M A I C
Calculating Z–Continuous Data 10
You can calculate a Z-value for any given value of x. Z is the

number of standard deviations which will fit between the mean and
the value of x. This is known as a Z-score.
X− µ
Z=
σ
µ 1σ 2σ 3σ 4σ
The Z-score is a way to transform any normal distribution

to the standard normal distribution. The standard normal
distribution has a mean of 0 and a standard deviation of 1.
This transformation allows us to compare two entirely
different processes on a common scale-that of standard
deviation units.
You must have a normal distribution: A stable process

prior to using this method. If not, you’ll need to use the
Discrete, DPMO method.
D M A I C
Calculating Capability 11
X = 8.5
s = 0.1
xx
Probability
Probability Probability
Probability
ofofaadefect
defect LSL
LSL USL
USL ofofaadefect
defect
less
lessthan
than greater
ZUSL greater
LSL
LSL ZLSL than
thanUSL
USL
Standard Deviations -4 -3 -2 -1 0 1 2 3 4
Units of Measure 8.1 8.2 8.3 8.4 8.5 8.6 8.7 8.8 8.9
ZUSL = USL - X = 8.7 - 8.5 = 0.2 = 2

s 0.1 0.1
ZLSL = X - LSL = 8.5 - 8.2 = 0.3 = 3

s 0.1 0.1
ZUSL = Look up Z = 2.0 => 0.0228
LSL = Look up Z = 3 => 0.00135
Area Total = 0.0228 + 0.00135 = .02415
Yield = 1-.02415 = 0.97585 = 97.6%
Go to Memory Jogger (or Abridged Sigma Table) and

look up a Yield of 97.6%
ZST = 3.4
D M A I C
Reading The Z Table 12
Yield
Probability
Of A Defective
Example = .028066
Specification
Limit
Z-score
Units of Measure
0.20 .420740315 1.71 .043632958 3.22 .000640954 4.73 .000001153
0.25 .401293634 1.76 .039203955 3.27 .000537758 4.78 .000000903
.382088486 1.81 .035147973 3.32 .000450127 4.83 .000000705
1.86 .031442864 3.37 .000375899 4.88 .000000550
1.91 .028066724 3.42 .000313179 4.93 .000000428
1.96 .024998022 3.47 .000260317 4.98 .000000332
2.01 .022215724 3.52 .000215873 5.03 .000000258
2.06
2.11 .019699396 3.57 .000178601 5.08 .000000199
2.16 .017429293 3.62 .000147419 5.13 .000000154
To the
2.21 .015386434 3.67 .000121399 5.18 .000000118
2.26 .013552660 3.72 .000099739 5.23 .000000091
2.31
2.36 .011910681 3.77 .000081753 5.28 .000000070
right of the
.010444106 3.82 .000066855 5.33 .000000053
.009137469 3.87 .000054545 5.38 .000000041
2.41 .007976235 3.92 .000044399 5.43 .000000031
2.46 .006946800 3.97 .000036057 5.48 .000000024
Z-score Z = 1.91
2.51
2.56
2.61
.006036485
.005233515
.004527002
4.02
4.07
4.12
.000029215
.000023617
.000019047
5.53
5.58
5.63
.000000018
.000000014
.000000010
you will 2.66

2.71
2.76
.003906912
.003364033
.002889938
4.17
4.22
4.27 Table Of
note the 1.35
1.40
.088507862
.080756531
2.81
2.86
2.91
.002476947
.002118083
.001807032
4.32
4.37
4.42 Area Under
1.45 .073529141 2.96 .001538097 4.47
tail area 1.50 .066807100 3.01 .001306156 4.52
The Normal
Curve
Yield = Probability of a non-defective
From the Z Table, we can get the probability of a

defective. 1-Probability of defective = Yield.
This is what we’ll need to eventually calculate Process

Capability.
D M A I C z
(Values of Z from 0.00 to 4.99)
Single–Tail Z Table 13
Z 0.00 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09
0.00 5.00e-001 4.96e-001 4.92e-001 4.88e-001 4.84e-001 4.80e-001 4.76e-001 4.72e-001 4.68e-001 4.64e-001
0.10 4.60e-001 4.56e-001 4.52e-001 4.48e-001 4.44e-001 4.40e-001 4.36e-001 4.33e-001 4.29e-001 4.25e-001
0.20 4.21e-001 4.17e-001 4.13e-001 4.09e-001 4.05e-001 4.01e-001 3.97e-001 3.94e-001 3.90e-001 3.86e-001
0.30 3.82e-001 3.78e-001 3.74e-001 3.71e-001 3.67e-001 3.63e-001 3.59e-001 3.56e-001 3.52e-001 3.48e-001
0.40 3.45e-001 3.41e-001 3.37e-001 3.34e-001 3.30e-001 3.26e-001 3.23e-001 3.19e-001 3.16e-001 3.12e-001
0.50 3.09e-001 3.05e-001 3.02e-001 2.98e-001 2.95e-001 2.91e-001 2.88e-001 2.84e-001 2.81e-001 2.78e-001
0.60 2.74e-001 2.71e-001 2.68e-001 2.64e-001 2.61e-001 2.58e-001 2.55e-001 2.51e-001 2.48e-001 2.45e-001
0.70 2.42e-001 2.39e-001 2.36e-001 2.33e-001 2.30e-001 2.27e-001 2.24e-001 2.21e-001 2.18e-001 2.15e-001
0.80 2.12e-001 2.09e-001 2.06e-001 2.03e-001 2.00e-001 1.98e-001 1.95e-001 1.92e-001 1.89e-001 1.87e-001
0.90 1.84e-001 1.81e-001 1.79e-001 1.76e-001 1.74e-001 1.71e-001 1.69e-001 1.66e-001 1.64e-001 1.61e-001
1.00 1.59e-001 1.56e-001 1.54e-001 1.52e-001 1.49e-001 1.47e-001 1.45e-001 1.42e-001 1.40e-001 1.38e-001
1.10 1.36e-001 1.33e-001 1.31e-001 1.29e-001 1.27e-001 1.25e-001 1.23e-001 1.21e-001 1.19e-001 1.17e-001
1.20 1.15e-001 1.13e-001 1.11e-001 1.09e-001 1.07e-001 1.06e-001 1.04e-001 1.02e-001 1.00e-001 9.85e-002
1.30 9.68e-002 9.51e-002 9.34e-002 9.18e-002 9.01e-002 8.85e-002 8.69e-002 8.53e-002 8.38e-002 8.23e-002
1.40 8.08e-002 7.93e-002 7.78e-002 7.64e-002 7.49e-002 7.35e-002 7.21e-002 7.08e-002 6.94e-002 6.81e-002
1.50 6.68e-002 6.55e-002 6.43e-002 6.30e-002 6.18e-002 6.06e-002 5.94e-002 5.82e-002 5.71e-002 5.59e-002
1.60 5.48e-002 5.37e-002 5.26e-002 5.16e-002 5.05e-002 4.95e-002 4.85e-002 4.75e-002 4.65e-002 4.55e-002
1.70 4.46e-002 4.36e-002 4.27e-002 4.18e-002 4.09e-002 4.01e-002 3.92e-002 3.84e-002 3.75e-002 3.67e-002
1.80 3.59e-002 3.51e-002 3.44e-002 3.36e-002 3.29e-002 3.22e-002 3.14e-002 3.07e-002 3.01e-002 2.94e-002
1.90 2.87e-002 2.81e-002 2.74e-002 2.68e-002 2.62e-002 2.56e-002 2.50e-002 2.44e-002 2.39e-002 2.33e-002
2.00 2.28e-002 2.22e-002 2.17e-002 2.12e-002 2.07e-002 2.02e-002 1.97e-002 1.92e-002 1.88e-002 1.83e-002
2.10 1.79e-002 1.74e-002 1.70e-002 1.66e-002 1.62e-002 1.58e-002 1.54e-002 1.50e-002 1.46e-002 1.43e-002
2.20 1.39e-002 1.36e-002 1.32e-002 1.29e-002 1.25e-002 1.22e-002 1.19e-002 1.16e-002 1.13e-002 1.10e-002
2.30 1.07e-002 1.04e-002 1.02e-002 9.90e-003 9.64e-003 9.39e-003 9.14e-003 8.89e-003 8.66e-003 8.42e-003
2.40 8.20e-003 7.98e-003 7.76e-003 7.55e-003 7.34e-003 7.14e-003 6.95e-003 6.76e-003 6.57e-003 6.39e-003
2.50 6.21e-003 6.04e-003 5.87e-003 5.70e-003 5.54e-003 5.39e-003 5.23e-003 5.08e-003 4.94e-003 4.80e-003
2.60 4.66e-003 4.53e-003 4.40e-003 4.27e-003 4.15e-003 4.02e-003 3.91e-003 3.79e-003 3.68e-003 3.57e-003
2.70 3.47e-003 3.36e-003 3.26e-003 3.17e-003 3.07e-003 2.98e-003 2.89e-003 2.80e-003 2.72e-003 2.64e-003
2.80 2.56e-003 2.48e-003 2.40e-003 2.33e-003 2.26e-003 2.19e-003 2.12e-003 2.05e-003 1.99e-003 1.93e-003
2.90 1.87e-003 1.81e-003 1.75e-003 1.69e-003 1.64e-003 1.59e-003 1.54e-003 1.49e-003 1.44e-003 1.39e-003
3.00 1.35e-003 1.31e-003 1.26e-003 1.22e-003 1.18e-003 1.14e-003 1.11e-003 1.07e-003 1.04e-003 1.00e-003
3.10 9.68e-004 9.35e-004 9.04e-004 8.74e-004 8.45e-004 8.16e-004 7.89e-004 7.62e-004 7.36e-004 7.11e-004
3.20 6.87e-004 6.64e-004 6.41e-004 6.19e-004 5.98e-004 5.77e-004 5.57e-004 5.38e-004 5.19e-004 5.01e-004
3.30 4.83e-004 4.66e-004 4.50e-004 4.34e-004 4.19e-004 4.04e-004 3.90e-004 3.76e-004 3.62e-004 3.49e-004
3.40 3.37e-004 3.25e-004 3.13e-004 3.02e-004 2.91e-004 2.80e-004 2.70e-004 2.60e-004 2.51e-004 2.42e-004
3.50 2.33e-004 2.24e-004 2.16e-004 2.08e-004 2.00e-004 1.93e-004 1.85e-004 1.78e-004 1.72e-004 1.65e-004
3.60 1.59e-004 1.53e-004 1.47e-004 1.42e-004 1.36e-004 1.31e-004 1.26e-004 1.21e-004 1.17e-004 1.12e-004
3.70 1.08e-004 1.04e-004 9.96e-005 9.57e-005 9.20e-005 8.84e-005 8.50e-005 8.16e-005 7.84e-005 7.53e-005
3.80 7.23e-005 6.95e-005 6.67e-005 6.41e-005 6.15e-005 5.91e-005 5.67e-005 5.44e-005 5.22e-005 5.01e-005
3.90 4.81e-005 4.61e-005 4.43e-005 4.25e-005 4.07e-005 3.91e-005 3.75e-005 3.59e-005 3.45e-005 3.30e-005
4.00 3.17e-005 3.04e-005 2.91e-005 2.79e-005 2.67e-005 2.56e-005 2.45e-005 2.35e-005 2.25e-005 2.16e-005
4.10 2.07e-005 1.98e-005 1.89e-005 1.81e-005 1.74e-005 1.66e-005 1.59e-005 1.52e-005 1.46e-005 1.39e-005
4.20 1.33e-005 1.28e-005 1.22e-005 1.17e-005 1.12e-005 1.07e-005 1.02e-005 9.77e-006 9.34e-006 8.93e-006
4.30 8.54e-006 8.16e-006 7.80e-006 7.46e-006 7.12e-006 6.81e-006 6.50e-006 6.21e-006 5.93e-006 5.67e-006
4.40 5.41e-006 5.17e-006 4.94e-006 4.71e-006 4.50e-006 4.29e-006 4.10e-006 3.91e-006 3.73e-006 3.56e-006
4.50 3.40e-006 3.24e-006 3.09e-006 2.95e-006 2.81e-006 2.68e-006 2.56e-006 2.44e-006 2.32e-006 2.22e-006
4.60 2.11e-006 2.01e-006 1.92e-006 1.83e-006 1.74e-006 1.66e-006 1.58e-006 1.51e-006 1.43e-006 1.37e-006
4.70 1.30e-006 1.24e-006 1.18e-006 1.12e-006 1.07e-006 1.02e-006 9.68e-007 9.21e-007 8.76e-007 8.34e-007
4.80 7.93e-007 7.55e-007 7.18e-007 6.83e-007 6.49e-007 6.17e-007 5.87e-007 5.58e-007 5.30e-007 5.04e-007
4.90 4.79e-007 4.55e-007 4.33e-007 4.11e-007 3.91e-007 3.71e-007 3.52e-007 3.35e-007 3.18e-007 3.02e-007
D M A I C z
(Values of Z from 5.00 to 9.99)
Single–Tail Z Table 14
Z 0.00 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09
5.00 2.87e-007 2.72e-007 2.58e-007 2.45e-007 2.33e-007 2.21e-007 2.10e-007 1.99e-007 1.89e-007 1.79e-007
5.10 1.70e-007 1.61e-007 1.53e-007 1.45e-007 1.37e-007 1.30e-007 1.23e-007 1.17e-007 1.11e-007 1.05e-007
5.20 9.96e-008 9.44e-008 8.95e-008 8.48e-008 8.03e-008 7.60e-008 7.20e-008 6.82e-008 6.46e-008 6.12e-008
5.30 5.79e-008 5.48e-008 5.19e-008 4.91e-008 4.65e-008 4.40e-008 4.16e-008 3.94e-008 3.72e-008 3.52e-008
5.40 3.33e-008 3.15e-008 2.98e-008 2.82e-008 2.66e-008 2.52e-008 2.38e-008 2.25e-008 2.13e-008 2.01e-008
5.50 1.90e-008 1.79e-008 1.69e-008 1.60e-008 1.51e-008 1.43e-008 1.35e-008 1.27e-008 1.20e-008 1.14e-008
5.60 1.07e-008 1.01e-008 9.55e-009 9.01e-009 8.50e-009 8.02e-009 7.57e-009 7.14e-009 6.73e-009 6.35e-009
5.70 5.99e-009 5.65e-009 5.33e-009 5.02e-009 4.73e-009 4.46e-009 4.21e-009 3.96e-009 3.74e-009 3.52e-009
5.80 3.32e-009 3.12e-009 2.94e-009 2.77e-009 2.61e-009 2.46e-009 2.31e-009 2.18e-009 2.05e-009 1.93e-009
5.90 1.82e-009 1.71e-009 1.61e-009 1.51e-009 1.43e-009 1.34e-009 1.26e-009 1.19e-009 1.12e-009 1.05e-009
6.00 9.87e-010 9.28e-010 8.72e-010 8.20e-010 7.71e-010 7.24e-010 6.81e-010 6.40e-010 6.01e-010 5.65e-010
6.10 5.30e-010 4.98e-010 4.68e-010 4.39e-010 4.13e-010 3.87e-010 3.64e-010 3.41e-010 3.21e-010 3.01e-010
6.20 2.82e-010 2.65e-010 2.49e-010 2.33e-010 2.19e-010 2.05e-010 1.92e-010 1.81e-010 1.69e-010 1.59e-010
6.30 1.49e-010 1.40e-010 1.31e-010 1.23e-010 1.15e-010 1.08e-010 1.01e-010 9.45e-011 8.85e-011 8.29e-011
6.40 7.77e-011 7.28e-011 6.81e-011 6.38e-011 5.97e-011 5.59e-011 5.24e-011 4.90e-011 4.59e-011 4.29e-011
6.50 4.02e-011 3.76e-011 3.52e-011 3.29e-011 3.08e-011 2.88e-011 2.69e-011 2.52e-011 2.35e-011 2.20e-011
6.60 2.06e-011 1.92e-011 1.80e-011 1.68e-011 1.57e-011 1.47e-011 1.37e-011 1.28e-011 1.19e-011 1.12e-011
6.70 1.04e-011 9.73e-012 9.09e-012 8.48e-012 7.92e-012 7.39e-012 6.90e-012 6.44e-012 6.01e-012 5.61e-012
6.80 5.23e-012 4.88e-012 4.55e-012 4.25e-012 3.96e-012 3.69e-012 3.44e-012 3.21e-012 2.99e-012 2.79e-012
6.90 2.60e-012 2.42e-012 2.26e-012 2.10e-012 1.96e-012 1.83e-012 1.70e-012 1.58e-012 1.48e-012 1.37e-012
7.00 1.28e-012 1.19e-012 1.11e-012 1.03e-012 9.61e-013 8.95e-013 8.33e-013 7.75e-013 7.21e-013 6.71e-013
7.10 6.24e-013 5.80e-013 5.40e-013 5.02e-013 4.67e-013 4.34e-013 4.03e-013 3.75e-013 3.49e-013 3.24e-013
7.20 3.01e-013 2.80e-013 2.60e-013 2.41e-013 2.24e-013 2.08e-013 1.94e-013 1.80e-013 1.67e-013 1.55e-013
7.30 1.44e-013 1.34e-013 1.24e-013 1.15e-013 1.07e-013 9.91e-014 9.20e-014 8.53e-014 7.91e-014 7.34e-014
7.40 6.81e-014 6.31e-014 5.86e-014 5.43e-014 5.03e-014 4.67e-014 4.33e-014 4.01e-014 3.72e-014 3.44e-014
7.50 3.19e-014 2.96e-014 2.74e-014 2.54e-014 2.35e-014 2.18e-014 2.02e-014 1.87e-014 1.73e-014 1.60e-014
7.60 1.48e-014 1.37e-014 1.27e-014 1.17e-014 1.09e-014 1.00e-014 9.30e-015 8.60e-015 7.95e-015 7.36e-015
7.70 6.80e-015 6.29e-015 5.82e-015 5.38e-015 4.97e-015 4.59e-015 4.25e-015 3.92e-015 3.63e-015 3.35e-015
7.80 3.10e-015 2.86e-015 2.64e-015 2.44e-015 2.25e-015 2.08e-015 1.92e-015 1.77e-015 1.64e-015 1.51e-015
7.90 1.39e-015 1.29e-015 1.19e-015 1.10e-015 1.01e-015 9.33e-016 8.60e-016 7.93e-016 7.32e-016 6.75e-016
8.00 6.22e-016 5.74e-016 5.29e-016 4.87e-016 4.49e-016 4.14e-016 3.81e-016 3.51e-016 3.24e-016 2.98e-016
8.10 2.75e-016 2.53e-016 2.33e-016 2.15e-016 1.98e-016 1.82e-016 1.68e-016 1.54e-016 1.42e-016 1.31e-016
8.20 1.20e-016 1.11e-016 1.02e-016 9.36e-017 8.61e-017 7.92e-017 7.28e-017 6.70e-017 6.16e-017 5.66e-017
8.30 5.21e-017 4.79e-017 4.40e-017 4.04e-017 3.71e-017 3.41e-017 3.14e-017 2.88e-017 2.65e-017 2.43e-017
8.40 2.23e-017 2.05e-017 1.88e-017 1.73e-017 1.59e-017 1.46e-017 1.34e-017 1.23e-017 1.13e-017 1.03e-017
8.50 9.48e-018 8.70e-018 7.98e-018 7.32e-018 6.71e-018 6.15e-018 5.64e-018 5.17e-018 4.74e-018 4.35e-018
8.60 3.99e-018 3.65e-018 3.35e-018 3.07e-018 2.81e-018 2.57e-018 2.36e-018 2.16e-018 1.98e-018 1.81e-018
8.70 1.66e-018 1.52e-018 1.39e-018 1.27e-018 1.17e-018 1.07e-018 9.76e-019 8.93e-019 8.17e-019 7.48e-019
8.80 6.84e-019 6.26e-019 5.72e-019 5.23e-019 4.79e-019 4.38e-019 4.00e-019 3.66e-019 3.34e-019 3.06e-019
8.90 2.79e-019 2.55e-019 2.33e-019 2.13e-019 1.95e-019 1.78e-019 1.62e-019 1.48e-019 1.35e-019 1.24e-019
9.00 1.13e-019 1.03e-019 9.40e-020 8.58e-020 7.83e-020 7.15e-020 6.52e-020 5.95e-020 5.43e-020 4.95e-020
9.10 4.52e-020 4.12e-020 3.76e-020 3.42e-020 3.12e-020 2.85e-020 2.59e-020 2.37e-020 2.16e-020 1.96e-020
9.20 1.79e-020 1.63e-020 1.49e-020 1.35e-020 1.23e-020 1.12e-020 1.02e-020 9.31e-021 8.47e-021 7.71e-021
9.30 7.02e-021 6.39e-021 5.82e-021 5.29e-021 4.82e-021 4.38e-021 3.99e-021 3.63e-021 3.30e-021 3.00e-021
9.40 2.73e-021 2.48e-021 2.26e-021 2.05e-021 1.86e-021 1.69e-021 1.54e-021 1.40e-021 1.27e-021 1.16e-021
9.50 1.05e-021 9.53e-022 8.66e-022 7.86e-022 7.14e-022 6.48e-022 5.89e-022 5.35e-022 4.85e-022 4.40e-022
9.60 4.00e-022 3.63e-022 3.29e-022 2.99e-022 2.71e-022 2.46e-022 2.23e-022 2.02e-022 1.83e-022 1.66e-022
9.70 1.51e-022 1.37e-022 1.24e-022 1.12e-022 1.02e-022 9.22e-023 8.36e-023 7.57e-023 6.86e-023 6.21e-023
9.80 5.63e-023 5.10e-023 4.62e-023 4.18e-023 3.79e-023 3.43e-023 3.10e-023 2.81e-023 2.54e-023 2.30e-023
9.90 2.08e-023 1.88e-023 1.70e-023 1.54e-023 1.39e-023 1.26e-023 1.14e-023 1.03e-023 9.32e-024 8.43e-024
D M A I C
Abridged Process Sigma Conversion Table 15
ST Process Defects Per Defects Per Defects Per Defects Per Defects Per
Long-Term Yield
Sigma 1,000,000 100,000 10,000 1,000 100
99.99966% 6.0 3.4 0.34 0.034 0.0034 0.00034
99.9995% 5.9 5 0.5 0.05 0.005 0.0005
99.9992% 5.8 8 0.8 0.08 0.008 0.0008
99.9990% 5.7 10 1 0.1 0.01 0.001
99.9980% 5.6 20 2 0.2 0.02 0.002
99.9970% 5.5 30 3 0.3 0.03 0.003
99.9960% 5.4 40 4 0.4 0.04 0.004
99.9930% 5.3 70 7 0.7 0.07 0.007
99.9900% 5.2 100 10 1.0 0.1 0.01
99.9850% 5.1 150 15 1.5 0.15 0.015
99.9770% 5.0 230 23 2.3 0.23 0.023
99.9670% 4.9 330 33 3.3 0.33 0.033
99.9520% 4.8 480 48 4.8 0.48 0.048
99.9320% 4.7 680 68 6.8 0.68 0.068
99.9040% 4.6 960 96 9.6 0.96 0.096
99.8650% 4.5 1,350 135 13.5 1.35 0.135
99.8140% 4.4 1,860 186 18.6 1.86 0.186
99.7450% 4.3 2,550 255 25.5 2.55 0.255
99.6540% 4.2 3,460 346 34.6 3.46 0.346
99.5340% 4.1 4,660 466 46.6 4.66 0.466
99.3790% 4.0 6,210 621 62.1 6.21 0.621
99.1810% 3.9 8,190 819 81.9 8.19 0.819
98.930% 3.8 10,700 1,070 107 10.7 1.07
98.610% 3.7 13,900 1,390 139 13.9 1.39
98.220% 3.6 17,800 1,780 178 17.8 1.78
97.730% 3.5 22,700 2,270 227 22.7 2.27
97.130% 3.4 28,700 2,870 287 28.7 2.87
96.410% 3.3 35,900 3,590 359 35.9 3.59
95.540% 3.2 44,600 4,460 446 44.6 4.46
94.520% 3.1 54,800 5,480 548 54.8 5.48
93.320% 3.0 66,800 6,680 668 66.8 6.68
91.920% 2.9 80,800 8,080 808 80.8 8.08
90.320% 2.8 96,800 9,680 968 96.8 9.68
88.50% 2.7 115,000 11,500 1,150 115 11.5
86.50% 2.6 135,000 13,500 1,350 135 13.5
84.20% 2.5 158,000 15,800 1,580 158 15.8
81.60% 2.4 184,000 18,400 1,840 184 18.4
78.80% 2.3 212,000 21,200 2,120 212 21.2
75.80% 2.2 242,000 24,200 2,420 242 24.2
72.60% 2.1 274,000 27,400 2,740 274 27.4
69.20% 2.0 308,000 30,800 3,080 308 30.8
65.60% 1.9 344,000 34,400 3,440 344 34.4
61.80% 1.8 382,000 38,200 3,820 382 38.2
58.00% 1.7 420,000 42,000 4,200 420 42
54.00% 1.6 460,000 46,000 4,600 460 46
50% 1.5 500,000 50,000 5,000 500 50
46% 1.4 540,000 54,000 5,400 540 54
43% 1.3 570,000 57,000 5,700 570 57
39% 1.2 610,000 61,000 6,100 610 61
35% 1.1 650,000 65,000 6,500 650 65
31% 1.0 690,000 69,000 6,900 690 69
28% 0.9 720,000 72,000 7,200 720 72
25% 0.8 750,000 75,000 7,500 750 75
22% 0.7 780,000 78,000 7,800 780 78
19% 0.6 810,000 81,000 8,100 810 81
16% 0.5 840,000 84,000 8,400 840 84
14% 0.4 860,000 86,000 8,600 860 86
12% 0.3 880,000 88,000 8,800 880 88
10% 0.2 900,000 90,000 9,000 900 90
8% 0.1 920,000 92,000 9,200 920 92
Note: Subtract
1.5 to get long-
term
Sigma level
D M A I C
Long-Term vs. Short-Term Data 16
Long-Term Data
Y
(Continuous)
Short-Term
Data
Time
How much variation we observe in a process is For now, and as a default, treat all data as long-term data.
influenced by whether we are looking at long-term data The convention for Six Sigma is to report short-term, as
or short-term data. you will see later.
Long-Term: Why do we report Short-Term Sigma?
Data collected over a long enough period of time and We want to standardize reporting of the best a process
over diverse enough conditions such that it is likely to can do, given its current capability. Therefore,
contain some process shifts and other special causes. Short-Term Sigma tells us this.
Short-Term:
Data collected over a short enough period of time so that

shifts and other special causes are unlikely
D M A I C
Reporting Sigma Values 17
Short-Term Sigma = Long-Term Sigma + Sigma Shift

If “Shift” is unknown, then assume 1.5
Assume that sigma calculated from project data is long-term sigma
A rational subgrouping sampling scheme for data collection (in the
Measurement Phase) must have been used if you are calculating a shift
(other than using 1.5.)
It is possible to calculate short-term, but for most Green Special Cause = Between subgroup variation due to
Belt projects we do not attempt to do so; instead we use assignable causes, non-random influences.
the 1.5 sigma shift rule.
Common Cause = Within subgroup variation,
In order to truly calculate short-term sigma, you have to inherent in a process, random influences.
use several rational subgroups. If you did not use a
rational sub-grouping sampling scheme for data To ensure you are sampling* properly, these 3 items must
collection in the Measure phase, it will not be possible to be true:
calculate actual short-term sigma. (Therefore, add 1.5 to – Representative Data
long-term sigma.)
– Enough Data (at least 20 subgroups)
– Measured with something that passes the Gage R&R
*Refer to the workbook for further details on sampling

(Tools > Measure phase)
D M A I C
Principles Of Rational Subgrouping 18
1. Never knowingly subgroup unlike things together

2. Minimize variation within each subgroup
– Group homogeneous units, within a logic, within a reason
3. Maximize variation between subgroups
– The Xbar shows differences between subgroups that are bigger than that
shown within subgroups
4. Treat the chart in accordance with the use of the data
– Subgroup frequency should reflect the process
– Use individuals with limited data
– Use subgroups when logical
Rational subgrouping refers to grouping the data for

analysis in a meaningful way to understand variation.
Rational subgrouping attempts to select groups of data
such that mainly common cause variation is within
groups, and mainly special cause variation is between
groups.
D M A I C
Generalizing The Correction 19
Six Sigma Centered

Process
Capability
.0005 ppm .0005 ppm

T
T
LSL
LSL ± 6σ USL
USL
USL
Six Sigma Shifted 1.5σ

µµ
3.4 ppm
TT
LSL USL
4.5σ
The 1.5 shift is used as a compensatory off-set in the

mean to generally account for dynamic non-random
variations in process centering. It represents the average
amount of change in a typical process over many cycles
of that process.
If the process is centered, then Six Sigma capability

means that 6 standard deviations fit between the mean
and the specification limit.
If the process center “shifts” by 1.5 sigma , then the

average is 4.5 sigma from either the LSL or USL.
Six Sigma Short-Term Capability = 4.5 Sigma Long-

Term Capability
Zero Sigma Short-Term Capability = -1.5 Sigma Long-

Term Capability
D M A I C
The Universal Equation For Z 20
. . . so what are the possibilities? SL - λ

Z=
σ
SL =USL
LSL λ= T (Target)
µ (Mean)
Z =
st (short-term)
lt (long-term) σst
σlt
and how do we choose the right one?
SL − µ The larger the ZShift , the greater the control problem.

