6.7K views

Uploaded by srikantharvikar

- Cssbb 2007 Final
- Six Sigma Green Belt
- Six Sigma Green Belt Training
- Six Sigma Green Belt Manual
- ASQ SSGB Handbook
- Six sigma green belt
- DMAIC Six Sigma Guide 3M
- Six Sigma Black Belt Wk1-Define & Measure
- Six Sigma Green Belt Training Handout_IIHMR
- 55997347-ASQ-Six-Sigma.pptx
- CSSGB Questions Rev01
- Six Sigma Exam
- Six Sigma Green and Black Belt Question Paper From Iibm
- 7 Measure Statistics
- Six Sigma Exam Preparation
- Lean Six Sigma Exam Practice Questions
- DMAIC Project Template
- GrnPgsII-V
- Green Bwlt Question Paper
- Six sigma questions

You are on page 1of 650

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

10 Break

45 Measure 2–Performance Standards 2

5 Break

120 Minitab Tutorial and Graphical

Analysis various

DMAIC GB A TX PG V4.2.0

D M A I C

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

15 Break

15 Analyze Overview 4,5,6

150 Analyze 4–Establish Process Capability 4

day

DMAIC GB A TX PG V4.2.0

D M A I C

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

day

DMAIC GB A TX PG V4.2.0

D M A I C

DMAIC GB A TX PG V4.2.0

D M A I C

Assessment Strategy

Introductions

1

1. Classroom

2. Project, and

3. Test

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

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.

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.

DMAIC GB B TX PG V 4.2.0

D M A I C

conditions under which the DFSS methodology would be more

appropriate.

Compare and contrast the DFSS design methodologies to

DMAIC.

DMAIC GB B TX PG V 4.2.0

D M A I C

and outputs for each phase.

DMAIC GB B TX PG V 4.2.0

D M A I C

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).

DMAIC GB B TX PG V 4.2.0

D M A I C

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.

DMAIC GB B TX PG V 4.2.0

D M A I C

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.

DMAIC GB B TX PG V 4.2.0

D M A I C

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

DMAIC GB B TX PG V 4.2.0

D M A I C

Identify the steps in the project approval process.

Enter a project into QPT.

DMAIC GB B TX PG V 4.2.0

D M A I C

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.

DMAIC GB B TX PG V 4.2.0

D M A I C

DMAIC GB B TX PG V 4.2.0

D M A I C

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.

DMAIC GB B TX PG V 4.2.0

D M A I C

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.

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.

DMAIC GB B TX PG V 4.2.0

D M A I C

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.

DMAIC GB B TX PG V 4.2.0

D M A I C

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.

DMAIC GB B TX PG V 4.2.0

D M A I C

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.

Describe the purpose and benefits of a Data Collection Plan.

Write a data collection strategy.

Define a clear strategy for collecting reliable data efficiently.

DMAIC GB B TX PG V 4.2.0

D M A I C

phases (DMA) and optimization phases (IC).

Define product characterization.

Define process optimization.

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.”

DMAIC GB B TX PG V 4.2.0

D M A I C

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.

DMAIC GB B TX PG V 4.2.0

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.

DMAIC GB B TX PG V 4.2.0

D M A I C

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.

DMAIC GB B TX PG V 4.2.0

D M A I C

(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.

DMAIC GB B TX PG V 4.2.0

D M A I C

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.

DMAIC GB B TX PG V 4.2.0

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.

DMAIC.

DMAIC GB B TX PG V 4.2.0

D M A I C

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.

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.

DMAIC GB B TX PG V 4.2.0

D M A I C

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.

DMAIC GB B TX PG V 4.2.0

D M A I C

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.

DMAIC GB B TX PG V 4.2.0

D M A I C

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.

DMAIC GB B TX PG V 4.2.0

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.

Describe value added/ non-value added analysis.

DMAIC GB B TX PG V 4.2.0

D M A I C

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.

DMAIC GB B TX PG V 4.2.0

D M A I C

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.

DMAIC GB B TX PG V 4.2.0

D M A I C

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.

DMAIC GB B TX PG V 4.2.0

D M A I C

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.

DMAIC GB B TX PG V 4.2.0

D M A I C

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.

DMAIC GB B TX PG V 4.2.0

D M A I C

normal/non-normal data.

Describe the use of the normality test.

Describe the use of Mood’s median for non-normal data.

DMAIC GB B TX PG V 4.2.0

D M A I C

elements.

parameters.

Develop a mathematical model of a proposed solution.

Determine the best configuration or combination of X’s

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.

DMAIC GB B TX PG V 4.2.0

D M A I C

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.

DMAIC GB B TX PG V 4.2.0

D M A I C

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.

DMAIC GB B TX PG V 4.2.0

D M A I C

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.

DMAIC GB B TX PG V 4.2.0

D M A I C

Sigma.

DMAIC GB B TX PG V 4.2.0

D M A I C

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.

DMAIC GB B TX PG V 4.2.0

D M A I C

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.

DMAIC GB B TX PG V 4.2.0

D M A I C

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.

DMAIC GB B TX PG V 4.2.0

D M A I C

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.

DMAIC GB B TX PG V 4.2.0

D M A I C

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.

DMAIC GB B TX PG V 4.2.0

D M A I C

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.

DMAIC GB B TX PG V 4.2.0

D M A I C

DMAIC GB B TX PG V 4.2.0

D M A I C

Introductions

Desired Outcome: Introduction of Classmates and CD Rom Learnings

Preparation yourself to your table-team members

(name and business only, at this

time.)

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.

Exercise report out

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.

© GE Capital, Inc., 2000

DMAIC GB C TX PG V 4.2.0

D M A I C

Classroom Session–Objectives 2

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

DMAIC GB C TX PG V 4.2.0

D M A I C

Iterative Process 3

Steps A,B,C

Steps

10,11,12 Steps 1,2,3

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.

© GE Capital, Inc., 2000

DMAIC GB C TX PG V 4.2.0

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

DMAIC GB C TX PG V 4.2.0

D M A I C

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

DMAIC GB C TX PG V 4.2.0

D M A I C

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

DMAIC GB C TX PG V 4.2.0

D M A I C

Meeting Skills 7

Roles:

Leader or Facilitator

Timekeeper

Scribe

Note Taker

Tools:

Agenda

Desired Outcomes

Ground Rules

Decision-Making Process

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

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.

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.

© GE Capital, Inc., 2000

DMAIC GB C TX PG V 4.2.0

D M A I C

Preparation and scribe

Team Name names

Using N/3*, prioritize your

brainstormed list

Identify your top name

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.

DMAIC GB C TX PG V 4.2.0

D M A I C

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

deliverables for Define phase

DMAIC GB D TX PG V 4.2.0

D M A I C

2

First Things First (10 minutes)

Preparation have experienced in the past in Team

which the change was unsuccessful. Member

Why did the change initiative fail?

ideas for why change initiatives

fail. What are the common themes?

Be prepared to report findings.

DMAIC GB D TX PG V 4.2.0

D M A I C

Techn

ical St

rate g y

rat egy

iza tional St

an

al/Org

Cultur

Change Initiative

(Target)

Tools To Achieve The Change Initiative

Q x A = E

acceptance as it is to have a plan for implementing the

solution.

Strategy are both necessary in order to effectively

achieve your change target. CAP gives us tools to

develop the Cultural/Organizational or Influence

Strategy.

© GE Capital, Inc., 2000

DMAIC GB D TX PG V 4.2.0

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

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.

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.

DMAIC GB D TX PG V 4.2.0

D M A I C

Shared Need Vision Commitment Last Progress

Threat vs. Of Frame Opportunity Matrix

Opportunity Matrix GRPI

Project Scope ARMI

Contract

Measure

Analyze

Last Checklist Last Checklist

Control Pans Control Pans

Measures And Measures And

Rewards Rewards

Control

Section of the Self-Paced Workbook. For further training,

you can attend the week-long course sponsored by

Learning Services.

DMAIC GB D TX PG V 4.2.0

D M A I C

Homework 6

Desired Outcome: Identify Cap Tools which you will use in your project

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

DMAIC GB D TX PG V 4.2.0

D M A I C

D M A I C

DEFINE

PHASE

OVERVIEW

Identify Project Develop Team Define

CTQ’s Charter Process Map

DMAIC GB D TX PG V 4.2.0

D M A I C

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

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.

Define A: Identify Project CTQ’s

Define B: Develop Team Charter

Define C: Define Process Map

DMAIC GB D TX PG V 4.2.0

D M A I C

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.

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.

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.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

DMAIC GB D TX PG V 4.2.0

D M A I C

10

Define B–Define 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.

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.

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

Identify Project Develop Team Define Process

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

DMAIC GB D TX PG V 4.2.0

D M A I C

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.

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.

C.1 Develop high-level Process Map

DEFINE STEP

OVERVIEW

Identify Project Develop Team Define Process

CTQ’s Charter Map

DMAIC GB D TX PG V 4.2.0

D M A I C

Desired Outcome: Share elevator speech for the project with members of your

team

Speech speech for their project. (See the description Team

of an elevator speech on the following page.) Member

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.

pp 197-198

DMAIC GB D TX PG V 4.2.0

D M A I C

Elevator Speech 13

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)

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?”

DMAIC GB D TX PG V 4.2.0

D M A I C

DMAIC Checklist–Define A 14

Have you considered both internal and external customers?

Have you focused on the most important customer segment(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?

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?

DMAIC GB D TX PG V 4.2.0

D M A I C

DMAIC Checklist–Define B 15

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?

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?

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?

DMAIC GB D TX PG V 4.2.0

D M A I C

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?

DMAIC GB D TX PG V 4.2.0

D M A I C

DMAIC Checklist–Define C 17

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

DMAIC GB D TX PG V 4.2.0

D M A I C

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

DMAIC GB D TX PG V 4.2.0

D M A I C

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.

Designer with questions, and for guidance

DMAIC GB D TX PG V 4.2.0

D M A I C

DMAIC GB D TX PG V 4.2.0

D M A I C

Start work on Green Belt project deliverables for Measure phase

DMAIC GB E TX PG V 4.2.0

D M A I C

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

DMAIC GB E TX PG V 4.2.0

D M A I C

3

The12-Step Process

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

DMAIC GB E TX PG V 4.2.0

D M A I C

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.

This phase is important because it ensures that accurate and reliable data is

collected to measure current process performance related to the customer CTQ.

