You are on page 1of 650

D M A I C

Classroom Session Agenda–Day 1 1

Total Topic Step


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

60 Measure 1–CTQ Tools continued 1


10 Break
45 Measure 2–Performance Standards 2
5 Break
120 Minitab Tutorial and Graphical
Analysis various

© GE Capital, Inc., 2000

DMAIC GB A TX PG V4.2.0
D M A I C

Classroom Session Agenda–Day 2* 2

Total Topic Step


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

60 Measurement System Analysis (MSA) 3


15 Break
15 Analyze Overview 4,5,6
150 Analyze 4–Establish Process Capability 4

* Elevator Speeches will be shared throughout the


day

© GE Capital, Inc., 2000

DMAIC GB A TX PG V4.2.0
D M A I C

Classroom Session Agenda–Day 3* 3

Total Topic Step


Min
30 Day 2–General Review various
15 Analyze 5–Define Performance Objective 5
15 Break
150 Analyze 6–Identify Variation Sources 6

30 Analyze 6–continued 6
15 Improve Preview 7,8,9
30 Improve 7 and 8–Design Of Experiment 7,8
15 Break
30 Improve 9–Statistical Tolerancing 9
15 Control Preview 10,11,12
60 Control 12–Control Charts 12

* Elevator Speeches will be shared throughout the


day

© GE Capital, Inc., 2000

DMAIC GB A TX PG V4.2.0
D M A I C

© GE Capital, Inc., 2000

DMAIC GB A TX PG V4.2.0
D M A I C

Assessment Strategy
Introductions
1

The Learning Objectives are divided in 3 categories:


1. Classroom
2. Project, and
3. Test

ƒ It is expected that a Green Belt be proficient in all learning


objectives, however, only those designated Test represent
the content of the certification exam.
ƒ Each phase of the DMAIC cycle is listed below and the
learning objectives have classified into the 3 categories
within each phase.

© GE Capital, Inc., 2000

DMAIC GB B TX PG V 4.2.0
D M A I C

Overview–Classroom Learning Objectives 2

1 Explain the benefits of Six Sigma to GE’s business.


ƒ Compare and contrast Six Sigma’s process improvement approach to quality with traditional
defect prevention strategies (i.e., inspection and testing).
ƒ 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.
ƒ Explain the relationship between increasing levels of process complexity and quality
improvement results.

2 Describe the Six Sigma Methodology for quality improvement.


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

© GE Capital, Inc., 2000

DMAIC GB B TX PG V 4.2.0
D M A I C

Overview–Project Learning Objectives 3

1. Determine if DMAIC is the right strategy by identifying the


conditions under which the DFSS methodology would be more
appropriate.
ƒ Compare and contrast the DFSS design methodologies to
DMAIC.

© GE Capital, Inc., 2000

DMAIC GB B TX PG V 4.2.0
D M A I C

Overview–Test Learning Objectives 4

1 Describe the 5 phases of DMAIC, including the purpose, tools,


and outputs for each phase.

© GE Capital, Inc., 2000

DMAIC GB B TX PG V 4.2.0
D M A I C

Define–Project Learning Objectives 5

1 Identify project CTQs.


ƒ Define “CTQ” (Critical to Quality Characteristic).
ƒ 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.
ƒ 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).

© GE Capital, Inc., 2000

DMAIC GB B TX PG V 4.2.0
D M A I C

Define–Project Learning Objectives (continued) 6

2 Use a process/product drill-down tree to: define the limits of a


project (project bounding); clarify what the project is and is not;
identify other areas for improvement.
ƒ Create a process/product drill-down tree.
ƒ Integrate measurements to clarify areas needing improvement.
ƒ Given an example of a process/product drill down tree, identify
viable Six Sigma projects.

© GE Capital, Inc., 2000

DMAIC GB B TX PG V 4.2.0
D M A I C

Define–Project Learning Objectives (continued) 7

3 Develop a team charter.


ƒ Describe the purpose of a charter.
ƒ Identify the five major elements of a charter.
ƒ 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.
ƒ Develop a problem statement.
ƒ Describe the customer’s pain.
ƒ Identify key considerations and potential pitfalls to consider when developing a
problem statement.
ƒ Develop a SMART goal statement (specific, measurable, attainable, relevant, time
bound).
ƒ Assess the scope of the project.
ƒ Identify the 8 steps for bounding a project.
ƒ Define project milestones.
ƒ Select a project team and define team roles.
ƒ Identify team roles and responsibilities.
ƒ Evaluate a proposed Six Sigma project.
ƒ Recognize characteristics of a “good” project.
ƒ Recognize characteristics of a “bad” project.

© GE Capital, Inc., 2000

DMAIC GB B TX PG V 4.2.0
D M A I C

Define–Project Learning Objectives (continued) 8

4 Map a business process.


ƒ Describe the goal of process mapping.
ƒ Identify the components of a process map (COPIS).
ƒ Describe the steps involved in creating a process map.
ƒ Define and name a process.
ƒ 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

© GE Capital, Inc., 2000

DMAIC GB B TX PG V 4.2.0
D M A I C

Define–Project Learning Objectives (continued) 9

5 Obtain approval for a Six Sigma project.


ƒ Identify the steps in the project approval process.
ƒ Enter a project into QPT.

© GE Capital, Inc., 2000

DMAIC GB B TX PG V 4.2.0
D M A I C

Define–Test Learning Objectives 10

1 Recognize the components of the 12 Step Process and how they


may be applied to a Six Sigma project.
2 Describe the purpose of the define phase and it’s key
deliverables: CTQs, team charter, and process map.
ƒ Identify the five key objectives of the Define Phase.
3 Describe the CAP tools and their connection to Six Sigma.

© GE Capital, Inc., 2000

DMAIC GB B TX PG V 4.2.0
D M A I C

Measure1–Classroom Learning Objectives 11

1 Describe and define the deliverables of Step 1.

© GE Capital, Inc., 2000

DMAIC GB B TX PG V 4.2.0
D M A I C

Measure 1–Project Learning Objectives 12

1 Identify the project Y.


ƒ Identify the tools that may be used to select the relevant CTQ
or Y on which to focus.
ƒ Explain the purpose Quality Function Deployment, Process
Map, and FMEA tools have.
ƒ Define performance standards for Y including specification
limits as well as defect and opportunity definitions.

© GE Capital, Inc., 2000

DMAIC GB B TX PG V 4.2.0
D M A I C

Measure 1–Project Learning Objectives (continued) 13

2 Select CTQ characteristics.


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

3 Select and apply appropriate tools to narrow the focus of a Six


Sigma project by identifying key areas for improvement.
ƒ Identify tools that may be used to narrow the focus of a project,
including Process Map, and FMEA.
ƒ Recognize the purpose and benefits of each tool.

© GE Capital, Inc., 2000

DMAIC GB B TX PG V 4.2.0
D M A I C

Measure 1–Project Learning Objectives (continued) 14

4 Use Process Mapping to identify potential breakdowns, rework loops, and


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

© GE Capital, Inc., 2000

DMAIC GB B TX PG V 4.2.0
D M A I C

Measure 1–Project Learning Objectives (continued) 15

5 Use a Failure Modes and Effects Analysis to identify the potential


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

© GE Capital, Inc., 2000

DMAIC GB B TX PG V 4.2.0
D M A I C

Measure 1–Project Learning Objectives (continued) 16

6 Conduct a test-retest study and analyze the results.


ƒ Describe the purpose and procedure for conducting a test-
retest study.
ƒ Plot and test-retest study data.
ƒ Use descriptive statistics to evaluate test-retest study data.

7 Establish a Data Collection Plan for a Six Sigma project.


ƒ Describe the purpose and benefits of a Data Collection Plan.
ƒ Write a data collection strategy.
ƒ Define a clear strategy for collecting reliable data efficiently.

© GE Capital, Inc., 2000

DMAIC GB B TX PG V 4.2.0
D M A I C

Measure–Test Learning Objectives 17

1 Recall the DMAIC 12 Step process, and distinguish between characterization


phases (DMA) and optimization phases (IC).
ƒ Define product characterization.
ƒ Define process optimization.

2 Recognize how statistics can be applied to the problem solving process.


ƒ Define the terms “precision” and “accuracy” as they relate to a Six Sigma
process.
ƒ Relate precision/variation and accuracy/mean to quality and customer
satisfaction.
ƒ State the goal of Six Sigma in statistical terms.
ƒ Define the term “Upper Specification Limit (USL).”
ƒ Define the term “Lower Specification Limit (LSL).”
ƒ Define the term “target (T).”
ƒ Define “σ.”

© GE Capital, Inc., 2000

DMAIC GB B TX PG V 4.2.0
D M A I C

Measure–Test Learning Objectives 18

3 Explain how statistics can be used to solve problems.


ƒ Identify project variables using the formula Y = f (X1,…..,Xn ).
ƒ Describe the relationship between any dependent variable (Y)
and independent variables (X).
ƒ Explain how the shape, mean, and standard deviation
characterize a process.
ƒ Express the capability of a process in terms of a standard
measure (z-value).
ƒ Define hidden factories and how capability impacts cycle time.

4 Identify the key deliverables of the Measure phase of DMAIC.

© GE Capital, Inc., 2000

DMAIC GB B TX PG V 4.2.0
D M A I C

Measure–Test Learning Objectives 19

5 Relate and apply the Quality Function Deployment (QFD) process to Six
Sigma.
ƒ Explain the purpose of QFD.
ƒ Describe the phases of QFD.
ƒ Explain QFD flowdown for product and service applications.
ƒ Generate/build a House of Quality (Product Planning Chart).
ƒ Identify what the customer wants (the “what’s”).
ƒ Identify the functions or processes that impact customer wants (the how’s).
ƒ Evaluate the impact of each function/process on customer wants.
ƒ Calculate the overall magnitude of the impact each function/process has on
customer wants (prioritize actions).
ƒ Analyze and diagnose a completed House of Quality.
ƒ Describe other QFD applications.
ƒ Determine when QFD is appropriate to use.
ƒ Recognize QFD pitfalls.
ƒ Describe an example of QFD from GE Medical Systems.

6 Describe and define the deliverables of Step 2.

© GE Capital, Inc., 2000

DMAIC GB B TX PG V 4.2.0
D M A I C

Measure–Test Learning Objectives 20

7 Define Performance Standards for a Six Sigma project.


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

8 Describe and define the deliverables of Step 3.

© GE Capital, Inc., 2000

DMAIC GB B TX PG V 4.2.0
D M A I C

Measure–Test Learning Objectives 21

9 Establish the accuracy of the measurement system and the data


(Analyze the measurement system).
ƒ Describe measurement as a process that includes
Measurement, Analysis, Improvement, and Control phases.
ƒ Describe measurement as a system that includes operators,
gages, and environment.
ƒ Define the terms “(gage) resolution,” “precision,” “accuracy,”
and “bias” as used in Measurements System Analysis (MSA).
ƒ Using the MSA checklist, document the existing measurement
system.
ƒ Recognize the sources of variation in a measurement system.

© GE Capital, Inc., 2000

DMAIC GB B TX PG V 4.2.0
D M A I C

Measure–Test Learning Objectives 22

10 Conduct a Gage R&R study and analyze the results.


ƒ Develop and implement a Data Collection Plan to collect Gage
R&R study data.
ƒ Describe equipment and appraiser sources of variation.
ƒ Describe the total R&R variation in terms of Reproducibility
(AV) and Repeatability (EV).
ƒ Set up, collect, and enter data into a Minitab data sheet.
ƒ Calculate both the appraiser variation (reproducibility) and
equipment variation (repeatability).
ƒ Describe the concepts of stability and linearity in gage studies.
ƒ Compare R&R variation to the tolerance (specification window).
ƒ Create graphs and charts (ANOVA method) to analyze study
results.
ƒ Recall and apply rules of thumb (guidelines) for analyzing R&R
study results.

© GE Capital, Inc., 2000

DMAIC GB B TX PG V 4.2.0
D M A I C

Measure–Test Learning Objectives 23

11 Describe and define the use of gage R& R for discrete data.
ƒ Describe the use of the attribute R & R spreadsheet for discrete
data.

12 Recall and list the deliverables of the Measure phase of


DMAIC.

© GE Capital, Inc., 2000

DMAIC GB B TX PG V 4.2.0
D M A I C

Analyze–Classroom Learning Objectives 24

1 Perform hypothesis testing for a continuous Y and discrete X.


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

2 Perform hypothesis testing for a discrete Y and discrete X.


ƒ Use chi-square testing to determine the goodness-of-fit and as
a test of independence.
ƒ Based on the chi-square test, determine whether or not to
accept the null hypothesis or the alternative hypothesis.

© GE Capital, Inc., 2000

DMAIC GB B TX PG V 4.2.0
D M A I C

Analyze–Classroom Learning Objective (continued) 25

3 Perform hypothesis testing for a continuous Y and continuous X.


ƒ Use a scatterplot to determine correlations between variables.
ƒ Use a linear regression analysis to quantify correlations and
predict values.
ƒ Determine process stability with run charts and other tools.
ƒ Determine whether or not to accept the null hypothesis or the
alternative hypothesis.
ƒ Describe the use of multiple regression for this type of data.
ƒ Describe the implications of multiple regression in statistical
analysis.

© GE Capital, Inc., 2000

DMAIC GB B TX PG V 4.2.0
D M A I C

Analyze–Project Learning Objectives 26

1 Use benchmarking to assist in developing project goals.


ƒ Describe the purpose of benchmarking.
ƒ Describe the uses of five different types of benchmarking, including
competitive benchmarking, product benchmarking, process benchmarking,
best practices benchmarking, strategic benchmarking, and parameter
benchmarking.
ƒ Apply benchmarking methodology to a variety of situations.
ƒ List potential sources of benchmarking data and how to access such
sources.
ƒ 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.
ƒ Describe the methods used to set project goals.
ƒ Use GE Standards for defect reduction in combination with benchmark
results to determine project goals.
ƒ Use the process entitlement to validate the achievability of the project goals.

© GE Capital, Inc., 2000

DMAIC GB B TX PG V 4.2.0
D M A I C

Analyze–Test Learning Objectives 27

3 Develop consensus within a project team on the acceptability of


the project goals.
ƒ 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.
ƒ Recognize the purpose and benefits of a Cause & Effect
diagram.
ƒ Write a problem statement.
ƒ Brainstorm categories appropriate to a problem.
ƒ Recognize the 4 Ps: policies, procedures, people and plant.
ƒ Brainstorm and analyze causes for each category to identify the
most likely cause(s) of a problem.
ƒ Determine which causes need to be verified with data.

© GE Capital, Inc., 2000

DMAIC GB B TX PG V 4.2.0
D M A I C

Analyze–Test Learning Objectives 28

4 Use a Pareto Chart to separate the vital few from the trivial many
in a process to determine where to focus improvement efforts.
ƒ Recognize the purpose and benefits of a Pareto Chart.
ƒ Describe the Pareto Principle.
ƒ Describe the steps involved in building a Pareto Chart:
ƒ Collect data.
ƒ Total results and arrange data in descending order.
ƒ Draw and label a Pareto Chart.
ƒ Analyze results.
ƒ Compare before and after Pareto Chart to evaluate
improvement effectiveness.

5 Describe process map analysis.


ƒ Describe value added/ non-value added analysis.

© GE Capital, Inc., 2000

DMAIC GB B TX PG V 4.2.0
D M A I C

Analyze–Test Learning Objectives 29

6 Identify variation sources.


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

© GE Capital, Inc., 2000

DMAIC GB B TX PG V 4.2.0
D M A I C

Analyze–Test Learning Objectives 30

1 Describe and define the deliverables of Step 4.

2 Apply statistical principles of the Standard Normal Probability Distribution to


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

© GE Capital, Inc., 2000

DMAIC GB B TX PG V 4.2.0
D M A I C

Analyze–Test Learning Objectives (continued) 31

3 Characterize a process using discrete data.


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

© GE Capital, Inc., 2000

DMAIC GB B TX PG V 4.2.0
D M A I C

Analyze–Test Learning Objectives (continued) 32

4 Use Process Centering strategies to perform a capabilities analysis.


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

© GE Capital, Inc., 2000

DMAIC GB B TX PG V 4.2.0
D M A I C

Analyze–Test Learning Objectives (continued) 33

5 Determine process capability.


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

© GE Capital, Inc., 2000

DMAIC GB B TX PG V 4.2.0
D M A I C

Analyze–Test Learning Objectives (continued) 34

6 Describe and define the deliverables of Step 5.

7 Describe and define the deliverables of Step 6.

8 Describe the statistical analysis tools and process for


normal/non-normal data.
ƒ Describe the use of the normality test.
ƒ Describe the use of Mood’s median for non-normal data.

© GE Capital, Inc., 2000

DMAIC GB B TX PG V 4.2.0
D M A I C

Improve–Project Learning Objectives 35

1 Characterize X’s as either operating parameters or critical


elements.

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


parameters.
ƒ Develop a mathematical model of a proposed solution.
ƒ Determine the best configuration or combination of X’s

3 Develop a strategy for those X’s identified as critical elements.


ƒ Optimize process flow issues.
ƒ Standardize the process.
ƒ Develop a practical solution.
ƒ Explain the needs and process to do screening experiments.
The implications for this in DOE is described.

© GE Capital, Inc., 2000

DMAIC GB B TX PG V 4.2.0
D M A I C

Improve–Project Learning Objectives 36

4 Perform optimizing experiments in order to develop a proposed


solution.
ƒ Identify factors for optimization experiments.
ƒ Identify factor levels for optimizing experiments.
ƒ Design optimizing experiment to include randomization and
replication.
ƒ Perform experiments and collect data.
ƒ Analyze data with various tools including regression analysis.

© GE Capital, Inc., 2000

DMAIC GB B TX PG V 4.2.0
D M A I C

Improve–Project Learning Objectives 37

5 Develop a proposed solution.


ƒ Interpret the outputs of various tools to determine the optimum
solution.
ƒ Determine if the optimum solution will meet project goals.
ƒ Present the proposed solution to management.
ƒ Use statistical tolerancing to define the control mechanisms for
implementation.

© GE Capital, Inc., 2000

DMAIC GB B TX PG V 4.2.0
D M A I C

Improve–Project Learning Objectives 38

6 Establish operating tolerances.


ƒ Describe the concept of tolerances and describe an example of
this concept.
ƒ Describe the use of simulation and the use of Crystal Ball.
ƒ Describe and show an example of Crystal Ball.

7 Pilot the proposed solution.

© GE Capital, Inc., 2000

DMAIC GB B TX PG V 4.2.0
D M A I C

Improve–Test Learning Objectives 39

1 Describe and define the deliverables of Steps 7 and 8 in Six


Sigma.

2 Describe and define the deliverables of Step 9 in Six Sigma.

© GE Capital, Inc., 2000

DMAIC GB B TX PG V 4.2.0
D M A I C

Control–Project Learning Objectives 40

1 Develop/modify and implement Quality Plans.


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

© GE Capital, Inc., 2000

DMAIC GB B TX PG V 4.2.0
D M A I C

Control–Project Learning Objectives 41

2 Develop and implement risk management strategies.


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

© GE Capital, Inc., 2000

DMAIC GB B TX PG V 4.2.0
D M A I C

Control–Project Learning Objectives 42

3 Develop and implement mistake proofing strategies.


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

© GE Capital, Inc., 2000

DMAIC GB B TX PG V 4.2.0
D M A I C

Control–Project Learning Objectives 43

4 Select and apply the appropriate Control Chart.


ƒ Recall and explain SPC Concepts, including controlled
variation, uncontrolled variation, common causes, and special
causes.
ƒ Recognize the five main uses of control charts.
ƒ Distinguish between variable, attribute, and process focused
control charts.
ƒ Determine control limits.
ƒ Distinguish between control limits and specification limits.
ƒ Recognize the four states of a process.

© GE Capital, Inc., 2000

DMAIC GB B TX PG V 4.2.0
D M A I C

Control–Test Learning Objectives 44

1 Describe and define the deliverables of Steps 10 and 11 in Six Sigma.

2 Describe and define the deliverables of Step 12 in Six Sigma.

3 Develop and implement Variable Control Charts.


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

© GE Capital, Inc., 2000

DMAIC GB B TX PG V 4.2.0
D M A I C

Control–Test Learning Objectives 45

4 Develop and implement Attribute Control Charts.


ƒ Describe the purpose of Attribute Control Charts.
ƒ Define and relate the terms “a defect” and “a defective.”
ƒ Recognize types of Attribute Control Charts, including C-
Charts, U-Charts, P-Charts, NP Charts.
ƒ Select the appropriate Attribute Control Chart.
ƒ Use Minitab to generate each type of Attribute Control Chart.
ƒ Determine the appropriate Attribute Chart subgroup size.

© GE Capital, Inc., 2000

DMAIC GB B TX PG V 4.2.0
D M A I C

Glossary And Minitab Primer 46

1 Define the typical terms used in Six Sigma methodologies.

2 Describe the use of the Minitab software program.

© GE Capital, Inc., 2000

DMAIC GB B TX PG V 4.2.0
D M A I C

Introduction Exercise (60 Minutes) 1


Introductions
Desired Outcome: Introduction of Classmates and CD Rom Learnings

What How Who Timing

Team ƒ Go around your tables and introduce All 2 mins.


Preparation yourself to your table-team members
(name and business only, at this
time.)

Exercise ƒ Discuss the importance of the All 7 mins.


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

Close ƒ Choose a spokesperson to 1 min.


Exercise report out

Report ƒ When asked by the instructor, 1 min.


Out each team member will introduce
themselves:
– Name
– How long you’ve been
with GE
– Your Title and what you do
– Previous Quality
Experience
– What Phase your project
is in
ƒ 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

ƒ Share and get feedback on project deliverables for Define and


Measure phases
ƒ Review Define phase
ƒ Review Measure phase
ƒ Learn how to apply CTQ tools in the Measure phase of a
project
ƒ Learn how to use Minitab for statistical and graphical analysis
during a project
ƒ Learn knowledge and skills in Analyze phase and apply to
Capital Logistics case
ƒ Preview Improve phase
ƒ Learn how to apply Design of Experiment (DOE) in the Improve
phase of a project
ƒ Preview Control phase
ƒ Learn how to apply control charts in the Control phase of a
project

© GE Capital, Inc., 2000

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

Steps 7,8,9 Steps 4,5,6

The Phases of the 12-Step Process Phase 4 (Improvement). This phase is usually initiated by
selecting those product performance characteristics which
Phase 1 (Define). This phase defines the project. It must be improved to achieve the goal. Once this is done, the
identifies customer CTQ’s and ties them to business characteristics are diagnosed to reveal the major sources of
needs. Further, it defines a project charter and the variation. Next, the key process variables are identified by
business process bounded by the project. way of statistically designed experiments. For each process
variable which proves to be significant, performance
Phase 2 (Measure). This phase is concerned with specifications are established.
selecting one or more product characteristics; i.e.,
dependent variables, mapping the respective process, Phase 5 (Control). This phase is related to ensuring that the
making sure the measurement system is valid, making
new process conditions are documented and monitored via
the necessary measurements, and recording the results.
statistical process control methods. After a “settling in”
period, the process capability would be reassessed.
Phase 3 (Analyze). This phase entails estimating the Depending upon the outcomes of such a follow-up analysis,
short and long-term process capabilities and it may be necessary to revisit one or more of the preceding
benchmarking the key product performance metrics. phases.
Following this, a gap analysis is often undertaken to
identify the common factors of successful performance;
i.e., what factors explain best-in-class performance.
© 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
ƒ 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
ƒ Data Integrity the mean and – Risk
– MSA ƒ Capability variability analysis
– Brainstorm – ZBench ST & LT – Test – Control
potential X’s – Model plans
– Sampling plan – Estimate

ƒ Collect data

The Practical-To-Statistical-To-Practical
Transformation Process

© GE Capital, Inc., 2000

DMAIC GB C TX PG V 4.2.0
D M A I C

The 12-Step Process 5

Step Description Focus Tools Deliverables


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

© GE Capital, Inc., 2000

DMAIC GB C TX PG V 4.2.0
D M A I C

The “Basics” For Effective Meetings 6

ƒ DMAIC is a powerful methodology which will allow teams to


significantly improve the processes on which they are working.
However, in order for teams to efficiently use the tools and
techniques associated with DMAIC, they must be able to work
together effectively. Although not the focus of this class, we will
briefly focus on the minimum requirements for effective
meetings.
ƒ Further training in effective meeting skills is provided in
“Facilitating Teams through Change Projects” Training

© GE Capital, Inc., 2000

DMAIC GB C TX PG V 4.2.0
D M A I C

Meeting Skills 7

The “Basics” For Effective Meetings

Roles:
ƒ Leader or Facilitator
ƒ Timekeeper
ƒ Scribe
ƒ Note Taker

Tools:
ƒ Agenda
ƒ Desired Outcomes
ƒ Ground Rules
ƒ Decision-Making Process
ƒ Brainstorming

Define These Roles For A Meeting Or Any Team Rules Of Brainstorming:


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

The effectiveness of any meeting can be improved simply


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

Ground rules, or the “code of conduct,” further increase


the effectiveness of your meeting by specifying the
behavior that is expected from all participants. For
ongoing teams, such as DMAIC teams, ground rules
should be established and agreed upon by all team
members early in the project.
© GE Capital, Inc., 2000

DMAIC GB C TX PG V 4.2.0
D M A I C

Practice Meeting (5 minutes) 7

Desired Outcome: Name your team

What How Who Timing

Team ƒ Choose a facilitator, timekeeper All 30 secs.


Preparation and scribe

Determine ƒ Brainstorm a list of possible team Facilitator 4 mins.


Team Name names
ƒ Using N/3*, prioritize your
brainstormed list
ƒ Identify your top name

Close ƒ Choose a spokesperson to All 30 secs.


Exercise report out

* N/3 is a prioritization technique used to narrow a large list down to the top
priorities. The number of ideas (N) is divided by 3 (N/3) and team members get
that number of votes to identify their top choices. All votes should not be put on
one choice.

© GE Capital, Inc., 2000

DMAIC GB C TX PG V 4.2.0
D M A I C

Define Review Module Objectives 1


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

„ Review and get feedback on Greenbelt project


deliverables for Define phase

© GE Capital, Inc., 2000

DMAIC GB D TX PG V 4.2.0
D M A I C

2
First Things First (10 minutes)

Desired Outcome: Learning the components for successful change

What How Who Timing

Team ƒ Think of a change initiative that you Each 5 mins.


Preparation have experienced in the past in Team
which the change was unsuccessful. Member
Why did the change initiative fail?

Prepare List ƒ On a flip chart, list your group’s Facilitator 5 mins.


ideas for why change initiatives
fail. What are the common themes?
Be prepared to report findings.

© GE Capital, Inc., 2000

DMAIC GB D TX PG V 4.2.0
D M A I C

Change Acceleration Process (CAP) 3

Techn
ical St
rate g y

rat egy
iza tional St
an
al/Org
Cultur

Change Initiative
(Target)

CAP Complements Technical Strategy With Cultural


Tools To Achieve The Change Initiative

Change can be depicted via the equation:

Q x A = E

Q = Quality of solution to be implemented

A = Acceptance of solution to be implemented

E = Effectiveness of the implemented solution

It is as important to have a strategy for developing


acceptance as it is to have a plan for implementing the
solution.

A Technical Strategy and a Cultural/Organizational


Strategy are both necessary in order to effectively
achieve your change target. CAP gives us tools to
develop the Cultural/Organizational or Influence
Strategy.
© 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

Changing Systems And Structures

Leading Change Making Change Last


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

Creating A Shared Need Monitoring Progress


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

© GE Capital, Inc., 2000

DMAIC GB D TX PG V 4.2.0
D M A I C

Cap Tools And Six Sigma 5

Creating A Shaping A Mobilizing Making Change Monitoring


Shared Need Vision Commitment Last Progress

Define ƒ SIPOC ƒ In Frame / Out ƒ Threat vs.


ƒ Threat vs. Of Frame Opportunity Matrix
Opportunity Matrix ƒ GRPI
ƒ Project Scope ƒ ARMI
Contract

Measure

Analyze

Improve ƒ Making Change ƒ Making Change


Last Checklist Last Checklist
ƒ Control Pans ƒ Control Pans
ƒ Measures And ƒ Measures And
Rewards Rewards

Control

The above-mentioned tools are covered in the Tools


Section of the Self-Paced Workbook. For further training,
you can attend the week-long course sponsored by
Learning Services.

© GE Capital, Inc., 2000

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

What How Who Timing

Prepare a ƒ Identify at least one CAP Tool that Each As


plan for you have used in your project Student Homework
CAP Tools thus far
ƒ If you have not used any CAP
Tools, then identify a CAP Tool that
you could, in retrospect, have
used
ƒ 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

© GE Capital, Inc., 2000

DMAIC GB D TX PG V 4.2.0
D M A I C

Define Phase Flowchart 7

D M A I C

DEFINE
PHASE
OVERVIEW

Define A: Define B: Define C:


Identify Project Develop Team Define
CTQ’s Charter Process Map

© GE Capital, Inc., 2000

DMAIC GB D TX PG V 4.2.0
D M A I C

Define Phase Overview 8

What is the Define phase?


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

Why is the Define phase important?


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

Steps involved in the Define phase:


Define A: Identify Project CTQ’s
Define B: Develop Team Charter
Define C: Define Process Map

© GE Capital, Inc., 2000

DMAIC GB D TX PG V 4.2.0
D M A I C

Define A–Identify Project CTQ’s 9

What does it mean to Identify Project CTQ’s?


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

Why is it important to Identify Project CTQ’s?


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

What are the project tasks for completing Define A?


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

DEFINE STEP
OVERVIEW

Define A: Define B: Define C:


Identify Project Develop Team Define Process
CTQ’s Charter Map

A.1 Identify Customer


A.2 Collect Voice of the Customer data to Identify Project CTQ’s
A.3 Build a process/product drill-down tree to Identify Project CTQ’s

© GE Capital, Inc., 2000

DMAIC GB D TX PG V 4.2.0
D M A I C

10
Define B–Define Team Charter

What does it mean to Develop a Team Charter?


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

Why is it important to Develop a Team Charter?


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

What are the project tasks for completing Define B?