Z LT =
σ LT Typical ZShift = 1.5
Long-term is what is actually going on in the process.

SL - m is the distance the specification is from the mean.
SL − T
ZST =
σ ST
Short -term is the best the process can do.

SL - T is the distance the specification is from the target.
ZST is bigger than
ZShift = ZST − Z LT
D M A I C
Z-Bench 21
Short-Term
Long-Term
P(d)LSL P(d)USL
_
LSL T x USL
Z-Long-Term Z-Short-Term
SL - µ SL - T
Zlt = Zst =
σ lt σst
Z-Bench-Long-Term Z-Bench-Short-Term
USL- µ µ - LSL USL- T T - LSL
ZUSL= ZLSL= ZUSL= ZLSL=
σ lt σ lt σ st σ st
P(d)USL = from P(d)LSL = from P(d)USL = from P(d)LSL = from

Z table Z table Z table Z table
P(d)Total = P(d)USL + P(d)LSL P(d)Total = P(d)USL + P(d)LSL
ZB-lt = from ZB-st = from

Z table Z table
In Six Sigma, it is common to report out process

capability using ZBench. As shown previously, ZBench
corresponds to yield, i.e., we look at the chance of being
outside the specification limit on either side. ZBench is
calculated for both short and long-term. The difference
between the two is ZShift.
D M A I C
Activity–Calculating Process Capability–Continuous Data 22
What is the process capability for a process that has:
Mean = 5
Standard Deviation = 2
Upper Spec. Limit = 9
You can use the hand calculation or you can open the
spreadsheet: SIGMACAL.xls.
Enter:
Xbar: 5
S: 2
USL: 9
LSL: (leave blank)
Note: This gives you ZST (the 1.5 shift was added to ZLT).
D M A I C
Graphically Analyze Data–Breakout Activity (20 minutes) 23
Desired Outcome: Graphical Analysis of Capital Logistics Project Y Data

Minitab File: GB Case Study.mtw
What How Who Timing
Run Use the Run Chart tool in Minitab to All 5 mins.

Chart investigate the variation in the
project Y data over time
Shape, Use the Normal Probability Plot in All 10 mins.

Normality, Minitab to analyze the shape of the
Central project Y data
Tendency And
Use the Descriptive Statistics tool in
Spread
Minitab to analyze the shape,
normality, central tendency and
spread of the project Y data
Use the Minitab Six Sigma Process
Report to calculate Process Sigma
Solutions You can check your answers using All 5 mins.

the solution sheets on the following
pages
We always analyze the data this way:
1. Look at Stability–Is the process Stable?
2. Look at Shape–Do I have a normal distribution?
3. Look at the Spread–What measure of dispersion should I use?
Recall from our Case Study:
Time *(minutes early or late) = (Target Arrival Time) – (Actual Arrival Time)
Date 1 = Date of Shipment
* Spec. for time = ± 60 min.
D M A I C
Solution–Run Chart 24
Stability Calculations for Case study data.
First Step, Check Stability
Stat > Quality Tools > Run Chart
D M A I C
1. Double click on
“C5”
2. Double click on “C1”

Each Date is a sub-
group
3. Click ”OK”
D M A I C
Each Date is a column of

data points.
Run Chart for time
100
time
0 Median
Each column has

it’s mean plotted as
-100 a red square
2 7 12 17
Subgroup Number
Number of runs about median: 10.0000 Number of runs up or dow n: 9.0000
Expected number of runs: 9.4706 Expected number of runs: 11.0000
Longest run about median: 4.0000 Longest run up or dow n: 4.0000
Approx P-Value for Clustering: 0.6050 Approx P-Value for Trends: 0.1118
Spread
Approx P-Value for Mixtures: 0.3950 Approx P-Value for Oscillation: 0.8882
If these numbers had a value

< .05, then the process is unstable.
This Run Chart indicates that the process is stable over time.
By using the Run Chart, I’m looking for stability.
How do I analyze this chart?

Look at p-values. P > .05 indicates stability. If the
process is unstable, look for the most recent stable
period and analyze with Minitab. If there is not a most
recent stable time period, investigate & remove special
causes.
Note: To manipulate data in Minitab:
MANIP > Subset worksheet > include/exclude row #’s
D M A I C
Solution–Normality Testing 27
In Minitab, you can perform a Normality Test (Stat-Basic Statistics-Normality Test).

Use the normality test to validate the shape of your data, when the shape is in question.
.999
.99
.95
Probability
.80
.50
.20
.05
.01
.001
-100 0 100
time
Average: -4.40316 Anderson-Darling Normality Test
N: 506 P-Value: 0.983
The data is Normal by this test

method if the P-value is > = .05
Second Step: Check Shape (Normality)
Stat > Basic Statistics > Normality Test (Variable =

Time)
The results of the Anderson-Darling Normality Test tell

you if the data is normal or not. A p-value of greater than
or equal to 0.05 means the data is normal. A p-value of
less than 0.05 means that the data is not normal. Step 6
will address p-values in more detail.
D M A I C
Solution–Normality Testing (continued) 28
Descriptive Statistics
Variable: time This
data is
normal
Anderson-Darling Normality Test
.
A-Squared: 0.130
P-Value: 0.983
Mean -4.4032
StDev 43.1714
Variance 1863.77
Skewness 7.65E-02
Kurtosis -1.2E-01
-100 -60 -20 20 60 100 N 506
Minimum -114.021
1st Quartile -35.085
Median -4.956
3rd Quartile 23.421
-8.174 -0.633
-10 -5 0 95% Confidence Interval for Sigma
40.665 46.009
-9.369 0.532
Third Step: Check Spread
The graphical summary of descriptive statistics (Stat–

Basic Statistics–Descriptive.)
Stat > Basic Statistics > Display Descriptive Statistics
Variable = Time
Click on “Graphs” & choose “Graphical Summary”
Test p-value. Minitab will always display a normal

curve based on the data’s mean and standard deviation
–this may not be a fit to the actual distribution.
D M A I C
Minitab Six Sigma Process Report 29
MINITAB FILE: GB case study.mtw
1. Double
Click C5 time
2. Double
click C1 to
use Date1
as the
Subgroup
size
3. Type in
Lower and
Upper
specs
4. Click OK
Now let’s look at how to calculate process capability for The Six Sigma Process Report is used to calculate the
continuous data in Minitab, using the Six Sigma Process long-term and short-term z-values of your process. In
Report. order to be accurate in the calculation of the z-values, it is
important that we enter all of the information available.
The subgroup size can be entered in two different ways: If a target is available, then enter the target value for the
Size of each subgroup –10 (subgroup size must be calculation of the z-values. If the upper and lower
constant) specification limits are entered and the target is left
blank, Minitab approximates the target value as the
ID column–oper 50 (subgroup size may be constant or midpoint of the specific range. If only one specification
not) limit is entered and the target is left blank, then Minitab
approximates the target as the mean of the data. Hence,
Warning: If you choose to use a subgroup size of 1, the when the target is available, always input the given value
short-term values calculated in the Six Sigma Process to avoid approximations made by Minitab.
Report will be invalid since there is not variation within
the groups.
D M A I C
Report 1: Executive Summary

Process Performance Process Demographics
Actual (LT)
Potential (ST) Date:
Reported by:
Project:
LSL USL
Department:
Process:
Characteristic:
Units:
-100 0 100 Upper Spec: 60

Lower Spec: -60
1,000,000 Nominal:
Actual (LT)
Potential (ST)
Opportunity:
100,000
10,000
Process Benchmarks
1000
Actual (LT) Potential (ST)
100
Sigma 0.97 0.98
(Z.Bench)
10
PPM 166781 164350
1
0 5 10 15
Report 1 contains the essence: Upper graph: show short (dotted) and long-term (full
Sigma potential (ST) = ZBench-ST: if you were to use line) distributions. By definition, the short-term
just one number to indicate the capability of your distribution is shown to be on target. In this case, the
process, it is this one. If a firm XY claims to be a 5σ long-term distribution is slightly below the target.
company, they talk about this value. If you only see Lower graph: shows cumulative PPM’s.
ZST, it means ZBench-ST. It is a convention to benchmark
other companies using this value. However, it is a
Note: When reporting ZST, determine if you will be using
good practice to report out both ZBench values, for the
the shift calculated from your data or the standard 1.5
short and long-term.
shift. If using the 1.5 shift, you will need to manually add
Sigma actual (LT) = ZBench-LT: this value corresponds the 1.5 to ZLT to get ZST. Minitab cannot accommodate a
to the actual performance of your process, and is easier predetermined shift. (When in doubt, use the 1.5 shift
to comprehend with PPM. and/or consult your BB/MBB).
PPM actual = DPMO actual: the actual number of
defects per million opportunities that the process
produces.
PPM potential (not so important): corresponds to
ZBench-ST, not commonly used.
D M A I C
Report 2: Process Capability for time
Xbar and S Chart Capability Indices

40
30 ST LT
1. Between 20 3.0SL=21.49
10
Subgroup 0 Mean 0.0000 -4.4032
X=-4.403
Variation -10
-20 StDev 43.1471 43.1714
-30 -3.0SL=-30.29
-40 Z.USL 1.3906 1.4918
-50
Z.LSL 1.3906 1.2878
Subgroup 0 5 10 15
80 Z.Bench 0.9767 0.9670
70 Z.Shift 0.0098 0.0098
2. Within 60 3.0SL=61.28
Subgroup 50 P.USL 0.082175 0.067876
Variation 40 S=42.70
P.LSL 0.082175 0.098905
30
20 -3.0SL=24.12 P.Total 0.164350 0.166781
10
Yield 83.5650 83.3219
3. Because variation is
not out of control, there is PPM 164350 166781
no LT Variation
Potential (ST) Capability Actual (LT) Capability Cp 0.46
Process Tolerance Process Tolerance Cpk 0.43
-129.508 129.508 -133.981 125.175 Pp 0.46
I I I I I I Ppk 0.43
4. This makes sense, we
only have 1 month’s worth I I I I I I
of data -60 60 -60 60 Data Source:
Specifications Specifications Time Span:
Data Trace:
Report 2 is more detailed: Note: When reporting ZST, determine if you will be using
The table to the right side contains all the details of the the shift calculated from your data or the standard 1.5
shift. If using the 1.5 shift, you will need to manually add
calculation of ZBench for short and long-term.
the 1.5 to ZLT to get ZST. Minitab cannot accommodate a
The graphs to the left side are called control charts. predetermined shift. (When in doubt, use the 1.5 shift
This subject will be covered in the “control” session of and/or consult your BB/MBB).
your training. In order to use Minitab’s calculated shift
(right column in the table), the “S” chart must be in
control (minimize variation within subgroup)
In this example, S chart is out-of-control (one point
outside control limits). It does not meet the condition
mentioned in the 2nd bullet, therefore, report ZLT =
0.9670. To get ZST: ZST = ZLT + ZShift
ZST = 0.9670 + 1.5 = 2.467
Refer to the next 2 pages for rules on using this
Process Capability Report.
D M A I C
Rules For Using The Six Sigma Process Report 32
Rules for Process Capability

1. Ensure the process is stable (Run Chart). If not, remove
special causes first.
2. Ensure normal distribution (if not, use DPMO method to
calculate capability)
3. For continuous data: use Six Sigma Process Report. If
subgroup size > 1 and the S-chart is in-control, then both
columns in the table are valid. If the S-chart is out-of control,
only the LT column is valid. Add 1.5 to the ZBenchLT to get ZST.
4. If the subgroup size = 1, use LT column only in the Process
Report
Use the Decision Flow on the next page to help you!
D M A I C
Process Capability Decision Flow 33
Is your
Process Yes Do you Yes Is the Yes Use Six Sigma> Process Report
Stable (use have Distribution (Input Data, Specification Limits
Run Chart)? Continuous Normal? and Target as appropriate)
Data?
No No No
Do you
Analyze for Special Six Sigma>Product Report, No Use Z Bench-LT only
have
Causes- DPMO, L1 (Input D, U and O) and add 1.5 Shift
Rational
to get Z Bench-ST
Should not •Sigma Conversion Table gives ZST Subgroups?
calculate process
•Z LT = Z ST - Z Shift (1.5)
capability until the
process is stable. •Product Report and L1 give Z Bench Yes
Unstable process •Z Bench = Z ST
data can only be
•Z LT = Z Bench - Z Shift (1.5) Is the S- No Use Z Bench-LT only
used to describe and add 1.5 Shift
Chart in- to get Z Bench-ST
the capability of the
control?
current data set, not
the future capability
of the process. Yes
Note: Z Bench-LT , Z Bench- ST

Z ST = Z LT + Zshift & Z Shift all valid on
Z ST = Z Bench-ST the Process
Z LT = Z Bench-LT Report
D M A I C
The Normal Curve And Capability 34
Poor Design Capability
High
Probability High
of Defective Probability of
Defective
Good Design
Capability
LSL USL
Low Low
Probability Probability
of Defective of Defective
LSL USL
D M A I C
Summary–Z-Value 35
Basic statistical summaries, histograms, dotplots, boxplots, and

run charts are used to visualize data and better understand a
process
The Z-score is a non-dimensional quantity that enables us to
compare different processes–it represents the process
capability
The Z-score is the number of standard deviations that will fit
between the mean and the respective specification limit of a
normal distribution
The Z-score corresponds to yield, or the area under the curve
inside the specification limits
Note on Capability:
Capability can be described as any of the following:
– ZBenchST = ZST = σST = Sigma CapabilityST =
SigmaST = ZBench
– ZBenchLT = ZLT = σLT = Sigma CapabilityLT =
SigmaLT
In the new BOK, σST and σLT are not recommended

ways to describe capability. σST = pooled standard
deviation and σLT = standard deviation.
D M A I C
Summary–Continuous Data Process Capability 36
The Minitab Six Sigma Process Report is used to describe capability

with continuous data
– Displays the actual capability relative to the target distribution
– By rationally subgrouping (subgroup size >1), long-term capability, short-term
capability, and shift are calculated.
Rational Subgrouping refers to grouping the data for analysis in a
meaningful way to understand variation. Rational Subgrouping
attempts to select groups of data such that mainly common cause
variation is within groups, and mainly special cause variation is
between groups.
– Special Cause = Between group variation, due to assignable causes, non-
random influences
– Common Cause = Within group variation, inherent in a process, random
influences
A subgroup that contains only common cause variation, or random
variation, represents the short-term capability of the process or
entitlement (ZBench)
D M A I C
Process Capability & Yield–Discrete Data 37
Calculate the distribution of defects, defects per million

opportunity (DPMO)
Utilize Z tables to convert DPMO to yield or sigma level
Understand the difference between classical, Throughput Yield,
and Rolled Throughput Yield
Calculate submitted, observed, and escaping defect levels
Classical Yield (YC) = Final Yield
Throughput Yield (YTP) = First Pass Yield
Rolled Throughput Yield (YRT)
D M A I C
Definitions 38
Unit (U)
The number of parts, sub-assemblies, assemblies, or systems
inspected or tested
– Squares: 4 units
Opportunity (OP)
A characteristic you inspect or test
– Circles: 5 opportunities per unit
Defect (D)
Anything that results in customer dissatisfaction. Anything that
results in a non-conformance.
– Black circles: 9 defects
Note: Black dots indicate a defect.
Clarification
Defect–Any single non-conformance (defect=black dot)
Defective–A unit with 1 or more defects. (defective=any

square with 1 or more black dots).
D M A I C
Formulas 39
Defects per Unit

DPU = D/U
9/4 = 2.25
Total Opportunities
TOP = U*OP
4*5 = 20
Defects per Opportunity (Probability of a Defect)

DPO = D/TOP
9/20 = .45
Defects per Million Opportunities

DPMO = DPO*1,000,000
.45*1,000,000 = 450,000
Once we calculate DPMO, we go to the Abridged Sigma

Table (next page) to get ZBenchST or ZBenchLT
D M A I C
ConvertingDPMO
Converting DPMOTo
ToZZST 40
Long term Actual Reported
ZST = Reported D PMO

500,000
Sigm a (long term )
0
Sigm a (s hort term )
1.5
460,172 0.1 1.6 ZST = 1.6 for 450,000 DPMO
Sigma (Short-Term) 420,740 0.2 1.7
382,089 0.3 1.8
344,578 0.4 1.9
Abridged Process 308,538
274,253
0.5
0.6
2
2.1
Sigma Conversion 241,964

211,855
0.7
0.8
2.2
2.3
184,060 0.9 2.4
Table 158,655 1 2.5
135,666 1.1 2.6
115,070 1.2 2.7
96,801 1.3 2.8
80,757 1.4 2.9 ZZST DPMO
DPMO
66,807 1.5 3 ST
54,799 1.6 3.1
44,565 1.7 3.2
35,930 1.8 3.3 2 308,538
28,716 1.9 3.4
22,750 2 3.5 3 66,807
17,864 2.1 3.6
13,903 2.2 3.7 4 6,210
10,724 2.3 3.8
8,198 2.4 3.9 5 233
6,210 2.5 4
4,661 2.6 4.1 6 3.4
3,467 2.7 4.2
2,555 2.8 4.3
1,866 2.9 4.4
1,350 3 4.5
968 3.1 4.6
687 3.2 4.7
483 3.3 4.8
337 3.4 4.9
233 3.5 5
159 3.6 5.1
108 3.7 5.2
72 3.8 5.3
48 3.9 5.4
32 4 5.5
21 4.1 5.6
13 4.2 5.7
9 4.3 5.8
5 4.4 5.9
3.4 4.5 6
To Calculate Process Sigma:

Go to the Memory Jogger–Abridged Sigma
Conversion Table (or use the table above)
Look up the DPMO value and report Short Term
Sigma value (always round to poorer performance if
between values in the table.)
In our example, we calculated 450,000 DPMO. This

value is not found in the table, therefore, we go to
460,000 and report 1.6 as the Short Term Sigma Value.
We’re always conservative. Therefore, report the lower
value for ZST or ZLT. And, you cannot interpolate, so
choose the lower sigma value.
D M A I C
DPMO Calculation 41
Example
Errors Detected
28 – Wrong amount
14 – Wrong address
12 – Improper accounting code
30 Invoices
Mailed Late
Prepare Review Fix Mail

Invoice Invoice Errors Invoice
500
Preliminary 470 Invoices
Invoices Mailed On-Time
446 Accurate Invoices
Customer CTQs
Invoice mailed on date specified
Invoice is error free
– Correct address
– Correct amount
How would the customer measure the process

performance?
First, determine the unit, defect opportunities per unit,

and the number of defects from each perspective.
For this example, assume the process of reviewing and

correcting invoices is 100% effective.
Use the worksheets on the following pages.
D M A I C
42
Calculating Process Sigma Discrete Data Method
1. Number of Units processed 500

U = __________
2. Number of Defect Opportunities Per Unit OP = __________

3
3. Total number of Defects made 72

D = __________
(include defects made and later fixed)
4. Solve for Defects Per Opportunity

D ( 72 )
DPO = U∗OP = ( 500 ) ( 3 ) = 0.048
5. Convert DPO to DPMO
DPMO = DPO ∗ 1,000,000 = 0.048 ∗ 1,000,000 = 48,000
6. Look up Process Sigma in

Abridged Process Sigma Conversion Table ZST = 3.1
The unit in this case is an invoice. Hence, there are 500

Now calculate the DPMO and determine the
units for this example.
capability.
Based on the CTQ’s listed for this example, there are 3

defect opportunities per unit–late, wrong amount, and
wrong address.
Therefore, the total number of defects is 30 late, 28

wrong amount, and 14 wrong address for a total of 72.
Even though the errors on the accounting code were

detected and corrected, they are not counted for this
process. The CTQ’s stated here are for our external
customers. They do not care if our accounting codes are
correct; therefore, this “error” is not one of their CTQ’s.
If we were to examine this from the perspective of our
internal customer, the list of CTQ’s would be different.
.
D M A I C
43
Exercise–Calculating Process Sigma Discrete Data
1. Number of Units processed U = __________

500
2. Number of Defect Opportunities Per Unit 3

OP = __________
3. Total number of Defects made 30

D = __________
4. Solve for Defects Per Opportunity
D ( )
DPO = = =
U∗OP ( )( )
5. Convert DPO to DPMO

DPMO = DPO ∗ 1,000,000 = ( ) ∗ 1,000,000 =
6. Look up Process Sigma in

Abridged Process Sigma Conversion Table ZST =
Let’s verify our answer using Minitab’s Product Report.
1. First open a blank worksheet and enter 3 column

headings and your data:
Defects Units Opportunities

30 500 3
2. Six Sigma > Product Report

Defects = Defects
Units = Units
Opportunities = Opportunities
3. Click “OK”
4. Verify ZBench = ZST (from your calculations)
D M A I C
44
Activity–Calculating Process Capability–Discrete Data
Project: Billing Product Improvement

CTQ: Complete Billing Information
Defect: Information missing from bill
Unit: A bill
Opportunities for defect per unit: 3 (items of information that
can be missing on each bill)
Out of 50 bills, you find 20 defects
What is the sigma level of the process?
What is the DPMO value for this process?
Now let’s look at how to calculate process capability for

discrete data in Minitab, using the Six Sigma Product
Report.
D M A I C
45
Six Sigma Product Report
D M A I C
46
Discrete Data Examples
1. Select Defects, Units 2, & Opportunities
2. Select OK.
D M A I C
47
Six Sigma Product Report Output
Report 7: Product Performance
Characteristic Defs Units Opps TotOpps DPU DPO PPM ZShift ZBench
U DPMO ZST
D
1 89 506 1 506 0.176 0.175889 175889 1.500 2.431
OP
Total 89 506 0.175889 175889 1.500 2.431
In addition to the Product Performance table, Minitab

provides two graphical charts for discrete data. The
product benchmarks allows one to see visually how the
ZBench values compare for each characteristic (8A).
To verify our answer:

D 89
DPO = = = 0.175889
U ∗ OP (506 )(1)
DPMO = DPO ∗ 1,000,000 = 175,889
Go to Abridged Process Sigma Conversion Table,

Z = 2.4
ST
D M A I C
48
Yield
Four parts are made

1. After first inspection: 1 passed, 3 failed
Rework 3 parts
2. After second inspection: 1 passed, 2 failed

Rework 2 parts
3. After third inspection: 1 passed, 1 scrapped
What is the Yield of this Process?
D M A I C
49
Types Of Yield
Classical Yield = YC = 3/4 = 75%

Classical yield is the percent of defect-free parts for the whole process divided
by the total number of parts inspected. If we say the yield is 3/4 or 75%, we lose
valuable data on the true performance of the process. This loss of insight
becomes a barrier to process improvement.
First-Time Yield = YFT = 1/4 = 25%

First-time yield is the percent of defect-free parts divided by the total number of
parts inspected for the first time. If we say the yield is 1/4 or 25%, we are really
talking about the First-Time Yield (FTY). This is the best yield estimate to drive
improvement. This is also Throughput Yield (YTP): the percent of units that
pass through an operation without any defects.
Classical Yield:
YC = ¾
# parts making it through final inspections = 3
Total # Parts = 4
First-Time Yield
YFT = ¼
# Parts that are acceptable after first inspection = 1
Total # Parts = 4
Probability of Single Unit Being Defect-Free (SDF)

SDF = P(0) = e-DPU = e-2.25 = .1054 = 10.54%
DPU = Defects per unit = 9/4 = 2.25
D M A I C
Extending The Concept 50
A given process has two operations. Each operation has a

throughput yield of 99 %. The rolled yield equals:
Process Centered Process Centered
Op 1 x Op 2 = Output
99% 99% 98%
Without Inspection or Test Without Inspection or Test Without Inspection or Test
. . . There is an 98% probability that any given unit of product could

pass through both operations defect free.
Rolled Throughput Yield = YRT = 98%
When the defects of the subsequent operations are

independent, rolled throughput yield (YRT) = .99 x .99 =
.9808
D M A I C
Rolled Throughput Yield (YRT) 51
Receive parts from Supplier
99.53% Yield (YTP)

Following Receiving
Inspection and Line Fall-out...
97.13% Yield (YTP)

From Machining Operations
94.52% Yield (YTP)

At Test Stands on
first attempt
Right
First
YRT = .9953*.9713 *.9452 = 91.38% Time
G. Reimer 12/21/94 - Charlotte NC
You can establish a yield for each step of your process. If

you have multiple processes, each time you go through
additional steps, you lose yield at the end of the process.
This is called Rolled Throughput Yield. YRT is calculated

by multiplying all the yields together.
The more complex the process, the worse the YRT

becomes.
We waste a lot of time scooping up the spilled paint.
D M A I C
Quantification Of Defects 52
Knowing the observed defect level and the test effectiveness, we can estimate
submitted defects and escaping defects.
For example:
Inspection
E = 0.8
Submitted Escaping
DPU Level DPU Level
(DPUS ) (DPUE )
DPUS = DPUO + DPUE

Observed DPU Level
DPUO = DPUS x E (DPUO )
DPUE = DPUS x (1-E)
Let n denote the number of defects going into an Example

inspection operation.
If we have a process that is submitting 100 defects (DPUS
Let E denote the proportion likelihood of detecting a =100) and our inspection process is 80% effective (E=.8)
defect. at finding these defects, then we could observe 80
(DPUO) and 20 would escape to the customer (DPUE).
Inspection catches E*n.
The number escaping to the customer is (1-E)*n.
If you submit parts with a DPUS level, inspection E = 0.8

observes (catches) a DPUO level of E*DPUS = Observed DPUS=100 DPUE=20
DPU.
The customer sees a product with an escaping DPUE =

(1-E)*DPUS.
Submitted DPUS = Observed DPUO + Escaping DPUE. DPUO=80

D M A I C
Quantification Of Defects (continued) 53
Given:
Observed Defects = DPU of 0.25 and E = .8
Then:
Submitted Defects = DPUS = DPUO /E = (0.25/0.8)
= DPUS = 0.31
Escaping Defects = DPUE = DPUS - DPUO = (0.31 - 0.25)
= DPUE = 0.06
Inspection
E = 0.8
Submitted Escaping
DPUS Level DPUE Level
= 0.31 (calc) = 0.06 (calc)
Observed DPUO Level = 0.25 (given)
The only thing that is real is the DPU observed–that is

what we caught at inspection. In this example DPUO =
.25.
A typical value for E = .9. Your operation will differ

slightly; in this example E = .8.
Review the calculations below:

DPUS*.8 = DPUO
DPUS = DPUO /.8
DPUE = DPUS- DPUO
D M A I C
Exercise (15 minutes) 54
Calculating Submitted, Observed and

Escaping Defect Levels
Given:
The requirement is to ship the product with no more than one
defect in 500 units shipped
The final test is to be performed using automatic test
equipment having an effectiveness of 0.95
Knowing the maximum level of escaping defects, we can estimate
the maximum defect/unit level:
Observed in the final test. (This sets a minimum throughput
yield limit.)
Submitted to the final test. (This sets a minimum on the
combination of defects created in model assembly and from the
escaping defect levels from prior tests.)
Use the worksheet on the following page to help you

work through this exercise.
D M A I C
Exercise (continued) 55
Calculating Submitted, Observed and Escaping Defect Levels
Inspection
E = 0.95
Submitted Escaping
DPUS Level DPUE Level
Observed DPUO Level

1. Expressed as a decimal, what is the given acceptable maximum escaping defect
level? ___________
2. What then is the acceptable maximum defect level for:

a. Units submitted to final test? ___________
b. Defects observed at final test? __________
3. What is the observed throughput yield figure? ______
D M A I C
Answers To Exercise 56
Calculating Submitted, Observed, and

Escaping Defect Levels
1. DPUE = 1 = 0.002
500
2. Maximum acceptable submitted defects =
DPUS = defects escaping = DPUE = 0.002 = 0.040 DPUS

1 - effectiveness 1 -E 0.05
2b. Maximum acceptable observed defects =
DPUO = DPUS - DPUE = 0.040 - 0.002 = 0.038 DPUO
3. YTP = 1-.038 = .962 = 96.2%
D M A I C
Summary–Discrete Data Process Capability 57
Define defects, units and opportunities with your team. Be sure

the definitions make sense and are consistent with similar
processes and customer definitions.
Defects will be stated as defects per million opportunities.
Discrete data is generally considered long-term data.
For discrete data, Minitab Six Sigma product report is used to
calculate capability from defects and opportunities
Determine DPMO (which is long-term), then determine the
corresponding Z–value (ST capability)
D M A I C
Summary–Z-Shift 58
Often, you must assume a shift value (default 1.5) to estimate

short-term capability
Our customers experience the long-term capability of the
process
To minimize shift, we need to reduce special cause variation
D M A I C
Analyze 5–Define Performance Objectives 1
What does it mean to Define Performance Objectives?