Measure 1: Select CTQ Characteristic

Measure 2: Define Performance Standards

Measure 3: Establish Data Collection Plan, Validate Measurement

System & Collect Data

DMAIC GB E TX PG V 4.2.0

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

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

DMAIC GB E TX PG V 4.2.0

D M A I C

Statistical Thinking 6

D M A I C

Practical Statistical Statistical Practical

Problem Problem Solution Solution

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

The Practical-To-Statistical-To-Practical

Transformation Process

DMAIC GB E TX PG V 4.2.0

D M A I C

1

Measure 1–Select CTQ Characteristics

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.

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.

1.1 Identify the measurable CTQ characteristic that will be improved (Project Y).

MEASURE STEP

OVERVIEW

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

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

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

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

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)

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).

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

DMAIC GB G TX PG V 4.2.0

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

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.

DMAIC GB G TX PG V 4.2.0

D M A I C

4

Levels Of A Process

Acquire New

Core Business

Buy Service Underwriting Contract Customer Service

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.

consumes the largest portion of total time, or which one

has the most variation or delays?

DMAIC GB G TX PG V 4.2.0

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.

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.

DMAIC GB G TX PG V 4.2.0

D M A I C

6

Outside-In Focus

Should the customer’s process be mapped?

Should you partner with the customer to

assess the processes?

DMAIC GB G TX PG V 4.2.0

D M A I C

7

Subprocess Mapping Techniques

Process flowchart

Alternate path method

Deployment or cross-functional map/flowchart

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.

DMAIC GB G TX PG V 4.2.0

D M A I C

8

Process Mapping Symbols

process mapping.

the following page.

DMAIC GB G TX PG V 4.2.0

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

uses four simple symbols: oval, rectangle, diamond, and

arrow.

simple, or when documenting the work done by a single

person or group.

DMAIC GB G TX PG V 4.2.0

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

reengineering efforts, where the mapping of very large

processes made “decision diamonds” more of a hindrance

than a help.

by split arrows. Teams can then note relative percentages

of times/incidences the process follows each path.

“icons” showing the tools or methods employed at

different points in the process.

of variation which maybe adding complexity to our

process.

DMAIC GB G TX PG V 4.2.0

D M A I C

11

Deployment Or Cross-Functional Flowchart

Sign File

Customer Negotiate

Contract Contract

Negotiate

Call Contract Contract Contract

Admin.

Review Review

Attorney

Contract Contract

Write File

Dealer Contract Contract

performs which tasks and which steps are done

concurrently.

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.)

“bands” on this type of map. (Because customers are

visible at the top, some people refer to this as a “Service

Blueprint.”)

DMAIC GB G TX PG V 4.2.0

D M A I C

12

Steps In Deployment Mapping

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.

DMAIC GB G TX PG V 4.2.0

D M A I C

13

Tips In Subprocess Mapping

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

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.

DMAIC GB G TX PG V 4.2.0

D M A I C

14

Verifying CTQ Elements For Process Mapping

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)

DMAIC GB G TX PG V 4.2.0

D M A I C

15

Subprocess Mapping–Breakout Activity (20 minutes)

Preparation timekeeper, scribe and/or note

taker

Subprocess your own project

Map

Brainstorm the challenges of

subprocess mapping on your

own projects

Exercise your results

DMAIC GB G TX PG V 4.2.0

D M A I C

16

Cause & Effect Diagrams–CTQ Tool #2

logically organize possible causes for a specific problem or effect.

Statement

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.

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.

DMAIC GB G TX PG V 4.2.0

D M A I C

17

Cause & Effect Diagrams–Example

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

Precipitation

Drop off & Hook up Freightliner Trucks

measure?” I can now focus on weight of freight or

distance or temperature, etc. (This is the output

characteristic.)

DMAIC GB G TX PG V 4.2.0

D M A I C

18

FMEA–CTQ Tool #3

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

Plan To Prevent Those Failures

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

phase of the CD Rom. This section is a brief introduction.

DMAIC GB G TX PG V 4.2.0

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

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

System Down 3 “ 4 72 6 3 4 72

Total Risk Priority Number 2,205 Resulting Risk Priority Number 821

accuracy becomes the highest priority defect type.

DMAIC GB G TX PG V 4.2.0

D M A I C

20

Failure Mode And Effects Analysis

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

DMAIC GB G TX PG V 4.2.0

D M A I C

21

FMEA: Calculating Risk Priority Number (RPN)

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

individual scores.

estimates of risk.

DMAIC GB G TX PG V 4.2.0

D M A I C

22

Using A FMEA In The Measure Phase

Helps to identify where and how a process may fail to meet a

CTQ (defects)

need to contain a problem (e.g., regulatory, non

compliance, legal). If this occurs, it raises the priority.

DMAIC GB G TX PG V 4.2.0

D M A I C

23

Pareto Charts–CTQ Tool #4

significant factor that

influences the process

C A E D B

Category of Defect

(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).

DMAIC GB G TX PG V 4.2.0

D M A I C

24

Pareto Charts (continued)

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

“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.

and this would become my project Y.

cause your focus to change.

Find out what is in “other”.

DMAIC GB G TX PG V 4.2.0

D M A I C

25

Using Pareto Charts In The Measure Phase

Graphically displays most frequent occurrence of outcomes

(little y’s)

Helps you get from Big Y to little y

DMAIC GB G TX PG V 4.2.0

D M A I C

26

QFD–CTQ Tool #5

and Translate Customer Needs

and Wants Into Measurable

Features and Characteristics of a Product or Service

DMAIC GB G TX PG V 4.2.0

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

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

to key process variables.

DMAIC GB G TX PG V 4.2.0

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

key process tasks.

DMAIC GB G TX PG V 4.2.0

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)

© GE Capital, Inc., 2000

DMAIC GB G TX PG V 4.2.0

D M A I C

High Positive

Positive

Lunch QFD High Negative

Negative

Percent carbohydrate provided

Percent nutrition provided

Product Requirement

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

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”

DMAIC GB G TX PG V 4.2.0

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

DMAIC GB G TX PG V 4.2.0

D M A I C

Measure 1–Select CTQ Characteristics

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

Project YY Measure 1

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.

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

DMAIC GB G TX PG V 4.2.0

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

Project YY Measure 1

Operational

Operational %

%Carbohydrate

Carbohydrate

High-Level Definition

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.

This information will come from

the VOC. ##of

ofDefect

Defect Measure 2

Opportunities

Opportunities

Per

Per Unit

Unit 11

DMAIC GB G TX PG V 4.2.0

D M A I C

34

Another Example

# 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

DMAIC GB G TX PG V 4.2.0

D M A I C

35

Points To Remember

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

numbers to H-M-L does not make the numbers and cut-

off’s absolute decision points).

DMAIC GB G TX PG V 4.2.0

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

High-Level Definition

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.

This information will come from

the VOC. ##of

ofDefect

Defect Measure 2

Opportunities

Opportunities 11

Per

Per Unit

Unit

© GE Capital, Inc., 2000

DMAIC GB G TX PG V 4.2.0

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

DMAIC GB G TX PG V 4.2.0

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.

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.

DMAIC GB G TX PG V 4.2.0

D M A I C

and timekeeper

Relationships House Of Quality on the next

page

Complete the relationship room

of the house

“How” ratings

Importance

Which “How” is most important

Ratings

to satisfy the customer?

DMAIC GB G TX PG V 4.2.0

D M A I C

40

QFD Activity

HOWS

Service Characteristics

or good on customer

Rep (CSR) excellent

resolved on first call

% 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

Courteous Service 3

Importance rating

Units % Hrs.

Hrs. %

days

Use L=1

M=3

H=9

rating on the CTQ.

DMAIC GB G TX PG V 4.2.0

D M A I C

41

CTQ Tools: Summary

characteristic for your project (Project Y)

Select the tool that is most appropriate for your project

DMAIC GB G TX PG V 4.2.0

D M A I C

DMAIC GB G TX PG V 4.2.0

D M A I C

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.

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

Select CTQ Define Establish Data

Characteristic Performance Collection, Validate

Standards MSA

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

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

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

measured and how they will be measured.

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.

DMAIC GB H TX PG V 4.2.0

D M A I C

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)

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

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

DMAIC GB H TX PG V 4.2.0

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

performance and the starting point for understanding how

to control process output.

DMAIC GB H TX PG V 4.2.0

D M A I C

Timely Call Pick-up

(Output characteristic) Established in Define, Step A

Hold Time

(Project Y Measure) Established in Measure, Step 1

Customer Need

on hold too long CTQ (Target) Established in Measure, Step 2

(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

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.

characteristic is to be quantified. The key metric for the

project.

were no variation in the product/service, this is the value

we would always achieve.

DMAIC GB H TX PG V 4.2.0

D M A I C

6

Mapping Tools To Generation Of CTQ Components

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

DMAIC GB H TX PG V 4.2.0

D M A I C

A performance standard

translates customer needs

into quantified requirements

for our product or process Output

Output Timely Resolution

Characteristic

Characteristic of a Call

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

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.

DMAIC GB H TX PG V 4.2.0

D M A I C

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

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

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

USL= Upper Specification Limit

DMAIC GB H TX PG V 4.2.0

D M A I C

Once the Project Y metric has been established, the target and

specification limits can be established.

Output

Output

Characteristic Timely response

Characteristic

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

DMAIC GB H TX PG V 4.2.0

D M A I C

Project Y 10

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

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.

DMAIC GB H TX PG V 4.2.0

D M A I C

Project Y Alignment 11

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

influence the Y

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.

DMAIC GB H TX PG V 4.2.0

D M A I C

Desired Outcome

Practice defining units and defect for your CTQ

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.

exercise results

DMAIC GB H TX PG V 4.2.0

D M A I C

1

Minitab and Graphical Analysis Module Objectives

Understand data entry and correct data structure for analysis in

Minitab

Review variation

Be able to create and interpret basic graphs in Minitab

Minitab and how to perform basic graphical analysis of

data. We are stepping away from the DMAIC Flow for a

moment.

view the data graphically. We will use Minitab to help us

view this data.

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.

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.

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.

(not shown) window.

shown) from either the menus or the session window (not

shown).

DMAIC GB I TX PG V 4.2.0

D M A I C

3

Data Window

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.

DMAIC GB I TX PG V 4.2.0

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

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)

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.

DMAIC GB I TX PG V 4.2.0

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

parameters:

X Y

Factor Response

Independent Dependent

Input variable

By variable variables

graph variables

DMAIC GB I TX PG V 4.2.0

D M A I C

6

Using Minitab: A Typical Session

2. Type data into the worksheet

3. Import data files from other compatible software packages (such

as Excel)

4. Paste data from other applications

DMAIC GB I TX PG V 4.2.0

D M A I C

7

Using Minitab: A Typical Session

heading

C1 C2 C3 C4 C5 C6 C7 C8 C9 C10 C11 C12 C13 C14

Type data

directly into

worksheet

directly into the worksheet. For your Green Belt project,

you will probably import or copy the data from Excel.