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

Define A: Define B: Define C:


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

© GE Capital, Inc., 2000

DMAIC GB D TX PG V 4.2.0
D M A I C

Define C–Define Process Map 11

What does it mean to Define a Process Map?


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

Why is it important to Define a Process Map?


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

What are the project tasks for completing Define C?


C.1 Develop high-level Process Map

DEFINE STEP
OVERVIEW

Define A: Define B: Define C:


Identify Project Develop Team Define Process
CTQ’s Charter Map

C.1 Develop high-level Process Map

© GE Capital, Inc., 2000

DMAIC GB D TX PG V 4.2.0
D M A I C

Elevator Speech–Homework Activity* 12

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

What How Who

Preparation ƒ Develop an elevator speech for your project. All

Elevator ƒ Each team member shares the elevator Each


Speech speech for their project. (See the description Team
of an elevator speech on the following page.) Member

Feedback ƒ Make a note of any feedback you receive All


from your class members. Did your elevator
speech provide a clear understanding of your
project?

Elevator speeches will be shared with the entire class throughout the 3-day workshop.

* For additional information on elevator speeches, see the Self-Paced Workbook,


pp 197-198

© GE Capital, Inc., 2000

DMAIC GB D TX PG V 4.2.0
D M A I C

Elevator Speech 13

ƒ Here is what our project is about (describe the problem or


issue)
ƒ Here is why it’s important to you (describe the benefit of doing
the project)
ƒ Here is what success will look like (describe the goal of the
project and where we currently are)

What is it? Why use it?

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

© GE Capital, Inc., 2000

DMAIC GB D TX PG V 4.2.0
D M A I C

DMAIC Checklist–Define A 14

A.1 Identify customer


ƒ Have you considered both internal and external customers?
ƒ Have you focused on the most important customer segment(s)?

A.2 Collect voice of the customer data to identify customer CTQ’s


ƒ Have you collected data to understand customer requirements?
ƒ Have you coordinated your data collection with others who are collecting similar
data, so as not to overwhelm the customer?
ƒ Have you validated the customer CTQ’s (Big Y) with the customer?
ƒ Have you verified a VOC Research method that you’ve selected that reflects the
true VOC?

A.3 Build a process/product drill-down tree to identify project CTQ’s


ƒ Is your project CTQ focused on what’s most important to the customer?
ƒ If you have multiple CTQ’s, have you used the appropriate tool to prioritize them?

© GE Capital, Inc., 2000

DMAIC GB D TX PG V 4.2.0
D M A I C

DMAIC Checklist–Define B 15

B.1 Define the business case


ƒ Is the business case compelling to the team?
ƒ Does this business case illustrate why this project needs to be done now ?
ƒ Have I considered intellectual infringements?

B.2 Develop problem statement


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

B.3 Develop goal statement


ƒ Is your goal statement SMART (Specific, Measurable, Attainable, Relevant, Time-
bound)?
ƒ Does your goal statement focus on what needs to be improved, and without
predetermining a solution?
ƒ How will you know the team is successful?

© GE Capital, Inc., 2000

DMAIC GB D TX PG V 4.2.0
D M A I C

DMAIC Checklist–Define B (continued) 16

B.4 Project scope


ƒ Does your team agree on the project scope?
ƒ Is your project scope within the team’s control?

B.5 Roles
ƒ Are all functions represented?
ƒ Does each team member fully represent their manager’s input on the project?

© GE Capital, Inc., 2000

DMAIC GB D TX PG V 4.2.0
D M A I C

DMAIC Checklist–Define C 17

C.1 Develop high-level process map


ƒ Does the COPIS scope (start and stop) match the scope definition in your team
charter?
ƒ Have you mapped the process from your customer's perspective?
ƒ Is this the “As-Is” process? Was the map validated with the people who actually
do the process

© GE Capital, Inc., 2000

DMAIC GB D TX PG V 4.2.0
D M A I C

Forms Of Intellectual Property 18

Patent
ƒ Best Protection against independent development by others
ƒ Gives right to exclude others from making, using selling, the invention of 20
years from filing
ƒ Can wait 2-3 years for patent to issue
ƒ Right granted by the government for new, useful, non-obvious inventions
Trade Secret
ƒ Rights exist only so long as actually kept secret
ƒ Protects against theft or misappropriation, not independent development
ƒ Does not require inventiveness in the patent sense
Copyright
ƒ Protects original works of authorship fixed in a tangible medium
ƒ Does not protect against independent development
Trademark
ƒ Defines sources of goods or services
ƒ Arbitrary, fanciful marks get stronger protection

© GE Capital, Inc., 2000

DMAIC GB D TX PG V 4.2.0
D M A I C

Intellectual Property Assessment in Chartering Phase 19

Throughout the project consider implications to Intellectual


Property Policies and Procedures through the following:
ƒ Commercial benefit for licensing–consider this in the
cost/benefit
ƒ Infringement Avoidance–GE has a policy covering avoiding
infringing on other’s IP-Assure no infringement
ƒ Identify and Capture IP–Throughout this project consider
whether IP is developed and take actions to capture and
protect

Note: When writing your Business Case be certain to consider and include IP
issues.

Consult the business’ IP Business Champion and IP


Designer with questions, and for guidance

© GE Capital, Inc., 2000

DMAIC GB D TX PG V 4.2.0
D M A I C

© GE Capital, Inc., 2000

DMAIC GB D TX PG V 4.2.0
D M A I C

Measure Module Objectives 1

ƒ Review steps in Measure phase


ƒ Start work on Green Belt project deliverables for Measure phase

© GE Capital, Inc., 2000

DMAIC GB E TX PG V 4.2.0
D M A I C

Measure Phase Flowchart 2

D M A I C

MEASURE
PHASE
OVERVIEW

Measure 3:
Measure 2: Establish Data
Measure 1:
Define Collection Plan,
Select CTQ
Performance Validate
Characteristic
Standards Measurement
System, & Collect
Data

© GE Capital, Inc., 2000

DMAIC GB E TX PG V 4.2.0
D M A I C

3
The12-Step Process

Step Description Focus Tools Deliverables


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

© GE Capital, Inc., 2000

DMAIC GB E TX PG V 4.2.0
D M A I C

Measure Phase Overview 4

What is the Measure phase?


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

Why is the Measure phase important?


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

Steps involved in the Measure phase:


Measure 1: Select CTQ Characteristic
Measure 2: Define Performance Standards
Measure 3: Establish Data Collection Plan, Validate Measurement
System & Collect Data

© GE Capital, Inc., 2000

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

High Level Need Project


Project YY Measure 1

(VOC) Operational
Operational QFD
QFD
C&E
Definition
Definition
C&E
C-O-P-I-S
C-O-P-I-S

Measure 1
QFD
QFD
CTQ
CTQ Project
Project YY
Measure
Process
ProcessMap
C&E
C&E
Map
Measure Pareto
Pareto

Measure 2
Specification
Specification VOC
Limits
Limits VOC
QFD
QFD

QFD Measure 2
Target
Target QFD
Pareto
Pareto
VOC
VOC

Measure 2
Defect
Defect FMEA
FMEA
C&E
C&E

##of
ofDefect
Defect Measure 2
FMEA
Opportunities
Opportunities FMEA
Process
ProcessMap
Map
Per
Per Unit
Unit

© GE Capital, Inc., 2000

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
ƒ 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
ƒ Data Integrity – Risk
– MSA ƒ Capability analysis
– Brainstorm – ZBench ST & LT – Control
potential X’s plans
– Sampling plan

ƒ Collect data

The Practical-To-Statistical-To-Practical
Transformation Process

© GE Capital, Inc., 2000

DMAIC GB E TX PG V 4.2.0
D M A I C

1
Measure 1–Select CTQ Characteristics

What does it mean to Select a CTQ Characteristic?


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

Why is it important to Select a CTQ Characteristic?


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

What are the project tasks for completing Measure 1?


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

MEASURE STEP
OVERVIEW

Measure 1: Measure 2: Measure 3:


Select CTQ Define Establish Data
Characteristic Performance Collection, Validate
Standards MSA

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

© GE Capital, Inc., 2000

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

High Level Need Project


Project YY Measure 1

(VOC) Operational
Operational QFD
QFD
C&E
Definition
Definition C&E
C-O-P-I-S
C-O-P-I-S

Measure 1
QFD
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

© GE Capital, Inc., 2000

DMAIC GB F TX PG V 4.2.0
D M A I C

1
CTQ Tools Module Objectives

ƒ Learn how to apply a variety of CTQ tools to help prioritize &


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

This group of CTQ tools can be used to do some


preliminary analysis of your process in order to identify
the sub-processes that contribute most to satisfying the
customer CTQ requirement. Most likely, you will use 1
or 2 of these tools to gather your Project Y (Measure,
Step 1 deliverable).

Note: FMEA’s and Pareto Charts can also be used in


other DMAIC phases and will be covered in detail in later
sections of the program.

© GE Capital, Inc., 2000

DMAIC GB G TX PG V 4.2.0
D M A I C

Process Mapping

© GE Capital, Inc., 2000

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

Applications „ Enables group to see the entire process as a team


„ Techniques for examining a process to determine „ Magnifies normally overlooked areas and displays
where and why major breakdowns occur their relevancy
„ Graphically displays steps, events, operations and
relationships of resources
„ Used to design an improved process
„ Benefits
„ Provides a structure for breaking down a complex
process
„ Uncovers problem spots
„ Determines what data to collect.Targets selected
improvements.

© GE Capital, Inc., 2000

DMAIC GB G TX PG V 4.2.0
D M A I C

4
Levels Of A Process

Acquire New
Core Business

Obtain Request to Conduct Prepare Initialize


Buy Service Underwriting Contract Customer Service

COPIS/SIPOC S Establish Prepare Negotiate Close


C
Terms Docs Contract Deal

Underwriters Customer/
Customer Service

Detailed
Subprocess
Map
Tasks Procedures

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

If the Project Y is a time measure, which of the blocks


consumes the largest portion of total time, or which one
has the most variation or delays?

© GE Capital, Inc., 2000

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.

Reconciling what the process map is, into what it really


is, is a second version of the process. These first two
versions of the process constitute what is referred to as
the “as is” process map. A thorough “as is” process map
is one of the short-term goals of good process mapping
and a deliverable for the Analyze phase.

© GE Capital, Inc., 2000

DMAIC GB G TX PG V 4.2.0
D M A I C

6
Outside-In Focus

ƒ Does your process overlap with the customer’s?


ƒ Should the customer’s process be mapped?
ƒ Should you partner with the customer to
assess the processes?

Continue to keep focused on the customer’s perspective.

© GE Capital, Inc., 2000

DMAIC GB G TX PG V 4.2.0
D M A I C

7
Subprocess Mapping Techniques

ƒ Process flowchart
ƒ Alternate path method
ƒ Deployment or cross-functional map/flowchart

There are a number of ways to draw a subprocess map.


Your team can choose the approach or approaches most
appropriate (or comfortable) for your use. Whatever type
of subprocess map is chosen, make sure that it accurately
reflects your process as it currently exists.

© GE Capital, Inc., 2000

DMAIC GB G TX PG V 4.2.0
D M A I C

8
Process Mapping Symbols

These symbols represent the standard symbols for


process mapping.

Most projects use the simplified symbol sets shown on


the following page.

© GE Capital, Inc., 2000

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

The business process map most people are familiar with


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

It is generally used when the process is fairly small and


simple, or when documenting the work done by a single
person or group.

© GE Capital, Inc., 2000

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

The “alternate path” process map method arose from


reengineering efforts, where the mapping of very large
processes made “decision diamonds” more of a hindrance
than a help.

In this technique, diverging or alternate “paths” are noted


by split arrows. Teams can then note relative percentages
of times/incidences the process follows each path.

Process mapping software tools also make it easier to add


“icons” showing the tools or methods employed at
different points in the process.

Showing the alternate path allows us to identify sources


of variation which maybe adding complexity to our
process.

© GE Capital, Inc., 2000

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

Contract Initiate Revise Revise File


Negotiate
Call Contract Contract Contract
Admin.

Review Review
Attorney
Contract Contract

Write File
Dealer Contract Contract

Who PROCESS FLOW

This approach to process mapping emphasizes who


performs which tasks and which steps are done
concurrently.

It provides a clear visual perspective of the hand-offs and


relationships between groups involved in the process.
(Note: Either of the mapping techniques described on the
previous pages can be placed on this type of map.)

Always include the process customer and key suppliers as


“bands” on this type of map. (Because customers are
visible at the top, some people refer to this as a “Service
Blueprint.”)

© GE Capital, Inc., 2000

DMAIC GB G TX PG V 4.2.0
D M A I C

12
Steps In Deployment Mapping

1. Identify the participants of the process


from start-to-top point. Place the
customer at the upper left corner.
2. Identify the “trigger” or initial step and
identify the “final” step
3. Identify who receives the output of the
initial step and what activity they
perform
4. Repeat by identifying who receives the
output from the second step and what
activity they perform
5. Continue identifying steps and align
vertically with participants

For Your Process/Subprocess: 3. Next, identify who receives the output of the second
step and what activity they perform. Write this activity
1. Identify the participants involved in the process, on a card and place it to the right of the second step on
write their names on cards, and pin or tape the cards the same row as the participant performing the activity.
in a vertical column on the wall with the customer at
4. Identify who receives the output of this activity and what
the top of the column.
activities that participant performs.

2. Identify the action that initiates the process, write it 5. Continue to create cards and position them in the same
on a card, and place it next to the appropriate manner until the process is complete (i.e., the customer
participant card. In many instances,the process is receives the final process output).
initiated by the customer but it is not always the case.
Also, identify the final step in the process and place in
the appropriate row at the end of the map. Here again,
the last step may be the customer’s action.

© GE Capital, Inc., 2000

DMAIC GB G TX PG V 4.2.0
D M A I C

13
Tips In Subprocess Mapping

ƒ Involve people who know (focus on) the “as is”


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

„ Respect the boundaries


„ Don’t start “problem-solving”
„ Validate and refine before analyzing

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

© GE Capital, Inc., 2000

DMAIC GB G TX PG V 4.2.0
D M A I C

14
Verifying CTQ Elements For Process Mapping

Using Process Mapping in the Measure phase allows us to:


ƒ Select an upstream step that is generating defects
ƒ Determine where to collect the data
ƒ Define the output measure (a good place to begin looking is a “No Exit” out
of Decision Diamonds)

© GE Capital, Inc., 2000

DMAIC GB G TX PG V 4.2.0
D M A I C

15
Subprocess Mapping–Breakout Activity (20 minutes)

Desired Outcome: Practice creating a subprocess map

What How Who Timing

Team ƒ Choose a facilitator, All 30 secs.


Preparation timekeeper, scribe and/or note
taker

Prepare ƒ Prepare the subprocess map for All 19 mins.


Subprocess your own project
Map
ƒ Brainstorm the challenges of
subprocess mapping on your
own projects

Close ƒ Choose a subprocess to report All 30 secs.


Exercise your results

© GE Capital, Inc., 2000

DMAIC GB G TX PG V 4.2.0
D M A I C

16
Cause & Effect Diagrams–CTQ Tool #2

A visual tool used by an improvement team to brainstorm and


logically organize possible causes for a specific problem or effect.

Machines Methods Materials

Potential High-Level Causes Problem


Statement

Measurement Mother Nature People

Procedure: 7. As a team, determine the three to five most


likely causes.
1. Draw a blank diagram on a flip chart.
8. Determine which likely causes you will need to
2. Define your problem statement. verify with data.

3. Label branches with categories appropriate to your Alternative procedure:


problem. The categories shown are often used, but any
categories can be used. „ Brainstorm causes independent of categories.

4. Brainstorm possible causes and attach them to „ Use the Affinity Diagram and brainstorm items into
appropriate branch. categories.

5. For each cause ask, “Why does this happen?”

6. Analyze results; do any causes repeat?

© GE Capital, Inc., 2000

DMAIC GB G TX PG V 4.2.0
D M A I C

17
Cause & Effect Diagrams–Example

Cause & Effect for Capital Logistics


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

Overnight Delivery Volvo Trucks

Unload Delivery Mack Trucks


Precipitation
Drop off & Hook up Freightliner Trucks

Environment Methods Machines

The C&E answers the question of “What am I going to


measure?” I can now focus on weight of freight or
distance or temperature, etc. (This is the output
characteristic.)

© GE Capital, Inc., 2000

DMAIC GB G TX PG V 4.2.0
D M A I C

18
FMEA–CTQ Tool #3

An FMEA is a structured approach to:


ƒ Identifying the ways in which a process can fail to meet critical customer
requirements
ƒ Estimating the risk of specific causes with regard to these failures
ƒ Evaluating the current control plan for preventing these failures from
occurring
ƒ Prioritizing the actions that should be taken to improve the process

Identify Ways The Product Or Process Can Fail, Then


Plan To Prevent Those Failures

Purpose & Benefits of FMEA Types of FMEA


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

FMEA’s will be covered in detail during the Improve


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

© GE Capital, Inc., 2000

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

FMEA Process Action Results

Occurrence
Occurrence
Responsibility

Detection

Detection
Severity

Severity
Potential Potential
Recommended And Target

RPN

RPN
Item/Process Potential Current Action
Effect(s) Of Cause(s) Of Completion
Step Failure Mode Controls Action Taken
Failure Failure Date
Enter
Amt Inaccurate Overbill 8 Wrong Ct 8 Rare 9 576 DMAIC Team E. Jones All DMAIC 8 4 2 64
Owed Wrong Price 5 “ 9 360 to Investigate 5/15/97 Tasks 8 3 2 48
Root Causes Complete
Underbill 5 Wrong Ct 8 Rare 9 360 of Count & 5 1 2 10
Wrong Price 5 “ 9 225 Price 5 1 3 15
Accessories
Missing No Payment 6 Sale Error 5 Reviewed 3 90 6 5 3 90
De Error 7 “ 3 126 6 7 3 126
De Error
Delay 3 Sale Error 5 Reviewed 3 45 3 5 3 45
De Error 7 “ 3 63 3 7 3 63
De Error
Delayed Late Bill 3 Sale “Too 7 Measured 4 84 3 7 4 84
Busy”
System Down 3 “ 4 36 3 3 4 36

No Bill 6 Sales Busy 7 Measured 4 168 6 7 4 168


System Down 3 “ 4 72 6 3 4 72

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

In this example,the “Wrong Ct” component of billing


accuracy becomes the highest priority defect type.

© GE Capital, Inc., 2000

DMAIC GB G TX PG V 4.2.0
D M A I C

20
Failure Mode And Effects Analysis

How it Works Function


Part/Process

Failure Mode

Effects Controls
Severity Detectability
(1-10) (1-10)
Causes
Occurrence
(1-10)
RPN
RPN
Risk
Risk Priority Number
Priority Number
RPN
RPN = S x O x D == 11 to
= S x O x D to 1000
1000

„ Review the product, service, or process „ Assign Severity, Occurrence, and Detection Factors
„ Determine failure modes „ Calculate RPN.Prioritize RPNs from high to low.
„ List one or more potential effects for each failure Identify the top issues, those with high RPN’s.
mode. Answer the question: “If the failure occurs, „ Determine preventive or remedial actions, especially
what are the consequences?” for high-priority issues.
„ Identify potential causes „ Recalculate RPN after actions have been implemented
„ List current controls

© GE Capital, Inc., 2000

DMAIC GB G TX PG V 4.2.0
D M A I C

21
FMEA: Calculating Risk Priority Number (RPN)

RPN = Severity x Occurrence x Detection


Severity Scale
Rating Criteria – A Failure Could:
Bad
10 Injure a customer or employee
9 Be illegal
8 Render product or service unfit for use
7 Cause extreme customer dissatisfaction
6 Result in partial malfunction
5 Cause a loss of performance which

4
is likely to result in a complaint
Cause minor performance loss
Occurrence Scale
3 Cause a minor nuisance, but be Rating Time Period Probability
overcome with no performance loss
10 More than once per day > 30%
Good
2 Be unnoticed and have only minor
9 Once every 3-4 days ≤ 30%
Detection Scale
effect on performance
1 Be unnoticed and not affect the 8 Once per week ≤ 5% Rating Definition
performance 7 Once per month ≤ 1% 10 Defect caused by failure is not detectable

6 Once every 3 months ≤ .03% 9 Occasional units are checked for defect
8 Units are systematically sampled and inspected
5 Once every 6 months ≤ 1 Per 10,000
7 All units are manually inspected
4 Once per year ≤ 6 Per 100,000
6 Units are manually inspected with mistake-proofing
3 Once every 1-3 years ≤ 6 Per Million modifications
2 Once every 3-6 years ≤ 3 Per 10 Million 5 Process is monitored (SPC) and manually inspected
1 Once every 6-100 years ≤ 2 Per Billion 4 SPC is used with an immediate reaction to out-of-
control conditions
3 SPC as above with 100% inspection surrounding out -
of-control conditions
2 All units are automatically inspected
1 Defect is obvious and can be kept from affecting the
customer

Total RPN is determined by multiplying all of the


individual scores.

Note: Occurrence, detection and severity are independent


estimates of risk.

© GE Capital, Inc., 2000

DMAIC GB G TX PG V 4.2.0
D M A I C

22
Using A FMEA In The Measure Phase

ƒ Helps to prioritize Project Y’s


ƒ Helps to identify where and how a process may fail to meet a
CTQ (defects)

In this process, you may uncover an issue where you


need to contain a problem (e.g., regulatory, non
compliance, legal). If this occurs, it raises the priority.

© GE Capital, Inc., 2000

DMAIC GB G TX PG V 4.2.0
D M A I C

23
Pareto Charts–CTQ Tool #4

Is there a defect that occurs frequently?

Segment data to look for a Frequency


significant factor that
influences the process

C A E D B
Category of Defect

The Pareto chart is a bar chart with the bars


(segmentation levels) arranged in descending order. It is
an essential tool to help prioritize improvement targets by
identifying the 20% of the problems that cause 80% of
the poor performance (Pareto principle).

© GE Capital, Inc., 2000

DMAIC GB G TX PG V 4.2.0
D M A I C

24
Pareto Charts (continued)

Example April 1 – June 30


200 100
Number Of Units Investigated: 5,000
180 *f = frequency Cumulative
90
Summation Line
160 (Cum Sum line)
80
f* of D+B+F

Cumulative Percentage
140 70
f* of D+B

120 60
Frequency

f* of D
100 50
LEGEND
80 40
A: Illegible
B: Bank Info Incomplete
60 30 C: Missing Signature
D: Personal Information
Incomplete
40 20 E: Employment History
Incomplete
20 10 F: Address Incomplete

0
Defect Type D B F A C E Other* Total
# Defects 105 40 20 10 5 3 25 208
% Defects 50.5 19.2 9.6 4.8 2.4 1.5 12.0 100
Cum % 50.5 69.7 79,3 84.1 86.5 88.0 100.0
Approximately 80% of Defects from Defects D + B + F

A fully documented Pareto chart will include the


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

In this example, I would likely focus on the ‘D” segment,


and this would become my project Y.

You should also create a Pareto showing $’s. This may


cause your focus to change.

* Other- Try to keep this category as small as possible.


Find out what is in “other”.

© GE Capital, Inc., 2000

DMAIC GB G TX PG V 4.2.0
D M A I C

25
Using Pareto Charts In The Measure Phase

ƒ Quantitative tool use to determine segmented areas of focus


ƒ Graphically displays most frequent occurrence of outcomes
(little y’s)
ƒ Helps you get from Big Y to little y

© GE Capital, Inc., 2000

DMAIC GB G TX PG V 4.2.0
D M A I C

26
QFD–CTQ Tool #5

Definition of QFD (Quality Function Deployment)

Structured Methodology to Identify


and Translate Customer Needs
and Wants Into Measurable
Features and Characteristics of a Product or Service

© GE Capital, Inc., 2000

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

House (HOW’s) Variables


Functional

Quality
(WHAT’s)

of

Characteristics
#1 (HOW’s)
Quality

(WHAT’s)
#2 Part House

Processes
of

(WHAT’s)
House
Quality

GE
of
Y #3
Quality
Y
#4
Key Functional
Requirements
Key Part
Characteristics X
Key
Manufacturing Key
Processes Process
Variables

Product application of QFD links functional requirements


to key process variables.

© GE Capital, Inc., 2000

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

Services application of QFD links project deliverables to


key process tasks.

© GE Capital, Inc., 2000

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

House Of Quality Example–Lunch Selection 30

Relationship O = HOLD CONSTANT


High Positive
Positive
Lunch QFD High Negative
Negative

Number of measured ingredients


Percent carbohydrate provided
Percent nutrition provided
Product Requirement

Number of dishes used

Cost of ingredients
Customer Needs Weight of portion

PB&J Sandwich
Time to prepare

Instant Soup
Importance

Fast Food
Direction for improvement
Fills us up 5 H M M 2 3 4
Is nutritious 4 M H L L 3 3 2
Tastes good 3 H L 2 3 4
Is easy to make 4 H H 2 1 4
Is easy to clean up 2 M H 2 3 5
Sticks with us 4 H M 2 2 4
Is inexpensive 1 L L L M 4 3 1
Is clean 2 M M H 2 3 4

Importance Rating 94* 52 59 48 60 10 18


Target (Obtained from VOC) 16 33 25 3 1 0.50 0
PB&J 2 33 25 3 1 0.25 0 Key
Instant Soup 8 15 10 5 3 0.50 1 H=9
Fast Food 16 40 73 10 0 3.00 0 M=3
Units Of Measure (Obtained from VOC) oz % % min $ L=1
*Importance Rating For “Weight Of Portion” = 5(9) + 4(3) + 4(9) + 1(1) = 94
Roof Relationship Symbols indicate if the quality goals of the CTQ’s are conflicting. For example, if I chose to focus on
“Weight of Portion” I need to be concerned about the negative impact on the “Number of Measured Ingredients”

© GE Capital, Inc., 2000

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

© GE Capital, Inc., 2000

DMAIC GB G TX PG V 4.2.0
D M A I C
Measure 1–Select CTQ Characteristics

CTQ And Performance Standard Worksheet 32

Define A
Here’s our next step: Output
Output Unit
Unit AA
Lunch
LunchPortion
Portion

Output
Output Define A

Characteristic
Characteristic Healthy
HealthyLunch
Lunch

Project
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.

You may not know Spec, Target Measure 2


or Defect at this time (in Step 1). Defect
Defect**
If you know it now, write it down. <<14
14oz.
oz.Or
Or>>18
18oz.
oz.
This information will come from
the VOC.
##of
ofDefect
Defect Measure 2
Opportunities
Opportunities
Per
Per Unit
Unit 11

© GE Capital, Inc., 2000

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

© GE Capital, Inc., 2000

DMAIC GB G TX PG V 4.2.0
D M A I C

34
Another Example

What are your needs for this class?

# Tools Used

# Completed
Assessment
Customer

Exam Pass
Importance

Knowledge

Project Per
In Project
Needs

Quarter
Score

Rate
Learn The
Material
5 9 3 3 1

Pass The
Exam
4 3 9 1 1

Apply This
Material To Project
2 1 1 9 3

Importance
Rating 59 53 37 15

Key
High = 9
Medium = 3
Low = 1

© GE Capital, Inc., 2000

DMAIC GB G TX PG V 4.2.0
D M A I C

35
Points To Remember

ƒ The process may look simple, but requires effort


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

QFD is a valuable decision support tool (assigning


numbers to H-M-L does not make the numbers and cut-
off’s absolute decision points).

© GE Capital, Inc., 2000

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

ƒ QFD on everything
– Set the “right” granularity
– Don’t apply to every last project
ƒ Lack of teamwork
– Wrong participants
– Lack of team skills
– Lack of support or commitment
ƒ Too much “chart focus”
ƒ “Hurry up and get done”
ƒ Failure to integrate and implement QFD

© GE Capital, Inc., 2000

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.

A team consisting of Capital Logistics customer service


representatives and Capital Logistics customers has been
assembled. The team’s objective is to use QFD to identify the sub-
processes that will be the focus of DMAIC improvement projects.

© GE Capital, Inc., 2000

DMAIC GB G TX PG V 4.2.0
D M A I C

QFD–Breakout Activity (20 minutes) 39

Desired Outcome: Practice constructing a House Of Quality

What How Who Timing

Preparation ƒ Choose a facilitator, scribe All 5 mins.


and timekeeper

Complete ƒ Review the partially completed All 10 mins.


Relationships House Of Quality on the next
page
ƒ Complete the relationship room
of the house

Calculate ƒ Calculate the “How” importance All 5 mins.


“How” ratings
Importance
ƒ Which “How” is most important
Ratings
to satisfy the customer?

© GE Capital, Inc., 2000

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

the day helpdesk is

% customers rating
Number of hours in

Customer Service
resolve problems

survey of (CSR).
courteousness
Cycle time to
% problems
Importance

available
Quick problem resolution 5
WHATS
Customer Wants

Available when needed 4

Courteous Service 3

Importance rating

Units % Hrs.
Hrs. %
days

Target 100 <3 24x7 100

ƒ To what extent will the “How’s” impact the “What”?

Use L=1
M=3
H=9

ƒ Add one more service characteristic (CTQ) and place


rating on the CTQ.

© GE Capital, Inc., 2000

DMAIC GB G TX PG V 4.2.0
D M A I C

41
CTQ Tools: Summary

ƒ There are a variety of tools available to identify the CTQ


characteristic for your project (Project Y)
ƒ Select the tool that is most appropriate for your project

© GE Capital, Inc., 2000

DMAIC GB G TX PG V 4.2.0
D M A I C

© GE Capital, Inc., 2000

DMAIC GB G TX PG V 4.2.0
D M A I C

Measure 2–Define Performance Standards 1

What does it mean to Define Performance Standards?


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

Why is it important to Define Performance Standards?


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

What are the project tasks for completing Measure 2?


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

Measure 1: Measure 2: Measure 3:


Select CTQ Define Establish Data
Characteristic Performance Collection, Validate
Standards MSA

2.1 Develop operational definition for process to be measured


2.2 Identify target performance
2.3 Set specification limit(s)
2.4 Define unit, defect and defect opportunity

© GE Capital, Inc., 2000

DMAIC GB H TX PG V 4.2.0
D M A I C

Operational Definitions 2

Defining The Measure


Definition An operational definition is a clear, concise
description of a measurement
and the process by which it is to
be collected

Purpose
To remove ambiguity
ƒ Everyone has a consistent understanding

To provide a clear way to measure the


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

Operational definitions guide what properties will be


measured and how they will be measured.