A performance objective is a statement of your project Y’s performance level that will
satisfy the project CTQ(s). It is the projected reduction in defects you plan to achieve
for your process or product. Typically, this is stated in terms of defects per million
opportunities (DPMO) reduction and a corresponding target Z-value. In Step 4 you
determined the current process performance. In Step 5 you will state what the end
results of the Six Sigma project will be by statistically defining the goal of the project.
In addition, an estimate of financial benefits is due in Analyze.
Why is it important to Define Performance Objectives?

It is important to identify your improvement goals in measurable terms in order to
define the level of improvement you wish to achieve and provide a focused target
toward which you can direct your efforts.

5.1 Identify Performance Objectives.
ANALYZE STEP
OVERVIEW

Establish Define Identify Variation
Process Capability Performance Sources
Objectives
5.1 Identify Performance Objectives.
DMAIC GB P TX PG V 4.2.0
D M A I C
How Do I Define My Performance Objectives? 2
Use benchmarking
Use other sources
D M A I C
Defining Performance Objectives With Benchmarking 3
If I benchmark, performance standards are based upon:

Closing the gap with the competition
Exceeded projected competitive performance
Similar performance in dissimilar businesses
Gathering best practices from multiple sources to become best
in class
Becoming as good or better than a substitute product/service
D M A I C
Defining Performance Objectives With Other Sources 4
If I don’t benchmark, performance objectives are based upon:

For a process with ZST ≤ 3 Sigma Capability, decrease %
defects by 10x and for ZST > 3, decrease % defects
by 2x
If your process is in statistical control (Run Chart), the next
improved performance objective comes from a capability
investment as in facilities, equipment, digitization, etc.
Corporate mandate
Compliance/legal
VOC data
D M A I C
Nature Of Benchmarking–Reference Material 5
Benchmarking is the process of continually searching for the best

methods, practices and processes, and either adopting or adapting
their good features and implementing them to become the “best of
the best”.
The remaining pages are for your reference on

Benchmarking.
There are different types of Benchmarking. These

include:
Competitive Benchmarking
Product Benchmarking
Process Benchmarking
Best Practices Benchmarking
D M A I C
What It Is And What It Isn’t 6
Benchmarking Is… Benchmarking Isn’t…

A continuous process A one-time event
A process of investigation that A process of investigation
provides valuable information that provides simple
A process of learning from answers
others; a pragmatic search for Copying, imitating
ideas Quick and easy
A time-consuming, labor-
intensive process requiring
discipline
A viable tool that provides
useful information for
improving virtually any
business process
D M A I C
Common Benchmarking Mistakes 7
1. Internal process(es) is unexamined

2. Site visits “feel good,” but don’t elicit data or ideas
3. Questions and goals are vague
4. The effort is too broad or has too many parameters
5. The focus is not on processes
6. The team is not committed to the effort
7. Homework and/or advanced research isn’t assigned
8. The wrong benchmarking partner is selected
9. The effort fails to look outside the industry (outside the box)
10. No follow-up action is taken
D M A I C
Benchmarking Six-Step Process 8
Identify Process to Benchmark

Select process and define defect and opportunities
Measure current process capability and establish goal
Understand detailed process that needs improvement
Select Organization to Benchmark

Outline industries/functions which perform your process
Formulate list of world class performers
Contact the organization and network through to key contact
Prepare for the Visit

Research the organization and ground yourself in their processes
Develop a detailed questionnaire to obtain desired information
Set up logistics and send preliminary documents to organization
Visit the Organization

Feel comfortable with and confident about your homework
Foster the right atmosphere to maximize results
Conclude in thanking organization and ensure follow-up if necessary
Debrief & Develop an Action Plan

Review team observations and compile report of visit
Compile list of best practices and match to improvement needs
Structure action items, identify owners and move into Improve phase
Retain and Communicate

Report out to business management and 6σ leaders
Post findings and/or visit report on local server/6σ bulletin board
Enter information on GE Intranet benchmarking project database
D M A I C
Sources Of Information 9
Library Database Company Watches

Internal Reviews Newspapers
Internal Publications Advertisements
Professional Associations Newsletters
Industry Publications Original Research
Special Industry Reports Customer Feedback
Functional Trade Publications Supplier Feedback
Seminars Telephone Surveys
Industry Data Firms Inquiry Service
Industry Experts Networks
University Sources Worldwide Web
D M A I C
Capital Logistics Case Study 10
Project Goal(s)
Reduction of deliveries that are too early or too late
Benchmarking Process
Identified 20 similar 3rd party Logistics providers from Logistician Magazine
Selected 3PL (2) that provide supply chain solutions to the petroleum industry
(Non-competitors–Capital Logistics strategically avoids supporting petroleum
industry because of risks)
Polled logistics managers (telephone polls) to determine delivery practices
Results
Identified one viable new delivery method…The Dynamic Routing System
– Computer system that selects routes based on individual driver performance for
load types and regional deliveries
Invited Petro-hauz to Stamford, CT for product demo with Capital Logistics
Engineering and Technology Teams
D M A I C
Benchmarking Example 11
For the Capital Logistics Case Study

– Target is the same = 0 minutes
– Best in Class performance is a s = 12 minutes
– Theoretical calculation of Sigma based on normal distribution would be
Report 7: Product Performance
Characteristic Defs Units Opps TotOpps DPU DPO PPM ZShift ZBench
1 60 25000 1 25000 0.002 0.002400 2400 1.500 4.320
D M A I C
Summary–Analyze 5 12
A performance objective is determined by using Z–short-term,

benchmarking, or defect reduction goals
Benchmarking is a process of identifying best practices,
measuring our own practices against those best practices, and
adapting the appropriate best practices to our own processes
Revenue & cost implications are also due for benefit analysis
D M A I C
Analyze 6–Identify Variation Sources 1
What does it mean to Identify Variation Sources?

In step 6 you develop a list of statistically significant X’s, chosen based on analysis of
historical data. This list is then prioritized to identify those X’s that have the most impact
on the project Y. The question in this step is “What are the variables that are preventing
us from reaching our goal?” You will identify all possible X’s before selecting the Critical
(or Vital Few) X’s in the next step.
Why is it important to Identify Variation Sources?

The output of a process (Y) is a function of the input sources of variation (X’s). In other
words, you can change the output of a process (Y) only by changing the input & process
variables (X’s). Therefore, in order to improve products and processes, you must shift
your focus from monitoring the outputs of a process (Y’s) to optimizing the inputs to the
process and correcting the root causes of defects (X’s). You should use data and
process analysis to identify potential X’s, and not make any assumptions.

6.1 Identify possible causes of variation
6.2 Narrow list of potential causes
ANALYZE STEP
OVERVIEW

Establish Define Identify Variation
Process Capability Performance Sources
Objectives
6.1 Identify possible causes of variation

6.2 Narrow list of potential causes
DMAIC GB Q TX PG V 4.2.0
D M A I C
Identify The Vital Few 2
Transfer Function
Y = f (X)
Project Y Relationship Process

that explains Y variables
in terms of X
Understanding The ƒ Gives Insight

Into The Vital Few X’s
The Project Y (or dependent variable) is determined by

the values of the process variables X (or independent
variables). Once we have a clearer understanding of the
relationship between the X and Y (f–transfer function),
we will be able to pinpoint the key X’s that drive the
variation in Y.
D M A I C
Identify The Vital Few 3
Process
Input
Measures Outputs
(X’s) (Y’s)
X X X X
Process Measures
(X’s)
Definitions: Process
Project Y (Dependent Variable) Provides context for improvement
Product or service produced or delivered by the Links outputs (Y’s) to inputs (X’s) and
process process variables (X’s)
Process X’s (Independent Variable)
Those variables that influence the output and are In Analyze 6 we will be analyzing upstream
generally controllable by those who operate the variables, or input and process variables (X’s), to
process determine how they affect output variables (Y’s)
and to what extent.
Input X’s (Independent Variable)
Materials and information used by the process to create Remember Y=f (X1, X2, ..., Xn)
the outputs. Inputs are often outside the control of the
process operator.
or in words….
The results we get (the Y’s) are a function of the
process and input variables (the X’s). In Analyze,
we will study the process and the data in order to
gain understanding of how it all fits together.
D M A I C
Analyze 6.1: Identify Possible Causes of Variation
D M A I C
Methods To Identify Possible Sources Of Variation 5
Methods To Identify Vital X’s

Process Map Analysis
Graphical
Analysis
Problem
Statement
As we work our way down into the process seeking the

vital few causes, we analyze both the data and the
process–gaining knowledge of what impacts the
output. How do the various X’s impact the Y’s? With
this knowledge, we will pursue a solution that will
improve the process.
We have already learned about graphical analysis,

process map analysis and cause and effect analysis in
previous modules. In step 6, we will learn how to
apply them in the Analyze phase.
D M A I C
Review: Graphical Analysis 6
Looking For Patterns In Data
Continuous Y Discrete Y
Boxplot Pareto Chart
Scatterplot
Histogram
We learned about these tools in the Minitab and

Graphical Analysis module. These graphical analysis
tools are used in the Analyze phase to look for patterns in
the project Y data that help to identify potential critical
X’s.
Let’s do an activity to briefly review these tools.
D M A I C
Review Activity–Which Graphical Analysis Tool Would You Use? 7
For each scenario described below, which tool would you use to investigate? For each
scenario, what pattern would you look for to make your conclusions?
1. (a) A Six Sigma project is being conducted in the field to improve the cycle time for
warranty repair returns. The warranty return cycle time was measured for a period of 6
weeks for 4 regions. The Green Belt wants to investigate whether there is any
difference in average warranty repair cycle time between each of the regions.
(b) The Green Belt would also like to know if there is any difference in the variance of
the cycle time between each of the regions.
2. Checks Are Us is a payroll processing firm. Timecard errors are routinely monitored
and recorded. A Black Belt investigating the errors wishes to determine if there are
any differences in the number of errors between five of its major customers. The
number of errors contained in a sample of 150 employees was recorded for five
weeks. How would you proceed with stratifying the data to make this investigation?
3. Tungsten steel erosion shields are fitted to the low pressure blading in steam turbines.
The most important feature of a shield is its resistance to wear. Resistance to wear
can be measured by abrasion loss, which is thought to be associated with the
hardness of steel. How would you investigate the relationship between hardness (the
X) and abrasion loss (the Y)?
D M A I C
Graphical Analysis–Breakout Activity (15 minutes) 8
Desired Outcome: Identify possible causes of variation in the Capital

Logistics case study data. File: Green Belt Case Study.mtw
What How Who Timing
Minitab Use the appropriate graphical All 15 mins.

Analysis analysis tool in Minitab to
investigate
– Differences in delivery cycle time
between the two different load
types (See note below)
– Differences in delivery time
between drivers
– Relationship between delivery
cycle time and distance
Report Be prepared to share your results All
Out with the class
Load types: D/H Drop & Hook–A driver unhooks their full trailer, and then hooks into an empty to take
back. Unload–The driver opens the trailer and allows the store to unload its freight, and brings back
the trailer.
To create Boxplots:
Stat > Basic Statistics: > Display Descriptive Statistics
variable: time
click: by variable: load type or driver
click Graph > Choose Boxplots
To create Scatter Plots:

Graph > Plot
X: Distance
Y: Time
D M A I C
Process Map Analysis 9
Types Of Analysis
Moments of truth–What does the customer feel?
Nature of work–What is of value to the customer?
Flow of work–What makes the customer wait?
The purpose of Process Map Analysis is to: The second, Nature of Work, is concerned with value
View our process and its performance from the analysis: understanding whether or not the work in your
customer’s perspective process is valued by the customer or by your internal
culture.
Identify current process problems and opportunities
Create a shared sense of urgency for improving the The third, Flow of Work, characterizes the work along
the dimension of time. The two latter components are
process clearly identify and document value-
focused on the “Profitability” piece of the GE Capital
added/nonvalue-added activities and process flow
Quality Vision.
measures and characteristics.
There are three key components to the Process Map

Analysis step:
The first, Moments of Truth, ensures that we are

“Completely Satisfying Customer Needs” through a
customer-focused process improvement perspective.
D M A I C
Moments Of Truth 10
A Typical Deployment Flow Chart
Initial Contact Service/Produce Delivery Follow-Up
Customer Moments
of Truth
Intermediaries
Or Front Line
Backroom
Activity
Support Or
Suppliers
A moment of truth is defined as anytime a customer “If you’re a service person and you get it wrong at your
draws a critical judgment, positive or negative, about the point in the customer’s chain of experience, you are very
service, based upon a service experience (or lack likely erasing from the customer’s mind all the memories
thereof). Certainly every time we “touch” the customer is of the good treatment he or she may have had up until
a moment of truth, but it isn’t limited to these cases. now.
But if you get it right, you have a chance to undo all the
From the customer’s view, the process is a set of service wrongs that may have happened before the customer got
encounters. In this case, the customers believe the steps to you. You really are the moment of truth.”
should be simple: apply for the loan and receive the
money.
In the best of cases, the customer is unaware of all the

hard work, delays, and hand-offs which go into making
their loan a reality. It occurs behind the scenes.
D M A I C
Nature Of Work–Value Analysis 11
Value-Added Work Nonvalue-Added Work
Steps That Are Steps That Are Considered

Essential Because They Non-Essential To Produce
Physically Change The And Deliver The Product Or
Product/Service, The Service To Meet The
Customer Is Willing To Customer’s Needs And
Pay For Them, And Requirements. The
They Are Done Right Customer Is Not Willing
The First Time. To Pay For Them.
Steps That Are Not

Essential To The
Customer, But That
Allow The Value-
Adding Tasks To Be
Done Better/Faster.
Value-Enabling Work
Your policies address only control issues

There are more managers than workers
People get rewarded for managing problems (e.g.,
Manager of Reissues, Special Expediting)
Rework is a common characteristic of most processes
– RE’s are things in your process that are done more
than one time (Redo, Recall, Reissue).
– RE’s usually cause you to loop back to an earlier
point in your process.
– RE’s consume time, add to the complexity, and use
additional resources.
OPTIONAL: Show the video: “Time the dimension of
Quality”
D M A I C
Types Of Nonvalue–Added Work 12
Internal Failure Delay
External Failure Preparation/Set-Up
Control/Inspection Move
What Does The Customer Value?
Internal Failure Delay

Steps that are related to correcting in-process errors due to Steps where work is waiting to be processed at that step.
failures in current or prior step in the process. Examples:
Example: Backlogs
Rework Queues
External Failure Storage
Steps which relate to fixing errors in the product that the
customer has found and has returned to you. Bottlenecks
Examples: Preparation/Set-Up
Steps that prepare work for a subsequent activity in the process.
Customer Follow-up
Examples:
Recall
Entering into a computer
Control/Inspection
Steps for internal process review often referred to as “the Retrieve policies/pricing
checker-checking-the checker.” Move
Examples: Steps that entail the physical transport or transmit of outputs
Approval/review between activities in a process.
Example:
Inspection
Fax/mail
Bureaucracy
D M A I C
Flow Of Work 13
Process Time
+ Delay Time
Cycle Time
To analyze the flow of work, break the process flow Terms

down to its lowest component, and analyze the movement
of the component through the process. For example, Cycle Time
The total time from the point in which a customer
“become an application” and trace its progress through requests a good or service until the good or service is
the process. It is similar to a process flowchart, but with delivered to the customer. Cycle time includes process
the value-added and nonvalue-added steps clearly time and delay time.
identified.
Process Time
The total time that a unit of work is having something
done to it other than time due to delays or waiting. It
includes the time taken for value-added steps, internal
failure, external failure, control, inspection,
preparation/set-up, and more.
Delay Time
Total time in which a unit of work is waiting to have
something done to it.
D M A I C
Flow Of Work–Process Disconnects 14
Gaps
Redundancies
Implicit or unclear requirements
Inefficient hand-offs
Conflicting objectives
Common problem areas
Process disconnects can often be found where there are Inefficient hand-offs: no check is in place to assure the
delays in the process. These are points in the process process is continuing without delays. For example,
where work can be disrupted or delayed, or where defects Department A sends something to Department B but
neither has a way to know if it was properly received.
can be created. They include:
Conflicting objectives: the goals of one group cause
Gaps: responsibility for a given step in the process is problems or errors for another. For example, one group
unclear, or the process seems to go off track. is focused on process speed while another is oriented to
Redundancies: duplication of efforts such as when two error reduction–the result may be that neither group
people or groups approve a document. Redundancies accomplishes its objectives.
occur when different groups take action that they are Common problem areas: occurs when steps are
unaware is being done somewhere else in the process. repeated in a variety of places in process. Noting these
areas may provide insight into potential solutions.
Implicit or unclear requirements: operational
definitions do not exist or clear differences exist in
perspective or interpretation. Your team should encode these “disconnects” and
highlight them directly on your process map.
D M A I C
Flow Of Work–”Be The Unit” 15
Unclear
requirements
1. Receive
Are we Yes
application in 11. Make loan extending
mail and open decision
Redundancy loan?
envelope
No 12. Generate
2. Place 10. Review for
Inefficient loan packet
application in completeness
hand-off 19. Generate
mail slot and make
turndown letter
decision
3. Move 9. Queue 13. Place in 18. Postman

application to application for out-box picks up
Entry Dept. credit review outbound mail
Unclear
4. Place requirements Inefficient 14. Move to 17. Place in
8. Score hand-off
application in mailroom outbound mail
application
in-box basket
5. Retrieve 7. Enter
Yes 15. Wait for 16. Post package
application and Is application application
postage or letter
review for complete? to computer
completeness system
No
6. Call to obtain
necessary
information
Cycle time is the total time taken from the point at which Constructing a time line helps identify areas of potential
the customer requests a good or service until the good or improvement by illustrating the time required for each
service is delivered to the customer. It is the sum of the step in a process.
process time, move time, inspect time, delay time, and
storage time. Remember:
Time data may not be available for all steps in a process.
Cycle time analysis often provides eye-opening insights. Some possible sources include:
People seldom think of work in this context. When an Time, schedule, and activity logs
Improvement Team presents cycle time data, Files, studies, and reports
management may become defensive. For example, when
one manager at a commercial insurer found out it was If data is not available:
taking 100 days, on average, to process the commercial
line of a new business, she resisted the analysis. After
Estimate and ten track to determine actual average
heated “discussion,” she suggested that the period
value
sampled did not reflect normal conditions. She
recommended looking at another period of time to get a
“better number.” The team obliged. The next sample Use minimum of 30 data points to determine an average
averaged 103 days! value when calculating averages, also record the highest
and lowest time so consistency is understood.
D M A I C
Linking Value Analysis With Process Flow 16
Summarized Analysis
% %
Process Step 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 Total Total Steps
Est. Avg. Time (Mins) 1 120 15 120 3 180 7 1 120 5 10 15 90 15 120 2 120 5 8 957 100%
Value-Added 9 9 99 9 99 48 5.0%
Nonvalue-Added
Internal Failure 9 180 18.8%
External Failure
Control/Inspection 9 9 8 .8%
Delay 9 9 9 9 9 9 690 72.1%
Prep/Set-Up
Move 9 9 30 3.1%
Value-Enabling 9 1 .1%
Total 957 100%
Linking value analysis with cycle time analysis creates a Conclusion: Example
compelling business case for change. 7 steps (-40% of total steps) provide 100% value and
take 48 minutes or 7% of total cycle time.
Displaying this information on process maps emphasizes 11 steps (-61% of steps) are non-value-added, and
focus areas for improvement. consume 94.9% of cycle time.
Tips/Traps
Don’t get stuck on deciding if a step is
value-added or nonvalue-added
– Allow the team to discuss it.
– Consensus with 2/3s majority vote.
Push the team to make the process map a true “as is.”
– The ratio for an average “as is” map is 20% value-
added, 80% nonvalue-added.
– Think of the process map from the point of view of
the product or service.
D M A I C
Review: Cause & Effect Diagrams 17
A visual tool used by an improvement team to brainstorm and

logically organize possible causes for a specific problem or effect.
Potential High-Level Causes

Problem
Statement
We learned about this tool in the Measure phase. In the

Analyze phase the Cause and Effect diagram is often
used to identify X’s.
D M A I C
Cause & Effect Diagrams–The Five Why’s 18
The “Five Why’s” drill deep into the process to identify

potential Root Cause(s)
Ask “why” five times to identify deeper causes
Use process data to answer each “why” question
Example 4. Why is the format confusing?

A project team has verified that the X (complicated form)
Because the directions are hard to read
accounts for the difference in cycle times between small
and medium loans. They use the Five Why’s to drill
5. Why are the directions hard to read?
deeper in the process.
Because the font size is too small
1. Why do complicated forms cause delays in the
underwriting steps?
Because underwriters receive incomplete
applications
2. Why do they receive incomplete applications?

Because customers don’t fill out the form
accurately or completely
3. Why don’t customers fill out the form accurately or

completely?
Because the format is confusing
D M A I C
Cause And Effect–A Breakout Activity (25 minutes) 19
Desired Outcome: Identify possible causes of variation in the Capital Logistics

case study data
What How Who Timing
Preparation Select a facilitator, scribe and timekeeper All 5 mins.
Construct Cause As a result of the QFD constructed in All 15 mins.

And Effect Measure 1 for the Capital Logistics
Diagram customer service desk, a project team
elects a project Y of cycle time to
resolve customer problems
In the Analyze phase of the project, the
team uses a cause and effect diagram
to brainstorm potential X’s
Construct a cause and effect diagram
for the problem statement “Why is it
taking too long to resolve customer
problems?”
Base cause and effect diagram on your
general knowledge of customer service
desks. What might be potential X’s?
Five Use the ask “why” approach to identify All

Why’s deeper causes as appropriate
Close Choose a spokesperson to share the All 5 mins.

results with the class
D M A I C
Prioritization Of X’s–Control/Impact Matrix 20
IMPACT
High Medium Low
C In Our
O Control
N
T
R
O
L
Out Of
Our Control
Always Verify With Data
Use your team’s process knowledge and business

experience to list possible X’s in a Control/Impact Matrix.
Then use process data to verify or disprove placement of
the X’s.
Prioritization Steps
1. Using the Control/Impact Matrix shown above, examine

each X in light of two questions:
What is the impact of this X on our process?
Is this X in our team’s control or out of our team’s
control?
2. With your team, place each X in the appropriate box on

the matrix.
3. The validated matrix is a guide to addressing the X’s.

Begin with the “High Impact/In Our Control” category.
D M A I C
Prioritization Of X’s–Control/Impact Matrix (continued) 21
Example
Why Is There
Difference In The
Variation In Cycle
Time Between Small
And Medium Loans?
Measure- Mother People
ments Nature
IMPACT
High Medium Low
Too many
defects
Complicated Too long for Complexity
C In Our form customer Evaluation
Control Too much number of risk
O review worthiness
N Duplication
of effort
T
R
Too long to Not enough
O Out Of get credit staff
report
Our Control Not well
L trained
The project team completed the Cause & Effect diagram

and the prioritization matrix. The X’s that are identified
as “in control/high impact” appear to be the top priority.
However, the team should discuss the other X’s to make
sure they have reached consensus for those priority X’s.
D M A I C
Control/Impact Matrix–Breakout Activity (15 minutes) 22
Desired Outcome: Practice using a Control/Impact Matrix to prioritize X’s
What How Who Timing
Preparation Choose a facilitator, scribe, All 2 mins.

timekeeper and/or note taker
Construct Review the potential X’s generated All

Cause for the customer service desk project
And in the cause and effect diagram
Effect activity
Diagram
Prioritize Use the Control/Impact Matrix on All 10 mins.

Potential X’s the next page to prioritize the X’s
identified by your team
Close Choose a spokesperson to report All 3 mins.

out on your high impact X’s. Are all
of these X’s within the control of the
team?
D M A I C
Prioritization Of X’s–Control/impact Matrix 23
IMPACT
High Medium Low
C In Our
O Control
N
T
R
O
L
Out Of
Our Control
D M A I C
24
Analyze 6.2
Narrow list of Potential Causes
D M A I C
The Idea Of Sampling 25
Based on the sample, we make decisions about the population.