DMAIC GB I TX PG V 4.2.0

D M A I C

8

Using Minitab: A Typical Session

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.

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.

DMAIC GB I TX PG V 4.2.0

D M A I C

9

Using Minitab: A Typical Session

1.In Excel:

Highlight Data (and Column Names) to be copied

Using Your Mouse

Edit > Copy (or CTRL-C on Your Keyboard)

3.Go to Minitab:

ALT > Tab

See example below.

Edit > Paste/Insert Cells (or Ctrl-V on the keyboard)

Insertion point

must make all data non-format specific

in Excel before you cut-paste.

DMAIC GB I TX PG V 4.2.0

D M A I C

10

Using Minitab: A Typical Session

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.)

these points in mind.

DMAIC GB I TX PG V 4.2.0

D M A I C

11

Using Minitab: A Typical Session

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 best format for analysis of data in Minitab is variable

columns.

point is that Minitab likes to have the data in columns.

DMAIC GB I TX PG V 4.2.0

D M A I C

12

Using Minitab: A Typical Session

File name: sales_data.xls

Sales 1999

Sales Office January February March April

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:

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.

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!

DMAIC GB I TX PG V 4.2.0

D M A I C

13

Using Minitab: A Typical Session

project are in the Stat and Graph menus.

DMAIC GB I TX PG V 4.2.0

D M A I C

14

Using Minitab: A Typical Session

DMAIC GB I TX PG V 4.2.0

D M A I C

15

Using Minitab: A Typical Session

Therefore, check both the session window & the graph (if

one exists).

DMAIC GB I TX PG V 4.2.0

D M A I C

16

Using Minitab: A Typical Session

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

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.

DMAIC GB I TX PG V 4.2.0

D M A I C

17

Using Minitab: A Typical Session

session window individually, or save an entire project.

DMAIC GB I TX PG V 4.2.0

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

DMAIC GB I TX PG V 4.2.0

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?

DMAIC GB I TX PG V 4.2.0

D M A I C

20

Review–Variation

Variation is a primary source of customer dissatisfaction

In order for our customers to “feel the quality” at GE, we must

reduce variation

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.

DMAIC GB I TX PG V 4.2.0

D M A I C

21

Using Data To Understand Variation

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

– Box Plot – Pie Chart – Control Chart – Run Chart

– Histogram

fluctuations in our processes.

description of the variation and its sources.

understanding variation is to plot the data.

DMAIC GB I TX PG V 4.2.0

D M A I C

22

Review–Continuous vs. Discrete Data

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

Upper Specification Limit = 30 Days

Continuous

Y = days to process

Discrete

Y = late/on-time USL

Actual Times

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

data:

– It gives us more information about our process.

– Smaller sample sizes are required when we use

continuous data.

– Which type of data would you want?

– Which type of data shows what you see?

would you prefer?

– Simply how many applications were processed

within the customer specification?

OR

–The exact time it took to process each application?

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0

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.

DMAIC GB I TX PG V 4.2.0

D M A I C

24

Two Types Of Variation

Characteristics

Always Present

Common Cause Expected

Predictable

Normal

Special Cause Unexpected

Unpredictable

Not Normal

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

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.

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.

DMAIC GB I TX PG V 4.2.0

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?

things:

– Center

– Spread

DMAIC GB I TX PG V 4.2.0

D M A I C

26

Statistics

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

DMAIC GB I TX PG V 4.2.0

D M A I C

27

Some Distributions

Uniform Distribution

P(X) Single roll of a die

are equally likely

Triangular Distribution

Sums of a pair of dice

P(X)

Rapidly descending

P(X), no tails

X

DMAIC GB I TX PG V 4.2.0

D M A I C

28

Distributions

Normal Distribution

P(X)

Process/repair times

operating point

X

Exponential Distribution

Time between arrivals

P(X)

Time between random

(unrelated) failures

to the next

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.

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?

DMAIC GB I TX PG V 4.2.0

D M A I C

29

Shape

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

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.

is of special interest.

DMAIC GB I TX PG V 4.2.0

D M A I C

30

The Normal Curve

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 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.

DMAIC GB I TX PG V 4.2.0

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%

68.26%

95.46%

99.73%

95.46% Fall Within +\- 2 Standard Deviation

99.73% Fall Within +\- 3 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.

standard deviation which represents approximately 34%

of whatever you are measuring.

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.

DMAIC GB I TX PG V 4.2.0

D M A I C

32

Normal Probability Plot

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

distribution is normal

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.

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0

D M A I C

33

Normal Probability Plot (continued)

Distribution Type:

Normal Bimodal curve Skewed Long-Tailed

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

(Stable Operations)

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.

bimodal, skewed, and long-tailed distribution?

DMAIC GB I TX PG V 4.2.0

D M A I C

34

What If Your Data Is Not Normal?

two general approaches:

Approach 1: Variance-based Thinking (VBT) Methodology

– possibly multiple processes embedded

– segmentation and stratification

– span reduction

Be Able To Do Approach 1

is important that you consult with your MBB before

proceeding.

DMAIC GB I TX PG V 4.2.0

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

Histogram: Variable: Cycle Time

Click OK

DMAIC GB I TX PG V 4.2.0

D M A I C

36

Minitab–Normal Probability Plot

Normal Probability Plot

Datafile: Fill_time.mtw

.999

.99

Question: Are the cycle time data .95

normally distributed?

Probability

.80

.50

.20

Tool: Normality Test

.05

.01

2 12 22 32

Average: 16.3921 Anderson-Darling Normality Test

StDev: 5.61675 A-Squared: 0.208

N: 240 P-Value: 0.864

Probability Plot: Statistics > Normality Test

Variable: Cycle Time

Test For Normality:

Choose Anderson-Darling

Click OK

standardized on this Anderson-Darling normality test. We

will discuss p-values later in the Analyze phase.

DMAIC GB I TX PG V 4.2.0

D M A I C

37

“Center” Or Central Tendency

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

visually examining a histogram.

be made using descriptive statistics such as the mean,

median, or quartile values. The distribution shape dictates

which metrics should be used.

DMAIC GB I TX PG V 4.2.0

D M A I C

38

“Center” Or Central Tendency (continued)

the most likely or expected value. The formula

for the mean is:

X=

∑X i

1. The sum of all data

values

values

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 :

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.

DMAIC GB I TX PG V 4.2.0

D M A I C

39

Variation

the process

Span Long-tailed Distribution

p=.01 p=.99

Skewed Distributions

Q1 Q3 Q1 Q3

“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.

basic descriptive statistics such as standard deviation,

span, or stability factor.

mean. Smaller is better.

median. Smaller is better.

1.0 is best.

DMAIC GB I TX PG V 4.2.0

D M A I C

Variation (continued) 40

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 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:

100-500 90%-10% P90, P10

501-2000 95%-5% P95, P5

2001-5000 98%-2% P98, P2

5001+ 99%-1% P99, P1

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0

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

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.

DMAIC GB I TX PG V 4.2.0

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

Deviation = measure of Variation, measure of Spread

SST = 8726.1 … large Deviations = large SST

s2 = Sample Variance = SST/9 = 969.6

σ2 = Population Variance = SST/10 = 872.6

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 population and a near average for a sample = SST/(n-1)

= SST/df.

same as larger sample size (n)), the near average and

average SST become essentially the same.

variance.

DMAIC GB I TX PG V 4.2.0

D M A I C

43

Summary–Measures

Summary–Measures Of Variation

Of Variation:

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 %

and unchanging.

i.e., within and between group.

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.

DMAIC GB I TX PG V 4.2.0

D M A I C

44

Variation For A Period Of Time

Descriptive Statistics Summary

(central tendency) (variation)

Normal Probability Plot for a Normal Distribution

ML Estimates

Mean:15.7224

StDev:1.74183

Standard

Percent

normal

Normal Probability Plot for an Exponential Distribution

ML Estimates

Mean: 15.0790

StDev:12.6232

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

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

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)

DMAIC GB I TX PG V 4.2.0

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

and described using basic statistical measures. The first

graphical display used for describing variation is the

histogram.

provide a clear picture of the current performance of a

process.

DMAIC GB I TX PG V 4.2.0

D M A I C

46

Displaying Variation For A Period Of Time (continued)

Box Plots

* Outlier

Highest Value

Each segment

represents

Median Q2

25% of the

data points First Quartile (25%) value Q1

Lowest Value

variation in sets of data.

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.

or outliers.

Outlier calculations

High: Q3 + 1.5 (Q3-Q1)

Low: Q1 - 1.5 (Q3-Q1)

DMAIC GB I TX PG V 4.2.0

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

in different graphical formats.

DMAIC GB I TX PG V 4.2.0

D M A I C

48

Minitab–Graphical Summary

Datafile: Fill_time.mtw

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

StDev

Variance

16.3921

5.6167

31.5478

Skewness 6.39E-02

Kurtosis -2.4E-01

N 240

4 10 16 22 28

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

15.7274 17.3441

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.

DMAIC GB I TX PG V 4.2.0

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

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.

order for run charts to be valid.

DMAIC GB I TX PG V 4.2.0

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

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

of type (A).

DMAIC GB I TX PG V 4.2.0

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

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).

Run Chart: Single Column: cycle time

Subgroup size: 1

p>0.05, there are no issues with Mixtures, Clusters,

Oscillations or Trends. Therefore, our process appears

stable (no special causes).

DMAIC GB I TX PG V 4.2.0

D M A I C

52

Two Types Of Variation

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

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.

DMAIC GB I TX PG V 4.2.0

D M A I C

53

Two Types Of Variation (continued)

How you interpret variation . . .

Common Causes Special Causes

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

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.

DMAIC GB I TX PG V 4.2.0

D M A I C

54

Graphical Analysis Tools

Continuous Y Discrete Y

Boxplot

Pareto Chart

Scatterplot

DMAIC GB I TX PG V 4.2.0

D M A I C

55

Box Plots

different shifts?

60

Measure

30

10

Shift 2 Shift 5

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?

DMAIC GB I TX PG V 4.2.0

D M A I C

56

Minitab–Box Plots

Datafile: Fill_time.mtw

in cycle time between agents?

30

Cycle Time 20

10

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?

DMAIC GB I TX PG V 4.2.0

D M A I C

57

Scatter Diagrams–Analyzing Relationships

Two Variables.