There is no single right way to write an operational


definition. There is only what people agree to for a
specific purpose. The critical factor is that any two
people using the operational definition will be measuring
the same thing.

© GE Capital, Inc., 2000

DMAIC GB H TX PG V 4.2.0
D M A I C

Operational Definitions For Output Measures 3

Features
ƒ What: Must have specific and concrete criteria
ƒ How: Must have a method to measure criteria
ƒ Must be useful to both you and the customer
(the “wing-to-wing” concept)

Example: Loan Application Cycle Time


What: Loan application cycle time is the number of hours from
receipt of a loan application, to successful notification of
decision for the loan application
How: The clock starts when the computer attaches the time of
application receipt at data entry

The clock stops when the phone caller notes time of completed
application decision notification in desk log

Must Have Specific And Concrete Criteria Example:


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

Project Y data is the first key to understanding process


performance and the starting point for understanding how
to control process output.

© GE Capital, Inc., 2000

DMAIC GB H TX PG V 4.2.0
D M A I C

Define Performance Standards 5

CTQ Elements (Performance Standards)–5 Major Elements ONLY


Timely Call Pick-up
(Output characteristic) Established in Define, Step A

Hold Time
(Project Y Measure) Established in Measure, Step 1
Customer Need

Customer waits 10 seconds or less


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

Not greater than 20 seconds


(specification/ tolerance limit) Established in Measure, Step 2
A CTQ applies
To each individual
output unit Any response taking More
than 20 seconds (Defect) Established in Measure, Step 2

CTQ’s Are The Bridge Between Our Process Output


And Customer Satisfaction

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

Project Y: a definition of how the product/service’s


characteristic is to be quantified. The key metric for the
project.

Target: where we “aim” our product/service. If there


were no variation in the product/service, this is the value
we would always achieve.

© GE Capital, Inc., 2000

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

High Level Need Project


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

© GE Capital, Inc., 2000

DMAIC GB H TX PG V 4.2.0
D M A I C

Establishing A Performance Standard 7

A performance standard
translates customer needs
into quantified requirements
for our product or process Output
Output Timely Resolution
Characteristic
Characteristic of a Call

Specific Needs Total time to resolve


Statement Project
Project YY inquiry minus initiation
time of inquiry

Quick
Quick
Answer CTQ
CTQ Target
Target
5 Minutes or
Answer less

Specification/
Specification/ Not Greater
tolerance
tolerance Than 60
limit(s)
limit(s) (USL/USL)
(USL/USL) Minutes (USL)

Any Response
Defect
Defect Taking More Than
60 Minutes

Output Characteristic: A word or phrase that describes


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

© GE Capital, Inc., 2000

DMAIC GB H TX PG V 4.2.0
D M A I C

Identify Project CTQ’s 8

Translating Needs into CTQ’s: Example 1–Pizza


ƒ Characteristic: temperature of pizza
ƒ Project Y metric: temperature (degrees F) of Pizza at customer door step
Need
Need == Hot
Hot
Pizza ƒ Target: 160 degrees
Pizza
ƒ Specification: LSL=140 degrees and USL=180 degrees
ƒ Defect: any pizza temperature falling outside of the specifications

Translating Needs into CTQ’s: Example 2–Customer Service


ƒ Characteristic: CSA Knowledge level
Need
Need == ƒ Project Y: training test score
Problem
Problem ƒ Target: 100 points
Solved on First
Solved on First ƒ Specification: LSL= 90 points
Call
Call ƒ Defect: any score less than 90 points

Translating Needs into CTQ’s: Example 3–Information


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

LSL=Lower Specification Limit


USL= Upper Specification Limit

© GE Capital, Inc., 2000

DMAIC GB H TX PG V 4.2.0
D M A I C

Specify Target And Specification Limits For Project Y 9

Once the Project Y metric has been established, the target and
specification limits can be established.

Output
Output
Characteristic Timely response
Characteristic

Specific Needs Variance between


Statement Project
Project YY customer request and
actual delivery
Customer
Customer
receives
receives approval
approval
on CTQ
CTQ Target
Customer
on customer
customer Target request date
request
request date
date

Specification/
Specification/ Variance not
tolerance
tolerance greater than
limit(s)
limit(s) one day

Any response
Defect
Defect greater than
1 day

Target/Nominal Value
„ Optimal value for Y
„ Specification/Tolerance Limit(s)
„ Window or range of acceptable output values for Y
„ Also called “performance standards”
„ Information about where to set target and specification
limits comes from the customer. This information can
be collected not only from VOC but also from:
– Contractual agreements
– Industry standards
– Customer observations
– Process data

© GE Capital, Inc., 2000

DMAIC GB H TX PG V 4.2.0
D M A I C

Project Y 10

Six Sigma Performance Definitions


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

„ The definitions listed above are based on the Project Y


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

© GE Capital, Inc., 2000

DMAIC GB H TX PG V 4.2.0
D M A I C

Project Y Alignment 11

Key output metrics that are


Business Big Y’s aligned with strategic
goals/objectives of the
business. Big Y’s provide a
direct measure of business
performance.
Process Y’s
Y
PROCESS
PROCESS Y
Key output metrics that
Y summarize process
Management
Management Y performance

Project Y
Key project metric defined
from the customer’s
perspective
X1 X2 X3

Any parameters that


influence the Y

Does Your Project Y Pass The Alignment Test?


„ All quality projects should be aligned to a Big Y and
CTQ for the business. This ensures that project activity
is organized and targeted to impact specific CTQ’s that
the customer will feel.
„ Not all Y’s are useful measures of business
performance or customer impact. Your Project Y
should be correlated to a higher level Business Big Y
or CTQ.
„ You should be able to describe the link between your
Project Y and the related Business Big Y or CTQ in
specific terms.

© GE Capital, Inc., 2000

DMAIC GB H TX PG V 4.2.0
D M A I C

6σ Performance Standards Definitions–Breakout Activity (5 minutes) 12

Desired Outcome
„ Practice defining units and defect for your CTQ

What How Who Timing


Team ƒ Choose a facilitator All
Preparation

Define units ƒ For your process output, agree on the Facilitator 5 mins.
and defect unit and defect for your Project Y. Use
the worksheet on page 17 of your GB
Workbook.

Close ƒ Choose a spokesperson to report your All


exercise results

© GE Capital, Inc., 2000

DMAIC GB H TX PG V 4.2.0
D M A I C

1
Minitab and Graphical Analysis Module Objectives

ƒ Understand the structure of Minitab*


ƒ Understand data entry and correct data structure for analysis in
Minitab
ƒ Review variation
ƒ Be able to create and interpret basic graphs in Minitab

In this module, you will focus on learning how to use


Minitab and how to perform basic graphical analysis of
data. We are stepping away from the DMAIC Flow for a
moment.

If you’ve been given some historical data, you can start to


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

In order to be fully effective in the Analyze phase, we


need to expose you to Minitab’s capabilities now. This
will help you think about the best way to collect data in
Measure
Step 3.

*Note: For another tutorial, the student can go to:


www.minitab.com

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

2
Minitab Windows

Menu
Menu Bar
Bar

Session
Session Window:
Window:
•• Analytical
Analytical Output
Output

Data
Data Window:
Window:
•• A
A Worksheet,
Worksheet, not
not aa
Spreadsheet
Spreadsheet

Graph
Graph Window
Window

Data Window–Stores data to be used for analysis. You can move through the windows using Ctrl-Tab, the
Discussed in more detail on next page. windows menu, or the following shortcuts.

Data Window Crtl+A


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

Info Window–Stores synopsis of your data


(not shown) window.

History Window–Stores all commands initiated (not


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

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

3
Data Window

Minimizes the Window Closes the Window

Data Entry Arrow


Maximizes
the Window

Column Names
Are Entered Here
Data is Entered
Here

Scroll Bars

ƒ Data window is a worksheet, not a spreadsheet. ƒ If the value you enter into the first row is something
ƒ 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).
ƒ Each column has a title area which is 31-characters or
less and this title must be unique.
ƒ Everything in a column is considered to be the same
variable.
ƒ Column names are above the column, not in the first
row.
ƒ Minitab cells contain values only, not formulas, like
Excel.
ƒ To perform mathematical functions, use
Calc > Calculator.
ƒ The orientation of the data arrow determines whether
the enter key will take you to the next row or the next
column.

© GE Capital, Inc., 2000

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

Calc Menu „ Calculate or generate data

Stat Menu „ Basic statistics and quality tools


„ Most often used by Green Belts
Graph Menu „ Contains the commands that you will use to do graphical analysis during
your project
Help Menu „ Minitab has a comprehensive Help system with detailed documentation of
all features, complete with examples of how all the menu commands are
used, and how to interpret graphical and statistical output which result from
the use of the commands
Window Menu „ Allows you to manage multiple graphs on the screen

Six Sigma* „ Product & process report can be used for determining performance levels
Menu (Calculating Sigma)

Notes on “Saving”: Notes on “Opening”:

Save Project As–When you save a project, you save all If you use the icon, only projects will be shown. You
the worksheet, session, info, history and graph windows must use the file > open worksheet to view all Minitab
that are currently open. Minitab will use the file extension worksheets.
of .mpj.
*Six Sigma add-on in Minitab can be obtained via a link
Save Current Worksheet As–When you save as a in support central.
worksheet, you save only the worksheet portion (you will
lose session window and graphs). Minitab will use the
file extension of .mtw.

© GE Capital, Inc., 2000

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

Command Parameters–Minitab sometimes calls these


parameters:

X Y
Factor Response
Independent Dependent
Input variable

By variable variables
graph variables

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

6
Using Minitab: A Typical Session

Step 1: Minitab allows you to enter data in four different ways:

1. Open an existing Minitab worksheet


2. Type data into the worksheet
3. Import data files from other compatible software packages (such
as Excel)
4. Paste data from other applications

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

7
Using Minitab: A Typical Session

Enter Data Manually

Only one row for


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

Type data
directly into
worksheet

One way to enter data in Minitab is to type the data


directly into the worksheet. For your Green Belt project,
you will probably import or copy the data from Excel.

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

8
Using Minitab: A Typical Session

Importing Data From Excel

1.3 How to import an Excel data file:


1. File > Open Worksheet
2. Select Files of Type: Excel
3. Highlight the file to be imported
4. Double-click or click Open

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

When you choose a file type from the drop-down list,


only those files containing the chosen file extension are
displayed in the file list. When you are importing files, it
is important to choose the specific file type you are
importing.

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

9
Using Minitab: A Typical Session

Paste data from Excel


1.In Excel:
Highlight Data (and Column Names) to be copied
Using Your Mouse

2.Copy the Data to the Windows Clipboard


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

3.Go to Minitab:
ALT > Tab

4. Position the Cursor where you want the data to fill


See example below.

5. Go to the Edit Menu:


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

Insertion point

Note: In order for this to work, you


must make all data non-format specific
in Excel before you cut-paste.

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

10
Using Minitab: A Typical Session

Tips for moving data back and forth:

ƒ Structure the data so that each variable is in a single column


ƒ Each column must have a title
ƒ The column title must have fewer than 31 characters and be
on a single line
ƒ All data must immediately follow the column names
ƒ Do not put empty rows between rows of data
ƒ 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.
ƒ After movement into Minitab, check column heading type
(D vs. T.)

If you want to use a file in both Minitab and Excel, keep


these points in mind.

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

11
Using Minitab: A Typical Session

Tables vs. Variable Columns


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

The format of data typically used in Excel is table format.


The best format for analysis of data in Minitab is variable
columns.

There are multiple ways to manipulate data. The major


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

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

12
Using Minitab: A Typical Session

Tips For Importing Data Into Minitab–Let’s practice


File name: sales_data.xls
Sales 1999
Sales Office January February March April

Central $387,980.00 $45,700.00 $456,789.00 $349,050.00

Southwest $578,990.00 $600,987.00 $456,789.00 $456,798.00

Northwest $435,800.00 $542,700.00 $345,988.00 $564,050.00

Southeast $497,050.00 $827,900.00 $456,789.00 $687,050.00

Northwest $613,242.00 $61,689.00 $456,789.00 $434,567.00

Total $2,513,062.00 $2,078,976.00 $2,173,144.00 $2,318,658.00


What is wrong with the format of this data in terms of its suitability for use in
Minitab?

Above is a set of data that has been captured and entered Note: For additional practice, see the Green Belt Training
into an Excel spreadsheet. Support Central Site.
To make the data suitable for use in Minitab:

1. Reformat the data in Excel (hide columns, change


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

Minitab also has several menu options that allow you to


reorganize and restructure the data once you have
imported it into Minitab. These options are available in the
Manip and Calc menus. You may want to first manipulate
in Excel prior to exporting the data!

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

13
Using Minitab: A Typical Session

Step 2: Select menu command

Most of the commands you use in your Green Belt


project are in the Stat and Graph menus.

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

14
Using Minitab: A Typical Session

Step 3: Enter parameters in dialog box

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

15
Using Minitab: A Typical Session

Step 4: View results in session or graph window

Note: Sometimes there is only a session window.


Therefore, check both the session window & the graph (if
one exists).

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

16
Using Minitab: A Typical Session

Step 5: Copy output to another application

Copying graphs:
1. Make sure the graph window is active in Minitab
2. Click right mouse button and select “Copy Graph”
3. Open the application into which you want to copy the graph or table,
e.g., Microsoft Word
4. Paste graph using Paste from the toolbar or Ctrl-V

Copying session window output:


1. Highlight the text lines you want to copy
2. Use the right mouse button to copy the text
3. Open the application into which you want to copy the text
4. Paste the text using Paste from the toolbar or Ctrl-V

Note: You can also copy graphs from the window >
manage graphs.

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

17
Using Minitab: A Typical Session

Steps 6 and 7: Print and Save

When you save a project, you can save the worksheet or


session window individually, or save an entire project.

© GE Capital, Inc., 2000

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

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

19
Graphical Analysis Of Data

Key Questions:
ƒ How is my data distributed (variation)?
ƒ What relationships exist between the
Y variable and X variables?

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

20
Review–Variation

ƒ All repetitive activities have variation (fluctuation)


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

A fundamental principle of process improvement is the


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

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

21
Using Data To Understand Variation

Plot The Data Using Variation Tools


Study Variation For A Period Of Time Study Variation Over Time
Histogram Run Chart

Measurement
Frequency

Measurement Time

For Continuous Data For Discrete Data For Continuous Data For Discrete Data

– Histogram – Bar Chart – Run Chart – Control Charts


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

Variation is a fact of life. There will always be


fluctuations in our processes.

Graphically displaying measures provides a basic


description of the variation and its sources.

Following any data collection effort, the first step toward


understanding variation is to plot the data.

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

22
Review–Continuous vs. Discrete Data

Reminder: Data Type Is Critical!


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

Application Cycle Time


Upper Specification Limit = 30 Days
Continuous
Y = days to process
Discrete
Y = late/on-time USL
Actual Times

No. Rec’d No. Late


28 23 13 34 24 29
21 16 24 11 49 21
30 2
21 25 26 27 27 29
30 29 30 20 10 30
12 11 27 23 24 28
17 9 30 29 29 28

Less variation information The most variation information

„ Continuous data holds several advantages over discrete


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

„ If you were the customer:


– Which type of data would you want?
– Which type of data shows what you see?

„ Consider this: If you were the process owner which


would you prefer?
– Simply how many applications were processed
within the customer specification?

OR
–The exact time it took to process each application?
© 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.

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

24
Two Types Of Variation

Common Versus Special Causes

Type of Variation Characteristics


Characteristics

Always Present
Common Cause Expected
Predictable
Normal

Not Always Present


Special Cause Unexpected
Unpredictable
Not Normal

To distinguish between common and special causes variation, use display


tools that study variation over time such as Run Charts and Control Charts.

There are two types of variation: The distinction between common and special causes is
important to determine the basic strategy for process
„ Common causes
improvement and control.
„ Special causes

Common causes are characteristics of the process and


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

Special causes are those causes that occur due to


occasional extraordinary circumstances. Special causes
are not always present, do not affect everyone working in
the process, do not affect all of the outcomes, and are not
predictable.

© GE Capital, Inc., 2000

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:
ƒ What is the shape of the distribution–symmetrical, skewed,
twin peaks, flat?
ƒ What is the central tendency (“center”)
of the distribution?
ƒ What is the variation (“spread”) of the distribution–wide or
narrow?

When describing variation, we usually focus on two


things:
– Center
– Spread

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

26
Statistics

ƒ Statistics is concerned with making inferences about general


populations and about characteristics of general populations
ƒ We study outcomes of random experiments
ƒ 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.
ƒ We call such a value a random variable

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

27
Some Distributions

ƒ A random variable can be expressed in terms of a distribution

Uniform Distribution
P(X) Single roll of a die

X All permissible values P(X)


are equally likely

Triangular Distribution
Sums of a pair of dice
P(X)

Rapidly descending
P(X), no tails
X

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

28
Distributions

Normal Distribution
P(X)
Process/repair times

Error fluctuations about an


operating point
X

Exponential Distribution
Time between arrivals
P(X)
Time between random
(unrelated) failures

X Events with no memory from one


to the next

Processes may have many differently shaped population


distributions. The shapes may be uniform, symmetric,
skewed, “ramped,” “camel-backed,” exponential, normal,
and non-normal. The shapes may be mixtures of normal
and non-normal distributions–or unmixed. Stable
distributions can have many different shapes.

Why does the process, or product, result in a particular


shape (distribution) and not some other shape? The
answers to these questions may provide a better
understanding of the process and how to improve it. The
particular shape is not as important as why this shape and
why not some other shape? Why this distribution, why
not some other distribution?

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

29
Shape

Shape is the distribution pattern exhibited by the data


Assess shape using a histogram, or more precisely with a Normal
Probability Plot
Roughly Normal Distribution Center Skewed Distribution

7 Center 20
6

Frequency
5
Frequency

4
10
3
2
1
0 0
12 13 14 15 16 17 18 19 0 10 20 30 40 50 60 70 80 90
Bimodal Distribution
Spread Spread
6 Center Center
5
Frequency

4
3
2
1
0
7 9 11 13 15 17 19 21 23

The shape of a data set can be determined by examining a


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

Because of its predictive qualities, the normal distribution


is of special interest.

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

30
The Normal Curve

Is The Data Distribution Normal?

Definition:
A probability distribution is where the most frequently occurring value is in the
middle and other probabilities tail off symmetrically in both directions.

Characteristics:
ƒ The curve does not reach zero
ƒ The curve can be divided in half with equal pieces falling either side of the most
frequently occurring value
ƒ The peak of the curve represents the center of the process
ƒ The area under the curve represents 100% of the product the process is capable of
producing

The normal curve is a probability distribution that forms


the basis for many decisions we will make about our
processes. The curve is noted by its “bell-shaped” nature,
where most of the values fall in the middle and fewer
values fall in either direction. The curve has several
important characteristics:
„ Infinity–The curve theoretically does not reach zero. It
goes out to infinity in either direction.
„ Symmetry–Roughly half of the values will be above
the average, and half below.
„ Centering–The peak of the curve tells us where the
process is centered.
„ Process Totality–The area under the curve represents
virtually 100% of whatever we are measuring.

© GE Capital, Inc., 2000

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%

-3s -2s -1s X +1s +2s +3s


68.26%

95.46%

99.73%

68.26% Fall Within +\- 1 Standard Deviation


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.

As you can see, the curve is first segmented into one


standard deviation which represents approximately 34%
of whatever you are measuring.

Because the curve is symmetrical, going one standard


deviation in the other direction represents approximately
68% of whatever it is you are measuring. Going out +/– 2
standard deviations is equal to approximately 95% of
whatever you are measuring and +/– 3 standard
deviations is equal to 99.73% of whatever you are
measuring.

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

32
Normal Probability Plot

We use a normal Probability Plot to determine if our data is


normal Normal Probability Plot

.999
.99
.95
Probability

.80
.50
.20
.05
.01
.001

2 12 22 32
Cycle Time
Average: 16.3921 Anderson-Darling Normality Test
StDev: 5.61675 A-Squared: 0.208
N: 240 P-Value: 0.864

Use this method to determine if If p>0.05, then data is normal


distribution is normal

What Is A Normality Test? When Should I Use A Normality Test?

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

Roughly Normal Distribution Bimodal Distribution Skewed Distribution Long-Tailed Distribution


7 6 20 6

6 5 5
Frequency

5
Frequency

Frequency
4

Frequency
4
3 10
3 3

2 2
2
1 1
1
0
0
12 13 14 15 16 17 18 19 0 0
7 9 11 13 15 17 19 21 23
0 10 20 30 40 50 60 70 80 90 7 9 11 13 15 17 19 21 23

Normal Probability Plot for a Normal Probability Plot for a Normal Probability Plot for a Normal Probability Plot for
Normal Distribution Bimodal Distribution Skewed Distribution Long-Tailed Distribution
99 99
ML Estimates ML Estimates
ML Estimates
Mean: 14.6382
Mean: Mean: 15.0790
95 95
StDev: 5.47084
StDev: 90 StDev: 12.6232 90

80 80
70
Percent

Percent
70
Percent

60 60
Percent

50 50
40 40
30 30
20 20

10 10

5 5

1 1

0 10 20 30 0 10 20 30

How Distribution Looks On Normality Curve:

Straight line Zig-zag Two lines “S” curve


(Stable Operations)

The Normal Probability Plot is another way besides the


histogram to plot data and look for normality. Normal
data, when plotted with the data value on the X-axis and
specially spaced percentiles of the normal distribution on
the Y-axis, will fall on a straight line. 95% confidence
limits are shown around the line.

In the top diagram on the left, we see that all of the data
points roughly form a straight line and fall within the
limits. We would conclude that there is no serious
departure from normality in this data. In the other
diagrams, we see that many of the data points fall outside
the limits and do not form a straight line. We would
therefore conclude that these data sets significantly depart
from the normal distribution.

What types of process might result in data that follows a


bimodal, skewed, and long-tailed distribution?

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

34
What If Your Data Is Not Normal?

If you conclude that Y is non-normally distributed, there are


two general approaches:
ƒ Approach 1: Variance-based Thinking (VBT) Methodology
– possibly multiple processes embedded
– segmentation and stratification
– span reduction

ƒ Approach 2: Non-parametric techniques (beyond the scope of this course)

Expectation: Green Belts Should


Be Able To Do Approach 1

If your data is non-normal, you may be able to proceed. It


is important that you consult with your MBB before
proceeding.

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

35
Minitab–Histogram

Recruitment Cycle Time Case Study


Datafile: Fill_time.mtw
Question: What does the distribution of cycle 30
time to fill positions look like?

Frequency
20
Tool: Histogram
10
Data Type: Continuous. The X axis is cycle time,
the Y axis is the number of times
cycle time fell within the range of each 0
bar or interval in the histogram. The Y 0 10 20 30
axis is calculated for you.
Cycle Time
Data A single column of data for each
Structure: histogram

To Make A Graph > Histogram


Histogram: Variable: Cycle Time
Click OK

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

36
Minitab–Normal Probability Plot

Recruitment Cycle Time Case Study


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

Data Type: Continuous Y .001

2 12 22 32

Data Structure: A single column of data Cycle Time


Average: 16.3921 Anderson-Darling Normality Test
StDev: 5.61675 A-Squared: 0.208
N: 240 P-Value: 0.864

To Make A Normal Stat > Basic


Probability Plot: Statistics > Normality Test
Variable: Cycle Time
Test For Normality:
Choose Anderson-Darling
Click OK

Since p > 0.05, the data is normal.

There are a few ways to determine normality. We have


standardized on this Anderson-Darling normality test. We
will discuss p-values later in the Analyze phase.

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

37
“Center” Or Central Tendency

Descriptive Statistics: Normal Distribution


Represents the nominal value
of the process.

ƒ Mean (X)
x
ƒ Median (“middle” data point)
Long-tailed Distribution
ƒ Quartile Values (Q1, Q3)

Skewed Distributions

Q1 Q3

A good guess of central tendency can be made by


visually examining a histogram.

A more precise estimate of central tendency can


be made using descriptive statistics such as the mean,
median, or quartile values. The distribution shape dictates
which metrics should be used.

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

38
“Center” Or Central Tendency (continued)

The Mean, sometimes called the average, is


the most likely or expected value. The formula
for the mean is:

X=
∑X i
1. The sum of all data
values

n 2. Divide by number of data


values

The Median is literally the middle of the data


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

„ Order the numbers from smallest to


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

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

39
Variation

Descriptive Statistics Normal Distribution

Represents the variation of s


the process

ƒ Standard Deviation (s)


ƒ Span Long-tailed Distribution

ƒ Stability Factor (SF) = Q1/Q3

p=.01 p=.99

Skewed Distributions

Q1 Q3 Q1 Q3

A question to be answered in describing variation is


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

A precise measure of variation is accomplished using


basic descriptive statistics such as standard deviation,
span, or stability factor.

Standard Deviation–measures average distance from the


mean. Smaller is better.

Span–measures unusually low & high distance from the


median. Smaller is better.

Stability Factor–ratio of first quartile and third quartile,


1.0 is best.

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

Variation (continued) 40

Normal Distributions: Skewed Or Stable Ops Distributions:


The standard deviation is the average distance, or The stability factor (SF) is the ratio of the first
quartile and third quartile values. The formula for
deviation, that a given point is away from the
Stability Factor is:
mean. The formula for the standard deviation is:
Xi-X
SF = Q1/Q3
As Q1 and Q3 get closer together, their ratio
decreases and the variation of the middle 50%
A- subtract decreases.
each data
value from
the mean Xi X
Q Q
1 3
E- take the

∑ (X − X )
square root of
2 B- square each
the result
i difference Q Q
1 3
n −1 C- sum the
squared
differences
D- divide by 1 less than the
number of data values
Q More Q
1 1 Less
0 Variatio 1 Variatio
Q n Q
3 3 n

Long-tailed Distributions: 0 More Variation Less Variation 1

Span is used to indicate the amount of variation in a


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

Sample Size Span P


100-500 90%-10% P90, P10
501-2000 95%-5% P95, P5
2001-5000 98%-2% P98, P2
5001+ 99%-1% P99, P1
© 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

The difference between population and sample:


„ population has all the data points, N
„ sample only has a portion of the total data points,
n<N
The divisor for the population variance is the population
size N, whereas the divisor for the sample variance is the
sample size minus one (n-1). The divisor (n-1) is used
rather than N because this leads to an unbiased estimate
for the population variance.

© GE Capital, Inc., 2000

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

Results & Take-Aways:


ƒ Deviation = measure of Variation, measure of Spread
ƒ SST = 8726.1 … large Deviations = large SST

ƒ Variance = “Average” SST = SST/df [df = deg. of freedom = n-1]


s2 = Sample Variance = SST/9 = 969.6
σ2 = Population Variance = SST/10 = 872.6

ƒ Standard Deviation = Square Root of Variance:


It’s All Variation… Deviation, SST, Variance,
Sample StDev = s = 969.6 = 31.1
Population SD = σ = 872.6 = 29.5
And Standard Deviation = Measures Of Spread

A key concept is that the variance is the average SST for


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

As the degrees of freedom (df) increase (which is the


same as larger sample size (n)), the near average and
average SST become essentially the same.

Finally, the standard deviation is the square root of the


variance.

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

43
Summary–Measures
Summary–Measures Of Variation
Of Variation:

ƒ Range = (Max - Min)


ƒ Deviation = (Xi - Xbar) [Xbar = mean or mu]
ƒ Sum-of-Squares (of the squared Deviations) = SST
ƒ Variance = Average SST = SST/df (df = degrees of freedom)
ƒ Std Deviation = SqrRoot (Variance) and
ƒ Coeff. of Variation = CV = (StDev/Xbar)* 100
i.e., ratio of StDev to Mean–expressed as %

SST is key concept in measuring variation.

The greater SST, the greater the variation.

SST is always conserved: given a set of numbers (data), it is fixed


and unchanging.

SST can be allocated to different COMPONENTS OF VARIATION–


i.e., within and between group.

Components Of Variation: The Key For Process


Centering And ANOVA

Note that for any given set of numbers (Xis), the value
for SST is fixed, unchanging. The only way to change
SST is to change one or more of the numbers–which
means that the spread around the mean will likely be
changed.

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

44
Variation For A Period Of Time
Descriptive Statistics Summary

Shape Normality Plot Center Spread


(central tendency) (variation)
Normal Probability Plot for a Normal Distribution
ML Estimates
Mean:15.7224
StDev:1.74183

Standard
Percent

Mean X( ) Deviation (s)


normal
Normal Probability Plot for an Exponential Distribution

ML Estimates
Mean: 15.0790
StDev:12.6232

Quartile Stability Factor


Percent

Q1 or Q3 (SF)
skewed
Normal Probability Plot for a Long-Tailed Distribution

99 ML Estimates
Mean: 14.6382

( )
95
StDev:5.47084

~
90
Percent

80
70
60
50 Median X Span
40
30
20
10
5

long-tailed 1

0 10 20 30
Normal Probability Plot for a Bimodal Distribution
99 ML Estimates

The different processes must be stratified


Mean: 14.6382
95
StDev:5.47084
90
Percent

80
70
60
50
40
before descriptive statistics
can be calculated.
30
20
10
5

bimodal 1

0 10 20 30

Q1= 25th Percentile


Q2 = Median,, 50th Percentile
Q3 = 75th Percentile
Range = Maximum Value–Minimum Value
Span = 99th Percentile–1st Percentile
SF = Q1/Q3
IQR = Inner Quartile Range = Q3-Q1 (middle 50th of
data)

© GE Capital, Inc., 2000

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

After data is collected, it should be displayed graphically


and described using basic statistical measures. The first
graphical display used for describing variation is the
histogram.

Histograms combined with basic descriptive statistics


provide a clear picture of the current performance of a
process.

© GE Capital, Inc., 2000

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

Third Quartile (75%) value Q3


Each segment
represents
Median Q2
25% of the
data points First Quartile (25%) value Q1

Lowest Value

Box Plots are graphical summaries of the patterns of


variation in sets of data.

The horizontal lines at the top and bottom of the plot


represent the highest and lowest values. The horizontal
line in the middle of the solid box represents the median,
and the solid box represents the middle 50% of the data,
with 25% of the data on the top side of the box and 25%
of the data on the bottom side of the box.