Population (Universe): A
set of characteristics that
defines membership in the
complete set.
Sample: A subset of
members that possesses
the same characteristics as
that of the population.
Why should we take a sample?
Should the sample be random?
Is it possible to have sampling error?
How many samples should be taken?
What are some everyday examples of sampling?
In most cases under study, it is impossible to measure

every element of a population. In particular, we want
samples that consist of observations that are independent
and random. These samples, called random samples, can
be used to make decisions about the population. This
involves a certain amount of risk. We can minimize this
risk, and, more importantly, understand what the risks are
if we use statistical tests.
If you were interested in finding the average height of all

of the people in the United States, what advantages would
exist in taking a sample?
D M A I C
Hypothesis Testing–Introduction 26
Refers to the use of statistical analysis to determine if observed

differences between two or more data samples are due to
random chance or to be true differences in the samples
Increase your confidence that probable X’s are statistically
significant
Used when you need to be confident that a statistical difference
exists
An assertion or conjecture about one or more parameters

of a population(s).
To determine whether it is true or false, we must
examine the entire population, this is impossible!!
Instead, use a random sample to provide evidence
that either supports or does not support the
hypothesis
The conclusion is then based upon statistical
significance
It is important to remember that this conclusion is
an inference about the population determined from
the sample data
D M A I C
Hypothesis Testing For Equal Means 27
The histograms below show the height of inhabitants of countries

A and B.
Both samples are of size 100, the scale is the same, and the unit
of measurement is inches.
Question: Is the population of country B, on average, taller than
that of country A?
Country A
Country B
60.0 62.0 64.0 66.0 68.0 70.0 72.0 74.0 76.0 78.0 80.0
[inch]
Issue: how conclusive is the evidence that the sample

results indicate a real, more-than-random effect in the
underlying population or process?
In the Analyze phase we will try to determine which X’s

have an effect on the Y. We can compare two sets of
data, with X set at different values, thereby determining if
that X has an effect.
Examples: Does a process perform better using

machine/material/fixture/tool A or B? Does the
purchased material conform to the desired specifications?
Is there a difference in performance between vendor A or
B? Is there a difference in your process after you make a
change? Is the process on target?
D M A I C
Concepts Of Hypothesis Testing 28
1. All processes have 2. Samples from one

variation. given process may
vary.
3. How can we differentiate

between sample–based
“chance” variation and
a true process
difference?
To improve processes, we need to identify factors which This way everyone makes the same decisions.
impact the mean or standard deviation.
Once we have identified these factors and made

adjustments for improvement, we need to validate actual
improvements in our processes.
Sometimes we cannot decide graphically or by using

calculated statistics (sample mean and standard deviation)
if there is a statistically significant difference between
processes.
In such cases the decision will be subjective.
We perform a formal statistical hypothesis test to decide

objectively whether there is a difference.
D M A I C
Kinds Of Differences 29
Continuous data:
Differences in averages
Differences in variation
Differences in distribution
“shape” of values
Discrete data:
Differences in proportions
There are a variety of different hypothesis tests. Each one

tests for a different kind of “difference.”
D M A I C
Hypothesis Testing 30
Guilty vs. Innocent Example
The American justice system can be used to illustrate the

concept of hypothesis testing.
In America, we assume innocence until proven guilty.

This corresponds to the null hypothesis.
It requires strong evidence “beyond a reasonable doubt”

to convict the defendant. This corresponds to rejecting the null
hypothesis and accepting the alternate hypothesis.
Ho: person is innocent

Ha: person is guilty
D M A I C
Nature Of Hypothesis 31
Null Hypothesis (Ho): Alternative Hypothesis (Ha):

Usually describes a status Usually describes a
quo difference
The one you assume Signs used in Minitab:
unless otherwise shown ≠ or < or >
The one you reject or fail
to reject based upon
evidence
Signs used in Minitab:
= or > or <
Note that we are not proving the hypothesis to be true or

false. We will reject or fail to reject the null hypothesis
based on the evidence from our samples. Failing to reject
the null hypothesis implies that the data does not provide
sufficient evidence to conclude that a difference exists.
On the other hand, rejection of the null hypothesis
implies that the sample data provides sufficient evidence
to say that a difference exists.
D M A I C
Activity–Hypothesis Statements (10 minutes) 32
Write the null and alternate hypothesis testing statements for each
Scenario 1: You
scenario have collected delivery time of supplier A and supplier B. You wish to test
below:
whether or not there is a difference in delivery time from supplier A and B.
Scenario 1: statement
Null hypothesis You have : collected delivery time of supplier A and supplier B.
You wish
Alternate to test statement:
hypothesis whether or not there is a difference in delivery time from
supplier A and B.
Scenario 2: You suspect that there is a difference in cycle time to process purchase orders in
site 1 of your company compared to site 2. You are going to perform a hypothesis test to verify
Null hypothesis statement :
your hypothesis.

Alternate hypothesis statement:
Scenario 2: You
Scenario 3: You have suspect
implemented that there
process is a difference
improvements in the
to reduce cycle time
cycle time to process
to process
purchase orders
purchase in your
orders incompany.
site 1 ofYou havecompany
your collected cycle time before
compared tothe process
site 2. You are
improvements and after the process improvement was implemented. You are going to perform a
going to perform
hypothesis a that
test to verify hypothesis
the processtest to verifyhave
improvements yourresulted
hypothesis.
in a reduction in cycle
time.
Null hypothesis
Null hypothesis statement
statement : :
Scenario 3: You have implemented process improvements to reduce the

cycle time to process purchase orders in your company. You have collected
cycle time before the process improvements and after the process
improvement was implemented. You are going to perform a hypothesis test
to verify that the process improvements have resulted in a reduction in
cycle time.
D M A I C
Guilty vs. Innocent Example

The only four possible outcomes:
1. An innocent person is set free. Correct decision
2. An innocent person is jailed. Type I error = α

– The probability of this type of error occurring we represent as
3. A guilty person is set free. Type II error = β

– The probability of this type of error occurring we represent as
4. A guilty person is jailed. Correct decision
Why is it important to minimize the chance of making a

Type I error in a six-sigma project?
D M A I C
Hypothesis Testing–Another View 34
Ho: Person is innocent. Truth

Ha: Person is guilty. Truth
Ho Ha
Innocent Guilty
Innocent, Guilty,
Ho
Set Free Set Free
Set Free Type II
β
Verdict
Verdict
Innocent, Guilty,
Ha Jailed Jailed
Jailed Type I
α
This is a visual way of looking at the four possible

outcomes of hypothesis testing.
D M A I C
The P-value is calculated by Minitab

The probability of getting the observed difference or greater
when the Ho is true. If p ≥ 0.05, then there is no statistical
evidence of a difference existing.
Ranges from 0.0 - 1.0
The alpha (α) level is usually set at 0.05. Alpha is the
probability of making a Type I Error (concluding there is a
statistical difference between samples when there really is no
difference).
P < α: Reject Ho
P > α : Accept Ho
The p (probability) value is the statistical measure for

the strength of H0, usually reported with a range between
0.0 and 1.0. The higher the p-value, the more evidence
we have to support H0, that there is no difference. Think
of the null hypothesis as a jury trial: the accused is
innocent until proven guilty. In hypothesis testing, the
samples are assumed equal until proven not. Since we are
usually doing a hypothesis test to prove there is a
difference, we are looking for p-values less than 0.05.
By convention if p >.05 accept H0 (no difference).

If p ≤.05 reject H0 (difference exists).
Reject Ho Accept Ho
Different Not Different
0.00 0.05 1.00
p-value
D M A I C
Statistical Tests In Minitab 36
Some basic statistical tests are shown below with the command for running each
test in Minitab.
What The Tool Tests Statistical Test Graphical Test

Histogram
Mean of population data is 1-Sample t-test
different from an established Stat > Basic Statistics
target. > 1-Sample t
Mean of population 1 is 2-Sample t-test

Histogram
different from mean of Stat > Basic Statistics

population 2. > 2-Sample t
The means of two or more 1-Way ANOVA

Histogram
populations is different. Stat > ANOVA > One-Way
Variance among two or more

Homogeneity of
populations is different. Scatter
Variance
Stat > ANOVA >
Plots
Homogeneity of Variance
Output (Y) changes as the Linear

input (X) changes. Regression Box
Stat > Regression >Fitted Plots
Line Plot
Chi-Square Test
Frequency
Output counts from two or Pareto

more subgroups differ. of Independence
Stat > Tables >
Cross Tabulation OR
Chi-Square Test C AB D E M NO
Category
Data is normally Normality Test

distributed Stat > Basic
Statistics
D M A I C
Select A Statistical Test 37
Hypothesis tests to find relationships between project Y and

potential X’s
Y
Continuous Discrete
Continuous Simple Linear

Regression
X 2 Sample t-Test
Discrete (Compare Means of two
samples)
Chi-Square Test
ANOVA (Compare means of
multiple samples)
Homgeneity of Variance
(Compare variances)
D M A I C
Hypothesis Test Summary 38
Normal Data Non-Normal Data

Variance Tests (Continuous Y) Variance Tests (Continuous Y)
X2 - Compares a sample variance to a known Homogeneity of Variance Levine’s–Compares two
population variance.
or more sample variances.
F-test- Compares two sample variances.
Median Tests (Continuous Y)
Homogeneity of Variance
Levine’s–Compares two or more sample Mood’s Median Test–Another test for two or more
variances medians. More robust to outliers in data.
Mean Tests (Continuous Y) Correlation–Tests linear relationship between two
T-test One-sample–Tests if sample mean is equal variables.
to a known mean or target. Proportion Tests (Continuous Y)
T-test Two-sample–Tests if two sample means are P-Test–Tests if two population proportions are
equal. equal.
ANOVA One-Way–Tests if two or more sample Chi-Square Test–Tests if three or more relative
means are equal. counts are equal.
ANOVA Two-Way–Tests if means from samples
classified by two categories are equal.
Correlation–Tests linear relationship between
two variables.
Regression–Defines the linear relationship
between a dependent and independent
variable.
(Here, “Normality” applies to the residuals of
the regression.)
There are a number of hypothesis tests for both normal

and non-normal data. You should consult a Black Belt or
Master Black Belt if you are not sure which test to use for
you project, or if your project involves non-normal data.
The next page shows a summary of the tests that we will

look at in detail during this training. We have already
looked at the Normality Test in previous modules.
D M A I C
We Always Look At Our data In This Way 39
1. Check Stability–Run Chart, p-values

2. Check Shape–Anderson-Darling Normality Test
3. Check Spread–Appropriate Hypothesis Test
4. Check Center–Appropriate Hypothesis Test
The remainder of this module will teach you how to

choose the correct hypothesis test. Not all material will be
covered in class. The instructor will choose the tests to
cover, based on time restrictions. Prior to taking the GB
certification exam you should review this section and go
through the Minitab commands.
D M A I C
Choosing The Correct Hypothesis Test 40
Are X’s YES Is the data NO

Discrete? normal?
Mood’s Median
HOV
NO YES
Regression
Are Y’s NO CHI
Continuous?
SQUARE (χ2)
YES
Comparing NO Multiple Groups

Only 2 ANOVA
Groups?
HOV
YES
Are We
Can I Match NO Comparing NO
X’s With X’s? To A 2 Sample t–Test
Standard? HOV
YES YES
Paired t 1 Sample t
Note: In order to use this chart, we are assuming our X’s Paired t-Test
are discrete. Otherwise, use Regression. Ho = Difference = 0
Ha = Difference ≠ 0
Moods Median Test
One Sample t-Test
Ho = Median1 = Median 2 = ...Median n
Ho = µ = Target
Ha = At least one Median is Different
Ha = µ ≠ Target
Homogeniety of Variance (Hov) – Levene’s Test
Two Sample t-test
Ho = σ12 = σ 22 = σ 32 ...σ 2n Ho = µ 1 = µ 2
Ha = At least σ 2 is Different Ha = µ 1 ≠ µ 2
Chi Square
Ho = χ 2 = 0
Ha = χ 2 ≠ 0
Analysis of Variance (ANOVA)

Ho = µ 1 = µ 2 = ...µ n
Ha = At least one µ is Different
D M A I C
Which Hypothesis Testing Tool Would You Use? 41
For each scenario described below, which hypothesis testing tool

would you use? Assume normal distribution, where appropriate
1. A six-sigma project is being conducted in the field to improve the cycle time for
warranty repair returns. The warranty return cycle time was measured for a
period of 6 weeks for 4 regions. The Green Belt suspects that there is a
difference in average warranty repair cycle time among each of the regions.
How would you test whether there is a statistically significant difference in mean
cycle time for the different regions?
2. Tungsten steel erosion shields are fitted to the low pressure blading in steam
turbines. The most important feature of a shield is its resistance to wear.
Resistance to wear can be measured by abrasion loss, which is thought to be
associated with the hardness of steel. How would you test whether there is a
statistically significant relationship between resistance to wear and abrasion
hardness of steel?
3. Your business purchases sheet stock from two different suppliers. It has found
an unacceptably large number of defects being caused by thickness beyond
tolerance levels. Data for overall mean thickness data was analyzed and found
to be on target. Data was collected that would identify a potential difference in
the variation of the thickness of the material by supplier.
4. Checks Are Us is a payroll processing firm. Timecard errors are routinely

monitored and recorded. A Black Belt investigating the errors wishes to
determine if there are any differences in the number of errors among five of its
major customers. The number of errors contained in a sample of 150 employees
was recorded for five weeks. How would you test if there is a statistically
significant difference in the number of errors among the customers?
D M A I C
Data Analysis 42
D M A I C
Stability–Minitab Run Chart 43
Let’s look at an example from our GB ANOVA.mpj.

Step 1: Check the stability of the processing time for SOUTH
Territory. Are there any trends in the data?
MINITAB FILE: GB ANOVA.mpj
File Description:
Territory (C1): Sales Territory
Zone (C2): city within the sales territory
Total Time (C3): Processing Time- stacked Territories

C4-C8– processing time per territory noted
Use Minitab to check the stability of your processing time

data by territory.
The Run chart is used to look at the stability of data in

time order or sequentially ordered.
Is the process stable over time for the South Territory?
D M A I C
Stability–Minitab
Stability–Minitab Run Chart
Run Chart 44 44
1. Double
Click.
2. Type in a “1.”
Process is Stable if all p-Values are > 0.05
It is important to note that the data should be in time Test for Trends: Ho: No fewer runs observed than
order for run charts to be valid. expected. Ha: More runs observed than expected. Trends
result from a gradual increase or decrease in the measure
A run is one or more consecutive points in the same so that there are fewer runs than expected by sampling
direction. A new run begins each time there is a change error alone.
in direction. Do not count points exactly on the median.
Test for Oscillations: Ho: No more runs observed than
Test for Clusters: Ho: No fewer runs observed than expected. Ha: More runs observed than expected.
expected. Ha: Fewer runs observed than expected. Oscillations are similar to mixtures in that they indicate
Clusters result from sampling from one process for a too many runs—more than expected
period of time and then from another process with a by sampling error alone.
different center for a period so there are fewer runs than
expected by sampling error alone.
Since all p-values are less than 0.05, the process is stable
for the south territory. You would also need to check
Test for Mixtures: Ho: No more runs observed than
each of the other territories for stability.
expected. Ha: More runs observed than expected.
Mixtures result from two or more processes with different
centers alternately sampled so there are too many runs
than expected by sampling error alone.
D M A I C
The P-Value Review 45
Alpha is the probability of making a Type I error

The p-value is the probability of getting the observed difference
or greater when Ho is true
Unless there is an exception based on engineering judgment,
we will set an acceptance level of a Type I error at a = 0.05
Thus, any p-value less than 0.05 means we reject the null
hypothesis
p < α: Reject Ho
p > α: Accept Ho
The alpha is another decision criteria. It will give you the

same conclusions as the confidence interval.
In general, comparing p-value to alpha is the decision

criteria we use most often in this course.
D M A I C
Data Analysis 46
D M A I C
Shape–Minitab Descriptive Statistics 47
Step 2: Check the shape of the data. Is the data normally

distributed?
MINITAB FILE: GB ANOVA.MPJ
Ho: The data is normally distributed

Ha: The data is not normally distributed
D M A I C
Descriptive Statistics–Input and Output 48
1. Double
Click
3. Click
OK 2. Click on Graphs,
Select Graphical Summary
Descriptive Statistics
Variable: South
Non-normal
Anderson-Darling Normality Test distribution!
A-Squared: 3.095
P-Value: 0.000
Mean 11.9348
StDev 11.1064
Variance 123.351
Skewness 1.75318
Kurtosis 2.69797
N 46
0 10 20 30 40
Minimum 1.0000
1st Quartile 4.0000
Median 9.0000
3rd Quartile 14.2500
8.6366 15.2330
7 8 9 10 11 12 13 14 15 16 95% Confidence Interval for Sigma
9.2120 13.9887
6.9173 12.0000
Data is Normal if P-Value is > 0.05 (Accept Ho)
If the p-value is > 0.05, then we Accept the Ho and the

data is normally distributed. In this case, our p-value is
0.000, therefore, we have a non-normal distribution.
D M A I C
Non-Parametric Tests 49
What if I don’t have normal data?
Perform Non-parametric Tests
Use HOV for spread

and
Mood’s Median for Center
Normal data allows a great variety of techniques to be

used in analyzing it
However, responding correctly is critical if the data is
not normal (which is not at all uncommon)
D M A I C
Non-Parametric Tests 50
One of the advantages of nonparametric tests is that they

assume no knowledge about the underlying distributions. They
often use an analysis of the ordered ranks of the data.
Nonparametric tests are more powerful for non-normal data
than the equivalent t-tests and ANOVA tests
The Mood’s Median and Homogeneity of Variance Test
(HOV) are nonparametric tests which are very similar to the
One-Way ANOVA test
Many times in commercial projects the data are not

normally distributed. Cycle time is one example. Cycle
time will often be skewed to the right.
Another example is ordinal data. This is data that is not
continuous but can be placed in ordered categories
such as good, better, and best. This often occurs with
survey data.
In these cases a test that does not depend on
assumptions of normality may be very useful
D M A I C
Spread–Test On Variances 51
Many times we want to know if we have succeeded in reducing the

variation of a process, or we might want to know if a change in a
variable (X) changes the variation in the output (Y).
Knowing if the variances of two (or more) samples are different is

also a prerequisite to the test on means.
Are these Variances Different?
Xb
Xa
The hypothesis test for comparing variances

is the Homogeneity of Variance Test
In the Analyze phase we will try to determine which X’s

have an effect on the Y. We can compare two sets of
data, with X set at different values, thereby determining if
that X has an effect.
Examples: Does a process perform better using

machine/material/ fixture/tool...A or B? Does the
purchased material conform to the desired specifications
(m & s)? Is there a difference in performance between
vendor A or B? Is there a difference in your process after
you make a change? Has a variation source been
removed?
D M A I C
Spread–Homogeneity Of Variance 52
Step 3: You want to compare the variability of the Sales Regions.

You have established that each process is stable but non-normally
distributed. Perform the Homogeneity of Variance Test .
D M A I C
Test On Variances, Example–Input 53
2 = 2 = 2 = 2 = 2
Ho : σEast σ West σ IFG σ Central σ South
Ha : at least one is different α = .05
D M A I C
Test On Variances, Examples–Output 54
2 = 2 = 2 = 2 = 2
Ho : σEast σ West σ IFG σ Central σ South
α = .05
Ha : at least one is different
Homogeneity of Variance Test for total time
95% Confidence Intervals for Sigmas Factor Levels
Central
Bartlett's Test
Test Statistic: 58.348

East
P-Value : 0.000
IFG
Levene's Test
South Test Statistic: 10.396

P-Value : 0.000
West
5 15 25 35 45 55
We have standardized on
The Levene’s test because
it is more robust to non-
p < α: Reject Ho normality of samples.
p > α: Accept Ho
If you want to compare the performance of two

territories, you should test their variances independently
of the other territories. This way the extra noise of the
other territories will not interfere with the analysis. You
may end up with different conclusions running them
independently, versus running them with all the other
territories.
The confidence intervals that are shown are intervals for

the standard deviation, even though we are comparing
variances (variance = σ2).
D M A I C
Center–Non-Normal Data Test 55
Mood’s Median Test

A Mood’s Median Test is used to test two or more population
medians. This test is robust to outliers or errors in data (shows
statistical significance).
This test will show statistical significance. It will not show
practical significance.
Just as with our earlier example, a picture is worth a

thousand words. Looking at a graphical depiction of the
data can help to verify whether or not findings (even
findings with statistical significance) have any practical
significance. By way of an analogy, even though you
could statistically prove that people are heavier before a
haircut than they are afterwards, nobody seeks out a
barber shop when they go on a diet. Even though a
statistical difference might be able to be proven, the
difference is negligible from a practical standpoint.
Viewing the data graphically helps to confirm how much
practical significance exists in what you’ve found.
D M A I C
Center–Mood’s Median Test 56
Step 4: Check the centers. Are the centers different by region?

Stat > Nonparametrics > Mood’s Median Test

– Response = Total Time
– Factor = Territory
Answer the following questions:

1.Are the medians different?
(interpret the p-value)
2.Which medians are different?
(interpret the confidence intervals)
~ ~ ~ ~ ~
H :X = X = X = X = X
o 1 2 3 4 5
H : At least one is different

a
Another type of analysis that may be investigated is

whether or not multiple samples of data can be shown to
be different from each other by a (statistically) significant
amount. The example shown here is questioning whether
similar territories are taking different amounts of time to
process loan applications as measured in days.
One test which can check for this condition in non-

normal data is called a Mood’s Median Test.
It should be noted that, unless it can proven otherwise,

the two groups are assumed to have the same median. In
other words, the null hypothesis for a Mood’s Median
Test is that the medians are equal.
D M A I C
Mood’s Median Test 57
D M A I C
Mood’s Median Test 58
Mood Median Test
Mood median test for total ti
Chi-Square = 108.88 DF = 4 P = 0.000
Individual 95.0% CIs

Territor N<= N> Median Q3-Q1 ---------+---------+---------+-------
Central 97 15 3.0 4.7 (+
East 47 1 10.0 8.8 (+)
IFG 2 11 28.0 56.5 ------+---------------------)
South 25 21 9.0 10.3 (-+)
West 22 91 15.0 17.5 (+-)
---------+---------+---------+-------
20 40 60
Overall median = 9.0
Recall, when two groups are equal, the confidence intervals overlap.
Ho: All medians are equal

Ha: At least one median is not equal
Notice that Minitab has converted the continuous data

into attribute data. For each level of the X, Minitab
supplies a count of how many values fall at or below
and above the overall median.
The Mood’s Median test then performs a simple Chi-

square test (which we’ll cover later) on this summary
table
− If the number above and below are proportional, Chi-
square will be low and p will be high, resulting in the
conclusion that the medians are equal.
− If the number above and below are not proportional, Chi-
square will be high and p will be low, resulting in the
conclusion that the medians are not equal
The results from manually performing a Chi-square

test using this summary table are the same as the
Mood’s Median test
D M A I C
Hypothesis Testing Procedure 59
1. Write the null hypothesis

Ho: There is no difference between Population A and B
µpop1 = µpop2
Team
preparation
2. Write the alternate hypothesis
Ha: There is a difference between Samples A and B
µpop1 ≠ µpop2
3. Decide on the alpha level α =.05 (typical for DMAIC projects)
4. Chose hypothesis test Choose the correct test, given the type of X and Y data.
5. Gather evidence and test/conduct analysis Collect data, run analysis, get p-value
6. Decide to Reject H0, or not reject H0, and If p ≥ 0.05 conclude, no difference between populations
draw conclusion If p < 0.05 conclude, the populations are different
The first step in hypothesis testing is to write a null

hypothesis for the population being investigated. State
that there is no difference between populations that you
are studying. The next step is to write the alternate
hypothesis (which is often what your team REALLY
thinks is true), which indicates that a difference does exist
between the population. The decision regarding a p-value
is usually straightforward. By convention, a p-value of
.05 is typical because that is sufficient to give us the
confidence we need to go forward whether we reject or
do not reject the null hypothesis. The team then collects
whatever additional data is necessary. The correct
hypothesis test can then be chosen and the analysis is
conducted. The team now has the necessary information
in order to know whether or not to reject or accept the
null hypothesis.
D M A I C
1-Sample Hypothesis 60
1. Ho : µ = constant = T
Ha : µ ≠ constant = T
Ho Ha
2. Ho : σ 2 = constant = T
Ha : σ 2 ≠ constant = T
There are 7 types of hypotheses that we will be covering.
We are standardizing the method by always trying to

accept, with a degree of confidence, the alternative
hypothesis when evidence supports the decision. We are
the prosecuting attorney. Set up the hypotheses according
to this standard convention.
D M A I C
1-Sample t-Test 61
Ho : µ = constant = T
Ha : µ ≠ constant = T
Minitab File: GB case study.mtw

Null Hypothesis: There is no difference in June’s delivery time
average with that of a target set by May’s delivery time average of
5 minute’s late.
Minitab Command: Stat > Basic Statistics > 1–Sample t–test
D M A I C
1-Sample t-Test 62
D M A I C
1-Sample t-Test 63
1. Double click
anywhere on the C5
line to select time as
a variable
2. Click Test Mean and

enter 5.0 as May’s
average, the
Alternative is that
they are not equal
3. Click Graphs to show

a pictorial
4. Click on
Histogram of data
5. Click OK
D M A I C
1-Sample t-Test 64
Histogram of time
(with Ho and 95% t-confidence interval for the mean)
50
40
Frequency
30
20
10
0 _
X Ho
[ ]
-100 0 100
time
The session window has our answer. If the p-value is less

than .05, we reject the Null Hypothesis and accept its
alternate, that there is a statistical difference in averages
between this month’s data and May’s data.
D M A I C
2-Sample Hypothesis 65
3. Ho :µ1 = µ2
Ha : µ1≠ µ2
4. Ho : µ1 ≤ µ2
µ1 µ2
Ha : µ1 > µ2
5. Ho : σ1 2 = σ22
Ha : σ1 2 ≠ σ22
σ1
σ2
D M A I C
2-Sample t-Test 66
Ho :µ1 = µ2
Ha : µ1≠ µ2

Null Hypothesis: There is no difference in the average delivery
time between the two different load types.
Minitab Command: Stat > Basic Statistics > 2–Sample t–Test
Let’s test the hypothesis that there is no difference in the

average delivery time between the two different load
types.
D/H–Drop & Hook–A driver unhooks their full trailer,

and then hooks into an empty trailer to take back.
Unload–The driver opens the trailer and allows the store

to unload it’s freight, and brings back the trailer.
D M A I C
2-Sample t-Test 67
D M A I C
2-Sample t-Test 68
1. Double click
anywhere on the
C5 line to select
time as a variable
for samples.
2. Click on C3 to chose
LoadType as the
variable for
subscripts.
3. Click Graphs to
show a pictorial.
4. Click on
Boxplots of data.
5. Click OK.
D M A I C
2-Sample t-test 69
Boxplots of time by LoadType

(means are indicated by solid circles)
100
time
-100
D/H Unload
LoadType

alternate, that there is a statistical difference in our Drop
& Hook and Unload averages.
D M A I C
Multi Sample Hypothesis 70
6. Ho : µ1 = µ2 = . . . = µn
Ha : at least one is not equal.
7. Ho : σ12 = σ22 = . . . = σn2
D M A I C
Analyze Of Variance (ANOVA) 71
Ho : µ1 = µ2 = . . . = µn

