Six Sigma Manual

D M A
Classroom Session AgendaDay 1 Total Min 60 20 20 10 10 90 60 10 45 5 120 Topic Introductions and Expectations Meeting Skills Define A,B and C Review and CAP Break Measure Overview Measure 1 Review CTQ Tools Measure 1CTQ Tools continued Break Measure 2Performance Standards Break Minitab Tutorial and Graphical Analysis various 2 1,2,3 1 1 Step N/A N/A A,B,C
GE Capital, Inc., 2000 DMAIC GB A TX PG V4.2.0
D M A
Classroom Session AgendaDay 2* Total Topic Min 30 30 15 15 45 75 60 15 15 150 Day 1General Review and Homework Minitabcontinued Measure 3Overview Break Measure 3Data Collection Plan Measure 3Sampling Measurement System Analysis (MSA) Break Analyze Overview Analyze 4Establish Process Capability 4,5,6 4 3 3 3 Step A,B,C,1,2 various 3
* Elevator Speeches will be shared throughout the day
D M A
Classroom Session AgendaDay 3* Total Topic Min 30 15 15 150 30 15 30 15 30 15 60 Day 2General Review Break Analyze 6Identify Variation Sources Analyze 6continued Improve Preview Improve 7 and 8Design Of Experiment Break Improve 9Statistical Tolerancing Control Preview Control 12Control Charts 9 10,11,12 12 6 6 7,8,9 7,8 Step various
Analyze 5Define Performance Objective 5
* Elevator Speeches will be shared throughout the day

D M A
D M A
Assessment Strategy Introductions The Learning Objectives are divided in 3 categories: 1. Classroom 2. Project, and 3. Test It is expected that a Green Belt be proficient in all learning objectives, however, only those designated Test represent the content of the certification exam. Each phase of the DMAIC cycle is listed below and the learning objectives have classified into the 3 categories within each phase.
GE Capital, Inc., 2000 DMAIC GB B TX PG V 4.2.0
D M A
OverviewClassroom Learning Objectives

1 Explain the benefits of Six Sigma to GEs business. Compare and contrast Six Sigmas process improvement approach to quality with traditional defect prevention strategies (i.e., inspection and testing). Identify the vital few CTQs that apply to all GE customers: responsiveness; marketplace competitiveness; on time, accurate and complete deliverables; and product/service technical performance. Explain the relationship between increasing levels of process complexity and quality improvement results. 2 Describe the Six Sigma Methodology for quality improvement. Define the term sigma (standard deviation) as it relates to the sigma capability (z value) of a business or manufacturing process. Recognize a Six Sigma level of quality (i.e., 99.99966% probability that defects will not be passed on to the customer). Define key Six Sigma terms and acronyms, including CTQ, opportunity, defect, DPMO, and Six Sigma capability (Z value). Explain the Master Back Belt (MBB), Black Belt (BB) and Green Belt (GB) roles in Six Sigma. Describe Six Sigmas focus on repeatable processes. Describe Six Sigmas focus on inputs (Xs) over outputs (Ys) using the formula Y=f (X). Describe the statistical objective of Six Sigma (i.e., reduce process variation). Describe the relationship between DPMO and process capability (i.e., as DPMO goes down, process capability goes up. Describe the financial benefits of Six Sigma to GE.
D M A
OverviewProject Learning Objectives 1. Determine if DMAIC is the right strategy by identifying the conditions under which the DFSS methodology would be more appropriate. Compare and contrast the DFSS design methodologies to DMAIC.
D M A
OverviewTest Learning Objectives 1 Describe the 5 phases of DMAIC, including the purpose, tools, and outputs for each phase.
D M A
DefineProject Learning Objectives 1 Identify project CTQs. Define CTQ (Critical to Quality Characteristic). Identify customer(s) in a quantifiable way.
Recognize the components of a process (i.e., supplier, input(s), subprocess, output(s), customer(s). Distinguish between internal and external customers.
Compile and evaluate customer CTQ data.

Distinguish between customer driven CTQs and process driven CTQs. Recognize sources of existing customer data. Assess customer requirements and expectations. Recall the vital few customer CTQs. Analyze the voice of the customer and its impact on CTQ data. Translate customer needs into requirements (CTQs).
D M A
DefineProject Learning Objectives (continued) 2 Use a process/product drill-down tree to: define the limits of a project (project bounding); clarify what the project is and is not; identify other areas for improvement. Create a process/product drill-down tree. Integrate measurements to clarify areas needing improvement. Given an example of a process/product drill down tree, identify viable Six Sigma projects.
D M A
DefineProject Learning Objectives (continued)

3 Develop a team charter. Describe the purpose of a charter. Identify the five major elements of a charter. Define the business case for a project in terms of its potential benefits, the consequences of not doing it, its relationship to other activities, and its fit with business initiatives/target. Develop a problem statement. Describe the customers pain. Identify key considerations and potential pitfalls to consider when developing a problem statement. Develop a SMART goal statement (specific, measurable, attainable, relevant, time bound). Assess the scope of the project. Identify the 8 steps for bounding a project. Define project milestones. Select a project team and define team roles. Identify team roles and responsibilities. Evaluate a proposed Six Sigma project. Recognize characteristics of a good project. Recognize characteristics of a bad project.
D M A
DefineProject Learning Objectives (continued) 4 Map a business process. Describe the goal of process mapping. Identify the components of a process map (COPIS). Describe the steps involved in creating a process map. Define and name a process. Given a business process, use brainstorming and storyboarding techniques to: identify its outputs, customers, suppliers, and inputs; identify customer requirements for primary outputs; and identify process steps
D M A
DefineProject Learning Objectives (continued) 5 Obtain approval for a Six Sigma project. Identify the steps in the project approval process. Enter a project into QPT.
D M A
DefineTest 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 its key deliverables: CTQs, team charter, and process map. Identify the five key objectives of the Define Phase. 3 Describe the CAP tools and their connection to Six Sigma.
D M A
Measure1Classroom Learning Objectives 1 Describe and define the deliverables of Step 1.
11
D M A
Measure 1Project Learning Objectives 1 Identify the project Y. Identify the tools that may be used to select the relevant CTQ or Y on which to focus. Explain the purpose Quality Function Deployment, Process Map, and FMEA tools have. Define performance standards for Y including specification limits as well as defect and opportunity definitions.
12
D M A
Measure 1Project Learning Objectives (continued) 2 Select CTQ characteristics. Select the Critical to Quality (CTQ) characteristic to be improved in a project. Narrow the focus of a project to an actionable level. Establish the project team and gained consensus on the project definition.
13
3 Select and apply appropriate tools to narrow the focus of a Six Sigma project by identifying key areas for improvement. Identify tools that may be used to narrow the focus of a project, including Process Map, and FMEA. Recognize the purpose and benefits of each tool.
D M A
Measure 1Project Learning Objectives (continued)