Points noted with an asterisk(*) represent extreme values


or outliers.

Outlier calculations
High: Q3 + 1.5 (Q3-Q1)
Low: Q1 - 1.5 (Q3-Q1)

© GE Capital, Inc., 2000

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

Box Plots and histograms provide similar information but


in different graphical formats.

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

48
Minitab–Graphical Summary

Recruitment Cycle Time Case Study


Datafile: Fill_time.mtw

Question: How can I examine the shape,


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

Tool: Descriptive Statistics Mean


StDev
Variance
16.3921
5.6167
31.5478
Skewness 6.39E-02
Kurtosis -2.4E-01
N 240
4 10 16 22 28

Data Type: Continuous Y Minimum


1st Quartile
3.0800
12.4067
Median 16.6400
3rd Quartile 20.3067
95% Confidence Interval for Mu Maximum 32.2667
95% Confidence Interval for Mu

To Make A Stat > Basic Statistics > Display 16.0 16.5 17.0 17.5
15.6779 17.1063
95% Confidence Interval for Sigma
Histogram: Descriptive Statistics: 5.1552 6.1698
95% Confidence Interval for Median

Variable: Cycle Time 95% Confidence Interval for Median


15.7274 17.3441

Click Graphs button


Select Graphical Summary
Click OK twice–once in each
dialog box

This graphical summary is one of the most widely used „ Skewness–A measure of asymmetry. A value more
charts in Minitab. than or less than zero indicates skewness in the data.
Useful Items: But, a zero does not necessarily indicate symmetry.
„ Anderson-Darling normality TEST–We use the p- „ Kurtosis–A measure of how different a distribution is
value to determine whether our distribution is normal from the normal distribution. A negative value
or non-normal. If p > 0.05, the distribution is normal. typically indicates a distribution more peaked than the
(we will discuss p-value & hypothesis testing in normal. A positive value typically indicates a
Analyze, Step 6.) distribution flatter than normal.
„ Mean–Average of the samples. „ N–sample size.
„ StDev–Standard deviation of the samples. „ Confidence Interval for Mu–Tells us the interval in
which the true population mean lies (with 95%
„ Variance–(Standard Deviation Squared.) A measure of confidence.)
how far the data are spread about the mean.

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

49
Variation Over Time

Run Chart
ƒ A graphical tool to monitor the “stability of Project Y
ƒ Allows observation of time order properties such as trend
ƒ Should be used before any detailed data analysis

Example of a Run
Chart

Median

Is the process stable over time?

An important basic tool for understanding variation is the


Run Chart. Run Charts are simple time-ordered plots of
data. On these plots one can perform tests for certain
patterns in the data. Presence of these patterns indicate
special causes.

It is important to note that the data should be in time


order for run charts to be valid.

© GE Capital, Inc., 2000

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

Minitab Run Chart Oscillating Pattern

The Minitab Run Chart tests for two kinds of Runs: Oscillation represents more than expected number of
(A) Consecutive points on the same side of the median. runs of type (B).
(Thus a new run starts when the median is crossed).
(B) Consecutive points in the same direction up or down. Trend Pattern
(Thus, a new run starts when the direction is changed).
A trend represents less than expected number of runs
of type (B).
Mixture Pattern
At this point in time, our biggest concern is whether
A mixture represents more than expected number of runs the process is stable or not. We use the above-
of type A (above). mentioned checks to indicate stability. Minitab will do
this for us.
Cluster Pattern

A cluster represents fewer than expected number of runs


of type (A).

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

51
Minitab–Run Chart

Recruitment Cycle Time Case Study


Datafile: Fill_time.mtw Run Chart for Cycle Time
Question: Does cycle time show any 32
patterned variation over
time?

Cycle Time
22

Tool: Run Chart 12

Data Type: Continuous Y 2

40 140 240
Observation
Data Structure: A single column of data is all that Number of runs about median: 124.000 Number of runs up or dow n: 151.000
is necessary. It is assumed that Expected number of runs:
Longest run about median:
121.000
9.000
Expected number of runs:
Longest run up or dow n:
159.667
4.000
the data is entered in the order in Approx P-Value for Clustering:
Approx P-Value for Mixtures:
0.651
0.349
Approx P-Value for Trends:
Approx P-Value for Oscillation:
0.091
0.909
which it is collected (time
order).

To Make A Stat > Quality Tools > Run Chart


Run Chart: Single Column: cycle time
Subgroup size: 1

The 4 p-values circled above indicate stability. Since


p>0.05, there are no issues with Mixtures, Clusters,
Oscillations or Trends. Therefore, our process appears
stable (no special causes).

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

52
Two Types Of Variation

Investigating Common vs. Special Causes


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

For new process data, a Run Chart can be used to detect


special cause situations. For processes that are already in
statistical control, Control Charts are the preferred
method. Control Charts are discussed in the Control
phase of DMAIC training. But, here in the Measure and
Analyze phase we suggest using the Run Chart to test for
stability.

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

53
Two Types Of Variation (continued)

Reacting To Common vs. Special Causes


How you interpret variation . . .
Common Causes Special Causes

Focus on systematic Mistake 1


Common Tampering
process change
Causes (increases variation)
True
variation
type... Investigate
Mistake 2
Special Under-reacting special causes
for possible
Causes (missed prevention)
quick-fixes

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

Treating common causes as special causes is essentially over-


reacting to individual data points and is often referred to as
“tampering.” The problem with making this type of mistake is
that it typically leads to increased variation, costs, and defects.
It’s like a novice driver who over-reacts to “slop” in the steering
of a car. The result is more variation in the path of the car.

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

54
Graphical Analysis Tools

Looking For Patterns In Data

Continuous Y Discrete Y
Boxplot
Pareto Chart
Scatterplot

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

55
Box Plots

What differences do you see between the output from the


different shifts?
60
Measure

30

10

Shift 1 Shift 3 Shift 4 Shift 6


Shift 2 Shift 5

Box plots can be used to identify differences in variation


between subgroups that exist in your data.
Look for 2 major items here:
Center–are the medians similar?
Spread–do the boxes and the “Whiskers”have similar
heights?

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

56
Minitab–Box Plots

Recruitment Cycle Time Case Study


Datafile: Fill_time.mtw

Question: Is there a difference in variation


in cycle time between agents?
30

Tool: Box Plot


Cycle Time 20

Data Type: Continuous Y, Discrete X


10

To Make: Graph > Box plot


A Boxplot
Graph variables: Y Cycle time 0
1 2 3 4
X Agent
Agent
Click OK
Additional Questions: Is there a difference in
variation in cycle time between sites? Between
types of resource?

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

57
Scatter Diagrams–Analyzing Relationships

Use Scatter Diagrams To Study The Relationship Between


Two Variables.

40
Cycle Time (Days) (Y)

35

30

25

20

15

10

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

Size Of Loan (X)

A Scatter Diagram is an important graphical tool for


exploring the relationship between two continuous
variables.

Here, it appears that size of Loan and Cycle Time have a


relationship (Larger Loans take longer to process–
positive relationship).

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

58
Warning! Correlation Does Not Imply Causation

Correlation Between Number Of Storks And Human


Population
100 200 300
80 80

70 70
Population
(In Thousands)

60 60

50 50
100 200 300

Number Of Storks
Source: Box, Hunter, Hunter. Statistics For Experimenters. New York, NY: John Wiley & Sons. 1978

This is a plot of the population of Oldenburg at the end of


each year against the number of storks observed in that
year, 1930-1936.

Even strong correlations do not imply causation. (For


example, there will likely be a positive correlation–but
not causation–between the occurrence of vaporlocks in
automobiles and the use of public swimming pools.)

© GE Capital, Inc., 2000

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

Computers Are Computer’s Computer’s System


Too Slow Response Time Load

Delays Unloading # Of People On Time To Unload


Airline Baggage Plane Baggage

One of the most common mistakes that occurs with Scatter


Diagrams is mixing up the X and Y variables.

ƒ The X variable is the potential cause and is plotted on


the horizontal axis.

ƒ The Y variable is the effect and is plotted on the vertical


axis.

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

60
Interpretation Of A Scatter Diagram

Look For:
ƒ Common patterns in the data
ƒ Range of the predictor variable (X)
ƒ Irregularities in the data pattern

Common Patterns In The Data


ƒ See whether the potential cause variable and the effect
variable are related to one another.

Range Of The Predictor Variable (X’s)


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

Irregularities In The Data Pattern


ƒ Check whether the data pattern indicates possible
problems in the data.

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

61
Interpreting A Scatter Diagram

Look For Patterns


1 3 5

Strong Positive Correlation Strong Negative Correlation No Correlation

2 4 6

Positive Correlation Negative Correlation Other Pattern

For all charts: Y = Participant satisfaction (scale: 1 – worst to 100 – best)


X = Trainer experience (# of hours)

Interpretation Of Patterns: As trainer experience increases, participant satisfaction decreases


Positive Relationship: X Increases/Y Increases No Relationship: For any value of X, there are many values of Y
ƒ Graphs 1 and 2–as X increases, Y increases ƒ Graph 5–there is no relationship between X and Y
ƒ Graph 1–Strong, positive relationship between ƒ No one line best describes the data
X and Y ƒ 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
ƒ Graph 2–Weaker, positive relationship between X and Y complex and cannot be summarized with a straight line.
– Scatter is wider ƒ 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
ƒ Graphs 2 and 3–as X increases, Y decreases
ƒ Graph 3–strong negative relationship between X and Y
– Not much scatter
– Points form a line with a negative slope
ƒ 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

Recruitment Cycle Time Case Study


Datafile: Fill_time.mtw

Question: Is there an association


between cycle time and
years experience? 30

Tool: Plot
Cycle Time 20

Data Type: Continuous Y, Continuous X 10

To Make A Graph > Plot


0
Scatter Plot: Y Cycle time
0 5 10 15 20 25
X Years
Yrs Exp

Experience Click OK

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

63
Pairs Exercise–Analyze Patterns

With A Partner:
ƒ Review scatter diagrams on following
pages as assigned
ƒ Determine the type of relationship you
observe (e.g., negative, positive, strong, weak, etc.)
ƒ Describe an example from your business
of this type of relationship

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

Common Scatter Diagram Patterns (continued) 64

Strong,
Plot +/- Other Weak, Other
Example

1
Effect

Potential Cause

2
Effect

Potential Cause

3
Effect

Potential Cause

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

Common Scatter Diagram Patterns (continued) 65

Strong, Weak,
Plot +/- Other Other
Example

4
Effect

Potential Cause

5
Effect

Potential Cause

6
Effect

Potential Cause

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

Common Scatter Diagram Patterns (continued) 66

Plot +/- Other Strong, Weak, Example


Other

7
Effect

Potential Cause

8
Effect

Potential Cause

9
Effect

Potential Cause

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

Common Scatter Diagram Patterns (continued) 67

Plot +/- Other Strong, Example


Weak, Other

10
Effect

Potential Cause

11
Effect

Potential Cause

12
Effect

Potential Cause

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

68
Pareto Charts

Is There A Defect That Occurs Frequently?

Frequency

C A E D B
Category of Defect

We were introduced to Pareto charts in the CTQ tools


module.

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

69
Minitab–Pareto Chart

Recruitment Cycle Time Case Study


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

By Variable: Cycle _Time_Defect Count


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

Independent Ordering of Bars


Click: OK

Additional Which site has the most cycle


Questions: time defects? Which type of
resource has the most
cycle time defects?

To find out which site has the most defects: Follow steps Alternate method (subsetting a worksheet)
above except, place “Site” in “Chart Defects Data In” Box. MANIP > SUBSET WORKSHEET
ƒ Name: No Transactions
Alternate Method (Splitting A Worksheet)
ƒ Include or Exclude: CLICK ON “Specify which rows
to include”
MANIP > SPLIT WORKSHEET
ƒ Specify which rows to include:
ƒ By variable: Cycle_Time_Defect CLICK ON “Rows that Match”
Click: OK
ƒ CLICK ON “Condition” Button
ƒ 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”
ƒ Activate the correct worksheet and then follow the steps ƒ Click “OK” twice
in the slide above
ƒ Now you have a worksheet with “N” transactions

© GE Capital, Inc., 2000

DMAIC GB I TX PG V 4.2.0
D M A I C

© GE Capital, Inc., 2000

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

What does it mean to Establish A Data Collection Plan, Validate the


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

Why is it important to Establish A Data Collection Plan, Validate the


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

What are the project tasks for completing Measure 3?


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

MEASURE STEP
OVERVIEW

Measure 1: Measure 2: Measure 3:


Select CTQ Define Establish Data
Characteristic Performance Collection, Validate
Standards MSA

3.1 Develop a plan to collect data


3.2 Validate the measurement system
3.3 Collect data per plan

© GE Capital, Inc., 1999


DMAIC GB J TX PG V 4.2.0
D M A I C
Measure 3–Plan data collection, validate measurement
system, and collect data

Data Collection
Define Review Plan Worksheet
Objectives
2

Data collection purpose:

What to measure (output measure and segmentation factors)


Data Type Operational
Measure Measure Range Of
(Discrete)/ Definition
(Name) Type (Y, X) Values
(Continuous)

How to measure:
Sampling Plan (Scheme, Frequency, Size)

Measurement Procedure
Refer to more detailed documentation, if necessary.

© GE Capital, Inc., 1999


DMAIC GB J TX PG V 4.2.0
D M A I C

Measurement System
Define Review Analysis (MSA) Worksheet
Objectives
3

Type of Gage R&R Conducted (Check One): Date


‰ Gage R&R (Continuous Data) Conducted:
Gage R&R Results
1) Two-Way ANOVA Significant?

Part p-value Y N
Oper p-value Y N
Oper & Part p-value Y N

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

Graphical Output OK?


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

‰ Attribute Gage R&R


1) Repeatability [ >90% ]
2) % Reproducibility [ >90% ]
3) % Accuracy [ >90% ]
_______________________________________________________________
1) Gage R&R Pass? Y N, If NO:
Plan for improvement:
_______________________________________________________________
_______________________________________________________________
© GE Capital, Inc., 1999
DMAIC GB J TX PG V 4.2.0
D M A I C

Define Review Objectives

© GE Capital, Inc., 1999


DMAIC GB J TX PG V 4.2.0
D M A I C
1

Plan For Data


Define Collection
Review Objectives
1

Develop
Establish Ensure Data Collect Data
1 Data Collection
Goals
2 Operational
Definitions And
Procedures
3 Consistency
And Stability 4 And Monitor
Consistency

ƒ Clarify purpose ƒ Write and pilot ƒTest and validate ƒTrain data
of data collection operational definitions measurement collectors
ƒ Identify what systems
ƒ Develop and pilot ƒ Pilot process
data to collect
data collection forms and make
and procedures adjustments

ƒ Establish a sampling ƒ Collect data


plan
ƒ Monitor data
accuracy and
consistency

Data Collection Is The First Step To Understanding The


Variation The Customer Feels

Data collection occurs multiple times throughout


DMAIC. The data collection plan described here
can be used as the guide for data collection
regardless of where you are in DMAIC or what
type of data you are collecting. Using this model
will help ensure that you collect useful, accurate
data that is needed to answer your process
questions.

© GE Capital, Inc., 1999

DMAIC GB K TX PG V 4.2.0
D M A I C

Step 1: Establish
Define Review Data Collection Goals
Objectives
2

In order to establish your data collection goals you must:


ƒ State the purpose of the data collection
ƒ Identify what data is required

Asking these questions may help you clarify your goals:


ƒ What do I need to know about my process?
ƒ What data do I need?
ƒ What is the plan for analysis once the data is collected?
ƒ What data is already available?

© GE Capital, Inc., 1999

DMAIC GB K TX PG V 4.2.0
D M A I C

Preparing For TheObjectives


Define Review Analyze Phase 3

ƒ 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
ƒ Collect Project Y data to identify patterns and performance
trends and establish current baseline defect rate
ƒ Collect segmentation factor data to be used for later analysis

Segmentation Helps Us Understand


Variation In Project Y

ƒ 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.
ƒ 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.
ƒ Segmentation factors include external variables
outside our control that may explain patterns in
the Project Y measures.

© GE Capital, Inc., 1999

DMAIC GB K TX PG V 4.2.0
D M A I C

Common Segmentation
Define Review Factors
Objectives
4

Segmenting by external factors will help us identify the


drivers of variation in the process
Possible categories:
ƒ Customer
ƒ Product
ƒ Market
ƒ Time
ƒ Geography
ƒ Size of account
ƒ Degree of dissatisfaction

What Data Is Needed? 2. Y = On-time delivery of services


Project Y data, collected over X time period Objective:
Examples: ƒ Measure on-time delivery performance for
1. Y = Application cycle time customer service center
Objective: ƒ Identify possible sources of poor or irregular
performance.
ƒ Understand cycle time performance of credit
approvals process. Data Required:
ƒ Identify potential sources of poor or irregular ƒ Deviations from target date/time for individual
performances. orders (minutes/hours, early or late)
Data Required: ƒ Data for segmenting factors [e.g., customer,
market, geography, time (day, month, hour),
ƒ Physical cycle time data (collect several request type]
business cycles) for individual credit
applications.
ƒ 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

In service-related processes, “When” elements to


pay special attention to include: relationship to
deadline (e.g., end of quarter), time/number in
queue, or timing versus systems changes.

Note: Your team will need to segment the data in


several different ways in order to uncover where
the most significant differences occur.

© GE Capital, Inc., 1999

DMAIC GB K TX PG V 4.2.0
D M A I C

How To Collect
Define Review Data For Segmentation
Objectives
6

ƒ Identify the factors for segmentation before you start collecting


data
ƒ Make sure the segmentation factors can be measured reliably
ƒ Record the segmentation factors for each Y data point
collected
ƒ Segmentation factors are typically easy to collect, so collect
more segmentation factors rather than fewer

© GE Capital, Inc., 1999

DMAIC GB K TX PG V 4.2.0
D M A I C

Step 1: Breakout
Define Review Activity (5 Minutes)
Objectives
7

Objective ƒ Determine Segmentation Method

ƒ For your Project Y:


Instructions
– Brainstorm a list of segmentation factors
– Remember to also segment on “unlikely” parameters

Time 5 Minutes

© GE Capital, Inc., 1999

DMAIC GB K TX PG V 4.2.0
D M A I C

Step 2: Develop
Define ReviewOperation Definitions and Procedures
Objectives
8

Clearly specify variables to be collected:


ƒ Operational definitions for all metrics
ƒ Specific descriptions of how to take the measurement
Specify the details of the data collection process:
ƒ How to collect the data
ƒ How to record the data
ƒ The period of time for data collection
ƒ The sampling plan to be followed

To help ensure consistency, clearly define each


metric and the process for collecting it. Attention to
these details will help ensure that the data you
collect will give you an accurate picture of the
variation in your process.

© GE Capital, Inc., 1999

DMAIC GB K TX PG V 4.2.0
D M A I C

Operational Definitions
Define Review Objectives
9

Defining The Measure


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

Always Pilot Your Operational Definitions

ƒ Operational definitions guide what properties


will be measured and how they will be
measured.
ƒ 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.

© GE Capital, Inc., 1999

DMAIC GB K TX PG V 4.2.0
D M A I C

Operational Definitions–Scale
Define Review Objectives Of Scrutiny
10

Choosing The Level Of Measurement


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

The scale of scrutiny is how finely you measure


your process. Measuring one level smaller than
your customer allows you to more fully understand
and capture the variation in the process.

© GE Capital, Inc., 1999

DMAIC GB K TX PG V 4.2.0
D M A I C
1

Sampling
Define Review Objectives
1

Sampling is the process of:


Collecting only a portion of the data that is available or could be
available, and drawing conclusions about the total population
(statistical inference)
Population Sample

x x x
x x
x x
x x x x x x
x x x x x
x
x x x
N = 5,000 n = 100

What is the From the sample,


average we infer that the
resolution time? average
resolution time
( ) is 1.2 days

Sampling is the process of collecting a portion or


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

In the example above, the population is all of the


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

A random sample of 100 inquiries is made and the


average resolution is estimated.

© GE Capital, Inc., 1999

DMAIC GB L TX PG V 4.2.0
D M A I C

When To Review
Define Sample Objectives 2

When to sample
ƒ Collecting all the data is impractical or too costly
ƒ Data collection can be a destructive process
ƒ When measuring a high-volume process
When not to sample
ƒ A subset of data may not accurately depict the process, leading
to a wrong conclusion (every unit is unique – e.g., structured
deals)

Statistically Sound Conclusions Can Often Be Drawn


From A Subset Of The Total Available Data

One of the first questions to ask is ‘Do I need to


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

If the circumstances surrounding the data collection


plan do not justify sampling, then sampling should
not be done. This is often the case in low volume
processes (e.g., deal processes).

© GE Capital, Inc., 1999

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:
ƒ All parts of the target population are represented
(i.e., selected for measurement) equally
ƒ The customer’s view is captured
How to guarantee a representative sample:
ƒ Design a sampling strategy
ƒ Understand special characteristics of the population before
sampling

Regardless of the situation, a sample must be


“representative” of the population.

For practical purposes a sample is representative if it


accurately represents the target population.
Consideration that may hinder collection of a
representative sample include:

ƒ The cost and ease of obtaining samples


ƒ Time constraints
ƒ Unknown characteristics of the population

Samples that are not representative of a target


population are called biased samples. Often, the
biases are not recognized until the collected data has
been analyzed.

© GE Capital, Inc., 1999

DMAIC GB L TX PG V 4.2.0
D M A I C

Sample
DefineSize For Continuous
Review ObjectivesData
4

How Do I Determine Sample Size?


Sample size (n) depends on three things
ƒ 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
ƒ Precision or accuracy (∆) required in the result, “The error bars or uncertainty
in my result”
– Precision increases as sample size increases
ƒ Standard deviation of the population (σ), “How much variation is in the
total data population”
– An estimate of standard deviation is needed to start
ƒ As standard deviation increases, a larger sample size is needed to obtain
reliable results 2
 1.96 × σ 
n= 
 ∆ 

In this equation, “1.96” represents a 95% confidence level

Consider the following example: Therefore, from the statistical theory we can answer
We want to estimate average call length in handling according to the formula:
customer inquiries, and we want our estimate to be
accurate to within + 1 minute.  1.96 × σ 
2

n= 
 ∆ 
Based on a small random sample of 30 inquiries we
know that the variation in call length, as measured
by a statistic called the standard deviation, is 5 Where n = sample size, u = standard deviation and
minutes. ∆ = degree of precision required. In our example,
the required sample size is:
We want to have 95% confidence that the estimate n = [(1.96x5)/1] 2 = 96.04
will be in the range of specified accuracy – i.e., + 1 or 96 samples
minute.

© GE Capital, Inc., 1999

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

In service-related processes, “When” elements to


pay special attention to include: relationship to
deadline (e.g., end of quarter), time/number in
queue, or timing versus systems changes.

Note: Your team will need to segment the data in


several different ways in order to uncover where
the most significant differences occur.

© GE Capital, Inc., 1999

DMAIC GB L TX PG V 4.2.0
D M A I C

Calculating A Sample
Define Review Size
Objectives
6

Continuous Data To Derive Sample Size Formula

– σ Standard error
CI –X X ± Zα/2 * n

Given X ± ∆
σ
Solve for n ∆ = Zα /2 n
Z α /2 *σ
n =
∆ 2
n = (∆ )
Zα /2 *σ

n = sample size
∆ = precision of the estimate. How much error is
ok? The smaller ∆, larger the sample size. May be a
business decision.
CI = Confidence Interval
Zα/2 = Z score we usually set at 1.96 for 96%
confidence.
σ = estimated standard deviation

© GE Capital, Inc., 1999

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
ƒ Use an existing Xbar or R chart
σˆ = R /d2 where d2 is control chart factor

(
σˆ = UCL − X /3 )
ƒ Collect a small pre-sample & calculate s (n = 30)
ƒ Ask subject matter experts to take an educated guess at the
plausible range of data

σˆ = (Highest known value - Lowest known value )/6

Control Chart Factor d2

n d2
2 1.128
3 1.693
4 2.059 σˆ = R /d 2
5 2.326
6 2.534

Range or moving range chart should be in


statistical control (stable) to use this estimate of σ

© GE Capital, Inc., 1999

DMAIC GB L TX PG V 4.2.0
D M A I C

Sample
DefineSize For Continuous
Review ObjectivesData (continued)
8

What Is The Performance For Delivery Time?


Population
Y = Delivery 2 Calculate sample size (n) based on:
Time  1.96 × σ  Precision (∆)
n= 
(Days)  ∆  95% confidence Level (1.96)
Standard deviation (σ)

Sample

n
Values
∆ ∆

⌧-∆ ⌧ ⌧+∆
Calculate average (⌧)
Conclusion:

I know with 95% confidence that the population mean is ⌧ + ∆

Whenever samples are taken to estimate a Confidence intervals are important when precise
population there will be differences between the estimates of populations are required, and the
“true” population values and the sample values. degree of precision in the estimate needs to be
known.
Through statistical theory we can determine the
amount of variation we can expect in our estimates. For those statistically inclined, an important
This is known as a confidence interval. Confidence statistical theory supporting confidence intervals is
intervals are stated in terms of an interval and a the “central limits theorem”. The central limit
confidence level. theorem states that regardless the shape of the
population, sample averages will always be
For example, when estimating average inquiry “normally” distributed and inversely proportioned
resolution time from samples, we may obtain a 95% to the square root of the sample size.
confidence interval for the average 1.2 - 2.3 days.
This means that if repeated samples were taken
from the same population, 95 times out of 100 we
would expect the sample average to be between 1.2
and 2.3 days.

© GE Capital, Inc., 1999

DMAIC GB L TX PG V 4.2.0
D M A I C

Finite Population
Define Review Correction
Objectives
9

If You Have A Finite Population


1. Calculate sample size (n)

2. If n/N > .05


OR Where n = sample size;
If n > N… N = population size

3. Calculate n finite
nfinite = n/(1+n/N)

ƒ Operational definitions guide what properties


will be measured and how they will be
measured.
ƒ 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.

© GE Capital, Inc., 1999

DMAIC GB L TX PG V 4.2.0
D M A I C

Sample
DefineSize For Discrete
Review Data
Objectives
10

How Do I Determine Sample Size?


Sample size (n) depends on three things:
ƒ 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
ƒ Precision or accuracy (∆) required in the result, “The error bars
or uncertainty in
my result”
– Precision increases as sample size increases
ƒ 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 )
 ∆ 

ƒ In this equation, “1.96” represents a 95% confidence interval

You need to know the approximate proportion


defective for the population to be sampled to calculate
the sample size. For example:
P ∆ n
.05 .02 456
.50 .02 2400
.95 .02 456

This calculation procedure is based on the binomial


model and should only be used when nP > 5.

As for continuous measurements, if the sample size (n)


is more than 5% of the population size (N), the finite
population correction should be used n (finite) = n/(1 +
n/N) and n (finite) should be used as the sample size.

© GE Capital, Inc., 1999

DMAIC GB L TX PG V 4.2.0
D M A I C

Sample
DefineSize For Discrete
Review Data (continued)
Objectives
11

What Is The Defect Rate (P) Of A Process?


Population

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


 1.96 
Defective n=  P(1 − P ) Precision (∆)
 ∆  95% confidence Level (2)
Estimated proportion defective (P)

Sample

n
Values
ƒConclusion:
ƒI know with 95% confidence that the population
proportion defective is P + ∆
Calculate Proportion
Defective (P)

ƒ Recalculate n* based on the calculated P. If the new required sample size (n*) is more than the number
of samples taken, take (n*-n) samples and recalculate P based on the full sample size. If it is not
practical to take more samples, then use the actual n and P to recalculate the actual precision (∆).

Minimum Sample Size

2
 1.96 
n=  P(1 − P )
 ∆ 

Example
We want to estimate the defect rate (P) within +
0.02 (I.e., ∆ = 0.02). We expect P to be
approximately 0.05.

2
 1.96 
n=  0.05(1 − 0.05 ) = 456
 0. 2 

© GE Capital, Inc., 1999

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

Sample Size & P

0 .5 1
P

© GE Capital, Inc., 1999

DMAIC GB L TX PG V 4.2.0
D M A I C

Sample
DefineSize Considerations
Review Objectives
13

Beyond The Formulas…


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

What Is The Impact On The Customer If Your Sample Size


Is Not Representative Of The Process?

© GE Capital, Inc., 1999

DMAIC GB L TX PG V 4.2.0
D M A I C

Sampling ExerciseObjectives
Define Review
14

Objective Apply Sample-Size Formulas

1. In your table-team, answer the assigned questions and be


Instructions
prepared to report your answers. Assume confidence level of
95%.
2. Determine the sample size needed if the following is known.
P = 0.20, ∆ = 0.0784
3. Give an estimated proportion defective guessed to be 5%, how
many observations should we take to estimate the proportion
defective within 2%?
4. We want to estimate the average cycle time within 2 days. A
preliminary estimate of the population standard deviation is 8
days. How many observations should we take?
5. We want to estimate average hold time at one of our cost
centers within ± 2 seconds. We will assume hold time standard
deviation is 10 seconds. How many calls do we need to
sample?

Time 10 Minutes

© GE Capital, Inc., 1999

DMAIC GB L TX PG V 4.2.0
D M A I C

Measurement System Analysis–Objectives 1

ƒ Recognize that observed variation of a product/process


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

One of the objectives of the Measure Phase is to validate


your measurement system. A Gage R&R Study will help
us do this!

The focus of this module is to review the methodology


and tools to validate your measurement system.

MSA = Measurement System Analysis

Gage R&R = Gage Repeatability & Reproducubility

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Possible Sources Of Variation 2

ƒ Observations
Inputs Process Outputs Inputs Measureme
nt Outputs ƒ Measuremen
ts
Process
ƒ Data

Observed Process Variation

Actual Process Variation Measurement Process


Measurement Process
Variation
Variation

Long-term Short- Variation Variation “Other”


Variation Variation “Other”
Process term due to due to sources
due to due to sources
gage operator (Environment
Variation Process gage operator (Environment
, etc.)
Variation , etc.)