Null Hypothesis: There is no difference in the average delivery
time between Capital Logistics drivers.
Minitab Command: Stat > ANOVA > One-Way

average delivery time between Capital Logistics drivers,
Temporary drivers working for us, and our 3 Third party
carriers we use in areas that are not supported by Capital
Logistics.
D M A I C
Analysis Of Variance (ANOVA) 72
D M A I C
1. Double click
anywhere on the C5
line to select time as
a response variable.
2. Double click C2
for Driver as a
factor.
3. Click Graphs
to show a
pictorial.
4. Click on
Boxplots of data.
5. Click OK.
D M A I C
Boxplots of time by Driver

(means are indicated by solid circles)
100
time
-100
Driver
Trans, Inc.
Temp Labor
Cap. Log.
JJ Truck
DNGY Inc.
1. Note the sample

number for each
group.
2. These 3 are
alike.
3. These 2 are
definitely
different.

alternate, that there is a difference in at least 1 average
between the 5 different groups of drivers.
D M A I C
Homogeneity Of Variance (HOV) 75
Ho : σ12 = σ 22 = ... = σn2
Ha : at least one is not equal

Null Hypothesis: There is no difference in the variation in delivery
time between Capital Logistics drivers.
Minitab Command: Stat > ANOVA > Homogeneity of Variance

variation in delivery times between Capital Logistics
drivers, Temporary drivers working for us, and our 3
Third party carriers we use in areas that are not supported
by Capital Logistics.
D M A I C
D M A I C
1. Double click anywhere

on the C5 line to select
time as a response
variable.
2. Double click C2 for

Driver as a factor.
3. Click OK.
D M A I C
1. We have standardized
our Levene’s Test
because it is more
robust for non-normal
data.
2. Our p-value is
less than 0.05.
The graphical output has our answer. If the p-value is

less than .05, we reject the Null Hypothesis and accept its
alternate, that there is a difference in standard deviation
between at least two of the drivers.
Always use the p-value for Levene’s Test.
D M A I C
Review: Scatter Plots 79
R-value
y r = –1.0 y r = +1.0
x x
r = +.7 r = –.7
y y
x x
y r=0 y r=0
x x
The correlation coefficient r measures the strength of

linear relationships.
–1 ≤r ≤1
When a relationship exists, the variables are said to be

correlated.
Perfect negative relationship: r = –1.0

No linear correlation: r = 0
Perfect positive relationship: r = +1.0
If no correlation is apparent, don’t stop. Use histograms,

regression, rational sub-grouping, etc. to try to extract
some meaning from the data.
D M A I C
Simple Linear Regression 80
We have shown/talked about positive and negative correlation

of two data sets
Regression analysis is a statistical technique used to build the
Y = ƒ(x) relationship between two or more variables. The model
is often used for prediction.
Regression is a hypothesis test. Ha: The “X” is a significant
predictor of the response.
It may be used to analyze relationships between the “X’s”, or
between “Y” and “X”
Regression is a powerful tool, but can never replace process
knowledge about trends
D M A I C
Y = b 0 + b 1X 1
Y Ha: The model is a significant predictor of the

response.
b0 = Predicted value of Y when X1 = 0
b1 = Slope of line
X change in Y per unit
change in X1

Null Hypothesis: There is no correlation between our
continuous Y metric (time) and a continuous X metric (distance)
Minitab Command: Stat > Regression > Fitted Line Plot
Let’s test the hypothesis that there is no correlation

between our continuous Y metric–time and a continuous
X metric–distance.
D M A I C
D M A I C
1. Double click anywhere on

the C5 line to select time
as a response variable.
2. Double click C6 for

Distance as a
predictor.
3. We believe there to
be a linear
relationship.
4. Click OK.
D M A I C
Regression Plot
Y = -29.1923 + 0.174971X
R-Sq = 10.5 %
100
2. This value means

that 10.5% of the
variation in Y - time
is explained by this X
time
0 - distance.
-100
1. There is a very weak

positive relationship 0 100 200 300 400
between distance
and timely delivery Distance
Distance is a
significant factor.
Entire model is
significant.

alternate, that there is a correlation between delivery time
and distance, but that it is a very weak one (as seen by
Low R-Sq value).
Hypothesis test for the model (p-value given for

“Regression”)
Ho: Model = O
Ha: Model ≠ O
D M A I C
Chi-Square Test 85
Y
Continuous Discrete
Continuous
X
Discrete
Chi-Square Test
D M A I C
Chi-Square Tests 86
The Chi-Square Tests

Used for:
1 - Goodness-of-Fit Test: To test if an observed
set of data fits a model (an expected set of
data)
2 - Test of Independence: To test hypothesis of
several proportions (contingency table)
It’s for discrete data, any

number of categories
For all cases, Ho: no difference in data
Ha: difference exists
χ2
The Chi-Square test provides a way to determine if a
given set of data has the hypothesized distribution. For
example, if we roll a die 600 times., the number 1 occurs
96 times, 2–88 times, 3–104 times, 4–107 times, 5–112
times, and 6–93 times. In theory, each number should
occur 100 times. The Chi-Square test enables us to
determine if the observed set of data, the output of 600
rolls of a die, fits an expected model.
The Chi-Square test will also let you know if any factor
has an impact on the output. If there is no difference in
the data (Ho), this means no factor(s) have an impact on
the output. If a difference exists (Ha), one or more
factors have an impact on the output.
D M A I C
The Nature And Use Of Chi-Square 87
g 2
fo - fe
χ =
2
Σ fe
j=1
g = number of categories
Suppose we flip a coin N
fo = observed frequency
= 100 times and observe
fe = expected frequency
63 heads and 37 tails.
Could this ratio of heads
to tails occur by chance or (f0-fe)2
should we conclude that fe
Penny
the coin is somehow
biased ?
Heads 63/50* 3.38
3.38
Tails 37/50*
*=Observed Expected
(f0) (fe) χ 2 = 6.76
The Goodness-of-Fit test examines the difference

between what is expected and what is observed. If this
difference is large, you should be suspicious that
something is influencing the output (one of the factors).
The c 2 value represents a test statistic used for
comparison to a critical value.
In this example, you would expect an unbiased coin to

have equal occurrences of heads and tails (50/50). We
know that we won’t always have this ratio of heads to
tails because of random chance.
If there is a large difference between what is expected

and what is observed, the chi-square value will be large
(see formula). If the observed value is close to the
expected value, the chi-square value will be small. How
large does chi-square have to be to say the coin is biased?
D M A I C
The Goodness Of Fit Test 88
To determine whether a sample comes from a population with a known

distribution, calculate chi-square as shown on the previous page.
H0 = good fit (observed distribution is consistent
with random outcomes from the expected
model)
Ha = sampled population does not have the
expected distribution
If the observed frequencies are close to the expected frequencies, chi-
square is small. So a low chi-square leads to rejecting Ha.
If chi-square is greater than a critical value, the null hypothesis is
rejected.
The critical values of chi-square for various confidence levels are given
in the table “Chi-Square Distribution.”
Each of the expected frequencies must be at least 5 for this test to be
valid.
D M A I C
Degrees Of Freedom 89
The chi-square distribution has “degrees of freedom” as a

parameter.
For a goodness-of-fit test, the number of degrees of freedom is k-

1, where k is the number of possible outcomes.
Thus for the penny example,

df = 2 - 1 = 1
D M A I C
The Statistical Test 90
For the penny example:

Critical value ( α = 0.05, dof = 1) = 3.841
χ2Calculated > χ2Table => There is a difference (I’m seeing something

unusual).
Since the statistic from the data, 6.76, is greater than the table
critical value, the alternative hypothesis is accepted: we are
confident that we saw unusual behavior with that penny.
Use the tables on the next two pages to determine the chi-
square critical value.
D M A I C
The Chi-Square Distribution 91
df=(# columns-1) x (# Rows-1)

df .250 .100 .050 .025 .010 .005 .001 Alpha
1 1.323 2.706 3.841 5.024 6.635 7.879 10.828
2 2.773 4.605 5.991 7.378 9.210 10.597 13.816
3 4.108 6.251 7.815 9.348 11.345 12.838 16.266
4 5.385 7.779 9.488 11.143 13.277 14.860 18.467
5 6.626 9.236 11.070 12.832 15.086 16.750 20.515
6 7.841 10.645 12.592 14.449 16.812 18.548 22.458

7 9.037 12.017 14.067 16.013 18.475 20.278 24.322
8 10.219 13.362 15.507 17.535 20.090 21.955 26.125
9 11.389 14.684 16.919 19.023 21.666 23.589 27.877
10 12.549 15.987 18.307 20.483 23.209 25.188 29.588
11 13.701 17.275 19.675 21.920 24.725 26.757 31.264

12 14.845 18.549 21.026 23.337 26.217 28.300 32.909
13 15.984 19.812 22.362 24.736 27.688 29.819 34.528
14 17.117 21.064 23.685 26.119 29.141 31.319 36.123
15 18.245 22.307 24.996 27.488 30.578 32.801 37.697
16 19.369 23.542 26.296 28.845 32.000 34.267 39.252

17 20.489 24.769 27.587 30.191 33.409 35.718 40.790
18 21.605 25.989 28.869 31.526 34.805 37.156 43.312
19 22.718 27.204 30.144 32.852 36.191 38.582 43.820
20 23.828 28.412 31.410 34.170 37.566 39.997 45.315
21 24.935 29.615 32.671 35.479 38.932 41.401 46.797

22 26.039 30.813 33.924 36.781 40.289 42.796 48.268
23 27.141 32.007 35.172 38.076 41.638 44.181 49.728
24 28.241 33.196 36.415 39.364 42.980 45.558 51.179
25 29.339 34.382 37.652 40.646 44.314 46.928 52.620
26 30.434 35.563 38.885 41.923 45.642 48.290 54.052

27 31.528 36.741 40.113 43.194 46.963 49.645 55.476
28 32.620 37.916 41.337 44.461 48.278 50.993 56.892
29 33.711 39.087 42.557 45.722 49.588 52.336 58.302
30 34.800 40.256 43.773 46.979 50.892 53.672 59.703
40 45.616 51.805 55.758 59.342 63.691 66.766 73.402

50 56.334 63.167 67.505 71.420 76.154 79.490 86.661
60 66.981 74.397 79.082 83.298 88.379 91.952 99.607
70 77.577 85.527 90.531 95.023 100.425 104.215 112.317

80 88.130 96.578 101.879 106.629 112.329 116.321 124.839
90 98.650 107.565 113.145 118.136 124.116 128.299 137.208
100 109.141 118.498 124.342 129.561 135.807 140.169 149.449
D M A I C
The Chi-Square Distribution 92
df .995 .990 .975 .950 .900 .750 .500 Alpha

1 .000039 .000160 .000980 .003930 .015800 .101500 .455000
2 0.010 0.020 0.051 0.103 0.211 0.575 1.386
3 0.072 0.115 0.216 0.352 0.584 1.213 2.366
4 0.207 0.297 0.484 0.711 1.064 1.923 3.357
5 0.412 0.554 0.831 1.145 1.610 2.675 4.351
6 0.676 0.872 1.237 1.635 2.204 3.455 5.348

7 0.989 1.239 1.690 2.167 2.833 4.255 6.346
8 1.344 1.646 2.180 2.733 3.490 5.071 7.344
9 1.735 2.088 2.700 3.325 4.168 5.899 8.343
10 2.156 2.558 3.247 3.940 4.865 6.737 9.342
11 2.603 3.053 3.816 4.575 5.578 7.584 10.341

12 3.074 3.571 4.404 5.226 6.304 8.438 11.340
13 3.565 4.107 5.009 5.892 7.042 9.299 12.340
14 4.075 4.660 5.629 6.571 7.790 10.165 13.339
15 4.601 5.229 6.262 7.261 8.547 11.036 14.339
16 5.142 5.812 6.908 7.962 9.312 11.912 15.338

17 5.697 6.408 7.564 8.672 10.085 12.792 16.338
18 6.265 7.015 8.231 9.390 10.865 13.675 17.338
19 6.844 7.633 8.907 10.117 11.651 14.562 18.338
20 7.434 8.260 9.591 10.851 12.443 15.452 19.337
21 8.034 8.897 10.283 11.591 13.240 16.344 20.337

22 8.643 9.542 10.982 12.338 14.041 17.240 21.337
23 9.260 10.196 11.688 13.091 14.848 18.137 22.337
24 9.886 10.856 12.401 13.848 15.659 19.037 23.337
25 10.520 11.524 13.120 14.611 16.473 19.939 24.337
26 11.160 12.198 13.844 15.379 17.292 20.843 25.336

27 11.808 12.879 14.573 16.151 18.114 21.749 26.336
28 12.461 13.565 15.308 16.928 18.939 22.657 27.336
29 13.121 14.256 16.047 17.708 19.768 23.567 28.336
30 13.787 14.953 16.791 18.493 20.599 24.478 29.336
40 20.707 22.164 24.433 26.509 29.051 33.660 39.335

50 27.991 29.707 32.357 34.764 37.689 42.942 49.335
60 35.535 37.485 40.482 43.188 46.459 52.294 59.335
70 43.275 45.442 48.758 51.739 55.329 61.698 69.334

80 51.172 53.540 57.153 60.391 64.278 71.145 79.334
90 59.196 61.754 65.647 69.126 73.291 80.625 89.334
100 67.328 70.065 74.222 77.929 82.358 90.133 99.334
D M A I C
The Statistical Test (continued) 93
The 5% risk column in the table already tells us there is a

difference. Alternately, a p-value (the probability of being wrong if
accepting the alternative hypothesis) can be computed in EXCEL
as follows:
= CHIDIST(statistic,dof)
= CHIDIST(6.76,1)
Excel will return the p-value, in this case 0.0093. Since the
probability of being wrong is less than 0.05, we accept the
alternative hypothesis: we are confident that we saw unusual
behavior with that penny.
Minitab does not have a Macro to accommodate the

Goodness-of-Fit test. It is most convenient to compute
the chi-square value in Minitab and then evaluate the
corresponding p-value in Excel.
One can obtain the p-value in Minitab, but it takes a little

work.
D M A I C
Test Of Independence 94
Suppose a bill has been introduced in the Government to raise the

speed limit. A polling organization polls 45 members of Political
Party 1 and 55 members of Political Party 2, and finds 30 Party 1
and 35 Party 2 members oppose the proposal. Is there a
significant effect of party affiliation on preference for the proposal?
Ho = no difference between parties
Ha = one party likes the proposal significantly more than the
other
We organize the data in a “contingency table”
Party 1 Party 2
Oppose Total = 65
fo = 30 fo = 35
Favor fo = 15 fo = 20 Total = 35
Total = 45 Total = 55
Use this example to demonstrate the test for

independence and contingency tables.
The test for independence asks the question: Does some

factor have an effect on the output? For example, does
car brand affect injury severity?
To test independence, we put the data in a contingency

table. We then test to see if the row and column method
for classification are independent.
D M A I C
Test Of Independence (continued) 95
Now calculate the expected distribution, given there is no party

preference. For example, the expected frequency of Party 1
opposing the proposal would be:
fe = Row Total x Column Total

Grand Total
= (65/100) x 45 = 29.25
Note that for each cell, this is the total frequency for the row, times
the total frequency for the column, divided by the total population
(see chart on following page).
You will be determining your expected frequency on the

ratio of number of people who oppose the proposal to the
number of people who favor the proposal (regardless of
party affiliation). Out of the total 100 (party affiliation
unknown) people, 65 oppose the proposal and 35 favor
the proposal. Expected ratios: 65/100 oppose; 35/100
favor. When you multiply these ratios by the total in a
party, you get the expected frequency.
If the population of Party 2 (or Party1) has a statistically

different (p < α) observed frequency than the expected
frequency (total or average) accept Ha–one party likes
the bill significantly more than the other.
D M A I C
Expected Frequency And Degrees Of Freedom 96
fe = frow x fcol
N
Party 1 Party 2
Oppose fe = 29.25 fe = 35.75 Total = 65
Favor fe = 15.75 fe = 19.25 Total = 35
Total = 45 Total = 55
When performing a chi-square test on a contingency table, the

number of degrees of freedom equals the number of rows in the
table minus one, times the number of columns in the table minus
one.
df = (2-1) x (2-1) = 1
Expected frequencies are calculated using the

methodology from the previous page.
We could use the expected and observed frequencies to

calculate the Chi-Square statistic like we have seen
before, and then compare it to the critical Chi-Square
value from the table using (alpha) and df. But, as usual,
we will have Minitab do this for us. See the next pages.
D M A I C
Minitab Example 97
MINITAB FILE: (Create Your Own)
H o: There is no difference between party affiliation and preference on the proposal.

H a: There is a difference between party affiliation and preference on the proposal.
For this example, Party 1 = Republicans

Party 2 = Democrats
D M A I C
Minitab Input And Results 98
1. Double click on
both columns.
2. Click OK.
Minitab calculates
the expected
frequency (fe)
Minitab calculates the expected frequencies.
The calculated chi-square is less than the critical value of

3.841, so we accept Ho.
The p-value is greater than .05, so we are within our 95%

confidence zone of good fit.
There is no apparent relationship between political party

affiliation and preference on the speed limit proposal.
D M A I C
Chi-Square Test–Case Study Example 99

Ho: There is no difference between Driver types and their on-time
delivery performance.
Ha: At least one driver type is better or worse at on-time delivery
performance.
Minitab Command: Stat > Tables > Cross–Tabulation
In our familiar case study example, with our data in rows,

we will let Minitab build the contingency table.
D M A I C
Chi-Square Test 100
MINITAB FILE: GB case study.MTW
D M A I C
Chi-Square Test 101
1. Double click on
both columns
C2 & C7..
2. Check the Chi-

Square Analysis
and Show Count
buttons.
3. Click OK.
Since our discrete data is in columns and not tabled, we

will first cross-tabulate and then calculate Chi-Square.
Since our p-value is less than .05, we reject the Ho and

accept it’s alternate, Ha.
Compare this p-value with the one done on the ANOVA

test for driver type & time. Are they the same?
D M A I C
Chi-Square Test–Summary 102
Goodness of Fit Test

– Does our data fit an expected distribution?
Test of Independence
– Does some factor have an effect on the output?
2
χ statistic 2
– χ ≤critical value Accept H o

2
– χ > critical value Reject Ho

p-value
– p < 0.05 Reject Ho
– p ≥ 0.05 Accept Ho
D M A I C
Hypothesis Testing–Breakout Activity (20 minutes) 103
Desired Outcome: Select and perform the appropriate statistical test for a given
project situation
What How Who Timing
Review Review the hypothesis testing scenarios All 5 mins.

Scenarios recruitment cycle time case study that
was introduced in the Minitab and
Graphical Analysis module.
Select For the scenario(s) assigned to your All 5 mins.

Test team, complete the hypothesis testing
worksheet to select the appropriate
statistical test
Perform Test Perform the test in Minitab All 10 mins.

In Minitab
Write the p-value from the test and your
conclusions on the hypothesis testing
worksheet
Close Be prepared to share your analysis and All

results with the class
D M A I C
Hypothesis Testing Scenarios 104
The following scenarios relate to the HR recruitment cycle time

case. In each case, perform the appropriate statistical test. (Data
File: fill_time.mtw)
1. The project team wants to determine if there is a statistical difference

in mean cycle time to fill technical positions depending on the type of
resource requested.
2. The project team wants to determine if there is a statistical difference

in cycle time variation, depending on the type of resource requested.
3. The project team wants to determine if there is statistical difference

between the number of cycle time defects for each type of resource.
4. The project team wants to determine if recruitment cycle time

increases as the number of years of experience of the person hired
increases.
D M A I C
Hypothesis Testing Worksheet 105
1. State the H0
2. State the Ha
3. State α level
4. What test should we use?
5. Will you reject or accept H0?
1. H0: ___________________________________________________________________
2. Ha: ___________________________________________________________________
3. α level: ________________________________________________________________
4. Test: _________________________________________________________________
5. Accept or Reject H0? _____________________________________________________
D M A I C
Hypothesis Testing Worksheet 106
1. State the H0
2. State the Ha
3. State α level
4. What test should we use?
5. Will you reject or accept H0?
1. H0: ___________________________________________________________________
2. Ha: ___________________________________________________________________
3. α level: ________________________________________________________________
4. Test: __________________________________________________________________
5. Accept or Reject H0? _____________________________________________________
D M A I C
1
Improve Phase Flowchart 1
D M A I C
IMPROVE
PHASE
OVERVIEW
Improve 8: Improve 9:
Improve 7: Discover Establish
Screen Variable Operating
Potential Relationships Tolerance & Pilot
Causes & Propose Solutions
Solution
DMAIC GB R TX PG V 4.2.0
D M A I C
Improve Phase Overview 2
What is the Improve phase?

The Improve phase is when your team:
Selects those product performance characteristics that must be improved to achieve
the improvement goal by identifying the major sources of variation in the process
Develops and pilots process improvements
Why is the Improve phase important?

This phase is important because it specifically identifies how the process should
be improved.
Steps involved in the Improve phase:

Improve 7: Screen Potential Causes
Improve 8: Discover Variable Relationships & Propose Solution
Improve 9: Establish Operating Tolerances & Pilot Solution
D M A I C
3
The12-Step Process

Define
Measure
Standards Project Y
R&R
Analyze
Graphical Analysis,
Hypothesis Tests
Improve
Relationships
Tolerances
Control
Capability
D M A I C
Improve 7–Screen Potential Causes 4
What does it mean to Screen Potential Causes?

Potential Causes are the possible Vital Few (or Critical) X’s discovered during Analyze 6
that have a significant effect on the Project Y. The Potential Causes are identified by the
improvement team. During this step, the Vital X’s discovered in Analyze 6 will be
statistically verified by obtaining more data, if necessary.
Why is it important to Screen Potential Causes?

Potential X’s are important because they help Green Belt teams focus their identification
of root causes. Once potential X’s are verified as having a significant effect on the
Project Y, then Green Belt teams will use the identified Vital X’s to focus improvement
efforts.
What are the project tasks for completing Improve 7?

7.1 Verify Vital Few X’s
IMPROVE STEP
OVERVIEW
Improve 7:
Discover Variable Establish
Screen Potential
Relationships & Operating
Causes
Propose Solution Tolerances & Pilot
Solution
7.1 Verify Vital Few X’s
D M A I C
Improve 8–Discover Variable Relationships and 5
Propose Solution
What does it mean to Discover Variable Relationships and Propose a Solution?

A Final Solution is the improvement strategy that the team develops based upon the
root causes verified in the previous step.The final solution will address the Vital X’s
while maintaining customer focus.
Why is it important to Discover Variable Relationships and Propose a Solution?

A Final Solution will eliminate or reduce the effect of the Vital Few X’s, thus improving
the process capability of the Project Y. The Final Solution must not only address the
Vital Few X’s, but you must also consider whether the solution satisfies the customer’s
CTQ’s. By using the FMEA tool, you can identify possible Failure Modes within the
Final Solution, identify what effect the Failure Modes would have on either the
customer’s CTQ’s or the new process, and create action plans to eliminate or reduce
the likelihood of that Failure Mode occurring.

8.1 Generate a transfer function to optimize the changes in Vital X’s (if necessary)
8.2 Propose solution
8.3 Refine solution
IMPROVE STEP
OVERVIEW
Improve 7:
Screen Potential
Causes
Solution
8.1 Generate a transfer function to optimize the changes in vital X’s (if necessary)
8.2 Propose solution
8.3 Refine solution
D M A I C
6
Improve 9–Operating Tolerances & Pilot Solution 6
What does it mean to establish Operating Tolerances and Pilot Solutions?

An Operating Tolerance is established when the X-Y transfer function is known and
the required specification for Y is known. The tolerance for X can be set by using the
specification for Y in the transfer function.
A pilot is a test of all or part of a proposed solution on a small scale in order to better
understand its effects and to learn about how to make the full scale implementation
more effective.
Why is it important to establish Operating Tolerances & Pilot Solution?

Operating tolerances allow you to control the settings of Vital X’s so that you can
better predict/control the output of the process – the Y. A pilot allows you to improve a
proposed solution to better meet customer CTQ’s, lower the risk of failure of a
proposed solution, confirm expected results of a proposed solution, and increase the
opportunity for feedback and buy-in for a proposed solution.

9.1 Establish operating tolerances for Vital X’s
9.2 Pilot solution
IMPROVE STEP
OVERVIEW
Improve 7:
Screen Potential
Causes
Solution
9.1 Establish operating tolerances for Vital X’s

9.2 Pilot solution
D M A I C
D M A I C
Collect data
D M A I C
D M A I C
1
Design Of Experiment: Learning Objectives
Define the purpose & benefits of conducting a Design of

Experiment
Describe screening DOE and optimization DOE and how they
are used in the Improve phase
Review full factorial DOE design
Describe the 5 steps in analyzing DOE data, the key
components of each step and the use of Minitab to analyze the
data
– Plot the raw data
– Plot the residual*
– Examine the factor effects
– Confirm impression with statistical procedures
– Summarize conclusions
Describe the use of fractional factorial design and explain the
benefits & the trade-off over full factorial design
* We don’t discuss residuals here, but to fully analyze

your DOE, seek assistance from your BB/MBB and/or
the “Book of Knowledge”

DMAIC GB S TX PG V 4.2.0
D M A I C
2
Design Of Experiments
With the use of Design of Experiments (DOE), we will be able

to:
Verify the “Vital Few” X’s that impact the quality of the Project Y
Identify the best combination X–values to optimize process
performance and meet customer CTQ’s
DOE is powerful in isolating the Vital Few causes from

the many factors affecting a process.
In identifying and testing solutions, a team can apply

DOE to compare the results of several possible solutions.
The team can develop a more effective final solution by
measuring effectiveness of the improvements in action.

D M A I C
3
DOE Terminology
Independent Variables–X’s
Also called factors Project Y
Factors or variables we select Dependent
in advance
Independent X
The causes (5 M’s and 1 P)
(x) (x) (x)

Dependent Variables–Y M M M
Also called responses
Project Y
The quantity (Y) that we
measure to determine the P M M
impact of the X’s (x) (x) (x)
The effect
Remember that the dependent variable (Y) is measured

from the customer’s perspective. DOE allows us to
identify the X’s that are the key drivers of variation in
the Y.