4 Use Process Mapping to identify potential breakdowns, rework loops, and sources of variation in a process. Use the C.O.P.I.S. model to illustrate a customer focused process. Identify the elements of a process (input, mechanism, control, output, process boundary). Identify and distinguish between internal and external process controls. Recognize the purpose and benefits of process mapping. Recognize the three types of process maps. Describe the process mapping process, including the following steps: Determine the scope. Determine the steps in the process. Arrange the steps in order. Recognize ISO 9000 standard symbols for process mapping. Validate a process map. Evaluate a process map.
14
D M A
Measure 1Project Learning Objectives (continued)
15
5 Use a Failure Modes and Effects Analysis to identify the potential failure modes of a process or product. Recognize the purpose and benefits of FMEA. Describe how FEMA works. Describe FMEA, including preparation, process, and improvement steps. Define the terms failure mode, cause, and effect, as they relate to FMEA, and recognize examples of each. Assign degree of severity, likelihood of occurrence, and ability to detect ratings, and calculate a risk priority number (RPN). Complete an FMEA form.
Recognize when and by whom an FMEA is prepared, updated, and completed.
D M A
Measure 1Project Learning Objectives (continued) 6 Conduct a test-retest study and analyze the results. Describe the purpose and procedure for conducting a testretest study. Plot and test-retest study data. Use descriptive statistics to evaluate test-retest study data. 7 Establish a Data Collection Plan for a Six Sigma project. Describe the purpose and benefits of a Data Collection Plan. Write a data collection strategy. Define a clear strategy for collecting reliable data efficiently.
16
D M A
MeasureTest Learning Objectives

1 Recall the DMAIC 12 Step process, and distinguish between characterization phases (DMA) and optimization phases (IC). Define product characterization. Define process optimization. 2 Recognize how statistics can be applied to the problem solving process. Define the terms precision and accuracy as they relate to a Six Sigma process. Relate precision/variation and accuracy/mean to quality and customer satisfaction. State the goal of Six Sigma in statistical terms. Define the term Upper Specification Limit (USL). Define the term Lower Specification Limit (LSL). Define the term target (T). Define .
17
D M A
MeasureTest Learning Objectives 3 Explain how statistics can be used to solve problems. Identify project variables using the formula Y = f (X1,..,Xn ). Describe the relationship between any dependent variable (Y) and independent variables (X). Explain how the shape, mean, and standard deviation characterize a process. Express the capability of a process in terms of a standard measure (z-value). Define hidden factories and how capability impacts cycle time. 4 Identify the key deliverables of the Measure phase of DMAIC.
18
D M A
19
5 Relate and apply the Quality Function Deployment (QFD) process to Six Sigma. Explain the purpose of QFD. Describe the phases of QFD. Explain QFD flowdown for product and service applications. Generate/build a House of Quality (Product Planning Chart). Identify what the customer wants (the whats). Identify the functions or processes that impact customer wants (the hows). Evaluate the impact of each function/process on customer wants. Calculate the overall magnitude of the impact each function/process has on customer wants (prioritize actions). Analyze and diagnose a completed House of Quality. Describe other QFD applications. Determine when QFD is appropriate to use. Recognize QFD pitfalls. Describe an example of QFD from GE Medical Systems. 6 Describe and define the deliverables of Step 2.
D M A
20
7 Define Performance Standards for a Six Sigma project. Describe the purpose and characteristics of a performance standard. Describe the purpose and characteristics of an operational definition. Define the term defect. Given an example of a problem or process, write an operational definition. Describe and distinguish between continuous and discrete data. Recognize the components of a performance standard, including product/process characteristic, measure, target value, specification limits, and defect. Given a CTQ type, identify performance standard sources and discrete/continuous data measurement methods. Given an example, define the measurable characteristic, determine whether it is continuous or discrete, determine the specification limit if applicable, identify a defect. 8 Describe and define the deliverables of Step 3.
D M A
21
9 Establish the accuracy of the measurement system and the data (Analyze the measurement system). Describe measurement as a process that includes Measurement, Analysis, Improvement, and Control phases. Describe measurement as a system that includes operators, gages, and environment. Define the terms (gage) resolution, precision, accuracy, and bias as used in Measurements System Analysis (MSA). Using the MSA checklist, document the existing measurement system. Recognize the sources of variation in a measurement system.
D M A
22
10 Conduct a Gage R&R study and analyze the results. Develop and implement a Data Collection Plan to collect Gage R&R study data. Describe equipment and appraiser sources of variation. Describe the total R&R variation in terms of Reproducibility (AV) and Repeatability (EV). Set up, collect, and enter data into a Minitab data sheet. Calculate both the appraiser variation (reproducibility) and equipment variation (repeatability). Describe the concepts of stability and linearity in gage studies. Compare R&R variation to the tolerance (specification window). Create graphs and charts (ANOVA method) to analyze study results. Recall and apply rules of thumb (guidelines) for analyzing R&R study results.
D M A
23
11 Describe and define the use of gage R& R for discrete data. Describe the use of the attribute R & R spreadsheet for discrete data. 12 Recall and list the deliverables of the Measure phase of DMAIC.
D M A
AnalyzeClassroom Learning Objectives 1 Perform hypothesis testing for a continuous Y and discrete X. Determine process stability with run charts and other tools Determine the data shape with histograms, normal probability plots, and Anderson-Darling tests. Select and use the appropriate tool to determine the p-value. Determine whether or not to accept the null hypothesis or the alternative hypothesis. 2 Perform hypothesis testing for a discrete Y and discrete X. Use chi-square testing to determine the goodness-of-fit and as a test of independence. Based on the chi-square test, determine whether or not to accept the null hypothesis or the alternative hypothesis.
24
D M A
AnalyzeClassroom Learning Objective (continued)
25
3 Perform hypothesis testing for a continuous Y and continuous X. Use a scatterplot to determine correlations between variables. Use a linear regression analysis to quantify correlations and predict values. Determine process stability with run charts and other tools. Determine whether or not to accept the null hypothesis or the alternative hypothesis. Describe the use of multiple regression for this type of data. Describe the implications of multiple regression in statistical analysis.
D M A
AnalyzeProject Learning Objectives