Accuracy
Accuracy Precision Discriminatio
Precision Discriminatio
(Bias)
(Bias) (Measureme nn
(Measureme (Resolution)
nt Error) (Resolution)
nt Error)

To Address Actual Process Variability; The Variation Due To


The Measurement System Must First Be Identified And
Separated From That Of The Process

Measurement System Variation Why is MSA important?

We will be evaluating the variation in our measurement „ Data is only as good as the process that
systems. We will include the variation due to the operator measures it
and gage (Accuracy, Precision and Discrimination). „ MSA identifies how much variation is present in
the measurement process
When collecting data, we are seeing the process through „ Understanding measurement variation is
the lens of our measurement system. We never really see necessary for identifying “true” process variation
the actual process variation-instead, we see the actual and maximizing true Y improvements
process variation and the measurement process variation. „ Without MSA, you run the risk of making
Therefore, the total observed variation can be broken decisions based on an inaccurate picture of your
down into 2 parts: the actual process variation and the process
variation (or measurement error) created by the
„ MSA helps direct efforts aimed at decreasing
measurement process.
measurement variation
„ Excessive measurement variation distorts our
understanding of what the customer feels

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Measurement Error vs. Process Variation 3

ƒ Call Taking Y= Time to Respond (Response Time)


– Process Variation: Time to answer a question varies for different
operators, locations, etc.
– Measurement Error: Error in capturing the time measurement due to
vague definition of when to stop.
ƒ 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.
ƒ 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.
ƒ 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

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Planning Your MSA 4

Questions to ask:
ƒ What are you measuring?
ƒ Who is measuring?
ƒ What do you use to measure (gage)?
ƒ How are you measuring? Do you always use the same
procedure?
ƒ Under what conditions are you measuring (6 m’s)?
ƒ What is the resolution of your measurement system? Does the
customer measure the same way?
ƒ Which analysis will you use (continuous, Attribute and/or
Destructive)?

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Performing An MSA 5

ƒ An MSA can be conducted with Continuous or Discrete Data


(we will review each method).
ƒ When conducting the MSA, the following guidelines are
suggested:
– Choose 2-3 “operators” who normally perform the measurements.
– For continuous data, choose 10 parts/samples (e.g., applications, calls,
deals, trucks, etc) to measure. The parts should be as dissimilar as
possible (within the normal measurement range). These same, ten parts
will be used throughout the entire MSA study.
– For discrete data, choose 30-40 parts/samples (higher is better).
– The parts must be labeled with a number. The part numbers will remain
constant throughout the MSA.
– Each operator will measure the parts 2-3 times. A blank data collection
form should be used for each trial.
– The Gage should be “calibrated” as per the calibration procedure prior to
conducting the MSA.

Choosing Parts for your Study: The “Parts” are NOT


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

Measuring the Parts: The operators should measure


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

Analyzing the Results: After all the data is collected,


you will run the MSA in Minitab (for Continuous data) or
in Excel (for Discrete data). The remaining pages in this
section explain the MSA output.

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Measurement System Analysis 6

Five Components Of An MSA:

Accuracy–the differences between observed average measurement and a


standard

Repeatability–variation when one person repeatedly measures the same unit


with the same measuring equipment

Reproducibility–variation when two or more people measure the same unit with
the same measuring equipment

Stability–variation obtained when the same person measures the same unit
with the same equipment over an extended period of time

Linearity–the consistency of the accuracy across the entire range of the


measurement system

Think in terms of shooting at a target.

Accuracy: How well do you hit the target? Do you hit


where you are aiming?

Repeatability: Can you keep hitting the target in the


same place?

Reproducibility: Can another person with the same gun


hit the target in the same place?

Stability: If you come back to shoot at other times are


you still able to shoot as accurately (consistency over
time)?

Linearity: Are you as accurate at 50 meters as you are at


250 meters?

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Process Flow For MSA (Measurement System Analysis) 7

Conduct MSA

Take Preventive And


OK?
Corrective Actions
No

Yes

Continue Process
Improvement
For Attribute R&R:

90% matches for each Repeatability, Reproducibility,


and Accuracy analysis.

Examples of Corrective Actions: Note: Attribute = Discrete


„ Review data and decisions
„ Institute stopgap measures

Examples of Preventive Actions:


„ Correct mechanical or definitional errors
„ Institute or upgrade training

Examples of Process Improvement:


„ Measure gage over time to address stability
„ Study linearity of gage

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Attribute Data AR&R’s 8

ƒ With Discrete Data, we will look at these things:


– Repeatability: Do repeated measures match within an operator?
– Reproducibility: Do repeated measures match between operators?
– Accuracy: Do the measures match the standard value?
ƒ An MSA helps us determine if our measurement system must
be improved, and if so, gives guidance as to how

Although it has been emphasized that it is best to find a Y


that is associated with continuous data, this is not always
possible. In some cases, the response will be discrete. The
measurement system must still be validated.

Attribute = Discrete

AR&R = Attribute Repeatability and Reproducibility


is the method we’ll use to analyze Discrete
Data.

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

AR&R Guidelines For Evaluation 9

ƒ Repeatability: 90% of repeated measures within an operator


match
ƒ Reproducibility: 90% of the repeated measures across
operators match
ƒ Accuracy: 90% of the individual measures match the standard

Guidelines for determining how to design an AR&R study: The examples that follow balance all
1. For accuracy: The sample size is the total number three characteristics of an AR&R study.
of measurements. (40 parts x 3 repeated measures
x 3 operators = 360 measures.)

2. For repeatability: The more repeated measures on


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

3. For reproducibility: The more operators, the better


the sensitivity. More units have a minimal effect on
sensitivity. (Rather than measure 40 parts with
3 operators, measure 10 parts with 12 operators.)

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

How To Determine Sample Size In The AR&R 10

How many samples should I take for an AR&R Study?


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

For example, if your process requires a Customer Service


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

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

AR&R Data Sheet 11

'True' Operator 1 Operator 2 Operator 3


Sample Answer Trial1 Trial2 Trial3 Trial1 Trial2 Trial3 Trial1 Trial2 Trial3
1 N N N N N N N N N N
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.

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

AR&R Example: Repeatability 12

Operator 1 Operator 2 Operator 3


Sample Tr1 Tr2 TR3 Match? Tr1 Tr2 Tr3 Match? Tr1 Tr2 Tr3 Match?
1 N N N Y N N N Y N N N Y
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

„ Repeatability of the measurement system is assessed


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

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

AR&R Example: Reproducibility 13

Operator 1 Operator 2 Operator 3


Sample Tr1 Tr2 TR3 Tr1 Tr2 Tr3 Tr1 Tr2 Tr3 Match?
1 N N N N N N N N N Y
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

„ Reproducibility (plus repeatability) of the measurement


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

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

AR&R Example: Accuracy 14

Std Operator 1 Operator 2 Operator 3 Not


Sample Value Tr1 Tr2 Tr3 Tr1 Tr2 Tr3 Tr1 Tr2 Tr3 Match?
1 N N N N N N N N N N 0
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

„ Accuracy of the measurement system is assessed by


determining the number of non-matches to the standard
for each individual measure
„ In this example,
– The “Non-Match?” column contains a count of non-
matches to the standard value (std value) for the unit
(row). For example, all operators matched the
standard for unit 2, so the number of non-matches is
0. For unit 3, operator 3 failed to match the standard
on trial 1, so the number of non-matches is 1.
– There are the 40 units *9 repeated measures, or 360
individual measurements. 40/360 = .1111 did not
match.
– The percent matched = 100 (1-.1111) = 88.89%
matched.
– Since 88.89% matched is very close to 90%
matched, gage accuracy technically fails but is
practically acceptable.
© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

AR&R Analysis Example: Accuracy (continued) 15

To summarize the AR&R MSA


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

Note: For Repeatability, 89.17% (and Accuracy


88.89%) we considered both of these as
“Passing” due to the closeness to 90%. Use your
judgment. If an improvement can be made, feel
free to improve the measurement system. The
90% rate is an approximation not a “hard” limit.

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Measurement System Analysis (MSA) Worksheet 16

Type of Gage R&R Conducted (Check One): Date


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

1) Two-Way ANOVA Significant?


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

Graphical Output OK?


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


‰ Attribute Gage R&R
1) % Repeatability [ >90% ] 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.

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Causes Of Measurement System Variation 17

If you fail the Gage R&R, here are some factors that could
cause measurement process variation (measurement error).

Repeatability
ƒ Operational definitions
ƒ Maintain stability

Reproducibility
ƒ Operational definitions
ƒ Consistent use of gage
ƒ Training
ƒ Varying work environment
ƒ AUDIT-follow-up on training
ƒ Human/physical characteristics
ƒ Performance measures
ƒ Unclear requirements

If the Gage R&R fails, you need to improve the


measurement system before moving on. Use these factors
above as places to begin looking for areas of variation in
your measurement system. By improving some of these
factors, your measurement system variation should be
reduced (or brought to an acceptable level).

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Causes Of Measurement System Variation (continued) 18

Accuracy
ƒ Error in master
ƒ Instrument used improperly by appraiser
ƒ Operational definition
ƒ Standard not understood

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Other Considerations 19

Temporal Effects
ƒ Gage R&R depends upon our ability to measure things multiple
times
ƒ If the item is an event, it may not happen the same way twice.
For this reason, Gage R&R may be impossible.
ƒ If the event is recorded, it may be possible to conduct a Gage
R&R study
ƒ If the event cannot be recorded, but multiple judges can
observe at once, reproducibility can be estimated, but
repeatability cannot

Video cameras or audio voice recordings can allow a


Gage R&R to be conducted. Use today’s technology to
help you in validating your measurement system.

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Other Considerations 20

Temporal Effects Example 1


ƒ 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.

ƒ For this same case, repeatability can be estimated only if the


competition is taped. A random sample of 10 performances
shown to five judges, two times each, would allow estimates of
both repeatability and reproducibility.

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Other Considerations 21

Temporal Effects Example 2


ƒ Call centers field service calls from around the world at
centralized locations. Call duration and call quality are both
recorded and tracked very closely.

ƒ Call quality is calculated based on a scoring sheet filled out by


a quality monitor. Calls are scored on a continuous scale from
1-100 and can be assumed to be continuous for the purposes
of a Gage R&R calculation.

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Other Considerations 22

Temporal Effects Example 2 (continued)


ƒ 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.

ƒ In another monitoring scenario, calls are recorded at random


and scored at a later date. For this recorded scenario, both
repeatability and reproducibility may be calculated.

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Continuous Data MSA, First Step 23

Test-Retest Study

© GE Capital, Inc., 2000

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.

Why: Determine the precision of the system, instrument, device, or gage–where


Precision = Measurement Error = Repeatability

How:
ƒ Repeatedly measure the same item
ƒ Same conditions, operator, device, and location on item–same, same,
same
ƒ Completely mount and dismount item for each measurement–exercise gage
through full range of normal use

Data: Twenty (20) or more measurements is an adequate sample size. If


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

Test–Retest studies are sometimes called Calibration


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

A Test–Retest Study will indicate if you have either a


precision or accuracy problem.

A Test–Retest Study is typically run using a certified or


working “Standard.” The same item is measured
repeatedly.

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Test–Retest Study Guidelines 25

ƒ Device precision should be less than 1/10 of the tolerance:

s < 1/10 x tolerance*

If s exceeds 1/10 x tolerance*, then the measurement system is


unacceptable because it lacks precision. Action is required to find
and remove the sources of this error, including the replacement of
the device.

ƒ Device accuracy may be estimated if you know the true value


of the test unit:

Inaccuracy = Bias = X - True Value

Precision is a term used to describe the amount of


measurement error (repeatability error) in a measurement
system.

Precision is also used to describe the term ∆ (delta) in


confidence intervals. There it indicates ½ width of a
confidence interval.

If your gage lacks precision or accuracy, you may need to


remove or replace the measurement device. Perhaps a
better or updated device can be used.

*For a one-sided specification you can: (1) Use the mean


minus the tolerance. Note: the new formula would be:
(½) S ≤ 1/10 (USL – X) or (½) S ≤ 1/10 ( X-LSL). (2) use
the natural boundary (for example, use 0 for a cycle time
measurement) as the other spec. (Tolerance = USL-0).

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Process Flow For MSA (Measurement System Analysis) 26

Conduct MSA

Take Preventive And


OK?
Corrective Actions
No

Yes

Continue Process
Improvement

For continuous data we’ll look at these major items: %Tolerance,


% Contribution, % Study and # Distinct Categories.

MSA is a set of methods for estimating the current


amount of variation in the Measurement System.

Examples of Corrective Actions:


„ Review data and decisions
„ Institute stopgap measures

Examples of Preventive Actions:


„ Correct mechanical or definitional errors
„ Institute or upgrade training

Examples of Process Improvement:


„ Measure gage over time to address stability
„ Study linearity of gage

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Precision & Accuracy 27

Target Analogy

== Xbar

True Value = Bull's Eye

I. Precise, not accurate


Xbar

==

True Value

II. Accurate, not precise


Xbar

==

True Value

III. Precise and accurate

Precision: What kind of repeatability does the whole


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

Accuracy: Does the average reading agree with the actual


size of the part? (mean)

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Measurement System 28

ƒ Observations
Inputs Outputs Inputs Measurement Outputs ƒ Measurements
Process Process ƒ Data

Product Measurement Total


Variation Variation Variation
(Actual (Observed
Example variation) variation)

#1

Example σ 2Actual (Part) + σ 2Meas. System = σ 2Observed (Total)


#2

Measurement
System Variability
- Investigated
through “R&R
Study”

The measurement system will also contribute some σ2 = variance


variation to the overall total observed variation.

Here are two examples of the effect of measurement


variation. This variation comes from both the gage and
the operator.

In a Gage R&R Study, we will determine how large the


measurement variation contributes to the overall variation
we’re observing in our process. The key will be to
minimize the measurement variation so that we are
mainly seeing process variation.

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Types Of Variation Estimated By The Gage R&R 29

Equipment Variation (EV)


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

Appraiser Variation (AV)


(Reproducibility)
(Source of variation from across the process)
Across Gages–Across Operators–
Across Parts/Process–Etc.
The variation introduced into the measurement process by
effects going across the measurement process–such as
different appraisers–different part configurations–different
checking methods.

Equipment variation–variation within a sample group.

Appraiser variation–variation between sample groups.

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Relationship Between EV, AV And R&R 30

R&R is the Reproducibility (AV) and Repeatability (EV) of the


Measurement System. It represents the total variation in the
Measurement System.

R&R

EV AV

σ 2Equipment + σ 2Appraiser = σ 2Total (R&R)

EV = Equipment Variation Repeatability: variation when one person


repeatedly measures the same unit with the same
AV = Appraiser Variation measuring equipment.

The variation due to the equipment and appraiser do not Reproducibility: variation when two or more
directly add up to determine the total (R&R) variation. people measure the same unit with the same
There is an overlap between EV and AV. You can use the measuring equipment.
Pythagorean theorem (right angles) to add to the
influence of each EV and AV to calculate the total
variation.
σ 2= variance

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Relating R&R To Specification Window 31

How much of the tolerance is used up by the Measurement


System variation?

99%

R&R
(5.15σ)

Lower Spec. Limit Upper Spec. Limit


Specification Window (tolerance)

About 50% Of The Tolerance In This Example Is Used Up By The


Measurement System Variation This Leaves Only 50%
For The Process Variation

We’re interested in how this variation compares to the


tolerance with which we are working.

If the measurement system takes up lots of variation, we


don’t have any room for process variation. Minitab calls
this variation % Tolerance. This is one or our 4 key
metrics in the Gage R&R Study.

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Gage R&R For Continuous Data: Example 32

GB Case Study: On Time Delivery of Shipments to Customers

Background: The dispatchers for “Capital Logistics” keep a


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

First, we will walk through the process to conduct a Gage


R&R Study in Minitab, analyze the results and draw
conclusions. We will use Capital Logistics’ study of the
delivery of shipments to customers.

Then, you will do an activity to conduct a Gage R&R


Study in Minitab with a new set of data from the same
Capital Logistics’ study.

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Gage Reproducibility & Repeatability (GR&R) Study: Steps 33

1. Set up a Minitab Worksheet


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

We have standardized on using the “ANOVA” method in


Minitab. ANOVA simply stands for Analysis Of
Variance. Minitab will partition the total variation and
allocate this variation to gage variation or process
variation. In the Gage R&R Study, we analyze the gage
variation and want to understand how much gage
variation is present. If we have too much measurement
variation, we’ll need to fix our measurement system prior
to collecting any data. We want to ensure the variation
we are observing is mainly process variation.

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Case Study Setup 34

Does my measurement system provide me a clear view of my


process...
To help answer this question the GB
established the following analysis:

1. The GB developed a master matrix which listed the times


each truck was supposed to arrive at the customer. The GB
gave a copy of the matrix to each of the three dispatchers.

2. The GB had each of the truck drivers radio a message


from a radio truck, identifying himself as one of the 10 units
listed in the matrix..

3. So that the GB could get repeatability & reproducibility


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

4. Therefore, both days, the dispatchers, all in the same


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

5. The GB then consolidated the data into a single Minitab


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

…or does it cloud what I see?

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Gage R&R For Continuous Data: Example 35

Step 1: Set up Minitab data sheet.


ƒ GB asked 3 dispatchers to record the truck driver call-in time
for the 10 different truck drivers.
ƒ 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).
ƒ Open the File: Gage R&R–Continuous Data.mtw

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Gage R&R For Continuous Data: Example (continued) 36

Step 2: Perform Study–Collect & Enter Data.


ƒ Truckers called in to report delivery and dispatchers recorded
the difference from target time.
ƒ The data was collected and recorded on a data sheet and was
input into Minitab (see Data column).

Minitab File: Gage R&R – Continuous Data.mtw

Use Minitab file: Gage R&R–Continuous Data.mtw Columns 1-3 in the Minitab worksheet represent
the data collected by the dispatchers as each
truck driver called on their delivery times.

1. Part/Truck–indicates which of the 10 trucks

2. Oper–indicates which one of the 3 dispatchers

3. Meas–indicates the difference, in minutes,


from the target time

4. Trial–indicates the trial number when the


operator took the measurement.

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Gage R&R For Continuous Data: Example (continued) 37

Step 3: Perform Calculations & Prepare Charts


Perform Gage R&R Study–ANOVA method

Now let’s perform the Gage R&R study on this In Minitab:


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

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Gage R&R For Continuous Data: Example (continued) 38

ƒ Minitab Output will generate the following items which we will


use to determine if our measurement system is acceptable:
1. Graphical Summary
2. Two-Way ANOVA Table with Interaction (p-values)
3. % Comparisons Table (% Contribution)
4. % Comparisons Table (% Tolerance)
5. % Comparisons Table (% Study)
6. Discrimination Index

We will cover each of these 6 outputs in this section.

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

The Central Question In A Gage R&R Study 39

There are 2 ways to (1) FUTURE STATUS: (2) PRESENT STATUS:


answer the question Will my measurement Is my measurement
“Is my measurement variation be too large variation too large right
variation too big?” when I reach the 6-Sigma now, compared to a
goal (i.e., with small process realistic estimate of current
spread, good capability)? process spread (which
may be narrow next year)?

Number
Numberof of R&R
R&R%%Study
Study
R&R
R&R Distinct
Distinct
R&R
R&R Variation
Variation
%Contribution
%Contribution Categories
Categories
%Tolerance
%Tolerance (Ratio
(Ratioof
ofStdDev’s
StdDev’s--
(Ratio
(Ratioof
ofVariances)
Variances) (Discrimination
(Discrimination --AIAG)
AIAG)
Index)
Index)
Red

30%
30% 8%
8% 44 30%
30%
Yellow

10%
10% 2%
2% 10
10 15%
15%
Green

CAUTION:
The magnitude of these
%’s are exaggerations

„ If the Gage R&R (as a percent of tolerance) is less than „ If the discrimination index (number of distinct
10%, the MS is green, or acceptable; if Gage R&R is categories) is greater than 10, the measurement system
10-30% , the MS is yellow, or marginal; if Gage R&R (MS) is green, or acceptable; if the discrimination
is greater than 30%, the MS is red, or unacceptable. index is 4-10, the MS is yellow, or marginal; if the
This estimate may be appropriate for evaluating how discrimination index is less than 4, the MS is red, or
well the measurement system can perform with respect unacceptable. The discrimination index is just a
to specifications. We can use it to look into the future transformation of %R&R contribution (for ease of
and ask “Will my measurement system variation be too interpretation.) It addresses the same issues and
large when I reach the 6 Sigma goal (with small contains the same information.
process spread, good capability)?” „ If Gage R&R (as a percent study) is less than 15%, the
„ If Gage R&R (as a percent contribution) is less than MS is green, or acceptable; if Gage R&R is 15-30%,
2%, the MS is green, or acceptable; if the Gage R&R is the MS is yellow, or marginal; if Gage R&R is greater
2-8%, the MS is yellow, or marginal; if Gage R&R is than 30%, the MS is red, or unacceptable.
greater than 8%, the MS is red, or unacceptable.
Note: the limits shown above can be derived from one
another. Some rounding was done for ease of
interpretation.

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Rules Of Thumb–Acceptable Ranges 40

Step 4: Analyzing Gage R&R Results

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

2. R&R 10% to 30%–May be acceptable–make decision


based on classification of Characteristic, Application,
Customer Input, etc.

3. R&R over 30%–Not acceptable. Find problem, re-visit the


Fishbone Diagram, remove Root Causes. Is there a better gage on the
market, is it worth the additional cost?

RULES OF THUMB:

1. Less than 10% = Good

2. 10% to 30% = Fair–improvements possible; look to


find opportunities in process and training.

3. Over 30% = Problem–measurement system not


acceptable; fix problem with training, improved
instruments, correct measuring process, and/or
improved operational definitions for data collection.
The gage is not adequate for Product/Process
Acceptance decisions.

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Rules Of Thumb (continued) 41

B. % Contribution (or Gage R&R StdDev):


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

Questions to ask for validating the measurement system:


„ Are we capturing the correct data? Does the data reflect
what is happening in the process?
„ How big is the measurement error?
„ Can we detect process improvement if and when it
happens?
„ What are the sources of measurement error?
„ Are the measurements being made with measurement
units which are small enough to properly reflect the
variation present?
„ Is the Measurement System stable over time
„ Is the Measurement System “capable” for this study?
„ How much uncertainty should be attached to a
measurement when interpreting it?
„ How do we improve the measurement system?

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Rules Of Thumb (continued) 42

C. Number of Distinct Categories


A “Signal-to-Noise” Ratio = (StdDevparts/process/StdDevGR&R) x 1.41 and
rounded
Guidelines:
< 2 =>no value for process control, parts all “look” the same
= 2 =>can see two groups–high/low, good/bad
= 3 =>can see three groups–high/mid/low
≥ 4 =>acceptable measurement system
(higher is better)

D. R&R % Study
1. % Study less than 15%–measurement system acceptable.
2. % Study 15% to 30%–may be acceptable (marginal)
3. % Study over 30%–not acceptable

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Minitab Gage R&R–Session Window Analysis 43

Here are the Results:


% Comparison Table–Found in our Minitab Session Window

Source %Contribution %Study Var %Tolerance

Total Gage R&R 10.67 Fails! 32.66 Fails! 171.53 Fails!


Repeatability 3.10 17.62 92.54
Reproducibility 7.56 27.50 144.43
Oper 2.19 14.81 77.76
Oper*Part/Truck 5.37 23.17 121.70
Part-To-Part 89.33 94.52 496.41
Total Variation 100.00 100.00 525.21

Number of distinct categories = 4 pass!

We fail the Gage R&R!

„ % Tolerance is 171.53% which fails, based on our „ % Study, probably the least important, (because %
criteria.This shows us the % of the Tolerance which is contribution tell us roughly the same thing) also fails
being taken up by the gage. In this case, 171.53% of based on our criteria. This gives us a measure of gage
our tolerance is taken up by the variation of the gage. variation compared to 99% of a normal distribution.
Note: You must put in the tolerance in Minitab in
order to get this data. If you do not enter a tolerance,
these figures will not appear. If you do not have a „ Number of distinct categories passes! This tells us the
tolerance, then % Contribution would be more number of divisions the measurement system can
important in your analysis accurately measure across the seen process variation.
.
„ % Contribution is 10.67% which fails our criteria. % „ For a more detailed explanation of output, refer to the
Contribution is based on overall measure of gage end of this section.
variation (regardless of the specification).

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Minitab Gage R&R–Graphical Analysis 44

Now Let’s Review the Graphical Summary:

Gage name:
Date of study :
Gage R&R (ANOVA) for Meas Reported by :
Tolerance:
Misc:

Xbar Chart by Oper Oper*Part/Truck Interaction


110 1 2 3 110 Oper
100 100 1
Sample Mean

90 3.0SL=87.96 90 2

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

R Chart by Oper By Oper


15 1 2 3 110
Sample Range

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

Components of Variation By Part/Truck


500 110
%Total Var 100
400 %Study Var 90
Percent

300 %Toler 80
3 200
70
60 6
100 50
0 40
Gage R&R Repeat Reprod Part-to-Part Part/Truck 1 2 3 4 5 6 7 8 9 10

Chart 4: Chart 5:
The chart shows the 10 parts and the average This chart shows the average reading for all parts by
measurement of those parts broken-out by operator. operator. The red circle is the average. This could show
Ideally, all 3 lines should be on top of each other. We can us if one operator is measuring higher/lower (on average)
see that there is some measurement variation in parts 4, 8 than the others.
and 10. We should investigate why.

Note: For more detailed explanation, refer to the end of


this section

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Next Steps For The Green Belt 45

ƒ Sit down with your BB and/or Mentor to ensure proper analysis


ƒ After an initial look at the Gage R&R Results for this example,
the Green Belt will investigate the following:
– Look at the feasibility of changing the gage to a digital clock
– Talk with the operators to find out why they are measuring differently
– Can a spreadsheet be developed to calculate time automatically?
– Can we implement a driver call-in procedure to reduce variation?
– Why was it harder to measure parts 4, 8 and 10?

Continue on to see the GB’s Next Steps…

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Causes Of Measurement System Variation 46

If you fail the Gage R&R, here are some factors that could
cause measurement process variation (measurement error).
Repeatability
ƒ Operational definitions
ƒ Calibrating too often
ƒ Granularity of the measure (e.g., nearest hour, minute, second)
ƒ Maintain stability

Reproducibility
ƒ Operational definitions
ƒ Consistent use of gage
ƒ Training
ƒ Varying work environment
ƒ AUDIT–follow-up on training
ƒ Human/physical characteristics
ƒ Performance measures
ƒ Unclear requirements

If the Gage R&R fails, you need to improve the


measurement system before moving on. Use these factors
above as places to begin looking for areas of variation in
your measurement system. By improving some of these
factors, your measurement system variation should be
reduced, (or brought to an acceptable level).

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Causes Of Measurement System Variation (continued) 47

Accuracy
ƒ Error in master
ƒ Instrument not calibrated
ƒ Worn components/trend or drift in master
ƒ Instrument used improperly by appraiser
ƒ Calibrating too often
ƒ Operational definition
ƒ Range of the master

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Causes Of Measurement System Variation 48

Linearity
ƒ Instrument calibrated incorrectly
ƒ Error in master
ƒ Worn instrument

Stability
ƒ Error in master
ƒ Worn instrument
ƒ Instrument measuring wrong characteristic
ƒ Instrument not calibrated properly
ƒ Instrument used improperly by appraiser

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Gage R&R Action Plan 49

Step 5: Investigate sources of variation in the measurement


system and make recommendations–keep, improve or replace the
measurement system.

After completing the analysis of the measurement system, the


Green Belt investigated the sources of variation. By observing the
process, the GB found:
ƒ 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.

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Gage R&R Action Plan 50

Step 5 (continued)
ƒ 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:
ƒ 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.
ƒ 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.

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Gage R&R For Continuous Data–Activity (15 minutes) 51

Desired Outcome: Determine if the improved measurement system


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

What How Who Timing


Preparation ƒ Choose a facilitator, scribe

Conduct ƒ Complete 5 steps for a Gage Team 10 mins.


MSA R&R study and record results
ƒ Use File: Gage R&R–Continuous
Data.mtw and use columns
C6-C9

Report ƒ Complete Report Summary Team 5 mins.


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

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Measurement System Analysis (MSA) Worksheet 52

Type of Gage R&R Conducted (Check One): Date


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

1) Two-Way ANOVA Significant?


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

Graphical Output OK?


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

‰ Attribute Gage R&R


1) Repeatability [>90%]
2) % Reproducibility [>90%]
3) % Accuracy [>90%]
Gage R&R Pass? Y N, IF NO:
Plan for improvement:________________________________________
__________________________________________________________

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Analyzing The Gage R&R Results 53

Rules of Thumb
1. R&R (% Tolerance)
ƒ Less than 10%–Measurement System is acceptable
ƒ 10% to 30%–maybe acceptable–make decision based on classification of
characteristic, hardware application, customer input, etc.
ƒ 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”

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Analyzing The Gage R&R Results (continued) 54

3. Number of Distinct Categories = a “Signal-To-Noise”


Ratio = (StdDevparts/StdDevGR&R) X 1.41 and rounded
Guidelines:
< 2–no value for process control, parts all “look” the same
= 2–can see two groups–high/low, good/bad
= 3–can see three groups–high/mid/low
> 4–acceptable measurement system (higher is better)

4. R&R % Study
1. % Study less than 15%–measurement system acceptable
2. % Study 15% to 30%–may be acceptable (marginal)
3. % Study over 30%–not acceptable

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Summary–Measurement System Analysis 55

ƒ Variation in the measurement system will contribute to the


observed variation in a process
ƒ 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.
ƒ Minimize the measurement process variation
ƒ 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
ƒ Understand how measurement error impacts your customer
ƒ Measurement error is always a bigger deal than you think!

Make Sure Your MSA Is Examining The Actual


Measurement System Itself

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Measure Step 3–Green Belt Project Activity (30 minutes) 56

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

What How Timing


Your Individual 1. Data Collection plan* 15 mins.
GB Project
ƒ Review and finalize your Data
Collection Plan for your GB project
Have you answered:
ƒ What data will you need to
collect?
ƒ How will you collect it? From
where?
ƒ Who will collect the data?
(can refer to the data collection
worksheet)

2 . MSA* 15 mins.
ƒ Develop an MSA for your project Y
data
ƒ List factors that might cause
measurement system variation and
how you would reduce the impact
of those factors

*Refer to the Self-Paced Workbook pp 43-49.