D M A I C
4
DOE Terminology (continued)
Levels For example: In our Capital Logistics GB Case Study:

Y = On-Time Delivery of Shipments
The test X1 = Truck Size
settings for X Levels = 62 ft. & 66 ft. truck
X2 = Tire Type
Levels = Brand X & Brand Y
Main Effects For example: Is the On-Time Delivery different for:
a) 62 ft. trucks vs. 66 ft. trucks?
Differences between b) Brand X or Brand Y tires?
each factor level
Interactions For example: Is On-Time Delivery different when:

a) 62 ft. truck, Brand X tire
Differences between b) 62 ft. truck, Brand Y tire
two or more
factor level c) 66 ft. truck, Brand X tire
d) 66 ft. truck, Brand Y tire
combinations
Main Effects Interactions

Deviation from required
62 ft.
truck
delivery time.
delivery time.
66 ft.
truck
Y
X
62 66 X Y
Truck Size Tire Type
delivery time.
Y
X
X Tire Type Y

D M A I C
5
Strategy For Choosing The Appropriate Design
The Knowledge Line

Current State of Process Knowledge
Low High
Type of Design Main Effect Fractional Full Response

/Screening Factorials Factorials Surface
Usual # of Factors >5 4-10 1-5 2-3
Purpose:
Identify Main effect Some Relationships Optimal factor
critical factors interactions among factor settings
- vital few
Estimate Crude direction Some All main effects Curvature in

for improvement interactions and interactions response,
empirical models
Screening DOE
Step 7
Step 8 Optimization DOE
Types of Experimental Design Types of Experimental Design
Screening DOE–experiments which are used to screen Fractional Factorials–provide a middle ground. They
out the vital few X variables out of many at a very low are used to expand on the study of the vital few X
variables at a low cost.
cost.
Can study the main effects
Can study main effects
Can study selected (and few in number)
Can’t study the interactions
interactions
Optimization DOE–Experiments which are used to Full Factorials–are used to study the vital few X
determine the proper settings for each vital X. variables and all possible interactions. They can be
quite costly when the number of vital few X’s are
large.
Can study all interactions

D M A I C
6
DOE Benefits
Benefits of Design of Experiment (DOE)

Can be used to identify “Vital Few” sources of variation
Defines the relationship between the inputs & outputs
Allows you to measure the influence of the “Vital Few” variables on a
response variable
– Also allows you to measure the interactions between the “Vital Few” variables
Is more effective & efficient than testing one factor at a time
Minimizes the number of test runs you have to make to draw valid
conclusions about X & Y linkages
Provides knowledge of best set-up conditions of X’s for improved Y
performance
Statistical Design Of Experiment

Raises Your Batting Average
Design of Experiment is an approach for effectively &

efficiently exploring the cause and effect relationship
between numerous process variables (X’s) and a process
performance variable (Y).

D M A I C
7
The Process Of Experimentation
1. Define Project 4. Determine Solutions

Identify responses 5. Record Results
2. Establish Current 6. Standardize
Situation 7. Determine Future Plans
3. Perform Analysis
Identify factors
Choose factor levels
Select design
Randomize runs
Collect data
Analyze data
Draw conclusions
Verify results

D M A I C
8
1. Define Project (problems)

The mileage on my new car is not up to advertised standards. I want to
improve my car’s mileage.
2. Establish Current Situation (& state hypothesis)

I currently do not get the manufacturer’s advertised gas mileage on my
car
Some combination of speed, gas octane, and tire pressure will provide
me with the optimum gas mileage

D M A I C
9
The Process Of Experimentation (continued)
3. Perform Analysis
Dependent Variable–Gas Mileage (Y)
Identify Factors (independent variables)
– Tire Pressure (X)
– Octane (X)
– Speed (X)
Choose Factor Levels
Independent Variables
(X’s) (Factors) Level - Level +
Tire Pressure (psig) 30 35
Octane 87 92
Speed (mph) 55 65

D M A I C
10
3. Perform Analysis (continued)

Select Design
– Number of test levels (Number of factors) = 23 = trials
Tire Pressure Octane Speed Gas Mileage

- - -
+ - -
- + -
+ + -
- - +
+ - +
- + +
+ + +
How many trials would be necessary if there were 4

variables each at 2 levels? How many variables would
there be before the cost became prohibitive?
What would be a different type of experiment that would

not require all possible runs?
The experimental trials tables represent all possible

combinations of the 3 factor levels.
Note: The table above shows the “standard order” for a

2-level, 3-factor design.

D M A I C
11

Randomize Runs
Tire Pressure Octane Speed Gas Mileage

- (30) - (87) - (55) 26
+ (35) - (87) - (55) 27
- (30) + (92) - (55) 30
+ (35) + (92) - (55) 33
- (30) - (87) + (65) 18
+ (35) - (87) + (65) 21
- (30) + (92) + (65) 19
+ (35) + (92) + (65) 22
What effects (main and interaction) seem to be significant?

– Main effect is the effect of one X on the Y
– Interaction effects are the combined effects of two or more
X’s on the Y
In this experiment, only one run of each experimental

trial was conducted. Additional runs would increase the
confidence level for the results.
Replication is a repeat of all experimental trials to obtain

additional data to increase the degree of belief in the
experimental results. Replications are used when:
The interactions are of critical importance

Data is tricky to get. Replications will supply an extra
data point when data from an experimental trial is lost
You need to increase the degree of confidence in the
experimental results
You need to reduce the risk when implementing
solutions

D M A I C
12

Collect & Analyze Data
Pareto Chart of the Effects
(Response is Mileage, Alpha = .05)
B
A: Pressure
A
B: Octane
BC
C: Speed
AC
ABC
AB
0 1 2 3 4 5 6 7 8 9
Effect Size
What factor has the biggest effect on gas mileage?
The Pareto Diagram bars show the relative importance of

each main effect and interaction in the experiment.
The length of each bar shows relative impact. Bars that

cross the dashed line have statistically significant impact
on the Y.
The dashed line is called the “reference line”. Any effects

that extend past this reference line are statistically
significant within our chosen confidence level.

D M A I C
13

Collect & Analyze Data (continued)
Main Effects Plot (data means for Mileage)
- + – + – +
28
26
Mileage
24
22
20
30 Pressure 35 87 Octane 92 55 Speed 65
What is the best speed to get the best gas mileage?
The Main Effects Plot show the average value for each of
the levels of the three independent variables. Each
average is based on 4 data points.
Extra credit: Can you look at the experimental trials table

and say why each level has a total of 4 data points?

D M A I C
14

Collect & Analyze Data (Continued)
Cube Plot (data means) for gas mileage

19 -, +, + +, +, +
22
Pressure, Octane, Speed
-, +, - +, +, -
30 33
+
-, -, + +, -, +
18 21
Octane +
Speed
– 27
–
26
– +
-, -, - Tire Pressure +, -, -
Each corner of the cube represents an experimental trial from the trial table
For example, when tire pressure is negative (-), octane is negative (-) and
speed is negative (-), the gas mileage is 26 (x,y,z is the convention)
Can you tell which combination of tire pressure, octane and speed is the
best by looking at the cube?
The cube shows the result of the interaction between tire

pressure, octane and speed.

D M A I C
15

Draw Conclusions
- Speed has an important effect on gas mileage
- Tire Pressure doesn’t significantly affect gas mileage
- Octane doesn’t significantly affect gas mileage
Verify Results
- Run additional trials at optimum settings
- Conduct MSA on data results
4. Determine Solutions
Drive at 55 mph to get the best gas mileage
Set tire pressure for best tire wear results
Buy the octane that optimizes cost or engine cleanliness
5. Record Results
6. Standardize
7. Determine Future Plans
Any ideas on how the results from this experiment could

be used for further experimentation?
Replicate the experiment to get a better indication of
the effects and to quantify variability
Test beyond the original limits of the X
Test at midpoint for nonlinear Y responses
Try additional potential X’s
Block by season, tire type, fuel type, etc.

D M A I C
16
Full Factorial DOE

D M A I C
17
Full Factorial DOE
Full Factorial Design of Experiment

Is used to study the Vital Few X variables and all possible
interactions
Can be quite costly when the number of Vital Few X’s is large

D M A I C
18
Full Factorial Layout: 3 Factors, 2 Levels
Std. Factor Factor Factor

Order 1 2 3
1 – – –
2 + – –
3 – + –
4 + + –
5 – – +
6 + – +
7 – + +
8 + + +
For 3 factors, each at 2 levels, there are 23 = 2 x 2 x 2 = 8

combinations of factor settings. (Levels Factors = Combinations)
Notice the pattern of factor settings in the standard order.
This is a 3–factor, 2–level full factorial layout arranged in Does this full factorial layout look familiar?
“standard order.”
Where did you see this before?
Each row of the full factorial layout represents a set of
experimental conditions. There are 8 different
experimental conditions in this design layout.

D M A I C
19
Visualizing The Experimental Space
2 Factors = A SQUARE +
Factor
2
–
– +
Factor 1
3 Factors = A CUBE
Factor 2
+
– –
– Factor 3
+
Factor 1
Review how the square is constructed. To add a third

factor, you need a three–dimensional cube plot.
A cube helps us visualize the experimental space covered

by the 3 factors.
Each corner represents 1 set of experimental conditions.
23 = (Two Levels)(Three Factors) = 8 experimental

conditions.

D M A I C
20
The Factorial Pattern Of Experimentation
k Factors Order X1 X2 X3 X4 X5 • • Xk
•
1 – – – – –
k=1 2 + – – – –
3 – + – – –
k=2 4 + + – – –
5 – – + – –
6 + – + – –
7 – + + – –
k=3 8 + + + – –
9 – – – + –
10 + – – + –
11 – + – + –
12 + + – + –
13 – – + + –
14 + – + + –
15 – + + + –
k=4 16 + + + + –
17 – – – – +
18 + – – – +
19 – + – – +
20 + + – – +
21 – – + – +
22 + – + – +
23 – + + – +
24 + + + – +
25 – – – + +
26 + – – + +
27 – + – + +
28 + + – + +
29 – – + + +
30 + – + + +
31 – + + + +
k=5 32 + + + + +
•
•
•
k 2k
The number of trials = (2 levels)(k factors)s = 2k .
Full factorial layouts can be generated for any number of

factors.
This table displays factorial designs arranged in standard

order.
Notice the pattern. A handy way to remember is by

thinking how the levels alternate in the columns.
For the first column, the levels alternate every 20 = 1 trial.
For the second column, the levels alternate every 21 = 2

trials.
For the Kth, the levels alternate every 2K-1 trials.

D M A I C
21
Replication vs. Repetition
Replication: Multiple execution of all or part of the experimental process

with the same factor settings
– Build different experimental units
– In our car mileage example, replicate different experiments at a different time
Repeat tests: 2 or more observations that have the same levels for all
factors
– Performed on the same experimental unit
DESIGN LAYOUT EXAMPLE
2 2
2 design with 2 replications 2 design with 2 repetitions
Std. Order Response Std. Order Obs 1 Obs 2 Avg
1 - - y1 1 - - y11 y21 y1
2 + - y2 2 + - y12 y22 y2
3 - + y3 3 - + y13 y23 y3
4 + + y4 4 + + y14 y24 y4
1 - - y5
2 + - y6
3 - + y7
4 + + y8
Replication has 2 important properties: Why Replication?

It allows an experimenter to obtain an estimate of To measure experimental variability.
experimental error. This estimate becomes the basic
unit of measurement for determining whether observed So we can decide whether the difference between
differences are statistically significant. responses is due to the change in factor levels (an induced
special cause) or to common cause variability. To see more
It yields more precise estimates of factor effects clearly whether or not a factor is important.
Repetition Replication provides the opportunity for factors that are

unknown or uncontrollable to balance out. Along with
May affect the precision of the measured response, but
randomization, replication acts as a bias–decreasing effect.
it does not affect the logical structure of the experiment

D M A I C
22
Randomization: The Experimenter’s Insurance
Definition of Randomization: To assign the order in which the

experimental trials will be run using a random mechanism.
– It is not the standard order
– It is not running in an order that is convenient
– Minitab will randomize for us
Why Randomization?
– Averages the effect of any lurking variables over all of the factors in the
experiment
What is a Lurking variable?

– A variable that has an important effect and yet is not included among the
factors under consideration
How and why should we randomize the order in which Safeguard:

we conduct an experiment? Randomize the order of the experimental trials to
protect against the effect of lurking variables
Randomization reduces the introduction of a systematic
Action:
bias into an experiment.
If the lurking variable creates a trend, it can be
We call this bias a “lurking variable.” compensated for in the numerical analysis
Conclusions can then be drawn from the original
Lurking Variables: factors that are not affected by such lurking
variables
Definition
A variable that has an important effect and yet is not
included among the factors under consideration
because:
– Its existence is unknown
– Its influence is thought to be negligible
– Data on it is unavailable

D M A I C
23
Design Of Experiment
Summary
Provides process performance data under different types of operating
parameters
Establishes cause and effect relationships between a process Y and
possible X’s:
– Verify the Vital Few X’s
– Determine X values for optimal performance
– Predict future process performance
Can be used in both incremental and exponential process improvement
studies
Caution: Prior To Implementing Solutions, Understand The

Difference Between Experimental And Actual
Business/Process Conditions

D M A I C
24
Reference Material–
Analyzing DOE Data
The following pages are reference only and will become

helpful if you are contemplating the use of a DOE in your
project. Sit down with your mentor after you review this
section and determine feasibility.

D M A I C
25
1. Define Project 4. Determine Solutions

Identify responses 5. Record Results
2. Establish Current Situation 6. Standardize
3. Perform Analysis 7. Determine Future Plans
3.1 Identify factors
3.2 Choose factor levels
3.3 Select design
3.4 Randomize runs
3.5 Collect data
3.6 Analyze data Focus of this module
3.7 Draw conclusions
3.8 Verify results
For review: The steps in Designing an Experiment.
In this module, we will focus on Step #3–Perform

Analysis. We will investigate how you analyze the data
collected from the experiment.
You saw in the car gas mileage example how we

analyzed the data from that full factorial experiment.
Now we will use the Capital Logistics Case Study on

“On-Time Delivery” to see how they used Design of
Experiment to identify the vital few X’s.

D M A I C
26
Steps in Analysis: Full Factorial, Replicated Designs
3.6 Analyze Data:
Diagnostics: 3.6.1. Plot the raw data. Minitab will

Is data OK? provide this
3.6.2. Plot the residuals*. for you.
3.6.3. Examine factor effects.

Analysis: 3.6.4. Confirm impressions with
Make inferences statistical procedures.
3.6.5. Summarize conclusions.
This is the “Analyze Data” step, which is one step in the After using Minitab to check the data, the next step is to
process of experimentation on the previous page. plot the residuals.
Steps in analyzing the data: Residuals are the “leftover” variation in the data after you
have accounted for the main cause of variation. Plotting
the residuals should show a normal distribution. If the
In steps 1 & 2, we begin by checking for problems with
residuals are not normally distributed, it is a sign that
the data before doing the DOE analysis on the data.
there is another variable present that has not been
Minitab will plot the data for you. This will help to: accounted for (a “lurking” variable).
Find defects, outlier or patterns in the data
Get a feel for the actual data * We don’t discuss residuals here, but to continue to Step
3, seek assistance from someone who knows how.
Get impressions of what the data is going to tell us, to
guide our analysis
Once the data has been validated and looks acceptable,
you are ready to examine the factor effects.
Let’s look at our Capital Logistics GB Case Study.

D M A I C
27
Background For DOE: Capital Logistics GB Case Study
In Step 6 of the Analyze Phase, the Green Belt and the project team at Capital Logistics determined,
based on the results of their Hypothesis Testing, that there were 2 X’s that had an impact on the on-
time delivery of shipments to customers. The 2 X’s were:
– Drivers
– Distance
The team determined that there wasn’t anything that they could do to change the distance that each
truck traveled to make their deliveries–the routes were automated by the system and were
determined to be the shortest routes. They decided to focus on Driver as a factor, as this was under
their immediate control.
The team used the Cause & Effect diagram to brainstorm potential root causes as to why
Driver/Trucking company influenced Time. The resulting list included:
– size of the vehicle – type of tire
– dispatch method – fuel capacity
– type of engine
After reviewing their data further, the team decided to conduct a Design of Experiment. They would
use Size of Vehicle and Tire Type as possible Vital X’s in their DOE. Each of these factors has 2
levels:
– Size of Vehicle is either 66 or 62 feet
– Tire Type is either X or Y
The DOE was set-up and conducted to determine the Vital X that would have the greatest impact on
the timeliness of shipment deliveries to their customers.

D M A I C
28
Examine Factor Effects
The 3rd Step in “Analyze Data”:

Y = f (X1, X2, X3, …, Xn)
Y (Response) = On-Time Delivery of Shipment
Xi (Factors) = Size of Truck, Tire Type
How do the factors affect the response?

How do the combinations (interactions) of factors affect the response?
– We can write the equation that answers these questions
This is your Y = f(X) Transfer function: On-Time Delivery = Constant +(-)

Size of Truck + Tire Type +(-) Size of Truck *Tire Type Interaction Effect.
We can simplify the factor names with abbreviations in

our DOE (see next page).

D M A I C
29
Factor Names And Interaction Symbols
It is useful to abbreviate factor names. There are several ways

to do this:
– Choose the first letter or first two letters of each factor
– Assign factors a letter: A, B, C, D, . . .
Denote the interaction between Factor A and Factor B as:
– A x B or AB
For two factors (A and B) there is one interaction: AB
For three factors there are
– Three two-way interactions: AB, AC, BC
– One three-way interaction: ABC
How many interactions are there for four factors (A,B,C,D)?
Interactions for 4 factors:
2-way: AB, AC, AD, BC, BD, CD, (6)
3-way: ABC, ACD, BCD, ABD (4)
4-way: ABCD (1)
Total (11)
Higher level interactions (3-way or more) are hard to plot

and explain. They are also not very likely to occur.

D M A I C
30
Examine Factor Effects–In Minitab
Create Factorial Design–in Minitab Data File: GB Case Study.mtw
Type of design = full factorial

D M A I C
31
Select Two Replicates

D M A I C
32
Name Factors
Stat > DOE > Create Factorial Design

Click on Factors.
Enter “Name”.
Click “OK” twice.

D M A I C
33
Inputting Responses into Minitab
Note: Ensure that when you

enter this data into your Minitab
Worksheet, you put the times in
the correct rows, based on the
correct truck & tire combination
of (+) and (-).
Note: Ensure that when you enter this data into your
Minitab Worksheet, you put the times in the correct rows.
Your experiment may have run in a different order and
you may have coded your factors differently.
Truck Size Tire Size Time

66 X (+) 126
(+) 66 X 140
66 Y (-) 30
66 Y 60
62 X (+) -50
62 X -80
(-) 62 Y -30
(-)
62 Y -50

D M A I C
34
Analyze Factorial Design

In Minitab: Stat > DOE > Analyze Factorial Design

D M A I C
35
Analyze Factorial Design

D M A I C
36
What does the Pareto Chart tell you about the factors &
interactions?
Is Truck Size significant?
Is Tire Type significant?
Is the interaction of Truck Size & Tire Type significant?

D M A I C
37
Examine Factor Effect–In Minitab
Analyze Factorial Plots

D M A I C
38
Set-up Main Effects
Stat > DOE > Factorial Plots

Click OK.

D M A I C
39
Set-up Main Effects
Stat > DOE > Factorial Plots

Click OK twice.

D M A I C
40
Main Effect Plots
In our GB Capital Logistics On-Time Delivery Case Study:
Main Effects Plot (data means) for Time
-1 1 -1 1
90
60
30
Time
-30
How do Truck Size & Tire Type affect Time?
A useful way to graphically show factor effects is by

using Main Effects Plot.
Simply plot the average response for each level (+ or -).

The steeper the line, the more significant the effect. The
slope of the line determines whether the effect is positive
or negative. A positive slope is a positive effect and vice
versa.

D M A I C
41
Interaction Effect
Differences between 2 or more factor level combinations

In our Capital Logistics GB Case Study: On-Time Delivery
If we are looking at 2 factors: Truck Size
Tire Type
Is On-Time Performance different when:
62 ft. X
62 ft. Y
66 ft. X
66 ft. Y
Question: Which of the following indicates Interactions between

factors?
–
–
Factor A Factor A + Factor A
Response
Response
Response
+
+ –
– + – + – +
Factor B Factor B
Factor B
Adapted from Lawson, John and John Erjavac, Basic Experimental Strategies
and Data Analysis. Provo, UT: Brigham Young University, p. 104.
We use the same calculation for Interaction Effects as we If the lines are not parallel, there is interaction between
did for Main Effects. The only difference is we first need the factors (variables).
to generate the interaction column. We then use this
column for the calculation of the Interaction Effect.
Why are interactions important?
Why is it important to determine the effect of

interactions?

D M A I C
42
Interaction Effect
In our GB Capital Logistics On-Time Delivery Case Study:

Interaction Plot (data means) for Time
Truck Size
-1
1
100
Late
50
Mean
On-Time
0
Early -50
-1 1
Tire Type
How do Truck Size & Tire Type interact?

How do the interactions affect time?
What does this tell us?
In this case, our goal was On-Time-Delivery. None of the

points give us OTD for mean=0. Therefore, we should go
with the (-) Tire Type and the (-) Truck Size (X Tire
Type and Truck Size of 62.) This will get us closest to a
mean of zero.

D M A I C
43
Steps In Analysis: Full Factorial, Replicated Designs
Diagnostics: 3.6.1. Plot the raw data.

Is data OK? 3.6.2. Plot the residuals.


D M A I C
44
Confirm Impressions
The 4th Step in “Analyze Data”:

By now you have a good idea about the influence of the factors in
the experiment. You can confirm these impressions statistically by
performing a Hypothesis Test.
Ho: Factor has no effect on the results
Ha: Factor has an effect on the results
p α: Accept Ho
p α: Reject Ho
Statistical Analysis Confirms The Results

Of The Graphical Analysis
Now we are determining the statistical significance of the

effects.

D M A I C
45
Confirm Impressions With Minitab
In our GB Case Study: What does the Minitab Session

Window tell you ?
What do the p-values for:

Truck Size
Tire Type
Truck Size * Tire Type
Indicate?
Does this statistical analysis confirm the graphical

analysis?

D M A I C
46
Steps In Analysis: Full Factorial, Replicated Designs
Diagnostics: 3.6.1. Plot the raw data.

Is data OK?
3.6.2. Plot the residuals.


D M A I C
47
Summarize Conclusions
The 5th Step in “Analyze Data”:

List all the conclusions you have made during the analysis
Interpret the meaning of these results
– For example, relate them to known physical properties or an expert’s personal
knowledge of the process
Make recommendations
Formulate and write conclusions in simple language
What conclusions would you make from the Capital

Logistics DOE study?

D M A I C
48
Summarize Conclusions (continued)
From our Capital Logistics GB Case Study on On-Time Delivery:

The Green Belt and the project team reviewed the results of the DOE. They
saw that there seemed to be a link between on-time performance and the size of
the truck unit, as well as an interaction effect between Tire Type and Size of
Truck on on-time performance.
They returned to their Fishbone (Cause & Effect) Diagram with greater
granularity, and asked why these factors would influence on-time performance.
They validated these reasons through field observations.
The team validated that the larger size trucks took longer at some locations. In
addition to the time involved with maneuvering in a cramped environment, on
certain routes these trucks had a higher incidence of tire failure (lateral
movement caused by tight turns increased the stress on the tires, increasing the
number of flat tires).
By using the data & results from the DOE, the team had determined the root
causes of the on-time performance problem and were ready to start to
brainstorm solutions to address/control these “Vital X’s”.

D M A I C
49
Reducing the Size of Experiments–

Fractional Factorial DOE

D M A I C
50
Reducing The Size Of Factorial Experiments
Many factors potentially impact

the quality of any Number Number
process/product of Factors of Runs
The factorial strategy is an
1 2
efficient approach to
2 4
experimentation, compared to For a 2–Level 3 8
“one-at-a-time” 4 16
When factors are investigated 5 32
at 2 levels, the number of Factorial 6 64
experimental runs is 2k (note: 7 128
the “k” needs to be raised, with k Factors….. 8 256
because it is 2 to the power of k 9 512
- “k” is the exponent) 10 1024
Number of runs • •
This can result in a large = 2k • •
number of runs, even with a • •
relatively small number of 15 32,768
factors • •
• •
Limited resources allow you time to run only 4 trials • •
in the allotted time. 20 1,048,576
Which 4 trials will you choose?
Full factorial is time-consuming and costly as the number From a Full Factorial (4 factors):
of factors increase.
Number
Also consider the number of 2-way and higher order
interactions as the number of factors increases. Overall Average 1
For 4 factors, let’s look at every possible effect (main & Main Effects: A B C D 4
interaction) and the overall average. Evaluating the
average is what the extra run is used for. 2-way interactions 6
(AB, AC, AD, BC, BD, CD)
3-way interactions 4
(ABC, ABD, ACD, BCD)
4-way interactions 1
(ABCD)
D M A I C
51
Fractional Factorials
Fractional Factorials*:
Provide a middle ground for design of experiment
Are used to expand on the study of the Vital Few X variables at
a low cost
Can study selected (and few in number) interactions
Fractional Factorials Allow You To Reduce The Size

Of An Experiment And Select Certain Runs
*Fractional Factorials are beyond the scope of this

course. Consult your BB/MBB or refer to the “Book of
Knowledge” for further information.

D M A I C
52
Strategy For Choosing The Appropriate Design
The Knowledge Line

Current State of Process Knowledge
Low High
Type of Design Main Effect Fractional Full Response

/Screening Factorials Factorials Surface
Usual # of Factors >5 4-10 1-5 2-3
Purpose:
Identify Main effect Some Relationships Optimal factor
critical factors interactions among factor settings
- vital few
Estimate Crude direction Some All main effects Curvature in

for improvement interactions and interactions response,
empirical models
Screening DOE
Step 7
Step 8 Optimization DOE
Types of Experimental Design Types of Experimental Design
Screening DOE–experiments which are used to screen Fractional Factorials–provide a middle ground. They
out the vital few X variables out of many at a very low are used to expand on the study of the vital few X
cost. variables at a low cost.
Can study the main effects Can study main effects
Can’t study the interactions Can study selected (and few in number)
interactions
Optimization DOE–Experiments which are used to
determine the proper settings for each vital X. Full Factorials–are used to study the vital few X
variables and all possible interactions. They can be
quite costly when the number of vital few X’s are
large.

D M A I C
Optimization Design Of Experiment Activity (40 minutes) 53
Desired Outcome: Practice how to set-up a DOE
What How Who Timing
Team Choose a facilitator, scribe, Team 2 mins.

Preparation timekeeper, and/or note taker
Determine the timing for each
activity below
Agree on the example you will use
for the exercise
– Your own project
– How to get a great lawn
– How to make the best chocolate
chip cookies
– How to lose weight
Set-up the For the example you selected, Facilitator 15 mins.

DOE identify 3 factors at 2 levels (+/-)
each
Using the worksheet on the following
pages, design the experiment up
through Step 6
Set-up Discuss what the experiment might Facilitator 15 mins.

Optimization demonstrate, were you to execute it
DOE Brainstorm your list of questions for
how to run this experiment
Report Choose a spokesperson to report All 8 mins.

Out out on the design of your
experiment, and the questions you
have about running it

D M A I C
Design Of Experiments–Breakout Activity Worksheet 54
1. Define the problem
2. State the hypothesis
3. Identify the dependent (Y) and independent (X) variables
Dependent Variable (Y) =
Independent Variables (X’s) =
4. Determine the test levels for each independent variable (X).
Independent Variables
(X’s) (Factors) Level - Level +
5. Calculate the number of trials to test all combinations of test levels.