1 Use benchmarking to assist in developing project goals. Describe the purpose of benchmarking. Describe the uses of five different types of benchmarking, including competitive benchmarking, product benchmarking, process benchmarking, best practices benchmarking, strategic benchmarking, and parameter benchmarking. Apply benchmarking methodology to a variety of situations. List potential sources of benchmarking data and how to access such sources. Describe the advantages and disadvantages of Internal, Competitive, and Functional benchmarking and their relationship to Best Practices.
26
2 Set realistic and achievable defect reduction goals based on current baseline, GE guidelines, benchmarking results, and the process entitlement. Describe the methods used to set project goals. Use GE Standards for defect reduction in combination with benchmark results to determine project goals. Use the process entitlement to validate the achievability of the project goals.
D M A
AnalyzeTest Learning Objectives
27
3 Develop consensus within a project team on the acceptability of the project goals. Use a Cause & Effect (Fishbone) diagram to identify Xs that may impact the Y that is important in a project and provide a visual display of all possible causes of a specific problem. Recognize the purpose and benefits of a Cause & Effect diagram. Write a problem statement. Brainstorm categories appropriate to a problem. Recognize the 4 Ps: policies, procedures, people and plant. Brainstorm and analyze causes for each category to identify the most likely cause(s) of a problem. Determine which causes need to be verified with data.
D M A
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. Recognize the purpose and benefits of a Pareto Chart. Describe the Pareto Principle. Describe the steps involved in building a Pareto Chart: Collect data. Total results and arrange data in descending order. Draw and label a Pareto Chart. Analyze results. Compare before and after Pareto Chart to evaluate improvement effectiveness. 5 Describe process map analysis. Describe value added/ non-value added analysis.
D M A

6 Identify variation sources. Brainstorm a list of potential vital Xs. Use a histogram to aid in determining variation, center, and shape of a process. Use a dot plot to aid in determining variation, center, and shape of a process. Use a box plot to aid in determining variation, center, and shape of a process. Use a run chart to determine process stability. Define the terms population and sample and relate the two to each other. Use statistical tests to validate sampling techniques. Define the theoretical framework for hypothesis testing. Define and follow the hypothesis testing protocol. Define the terms null hypothesis and alternative hypothesis. Develop the null hypothesis for your project. Develop the alternative hypothesis for your project. Define type I and type II errors in relation to hypothesis testing. Define the relationship between the confidence interval and the p-value.
29
D M A

1 Describe and define the deliverables of Step 4.
30
2 Apply statistical principles of the Standard Normal Probability Distribution to predict the probability of a defect and process capability. Use continuous data to describe a process by its average, standard deviation, and normal curve. Define the term random variable. Interpret uniform, triangular, normal, and exponential distributions. Relate probability to distribution curves. Define the terms mean and standard deviation as they relates to a normal distribution curve. Recognize and distinguish between population and sample computational equations. Use Minitab to calculate a mean and standard deviation. Recognize the Descriptive Statistics tool as a method for validating calculations. Calculate statistical measures of variation, including range, deviation, sum-of-square, standard deviation, and coefficient of variation. Calculate capability (Z value). Perform basic statistic calculations using Minitab. Describe the purpose and characteristics of Descriptive Statistics tools, including Histogram, Dot Plot, Box and Whisker Plot, Run Chart. Be able to distinguish a normal distribution from other common non-normal distributions.
D M A
AnalyzeTest Learning Objectives (continued) 3 Characterize a process using discrete data. Define the terms unit (U), opportunity (OP), and defect (D). Recognize formulas for DPU, TOP, DPO, and DPMO. Use Z tables to convert DPMO to Z. Run and interpret a Minitab Product Report. Compare and contrast Classical Yield (Yc), Throughput Yield (YTP), and Rolled Yield (YRT). Calculation the distribution of defects for a given DPU. Calculate submitted, observed, and escaping defect levels. Recall DPU application rules. Determine how DPU controls Throughput Yield (YTP). Explain how complexity impacts quality.
31
D M A
AnalyzeTest Learning Objectives (continued)