40

Cycle Time (Days) (Y)

35

30

25

20

15

10

5

1K 2K 3K 4K 5K 6K 7K 8K 9K 10K

exploring the relationship between two continuous

variables.

relationship (Larger Loans take longer to process–

positive relationship).

DMAIC GB I TX PG V 4.2.0

D M A I C

58

Warning! Correlation Does Not Imply Causation

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

each year against the number of storks observed in that

year, 1930-1936.

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.)

DMAIC GB I TX PG V 4.2.0

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

Too Slow Response Time Load

Airline Baggage Plane Baggage

Diagrams is mixing up the X and Y variables.

the horizontal axis.

axis.

DMAIC GB I TX PG V 4.2.0

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

See whether the potential cause variable and the effect

variable are related to one another.

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.

Check whether the data pattern indicates possible

problems in the data.

DMAIC GB I TX PG V 4.2.0

D M A I C

61

Interpreting A Scatter Diagram

1 3 5

2 4 6

X = Trainer experience (# of hours)

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

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0

D M A I C

62

Minitab–Scatter Plot

Datafile: Fill_time.mtw

between cycle time and

years experience? 30

Tool: Plot

Cycle Time 20

0

Scatter Plot: Y Cycle time

0 5 10 15 20 25

X Years

Yrs Exp

Experience Click OK

DMAIC GB I TX PG V 4.2.0

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

DMAIC GB I TX PG V 4.2.0

D M A I C

Strong,

Plot +/- Other Weak, Other

Example

1

Effect

Potential Cause

2

Effect

Potential Cause

3

Effect

Potential Cause

DMAIC GB I TX PG V 4.2.0

D M A I C

Strong, Weak,

Plot +/- Other Other

Example

4

Effect

Potential Cause

5

Effect

Potential Cause

6

Effect

Potential Cause

DMAIC GB I TX PG V 4.2.0

D M A I C

Other

7

Effect

Potential Cause

8

Effect

Potential Cause

9

Effect

Potential Cause

DMAIC GB I TX PG V 4.2.0

D M A I C

Weak, Other

10

Effect

Potential Cause

11

Effect

Potential Cause

12

Effect

Potential Cause

DMAIC GB I TX PG V 4.2.0

D M A I C

68

Pareto Charts

Frequency

C A E D B

Category of Defect

module.

DMAIC GB I TX PG V 4.2.0

D M A I C

69

Minitab–Pareto Chart

Datafile: Fill_time.mtw

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

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

Click: OK

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

DMAIC GB I TX PG V 4.2.0

D M A I C

DMAIC GB I TX PG V 4.2.0

D M A I C

1

Measure

Define 3–Plan

ReviewData Collection And Collect Data

Objectives 1

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.

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

Select CTQ Define Establish Data

Characteristic Performance Collection, Validate

Standards MSA

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 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.

DMAIC GB J TX PG V 4.2.0

D M A I C

Measurement System

Define Review Analysis (MSA) Worksheet

Objectives

3

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

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

1) Repeatability [ >90% ]

2) % Reproducibility [ >90% ]

3) % Accuracy [ >90% ]

_______________________________________________________________

1) Gage R&R Pass? Y N, If NO:

Plan for improvement:

_______________________________________________________________

_______________________________________________________________

© GE Capital, Inc., 1999

DMAIC GB J TX PG V 4.2.0

D M A I C

DMAIC GB J TX PG V 4.2.0

D M A I C

1

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

plan

Monitor data

accuracy and

consistency

Variation The Customer Feels

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

State the purpose of the data collection

Identify what data is required

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?

DMAIC GB K TX PG V 4.2.0

D M A I C

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

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.

DMAIC GB K TX PG V 4.2.0

D M A I C

Common Segmentation

Define Review Factors

Objectives

4

drivers of variation in the process

Possible categories:

Customer

Product

Market

Time

Geography

Size of account

Degree of dissatisfaction

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).

© GE Capital, Inc., 1999

DMAIC GB K TX PG V 4.2.0

D M A I C

Common Segmentation

Define Review Factors

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

pay special attention to include: relationship to

deadline (e.g., end of quarter), time/number in

queue, or timing versus systems changes.

several different ways in order to uncover where

the most significant differences occur.

DMAIC GB K TX PG V 4.2.0

D M A I C

How To Collect

Define Review Data For Segmentation

Objectives

6

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

DMAIC GB K TX PG V 4.2.0

D M A I C

Step 1: Breakout

Define Review Activity (5 Minutes)

Objectives

7

Instructions

– Brainstorm a list of segmentation factors

– Remember to also segment on “unlikely” parameters

Time 5 Minutes

DMAIC GB K TX PG V 4.2.0

D M A I C

Step 2: Develop

Define ReviewOperation Definitions and Procedures

Objectives

8

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

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.

DMAIC GB K TX PG V 4.2.0

D M A I C

Operational Definitions

Define Review Objectives

9

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

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.

DMAIC GB K TX PG V 4.2.0

D M A I C

Operational Definitions–Scale

Define Review Objectives Of Scrutiny

10

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

your process. Measuring one level smaller than

your customer allows you to more fully understand

and capture the variation in the process.

DMAIC GB K TX PG V 4.2.0

D M A I C

1

Sampling

Define Review Objectives

1

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

average we infer that the

resolution time? average

resolution time

( ) is 1.2 days

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.

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.

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)

From A Subset Of The Total Available Data

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.

plan do not justify sampling, then sampling should

not be done. This is often the case in low volume

processes (e.g., deal processes).

DMAIC GB L TX PG V 4.2.0

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

“representative” of the population.

accurately represents the target population.

Consideration that may hinder collection of a

representative sample include:

Time constraints

Unknown characteristics of the population

population are called biased samples. Often, the

biases are not recognized until the collected data has

been analyzed.

DMAIC GB L TX PG V 4.2.0

D M A I C

Sample

DefineSize For Continuous

Review ObjectivesData

4

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=

∆

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.

DMAIC GB L TX PG V 4.2.0

D M A I C

Common Segmentation

Define Review Factors

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

pay special attention to include: relationship to

deadline (e.g., end of quarter), time/number in

queue, or timing versus systems changes.

several different ways in order to uncover where

the most significant differences occur.

DMAIC GB L TX PG V 4.2.0

D M A I C

Calculating A Sample

Define Review Size

Objectives

6

– 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

DMAIC GB L TX PG V 4.2.0

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

n d2

2 1.128

3 1.693

4 2.059 σˆ = R /d 2

5 2.326

6 2.534

statistical control (stable) to use this estimate of σ

DMAIC GB L TX PG V 4.2.0

D M A I C

Sample

DefineSize For Continuous

Review ObjectivesData (continued)

8

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:

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.

DMAIC GB L TX PG V 4.2.0

D M A I C

Finite Population

Define Review Correction

Objectives

9

1. Calculate sample size (n)

OR Where n = sample size;

If n > N… N = population size

3. Calculate n finite

nfinite = n/(1+n/N)

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.

DMAIC GB L TX PG V 4.2.0

D M A I C

Sample

DefineSize For Discrete

Review Data

Objectives

10

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 )

∆

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

model and should only be used when nP > 5.

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.

DMAIC GB L TX PG V 4.2.0

D M A I C

Sample

DefineSize For Discrete

Review Data (continued)

Objectives

11

Population

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 (∆).

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

DMAIC GB L TX PG V 4.2.0

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

0 .5 1

P

DMAIC GB L TX PG V 4.2.0

D M A I C

Sample

DefineSize Considerations

Review Objectives

13

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

Is Not Representative Of The Process?

DMAIC GB L TX PG V 4.2.0

D M A I C

Sampling ExerciseObjectives

Define Review

14

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

DMAIC GB L TX PG V 4.2.0

D M A I C

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

your measurement system. A Gage R&R Study will help

us do this!

and tools to validate your measurement system.

DMAIC GB M TX PG V 4.2.0

D M A I C

Observations

Inputs Process Outputs Inputs Measureme

nt Outputs Measuremen

ts

Process

Data

Measurement Process

Variation

Variation

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)

The Measurement System Must First Be Identified And

Separated From That Of The Process

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

DMAIC GB M TX PG V 4.2.0

D M A I C

– 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

DMAIC GB M TX PG V 4.2.0

D M A I C

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)?

DMAIC GB M TX PG V 4.2.0

D M A I C

Performing An MSA 5

(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.

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.

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.

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.

DMAIC GB M TX PG V 4.2.0

D M A I C

standard

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

measurement system

where you are aiming?

same place?

hit the target in the same place?

you still able to shoot as accurately (consistency over

time)?

250 meters?

DMAIC GB M TX PG V 4.2.0

D M A I C

Conduct MSA

OK?

Corrective Actions

No

Yes

Continue Process

Improvement

For Attribute R&R:

and Accuracy analysis.

Review data and decisions

Institute stopgap measures

Correct mechanical or definitional errors

Institute or upgrade training

Measure gage over time to address stability

Study linearity of gage

DMAIC GB M TX PG V 4.2.0

D M A I C

– 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

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

is the method we’ll use to analyze Discrete

Data.

DMAIC GB M TX PG V 4.2.0

D M A I C

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.)

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.)

the sensitivity. More units have a minimal effect on

sensitivity. (Rather than measure 40 parts with

3 operators, measure 10 parts with 12 operators.)

DMAIC GB M TX PG V 4.2.0

D M A I C

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.

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.

DMAIC GB M TX PG V 4.2.0

D M A I C

Sample Answer Trial1 Trial2 Trial3 Trial1 Trial2 Trial3 Trial1 Trial2 Trial3

1 N N N N N N N N N N

2 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

6 N N N N N N N N N N

7 D N D N D D D D D D

8 N N N N D N D N N N

9 N N N N N N N N N N

10 N N N N N N N D N D

11 D D D D D D D D D D

12 D N N N D D D D D D

13 D D D D D D D D D D

14 N N N N N N N N N N

15 D D D D D D D D D D

16 N D D D N N N N N N

17 N N N N N N N N N N

18 N N N N 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

22 N N N N D D D N N N

23 N N N N D D D N N N

24 N N N N N N N D D D

25 N N N N N N N N N N

26 D D D D D D D D D D

27 N N N N N N N N N N

28 N N N N N N N N N N

29 N N N N N N N N N N

30 D N D N D D D D D D

31 D D D D D D D D D D

32 N N N N N N N N N N

33 N D N N D D N N N N

34 N N N N N N N N N N

35 N N N N N N N N D N

36 D D D D D D D D D D

37 N N N N N N N D N N

38 N N N N N N N N N D

39 N N N N N N N N N N

40 N N N N D D D N N N

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.