# of independent variables
# of trials = # of test levels

D M A I C
55
Design Of Experiments–Breakout Activity Worksheet (Continued)
6. Construct the experimental trials table
Independent Variable (X) Independent Variable (X) Independent Variable (X) Dependent
#1 #2 #3 Variable (Y)
Independent Variable (X) Independent Variable (X) Independent Variable (X) Dependent
#1 #2 #3 Variable (Y)

D M A I C
56
Appendix: Example Of Designed Experiment: GE Card Services

D M A I C
57
DOE Example: Defining The CTQ
Overall
Quality
EX/VG - 57%
.26 .25 .11 .12 .16 * * .15

Financial & New Account Card Billing Payment Customer
Authorizations Marketing Rep
Marketing Info Approval Production Statements Processing Service
EX/VG - 60% EX/VG - 54%
EX/VG - 30% EX/VG - 34% EX/VG - 47% EX/VG - 40% EX/VG - 54% EX/VG - 44%
Impact Performance Impact Performance
Overall 18% 16% Overall 13% 34%
0.38 Timeliness
0.59 Speed Of 10% 33%
24% 25% Of Delivery
Approval 0.28
0.23 Accuracy 20% 41%
Amount/
18% 26% 0.24
Type Of Info Insert
12% 47%
0.2 Credit Lines 0.17 Capability
16% 25%
Granted Card Quality 23% 53%
1 0.8 0.6 0.4 0.2 0 1 0.8 0.6 0.4 0.2 0
0% 20% 40% 60% 80% 100% 0% 20% 40% 60% 80% 100%
Excellent Very Good Excellent Very Good
Inseparable CTQ’s
#1 Speed #1 Speed
One Experience To The Customer #2 Accuracy
Speed Is The #1 CTQ Of The New Account

And Card Experience
A project team was working to improve customer When examining the process map, it became clear that
satisfaction with the process of approval for and issuing many faxed and mailed applications needed additional
of new credit cards. information before they could be processed. This often
involved phone or written correspondence with the
One of the key drivers of satisfaction was identified as applicant, which slowed the process considerably.
time to issue cards, i.e., the cycle time between
a consumer’s approval for a credit card, and the time
he/she received it.
Through segmentation, the team found that applications

mailed or faxed in took considerably longer than other
applications where the application information was
entered electronically at the store.

D M A I C
58
DOE Example: Causes Of Time Consuming Delays
79% of queuing for non-fraud! Why?

Non-Fraud Queue Pareto
50%
40%
Frequency %
30%
20%
10%
0%
Msg Wrng Msg/Wrng Msg/ Unknown CBR Msg/ Duplicate
Home Promo Brnch/Dlr/ Wrng Issues Wrng
#/AC Cd Str # SS# Zip
Missing and illegible information is a major cause of delays.

What can be done to reduce missing and illegible information?
Many felt that little could be done to improve the situation
for handwritten applications.
The team discovered that 79% if the applications which

were “in queue”–i.e., on hold before being processed –
were in queue for reasons other than potential fraud
issues. The team analyzed the reasons for non-fraud
queuing and created a Pareto chart.

D M A I C
59
DOE Example: The Designed Experiment
Factor Name Explanation Standard Level New Level

App. Type Terms of the credit agreement Present (Full) Absent (Naked)
printed on the application
Instructions Written instructions on how to No Yes
complete the form included
Flags Critical information without which No Yes
the application can’t be processed
is specially noted
Boxes Delineated boxes to hold numeric No Yes
information, e.g., phone
( )- -
Insurance Location on the form for the purchase Not Body* Body*
of the optional credit card insurance
Page Layout Orientation of the form Portrait Landscape
The only influence that the business has on legibility and

completeness is through the application itself. The team
then decided to study what they could do to the
application to make it easier for their customers to fill
out.
The factors shown above were chosen for study.
*Body/Not Body–refers to location of the insurance text

on the application form. On a landscape layout, “not
body” applications had text to the right of the application.
On a portrait layout, “not body” applications had text at
the bottom of the application.

D M A I C
60
DOE Example: Experimental Complications
Complications The difficulty in tracking many (in this case, 16)

different applications through the process
Large amount of person-to-person variation with
respect to legibility and completeness, making it
difficult to assess the effect of changes
Remedies Test the applications outside the normal process

Have each tester fill out two applications
The applications were paired (eight pairs of applications) The completed applications were sent for imaging, and
and customers were asked to fill out a pair of applications the imaged applications were reviewed by an associate
and provide feedback regarding the two application to record the legibility and completeness of each field
versions. The applications were paired so that customers on each application. A legibility score was assigned
experienced the factors at both levels. using a 5-point scale; missing information was given a
“0” and fields not relevant to the applicant were
Eighty customers filled out a pair of applications in return recorded as not being applicable. Preferred applications
for a $10.00 gift certificate. The order which each paired were given a score of “1,” not preferred “0,” and no
application was presented to the customer was preference was indicated by a “0.5.”
randomized throughout the test to balance the effect of
order on the results. Customers were asked whether they
preferred one application version over another and why.
Applicants signing for credit insurance were asked
whether they understood that they had requested
insurance coverage.

D M A I C
61
DOE Example: Responses
Two Key Responses Were Studied:

A combined completeness and legibility score, which was
based on the percent of necessary information that was both
present and legible
A preference score, which told which of the two applications the
customer preferred

D M A I C
62
DOE Example: Analysis
Main Effects Plot – Means For Legibility And Completeness
0.849
Completeness
Legibility and
0.843
0.837
0.831
0.825
Naked Full No Yes No Yes No Yes Body Bottom Landscape
Portrait
App. Type Instruct Flags Boxes Insurance Pg. Layout
* * *
Main Effects Plot – Means For Preference
0.67
0.59
Preference
0.51
0.43
0.35
Naked Full No Yes No Yes No Yes Body Bottom Landscape Portrait
App. Type Instruct Flags Boxes Insurance Pg. Layout
* Effect was significant

* * * * *

D M A I C
63
DOE Example: Results
Effect On Legibility Effect On

Factor And Completeness Preference Recommendation
Application Full Naked Full Naked
Type 82.8% 84.8% 34% 66% Naked
Instructions H H No Yes
57% 43% Yes1
Flags No Yes H H
82.9% 84.8% Yes
Boxes H H No Yes
59% 41% No
Insurance Not Body Body Not Body Body

85.3% 82.4% 56% 44% Not Body
Page H H Portrait Landscape

Layout 36% 64% Landscape
1 The data does not show that instructions improve legibility and completeness, but it is felt that instructions should be provided for clarification
purposes. Also, consideration should be given to improving the effectiveness of the instructions provided on the application.
+ Effect not significant

D M A I C

D M A I C
1
Step 9
Step 8 provided the experimental techniques to establish the

relationship between the measurable Y characteristics and the
controlling X factors. In step 9, those relationships (transfer
functions) wil be used to define the key operating parameters and
tolerances to achieve the desired performance of the CTQ’s/ One
should be able to set the tolerance of X factors if the X-Y
relationship and the specifications of Y are given.
DMAIC GB T TX PG V 4.2.0
D M A I C
2
Principle Of Tolerancing
Tolerance: the allowable range of variation of X while still meeting

the requirements of Y.
Establish tolerance of X based on the requirements of Y, or
often the specification limits via the Transfer function
The tolerance should allow the project to reach the project
objectives established in step 5
When more than one CTQ is involved, be aware of the trade-
offs among CTQ’s
Be aware of the variations due to the measurement of X’s and
Y’s Y
USL
T Tolerance:
LSL the allowable
Y = f (X) range of X
X
Lower Limit Upper Limit
of Tolerance of Tolerance
It is complicated to calculate tolerance for ƒ (X) with

multiple X’s.
Use:
Crystal Ball
Calculus
It will be covered in detail in the GE-DFSS course
D M A I C
3
Consideration Of Measurement Variations
USL
The measurement T New tolerance

variation of Y
considering
LSL Y = f (X) the variation
of Y
measurement
X
Lower Limit Upper Limit
USL
The measurement T New tolerance

variation of Y
considering
LSL Y = f (X) the variation
of X + Y
measurement
The measurement
variation of X Lower Limit Upper Limit
Estimate the gage error for Y using the σe from the MSA
study in Step 3.
The width of the measurement error distribution is 5.15σe
Establish the gage error for X in Step 10. Use the 5.15σe
from the MSA of X.
D M A I C
Example: Weight Loss Wellness Program 4
lbs.
week
USL
3
Target
Weight 2 Transfer
Loss Function
1
LSL
0
1 1.4 2 2.7 3 hrs./week
CTQ requirements: 4
Target to lose 2 lbs/week
At least need to lose 1 lb/week
No more than 3 lbs/week due to health concerns
These goals do not take into consideration any error in

the Y or X due to measurement error (MSerror)
Since we have performed a Gage R&R on both the Y
and X measurement system we know the GRR %
Tolerance for Y and X
Lets use the information to develop the operating limit
for our actual X
D M A I C
5
Example: Weight Loss Wellness Program (continued)
lbs.
The measurement
variation of Y week
USL
3
Target
Weight 2
Loss
1
LSL
New operating limits
on X due to variability
0 in the Y and X
measurements
1 2 hr/week
Exercise
1.55 2.55
To get a more conservative tolerance for X, taking into

account measurement error (MSerror) on Y, we center the
width of the 5.15 MSerror distribution on both the USL
and LSL for Y.
Q: Why center the measurement error distribution on the

spec?
A: The error in measurement system will effect us most

at the ends of the tolerance
Q: Why start from within the tolerance?
A: This results in a more conservative tolerance for the X

factor operating within. The new operating limits of X
will ensure you stay in spec for Y.
D M A I C
Example: Weight Loss Wellness Program (continued) 6
The measurement lbs.

variation of Y
week
USL
3
Target
Weight 2
Loss
1
LSL
0
1 1.5 2 2.6 3 hrs./week
Exercise
New operating limits on X due to
variability in the Y measurements
How to Determine the values of the new Y start points

Given Y Tolerance = 2; GR&R = R&R % Tolerance =10%
1. Calculate MSerror in Y units
MSerror = (R&R % Tolerance) ‫( ٭‬Tolerance)
MSerror = (10%) ‫( ٭‬2) = .2
2. Take ½ MSerror
(½) ‫( ٭‬MSerror) = ½ (.2) = .1
3. Calculate new start points for Y axis
New Y USL start point
Y USL- (½ MSerror) = 3 - .1 = 2.9
New Y LSL start point
Y LSL + (½ MSerror) = 1 + .1 = 1.1
Our purpose is to get an operating tolerance on the X but

we first should adjust the Y tolerance due to
measurement error.
D M A I C
Example: Weight Loss Wellness Program (Continued) 7
The measurement lbs.

variation of Y
week
USL
3
Target
Weight 2
Loss
1
LSL
0
1 1.5 2 2.6 3 hrs./week
Exercise
New operating limits on X due to
variability in the Y measurements
Given: X is R&R % Tolerance = 20%

New X Tolerance = USL – LSL (from last page)
New X Tolerance = 2.6 –1.5 = 1.1
Calculate MSerror in X units
1. MSerror = (20%) ‫(٭‬1.1) = .22
2. Take (½) ‫(٭‬MSerror) = (½) ‫(٭‬.22) = .11
3. New X USL = USLx - (½ (MSerror))
= 2.6 - .11 = 2.49
New X LSL = LSLx - (½ (MSerror))
= 1.5 + .11 = 1.61
Correct X USL and LSL for the MSerror of the MSA

system used to measure X using same methods as
with Y.
This has the effect of further tightening the X
Tolerance
If you operate in these new operating limits of X you
should be ensured of staying in spec on Y. That means
no defects.
D M A I C
8
Summary
Set the process tolerances based on product specifications

(exercise time needed to achieve desired weight loss goal).
Adjusted our process tolerances to account for product
measurement variation
The tolerance principle is relatively simple and the application is
relatively straightforward when:
1) The measurement variation of of Y and X are sufficiently small
to be ignored
2) Only one X factor involved
However, when measurement variations of Y and X needs to be
considered and multiple X factors are involved, the math techniques
could be more complicated. Courses on DFSS Statistical Design Method
or DFSS Statistical Tolerance are offered in various GE businesses.
D M A I C
Control Phase Flowchart 1
D M A I C
CONTROL
PHASE
OVERVIEW
Control 11: Control 12:

Control 10:
Determine Implement
Validate
Process Process Control
Measurement
Capability System & Project
System
Closure
DMAIC GB U TX PG V 4.2.0
D M A I C
Control Phase Overview 2
What is the Control phase?

The Control phase is when your team:
Ensures that the new process conditions are documented and monitored via
statistical process control methods
After a “settling–in” period, reassesses the process capability
Closes the project
Why is the Control phase important?

This phase is important because it ensures that process improvements are
implemented together with appropriate process controls to ensure that the
process changes are sustained into the future. It also is an opportunity for the
team to “prove” that the project process performance goal has been met.
Steps involved in the Control phase

Control 10: Validate Measurement System
Control 11: Determine Process Capability
Control 12: Implement Process Control System & Project Closure
D M A I C
3
The12-Step Process
Define
Measure
Standards Project Y
R&R
Analyze
Graphical Analysis,
Hypothesis Tests
Improve
Relationships
Tolerances
Control
Capability
D M A I C
Control 10–Validate Measurement System 4
What does it mean to Validate Measurement System?

In the Measure phase, you focused on validating the measurement system for the
Project Y. In this step, you should now validate the measurement system for the Vital
X’s. A Measurement System Analysis (MSA) is the overall process to test your data for
its validity. Before actually performing the MSA, you will have to develop a data
collection plan for the Vital X’s. You should focus on the Vital X’s at this point because
you want to develop Control Plans for each Vital X.
Why is it important to Validate Measurement System?

Validating the Measurement System is important because you need to have data that
you can trust in order to make decisions with that data. If your data is not telling you the
truth, you are likely to make a decision with that data that will add more variation to the
process. As part of your Process Control System, you will monitor the Vital X’s to make
sure they stay in control. Therefore, you must validate the Measurement system used
to collect the Vital X data.
What are the project tasks for completing Control 10?

10.1 Validate Measurement System
10.2 Determine whether Measurement System is adequate to measure Vital Few X’s
CONTROL STEP
OVERVIEW
Control 10: Control 11: Control 12:

Validate Determine Implement Process
Measurement Process Control System &
System Capability Project Closure
10.1 Validate Measurement System

10.2 Determine whether Measurement System is adequate to measure Vital Few X’s
D M A I C
4
Control 11–Determine Process Capability 5
What does it mean to Determine Process Capability?

Process Capability is the measure of process performance relative to the Customer’s
CTQ. It is measured again at this point in the process in order to see if there was a
process improvement. If a pilot is conducted, the improvement team should measure
process capability from the pilot and then, once full implementation takes place (after
Step 12), measure process capability again.
Why is it important to Determine Process Capability?

Process Capability is important at this stage of the project because you want to be able
to compare baseline performance before improvements to the new Process Capability
once improvements have been implemented.
It is also important to statistically validate the improvement in the process through the
use of a Hypothesis Test, such as the 2-Sample t-Test.

11.1 Evaluate Process Capability of the improved process
11.2 Confirm statistically that the improvement goal has been met
CONTROL STEP
OVERVIEW

11.1 Evaluate Process Capability of the improved process

11.2 Confirm statistically that the improvement goal has been met
D M A I C
Control 12–Implement Process Control Systems 6
What does it mean to Implement Process Control System?

A Process Control System is a system of activities whose purpose is to maintain
process performance at a level that satisfies customers’ needs and drives the ongoing
improvement of process performance. A Process Control System consists of three
parts: Documentation, a Monitoring Plan and a Response Plan. During this step of the
project, you are also ready for project implementation. Once the improvement is
implemented and the process improvement is verified, the final step of the project is
Project Closure. During Project Closure, you will finalize project documentation,
leverage learning to other areas of the business and celebrate project completion with
your team.
Why is it important to Implement Process Control?

A Process Control System will help maintain the gain that occurred with the process
improvement. Without control of an improved process, the process tends to revert to
its old performance.

12.1 Implement a Control strategy for each Vital X
12.2 Prepare Process Control plan
12.3 Implement a solution
12.4 Close the project
CONTROL STEP
OVERVIEW

12.1 Implement a Control strategy for each Vital X

12.2 Prepare Process Control plan
12.3 Implement a solution
12.4 Close the project
D M A I C
D M A I C
Collect data
8
Process Control Plan Worksheet
Documentation Monitoring Response Plan
Plan
PlanFor
ForDoing
DoingThe
TheWork
Work Checking
CheckingThe
TheWork
Work Response
ResponseTo
ToSignals
Signalsof
ofChange
Change
Process Step Procedure Key Process Method For Corrective Action Preventive Action
Monitoring
(From Process (From & Output Recording For Containment Procedure For Process
Standards
Map Attached) SOPs) Measures Data Improvement
Refer To Diagram Refers To Root Cause X’s For Each For Each Identifies Who Describes What Must Be Done When
That Illustrates Document That Measure, Measure, Should Do What We Fail Process Sigma Standard.
Project Y’s
Process Steps Describes How Describes Any Describes How With The
Describes What Must Be Done To
Separated By The Task Should Process Sigma Target, Numeric The Monitored Defective Output.
Prevent Special Causes That Worsen
Function. Shows Be Done, Or For Each Limits, Or Data Should Be
Describes What Performance Or To Incorporate
Transfers Between Refers To A Project Y Tolerances To Recorded, Who
Should Be Done These Special Causes That Improve
Functions, And Document That Which A Process Should Record
For Those Ill- Performance.
Which Function Is Describes The Should Conform The Data And
Served By These
Responsible At Step. How.
Defects.
Each Step.
D M A I C
Forms Of Intellectual Property 9
Patent
Best Protection against independent development by others
Gives right to exclude others from making, using selling, the invention or 20
years from filing
Can wait 2-3 years for patent to issue
Right granted by the government for new, useful, nonobvious inventions
Trade Secret
Rights exist only so long as actually kept secret
Protects against theft or misappropriation, not independent development
Does not require inventiveness in the patent sense
Copyright
Protects original works of authorship fixed in a tangible medium
Does not protect against independent development
Trademark
Defines source of goods or services
Arbitrary, fanciful marks get stronger protection
D M A I C
10
Intellectual Property Infringement Avoidance Clearance And Assessment
It is critical to assure that improvement or new designs developed in the course of the
project do not infringe on the Intellectual Property of others, external to GE. GE has a
clear policy on infringement avoidance. “Intellectual Property Infringement Avoidance
Process”. The team must conduct the necessary reviews to assure no infringement
(consult the IP Business Champion for guidance). In general, if the answer is “yes” to
any of the questions below, consult with the IP Business Champion and IP Designee
for your business to obtain clearance before proceeding.
1) Is the team aware of any non-GE businesses who use substantially the same process or
sell substantially the same process?
2) Will the project result in a product or process that is significantly different than a current GE
product or process, and will it be commercially significant?
3) Have or will any services or products been obtained from an outside party I connection with
this project, and is the project result commercially significant?
4) Will the product or process be exposed to non-GE parties, and be commercially
significant?
5) Will the product or process be exposed to the public, and be commercially significant?
If the project obtained clearance from the Infringement Avoidance screen, then
consider whether there is Intellectual Property to Capture and Protect, via Patent,
Trade Secret, Copyright or Trademark. If the project results in a process or product
that is new, non-obvious and useful and will be of significant commercial value. It may
be patentable. Consult the IP Business Champion and IP Designee for your business
and submit the new process or product for evaluation through the IDEAS website.
D M A I C
Statistical Process Control–

Overview
DMAIC GB V TX PG V 4.2.0
D M A I C
Statistical Process Control: Learning Objectives 2
Describe the two types of Control Charts

Describe and use Variable and Attribute Control Charts
Select the appropriate Control Chart for a given process based
on data type
Define the control limits and how they differ from specification
limits
Explain the decision errors in SPC
Explain the rules (Minitab and Western Electric) to determine
whether a process is out of control
Describe and practice the use of Minitab to generate I&MR,
X & R, U, C, P and NP chart
D M A I C
Statistical Process Control 3
“The Application of Statistical Techniques

for Measuring and Analyzing
the Variation in Processes”
Control Charts Are The Primary Tools In SPC
First developed by Dr. Walter Shewhart, control charts

serve as the best way to determine if a process is “in
control.”
By “in control,” we mean that the process is consistent

and predictable. When a process is out of control,
something about the process has changed. A cause for the
change in the process should be sought.
D M A I C
Five Main Uses Of Control Charts 4
To reduce scrap and rework and for improving productivity

Defect prevention. In control means less chance of
nonconforming units produced.
Prevents unnecessary process adjustments by
distinguishing between common cause variation and special or
assignable cause variation
Provides diagnostic information so that an experienced
operator can determine the state of the process by looking at
patterns within the data. The operator can then make the
necessary changes to improve the process performance.
Provides information about important process parameters
over time
D M A I C
5
Process States
The Four States of a Process
1. Chaos
Process out-of-control, producing non-conforming product
Even the level of nonconformance is unstable
Assignable causes dominate the output
Fixes don’t work for very long
How To Begin To Sort Out

The Problems To Be Solved?
D M A I C
6
Process States (continued)
2. The Brink of Chaos

Process unstable, some nonconforming product is being produced
Instability will continually change product characteristics
Process output is influenced by assignable causes
No assurance the next piece produced will be conforming
How To Determine Existence Of Assignable Causes?
D M A I C
7
Process States (continued)
3. The Threshold State

Process inherently stable over time, but producing some nonconforming
product (process sigma is less than process sigma goal/standard
Proportion nonconforming predictable
Some non-conformances will be shipped
If process natural spread is greater than the tolerance, common causes
must be reduced/removed
Process must be monitored to assure desired effect is achieved
How To Be Certain That The Process Has Improved?
D M A I C
Process States (continued) 8
4. The Ideal State

The process is inherently stable over time (process sigma > process
goal/standard)
Operating conditions are not changed arbitrarily (follow the process plan)
The process average is set and kept at the proper level
The natural spread of the process is less than the specified tolerance
How To Be Certain That None Of These Conditions

Change Or Degrade?
D M A I C
Variation And Control Charts 9
40
Upper Control Limit*
(UCL)
30
Measurement
20 Average
10
Lower Control Limit*

0 (LCL)
0 10 20
Time Order of Sample
Control charts are used to distinguish between common

and special causes of variation and use that understanding
to control and improve processes.
Control charts are characterized by two things:
1. The average (mean), or centerline, which represents

the middle point about which plotted measures are
expected to vary randomly.
2. Control limits, both upper and lower, which represent

the performance boundaries you can expect for the
process. Although measures vary, one would not
expect to see plotted measures outside of these
boundaries if the process operated predictably.
*Note: Minitab displays UCL as 3.0SL (Three sigma

limits) Minitab displays LCL as–3.0SL
D M A I C
Determining If Your Process Is “Out Of Control” 10
Upper Control Limit

Zone A (UCL)
Zone B
Zone C
Average
Zone C
Zone B
Zone A
Lower Control Limit
(LCL)
Source: Memory Jogger Plus, ©1994 GOAL/QPC
Dividing the control chart into zones can aid in detecting

special cause variation. Each of the zones represent
standard deviations from the mean. Zone C, for example,
is + and - one standard deviation from the average.
D M A I C
Minitab Rules 11
1. One point beyond zone A (Western Electric)

2. Nine points in a row in zone C or beyond (All on one side.) (Western
Electric)
3. Six points in a row, all increasing or decreasing
4. Fourteen points in a row, alternating up and down
5. Two out of three points in a row in zone A or beyond (Western Electric)
6. Four out of five points in a row in zone B or beyond (Western Electric)
7. Fifteen points in a row in zone C, above or below center
8. Eight points in a row beyond zone C, above or below center
+3σ
+2σ A
B
+1σ
C
C
-1σ
B
-2σ
A
-3σ
These are the rules available in Minitab. You can find

these rules under the “TESTS” button inside the Control
Chart screen. GE has standardized on the above-
mentioned rules to identify special cause variation..
You can select the rules you want Minitab to use.

One thing to be careful with all these rules is that
you may not want to react too quickly. Several
factors need to be considered:
Are consequences high (i.e. cost, safety, etc) if we
do not react to a potentially unstable system using
the 8 rules?
Can we potentially make a process unstable if we
react too quickly to the control chart, when in reality
the process is not out of control (i.e., “Do not fix it,
if it is not broken”)?
D M A I C
Control Limits vs. Specification Limits 12
Control Limits Customer Specification

Defined based on process Limits
performance (+/- 3 estimated Defined based on feedback
standard deviations from the from the customer(s)
mean) Help determine if your process
Help determine if your process is producing defects
is “in control” (without special Plotted on histograms (not
cause variation) control charts)
Plotted on control charts Change when your customers
Change when there is a say they do!
verified, significant change to Represent the voice of the
your process customer
Represent the voice of the
process
Control limits are not the same as tolerances. While the data collected for the control charts
Control limits are what you have demonstrates:
Tolerances are what you want
Mean = 40 days
Example:
UCL= 46 days
Customer may define the following for shipment delivery

LCL= 39 days
from time of order:
The UCL and LCL are calculated based on the data
Target= 20 days collected–the UCL and LCL are NOT related at all to
spec. limits.
USL= 45 days
D M A I C
Control Limits vs. Specification Limits (continued) 13
It is possible to have a stable (in control) process that has unacceptable

variation
Assume both process A and B are statistically performing “in control”
PROCESS A PROCESS B
Lower Upper Lower Upper
Spec. Limit Spec. Limit Spec. Limit Spec. Limit
Process A has acceptable Process B has unacceptable

variation when evaluated against variation when evaluated against
customer specification limits customer specification limits
When a process is in statistical control and has unacceptable variation,

work on the reduction of variation due to common causes
To reduce common cause variation, make improvements to the Vital Few X
variables
D M A I C
Selecting Measures For Control Charts 14
What Do I Monitor? Process
Input Output
Variables (Y’s)
(X’s)
Key input and Key output measures

process measures (Y) from the
(X) that track X customer’s
X X X
variables identified perspective
in your project as Process Variables
key drivers of
Project (Y) variables (X’s)
Measure The Process And Not The People
The measures that you choose to monitor in the control

plan are the critical measures for your process. These are
the process management measures that the process owner
will rely on to track process performance over time.
D M A I C
Selecting Measures For Control Charts (continued) 15
How Do I Monitor The Control Plan?
Establish Develop
Ensure Data Collect Data
Data Operational
Consistency & Monitor
Collection Definitions
& Stability Consistency
Goals & Procedures
Use the 4-step data collection plan in order to develop a process

that will protect the integrity of the data.
Following the 4-step data collection plan will ensure that

the data collected on an ongoing basis will accurately
reflect the variation in the process. Since this data will be
collected over a long period of time, measurement
systems analysis to ensure consistency should be a major
concern for both the project team as well as the process
owner after the hand-off. Make sure that periodic audits
of the data collection are an integral part of the overall
control plan.
D M A I C
Types Of Errors In Control Charts 16
3 σ-level Control Limits

– Created by Dr. Walter Shewhart to minimize two types of mistakes
– Placed empirically because they minimize the two types of mistakes
– Are not probability limits
Two types of Mistakes:
– Calling a special cause of variation a common cause of variation (Missing
a chance to identify a change in the process)
– Calling a common cause of variation a special cause of variation
(Interfering with a stable process, wasting resources looking for special
causes of variation that do not exist)
Control limits vs. Specification limits
– Control limits are determined by process data
– Specification limits are determined by customer needs
D M A I C
Two Types Of Control Charts 17
VARIABLE CHART
Uses Measured Values

– Cycle Time, Lengths, Diameters, etc.
Generally One Characteristic Per Chart
More Expensive, But More Information
ATTRIBUTE CHART
Pass/Fail, Good/Bad, Go/No-Go Information

Can Be Many Characteristics Per Chart
Less Expensive, But Less Information
Variable = continuous data

Attribute = discrete data
D M A I C
Selecting the Appropriate Control Chart 18
Variable or
Attribute Data?*
Variable
Variable Attribute
Attribute
No Rational Yes No Constant Yes

Subgrouping
Possible? Lot Size?
Defects or Defects or
Defective Defective
Individuals
Individuals &
& Xbar
Xbar &
&
Moving
Moving Range
Range uu
Range
Range pp cc np
np
There are many different types of control charts. The If the data being analyzed is discrete (attribute), you
chart pictured here is a helpful guide to decide which type would need to determine:
of chart to use and when. The primary determinant of If you have a constant lot (sample) size
which type of control chart to use is the type of data or if the lot (sample) size is not constant
being analyzed. If you are counting defects, you may have
multiple defects per unit. If you are counting
If the data being analyzed is continuous (variable), you defectives, the unit either is or is not defective.
would use either: In this module we will study the different types of control
The “Individuals & Moving Range” (I-MR): use if charts and when to use each one.
Rational Subgrouping is not possible (you want to see
each transaction on the control chart) What you want the Control Chart to tell you is:
The Xbar & Range (X - R) use if Rational Subgrouping
is possible. Typically subgroup size will be between 3 - 5. Is my level of non-conformance consistent?
Use (X - S) if subgroup size is 7 or greater Am I controlling my significant X’s?
TM
* Refer to Memory Jogger for further details.
D M A I C
Interpreting Control Charts–Activity (20 minutes) 19
Desired Outcome: Practice deciding which control chart is the appropriate

choice for given data examples
What How Who Timing
Preparation Choose a partner for the exercise All 1 min.
Review and For each of the examples listed on Partners 9 mins.