4 Use Process Centering strategies to perform a capabilities analysis. Explain the concept of Process Centering. Distinguish between special (assignable) and common (random) cause variation. Choose rational subgroups for proper sampling and analysis. Differentiate between entitlement, short term process capability, and long term process capability. Interpret Minitab hand calculations, histogram, and box plots. Calculate the long and short term standard deviation and Z-values. Explain the general long term 1.5 Z shift. Use Minitab Six Sigma Process Report to obtain short and long term process capability measures - ZST, ZLT, ZbenchLT, Zshift, DPMOST, DPMOLT. Using capability measures and a 2x2 matrix, determine if there is a control problem or a technology problem.
32
D M A
AnalyzeTest Learning Objectives (continued)
33
5 Determine process capability. Define the term process entitlement as it relates to process capability. Define and provide examples of common cause and special cause variation. Define and use rational subgrouping of data. Define shift and drift of a process. Describe the components of variation. Calculate variation for a given process. Calculate the standard deviation for a process. Calculate process capability. Define the difference between long term and short term capability and their uses in a six sigma project. Define the use of the sum of squares and the standard deviation. Use the universal equation for Z to calculate Z scores. Define, derive, and use the Z-Bench for a process. Relate and convert between the Z score and Defects per Million Opportunities. Use data collected in the Measure phase to generate a process capability chart for a process. Interpret the results generated by the process capability report to determine the short term and long term process capability of a process. Use process capabilities to compare your process with a benchmark process. Determine whether the deficiencies in a process are due to control problems or technology problems.
D M A
AnalyzeTest Learning Objectives (continued) 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. Describe the use of the normality test. Describe the use of Moods median for non-normal data.
34
D M A
ImproveProject Learning Objectives 1 Characterize Xs as either operating parameters or critical elements. 2 Develop a strategy for those Xs identified as operating parameters. Develop a mathematical model of a proposed solution. Determine the best configuration or combination of Xs 3 Develop a strategy for those Xs identified as critical elements. Optimize process flow issues. Standardize the process. Develop a practical solution. Explain the needs and process to do screening experiments. The implications for this in DOE is described.
35
D M A
ImproveProject Learning Objectives 4 Perform optimizing experiments in order to develop a proposed solution. Identify factors for optimization experiments. Identify factor levels for optimizing experiments. Design optimizing experiment to include randomization and replication. Perform experiments and collect data. Analyze data with various tools including regression analysis.
36
D M A
ImproveProject Learning Objectives
37
5 Develop a proposed solution. Interpret the outputs of various tools to determine the optimum solution. Determine if the optimum solution will meet project goals. Present the proposed solution to management. Use statistical tolerancing to define the control mechanisms for implementation.
D M A
ImproveProject Learning Objectives
38
6 Establish operating tolerances. Describe the concept of tolerances and describe an example of this concept. Describe the use of simulation and the use of Crystal Ball. Describe and show an example of Crystal Ball. 7 Pilot the proposed solution.
D M A
ImproveTest Learning Objectives 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.
39
D M A
ControlProject Learning Objectives
40
1 Develop/modify and implement Quality Plans. Describe the purpose and characteristics of a Quality Plan. Recognize the components of a Quality Plan. Plan ongoing process controls, including monitoring and auditing strategies. Explain the benefits of monitoring as compared to First Article Inspection (FAI) and Information Management methods. Determine what to monitor for a given process. Determine the appropriate amount of monitoring data to collect, and how frequently the monitoring should occur. Recognize methods for detecting changes in a process. Recognize the steps that should be taken if a process change is detected. Explain the purpose of auditing. Describe guidelines for effective auditing. Compare and contrast manufacturing control methods. Explain the purpose and process of variable data charting (SPC). Explain the purpose and process of process management charting.
D M A

2 Develop and implement risk management strategies. Explain the value of a risk management process. Define the terms risk and risk management as they relate to DMAIC/Six Sigma. Determine when to use risk management. Recognize different types of risks. Recognize the steps involved in risk management, including identifying risks, rating risks, abating risks, and executing risk management plans. Recognize methods for identifying risks. Describe methods and tools for rating risks. Using the Probability of Occurrence Rating guide and Consequence of Occurrence/Risk Impact chart, prioritize risks according to risk factor score. Determine when and how to implement a Risk Abatement plan. Integrate lessons learned from prior risk management efforts. Describe the formal risk review process. Explain the criticality of tracking and executing risk abatement plans.
41
D M A
42
3 Develop and implement mistake proofing strategies. Recognize examples of mistake proofing. Describe principles underlying the process of mistake proofing. Recognize the difference between errors and defects. Explain how defects originate. Identify ten types of human error. Recognize human error-provoking conditions. Identify the three key mistake proofing techniques: shutdown, control, and warning. Distinguish between prediction/prevention and detection methods of mistake proofing. Recognize typical mistake proofing tools. Describe the 5 steps involved in mistake proofing, including identifying problems, prioritizing problems, finding the root cause, creating solutions, and measuring results. Recognize the advantages of mistake proofing as a proactive tool. Explain how mistake proofing fits into the Six Sigma process.
D M A
ControlProject Learning Objectives 4 Select and apply the appropriate Control Chart. Recall and explain SPC Concepts, including controlled variation, uncontrolled variation, common causes, and special causes. Recognize the five main uses of control charts. Distinguish between variable, attribute, and process focused control charts. Determine control limits. Distinguish between control limits and specification limits. Recognize the four states of a process.
43
D M A
ControlTest Learning Objectives