DMAIC GB M TX PG V 4.2.0

D M A I C

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

2 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

6 N N N Y N N N Y N N N Y

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

9 N N N Y N N N Y N N N Y

10 N N N Y N N N Y D N D N

11 D D D Y D D D Y D D D Y

12 N N N Y D D D Y D D D Y

13 D D D Y D D D Y D D D Y

14 N N N Y N N N Y N N N Y

15 D D D Y D D D Y D D D Y

16 D D D Y N N N Y N N N Y

17 N N N Y N N N Y N N N Y

18 N N N 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

22 N N N Y D D D Y N N N Y

23 N N N Y D D D Y N N N Y

24 N N N Y N N N Y D D D Y

25 N N N Y N N N Y N N N Y

26 D D D Y D D D Y D D D Y

27 N N N Y N N N Y N N N Y

28 N N N Y N N N Y N N N Y

29 N N N Y N N N Y N N N Y

30 N D N N D D D Y D D D Y

31 D D D Y D D D Y D D D Y

32 N N N Y N N N Y N N N Y

33 D N N N D D N N N N N Y

34 N N N Y N N N Y N N N Y

35 N N N Y N N N Y N D N N

36 D D D Y D D D Y D D D Y

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

39 N N N Y N N N Y N N N Y

40 N N N Y D D D Y N N N Y

Operator 1 0.925 Operator 2 0.950 Operator 3 0.800

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.

DMAIC GB M TX PG V 4.2.0

D M A I C

Sample Tr1 Tr2 TR3 Tr1 Tr2 Tr3 Tr1 Tr2 Tr3 Match?

1 N N N N N N N N N Y

2 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

6 N N N N N N N N N Y

7 N D N D D D D D D N

8 N N N D N D N N N N

9 N N N N N N N N N Y

10 N N N N N N D N D N

11 D D D D D D D D D Y

12 N N N D D D D D D N

13 D D D D D D D D D Y

14 N N N N N N N N N Y

15 D D D D D D D D D Y

16 D D D N N N N N N N

17 N N N N N N N N N Y

18 N N N N N N N N N Y

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

22 N N N D D D N N N N

23 N N N D D D N N N N

24 N N N N N N D D D N

25 N N N N N N N N N Y

26 D D D D D D D D D Y

27 N N N N N N N N N Y

28 N N N N N N N N N Y

29 N N N N N N N N N Y

30 N D N D D D D D D N

31 D D D D D D D D D Y

32 N N N N N N N N N Y

33 D N N D D N N N N N

34 N N N N N N N N N Y

35 N N N N N N N D N N

36 D D D D D D D D D Y

37 N N N N N N D N N N

38 N N N N N N N N D N

39 N N N N N N N N N Y

40 N N N D D D N N N N

Operator 1 Operator 2 Operator 3 0.525

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.

DMAIC GB M TX PG V 4.2.0

D M A I C

Sample Value Tr1 Tr2 Tr3 Tr1 Tr2 Tr3 Tr1 Tr2 Tr3 Match?

1 N N N N N N N N N N 0

2 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

6 N N N N N N N N N N 0

7 D N D N D D D D D D 2

8 N N N N D N D N N N 2

9 N N N N N N N N N N 0

10 N N N N N N N D N D 2

11 D D D D D D D D D D 0

12 D N N N D D D D D D 3

13 D D D D D D D D D D 0

14 N N N N N N N N N N 0

15 D D D D D D D D D D 0

16 N D D D N N N N N N 3

17 N N N N N N N N N N 0

18 N N N N N N N N N N 0

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

22 N N N N D D D N N N 3

23 N N N N D D D N N N 3

24 N N N N N N N D D D 3

25 N N N N N N N N N N 0

26 D D D D D D D D D D 0

27 N N N N N N N N N N 0

28 N N N N N N N N N N 0

29 N N N N N N N N N N 0

30 D N D N D D D D D D 2

31 D D D D D D D D D D 0

32 N N N N N N N N N N 0

33 N D N N D D N N N N 3

34 N N N N N N N N N N 0

35 N N N N N N N N D N 1

36 D D D D D D D D D D 0

37 N N N N N N N D N N 1

38 N N N N N N N N N D 1

39 N N N N N N N N N N 0

40 N N N N D D D N N N 3

# of N 5 2 4 6 5 5 5 4 4 40

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.

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0

D M A I C

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!)

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.

DMAIC GB M TX PG V 4.2.0

D M A I C

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

Gage R&R Results

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

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% ] 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.

DMAIC GB M TX PG V 4.2.0

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

Operational definitions

Maintain stability

Reproducibility

Operational definitions

Consistent use of gage

Training

Varying work environment

AUDIT-follow-up on training

Human/physical characteristics

Performance measures

Unclear requirements

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).

DMAIC GB M TX PG V 4.2.0

D M A I C

Accuracy

Error in master

Instrument used improperly by appraiser

Operational definition

Standard not understood

DMAIC GB M TX PG V 4.2.0

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

Gage R&R to be conducted. Use today’s technology to

help you in validating your measurement system.

DMAIC GB M TX PG V 4.2.0

D M A I C

Other Considerations 20

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.

competition is taped. A random sample of 10 performances

shown to five judges, two times each, would allow estimates of

both repeatability and reproducibility.

DMAIC GB M TX PG V 4.2.0

D M A I C

Other Considerations 21

Call centers field service calls from around the world at

centralized locations. Call duration and call quality are both

recorded and tracked very closely.

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.

DMAIC GB M TX PG V 4.2.0

D M A I C

Other Considerations 22

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.

and scored at a later date. For this recorded scenario, both

repeatability and reproducibility may be calculated.

DMAIC GB M TX PG V 4.2.0

D M A I C

Test-Retest Study

DMAIC GB M TX PG V 4.2.0

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.

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

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.

Tests. This study is done prior to a Gage R&R.

precision or accuracy problem.

working “Standard.” The same item is measured

repeatedly.

DMAIC GB M TX PG V 4.2.0

D M A I C

unacceptable because it lacks precision. Action is required to find

and remove the sources of this error, including the replacement of

the device.

of the test unit:

measurement error (repeatability error) in a measurement

system.

confidence intervals. There it indicates ½ width of a

confidence interval.

remove or replace the measurement device. Perhaps a

better or updated device can be used.

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).

DMAIC GB M TX PG V 4.2.0

D M A I C

Conduct MSA

OK?

Corrective Actions

No

Yes

Continue Process

Improvement

% Contribution, % Study and # Distinct Categories.

amount of variation in the Measurement System.

Review data and decisions

Institute stopgap measures

Correct mechanical or definitional errors

Institute or upgrade training

Measure gage over time to address stability

Study linearity of gage

DMAIC GB M TX PG V 4.2.0

D M A I C

Target Analogy

== Xbar

Xbar

==

True Value

Xbar

==

True Value

system have? What is the scatter? (standard deviation)

size of the part? (mean)

DMAIC GB M TX PG V 4.2.0

D M A I C

Measurement System 28

Observations

Inputs Outputs Inputs Measurement Outputs Measurements

Process Process Data

Variation Variation Variation

(Actual (Observed

Example variation) variation)

#1

#2

Measurement

System Variability

- Investigated

through “R&R

Study”

variation to the overall total observed variation.

variation. This variation comes from both the gage and

the operator.

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.

DMAIC GB M TX PG V 4.2.0

D M A I C

(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.

(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.

DMAIC GB M TX PG V 4.2.0

D M A I C

Measurement System. It represents the total variation in the

Measurement System.

R&R

EV AV

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

DMAIC GB M TX PG V 4.2.0

D M A I C

System variation?

99%

R&R

(5.15σ)

Specification Window (tolerance)

Measurement System Variation This Leaves Only 50%

For The Process Variation

tolerance with which we are working.

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.

DMAIC GB M TX PG V 4.2.0

D M A I C

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.

R&R Study in Minitab, analyze the results and draw

conclusions. We will use Capital Logistics’ study of the

delivery of shipments to customers.

Study in Minitab with a new set of data from the same

Capital Logistics’ study.

DMAIC GB M TX PG V 4.2.0

D M A I C

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.

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.

DMAIC GB M TX PG V 4.2.0

D M A I C

process...

To help answer this question the GB

established the following analysis:

each truck was supposed to arrive at the customer. The GB

gave a copy of the matrix to each of the three dispatchers.

from a radio truck, identifying himself as one of the 10 units

listed in the matrix..

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.

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.

file for analysis. (Gage R&R–continuous data.mtw).

DMAIC GB M TX PG V 4.2.0

D M A I C

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

DMAIC GB M TX PG V 4.2.0

D M A I C

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).

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.

from the target time

operator took the measurement.

DMAIC GB M TX PG V 4.2.0

D M A I C

Perform Gage R&R Study–ANOVA method

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).

DMAIC GB M TX PG V 4.2.0

D M A I C

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

DMAIC GB M TX PG V 4.2.0

D M A I C

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.

DMAIC GB M TX PG V 4.2.0

D M A I C

A. R&R% of Tolerance

1. R&R less than 10%–Measurement System “acceptable”

based on classification of Characteristic, Application,

Customer Input, etc.

Fishbone Diagram, remove Root Causes. Is there a better gage on the

market, is it worth the additional cost?

RULES OF THUMB:

find opportunities in process and training.

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.

DMAIC GB M TX PG V 4.2.0

D M A I C

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”

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?

DMAIC GB M TX PG V 4.2.0

D M A I C

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

DMAIC GB M TX PG V 4.2.0

D M A I C

% Comparison Table–Found in our Minitab Session Window

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

% 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).

DMAIC GB M TX PG V 4.2.0

D M A I C

Gage name:

Date of study :

Gage R&R (ANOVA) for Meas Reported by :

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

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

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

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.

this section

DMAIC GB M TX PG V 4.2.0

D M A I C

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?

DMAIC GB M TX PG V 4.2.0

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

Operational definitions

Calibrating too often

Granularity of the measure (e.g., nearest hour, minute, second)

Maintain stability

Reproducibility

Operational definitions

Consistent use of gage

Training

Varying work environment

AUDIT–follow-up on training

Human/physical characteristics

Performance measures

Unclear requirements

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).

DMAIC GB M TX PG V 4.2.0

D M A I C

Accuracy

Error in master

Instrument not calibrated

Worn components/trend or drift in master

Instrument used improperly by appraiser

Calibrating too often

Operational definition

Range of the master

DMAIC GB M TX PG V 4.2.0

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

Instrument used improperly by appraiser

DMAIC GB M TX PG V 4.2.0

D M A I C

system and make recommendations–keep, improve or replace the

measurement system.