Feedback the following page
– Determine the appropriate type of
control chart
– Determine how you would label
the X and Y axes
Close Keep notes on your responses for Partners 10 mins.

report-out
D M A I C
Selecting Control Charts-Partner Activity (continued) 20
1. A team from Marketing is tracking response rates (yes, no) on telemarketing

calls. Each day, they randomly sample 100 responses from their outgoing
calls. The number of positive responses is plotted on the chart.
2. The technical support area is monitoring average response time on help
requests. Each hour, 5 calls are sampled.
3. A team is monitoring the number of applications received each day with
incomplete fields. All applications are being audited. Each day, 200-250
applications are received.
4. A team is monitoring the cycle time for the underwriters’ loan decisions. The
data comes in slowly; that is, one loan is completed daily.
5. A team is tracking the number of fields left blank on an application. Each day,
a sample of 100 applications is audited.
6. Another team is doing the same thing as the team in example 5 (counting the
number of fields left blank on an application), only they are looking at all the
applications that come in daily. Each day, 50-100 applications are turned in.
7. The team is tracking the number of abandoned calls to a customer call center
in a shift. Each shift takes from 200-400 calls.
D M A I C
21
Variable Control Charts–

Xbar R-Chart
In class we will show you how to construct in Minitab

two continuous Control Charts (Xbar and R Chart and
I/MR Chart) and one Discrete Chart (p Chart). The
remainder will be reference material. Note: All Control
Chart interpretation is identical.
D M A I C
Selecting The Appropriate Control Chart 22
Variable or
Attribute Data?*
Variable
Variable Attribute
Attribute

Subgrouping
Possible? Lot Size?
Defective Defective
Individuals
Individuals &
& Xbar
Xbar &
&
Moving
Moving Range
Range uu
Range
Range pp cc np
np
We will now focus on the Variable Control Chart with

Rational Subgrouping: the Xbar & Range Chart.
D M A I C
A Valid Variable Control Chart Has… 23
Data in time or production sequence

– To show stability, time-to-time variation
A measure of central tendency
– To portray behavior of process center
A measure of variability
Control limits
– To allow separating common cause from assignable (special) cause
Xbar & Range Charts
Xbar Chart: a plot of the sample means over time

R-Chart: a plot of the range (difference between highest and
lowest values) of a sample over time
General Comments on Xbar & Range Charts: In Minitab:

Stat > Control Charts > Xbar-R
Cluster or periodic measurements of characteristic
Frequency depends on speed and stability of process.
Subgroup averages are plotted on Xbar chart
Xbar chart monitors central tendency of a process over
time
Subgroup ranges plotted on Range Chart
R-Chart monitors the variability of a process over time
Provides data-smoothing effect
D M A I C
24
Variable Control Charts–

I & MR Chart
D M A I C
Variable or
Attribute Data?*
Variable
Variable Attribute
Attribute

Subgrouping
Possible? Lot Size?
Defective Defective
Individuals
Individuals &
& Xbar
Xbar &
&
Moving
Moving Range
Range uu
Range
Range pp cc np
np
D M A I C
Individuals & Moving Range Charts 26
Can use in intermittent operations

Similar to Xbar & R-Charts, Except...
– Single Values, Not Subgroups
– Range Values must be artificially constructed
– Somewhat “Noisier”, since you track each individual
value (transaction,etc.)
Individuals Chart: a plot of the individual values over time
Moving Range Chart: a plot of the moving range (for two
samples |Xi –( Xi-1)|over time)
General Comments on I & MR Charts: In Minitab:

Range value artificially constructed from successive Stat > Control Charts > I-MR
readings
Subgroup size for X is n = 1; for MR is usually 2
Some correlation between charts is possible
More “noise” in chart–tougher to spot true process
shift
Displays the variability between individual
observations over time
Assumes that past and present data is equally important
D M A I C
Building An Individuals And Moving Range Chart 27
Individual Data Moving Range

55 N/A
56 ABS(55-56) = 1
59 ABS(56-59) = 3
55 ABS(59-55) = 4
Individuals
4 Moving Range
3
2
1
0
An Individuals chart plots actual data–each transaction.
A Moving Range chart calculates ranges between

consecutive data points (absolute value). You lose one
data point, since the range is not calculated for the first
point.
D M A I C
Take Aways–Variable Control Charts 28
Variable control charts can be used with continuous data to tell

when a process is:
– Experiencing only common cause variation and working at its intended
best
– When the process is disturbed and needs corrective action
Control charts:
– Time ordered plot of data
– Reflect the expected range of variation of the data
– Identify when a special cause appears to be influencing the data
Xbar & R-Charts are used for plotting means and ranges of
subgroups over time
I & MR charts are used for plotting individual values and
moving ranges over time
D M A I C
Take Aways–Variable Control Charts 29
Control limits are typically calculated as + 3 standard deviations

away from the mean of the process
Control limits and specification limits are not the same
– Control limits are calculated from the sample data; they are internal to the
process
– Specification limits are determined by your performance standard; they
are external to the process (generally come from the customer)
Know when a process is out of control: Western Electric Rules
Control charts are only as good as the actions that you take to
keep the process in control
D M A I C
Variable Control Charts: Activity (10 minutes) 30
Objective Practice using Minitab to generate an Xbar and R Chart.

Think about the correct Data File: Variable Control Charts.mtw
rational subgroup.
Instructions The quality team has 25 weeks of historical data, contained in the
file listed above. You have been asked to examine their data and
believe an Xbar and R Chart, or an I/MR Chart would be the
appropriate tool to use.
1. Plot the data using an I and MR chart.
2. Plot the data as Xbar/R-charts, forming subgroups by shift, day,

and week.
3. For each chart created in 1 and 2 above, answer the following:

A. Is the chart for spread stable?
B. Is the chart for center stable?
Time 10 minutes
Stat > Control Charts > Xbar-R.
D M A I C
31
Minitab Output
Stat > Control Chart > I-MR

Variable: Response Time
Click on TESTS: Perform “All 8 Tests”
Response Time
I and MR Chart for Response
600 1
Individual Value
500
400
1 1 1 11 1
300 1 1
1 11 1 1 1 11 1 11 1
1 1 1
200 5 5 5
55 3.0SL=205.2
2 44 7 4
100 7 44
X=84.00
4 2 4 77 7
0 6 2
-3.0SL=-37.19
-100
Subgroup 0 500 1000
500 1
1
Moving Range
400
300
1 1 11
1 1 1 1
1
200 1 11 1 11 1 1 11 1
2 1 1 1 1 111 1
3.0SL=148.9
100
R=45.57
0 2 2 2 2
22 -3.0SL=0.00E+00
22
22 2 222 2
D M A I C
Minitab Output (continued) 32
Stat > Control Charts > Xbar-R

Single Cause: Response Time
Subgroup Size: Day
Click on > Tests: “Perform All 8 Tests”
SUBGRP.MTW: Xbar/R for Response Time
Xbar/R Chart for Response

Time: Day
D M A I C
33
Attribute Control Charts
D M A I C
Selecting The Appropriate Control Charts 34
Attribute Data
Variable or
Attribute Data?*
Variable
Variable Attribute
Attribute

Subgrouping
Possible? Lot Size?
Defective Defective
Individuals
Individuals &
& Xbar
Xbar &
&
Moving
Moving Range
Range uu
Range
Range pp cc np
np
If the data being analyzed is discrete (attribute), you

would
need to determine:
If you have a constant lot (sample) size
or if the lot (sample) size is not constant
If you are counting defects you may have multiple
defects per unit. If you are counting defectives, the unit
either is or is not defective.
In this module, we will study the different types of

control charts and when to use each one.
What you want the Control Chart to tell you is:

Is my level of non-conformance consistent?
Am I controlling my significant X’s?
TM
* Refer to Memory Jogger for further details.
D M A I C
Important Definitions 35
A Defect
A single characteristic that does not meet requirements
A Defective
A unit that contains one or more defects
Attribute Charts Can Consider Either Case,

Depending On The Chart Type Chosen
Other terminology may also be used:
A defect is also known as a nonconformance

A defective is known as a nonconforming
D M A I C
Classification Of Attribute Chart Types 36
Constant Variable
Sample Size Sample Size
Defects c u Can not count non-occurrences
Defective Can count occurrences

np p
To choose the appropriate attribute control chart, use the

following factors:
1. Is it defects or defectives that you are investigating in

the data you are collecting?
For example:
Is it pass/fail for the part or process?
Or is it several sub-components or sub-processes
passed/failed for the part or process)?
2. Is it a constant lot (sample) size or variable lot (sample)

size?
D M A I C
37

P-Chart
D M A I C
P-Chart
Variable or
Attribute Data?*
Variable
Variable Attribute
Attribute

Subgrouping
Possible? Lot Size?
Defective Defective
Individuals
Individuals &
& Xbar
Xbar &
&
Moving
Moving Range
Range uu
Range
Range pp cc np
np
D M A I C
P-Chart 39
Chart of proportion defective

Variable subgroup/lot size (n)
P–chart shows the Proportion Defective–the proportion

that do not conform
The units may have one or more defects
D M A I C
Desired Outcome: Practice interpreting P-Chart
What How Who Timing
Preparation In your team, read background Team 5 mins.

information on the next page
Interpret Interpret P-Chart results and Team 5 mins.

Control Charts answer questions

Exercise out on your responses to the
questions
D M A I C
Interpreting Control Charts–Activity 41
The loan department has made a change to the loan application process. They
think the change will help weed out the bad loan risks before the applications
are actually submitted. If they are right, the approval rate for submitted
applications should increase. They want to check to see if this change in the
process has had the desired effect.
1. Is there evidence that the new process made a difference?
2. What is the number of opportunities for defect in this process measure?
3. Is this an input, process, or output variable? Why?
3.0SL = 97.71
New Process
•
P-Chart Of Loan Approval Rate
• 93.79
% Applications Approved
Implemented
May 1 To November 5 •
• •
90
• • • •• •
89.86
•• •• • •
• • • P = 85.94
•
• 82.02
80 •
• •
• • 78.09
-3.0SL = 74.18
70
0 10 20 30
Week (Time Order)
Note: “SL” is the standard Minitab notation for “Sigma Limits” which are the same as Control Limits.
To create a p-chart in Minitab:
STAT > CONTROL CHARTS > P
D M A I C
Reference Materials–Attribute Control Charts C-Chart 42
Reference Materials–
C-Chart
Pages 43-60 cover c, np and u charts. All Attribute Charts are

interpreted as discussed in the previous pages for p-charts.
Review these charts with your mentor for additional exercises.
Proceed to page 61
Stat > Control Charts:
All 4 Attribute Charts: (p,np,c,u) can be generated from

this menu.
D M A I C
Attribute Data
Variable or
Attribute Data?*
Variable
Variable Attribute
Attribute

Subgrouping
Possible? Lot Size?
Defective Defective
Individuals
Individuals &
& Xbar
Xbar &
&
Moving
Moving Range
Range uu
Range
Range pp cc np
np
D M A I C
C-Chart 44
Chart for counts

Based on Poisson distribution
– High probability of finding defect of some
type. Large samples are needed if defect
probabilities are low.
– Lower probability of a defect of a given
type
Works best on complex unit of product
Constant subgroup/lot size
C bar = DPU = number of defects detected/subgroup
A subgroup may contain one or more “physical” units.
You need C bar up around 5. If it is less than 5, your

distribution will be skewed because it will be
truncated by 0.
Increase your subgroup size if you need to increase C bar

or collapse existing categories into meaningful groups.
More specifically, combine two or more categories so
that the frequencies are greater than 5.
D M A I C
Desired Outcome: Practice interpreting C-Chart
What How Who Timing
Preparation In your team, read background All 5 mins.

information. Review chart on next
2 pages
Interpret Interpret C-Chart by answering the Facilitator 5 mins.

Control Charts questions on the page (Sub-group)

questions
D M A I C
Interpreting Control Charts–Activity (continued) 46
An applications group of a large insurer recently made some changes to the design of
their application. The old applications had 10 fields for the customer to fill out; the new
application has 5 fields for the customer to fill out. They are now monitoring the incoming
applications to ensure the new design has had an impact on completeness. A random
sample of 100 applications are reviewed each day and the total number of incomplete
fields are plotted.
The chart for this data is on the following page.
1. Why is the control chart a C-Chart?
2. Is the number of errors in a sample stable over the 20–day sample?
3. What questions does the chart raise?
Extra Credit!! Calculate baseline sigma for the first 20 days of service for the new
application.
Day Data Day Data Day Data Day Data
1 5 6 9 11 6 16 7
2 5 7 9 12 6 17 7
3 11 8 8 13 7 18 8
4 5 9 8 14 5 19 12
5 8 10 7 15 6 20 13
D M A I C
3.0SL=15.87
15
Number of Incomplete Fields
10
C=7.600
0 -3.0SL=0.00E+00
0 10 20
Day of Sample
D M A I C
48

U-Chart
D M A I C
Attribute Data
Variable or
Attribute Data?*
Variable
Variable Attribute
Attribute

Subgrouping
Possible? Lot Size?
Defective Defective
Individuals
Individuals &
& Xbar
Xbar &
&
Moving
Moving Range
Range uu
Range
Range pp cc np
np
D M A I C
U-Chart 50
Chart for defects per unit, with variable lot (subgroup) size
Same logic as C-Chart, except variable lot (subgroup) size (n)
Compare to C-Chart In Minitab:

Stat > Control Charts > I
D M A I C
U-Chart Activity (20 minutes) 51
Desired Outcome: Practice constructing and interpreting U-Chart
What How Who Timing
Preparation In your team, prepare to create a All 1 min.

U-Chart in Minitab
Construct And In the U-Chart.mtw, column errors Team 10 mins.

Interpret The contains data (in time sequence)
U-Chart of the defects found each day on
customer orders. A defect is
defined to be inaccurate
information found on a customer
order. Both the number of defects
and the daily number of orders are
recorded.
Using Minitab, construct a
U-Chart of the data
What are your observations?

out your conclusions
Note: Do not turn the page until you have completed the activity.
D M A I C
U-Chart Example Minitab Menu Commands 52
MINITAB FILE: U_Chart.mtw
D M A I C
U-Chart Minitab Input & Output 53
U Chart for errors
3
Sample Count
3.0SL=2.114
2
U=1.764
-3.0SL=1.415
0 10 20 30
Sample Number
CONTROL: The process does not look stable–there are

indications pf special cause variations.
Why do the control limits vary? Lot (Sample) size varies.
D M A I C
54

NP-Chart
D M A I C
NP-Chart
Variable or
Attribute Data?*
Variable
Variable Attribute
Attribute

Subgrouping
Possible? Lot Size?
Defective Defective
Individuals
Individuals &
& Xbar
Xbar &
&
Moving
Moving Range
Range uu
Range
Range pp cc np
np
D M A I C
NP-Chart 56
NP-chart: number of defectives (non-conforming items) in

subgroup
Same logic as the P-Chart, except constant subgroup/lot
size (n)
D M A I C
Desired Outcome: Practice interpreting NP-Chart
What How Who Timing

Preparation In your team, read background Team 5 mins.
information on next 2 pages
Interpret Interpret NP-Chart results and Team 5 mins.

Control Charts answer questions

questions
D M A I C
An applications group of a large insurer recently made some changes to the design of
their application. They are now monitoring the incoming applications to ensure the new
design has had an impact on completeness. 150 applications are chosen randomly and
reviewed each day. The number that are incomplete are plotted each day
The chart for this data is on the following page.

1. Why is the chart an NP–chart?
2. What questions does the chart raise about the completeness of the insurance
application?
3. What is the number of opportunities for defect in this measure?
4. What is the difference between the value plotted on a P–and an NP–Chart?
Day Data Day Data Day Data Day Data
1 12 6 14 11 2 16 11
2 22 7 16 12 15 17 17
3 14 8 19 13 8 18 12
4 15 9 11 14 12 19 11
5 12 10 1 15 12 20 21
D M A I C
25
3.0SL=23.13
20
# of Incomplete Applications
15
NP=12.85
10
Day of Sample
5
-3.0SL=2.567
0 10 20
D M A I C
Attribute Chart Subgroup Size 60
Rule of Thumb:
– Select a subgroup size that will provide an average defect/defective of
approximately
C, U, NP > 5.0
To Make UCL & LCL Nearly Symmetrical Around the Mean
– For NP-charts, to select the appropriate sample size such that 95% of the
subgroups will have at least one defective, use the relationship
n= 3
p
D M A I C
Summary Of Attribute Charts 61
Useful when variable data not available

Use count/classification data–pass/fail, good/bad
Same general rules for interpretation as variable charts
Useful as end-to-end overview; use variable charts for further
study of problems
Can use data gathered for other purposes
Generally less expensive to administer, but tell you less
Shortcomings
Including too many variables makes interpretation difficult
Must fit the parameters you are evaluating to theoretical
distribution [Poisson (C, U-Charts), Binomial (P, NP-Charts)]
Need to evaluate whether constant/non-constant lot size will
help you with root cause analysis
Sensitivity is dependent on magnitude of defect level
Variable or
Attribute Data?*
Variable
Variable Attribute
Attribute
Rational Yes
No
Subgrouping
No Constant Yes
Possible? Lot Size?
Defective Defective
Individuals
Individuals&& Xbar
Xbar&&
Moving
Moving Range
Range
Range
Range uu pp cc np
np
D M A I C
Take Aways Attribute Control Charts 62
Attribute control charts are used to monitor the level of

nonconformance of a process
Select the appropriate attribute control chart based upon
– Constant vs. variable lot (sample) size
– Defects vs. defectives
Defect
– A single characteristic that does not meet requirements
Defective
– A unit that contains one or more DEFECTS
D M A I C
Summary–Control Charts 63
All control charts have the same form

– Time plot of data
– Statistical limits placed +/- 3s from center line
Choose the control chart according to:
– Data type
– Sample size
Interpret and act on the control chart
– Investigate special causes of variation
– Assess baseline sigma in order to understand if we are performing to
customer CTQ’s
– Shrink common cause variation by making fundamental changes in the
Vital Few X variables
Maintain the control chart
– Make notes on the control chart to indicate problems, changes, or
important events
D M A I C
64
Calculating Control Limits

(Optional)
D M A I C
Control Limit Calculations 65
The formulas for calculating control limits are shown on the

following pages
There are different tables for continuous and discrete data
Both continuous and discrete data control charts have control
limits that are placed +/- 3 estimated sigma from the average
line
D M A I C
Control Limit Calculations–Formulas 66
Continuous Data
Control limits are calculated using control chart factors and the
Range-Bar (an estimate of short-term sigma)
Control chart factors were invented by Shewart in the 1920’s to avoid
long-hand calculation
Control chart factors are shown in the table according to the sample
size for each subgroup
Individual control charts are considered to have a sample of size 2,
the number of data points make up the moving range
Discrete Data
Control limits are calculated using a formula that estimates sigma
without
the necessity to transform the data
The normal estimate for sigma (under radical) is then multiplied by
three
D M A I C
Control Limit Calculations–Continuous Data Table 67
Type Control Sample Central Line* Control Limits

Chart Size n
(X1 + X2 +…Xk) UCLx = X + A2R

Average and Range X= k LCLx = X - A2R
< 10, but
usually
3 to 5
(R1 + R2 +…Rk) UCLR = D4R
X and R R= k LCLR = D3R
Average and (X1 + X2 + … Xk) UCLx = X + A3s

Standard X= k LCLx = X - A3s
Deviation Usually
≥ 10
(s1 + s2 +…sk) UCLs = B4s
X and s s= k LCLs = B3s
Median ˜ 1 + X˜ 2 + … X˜ k)
(X ˜ + A2R
UCLx = X
and Range X
˜= k LCLx = X
˜ - A2R
< 10, but
usually
3 to 5 (R1 + R2 +…Rk) UCLR = D4R
X
˜ and R R= k LCLR = D3R
(X1 + X2 + … Xk) UCLx = X + E2Rm

Individuals and X= k LCLx = X - E2Rm
Moving Range
1 Rm = I(Xi+1 - Xi)I
UCLRm = D4Rm
X and Rm (R1 + R2 +…Rk-1) LCLRm = D3Rm
Rm = k-1
˜
k = # of subgroups, X = median value within each subgroup
ΣXi
*X = n
D M A I C
Control Limit Calculations–Table Of Contents 68
Sample X and R Chart X and s Chart

Size
n
A2 D3 D4 A3 B3 B4 c4*
2 1.880 0 3.267 2.659 0 3.267 .7979
3 1.023 0 2.574 1.954 0 2.568 .8862
4 0.729 0 2.282 1.628 0 2.266 .9213
5 0.577 0 2.114 1.427 0 2.089 .9400
6 0.483 0 2.004 1.287 0.030 1.970 .9515
7 0.419 0.076 1.924 1.182 0.118 1.882 .9594
8 0.373 0.136 1.864 1.099 0.185 1.815 .9650
9 0.337 0.184 1.816 1.032 0.239 1.761 .9693
10 0.308 0.223 1.777 0.975 0.284 1.716 .9727
Sample ˜ and R Chart

X X and Rm Chart
Size
n ˜2
A D3 D4 E2 D3 D4 d2*
2 – 0 3.267 2.659 0 3.267 1.128
3 1.187 0 2.574 1.772 0 2.574 1.693
4 – 0 2.282 1.457 0 2.282 2.059
5 0.691 0 2.114 1.290 0 2.114 2.326
6 – 0 2.004 1.184 0 2.004 2.534
7 0.509 0.076 1.924 1.109 0.076 1.924 2.704
8 – 0.136 1.864 1.054 0.136 1.864 2.847
9 0.412 0.184 1.816 1.010 0.184 1.816 2.970
10 – 0.223 1.777 0.975 0.223 1.777 3.078
D M A I C
Control Limit Calculations–Discrete Data Table 69
Type Control Sample Control Limits

Central Line
Chart Size
Fraction
p(1 - p)
Defective For each subgroup: *UCLp = p + 3
Variable, n
p = np/n
usually ≥ 50 For all subgroups:
p-Chart p = ∑np/ ∑n *LCLp = p - 3 p(1 - p)
n
UCLnp = np + 3
Number For each subgroup: np(1 - p)
Defective Constant, usually np = # defectives
≥ 50 For all subgroups:
np = ∑np/k LCLnp = np - 3
np-Chart np(1 - p)
Number of For each subgroup: UCLc = c + 3

Defects c = # defects c
Constant For all subgroups: LCLc = c - 3
c-Chart c = ∑c/k c
Number of For each subgroup: *UCLu = u + 3 u

Defects Per u = c/n n
Unit Variable For all subgroups:
u = ∑c/ ∑n *LCLu = u - 3
u-Chart u
n
np = # defectives * This formula creates changing control

c = # of defects limits. To avoid this, use average sample
n = sample size within sizes n for those samples that are within
each subgroup ±20% of the average sample size.
k = # of subgroups Calculate individual limits for the samples
exceeding ± 20%.
If the Lower Control Limit (LCL) is a

negative number, set the LCL to zero.
D M A I C

Dmaic - Book of Knowledge - Green Belt11ea60e023f

Uploaded by

Document Information

Original Description:

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

Dmaic - Book of Knowledge - Green Belt11ea60e023f

Uploaded by

Copyright:

Available Formats

D M A I C

Classroom Session Agenda–Day 1 1

Total Topic Step

60 Measure 1–CTQ Tools continued 1

© GE Capital, Inc., 2000

Classroom Session Agenda–Day 2* 2

Total Topic Step

60 Measurement System Analysis (MSA) 3

* Elevator Speeches will be shared throughout the

© GE Capital, Inc., 2000

Classroom Session Agenda–Day 3* 3

Total Topic Step

* Elevator Speeches will be shared throughout the

© GE Capital, Inc., 2000

© GE Capital, Inc., 2000

The Learning Objectives are divided in 3 categories:

 It is expected that a Green Belt be proficient in all learning

© GE Capital, Inc., 2000

Overview–Classroom Learning Objectives 2

1 Explain the benefits of Six Sigma to GE’s business.

2 Describe the Six Sigma Methodology for quality improvement.

© GE Capital, Inc., 2000

Overview–Project Learning Objectives 3

1. Determine if DMAIC is the right strategy by identifying the

© GE Capital, Inc., 2000

Overview–Test Learning Objectives 4

1 Describe the 5 phases of DMAIC, including the purpose, tools,

© GE Capital, Inc., 2000

Define–Project Learning Objectives 5

1 Identify project CTQs.

© GE Capital, Inc., 2000

Define–Project Learning Objectives (continued) 6

2 Use a process/product drill-down tree to: define the limits of a

© GE Capital, Inc., 2000

Define–Project Learning Objectives (continued) 7

3 Develop a team charter.

© GE Capital, Inc., 2000

Define–Project Learning Objectives (continued) 8

4 Map a business process.

© GE Capital, Inc., 2000

Define–Project Learning Objectives (continued) 9

5 Obtain approval for a Six Sigma project.

© GE Capital, Inc., 2000

Define–Test Learning Objectives 10

1 Recognize the components of the 12 Step Process and how they

© GE Capital, Inc., 2000

Measure1–Classroom Learning Objectives 11

1 Describe and define the deliverables of Step 1.

© GE Capital, Inc., 2000

Measure 1–Project Learning Objectives 12

1 Identify the project Y.

© GE Capital, Inc., 2000

Measure 1–Project Learning Objectives (continued) 13

2 Select CTQ characteristics.

3 Select and apply appropriate tools to narrow the focus of a Six

© GE Capital, Inc., 2000

Measure 1–Project Learning Objectives (continued) 14

4 Use Process Mapping to identify potential breakdowns, rework loops, and

© GE Capital, Inc., 2000

Measure 1–Project Learning Objectives (continued) 15

5 Use a Failure Modes and Effects Analysis to identify the potential

© GE Capital, Inc., 2000

Measure 1–Project Learning Objectives (continued) 16

It is expected that a Green Belt be proficient in all learning