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Gage R&R Details–Optional Information 57

Pages 58-69 contain Optional/Additional


information that will not be covered in
the 3-Day Workshop

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Full Gage R&R Details–Optional Information 58

Graphical Output
•ANOVA
•ANOVAsimplysimplystands
standsfor
for““Analysis
AnalysisofofVariance”
Variance”
•It•Itisisaatool
tool used to analyzethe
used to analyze thetotal
totalvariability
variabilityamong
amongdifferent
differentsources.
sources.
•It•Itpartitions
partitions the total variation into “buckets”, and then allocateseach
the total variation into “buckets”, and then allocates eachbucket
buckettotoeach
eachsource
source
(part,
(part,operator,
operator,oper*part
oper*partinteraction).
interaction).
Gage name:
Date of study :
Gage R&R (ANOVA) for Meas Reported by :
Tolerance:
Misc:

Xbar Chart by Oper Oper*Part/Truck Interaction


110 1 2 3 110 Oper
100 100 1
Sample Mean

90 3.0SL=87.96 90
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

R Chart by Oper By Oper


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

Components of Variation By Part/Truck


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.)

3 “D” should be largest, then A, B & C approximately


equal.

4 Parallel lines will show no problems with certain parts


(trucks.)

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Minitab Gage R&R ANOVA Graphical Output 59

Average of 2 measurements
taken by operator 1. Gage name:
Date of study :
Gage R&R (ANOVA) for Meas Reported by :
Tolerance:
Misc:

Xbar Chart by Oper Oper*Part/Truck Interaction


110 1 2 3 110 Oper
100 100 1
Sample Mean

90 3.0SL=87.96 90 2

Average
80 X=80.75 80
-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

R Chart by Oper By Oper


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

Components of Variation By Part/Truck


500 110
%Total Var 100
Range of 2 measurements 400 %Study Var 90
Percent

taken by operator 1. 300 %Toler 80


70
200
60
100 50
0 40
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.

In this example, the Xbar chart looks OK (more than 50%


of the points are outside the control limits.)

R Chart by Operator:
„ Plots the range of each average point in the Xbar chart.
This is the difference between the highest and lowest
measurement for all 10 parts for each operator.
© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Minitab Gage R&R ANOVA Graphical Output 60

Gage name:
Date of study :
Gage R&R (ANOVA) for Meas Reported by :
Tolerance:
Misc:

Xbar Chart by Oper Oper*Part/Truck Interaction


110 1 2 3 110 Oper
100 100 1
Sample Mean

90 3.0SL=87.96 90 2

Average
80 X=80.75 80
-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

R Chart by Oper By Oper


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

Components of Variation By Part/Truck


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

Graphical Display MSA Indicators by Components of


Variation:
„ This is a graphical display of the components of
variation. You can view the actual data in the session
window.
„ % Total Var ==> same as saying % Contribution
„ % Toler ==> % Tolerance

Remember, we want the variation in the parts (Part-to-


Part) to be the major piece of variation and we want the
variation from the gage itself to be small. Therefore, we
want “A-C” to be small and “D” to be largest.

In this case, the total variation of the parts is contributing


89.33% and the variation in the gage is contributing
10.67% (can be found in the session window). We’ll
learn more later on % Contribution and % Tolerance.

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Minitab Gage R&R ANOVA Graphical Output 61

Gage name:
Date of study :
Gage R&R (ANOVA) for Meas Reported by :
Tolerance:
Misc:

Xbar Chart by Oper Oper*Part/Truck Interaction


110 1 2 3 110 Oper
100 100 1
Sample Mean

90 3.0SL=87.96 90 2

Average
80 X=80.75 80
-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

R Chart by Oper By Oper


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

Components of Variation By Part/Truck


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.

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Minitab Gage R&R ANOVA Table 62

ANOVA TABLE
Two-Way ANOVA Table With Interaction

Source DF SS MS F P

Part/Truck 9 20587.1 2287.45 39.7178 0.00000


Sources of Oper 2 480.0 240.00 4.1672 0.03256
Variability Oper*Part/Truck 18 036.7 7.59 4.4588 0.00016
Repeatability 30 387.5 12.92
Total 59 22491.2
P< 0.05 indicates statistically
significant!

Are any of these “p-values” less than 0.05?


If so, their sources of variation can be considered statistically significant (i.e,
Active, Influential).

For this gage... parts, operators and the interaction between parts & operators
are statistically significant sources of variation!

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

Oper: p=0.03256; This is telling us that the Operators are


statistically significant. I need to look further to find out
if it’s a repeatability or a reproducibility issue (Minitab
does this for us in both the Comparisons Table and in the
charts. We’ll show you later!)

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Minitab Gage R&R % Tolerance 63

% 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

If you wanted to calculate %Tolerance by hand:


5 .1 5 • σ R & R ( M e a s . S y s t .)
% T o le r a n c e = ∗1 0 0 %
T o le r a n c e

This shows us the % of the Tolerance which is being „ If % Tolerance is unacceptable now, then once you
taken up by the gage. For instance, in this case, 171.53% reach a Six Sigma Process, the measurement variation
of our tolerance is taken up by the variation of the gage. will be too large compared to the process variation.
Refer to the next page for a visual explanation of % „ In this case, the gage fails % Tolerance.
Tolerance. Note: You must put in the tolerance in
Minitab in order to get this data. If you do not enter a
tolerance, these figures will not appear. If you do not
have a tolerance, then % Contribution would be more
important in your analysis.

% Tolerance = A good indicator for Future Status


„ For this Gage: %Tolerance GRR = 171.53%
„ This Gage Fails the Criteria for %Tolerance for the
Total Gage RR Piece.

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Minitab Gage R&R % Tolerance Graphically 64

%Tolerance: Graphically

LSL USL

%Tolerance = 10%

%Tolerance = 30%

%Tolerance = 70%

Graphically, this shows the amount of the engineering


tolerance used up just from the measurement system.

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Minitab Gage R&R % Contribution 65

% Comparisons Table Variation due to the gage


+ Variation due to the parts
= 100%

Source %Contribution %Study Var %Tolerance


%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

Total Gage R&R 10.67


Repeatability 3.10 3.10 + 7.56 = 10.67
Reproducibility 7.56
% Contribution
Oper 2.19 Now, we know that
Acceptable Ranges Oper*Part/Truck 5.37 Reproducibility is our bigger
Part-To-Part 89.33 issue (contributing more than
Not Acceptable
8.0% Total Variation 100.00 Repeatability).
Questionable
2.0%
2.19 + 5.37 = 7.56
Acceptable
Now, we know that of the Reproducibility, the Oper to Part/Truck
interaction is our bigger issue than the Reproducibility of
Operators. Check with the Oper*Part/Truck Interaction Graph to
find out which parts may be the issue.

% Contribution = a good indicator for Present Status „ Note: If you don’t have a tolerance, then, %
Contribution becomes a very critical measurement for
„ The measurement system (gage) is contributing to
Gage R&R. It gives you a good indication of the
10.67% of the total variation and the parts are
measurement variation, now, compared to a realistic
contributing 89.33%.
estimate of current process spread.
„ This Gage Fails the Criteria for % Contribution, but it
is close to passing.
If you wanted to calculate %R&R Contribution by hand:

 σ 2

% R&R  R & R

=
( M e a s . S y s t .)
  × 1 0 0 %
Contribution  σ 2
T o t a l 
 

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Minitab Gage R&R Discrimination Index 66

Discrimination Index
Number of Distinct Categories = 4
Discrimination
DiscriminationIndex
Index
For this gage: the Discrimination Index = 4
• This says that the gage has reached the minimum acceptable
level.

Discrimination Index:
Discrimination Index:
•Provides the number of divisions that the Measurement System can accurately
•Provides the number of divisions that the Measurement System can accurately
measure
measureacross
acrossthe
theprocess
processvariation.
variation.
“Number
“Numberof
ofDistinct
Distinct •Indicates
•Indicateshow
how well
wellaagage
gagecan
candetect
detectpart-to-part
part-to-partvariation
variation---- process
processshifts
shiftsand
and
Categories”
Categories” improvement.
improvement.

Discrimination
DiscriminationIndex
IndexAcceptable
AcceptableValues
Values
•Less than 2, inadequate
•Equal to 2, equal to a go/nogo gauge.
•Minimum acceptable value = 4.
•Optimal = 10 or more.
If you wanted to calculate
Discrimination Index by hand:
 
 σ2 
Std Dev (Parts) D is c rim = 2 •  2 T o ta l  −1
x 1.41  σ R &R 
Std Dev (GRR)  ( M e a s .S y s t . ) 

Discrimination Index will tell us how many categories or


“buckets” the measurement device is able to see or
discriminate. If you’ve chosen parts that are different
(eg, in size, length, width, etc), the Discrimination Index
will indicate how many distinct categories are present. In
our example, Minitab is able to see 4 distinct categories.
Note: If you’ve designed and ran your study with similar
parts (on purpose), this reading would be meaningless to
you.

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Minitab Gage R&R Key Indices 67

33Key
KeyIndices
Indicesfor
forMeasurement
MeasurementSystem
SystemCapability
Capability

11 22 33

P/T Discrimination
DiscriminationIndex
P/TRatio
Ratio %R&R
%R&RContribution
Contribution Index
(Precision-to-Tolerance
(Precision-to-ToleranceRatio)
Ratio)

“R&R
“R&Rasasaa%
%ofof “R&R
“R&Ras
asaa%%of
ofTotal
Total “Number
“Numberof
ofDistinct
Distinct
Tolerance”
Tolerance” Measure Variation”
Measure Variation” Categories”
Categories”

%Tolerance GRR = 171.53% % R&R Contribution = 10.67% Discrimination Index = 4

The next question to be answered: What do I do now „ We also know that Operators and the
that the % Tolerance & % Contribution has failed? Operator*Part/Truck Interaction is significant
(from p-value). This is an indication that the
Well, this is what we have learned so far about the break- operators are measuring differently. It also tells me
down of % Tolerance and % Contribution: there is some variation when certain operators
„ For % Tolerance and % Contribution, we know the measure certain trucks. Again, the graphs indicate
bigger issue is reproducibility. And, with which trucks and which operators.
Reproducibility, the bigger issue is the interaction
between parts and operators. By looking at the I would investigate those areas mentioned above.
graphical summary, it appears that trucks 4 and 10 Depending on your findings, you may need to implement
may be the issue. Investigate why there is a a new training program with your operators on how to
difference. Is there something unique about those measure/record the time. You also may need to do
truck routes? Is it harder to measure the time for something like having the computer measure the time
Trucks 4 & 10 due to time length? Should there be an when we measure parts similar to Parts 4 & 10. Once
Operational Definition or Procedure on how to you’ve done those items, go back and re-run the Gage
measure? Was there a problem with calculating the R&R using your implemented changes.
time from target?

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Gage R&R Acceptable Ranges 68

Acceptable
Acceptable Ranges–ANOVA
Ranges–ANOVA METHOD
METHOD
This tells us the ratio of each
This tells us how much of the component of variation (the
This tells us how much
tolerance is being taken up by standard deviation for each
variation the gage itself is
the variation the gage. component divided by the total
contributing to the variation.
standard deviation).

%Tolerance % Discrimination Index


%Contribution

Not Acceptable
8.0% 30% 4 30%
Questionable
2.0% 10% 10 15%
Acceptable

• Provides the number of divisions that the Measurement System can accurately
measure across the process variation.
• Indicates how well a gage can detect part-to-part variation--process shifts and
improvement.

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

Measurement System Analysis (MSA) Worksheet 69

Type of Gage R&R Conducted (Check One): Date


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

1) Two-Way ANOVA Significant?


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

Graphical Output OK?


1) Effective Resolution [ >50% ] Y N
2) Stability [R Chart] Y N
3) Consistency Between Xbar consistency Y N
Between Oper
4) Systematic Shift [Oper/Part Inter. Plot] Y N
_______________________________________________________________
Gage R&R Pass? Y N, If NO:
Plan for improvement:
ƒ Look at the gage and possibly change to a digital clock.
ƒ Review/Implement a driver call-in procedure. Suspect there is variation between drivers.
ƒ Review/develop a new spreadsheet for calculating the time.
ƒ Talk with operators to determine why Oper 2 measures differently than Operators 1 & 3.
ƒ Talk with customers about the right specification, if possible.
ƒ Train all operators and truck drives on new procedures.
ƒ 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

© GE Capital, Inc., 2000

DMAIC GB M TX PG V 4.2.0
D M A I C

1
Analyze Module Objectives

ƒ Learn and apply the tools and concepts


in the Analyze phase

You will learn the concepts and tools for the Analyze
phase during this classroom session. After the classroom
session, you can use the Six Sigma training CD to
review how the Analyze phase is applied.

© GE Capital, Inc., 2000

DMAIC GB N TX PG V 4.2.0
D M A I C

2
Analyze Phase Flowchart

D M A I C

ANALYZE
PHASE
OVERVIEW

Analyze 4: Analyze 5: Analyze 6:


Establish Define Identify
Process Performance Variation
Capability Objectives Sources

© GE Capital, Inc., 2000

DMAIC GB N TX PG V 4.2.0
D M A I C

3
The12-Step Process

Step Description Focus Tools Deliverables


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

© GE Capital, Inc., 2000

DMAIC GB N TX PG V 4.2.0
D M A I C

4
Analyze Phase Overview

What is the Analyze phase?


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

Why is the Analyze phase important?


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

Steps involved in the Analyze phase


Analyze 4: Establish process capability
Analyze 5: Define performance objectives
Analyze 6: Identify variation sources

© GE Capital, Inc., 2000

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
ƒ 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
ƒ Data Integrity the mean and – Risk
– MSA ƒ Capability variability analysis
– Brainstorm – ZBench ST & LT – Test – Control
potential X’s – Model plans
– Sampling plan – Estimate

ƒ Collect data

The Practical-To-Statistical-To-Practical
Transformation Process

© GE Capital, Inc., 2000

DMAIC GB N TX PG V 4.2.0
D M A I C

© GE Capital, Inc., 2000

DMAIC GB N TX PG V 4.2.0
D M A I C

Analyze 4–Establish Process Capability 1

What does it mean to Establish Process Capability?


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

Why is it important to Establish Process Capability?


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

What are the project tasks for completing Analyze 4?


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

ANALYZE STEP
OVERVIEW

Analyze 4: Analyze 5: Analyze 6:


Establish Define Identify
Process Performance Variation
Capability Objectives Sources

4.1 Graphically analyze data for project Y (continuous data only)


4.2 Calculate baseline sigma for project Y

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0
D M A I C

Step 4.1: Graphically Analyze Data For


Project Y

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0
D M A I C

Review: Describing Variation 3

Prior to Calculating Capability, we need to know:


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

These attributes of your data will help determine which


statistical calculations are better descriptors of your data
set.

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0
D M A I C

Graphically Analyze Data 4

We can use the following example:


Data set: Holdtime.MPJ
1. First, look at Stability:
STAT > Quality Tools > Run Chart
- Approx p-value for clustering = 0.1139
- Approx p-value for mixtures = 0.8861
- Approx p-value for trends = 0.2883
- Approx p-value for oscillator = 0.7117
ƒ 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
ƒ p ≥ 0.05 indicates normal data. Therefore, will use mean and standard deviation to
describe data.

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0
D M A I C

Steps 4.2: Calculate Baseline Sigma

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0
D M A I C

What Is Process Capability? 6

A Measurement Scale Which Compares


the Output of a Process to
the Performance Standard

Process capability allows us to compare how “capable”


the process is of meeting customer CTQ’s by looking at
the probability that the process will produce a defect.

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0
D M A I C

Common Metric For Comparison 7

Process Performance
Purchase Order 98% accuracy
Generation
Accounts Receivable 33 days average aging
Customer Service 82% rated 4 or 5 on responsiveness
Supplier Delivery 95% on-time delivery

Which process is performing best?

It’s difficult to compare processes with very different


kinds of measure. By expressing process capability as a
“sigma” value we can compare different types of
processes.

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0
D M A I C

Data Analysis Roundup 8

Process Capability Tools and Terminology


e.g Cycle Time, Length,
e.g Light off Weight…

e.g Light On

Discrete
DiscreteData
Data Continuous
Continuous
Data
Data
Defects per Six Sigma Process
Opportunity Report:
Defects per Million ZLong term
Opportunities
ZShort term = ZBench =
Six Sigma Product reported yield
Report
ZShift

In this step you will learn how to calculate discrete and


continuous data.

A note on terms to describe capability. All the following


are synonyms:

ZLT = ZBenchLT = Sigma CapabilityLT (or, you may see


capability written as σLT or SigmaLT. This is no longer
standard usage to denote capability).

ZST = ZBenchST = Sigma CapabilityST = ZBench


(σ or SigmaST is no longer recommended).

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0
D M A I C

Process Capability Continuous Data 9

ƒ Verify we have a normal distribution


ƒ Calculate ZLSL and/or ZUSL
ƒ Determine probability of a defective
ƒ Determine ZBench

We will look at how to do each of these steps in detail!

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0
D M A I C

Calculating Z–Continuous Data 10

You can calculate a Z-value for any given value of x. Z is the


number of standard deviations which will fit between the mean and
the value of x. This is known as a Z-score.

X− µ
Z=
σ

µ 1σ 2σ 3σ 4σ

The Z-score is a way to transform any normal distribution


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

You must have a normal distribution: A stable process


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

© GE Capital, Inc., 2000

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

ZUSL = USL - X = 8.7 - 8.5 = 0.2 = 2


s 0.1 0.1

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


s 0.1 0.1

ZUSL = Look up Z = 2.0 => 0.0228

LSL = Look up Z = 3 => 0.00135

Area Total = 0.0228 + 0.00135 = .02415

Yield = 1-.02415 = 0.97585 = 97.6%

Go to Memory Jogger (or Abridged Sigma Table) and


look up a Yield of 97.6%

ZST = 3.4

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0
D M A I C

Reading The Z Table 12

Yield

Probability
Of A Defective
Example = .028066

Specification
Limit

Z-score
Units of Measure
0.20 .420740315 1.71 .043632958 3.22 .000640954 4.73 .000001153
0.25 .401293634 1.76 .039203955 3.27 .000537758 4.78 .000000903
.382088486 1.81 .035147973 3.32 .000450127 4.83 .000000705
1.86 .031442864 3.37 .000375899 4.88 .000000550
1.91 .028066724 3.42 .000313179 4.93 .000000428
1.96 .024998022 3.47 .000260317 4.98 .000000332
2.01 .022215724 3.52 .000215873 5.03 .000000258
2.06
2.11 .019699396 3.57 .000178601 5.08 .000000199
2.16 .017429293 3.62 .000147419 5.13 .000000154

To the
2.21 .015386434 3.67 .000121399 5.18 .000000118
2.26 .013552660 3.72 .000099739 5.23 .000000091
2.31
2.36 .011910681 3.77 .000081753 5.28 .000000070

right of the
.010444106 3.82 .000066855 5.33 .000000053
.009137469 3.87 .000054545 5.38 .000000041
2.41 .007976235 3.92 .000044399 5.43 .000000031
2.46 .006946800 3.97 .000036057 5.48 .000000024
Z-score Z = 1.91
2.51
2.56
2.61
.006036485
.005233515
.004527002
4.02
4.07
4.12
.000029215
.000023617
.000019047
5.53
5.58
5.63
.000000018
.000000014
.000000010

you will 2.66


2.71
2.76
.003906912
.003364033
.002889938
4.17
4.22
4.27 Table Of
note the 1.35
1.40
.088507862
.080756531
2.81
2.86
2.91
.002476947
.002118083
.001807032
4.32
4.37
4.42 Area Under
1.45 .073529141 2.96 .001538097 4.47
tail area 1.50 .066807100 3.01 .001306156 4.52
The Normal
Curve

Yield = Probability of a non-defective

From the Z Table, we can get the probability of a


defective. 1-Probability of defective = Yield.

This is what we’ll need to eventually calculate Process


Capability.

© GE Capital, Inc., 2000

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

© GE Capital, Inc., 2000

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

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0
D M A I C

Abridged Process Sigma Conversion Table 15

ST Process Defects Per Defects Per Defects Per Defects Per Defects Per
Long-Term Yield
Sigma 1,000,000 100,000 10,000 1,000 100
99.99966% 6.0 3.4 0.34 0.034 0.0034 0.00034
99.9995% 5.9 5 0.5 0.05 0.005 0.0005
99.9992% 5.8 8 0.8 0.08 0.008 0.0008
99.9990% 5.7 10 1 0.1 0.01 0.001
99.9980% 5.6 20 2 0.2 0.02 0.002
99.9970% 5.5 30 3 0.3 0.03 0.003
99.9960% 5.4 40 4 0.4 0.04 0.004
99.9930% 5.3 70 7 0.7 0.07 0.007
99.9900% 5.2 100 10 1.0 0.1 0.01
99.9850% 5.1 150 15 1.5 0.15 0.015
99.9770% 5.0 230 23 2.3 0.23 0.023
99.9670% 4.9 330 33 3.3 0.33 0.033
99.9520% 4.8 480 48 4.8 0.48 0.048
99.9320% 4.7 680 68 6.8 0.68 0.068
99.9040% 4.6 960 96 9.6 0.96 0.096
99.8650% 4.5 1,350 135 13.5 1.35 0.135
99.8140% 4.4 1,860 186 18.6 1.86 0.186
99.7450% 4.3 2,550 255 25.5 2.55 0.255
99.6540% 4.2 3,460 346 34.6 3.46 0.346
99.5340% 4.1 4,660 466 46.6 4.66 0.466
99.3790% 4.0 6,210 621 62.1 6.21 0.621
99.1810% 3.9 8,190 819 81.9 8.19 0.819
98.930% 3.8 10,700 1,070 107 10.7 1.07
98.610% 3.7 13,900 1,390 139 13.9 1.39
98.220% 3.6 17,800 1,780 178 17.8 1.78
97.730% 3.5 22,700 2,270 227 22.7 2.27
97.130% 3.4 28,700 2,870 287 28.7 2.87
96.410% 3.3 35,900 3,590 359 35.9 3.59
95.540% 3.2 44,600 4,460 446 44.6 4.46
94.520% 3.1 54,800 5,480 548 54.8 5.48
93.320% 3.0 66,800 6,680 668 66.8 6.68
91.920% 2.9 80,800 8,080 808 80.8 8.08
90.320% 2.8 96,800 9,680 968 96.8 9.68
88.50% 2.7 115,000 11,500 1,150 115 11.5
86.50% 2.6 135,000 13,500 1,350 135 13.5
84.20% 2.5 158,000 15,800 1,580 158 15.8
81.60% 2.4 184,000 18,400 1,840 184 18.4
78.80% 2.3 212,000 21,200 2,120 212 21.2
75.80% 2.2 242,000 24,200 2,420 242 24.2
72.60% 2.1 274,000 27,400 2,740 274 27.4
69.20% 2.0 308,000 30,800 3,080 308 30.8
65.60% 1.9 344,000 34,400 3,440 344 34.4
61.80% 1.8 382,000 38,200 3,820 382 38.2
58.00% 1.7 420,000 42,000 4,200 420 42
54.00% 1.6 460,000 46,000 4,600 460 46
50% 1.5 500,000 50,000 5,000 500 50
46% 1.4 540,000 54,000 5,400 540 54
43% 1.3 570,000 57,000 5,700 570 57
39% 1.2 610,000 61,000 6,100 610 61
35% 1.1 650,000 65,000 6,500 650 65
31% 1.0 690,000 69,000 6,900 690 69
28% 0.9 720,000 72,000 7,200 720 72
25% 0.8 750,000 75,000 7,500 750 75
22% 0.7 780,000 78,000 7,800 780 78
19% 0.6 810,000 81,000 8,100 810 81
16% 0.5 840,000 84,000 8,400 840 84
14% 0.4 860,000 86,000 8,600 860 86
12% 0.3 880,000 88,000 8,800 880 88
10% 0.2 900,000 90,000 9,000 900 90
8% 0.1 920,000 92,000 9,200 920 92
Note: Subtract
1.5 to get long-
term
Sigma level

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0
D M A I C

Long-Term vs. Short-Term Data 16

Long-Term Data

Y
(Continuous)

Short-Term
Data

Time

How much variation we observe in a process is For now, and as a default, treat all data as long-term data.
influenced by whether we are looking at long-term data The convention for Six Sigma is to report short-term, as
or short-term data. you will see later.

Long-Term: Why do we report Short-Term Sigma?

Data collected over a long enough period of time and We want to standardize reporting of the best a process
over diverse enough conditions such that it is likely to can do, given its current capability. Therefore,
contain some process shifts and other special causes. Short-Term Sigma tells us this.

Short-Term:

Data collected over a short enough period of time so that


shifts and other special causes are unlikely

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0
D M A I C

Reporting Sigma Values 17

Short-Term Sigma = Long-Term Sigma + Sigma Shift


ƒ If “Shift” is unknown, then assume 1.5
ƒ Assume that sigma calculated from project data is long-term sigma
ƒ A rational subgrouping sampling scheme for data collection (in the
Measurement Phase) must have been used if you are calculating a shift
(other than using 1.5.)

It is possible to calculate short-term, but for most Green „ Special Cause = Between subgroup variation due to
Belt projects we do not attempt to do so; instead we use assignable causes, non-random influences.
the 1.5 sigma shift rule.
„ Common Cause = Within subgroup variation,
In order to truly calculate short-term sigma, you have to inherent in a process, random influences.
use several rational subgroups. If you did not use a
rational sub-grouping sampling scheme for data To ensure you are sampling* properly, these 3 items must
collection in the Measure phase, it will not be possible to be true:
calculate actual short-term sigma. (Therefore, add 1.5 to – Representative Data
long-term sigma.)
– Enough Data (at least 20 subgroups)
– Measured with something that passes the Gage R&R

*Refer to the workbook for further details on sampling


(Tools > Measure phase)

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0
D M A I C

Principles Of Rational Subgrouping 18

1. Never knowingly subgroup unlike things together


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

Rational subgrouping refers to grouping the data for


analysis in a meaningful way to understand variation.
Rational subgrouping attempts to select groups of data
such that mainly common cause variation is within
groups, and mainly special cause variation is between
groups.

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0
D M A I C

Generalizing The Correction 19

Six Sigma Centered


Process
Capability

.0005 ppm .0005 ppm


T
T
LSL
LSL ± 6σ USL
USL
USL

Six Sigma Shifted 1.5σ


µµ

3.4 ppm
TT

LSL USL
4.5σ

The 1.5 shift is used as a compensatory off-set in the


mean to generally account for dynamic non-random
variations in process centering. It represents the average
amount of change in a typical process over many cycles
of that process.

If the process is centered, then Six Sigma capability


means that 6 standard deviations fit between the mean
and the specification limit.

If the process center “shifts” by 1.5 sigma , then the


average is 4.5 sigma from either the LSL or USL.

Six Sigma Short-Term Capability = 4.5 Sigma Long-


Term Capability

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


Term Capability

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0
D M A I C

The Universal Equation For Z 20

. . . so what are the possibilities? SL - λ


Z=
σ

SL =USL
LSL λ= T (Target)
µ (Mean)

Z =
st (short-term)
lt (long-term) σst
σlt

and how do we choose the right one?

SL − µ The larger the ZShift , the greater the control problem.


Z LT =
σ LT Typical ZShift = 1.5

Long-term is what is actually going on in the process.


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

SL − T
ZST =
σ ST

Short -term is the best the process can do.


SL - T is the distance the specification is from the target.
ZST is bigger than
ZShift = ZST − Z LT

© GE Capital, Inc., 2000

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

P(d)USL = from P(d)LSL = from P(d)USL = from P(d)LSL = from


Z table Z table Z table Z table

P(d)Total = P(d)USL + P(d)LSL P(d)Total = P(d)USL + P(d)LSL

ZB-lt = from ZB-st = from


Z table Z table

In Six Sigma, it is common to report out process


capability using ZBench. As shown previously, ZBench
corresponds to yield, i.e., we look at the chance of being
outside the specification limit on either side. ZBench is
calculated for both short and long-term. The difference
between the two is ZShift.

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0
D M A I C

Activity–Calculating Process Capability–Continuous Data 22

What is the process capability for a process that has:

Mean = 5
Standard Deviation = 2
Upper Spec. Limit = 9

You can use the hand calculation or you can open the
spreadsheet: SIGMACAL.xls.

Enter:

Xbar: 5

S: 2

USL: 9

LSL: (leave blank)

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

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0
D M A I C

Graphically Analyze Data–Breakout Activity (20 minutes) 23

Desired Outcome: Graphical Analysis of Capital Logistics Project Y Data


Minitab File: GB Case Study.mtw

What How Who Timing

Run ƒ Use the Run Chart tool in Minitab to All 5 mins.


Chart investigate the variation in the
project Y data over time

Shape, ƒ Use the Normal Probability Plot in All 10 mins.


Normality, Minitab to analyze the shape of the
Central project Y data
Tendency And
ƒ Use the Descriptive Statistics tool in
Spread
Minitab to analyze the shape,
normality, central tendency and
spread of the project Y data
ƒ Use the Minitab Six Sigma Process
Report to calculate Process Sigma

Solutions ƒ You can check your answers using All 5 mins.


the solution sheets on the following
pages
We always analyze the data this way:
1. Look at Stability–Is the process Stable?
2. Look at Shape–Do I have a normal distribution?
3. Look at the Spread–What measure of dispersion should I use?
Recall from our Case Study:
Time *(minutes early or late) = (Target Arrival Time) – (Actual Arrival Time)
Date 1 = Date of Shipment
* Spec. for time = ± 60 min.

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0
D M A I C

Solution–Run Chart 24

Stability Calculations for Case study data.

First Step, Check Stability

Stat > Quality Tools > Run Chart

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0
D M A I C

Solution–Run Chart 25

1. Double click on
“C5”

2. Double click on “C1”


Each Date is a sub-
group

3. Click ”OK”

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0
D M A I C

Solution–Run Chart 26

Each Date is a column of


data points.
Run Chart for time

100
time

0 Median

Each column has


it’s mean plotted as
-100 a red square

2 7 12 17
Subgroup Number
Number of runs about median: 10.0000 Number of runs up or dow n: 9.0000
Expected number of runs: 9.4706 Expected number of runs: 11.0000
Longest run about median: 4.0000 Longest run up or dow n: 4.0000
Approx P-Value for Clustering: 0.6050 Approx P-Value for Trends: 0.1118
Spread
Approx P-Value for Mixtures: 0.3950 Approx P-Value for Oscillation: 0.8882

If these numbers had a value


< .05, then the process is unstable.