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.

DMAIC GB M TX PG V 4.2.0

D M A I C

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.

DMAIC GB M TX PG V 4.2.0

D M A I C

utilized by the Capital Logistics dispatchers is “acceptable” or not

Preparation Choose a facilitator, scribe

MSA R&R study and record results

Use File: Gage R&R–Continuous

Data.mtw and use columns

C6-C9

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?

DMAIC GB M TX PG V 4.2.0

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

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

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

1) Repeatability [>90%]

2) % Reproducibility [>90%]

3) % Accuracy [>90%]

Gage R&R Pass? Y N, IF NO:

Plan for improvement:________________________________________

__________________________________________________________

DMAIC GB M TX PG V 4.2.0

D M A I C

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”

DMAIC GB M TX PG V 4.2.0

D M A I C

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

DMAIC GB M TX PG V 4.2.0

D M A I C

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!

Measurement System Itself

DMAIC GB M TX PG V 4.2.0

D M A I C

Desired Outcome: Outline of a Data Collection Plan and MSA for your

GB project

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

DMAIC GB M TX PG V 4.2.0

D M A I C

information that will not be covered in

the 3-Day Workshop

DMAIC GB M TX PG V 4.2.0

D M A I C

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 :

Gage R&R (ANOVA) for Meas Reported by :

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

Gage R&R Repeat Reprod Part-to-Part Part/Truck 1 2 3 4 5 6 7 8 9 10

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.)

equal.

(trucks.)

DMAIC GB M TX PG V 4.2.0

D M A I C

Average of 2 measurements

taken by operator 1. Gage name:

Date of study :

Gage R&R (ANOVA) for Meas Reported by :

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

70

200

60

100 50

0 40

Gage R&R Repeat Reprod Part-to-Part Part/Truck 1 2 3 4 5 6 7 8 9 10

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.

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.

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0

D M A I C

Gage name:

Date of study :

Gage R&R (ANOVA) for Meas Reported by :

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

Gage R&R Repeat Reprod Part-to-Part Part/Truck 1 2 3 4 5 6 7 8 9 10

A B C D

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

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.

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.

DMAIC GB M TX PG V 4.2.0

D M A I C

Gage name:

Date of study :

Gage R&R (ANOVA) for Meas Reported by :

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

Gage R&R Repeat Reprod Part-to-Part Part/Truck 1 2 3 4 5 6 7 8 9 10

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.

DMAIC GB M TX PG V 4.2.0

D M A I C

ANOVA TABLE

Two-Way ANOVA Table With Interaction

Source DF SS MS F P

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!

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).

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!)

DMAIC GB M TX PG V 4.2.0

D M A I C

% Comparisons Table

Source %Contribution %Study Var %Tolerance

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

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

“R&R

“R&Rasasaa%

%ofof

Tolerance”

Tolerance”

% Tolerance

Acceptable Ranges

Not Acceptable

30%

Questionable

10%

Acceptable

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.

For this Gage: %Tolerance GRR = 171.53%

This Gage Fails the Criteria for %Tolerance for the

Total Gage RR Piece.

DMAIC GB M TX PG V 4.2.0

D M A I C

%Tolerance: Graphically

LSL USL

%Tolerance = 10%

%Tolerance = 30%

%Tolerance = 70%

tolerance used up just from the measurement system.

DMAIC GB M TX PG V 4.2.0

D M A I C

+ 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

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

“R&R

“R&Ras

asaa%%of

ofTotal

TotalProcess

Process

Variation”

Variation” Further Explanation:

Source %Contribution

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

DMAIC GB M TX PG V 4.2.0

D M A I C

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 . )

“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.

DMAIC GB M TX PG V 4.2.0

D M A I C

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”

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?

DMAIC GB M TX PG V 4.2.0

D M A I C

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).

%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.

DMAIC GB M TX PG V 4.2.0

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

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

_______________________________________________________________

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.

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0

D M A I C

DMAIC GB M TX PG V 4.2.0

D M A I C

1

Analyze Module Objectives

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

Establish Define Identify

Process Performance Variation

Capability Objectives Sources

DMAIC GB N TX PG V 4.2.0

D M A I C

3

The12-Step Process

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

DMAIC GB N TX PG V 4.2.0

D M A I C

4

Analyze Phase Overview

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

This phase is important because it clearly defines how well the process is

currently performing and identifies how much the process will be improved.

Analyze 4: Establish process capability

Analyze 5: Define performance objectives

Analyze 6: Identify variation sources

DMAIC GB N TX PG V 4.2.0

D M A I C

5

Statistical Thinking

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

critical X’s on gains

– Variation

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

DMAIC GB N TX PG V 4.2.0

D M A I C

DMAIC GB N TX PG V 4.2.0

D M A I C

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.

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.

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

4.2 Calculate baseline sigma for project Y

DMAIC GB O TX PG V 4.2.0

D M A I C

Project Y

DMAIC GB O TX PG V 4.2.0

D M A I C

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?

statistical calculations are better descriptors of your data

set.

DMAIC GB O TX PG V 4.2.0

D M A I C

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.

DMAIC GB O TX PG V 4.2.0

D M A I C

DMAIC GB O TX PG V 4.2.0

D M A I C

the Output of a Process to

the Performance Standard

the process is of meeting customer CTQ’s by looking at

the probability that the process will produce a defect.

DMAIC GB O TX PG V 4.2.0

D M A I C

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

kinds of measure. By expressing process capability as a

“sigma” value we can compare different types of

processes.

DMAIC GB O TX PG V 4.2.0

D M A I C

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

continuous data.

are synonyms:

capability written as σLT or SigmaLT. This is no longer

standard usage to denote capability).

(σ or SigmaST is no longer recommended).

DMAIC GB O TX PG V 4.2.0

D M A I C

Calculate ZLSL and/or ZUSL

Determine probability of a defective

Determine ZBench

DMAIC GB O TX PG V 4.2.0

D M A I C

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

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.

prior to using this method. If not, you’ll need to use the

Discrete, DPMO method.

DMAIC GB O TX PG V 4.2.0

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

s 0.1 0.1

s 0.1 0.1

look up a Yield of 97.6%

ZST = 3.4

DMAIC GB O TX PG V 4.2.0

D M A I C

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

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

defective. 1-Probability of defective = Yield.

Capability.

DMAIC GB O TX PG V 4.2.0

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

DMAIC GB O TX PG V 4.2.0

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

DMAIC GB O TX PG V 4.2.0

D M A I C

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

DMAIC GB O TX PG V 4.2.0

D M A I C

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.

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:

shifts and other special causes are unlikely

DMAIC GB O TX PG V 4.2.0

D M A I C

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

(Tools > Measure phase)

DMAIC GB O TX PG V 4.2.0

D M A I C

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

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.

DMAIC GB O TX PG V 4.2.0

D M A I C

Process

Capability

T

T

LSL

LSL ± 6σ USL

USL

USL

µµ

3.4 ppm

TT

LSL USL

4.5σ

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.

means that 6 standard deviations fit between the mean

and the specification limit.

average is 4.5 sigma from either the LSL or USL.

Term Capability

Term Capability

DMAIC GB O TX PG V 4.2.0

D M A I C

Z=

σ

SL =USL

LSL λ= T (Target)

µ (Mean)

Z =

st (short-term)

lt (long-term) σst

σlt

Z LT =

σ LT Typical ZShift = 1.5

SL - m is the distance the specification is from the mean.

SL − T

ZST =

σ ST

SL - T is the distance the specification is from the target.

ZST is bigger than

ZShift = ZST − Z LT

DMAIC GB O TX PG V 4.2.0

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

Z table Z table Z table Z table

Z table Z table

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.

DMAIC GB O TX PG V 4.2.0

D M A I C

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

Note: This gives you ZST (the 1.5 shift was added to ZLT).

DMAIC GB O TX PG V 4.2.0

D M A I C

Minitab File: GB Case Study.mtw

Chart investigate the variation in the

project Y data over time

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

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.

DMAIC GB O TX PG V 4.2.0

D M A I C

Solution–Run Chart 24

DMAIC GB O TX PG V 4.2.0

D M A I C

Solution–Run Chart 25

1. Double click on

“C5”

Each Date is a sub-

group

3. Click ”OK”

DMAIC GB O TX PG V 4.2.0

D M A I C

Solution–Run Chart 26

data points.

Run Chart for time

100

time

0 Median

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

< .05, then the process is unstable.

This Run Chart indicates that the process is stable over time.

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.

DMAIC GB O TX PG V 4.2.0

D M A I C

Solution–Normality Testing 27

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

StDev: 43.1714 A-Squared: 0.130

N: 506 P-Value: 0.983

method if the P-value is > = .05

Time)

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.

DMAIC GB O TX PG V 4.2.0

D M A I C

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

95% Confidence Interval for Mu Maximum 120.046

95% Confidence Interval for Mu

-8.174 -0.633

-10 -5 0 95% Confidence Interval for Sigma

40.665 46.009

95% Confidence Interval for Median

95% Confidence Interval for Median

-9.369 0.532

Basic Statistics–Descriptive.)

Variable = Time

Click on “Graphs” & choose “Graphical Summary”

curve based on the data’s mean and standard deviation

–this may not be a fit to the actual distribution.

DMAIC GB O TX PG V 4.2.0

D M A I C

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.

DMAIC GB O TX PG V 4.2.0

D M A I C

Process Performance Process Demographics

Actual (LT)

Potential (ST) Date:

Reported by:

Project:

LSL USL

Department:

Process:

Characteristic:

Units:

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.

DMAIC GB O TX PG V 4.2.0

D M A I C

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.

DMAIC GB O TX PG V 4.2.0

D M A I C

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

DMAIC GB O TX PG V 4.2.0

D M A I C

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

Z ST = Z LT + Zshift & Z Shift all valid on

Z ST = Z Bench-ST the Process

Z LT = Z Bench-LT Report

DMAIC GB O TX PG V 4.2.0

D M A I C

High

Probability High

of Defective Probability of

Defective

Good Design

Capability

LSL USL

Low Low

Probability Probability

of Defective of Defective

LSL USL

DMAIC GB O TX PG V 4.2.0

D M A I C

Summary–Z-Value 35

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

ways to describe capability. σST = pooled standard

deviation and σLT = standard deviation.

DMAIC GB O TX PG V 4.2.0

D M A I C

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)

DMAIC GB O TX PG V 4.2.0

D M A I C

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

DMAIC GB O TX PG V 4.2.0

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

Clarification

Defect–Any single non-conformance (defect=black dot)

square with 1 or more black dots).