1 Describe and define the deliverables of Steps 10 and 11 in Six Sigma. 2 Describe and define the deliverables of Step 12 in Six Sigma. 3 Develop and implement Variable Control Charts. Describe the purpose of Statistical Process Control Charts. Given a control chart, recognize when a special cause is acting on a process. Recognize other types of variable control charts, including X Bar Chart, R Chart, Individuals Chart, and Moving Range Chart. State the five main uses of control charts. Describe data collection and sampling techniques. Establish and maintain control limits. Select the appropriate Variable Control Chart. Distinguish between control limits and specification limits. Determine whether a process is in control or out of control. Recognize Western Electric rules for identifying an out of control process. Recognize and apply Minitab rules. Build an Individuals and Moving Range chart. Use knowledge of the process to eliminate or reduce assignable/special causes.
44
D M A
ControlTest Learning Objectives 4 Develop and implement Attribute Control Charts. Describe the purpose of Attribute Control Charts. Define and relate the terms a defect and a defective. Recognize types of Attribute Control Charts, including CCharts, U-Charts, P-Charts, NP Charts. Select the appropriate Attribute Control Chart. Use Minitab to generate each type of Attribute Control Chart. Determine the appropriate Attribute Chart subgroup size.
45
D M A
Glossary And Minitab Primer 1 Define the typical terms used in Six Sigma methodologies. 2 Describe the use of the Minitab software program.
46
D M A
Introduction Exercise (60 Minutes)
Introductions
What Team Preparation
Desired Outcome: Introduction of Classmates and CD Rom Learnings How Go around your tables and introduce yourself to your table-team members (name and business only, at this time.) Discuss the importance of the DMAIC Step that your team has been assigned. Create a list of items that describe the deliverables for the step and why this step is important in the DMAIC cycle. Use your knowledge from your CD Rom Learnings. Close Exercise Report Out Who All Timing 2 mins.
Exercise
All
7 mins.
Choose a spokesperson to report out When asked by the instructor, each team member will introduce themselves: Name How long youve been with GE Your Title and what you do Previous Quality Experience What Phase your project is in After last team member has introduced himself/herself, the spokesperson will discuss the deliverables of the abovementioned exercise.
1 min.
1 min.
GE Capital, Inc., 2000 DMAIC GB C TX PG V 4.2.0
D M A
Classroom SessionObjectives
Share and get feedback on project deliverables for Define and Measure phases Review Define phase Review Measure phase Learn how to apply CTQ tools in the Measure phase of a project Learn how to use Minitab for statistical and graphical analysis during a project Learn knowledge and skills in Analyze phase and apply to Capital Logistics case Preview Improve phase Learn how to apply Design of Experiment (DOE) in the Improve phase of a project Preview Control phase Learn how to apply control charts in the Control phase of a project
D M A
Iterative Process
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 1 (Define). This phase defines the project. It identifies customer CTQs and ties them to business needs. Further, it defines a project charter and the business process bounded by the project. Phase 2 (Measure). This phase is concerned with selecting one or more product characteristics; i.e., dependent variables, mapping the respective process, making sure the measurement system is valid, making the necessary measurements, and recording the results. Phase 3 (Analyze). This phase entails estimating the short and long-term process capabilities and benchmarking the key product performance metrics. 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.
DMAIC GB C TX PG
Phase 4 (Improvement). This phase is usually initiated by selecting those product performance characteristics which must be improved to achieve the goal. Once this is done, the characteristics are diagnosed to reveal the major sources of variation. Next, the key process variables are identified by way of statistically designed experiments. For each process variable which proves to be significant, performance specifications are established. Phase 5 (Control). This phase is related to ensuring that the new process conditions are documented and monitored via statistical process control methods. After a settling in period, the process capability would be reassessed. Depending upon the outcomes of such a follow-up analysis, it may be necessary to revisit one or more of the preceding phases.
GE Capital, Inc., 2000 V 4.2.0
D M A
Statistical Thinking
D M Practical Problem
Problem statement Project Y Magnitude Impact
A Statistical Problem Statistical Solution

I C Practical Solution
Verify critical Xs and (x) Change process Control the gains Risk analysis Control plans
Characterize the Root cause process analysis Stability Critical Xs Shape Measure the influence of the Center critical Xs on Variation Data Integrity the mean and variability MSA Capability Test Brainstorm ZBench ST & LT potential Xs Model Sampling plan Estimate
Collect data
The Practical-To-Statistical-To-Practical Transformation Process
D M A
The 12-Step Process
Step Define A B C Measure 1 2 3
Description Identify Project CTQs Develop Team Charter Define Process Map Select CTQ Characteristics Define Performance Standards Measurement System Analysis
Focus Tools
Deliverables Project CTQs Approved Charter High Level Process Map
Y Y Y
Customer, QFD, FMEA Project Y Customer, Blueprints Performance Standard for Project Y Continuous Gage R&R, Data Collection Plan & MSA test/Retest, Attribute Data for Project Y R&R Capability Indices Team, Benchmarking Process Analysis, Graphical Analysis, Hypothesis Tests DOE-Screening Factorial Designs Process Capability for Project Y Improvement Goal for Project Y Prioritized List of all Xs
Analyze 4 Establish Process Capabilities 5 Define Performance Objectives 6 Identify Variation Sources
Y Y X
Improve 7 Screen Potential Causes 8 Discover Variable Relationships 9 Establish Operating Tolerances Control 10 Define & Validate Measurement System on Xs in Actual Application 11 Determine Process Capability 12 Implement Process Control
X X
List of Vital Few Xs Proposed Solution Piloted Solution
Y, X Simulation
Y, X Continuous Gage R&R, MSA Test/Retest, Attribute R&R Y, X Capability Indices Process Capability Y, X 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
The Basics For Effective Meetings
DMAIC is a powerful methodology which will allow teams to significantly improve the processes on which they are working. However, in order for teams to efficiently use the tools and techniques associated with DMAIC, they must be able to work together effectively. Although not the focus of this class, we will briefly focus on the minimum requirements for effective meetings. Further training in effective meeting skills is provided in Facilitating Teams through Change Projects Training
D M A
Meeting Skills The Basics For Effective Meetings Roles:

Leader or Facilitator Timekeeper Scribe Note Taker
Tools:
Agenda Desired Outcomes Ground Rules Decision-Making Process Brainstorming
Define These Roles For A Meeting Or Any Team During this training, you will practice these roles so that you can use them in your project team meetings to improve meeting effectiveness. 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.
Rules Of Brainstorming:

No judgment of ideas Record all ideas No discussion of ideas during brainstorming
D M A
Practice Meeting (5 minutes)

Desired Outcome: Name your team What Team Preparation How Choose a facilitator, timekeeper and scribe Brainstorm a list of possible team names Using N/3*, prioritize your brainstormed list Identify your top name Who All Timing 30 secs.
Determine Team Name
Facilitator
4 mins.
Close Exercise
Choose a spokesperson to report out
All
30 secs.
* N/3 is a prioritization technique used to narrow a large list down to the top priorities. The number of ideas (N) is divided by 3 (N/3) and team members get that number of votes to identify their top choices. All votes should not be put on one choice.
D M A
Define Review Module Objectives
Define Review Objectives
Review CAP Tools Review steps in Define phase Review and get feedback on Green Belt project deliverables for Define phase
Review Steps in Define phase and get feedback on Greenbelt project deliverables for Define phase
Review
GE Capital, Inc., 2000 DMAIC GB D TX PG V 4.2.0
D M A
First Things First (10 minutes)

Desired Outcome: Learning the components for successful change What Team Preparation How Think of a change initiative that you have experienced in the past in which the change was unsuccessful. Why did the change initiative fail? Who Each Team Member Timing 5 mins.
Prepare List
On a flip chart, list your groups ideas for why change initiatives fail. What are the common themes? Be prepared to report findings.
Facilitator
5 mins.
D M A
Change Acceleration Process (CAP)
Techn ical St ra
te g y
rategy
nal St o i t a z i an al/Org Cultur
Change Initiative (Target)
CAP Complements Technical Strategy With Cultural Tools To Achieve The Change Initiative
Change can be depicted via the equation: Q x A = E Q = Quality of solution to be implemented A = Acceptance of solution to be implemented E = Effectiveness of the implemented solution It is as important to have a strategy for developing acceptance as it is to have a plan for implementing the solution. A Technical Strategy and a Cultural/Organizational Strategy are both necessary in order to effectively achieve your change target. CAP gives us tools to develop the Cultural/Organizational or Influence Strategy.
D M A
Cap Model
Leading Change
Creating Creating A A Shared Shared Need Need Shaping A Vision Mobilizing Mobilizing Commitment Commitment
Current State Transition State Improved 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 sponsors the change if they are to be successful.
Once change is started, it endures and flourishes, and learnings are transferred throughout the organization.
Creating A Shared Need

Monitoring Progress
The reason to change, whether driven by threat or opportunity, is instilled within the organization and widely shared through data, demonstration, demand, or diagnosis. The need for change must exceed the resistance to change.
Progress is real; benchmarks are set and realized; indicators are established to guarantee accountability.
Changing Systems And Structures

Shaping A Vision
The desired outcome of change is clear, legitimate, widely understood, and shared.
Management practices are used to complement and reinforce change.
Mobilizing Commitment
Did the themes that youve developed in your last exercise align with our CAP Model? This CAP Model helps us to focus on the organizational changes aspect of projects.
There is a strong commitment from key constituents to invest in the change, make it work, and demand and receive management attention.
D M A
Cap Tools And Six Sigma
Creating A Shared Need Define

SIPOC Threat vs.
Opportunity Matrix Project Scope Contract
Shaping A Vision
Mobilizing Commitment
Making Change Last
Monitoring Progress
In Frame / Out Threat vs.

Of Frame Opportunity Matrix
GRPI ARMI
Measure
Analyze
Improve
Making Change
Last Checklist Control Pans Measures And Rewards
Making Change
Last Checklist
Control Pans Measures And

Rewards
Control
The above-mentioned tools are covered in the Tools Section of the Self-Paced Workbook. For further training, you can attend the week-long course sponsored by Learning Services.
D M A
Homework
Desired Outcome: Identify Cap Tools which you will use in your project What Prepare a plan for CAP Tools How Identify at least one CAP Tool that you have used in your project thus far If you have not used any CAP Tools, then identify a CAP Tool that you could, in retrospect, have used For tomorrow, be prepared to share with the class which tool you used, how you used it, and how it helped you with your project Who Each Student Timing As Homework
D M A
Define Phase Flowchart
A
DEFINE PHASE OVERVIEW
Define A: Identify Project CTQs
Define B: Develop Team Charter
Define C: Define Process Map
D M A
Define Phase Overview

What is the Define phase? The Define phase is when your team identifies:
Who your customers are and what their requirements are for your products and services The reason for doing the project and project boundaries The project team members and how they will work together What process you are trying to improve and what the process map looks like
Why is the Define phase important? This phase is important because it clearly and precisely describes the goals of the project, aligns the project with organizational priorities and lays the groundwork that will allow the team to remain focused. Steps involved in the Define phase: Define A: Identify Project CTQs Define B: Develop Team Charter Define C: Define Process Map
D M A
Define AIdentify Project CTQs

What does it mean to Identify Project CTQs? Critical to Quality Characteristics, CTQs, 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 CTQs of a process or product. Why is it important to Identify Project CTQs? Project CTQs 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 CTQs 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 CTQs
DEFINE STEP OVERVIEW
A.1 Identify Customer A.2 Collect Voice of the Customer data to Identify Project CTQs A.3 Build a process/product drill-down tree to Identify Project CTQs
D M A
Define BDefine Team Charter
10
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
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
D M A
Define CDefine 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 highlevel 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
C.1 Develop high-level Process Map
D M A
Elevator SpeechHomework Activity*