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

By using the Run Chart, I’m looking for stability.

How do I analyze this chart?


„ Look at p-values. P > .05 indicates stability. If the
process is unstable, look for the most recent stable
period and analyze with Minitab. If there is not a most
recent stable time period, investigate & remove special
causes.

Note: To manipulate data in Minitab:

MANIP > Subset worksheet > include/exclude row #’s

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0
D M A I C

Solution–Normality Testing 27

In Minitab, you can perform a Normality Test (Stat-Basic Statistics-Normality Test).


Use the normality test to validate the shape of your data, when the shape is in question.

Normal Probability Plot

.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

The data is Normal by this test


method if the P-value is > = .05

Second Step: Check Shape (Normality)

Stat > Basic Statistics > Normality Test (Variable =


Time)

The results of the Anderson-Darling Normality Test tell


you if the data is normal or not. A p-value of greater than
or equal to 0.05 means the data is normal. A p-value of
less than 0.05 means that the data is not normal. Step 6
will address p-values in more detail.

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0
D M A I C

Solution–Normality Testing (continued) 28

Descriptive Statistics
Variable: time This
data is
normal
Anderson-Darling Normality Test
.
A-Squared: 0.130
P-Value: 0.983

Mean -4.4032
StDev 43.1714
Variance 1863.77
Skewness 7.65E-02
Kurtosis -1.2E-01
-100 -60 -20 20 60 100 N 506

Minimum -114.021
1st Quartile -35.085
Median -4.956
3rd Quartile 23.421
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

Third Step: Check Spread

The graphical summary of descriptive statistics (Stat–


Basic Statistics–Descriptive.)

Stat > Basic Statistics > Display Descriptive Statistics

Variable = Time
Click on “Graphs” & choose “Graphical Summary”

Test p-value. Minitab will always display a normal


curve based on the data’s mean and standard deviation
–this may not be a fit to the actual distribution.

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0
D M A I C

Minitab Six Sigma Process Report 29

MINITAB FILE: GB case study.mtw

1. Double
Click C5 time

2. Double
click C1 to
use Date1
as the
Subgroup
size
3. Type in
Lower and
Upper
specs
4. Click OK

Now let’s look at how to calculate process capability for The Six Sigma Process Report is used to calculate the
continuous data in Minitab, using the Six Sigma Process long-term and short-term z-values of your process. In
Report. order to be accurate in the calculation of the z-values, it is
important that we enter all of the information available.
The subgroup size can be entered in two different ways: If a target is available, then enter the target value for the
„ Size of each subgroup –10 (subgroup size must be calculation of the z-values. If the upper and lower
constant) specification limits are entered and the target is left
blank, Minitab approximates the target value as the
„ ID column–oper 50 (subgroup size may be constant or midpoint of the specific range. If only one specification
not) limit is entered and the target is left blank, then Minitab
approximates the target as the mean of the data. Hence,
Warning: If you choose to use a subgroup size of 1, the when the target is available, always input the given value
short-term values calculated in the Six Sigma Process to avoid approximations made by Minitab.
Report will be invalid since there is not variation within
the groups.

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0
D M A I C

Minitab Six Sigma Process Report 30

Report 1: Executive Summary


Process Performance Process Demographics
Actual (LT)
Potential (ST) Date:
Reported by:
Project:
LSL USL
Department:
Process:
Characteristic:
Units:

-100 0 100 Upper Spec: 60


Lower Spec: -60
1,000,000 Nominal:
Actual (LT)
Potential (ST)
Opportunity:
100,000

10,000
Process Benchmarks
1000
Actual (LT) Potential (ST)

100
Sigma 0.97 0.98
(Z.Bench)
10
PPM 166781 164350
1

0 5 10 15

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

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0
D M A I C

Minitab Six Sigma Process Report 31

Report 2: Process Capability for time

Xbar and S Chart Capability Indices


40
30 ST LT
1. Between 20 3.0SL=21.49
10
Subgroup 0 Mean 0.0000 -4.4032
X=-4.403
Variation -10
-20 StDev 43.1471 43.1714
-30 -3.0SL=-30.29
-40 Z.USL 1.3906 1.4918
-50
Z.LSL 1.3906 1.2878
Subgroup 0 5 10 15
80 Z.Bench 0.9767 0.9670
70 Z.Shift 0.0098 0.0098
2. Within 60 3.0SL=61.28
Subgroup 50 P.USL 0.082175 0.067876
Variation 40 S=42.70
P.LSL 0.082175 0.098905
30
20 -3.0SL=24.12 P.Total 0.164350 0.166781
10
Yield 83.5650 83.3219
3. Because variation is
not out of control, there is PPM 164350 166781
no LT Variation
Potential (ST) Capability Actual (LT) Capability Cp 0.46
Process Tolerance Process Tolerance Cpk 0.43
-129.508 129.508 -133.981 125.175 Pp 0.46
I I I I I I Ppk 0.43
4. This makes sense, we
only have 1 month’s worth I I I I I I
of data -60 60 -60 60 Data Source:
Specifications Specifications Time Span:
Data Trace:

Report 2 is more detailed: Note: When reporting ZST, determine if you will be using
„ The table to the right side contains all the details of the the shift calculated from your data or the standard 1.5
shift. If using the 1.5 shift, you will need to manually add
calculation of ZBench for short and long-term.
the 1.5 to ZLT to get ZST. Minitab cannot accommodate a
„ The graphs to the left side are called control charts. predetermined shift. (When in doubt, use the 1.5 shift
This subject will be covered in the “control” session of and/or consult your BB/MBB).
your training. In order to use Minitab’s calculated shift
(right column in the table), the “S” chart must be in
control (minimize variation within subgroup)
„ In this example, S chart is out-of-control (one point
outside control limits). It does not meet the condition
mentioned in the 2nd bullet, therefore, report ZLT =
0.9670. To get ZST: ZST = ZLT + ZShift
ZST = 0.9670 + 1.5 = 2.467
„ Refer to the next 2 pages for rules on using this
Process Capability Report.

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0
D M A I C

Rules For Using The Six Sigma Process Report 32

Rules for Process Capability


1. Ensure the process is stable (Run Chart). If not, remove
special causes first.
2. Ensure normal distribution (if not, use DPMO method to
calculate capability)
3. For continuous data: use Six Sigma Process Report. If
subgroup size > 1 and the S-chart is in-control, then both
columns in the table are valid. If the S-chart is out-of control,
only the LT column is valid. Add 1.5 to the ZBenchLT to get ZST.
4. If the subgroup size = 1, use LT column only in the Process
Report

Use the Decision Flow on the next page to help you!

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0
D M A I C

Process Capability Decision Flow 33

Is your
Process Yes Do you Yes Is the Yes Use Six Sigma> Process Report
Stable (use have Distribution (Input Data, Specification Limits
Run Chart)? Continuous Normal? and Target as appropriate)
Data?

No No No
Do you
Analyze for Special Six Sigma>Product Report, No Use Z Bench-LT only
have
Causes- DPMO, L1 (Input D, U and O) and add 1.5 Shift
Rational
to get Z Bench-ST
Should not •Sigma Conversion Table gives ZST Subgroups?
calculate process
•Z LT = Z ST - Z Shift (1.5)
capability until the
process is stable. •Product Report and L1 give Z Bench Yes
Unstable process •Z Bench = Z ST
data can only be
•Z LT = Z Bench - Z Shift (1.5) Is the S- No Use Z Bench-LT only
used to describe and add 1.5 Shift
Chart in- to get Z Bench-ST
the capability of the
control?
current data set, not
the future capability
of the process. Yes

Note: Z Bench-LT , Z Bench- ST


Z ST = Z LT + Zshift & Z Shift all valid on
Z ST = Z Bench-ST the Process
Z LT = Z Bench-LT Report

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0
D M A I C

The Normal Curve And Capability 34

Poor Design Capability

High
Probability High
of Defective Probability of
Defective
Good Design
Capability

LSL USL

Low Low
Probability Probability
of Defective of Defective

LSL USL

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0
D M A I C

Summary–Z-Value 35

ƒ Basic statistical summaries, histograms, dotplots, boxplots, and


run charts are used to visualize data and better understand a
process
ƒ The Z-score is a non-dimensional quantity that enables us to
compare different processes–it represents the process
capability
ƒ The Z-score is the number of standard deviations that will fit
between the mean and the respective specification limit of a
normal distribution
ƒ The Z-score corresponds to yield, or the area under the curve
inside the specification limits

Note on Capability:
„ Capability can be described as any of the following:
– ZBenchST = ZST = σST = Sigma CapabilityST =
SigmaST = ZBench
– ZBenchLT = ZLT = σLT = Sigma CapabilityLT =
SigmaLT

„ In the new BOK, σST and σLT are not recommended


ways to describe capability. σST = pooled standard
deviation and σLT = standard deviation.

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0
D M A I C

Summary–Continuous Data Process Capability 36

ƒ The Minitab Six Sigma Process Report is used to describe capability


with continuous data
– Displays the actual capability relative to the target distribution
– By rationally subgrouping (subgroup size >1), long-term capability, short-term
capability, and shift are calculated.
ƒ 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
ƒ A subgroup that contains only common cause variation, or random
variation, represents the short-term capability of the process or
entitlement (ZBench)

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0
D M A I C

Process Capability & Yield–Discrete Data 37

ƒ Calculate the distribution of defects, defects per million


opportunity (DPMO)
ƒ Utilize Z tables to convert DPMO to yield or sigma level
ƒ Understand the difference between classical, Throughput Yield,
and Rolled Throughput Yield
ƒ Calculate submitted, observed, and escaping defect levels

Classical Yield (YC) = Final Yield

Throughput Yield (YTP) = First Pass Yield

Rolled Throughput Yield (YRT)

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0
D M A I C

Definitions 38

Unit (U)
ƒ The number of parts, sub-assemblies, assemblies, or systems
inspected or tested
– Squares: 4 units
Opportunity (OP)
ƒ A characteristic you inspect or test
– Circles: 5 opportunities per unit
Defect (D)
ƒ Anything that results in customer dissatisfaction. Anything that
results in a non-conformance.
– Black circles: 9 defects

Note: Black dots indicate a defect.

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

Defective–A unit with 1 or more defects. (defective=any


square with 1 or more black dots).

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0
D M A I C

Formulas 39

Defects per Unit


DPU = D/U
9/4 = 2.25

Total Opportunities
TOP = U*OP
4*5 = 20

Defects per Opportunity (Probability of a Defect)


DPO = D/TOP
9/20 = .45

Defects per Million Opportunities


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

Once we calculate DPMO, we go to the Abridged Sigma


Table (next page) to get ZBenchST or ZBenchLT

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0
D M A I C

ConvertingDPMO
Converting DPMOTo
ToZZST 40

Long term Actual Reported

ZST = Reported D PMO


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

Sigma Conversion 241,964


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

To Calculate Process Sigma:


„ Go to the Memory Jogger–Abridged Sigma
Conversion Table (or use the table above)
„ Look up the DPMO value and report Short Term
Sigma value (always round to poorer performance if
between values in the table.)

In our example, we calculated 450,000 DPMO. This


value is not found in the table, therefore, we go to
460,000 and report 1.6 as the Short Term Sigma Value.
We’re always conservative. Therefore, report the lower
value for ZST or ZLT. And, you cannot interpolate, so
choose the lower sigma value.

© GE Capital, Inc., 2000

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

Prepare Review Fix Mail


Invoice Invoice Errors Invoice
500
Preliminary 470 Invoices
Invoices Mailed On-Time
446 Accurate Invoices

Customer CTQs
„ Invoice mailed on date specified
„ Invoice is error free
– Correct address
– Correct amount

How would the customer measure the process


performance?

First, determine the unit, defect opportunities per unit,


and the number of defects from each perspective.

For this example, assume the process of reviewing and


correcting invoices is 100% effective.

Use the worksheets on the following pages.

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0
D M A I C

42
Calculating Process Sigma Discrete Data Method

1. Number of Units processed 500


U = __________

2. Number of Defect Opportunities Per Unit OP = __________


3

3. Total number of Defects made 72


D = __________
(include defects made and later fixed)

4. Solve for Defects Per Opportunity


D ( 72 )
DPO = U∗OP = ( 500 ) ( 3 ) = 0.048

5. Convert DPO to DPMO

DPMO = DPO ∗ 1,000,000 = 0.048 ∗ 1,000,000 = 48,000

6. Look up Process Sigma in


Abridged Process Sigma Conversion Table ZST = 3.1

The unit in this case is an invoice. Hence, there are 500


Now calculate the DPMO and determine the
units for this example.
capability.

Based on the CTQ’s listed for this example, there are 3


defect opportunities per unit–late, wrong amount, and
wrong address.

Therefore, the total number of defects is 30 late, 28


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

Even though the errors on the accounting code were


detected and corrected, they are not counted for this
process. The CTQ’s stated here are for our external
customers. They do not care if our accounting codes are
correct; therefore, this “error” is not one of their CTQ’s.
If we were to examine this from the perspective of our
internal customer, the list of CTQ’s would be different.
.
© 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

1. Number of Units processed U = __________


500

2. Number of Defect Opportunities Per Unit 3


OP = __________

3. Total number of Defects made 30


D = __________

4. Solve for Defects Per Opportunity

D ( )
DPO = = =
U∗OP ( )( )

5. Convert DPO to DPMO


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

6. Look up Process Sigma in


Abridged Process Sigma Conversion Table ZST =

Let’s verify our answer using Minitab’s Product Report.

1. First open a blank worksheet and enter 3 column


headings and your data:

Defects Units Opportunities


30 500 3

2. Six Sigma > Product Report


Defects = Defects
Units = Units
Opportunities = Opportunities

3. Click “OK”

4. Verify ZBench = ZST (from your calculations)

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0
D M A I C

44
Activity–Calculating Process Capability–Discrete Data

ƒ Project: Billing Product Improvement


ƒ CTQ: Complete Billing Information
ƒ Defect: Information missing from bill
ƒ Unit: A bill
ƒ Opportunities for defect per unit: 3 (items of information that
can be missing on each bill)
ƒ Out of 50 bills, you find 20 defects
ƒ What is the sigma level of the process?

What is the DPMO value for this process?

Now let’s look at how to calculate process capability for


discrete data in Minitab, using the Six Sigma Product
Report.

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0
D M A I C

45
Six Sigma Product Report

MINITAB FILE: GB case study.mtw

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0
D M A I C

46
Discrete Data Examples

1. Select Defects, Units 2, & Opportunities

2. Select OK.

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0
D M A I C

47
Six Sigma Product Report Output

Report 7: Product Performance

Characteristic Defs Units Opps TotOpps DPU DPO PPM ZShift ZBench

U DPMO ZST
D

1 89 506 1 506 0.176 0.175889 175889 1.500 2.431

OP

Total 89 506 0.175889 175889 1.500 2.431

In addition to the Product Performance table, Minitab


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

To verify our answer:


D 89
ƒ DPO = = = 0.175889
U ∗ OP (506 )(1)
ƒ DPMO = DPO ∗ 1,000,000 = 175,889

ƒ Go to Abridged Process Sigma Conversion Table,


Z = 2.4
ST

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0
D M A I C

48
Yield

Four parts are made


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

2. After second inspection: 1 passed, 2 failed


Rework 2 parts

3. After third inspection: 1 passed, 1 scrapped

What is the Yield of this Process?

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0
D M A I C

49
Types Of Yield

Classical Yield = YC = 3/4 = 75%


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

First-Time Yield = YFT = 1/4 = 25%


First-time yield is the percent of defect-free parts divided by the total number of
parts inspected for the first time. If we say the yield is 1/4 or 25%, we are really
talking about the First-Time Yield (FTY). This is the best yield estimate to drive
improvement. This is also Throughput Yield (YTP): the percent of units that
pass through an operation without any defects.

Classical Yield:
YC = ¾
# parts making it through final inspections = 3
Total # Parts = 4

First-Time Yield
YFT = ¼
# Parts that are acceptable after first inspection = 1
Total # Parts = 4

Probability of Single Unit Being Defect-Free (SDF)


SDF = P(0) = e-DPU = e-2.25 = .1054 = 10.54%
DPU = Defects per unit = 9/4 = 2.25

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0
D M A I C

Extending The Concept 50

A given process has two operations. Each operation has a


throughput yield of 99 %. The rolled yield equals:
Process Centered Process Centered
Op 1 x Op 2 = Output

99% 99% 98%

Without Inspection or Test Without Inspection or Test Without Inspection or Test

. . . There is an 98% probability that any given unit of product could


pass through both operations defect free.

Rolled Throughput Yield = YRT = 98%

When the defects of the subsequent operations are


independent, rolled throughput yield (YRT) = .99 x .99 =
.9808

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0
D M A I C

Rolled Throughput Yield (YRT) 51

Receive parts from Supplier

99.53% Yield (YTP)


Following Receiving
Inspection and Line Fall-out...

97.13% Yield (YTP)


From Machining Operations

94.52% Yield (YTP)


At Test Stands on
first attempt

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

G. Reimer 12/21/94 - Charlotte NC

You can establish a yield for each step of your process. If


you have multiple processes, each time you go through
additional steps, you lose yield at the end of the process.

This is called Rolled Throughput Yield. YRT is calculated


by multiplying all the yields together.

The more complex the process, the worse the YRT


becomes.

We waste a lot of time scooping up the spilled paint.

© GE Capital, Inc., 2000

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 )

DPUS = DPUO + DPUE


Observed DPU Level
DPUO = DPUS x E (DPUO )
DPUE = DPUS x (1-E)

Let n denote the number of defects going into an Example


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

The number escaping to the customer is (1-E)*n.

If you submit parts with a DPUS level, inspection E = 0.8


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

The customer sees a product with an escaping DPUE =


(1-E)*DPUS.

Submitted DPUS = Observed DPUO + Escaping DPUE. DPUO=80


© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0
D M A I C

Quantification Of Defects (continued) 53

Given:
Observed Defects = DPU of 0.25 and E = .8
Then:
Submitted Defects = DPUS = DPUO /E = (0.25/0.8)
= DPUS = 0.31
Escaping Defects = DPUE = DPUS - DPUO = (0.31 - 0.25)
= DPUE = 0.06

Inspection
E = 0.8
Submitted Escaping
DPUS Level DPUE Level
= 0.31 (calc) = 0.06 (calc)

Observed DPUO Level = 0.25 (given)

The only thing that is real is the DPU observed–that is


what we caught at inspection. In this example DPUO =
.25.

A typical value for E = .9. Your operation will differ


slightly; in this example E = .8.

Review the calculations below:


DPUS*.8 = DPUO

DPUS = DPUO /.8

DPUE = DPUS- DPUO

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0
D M A I C

Exercise (15 minutes) 54

Calculating Submitted, Observed and


Escaping Defect Levels
Given:
ƒ The requirement is to ship the product with no more than one
defect in 500 units shipped
ƒ 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:
ƒ Observed in the final test. (This sets a minimum throughput
yield limit.)
ƒ Submitted to the final test. (This sets a minimum on the
combination of defects created in model assembly and from the
escaping defect levels from prior tests.)

Use the worksheet on the following page to help you


work through this exercise.

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0
D M A I C

Exercise (continued) 55

Calculating Submitted, Observed and Escaping Defect Levels

Inspection
E = 0.95
Submitted Escaping
DPUS Level DPUE Level

Observed DPUO Level


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

2. What then is the acceptable maximum defect level for:


a. Units submitted to final test? ___________
b. Defects observed at final test? __________

3. What is the observed throughput yield figure? ______

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0
D M A I C

Answers To Exercise 56

Calculating Submitted, Observed, and


Escaping Defect Levels

1. DPUE = 1 = 0.002
500
2. Maximum acceptable submitted defects =

DPUS = defects escaping = DPUE = 0.002 = 0.040 DPUS


1 - effectiveness 1 -E 0.05

2b. Maximum acceptable observed defects =

DPUO = DPUS - DPUE = 0.040 - 0.002 = 0.038 DPUO

3. YTP = 1-.038 = .962 = 96.2%

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0
D M A I C

Summary–Discrete Data Process Capability 57

ƒ Define defects, units and opportunities with your team. Be sure


the definitions make sense and are consistent with similar
processes and customer definitions.
ƒ Defects will be stated as defects per million opportunities.
Discrete data is generally considered long-term data.
ƒ For discrete data, Minitab Six Sigma product report is used to
calculate capability from defects and opportunities
ƒ Determine DPMO (which is long-term), then determine the
corresponding Z–value (ST capability)

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0
D M A I C

Summary–Z-Shift 58

ƒ Often, you must assume a shift value (default 1.5) to estimate


short-term capability
ƒ Our customers experience the long-term capability of the
process
ƒ To minimize shift, we need to reduce special cause variation

© GE Capital, Inc., 2000

DMAIC GB O TX PG V 4.2.0
D M A I C

Analyze 5–Define Performance Objectives 1

What does it mean to Define Performance Objectives?


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

Why is it important to Define Performance Objectives?


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

What are the project tasks for completing Analyze 5?


5.1 Identify Performance Objectives.

ANALYZE STEP
OVERVIEW

Analyze 4: Analyze 5: Analyze 6:


Establish Define Identify Variation
Process Capability Performance Sources
Objectives

5.1 Identify Performance Objectives.

© GE Capital, Inc., 2000

DMAIC GB P TX PG V 4.2.0
D M A I C

How Do I Define My Performance Objectives? 2

ƒ Use benchmarking
ƒ Use other sources

© GE Capital, Inc., 2000

DMAIC GB P TX PG V 4.2.0
D M A I C

Defining Performance Objectives With Benchmarking 3

If I benchmark, performance standards are based upon:


ƒ Closing the gap with the competition
ƒ Exceeded projected competitive performance
ƒ Similar performance in dissimilar businesses
ƒ Gathering best practices from multiple sources to become best
in class
ƒ Becoming as good or better than a substitute product/service

© GE Capital, Inc., 2000

DMAIC GB P TX PG V 4.2.0
D M A I C

Defining Performance Objectives With Other Sources 4

If I don’t benchmark, performance objectives are based upon:


ƒ For a process with ZST ≤ 3 Sigma Capability, decrease %
defects by 10x and for ZST > 3, decrease % defects
by 2x
ƒ 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.
ƒ Corporate mandate
ƒ Compliance/legal
ƒ VOC data

© GE Capital, Inc., 2000

DMAIC GB P TX PG V 4.2.0
D M A I C

Nature Of Benchmarking–Reference Material 5

Benchmarking is the process of continually searching for the best


methods, practices and processes, and either adopting or adapting
their good features and implementing them to become the “best of
the best”.

The remaining pages are for your reference on


Benchmarking.

There are different types of Benchmarking. These


include:
ƒ Competitive Benchmarking
ƒ Product Benchmarking
ƒ Process Benchmarking
ƒ Best Practices Benchmarking

© GE Capital, Inc., 2000

DMAIC GB P TX PG V 4.2.0
D M A I C

What It Is And What It Isn’t 6

Benchmarking Is… Benchmarking Isn’t…


ƒ A continuous process ƒ A one-time event
ƒ A process of investigation that ƒ A process of investigation
provides valuable information that provides simple
ƒ A process of learning from answers
others; a pragmatic search for ƒ Copying, imitating
ideas ƒ Quick and easy
ƒ A time-consuming, labor-
intensive process requiring
discipline
ƒ A viable tool that provides
useful information for
improving virtually any
business process

© GE Capital, Inc., 2000

DMAIC GB P TX PG V 4.2.0
D M A I C

Common Benchmarking Mistakes 7

1. Internal process(es) is unexamined


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

© GE Capital, Inc., 2000

DMAIC GB P TX PG V 4.2.0
D M A I C

Benchmarking Six-Step Process 8

Identify Process to Benchmark


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

Select Organization to Benchmark


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

Prepare for the Visit


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

Visit the Organization


„ Feel comfortable with and confident about your homework
„ Foster the right atmosphere to maximize results
„ Conclude in thanking organization and ensure follow-up if necessary

Debrief & Develop an Action Plan


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

Retain and Communicate


„ Report out to business management and 6σ leaders
„ Post findings and/or visit report on local server/6σ bulletin board
„ Enter information on GE Intranet benchmarking project database

© GE Capital, Inc., 2000

DMAIC GB P TX PG V 4.2.0
D M A I C

Sources Of Information 9

Library Database Company Watches


Internal Reviews Newspapers
Internal Publications Advertisements
Professional Associations Newsletters
Industry Publications Original Research
Special Industry Reports Customer Feedback
Functional Trade Publications Supplier Feedback
Seminars Telephone Surveys
Industry Data Firms Inquiry Service
Industry Experts Networks
University Sources Worldwide Web

© GE Capital, Inc., 2000

DMAIC GB P TX PG V 4.2.0
D M A I C

Capital Logistics Case Study 10

Project Goal(s)
ƒ Reduction of deliveries that are too early or too late

Benchmarking Process
ƒ Identified 20 similar 3rd party Logistics providers from Logistician Magazine
ƒ Selected 3PL (2) that provide supply chain solutions to the petroleum industry
(Non-competitors–Capital Logistics strategically avoids supporting petroleum
industry because of risks)
ƒ Polled logistics managers (telephone polls) to determine delivery practices

Results
ƒ 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
ƒ Invited Petro-hauz to Stamford, CT for product demo with Capital Logistics
Engineering and Technology Teams

© GE Capital, Inc., 2000

DMAIC GB P TX PG V 4.2.0
D M A I C

Benchmarking Example 11

ƒ For the Capital Logistics Case Study


– Target is the same = 0 minutes
– Best in Class performance is a s = 12 minutes
– Theoretical calculation of Sigma based on normal distribution would be

Report 7: Product Performance

Characteristic Defs Units Opps TotOpps DPU DPO PPM ZShift ZBench

1 60 25000 1 25000 0.002 0.002400 2400 1.500 4.320

© GE Capital, Inc., 2000

DMAIC GB P TX PG V 4.2.0
D M A I C

Summary–Analyze 5 12

ƒ A performance objective is determined by using Z–short-term,


benchmarking, or defect reduction goals
ƒ 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
ƒ Revenue & cost implications are also due for benefit analysis

© GE Capital, Inc., 2000

DMAIC GB P TX PG V 4.2.0
D M A I C

Analyze 6–Identify Variation Sources 1

What does it mean to Identify Variation Sources?


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

Why is it important to Identify Variation Sources?


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

What are the project tasks for completing Analyze 6?


6.1 Identify possible causes of variation
6.2 Narrow list of potential causes

ANALYZE STEP
OVERVIEW

Analyze 4: Analyze 5: Analyze 6:


Establish Define Identify Variation
Process Capability Performance Sources
Objectives

6.1 Identify possible causes of variation


6.2 Narrow list of potential causes

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

Identify The Vital Few 2

Transfer Function

Y = f (X)

Project Y Relationship Process


that explains Y variables
in terms of X

Understanding The ƒ Gives Insight


Into The Vital Few X’s

The Project Y (or dependent variable) is determined by


the values of the process variables X (or independent
variables). Once we have a clearer understanding of the
relationship between the X and Y (f–transfer function),
we will be able to pinpoint the key X’s that drive the
variation in Y.

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

Identify The Vital Few 3

Process

Input
Measures Outputs
(X’s) (Y’s)

X X X X

Process Measures
(X’s)

Definitions: Process
„ Project Y (Dependent Variable) „ Provides context for improvement
„ Product or service produced or delivered by the „ Links outputs (Y’s) to inputs (X’s) and
process process variables (X’s)
„ Process X’s (Independent Variable)
„ Those variables that influence the output and are In Analyze 6 we will be analyzing upstream
generally controllable by those who operate the variables, or input and process variables (X’s), to
process determine how they affect output variables (Y’s)
and to what extent.
„ Input X’s (Independent Variable)
„ Materials and information used by the process to create Remember Y=f (X1, X2, ..., Xn)
the outputs. Inputs are often outside the control of the
process operator.
or in words….
The results we get (the Y’s) are a function of the
process and input variables (the X’s). In Analyze,
we will study the process and the data in order to
gain understanding of how it all fits together.

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

Analyze 6.1: Identify Possible Causes of Variation

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

Methods To Identify Possible Sources Of Variation 5

Methods To Identify Vital X’s


Process Map Analysis
Graphical
Analysis

Machines Methods Materials

Problem
Statement

Measurement Mother Nature People

As we work our way down into the process seeking the


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

We have already learned about graphical analysis,


process map analysis and cause and effect analysis in
previous modules. In step 6, we will learn how to
apply them in the Analyze phase.

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

Review: Graphical Analysis 6

Looking For Patterns In Data

Continuous Y Discrete Y
Boxplot Pareto Chart
Scatterplot
Histogram

We learned about these tools in the Minitab and


Graphical Analysis module. These graphical analysis
tools are used in the Analyze phase to look for patterns in
the project Y data that help to identify potential critical
X’s.

Let’s do an activity to briefly review these tools.

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

Review Activity–Which Graphical Analysis Tool Would You Use? 7

For each scenario described below, which tool would you use to investigate? For each
scenario, what pattern would you look for to make your conclusions?

1. (a) A Six Sigma project is being conducted in the field to improve the cycle time for
warranty repair returns. The warranty return cycle time was measured for a period of 6
weeks for 4 regions. The Green Belt wants to investigate whether there is any
difference in average warranty repair cycle time between each of the regions.

(b) The Green Belt would also like to know if there is any difference in the variance of
the cycle time between each of the regions.

2. Checks Are Us is a payroll processing firm. Timecard errors are routinely monitored
and recorded. A Black Belt investigating the errors wishes to determine if there are
any differences in the number of errors between five of its major customers. The
number of errors contained in a sample of 150 employees was recorded for five
weeks. How would you proceed with stratifying the data to make this investigation?

3. Tungsten steel erosion shields are fitted to the low pressure blading in steam turbines.
The most important feature of a shield is its resistance to wear. Resistance to wear
can be measured by abrasion loss, which is thought to be associated with the
hardness of steel. How would you investigate the relationship between hardness (the
X) and abrasion loss (the Y)?