DMAIC GB O TX PG V 4.2.0

D M A I C

Formulas 39

DPU = D/U

9/4 = 2.25

Total Opportunities

TOP = U*OP

4*5 = 20

DPO = D/TOP

9/20 = .45

DPMO = DPO*1,000,000

.45*1,000,000 = 450,000

Table (next page) to get ZBenchST or ZBenchLT

DMAIC GB O TX PG V 4.2.0

D M A I C

ConvertingDPMO

Converting DPMOTo

ToZZST 40

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

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

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.)

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.

DMAIC GB O TX PG V 4.2.0

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

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

performance?

and the number of defects from each perspective.

correcting invoices is 100% effective.

DMAIC GB O TX PG V 4.2.0

D M A I C

42

Calculating Process Sigma Discrete Data Method

U = __________

3

D = __________

(include defects made and later fixed)

D ( 72 )

DPO = U∗OP = ( 500 ) ( 3 ) = 0.048

Abridged Process Sigma Conversion Table ZST = 3.1

Now calculate the DPMO and determine the

units for this example.

capability.

defect opportunities per unit–late, wrong amount, and

wrong address.

wrong amount, and 14 wrong address for a total of 72.

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.

.

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0

D M A I C

43

Exercise–Calculating Process Sigma Discrete Data

500

OP = __________

D = __________

D ( )

DPO = = =

U∗OP ( )( )

DPMO = DPO ∗ 1,000,000 = ( ) ∗ 1,000,000 =

Abridged Process Sigma Conversion Table ZST =

headings and your data:

30 500 3

Defects = Defects

Units = Units

Opportunities = Opportunities

3. Click “OK”

DMAIC GB O TX PG V 4.2.0

D M A I C

44

Activity–Calculating Process Capability–Discrete Data

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?

discrete data in Minitab, using the Six Sigma Product

Report.

DMAIC GB O TX PG V 4.2.0

D M A I C

45

Six Sigma Product Report

DMAIC GB O TX PG V 4.2.0

D M A I C

46

Discrete Data Examples

2. Select OK.

DMAIC GB O TX PG V 4.2.0

D M A I C

47

Six Sigma Product Report Output

Characteristic Defs Units Opps TotOpps DPU DPO PPM ZShift ZBench

U DPMO ZST

D

OP

provides two graphical charts for discrete data. The

product benchmarks allows one to see visually how the

ZBench values compare for each characteristic (8A).

D 89

DPO = = = 0.175889

U ∗ OP (506 )(1)

DPMO = DPO ∗ 1,000,000 = 175,889

Z = 2.4

ST

DMAIC GB O TX PG V 4.2.0

D M A I C

48

Yield

1. After first inspection: 1 passed, 3 failed

Rework 3 parts

Rework 2 parts

DMAIC GB O TX PG V 4.2.0

D M A I C

49

Types Of Yield

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 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

SDF = P(0) = e-DPU = e-2.25 = .1054 = 10.54%

DPU = Defects per unit = 9/4 = 2.25

DMAIC GB O TX PG V 4.2.0

D M A I C

throughput yield of 99 %. The rolled yield equals:

Process Centered Process Centered

Op 1 x Op 2 = Output

pass through both operations defect free.

independent, rolled throughput yield (YRT) = .99 x .99 =

.9808

DMAIC GB O TX PG V 4.2.0

D M A I C

Following Receiving

Inspection and Line Fall-out...

From Machining Operations

At Test Stands on

first attempt

Right

First

YRT = .9953*.9713 *.9452 = 91.38% Time

you have multiple processes, each time you go through

additional steps, you lose yield at the end of the process.

by multiplying all the yields together.

becomes.

DMAIC GB O TX PG V 4.2.0

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 )

Observed DPU Level

DPUO = DPUS x E (DPUO )

DPUE = DPUS x (1-E)

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.

observes (catches) a DPUO level of E*DPUS = Observed DPUS=100 DPUE=20

DPU.

(1-E)*DPUS.

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0

D M A I C

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)

what we caught at inspection. In this example DPUO =

.25.

slightly; in this example E = .8.

DPUS*.8 = DPUO

DMAIC GB O TX PG V 4.2.0

D M A I C

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.)

work through this exercise.

DMAIC GB O TX PG V 4.2.0

D M A I C

Exercise (continued) 55

Inspection

E = 0.95

Submitted Escaping

DPUS Level DPUE Level

1. Expressed as a decimal, what is the given acceptable maximum escaping defect

level? ___________

a. Units submitted to final test? ___________

b. Defects observed at final test? __________

DMAIC GB O TX PG V 4.2.0

D M A I C

Answers To Exercise 56

Escaping Defect Levels

1. DPUE = 1 = 0.002

500

2. Maximum acceptable submitted defects =

1 - effectiveness 1 -E 0.05

DMAIC GB O TX PG V 4.2.0

D M A I C

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)

DMAIC GB O TX PG V 4.2.0

D M A I C

Summary–Z-Shift 58

short-term capability

Our customers experience the long-term capability of the

process

To minimize shift, we need to reduce special cause variation

DMAIC GB O TX PG V 4.2.0

D M A I C

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.

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

DMAIC GB P TX PG V 4.2.0

D M A I C

Use benchmarking

Use other sources

DMAIC GB P TX PG V 4.2.0

D M A I C

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

DMAIC GB P TX PG V 4.2.0

D M A I C

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

DMAIC GB P TX PG V 4.2.0

D M A I C

methods, practices and processes, and either adopting or adapting

their good features and implementing them to become the “best of

the best”.

Benchmarking.

include:

Competitive Benchmarking

Product Benchmarking

Process Benchmarking

Best Practices Benchmarking

DMAIC GB P TX PG V 4.2.0

D M A I C

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

DMAIC GB P TX PG V 4.2.0

D M A I C

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

DMAIC GB P TX PG V 4.2.0

D M A I C

Select process and define defect and opportunities

Measure current process capability and establish goal

Understand detailed process that needs improvement

Outline industries/functions which perform your process

Formulate list of world class performers

Contact the organization and network through to key contact

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

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

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

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

DMAIC GB P TX PG V 4.2.0

D M A I C

Sources Of Information 9

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

DMAIC GB P TX PG V 4.2.0

D M A I C

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

DMAIC GB P TX PG V 4.2.0

D M A I C

Benchmarking Example 11

– 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

Characteristic Defs Units Opps TotOpps DPU DPO PPM ZShift ZBench

DMAIC GB P TX PG V 4.2.0

D M A I C

Summary–Analyze 5 12

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

DMAIC GB P TX PG V 4.2.0

D M A I C

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.

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

6.2 Narrow list of potential causes

DMAIC GB Q TX PG V 4.2.0

D M A I C

Transfer Function

Y = f (X)

that explains Y variables

in terms of X

Into The Vital Few X’s

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.

DMAIC GB Q TX PG V 4.2.0

D M A I C

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.

DMAIC GB Q TX PG V 4.2.0

D M A I C

DMAIC GB Q TX PG V 4.2.0

D M A I C

Process Map Analysis

Graphical

Analysis

Problem

Statement

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.

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.

DMAIC GB Q TX PG V 4.2.0

D M A I C

Continuous Y Discrete Y

Boxplot Pareto Chart

Scatterplot

Histogram

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.

DMAIC GB Q TX PG V 4.2.0

D M A I C

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)?

DMAIC GB Q TX PG V 4.2.0

D M A I C

Logistics case study data. File: Green Belt Case Study.mtw

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

Graph > Plot

X: Distance

Y: Time

DMAIC GB Q TX PG V 4.2.0

D M A I C

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.

Analysis step:

“Completely Satisfying Customer Needs” through a

customer-focused process improvement perspective.

DMAIC GB Q TX PG V 4.2.0

D M A I C

Moments Of Truth 10

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.

hard work, delays, and hand-offs which go into making

their loan a reality. It occurs behind the scenes.

DMAIC GB Q TX PG V 4.2.0

D M A I C

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.

Essential To The

Customer, But That

Allow The Value-

Adding Tasks To Be

Done Better/Faster.

Value-Enabling Work

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”

DMAIC GB Q TX PG V 4.2.0

D M A I C

Control/Inspection Move

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

DMAIC GB Q TX PG V 4.2.0

D M A I C

Flow Of Work 13

Process Time

+ Delay Time

Cycle Time

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.

DMAIC GB Q TX PG V 4.2.0

D M A I C

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.

DMAIC GB Q TX PG V 4.2.0

D M A I C

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

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.

DMAIC GB Q TX PG V 4.2.0

D M A I C

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

External Failure

Control/Inspection 9 9 8 .8%

Prep/Set-Up

Move 9 9 30 3.1%

Value-Enabling 9 1 .1%

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.

DMAIC GB Q TX PG V 4.2.0

D M A I C

logically organize possible causes for a specific problem or effect.

Problem

Statement

Analyze phase the Cause and Effect diagram is often

used to identify X’s.

DMAIC GB Q TX PG V 4.2.0

D M A I C

potential Root Cause(s)

Ask “why” five times to identify deeper causes

Use process data to answer each “why” question

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

Because customers don’t fill out the form

accurately or completely

completely?

Because the format is confusing

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0

D M A I C

case study data

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?

Why’s deeper causes as appropriate

results with the class

DMAIC GB Q TX PG V 4.2.0

D M A I C

IMPACT

C In Our

O Control

N

T

R

O

L

Out Of

Our Control

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

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?

the matrix.

Begin with the “High Impact/In Our Control” category.

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0

D M A I C

Example

Machines Methods Materials

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

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.

DMAIC GB Q TX PG V 4.2.0

D M A I C

timekeeper and/or note taker

Cause for the customer service desk project

And in the cause and effect diagram

Effect activity

Diagram

Potential X’s the next page to prioritize the X’s

identified by your team

out on your high impact X’s. Are all

of these X’s within the control of the

team?

DMAIC GB Q TX PG V 4.2.0

D M A I C

IMPACT

C In Our

O Control

N

T

R

O

L

Out Of

Our Control

DMAIC GB Q TX PG V 4.2.0

D M A I C

24

Analyze 6.2

Narrow list of Potential Causes

DMAIC GB Q TX PG V 4.2.0

D M A I C

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.

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.

of the people in the United States, what advantages would

exist in taking a sample?

DMAIC GB Q TX PG V 4.2.0

D M A I C

Hypothesis Testing–Introduction 26

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

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

DMAIC GB Q TX PG V 4.2.0

D M A I C

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]

results indicate a real, more-than-random effect in the

underlying population or process?

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.

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?