Desired Outcome: Share elevator speech for the project with members of your team What Preparation Elevator Speech How Develop an elevator speech for your project. Each team member shares the elevator speech for their project. (See the description of an elevator speech on the following page.) Who All Each Team Member
12
Feedback
Make a note of any feedback you receive from your class members. Did your elevator speech provide a clear understanding of your project?
All
Elevator speeches will be shared with the entire class throughout the 3-day workshop. * For additional information on elevator speeches, see the Self-Paced Workbook, pp 197-198
D M A
Elevator Speech Here is what our project is about (describe the problem or issue) Here is why its important to you (describe the benefit of doing the project) Here is what success will look like (describe the goal of the project and where we currently are)
13
What is it? This is a tool that helps the team members put together a short sell pitch for the project. Being able to clearly and simply state the need for change and describe the future state is essential for rallying the support and commitment of key constituents. The metaphor of the elevator is useful in challenging the team to be clear, precise and simple. Imagine a chance meeting of a team member and a key stakeholder in an empty elevator with a 90-second ride. 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?
Why use it? It helps the team link the need for change with the vision of the future. All the team members will be using a common sell pitch. For teams that have thoroughly debated and documented both the need and vision, this is the synthesis event whereby they distill the essence of the project. For teams struggling to get started in terms of need and vision, it can be the place to begin to bring some focus to the teams more rambling discussion of need and/or vision. The elevator speech should be able to deal with the questions that will arise once the project is announced to the broader constituent base.
D M A
DMAIC ChecklistDefine A
A.1 Identify customer
Have you considered both internal and external customers? Have you focused on the most important customer segment(s)?
14
A.2 Collect voice of the customer data to identify customer CTQs

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

Is your project CTQ focused on whats most important to the customer? If you have multiple CTQs, have you used the appropriate tool to prioritize them?
D M A
DMAIC ChecklistDefine B
B.1 Define the business case
Is the business case compelling to the team? Does this business case illustrate why this project needs to be done now ? Have I considered intellectual infringements?
15
B.2 Develop problem statement

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

Is your goal statement SMART (Specific, Measurable, Attainable, Relevant, Timebound)? Does your goal statement focus on what needs to be improved, and without predetermining a solution? How will you know the team is successful?
D M A
DMAIC ChecklistDefine B (continued)

B.4 Project scope
Does your team agree on the project scope? Is your project scope within the teams control?
16
B.5 Roles
Are all functions represented? Does each team member fully represent their managers input on the project?
D M A
DMAIC ChecklistDefine C
C.1 Develop high-level process map
Does the COPIS scope (start and stop) match the scope definition in your team charter? Have you mapped the process from your customer's perspective? Is this the As-Is process? Was the map validated with the people who actually do the process
17
D M A
Forms Of Intellectual Property
18
Patent Best Protection against independent development by others Gives right to exclude others from making, using selling, the invention of 20 years from filing Can wait 2-3 years for patent to issue Right granted by the government for new, useful, non-obvious inventions Trade Secret Rights exist only so long as actually kept secret Protects against theft or misappropriation, not independent development Does not require inventiveness in the patent sense Copyright Protects original works of authorship fixed in a tangible medium Does not protect against independent development Trademark Defines sources of goods or services Arbitrary, fanciful marks get stronger protection
D M A
Intellectual Property Assessment in Chartering Phase Throughout the project consider implications to Intellectual Property Policies and Procedures through the following: Commercial benefit for licensingconsider this in the cost/benefit Infringement AvoidanceGE has a policy covering avoiding infringing on others IP-Assure no infringement Identify and Capture IPThroughout 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
19
D M A
D M A
C
1
Measure Module Objectives Review steps in Measure phase
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
C
2
Measure Phase Flowchart
A
MEASURE PHASE OVERVIEW
Measure 1: Select CTQ Characteristic
Measure 2: Define Performance Standards
Measure 3: Establish Data Collection Plan, Validate Measurement System, & Collect Data
D M A
C
3
The12-Step Process
Step Define A B C Measure 1 2 3 Select CTQ Characteristics Define Performance Standards Measurement System Analysis Y Y Y Description Identify Project CTQs Develop Team Charter Define Process Map Focus Tools Deliverables Project CTQs Approved Charter
High Level Process Map Customer, QFD, FMEA Project Y Customer, Blueprints Performance Standard for Project Y Continuous Gage R&R, Data Collection Plan & MSA test/Retest, Attribute Data for Project Y R&R Capability Indices Team, Benchmarking Process Analysis, Graphical Analysis, Hypothesis Tests DOE-Screening Factorial Designs Process Capability for Project Y Improvement Goal for Project Y Prioritized List of all Xs
Y Y X
X X
Y, X Simulation
DMAIC GB E TX PG
V 4.2.0
D M A
Measure Phase Overview

What is the Measure phase? This phase is concerned with selecting one or more product characteristics to measure, defining how the characteristics will be measured, planning data collection, and collecting data.
Why is the Measure phase important? This phase is important because it ensures that accurate and reliable data is collected to measure current process performance related to the customer CTQ. Steps involved in the Measure phase: Measure 1: Select CTQ Characteristic Measure 2: Define Performance Standards Measure 3: Establish Data Collection Plan, Validate Measurement System & Collect Data
D M A
C
5
Mapping Tools To Generation Of CTQ Components

Output Output Unit Unit
C-O-P-I-S C-O-P-I-S Define A
Output Output Characteristic Characteristic
Define A C-O-P-I-S C-O-P-I-S
High Level Need (VOC)
Project Project Y Y Operational Operational Definition Definition
Measure 1 QFD QFD C&E C&E C-O-P-I-S C-O-P-I-S
CTQ CTQ
Project Project Y Y Measure Measure
QFD QFD Process ProcessMap Map C&E C&E Pareto Pareto
Measure 1
Specification Specification Limits Limits
Measure 2 VOC VOC QFD QFD
Target Target
QFD QFD Pareto Pareto VOC VOC
Measure 2
Defect Defect
FMEA FMEA C&E C&E
Measure 2
# #of ofDefect Defect Opportunities Opportunities Per Per Unit Unit
FMEA FMEA Process ProcessMap Map
Measure 2
D M A
C
6
A Statistical Problem
Characterize the process Stability Shape Center Variation
Statistical Solution