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

Graphical Analysis–Breakout Activity (15 minutes) 8

Desired Outcome: Identify possible causes of variation in the Capital


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

What How Who Timing

Minitab ƒ Use the appropriate graphical All 15 mins.


Analysis analysis tool in Minitab to
investigate
– Differences in delivery cycle time
between the two different load
types (See note below)
– Differences in delivery time
between drivers
– Relationship between delivery
cycle time and distance
Report ƒ Be prepared to share your results All
Out with the class

Load types: D/H Drop & Hook–A driver unhooks their full trailer, and then hooks into an empty to take
back. Unload–The driver opens the trailer and allows the store to unload its freight, and brings back
the trailer.

To create Boxplots:
Stat > Basic Statistics: > Display Descriptive Statistics
variable: time
click: by variable: load type or driver
click Graph > Choose Boxplots

To create Scatter Plots:


Graph > Plot
X: Distance
Y: Time

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

Process Map Analysis 9

Types Of Analysis
ƒ Moments of truth–What does the customer feel?
ƒ Nature of work–What is of value to the customer?
ƒ Flow of work–What makes the customer wait?

The purpose of Process Map Analysis is to: The second, Nature of Work, is concerned with value
„ View our process and its performance from the analysis: understanding whether or not the work in your
customer’s perspective process is valued by the customer or by your internal
culture.
„ Identify current process problems and opportunities
„ Create a shared sense of urgency for improving the The third, Flow of Work, characterizes the work along
the dimension of time. The two latter components are
process clearly identify and document value-
focused on the “Profitability” piece of the GE Capital
added/nonvalue-added activities and process flow
Quality Vision.
measures and characteristics.

There are three key components to the Process Map


Analysis step:

The first, Moments of Truth, ensures that we are


“Completely Satisfying Customer Needs” through a
customer-focused process improvement perspective.

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

Moments Of Truth 10

A Typical Deployment Flow Chart

Initial Contact Service/Produce Delivery Follow-Up

Customer Moments
of Truth

Intermediaries
Or Front Line

Backroom
Activity

Support Or
Suppliers

A moment of truth is defined as anytime a customer “If you’re a service person and you get it wrong at your
draws a critical judgment, positive or negative, about the point in the customer’s chain of experience, you are very
service, based upon a service experience (or lack likely erasing from the customer’s mind all the memories
thereof). Certainly every time we “touch” the customer is of the good treatment he or she may have had up until
a moment of truth, but it isn’t limited to these cases. now.
But if you get it right, you have a chance to undo all the
From the customer’s view, the process is a set of service wrongs that may have happened before the customer got
encounters. In this case, the customers believe the steps to you. You really are the moment of truth.”
should be simple: apply for the loan and receive the
money.

In the best of cases, the customer is unaware of all the


hard work, delays, and hand-offs which go into making
their loan a reality. It occurs behind the scenes.

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

Nature Of Work–Value Analysis 11

Value-Added Work Nonvalue-Added Work

Steps That Are Steps That Are Considered


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

Steps That Are Not


Essential To The
Customer, But That
Allow The Value-
Adding Tasks To Be
Done Better/Faster.

Value-Enabling Work

„ Your policies address only control issues


„ There are more managers than workers
„ People get rewarded for managing problems (e.g.,
Manager of Reissues, Special Expediting)
„ Rework is a common characteristic of most processes
– RE’s are things in your process that are done more
than one time (Redo, Recall, Reissue).
– RE’s usually cause you to loop back to an earlier
point in your process.
– RE’s consume time, add to the complexity, and use
additional resources.
„ OPTIONAL: Show the video: “Time the dimension of
Quality”

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

Types Of Nonvalue–Added Work 12

Internal Failure Delay

External Failure Preparation/Set-Up

Control/Inspection Move

What Does The Customer Value?

Internal Failure Delay


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

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

Flow Of Work 13

Process Time

+ Delay Time

Cycle Time

To analyze the flow of work, break the process flow Terms


down to its lowest component, and analyze the movement
of the component through the process. For example, Cycle Time
The total time from the point in which a customer
“become an application” and trace its progress through requests a good or service until the good or service is
the process. It is similar to a process flowchart, but with delivered to the customer. Cycle time includes process
the value-added and nonvalue-added steps clearly time and delay time.
identified.
Process Time
The total time that a unit of work is having something
done to it other than time due to delays or waiting. It
includes the time taken for value-added steps, internal
failure, external failure, control, inspection,
preparation/set-up, and more.

Delay Time
Total time in which a unit of work is waiting to have
something done to it.

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

Flow Of Work–Process Disconnects 14

ƒ Gaps
ƒ Redundancies
ƒ Implicit or unclear requirements
ƒ Inefficient hand-offs
ƒ Conflicting objectives
ƒ 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.

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

Flow Of Work–”Be The Unit” 15

Unclear
requirements

1. Receive
Are we Yes
application in 11. Make loan extending
mail and open decision
Redundancy loan?
envelope

No 12. Generate
2. Place 10. Review for
Inefficient loan packet
application in completeness
hand-off 19. Generate
mail slot and make
turndown letter
decision

3. Move 9. Queue 13. Place in 18. Postman


application to application for out-box picks up
Entry Dept. credit review outbound mail

Unclear
4. Place requirements Inefficient 14. Move to 17. Place in
8. Score hand-off
application in mailroom outbound mail
application
in-box basket

5. Retrieve 7. Enter
Yes 15. Wait for 16. Post package
application and Is application application
postage or letter
review for complete? to computer
completeness system

No

6. Call to obtain
necessary
information

Cycle time is the total time taken from the point at which Constructing a time line helps identify areas of potential
the customer requests a good or service until the good or improvement by illustrating the time required for each
service is delivered to the customer. It is the sum of the step in a process.
process time, move time, inspect time, delay time, and
storage time. Remember:
Time data may not be available for all steps in a process.
Cycle time analysis often provides eye-opening insights. Some possible sources include:
People seldom think of work in this context. When an „ Time, schedule, and activity logs
Improvement Team presents cycle time data, „ Files, studies, and reports
management may become defensive. For example, when
one manager at a commercial insurer found out it was If data is not available:
taking 100 days, on average, to process the commercial
line of a new business, she resisted the analysis. After
„ Estimate and ten track to determine actual average
heated “discussion,” she suggested that the period
value
sampled did not reflect normal conditions. She
recommended looking at another period of time to get a
“better number.” The team obliged. The next sample Use minimum of 30 data points to determine an average
averaged 103 days! value when calculating averages, also record the highest
and lowest time so consistency is understood.

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

Linking Value Analysis With Process Flow 16

Summarized Analysis
% %
Process Step 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 Total Total Steps

Est. Avg. Time (Mins) 1 120 15 120 3 180 7 1 120 5 10 15 90 15 120 2 120 5 8 957 100%

Value-Added 9 9 99 9 99 48 5.0%

Nonvalue-Added

Internal Failure 9 180 18.8%

External Failure

Control/Inspection 9 9 8 .8%

Delay 9 9 9 9 9 9 690 72.1%

Prep/Set-Up

Move 9 9 30 3.1%

Value-Enabling 9 1 .1%

Total 957 100%

Linking value analysis with cycle time analysis creates a Conclusion: Example
compelling business case for change. „ 7 steps (-40% of total steps) provide 100% value and
take 48 minutes or 7% of total cycle time.
Displaying this information on process maps emphasizes „ 11 steps (-61% of steps) are non-value-added, and
focus areas for improvement. consume 94.9% of cycle time.

Tips/Traps
„ Don’t get stuck on deciding if a step is
value-added or nonvalue-added
– Allow the team to discuss it.
– Consensus with 2/3s majority vote.
„ Push the team to make the process map a true “as is.”
– The ratio for an average “as is” map is 20% value-
added, 80% nonvalue-added.
– Think of the process map from the point of view of
the product or service.

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

Review: Cause & Effect Diagrams 17

A visual tool used by an improvement team to brainstorm and


logically organize possible causes for a specific problem or effect.

Machines Methods Materials

Potential High-Level Causes


Problem
Statement

Measurement Mother Nature People

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


Analyze phase the Cause and Effect diagram is often
used to identify X’s.

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

Cause & Effect Diagrams–The Five Why’s 18

The “Five Why’s” drill deep into the process to identify


potential Root Cause(s)
ƒ Ask “why” five times to identify deeper causes
ƒ Use process data to answer each “why” question

Example 4. Why is the format confusing?


A project team has verified that the X (complicated form)
ƒ 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.
ƒ Because the font size is too small
1. Why do complicated forms cause delays in the
underwriting steps?
ƒ Because underwriters receive incomplete
applications

2. Why do they receive incomplete applications?


ƒ Because customers don’t fill out the form
accurately or completely

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


completely?
ƒ Because the format is confusing
© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

Cause And Effect–A Breakout Activity (25 minutes) 19

Desired Outcome: Identify possible causes of variation in the Capital Logistics


case study data

What How Who Timing

Preparation ƒ Select a facilitator, scribe and timekeeper All 5 mins.

Construct Cause ƒ As a result of the QFD constructed in All 15 mins.


And Effect Measure 1 for the Capital Logistics
Diagram customer service desk, a project team
elects a project Y of cycle time to
resolve customer problems
ƒ In the Analyze phase of the project, the
team uses a cause and effect diagram
to brainstorm potential X’s
ƒ Construct a cause and effect diagram
for the problem statement “Why is it
taking too long to resolve customer
problems?”
ƒ Base cause and effect diagram on your
general knowledge of customer service
desks. What might be potential X’s?

Five ƒ Use the ask “why” approach to identify All


Why’s deeper causes as appropriate

Close ƒ Choose a spokesperson to share the All 5 mins.


results with the class

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

Prioritization Of X’s–Control/Impact Matrix 20

IMPACT

High Medium Low

C In Our
O Control
N
T
R
O
L
Out Of
Our Control

Always Verify With Data

Use your team’s process knowledge and business


experience to list possible X’s in a Control/Impact Matrix.
Then use process data to verify or disprove placement of
the X’s.

Prioritization Steps

1. Using the Control/Impact Matrix shown above, examine


each X in light of two questions:
ƒ What is the impact of this X on our process?
ƒ Is this X in our team’s control or out of our team’s
control?

2. With your team, place each X in the appropriate box on


the matrix.

3. The validated matrix is a guide to addressing the X’s.


Begin with the “High Impact/In Our Control” category.
© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

Prioritization Of X’s–Control/Impact Matrix (continued) 21

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

The project team completed the Cause & Effect diagram


and the prioritization matrix. The X’s that are identified
as “in control/high impact” appear to be the top priority.
However, the team should discuss the other X’s to make
sure they have reached consensus for those priority X’s.

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

Control/Impact Matrix–Breakout Activity (15 minutes) 22

Desired Outcome: Practice using a Control/Impact Matrix to prioritize X’s

What How Who Timing

Preparation ƒ Choose a facilitator, scribe, All 2 mins.


timekeeper and/or note taker

Construct ƒ Review the potential X’s generated All


Cause for the customer service desk project
And in the cause and effect diagram
Effect activity
Diagram

Prioritize ƒ Use the Control/Impact Matrix on All 10 mins.


Potential X’s the next page to prioritize the X’s
identified by your team

Close ƒ Choose a spokesperson to report All 3 mins.


out on your high impact X’s. Are all
of these X’s within the control of the
team?

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

Prioritization Of X’s–Control/impact Matrix 23

IMPACT

High Medium Low

C In Our
O Control
N
T
R
O
L
Out Of
Our Control

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

24

Analyze 6.2
Narrow list of Potential Causes

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

The Idea Of Sampling 25

Based on the sample, we make decisions about the population.


Population (Universe): A
set of characteristics that
defines membership in the
complete set.

Sample: A subset of
members that possesses
the same characteristics as
that of the population.

Why should we take a sample?

Should the sample be random?

Is it possible to have sampling error?

How many samples should be taken?

What are some everyday examples of sampling?

In most cases under study, it is impossible to measure


every element of a population. In particular, we want
samples that consist of observations that are independent
and random. These samples, called random samples, can
be used to make decisions about the population. This
involves a certain amount of risk. We can minimize this
risk, and, more importantly, understand what the risks are
if we use statistical tests.

If you were interested in finding the average height of all


of the people in the United States, what advantages would
exist in taking a sample?

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

Hypothesis Testing–Introduction 26

ƒ Refers to the use of statistical analysis to determine if observed


differences between two or more data samples are due to
random chance or to be true differences in the samples
ƒ Increase your confidence that probable X’s are statistically
significant
ƒ Used when you need to be confident that a statistical difference
exists

An assertion or conjecture about one or more parameters


of a population(s).
„ To determine whether it is true or false, we must
examine the entire population, this is impossible!!
„ Instead, use a random sample to provide evidence
that either supports or does not support the
hypothesis
„ The conclusion is then based upon statistical
significance
„ It is important to remember that this conclusion is
an inference about the population determined from
the sample data

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

Hypothesis Testing For Equal Means 27

The histograms below show the height of inhabitants of countries


A and B.
Both samples are of size 100, the scale is the same, and the unit
of measurement is inches.
Question: Is the population of country B, on average, taller than
that of country A?

Country A

Country B
60.0 62.0 64.0 66.0 68.0 70.0 72.0 74.0 76.0 78.0 80.0
[inch]

Issue: how conclusive is the evidence that the sample


results indicate a real, more-than-random effect in the
underlying population or process?

In the Analyze phase we will try to determine which X’s


have an effect on the Y. We can compare two sets of
data, with X set at different values, thereby determining if
that X has an effect.

Examples: Does a process perform better using


machine/material/fixture/tool A or B? Does the
purchased material conform to the desired specifications?
Is there a difference in performance between vendor A or
B? Is there a difference in your process after you make a
change? Is the process on target?

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

Concepts Of Hypothesis Testing 28

1. All processes have 2. Samples from one


variation. given process may
vary.

3. How can we differentiate


between sample–based
“chance” variation and
a true process
difference?

To improve processes, we need to identify factors which This way everyone makes the same decisions.
impact the mean or standard deviation.

Once we have identified these factors and made


adjustments for improvement, we need to validate actual
improvements in our processes.

Sometimes we cannot decide graphically or by using


calculated statistics (sample mean and standard deviation)
if there is a statistically significant difference between
processes.

In such cases the decision will be subjective.

We perform a formal statistical hypothesis test to decide


objectively whether there is a difference.

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

Kinds Of Differences 29

Continuous data:
ƒ Differences in averages
ƒ Differences in variation
ƒ Differences in distribution
“shape” of values
Discrete data:
ƒ Differences in proportions

There are a variety of different hypothesis tests. Each one


tests for a different kind of “difference.”

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

Hypothesis Testing 30

Guilty vs. Innocent Example

The American justice system can be used to illustrate the


concept of hypothesis testing.

In America, we assume innocence until proven guilty.


This corresponds to the null hypothesis.

It requires strong evidence “beyond a reasonable doubt”


to convict the defendant. This corresponds to rejecting the null
hypothesis and accepting the alternate hypothesis.

Ho: person is innocent


Ha: person is guilty

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

Nature Of Hypothesis 31

Null Hypothesis (Ho): Alternative Hypothesis (Ha):


ƒ Usually describes a status ƒ Usually describes a
quo difference
ƒ The one you assume ƒ Signs used in Minitab:
unless otherwise shown ≠ or < or >
ƒ The one you reject or fail
to reject based upon
evidence
ƒ Signs used in Minitab:
= or > or <

Note that we are not proving the hypothesis to be true or


false. We will reject or fail to reject the null hypothesis
based on the evidence from our samples. Failing to reject
the null hypothesis implies that the data does not provide
sufficient evidence to conclude that a difference exists.
On the other hand, rejection of the null hypothesis
implies that the sample data provides sufficient evidence
to say that a difference exists.

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

Activity–Hypothesis Statements (10 minutes) 32

Write the null and alternate hypothesis testing statements for each
Scenario 1: You
scenario have collected delivery time of supplier A and supplier B. You wish to test
below:
whether or not there is a difference in delivery time from supplier A and B.

Scenario 1: statement
Null hypothesis You have : collected delivery time of supplier A and supplier B.
You wish
Alternate to test statement:
hypothesis whether or not there is a difference in delivery time from
supplier A and B.
Scenario 2: You suspect that there is a difference in cycle time to process purchase orders in
site 1 of your company compared to site 2. You are going to perform a hypothesis test to verify
Null hypothesis statement :
your hypothesis.

Null hypothesis statement :


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:

Scenario 3: You have implemented process improvements to reduce the


cycle time to process purchase orders in your company. You have collected
cycle time before the process improvements and after the process
improvement was implemented. You are going to perform a hypothesis test
to verify that the process improvements have resulted in a reduction in
cycle time.

Null hypothesis statement :

Alternate hypothesis statement:

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

Hypothesis Testing 33

Guilty vs. Innocent Example


The only four possible outcomes:
1. An innocent person is set free. Correct decision

2. An innocent person is jailed. Type I error = α


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

3. A guilty person is set free. Type II error = β


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

4. A guilty person is jailed. Correct decision

Why is it important to minimize the chance of making a


Type I error in a six-sigma project?

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

Hypothesis Testing–Another View 34

Ho: Person is innocent. Truth


Ha: Person is guilty. Truth
Ho Ha
Innocent Guilty

Innocent, Guilty,
Ho
Set Free Set Free
Set Free Type II
β
Verdict
Verdict

Innocent, Guilty,
Ha Jailed Jailed
Jailed Type I
α

This is a visual way of looking at the four possible


outcomes of hypothesis testing.

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

Hypothesis Testing 35

The P-value is calculated by Minitab


ƒ 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.
ƒ Ranges from 0.0 - 1.0
ƒ The alpha (α) level is usually set at 0.05. Alpha is the
probability of making a Type I Error (concluding there is a
statistical difference between samples when there really is no
difference).

P < α: Reject Ho
P > α : Accept Ho

The p (probability) value is the statistical measure for


the strength of H0, usually reported with a range between
0.0 and 1.0. The higher the p-value, the more evidence
we have to support H0, that there is no difference. Think
of the null hypothesis as a jury trial: the accused is
innocent until proven guilty. In hypothesis testing, the
samples are assumed equal until proven not. Since we are
usually doing a hypothesis test to prove there is a
difference, we are looking for p-values less than 0.05.

By convention if p >.05 accept H0 (no difference).


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

Reject Ho Accept Ho
Different Not Different
0.00 0.05 1.00
p-value

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

Statistical Tests In Minitab 36

Some basic statistical tests are shown below with the command for running each
test in Minitab.

What The Tool Tests Statistical Test Graphical Test


Histogram
Mean of population data is 1-Sample t-test
different from an established Stat > Basic Statistics
target. > 1-Sample t

Mean of population 1 is 2-Sample t-test


Histogram

different from mean of Stat > Basic Statistics


population 2. > 2-Sample t

The means of two or more 1-Way ANOVA


Histogram

populations is different. Stat > ANOVA > One-Way

Variance among two or more


Homogeneity of
populations is different. Scatter
Variance
Stat > ANOVA >
Plots
Homogeneity of Variance

Output (Y) changes as the Linear


input (X) changes. Regression Box
Stat > Regression >Fitted Plots
Line Plot

Chi-Square Test
Frequency

Output counts from two or Pareto


more subgroups differ. of Independence
Stat > Tables >
Cross Tabulation OR
Chi-Square Test C AB D E M NO
Category

Data is normally Normality Test


distributed Stat > Basic
Statistics

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

Select A Statistical Test 37

Hypothesis tests to find relationships between project Y and


potential X’s

Y
Continuous Discrete

Continuous Simple Linear


Regression
X 2 Sample t-Test
Discrete (Compare Means of two
samples)
Chi-Square Test
ANOVA (Compare means of
multiple samples)
Homgeneity of Variance
(Compare variances)

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

Hypothesis Test Summary 38

Normal Data Non-Normal Data


Variance Tests (Continuous Y) Variance Tests (Continuous Y)
X2 - Compares a sample variance to a known Homogeneity of Variance Levine’s–Compares two
population variance.
or more sample variances.
F-test- Compares two sample variances.
Median Tests (Continuous Y)
Homogeneity of Variance
Levine’s–Compares two or more sample Mood’s Median Test–Another test for two or more
variances medians. More robust to outliers in data.
Mean Tests (Continuous Y) Correlation–Tests linear relationship between two
T-test One-sample–Tests if sample mean is equal variables.
to a known mean or target. Proportion Tests (Continuous Y)
T-test Two-sample–Tests if two sample means are P-Test–Tests if two population proportions are
equal. equal.
ANOVA One-Way–Tests if two or more sample Chi-Square Test–Tests if three or more relative
means are equal. counts are equal.
ANOVA Two-Way–Tests if means from samples
classified by two categories are equal.
Correlation–Tests linear 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.)

There are a number of hypothesis tests for both normal


and non-normal data. You should consult a Black Belt or
Master Black Belt if you are not sure which test to use for
you project, or if your project involves non-normal data.

The next page shows a summary of the tests that we will


look at in detail during this training. We have already
looked at the Normality Test in previous modules.

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

We Always Look At Our data In This Way 39

1. Check Stability–Run Chart, p-values


2. Check Shape–Anderson-Darling Normality Test
3. Check Spread–Appropriate Hypothesis Test
4. Check Center–Appropriate Hypothesis Test

The remainder of this module will teach you how to


choose the correct hypothesis test. Not all material will be
covered in class. The instructor will choose the tests to
cover, based on time restrictions. Prior to taking the GB
certification exam you should review this section and go
through the Minitab commands.

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

Choosing The Correct Hypothesis Test 40

Are X’s YES Is the data NO


Discrete? normal?
Mood’s Median
HOV
NO YES

Regression
Are Y’s NO CHI
Continuous?
SQUARE (χ2)

YES

Comparing NO Multiple Groups


Only 2 ANOVA
Groups?
HOV
YES

Are We
Can I Match NO Comparing NO
X’s With X’s? To A 2 Sample t–Test
Standard? HOV
YES YES

Paired t 1 Sample t

Note: In order to use this chart, we are assuming our X’s Paired t-Test
are discrete. Otherwise, use Regression. Ho = Difference = 0
Ha = Difference ≠ 0
Moods Median Test
One Sample t-Test
Ho = Median1 = Median 2 = ...Median n
Ho = µ = Target
Ha = At least one Median is Different
Ha = µ ≠ Target
Homogeniety of Variance (Hov) – Levene’s Test
Two Sample t-test
Ho = σ12 = σ 22 = σ 32 ...σ 2n Ho = µ 1 = µ 2
Ha = At least σ 2 is Different Ha = µ 1 ≠ µ 2
Chi Square
Ho = χ 2 = 0
Ha = χ 2 ≠ 0

Analysis of Variance (ANOVA)


Ho = µ 1 = µ 2 = ...µ n
Ha = At least one µ is Different

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

Which Hypothesis Testing Tool Would You Use? 41

For each scenario described below, which hypothesis testing tool


would you use? Assume normal distribution, where appropriate

1. A six-sigma project is being conducted in the field to improve the cycle time for
warranty repair returns. The warranty return cycle time was measured for a
period of 6 weeks for 4 regions. The Green Belt suspects that there is a
difference in average warranty repair cycle time among each of the regions.
How would you test whether there is a statistically significant difference in mean
cycle time for the different regions?

2. Tungsten steel erosion shields are fitted to the low pressure blading in steam
turbines. The most important feature of a shield is its resistance to wear.
Resistance to wear can be measured by abrasion loss, which is thought to be
associated with the hardness of steel. How would you test whether there is a
statistically significant relationship between resistance to wear and abrasion
hardness of steel?

3. Your business purchases sheet stock from two different suppliers. It has found
an unacceptably large number of defects being caused by thickness beyond
tolerance levels. Data for overall mean thickness data was analyzed and found
to be on target. Data was collected that would identify a potential difference in
the variation of the thickness of the material by supplier.

4. Checks Are Us is a payroll processing firm. Timecard errors are routinely


monitored and recorded. A Black Belt investigating the errors wishes to
determine if there are any differences in the number of errors among five of its
major customers. The number of errors contained in a sample of 150 employees
was recorded for five weeks. How would you test if there is a statistically
significant difference in the number of errors among the customers?

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

Data Analysis 42

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

Stability–Minitab Run Chart 43

Let’s look at an example from our GB ANOVA.mpj.


Step 1: Check the stability of the processing time for SOUTH
Territory. Are there any trends in the data?
MINITAB FILE: GB ANOVA.mpj

File Description:

Territory (C1): Sales Territory

Zone (C2): city within the sales territory

Total Time (C3): Processing Time- stacked Territories


C4-C8– processing time per territory noted

Use Minitab to check the stability of your processing time


data by territory.

The Run chart is used to look at the stability of data in


time order or sequentially ordered.

Is the process stable over time for the South Territory?

© GE Capital, Inc., 2000

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.”

Process is Stable if all p-Values are > 0.05

It is important to note that the data should be in time Test for Trends: Ho: No fewer runs observed than
order for run charts to be valid. expected. Ha: More runs observed than expected. Trends
result from a gradual increase or decrease in the measure
A run is one or more consecutive points in the same so that there are fewer runs than expected by sampling
direction. A new run begins each time there is a change error alone.
in direction. Do not count points exactly on the median.
Test for Oscillations: Ho: No more runs observed than
Test for Clusters: Ho: No fewer runs observed than expected. Ha: More runs observed than expected.
expected. Ha: Fewer runs observed than expected. Oscillations are similar to mixtures in that they indicate
Clusters result from sampling from one process for a too many runs—more than expected
period of time and then from another process with a by sampling error alone.
different center for a period so there are fewer runs than
expected by sampling error alone.
Since all p-values are less than 0.05, the process is stable
for the south territory. You would also need to check
Test for Mixtures: Ho: No more runs observed than
each of the other territories for stability.
expected. Ha: More runs observed than expected.
Mixtures result from two or more processes with different
centers alternately sampled so there are too many runs
than expected by sampling error alone.
© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

The P-Value Review 45

ƒ Alpha is the probability of making a Type I error


ƒ The p-value is the probability of getting the observed difference
or greater when Ho is true
ƒ Unless there is an exception based on engineering judgment,
we will set an acceptance level of a Type I error at a = 0.05
ƒ Thus, any p-value less than 0.05 means we reject the null
hypothesis

p < α: Reject Ho
p > α: Accept Ho

The alpha is another decision criteria. It will give you the


same conclusions as the confidence interval.

In general, comparing p-value to alpha is the decision


criteria we use most often in this course.

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

Data Analysis 46

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

Shape–Minitab Descriptive Statistics 47

Step 2: Check the shape of the data. Is the data normally


distributed?
MINITAB FILE: GB ANOVA.MPJ

Ho: The data is normally distributed


Ha: The data is not normally distributed

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

Descriptive Statistics–Input and Output 48

1. Double
Click

3. Click
OK 2. Click on Graphs,
Select Graphical Summary

Descriptive Statistics
Variable: South
Non-normal
Anderson-Darling Normality Test distribution!
A-Squared: 3.095
P-Value: 0.000

Mean 11.9348
StDev 11.1064
Variance 123.351
Skewness 1.75318
Kurtosis 2.69797
N 46
0 10 20 30 40
Minimum 1.0000
1st Quartile 4.0000
Median 9.0000
3rd Quartile 14.2500
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 Normal if P-Value is > 0.05 (Accept Ho)

If the p-value is > 0.05, then we Accept the Ho and the


data is normally distributed. In this case, our p-value is
0.000, therefore, we have a non-normal distribution.

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

Non-Parametric Tests 49

What if I don’t have normal data?

Perform Non-parametric Tests

Use HOV for spread


and
Mood’s Median for Center

ƒ Normal data allows a great variety of techniques to be


used in analyzing it
ƒ However, responding correctly is critical if the data is
not normal (which is not at all uncommon)

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

Non-Parametric Tests 50

ƒ One of the advantages of nonparametric tests is that they


assume no knowledge about the underlying distributions. They
often use an analysis of the ordered ranks of the data.
ƒ Nonparametric tests are more powerful for non-normal data
than the equivalent t-tests and ANOVA tests
ƒ The Mood’s Median and Homogeneity of Variance Test
(HOV) are nonparametric tests which are very similar to the
One-Way ANOVA test

ƒ Many times in commercial projects the data are not


normally distributed. Cycle time is one example. Cycle
time will often be skewed to the right.
ƒ 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.
ƒ In these cases a test that does not depend on
assumptions of normality may be very useful

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

Spread–Test On Variances 51

Many times we want to know if we have succeeded in reducing the


variation of a process, or we might want to know if a change in a
variable (X) changes the variation in the output (Y).

Knowing if the variances of two (or more) samples are different is


also a prerequisite to the test on means.

Are these Variances Different?

Xb

Xa

The hypothesis test for comparing variances


is the Homogeneity of Variance Test

In the Analyze phase we will try to determine which X’s


have an effect on the Y. We can compare two sets of
data, with X set at different values, thereby determining if
that X has an effect.

Examples: Does a process perform better using


machine/material/ fixture/tool...A or B? Does the
purchased material conform to the desired specifications
(m & s)? Is there a difference in performance between
vendor A or B? Is there a difference in your process after
you make a change? Has a variation source been
removed?

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

Spread–Homogeneity Of Variance 52

Step 3: You want to compare the variability of the Sales Regions.


You have established that each process is stable but non-normally
distributed. Perform the Homogeneity of Variance Test .

MINITAB FILE: GB ANOVA.MPJ

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

Test On Variances, Example–Input 53

2 = 2 = 2 = 2 = 2
Ho : σEast σ West σ IFG σ Central σ South
Ha : at least one is different α = .05

© GE Capital, Inc., 2000

DMAIC GB Q TX PG V 4.2.0
D M A I C

Test On Variances, Examples–Output 54

2 = 2 = 2 = 2 = 2
Ho : σEast σ West σ IFG σ Central σ South
α = .05
Ha : at least one is different
Homogeneity of Variance Test for total time
95% Confidence Intervals for Sigmas Factor Levels

Central

Bartlett's Test

Test Statistic: 58.348


East
P-Value : 0.000

IFG

Levene's Test

South Test Statistic: 10.396


P-Value : 0.000

West

5 15 25 35 45 55

We have standardized on
The Levene’s test because
it is more robust to non-