DMAIC GB Q TX PG V 4.2.0

D M A I C

variation. given process may

vary.

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.

adjustments for improvement, we need to validate actual

improvements in our processes.

calculated statistics (sample mean and standard deviation)

if there is a statistically significant difference between

processes.

objectively whether there is a difference.

DMAIC GB Q TX PG V 4.2.0

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

tests for a different kind of “difference.”

DMAIC GB Q TX PG V 4.2.0

D M A I C

Hypothesis Testing 30

concept of hypothesis testing.

This corresponds to the null hypothesis.

to convict the defendant. This corresponds to rejecting the null

hypothesis and accepting the alternate hypothesis.

Ha: person is guilty

DMAIC GB Q TX PG V 4.2.0

D M A I C

Nature Of Hypothesis 31

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 <

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.

DMAIC GB Q TX PG V 4.2.0

D M A I C

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:

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 : :

Alternate hypothesis statement:

Alternate hypothesis statement:

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.

DMAIC GB Q TX PG V 4.2.0

D M A I C

Hypothesis Testing 33

The only four possible outcomes:

1. An innocent person is set free. Correct decision

– The probability of this type of error occurring we represent as

– The probability of this type of error occurring we represent as

Type I error in a six-sigma project?

DMAIC GB Q TX PG V 4.2.0

D M A I C

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

outcomes of hypothesis testing.

DMAIC GB Q TX PG V 4.2.0

D M A I C

Hypothesis Testing 35

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 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.

If p ≤.05 reject H0 (difference exists).

Reject Ho Accept Ho

Different Not Different

0.00 0.05 1.00

p-value

DMAIC GB Q TX PG V 4.2.0

D M A I C

Some basic statistical tests are shown below with the command for running each

test in Minitab.

Histogram

Mean of population data is 1-Sample t-test

different from an established Stat > Basic Statistics

target. > 1-Sample t

Histogram

population 2. > 2-Sample t

Histogram

Homogeneity of

populations is different. Scatter

Variance

Stat > ANOVA >

Plots

Homogeneity of Variance

input (X) changes. Regression Box

Stat > Regression >Fitted Plots

Line Plot

Chi-Square Test

Frequency

more subgroups differ. of Independence

Stat > Tables >

Cross Tabulation OR

Chi-Square Test C AB D E M NO

Category

distributed Stat > Basic

Statistics

DMAIC GB Q TX PG V 4.2.0

D M A I C

potential X’s

Y

Continuous Discrete

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)

DMAIC GB Q TX PG V 4.2.0

D M A I C

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 relation- ship between

two variables.

Regression–Defines the linear relationship

between a dependent and independent

variable.

(Here, “Normality” applies to the residuals of

the regression.)

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.

look at in detail during this training. We have already

looked at the Normality Test in previous modules.

DMAIC GB Q TX PG V 4.2.0

D M A I C

2. Check Shape–Anderson-Darling Normality Test

3. Check Spread–Appropriate Hypothesis Test

4. Check Center–Appropriate Hypothesis Test

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.

DMAIC GB Q TX PG V 4.2.0

D M A I C

Discrete? normal?

Mood’s Median

HOV

NO YES

Regression

Are Y’s NO CHI

Continuous?

SQUARE (χ2)

YES

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

Ho = µ 1 = µ 2 = ...µ n

Ha = At least one µ is Different

DMAIC GB Q TX PG V 4.2.0

D M A I C

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.

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?

DMAIC GB Q TX PG V 4.2.0

D M A I C

Data Analysis 42

DMAIC GB Q TX PG V 4.2.0

D M A I C

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:

C4-C8– processing time per territory noted

data by territory.

time order or sequentially ordered.

DMAIC GB Q TX PG V 4.2.0

D M A I C

Stability–Minitab

Stability–Minitab Run Chart

Run Chart 44 44

1. Double

Click.

2. Type in a “1.”

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.

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0

D M A I C

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

same conclusions as the confidence interval.

criteria we use most often in this course.

DMAIC GB Q TX PG V 4.2.0

D M A I C

Data Analysis 46

DMAIC GB Q TX PG V 4.2.0

D M A I C

distributed?

MINITAB FILE: GB ANOVA.MPJ

Ha: The data is not normally distributed

DMAIC GB Q TX PG V 4.2.0

D M A I C

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

95% Confidence Interval for Mu Maximum 45.0000

95% Confidence Interval for Mu

8.6366 15.2330

7 8 9 10 11 12 13 14 15 16 95% Confidence Interval for Sigma

9.2120 13.9887

95% Confidence Interval for Median

95% Confidence Interval for Median

6.9173 12.0000

data is normally distributed. In this case, our p-value is

0.000, therefore, we have a non-normal distribution.

DMAIC GB Q TX PG V 4.2.0

D M A I C

Non-Parametric Tests 49

and

Mood’s Median for Center

used in analyzing it

However, responding correctly is critical if the data is

not normal (which is not at all uncommon)

DMAIC GB Q TX PG V 4.2.0

D M A I C

Non-Parametric Tests 50

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

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

DMAIC GB Q TX PG V 4.2.0

D M A I C

Spread–Test On Variances 51

variation of a process, or we might want to know if a change in a

variable (X) changes the variation in the output (Y).

also a prerequisite to the test on means.

Xb

Xa

is the Homogeneity of Variance Test

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.

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?

DMAIC GB Q TX PG V 4.2.0

D M A I C

Spread–Homogeneity Of Variance 52

You have established that each process is stable but non-normally

distributed. Perform the Homogeneity of Variance Test .

DMAIC GB Q TX PG V 4.2.0

D M A I C

2 = 2 = 2 = 2 = 2

Ho : σEast σ West σ IFG σ Central σ South

Ha : at least one is different α = .05

DMAIC GB Q TX PG V 4.2.0

D M A I C

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

East

P-Value : 0.000

IFG

Levene's Test

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: Accept Ho

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 standard deviation, even though we are comparing

variances (variance = σ2).

DMAIC GB Q TX PG V 4.2.0

D M A I C

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.

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.

DMAIC GB Q TX PG V 4.2.0

D M A I C

MINITAB FILE: GB ANOVA.MPJ

– Response = Total Time

– Factor = Territory

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

a

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.

normal data is called a Mood’s Median Test.

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.

DMAIC GB Q TX PG V 4.2.0

D M A I C

DMAIC GB Q TX PG V 4.2.0

D M A I C

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.

Ha: At least one median is not equal

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.

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

test using this summary table are the same as the

Mood’s Median test

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0

D M A I C

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

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

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.

DMAIC GB Q TX PG V 4.2.0

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

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.

DMAIC GB Q TX PG V 4.2.0

D M A I C

1-Sample t-Test 61

Ho : µ = constant = T

Ha : µ ≠ constant = T

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

DMAIC GB Q TX PG V 4.2.0

D M A I C

1-Sample t-Test 62

DMAIC GB Q TX PG V 4.2.0

D M A I C

1-Sample t-Test 63

1. Double click

anywhere on the C5

line to select time as

a variable

enter 5.0 as May’s

average, the

Alternative is that

they are not equal

a pictorial

4. Click on

Histogram of data

5. Click OK

DMAIC GB Q TX PG V 4.2.0

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

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.

DMAIC GB Q TX PG V 4.2.0

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

DMAIC GB Q TX PG V 4.2.0

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

average delivery time between the two different load

types.

and then hooks into an empty trailer to take back.

to unload it’s freight, and brings back the trailer.

DMAIC GB Q TX PG V 4.2.0

D M A I C

2-Sample t-Test 67

DMAIC GB Q TX PG V 4.2.0

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.

DMAIC GB Q TX PG V 4.2.0

D M A I C

2-Sample t-test 69

(means are indicated by solid circles)

100

time

-100

D/H Unload

LoadType

than .05, we reject the Null Hypothesis and accept its

alternate, that there is a statistical difference in our Drop

& Hook and Unload averages.

DMAIC GB Q TX PG V 4.2.0

D M A I C

6. Ho : µ1 = µ2 = . . . = µn

DMAIC GB Q TX PG V 4.2.0

D M A I C

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.

DMAIC GB Q TX PG V 4.2.0

D M A I C

DMAIC GB Q TX PG V 4.2.0

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.

DMAIC GB Q TX PG V 4.2.0

D M A I C

(means are indicated by solid circles)

100

time

-100

Driver

Trans, Inc.

Temp Labor

Cap. Log.

JJ Truck

DNGY Inc.

number for each

group.

2. These 3 are

alike.

3. These 2 are

definitely

different.

than .05, we reject the Null Hypothesis and accept its

alternate, that there is a difference in at least 1 average

between the 5 different groups of drivers.

DMAIC GB Q TX PG V 4.2.0

D M A I C

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.

DMAIC GB Q TX PG V 4.2.0

D M A I C

DMAIC GB Q TX PG V 4.2.0

D M A I C

on the C5 line to select

time as a response

variable.

Driver as a factor.

3. Click OK.

DMAIC GB Q TX PG V 4.2.0

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.

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.

DMAIC GB Q TX PG V 4.2.0

D M A I C

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

linear relationships.

–1 ≤r ≤1

correlated.

No linear correlation: r = 0

Perfect positive relationship: r = +1.0

regression, rational sub-grouping, etc. to try to extract

some meaning from the data.

DMAIC GB Q TX PG V 4.2.0

D M A I C

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

DMAIC GB Q TX PG V 4.2.0

D M A I C

Y = b 0 + b 1X 1

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

between our continuous Y metric–time and a continuous

X metric–distance.

DMAIC GB Q TX PG V 4.2.0

D M A I C

DMAIC GB Q TX PG V 4.2.0

D M A I C

the C5 line to select time

as a response variable.

Distance as a

predictor.

3. We believe there to

be a linear

relationship.

4. Click OK.

DMAIC GB Q TX PG V 4.2.0

D M A I C

Regression Plot

Y = -29.1923 + 0.174971X

R-Sq = 10.5 %

100

that 10.5% of the

variation in Y - time

is explained by this X

time

0 - distance.

-100

positive relationship 0 100 200 300 400

between distance

and timely delivery Distance

Distance is a

significant factor.

Entire model is

significant.

than .05, we reject the Null Hypothesis and accept its

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).

“Regression”)

Ho: Model = O

Ha: Model ≠ O

DMAIC GB Q TX PG V 4.2.0

D M A I C

Chi-Square Test 85

Y

Continuous Discrete

Continuous

X

Discrete

Chi-Square Test

DMAIC GB Q TX PG V 4.2.0

D M A I C

Chi-Square Tests 86

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)

number of categories

For all cases, Ho: no difference in data

Ha: difference exists