Data Integrity MSA Capability Brainstorm ZBench ST & LT potential Xs Sampling plan
Collect data
D M A
C
1
Measure 1Select 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 3: Establish Data Collection, Validate 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
C
2

C-O-P-I-S C-O-P-I-S Define A
Define A
C-O-P-I-S C-O-P-I-S
High Level Need (VOC)
Measure 1 QFD QFD C&E C&E C-O-P-I-S C-O-P-I-S
CTQ CTQ
QFD QFD Process ProcessMap Map C&E C&E Pareto Pareto
Measure 1
Measure 2 VOC VOC QFD QFD
Target Target
QFD QFD Pareto Pareto VOC VOC
Measure 2
Defect Defect
Measure 2 FMEA FMEA C&E C&E
Measure 2 FMEA FMEA Process ProcessMap Map
GE Capital, Inc., 2000 DMAIC GB F TX PG V 4.2.0
D M A
C
1
CTQ Tools Module Objectives Learn how to apply a variety of CTQ tools to help prioritize & define CTQ elements to be improved If you dont 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: FMEAs 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
C
2
Process Mapping
D M A
C
3
Process MappingCTQ Tool #1
A Graphical Representation of Steps, Events, Operations, and Relationships of Resources Within a Process
Applications

Techniques for examining a process to determine where and why major breakdowns occur 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.
Enables group to see the entire process as a team Magnifies normally overlooked areas and displays their relevancy
D M A
C
4
Levels Of A Process
Core
Acquire New Business
Obtain Request to Buy Service
Conduct Underwriting
Prepare Contract
Initialize Customer Service
COPIS/SIPOC
S
Underwriters
Establish Terms
Prepare Docs
Negotiate Contract
Close Deal
C
Customer/ Customer Service
Detailed Subprocess Map

Tasks Procedures
Earlier in Define, you developed a high-level or COPIS process map. By looking at a process from a big picture perspective, you evaluated customer needs and supplier inputs, and determined initial measurement objectives. Now, you will look in more detail at the subprocesses defined in the COPIS map. Subprocess maps provide specifics on the process flow that you can then analyze using several useful techniques. Choose what to subprocess map by determining which of the major steps in the COPIS have the biggest impact on the output (Ys). The block (or blocks) selected is the one on which you create a subprocess mapusing 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?
If the Project Y is a cost measure, which of the blocks adds the most cost? If the Project Y is a function measure, which block has the most errors or problems? Like working with a puzzle, you begin to assemble the pieces of an area on which it makes sense to focus our efforts.
D M A
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 processdocumenting the flow of the process. But, there are at least four major versions of a process map that you can draw. First, you can document what individuals who touch the process think it is. Certainly with the daily contact, you would expect people to know how the process works. But many people can easily explain how things work when things go right. You need to know how the process works 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.
Reconciling the map requires your team to observe the processmaking sure you have recorded all the existing steps. As the team moves forward and does process analysis and problem-solving, eventually you will move toward the third version of the process mapthe should be map created as part of their Improvement plan. At this point, a check must be made as to whether customer CTQs have been met or exceeded.
D M A
C
6
Outside-In Focus Does your process overlap with the customers? Should the customers process be mapped? Should you partner with the customer to assess the processes?
Continue to keep focused on the customers perspective.
D M A
C
7
Subprocess Mapping Techniques Process flowchart Alternate path method Deployment or cross-functional map/flowchart
There are a number of ways to draw a subprocess map. Your team can choose the approach or approaches most appropriate (or comfortable) for your use. Whatever type of subprocess map is chosen, make sure that it accurately reflects your process as it currently exists.
D M A
C
8
Process Mapping Symbols
These symbols represent the standard symbols for process mapping. Most projects use the simplified symbol sets shown on the following page.
D M A
C
9
Process Flowchart
Customer Calls
Automated System Answers
TouchTone Phone
No
Yes
Customer Chooses Routing Option
Call Placed In Criteria Que, On Met? Hold
Customer Prepare Waits? Loan Offer
Yes
Rep Answers Phone
Call Gets Routed To Voice-Activated System
No
Call Ends
Key:
Start End
Task
Review Or Decision
Direction
The business process map most people are familiar with uses four simple symbols: oval, rectangle, diamond, and arrow. It is generally used when the process is fairly small and simple, or when documenting the work done by a single person or group.
D M A
C
10
Alternate Path Flowchart
Customer Calls
Automated System Answers
TouchTone Phone
15 No
85 Yes
Customer Chooses Routing Option
Call Placed In Criteria Que, On Met? Hold
Customer Prepare Waits? Loan Offer
80 Yes
Rep Answers Phone
Call Gets Routed To Voice-Activated System
20 No
Call Ends
Key: 60 40 Task Direction Percentages Decision
The alternate path process map method arose from reengineering efforts, where the mapping of very large processes made decision diamonds more of a hindrance than a help. In this technique, diverging or alternate paths are noted by split arrows. Teams can then note relative percentages of times/incidences the process follows each path. Process mapping software tools also make it easier to add icons showing the tools or methods employed at different points in the process. Showing the alternate path allows us to identify sources of variation which maybe adding complexity to our process.
D M A
C
11
Deployment Or Cross-Functional Flowchart
Customer
Negotiate
Sign Contract
File Contract
Contract Admin.
Initiate Call
Negotiate
Revise Contract
Revise Contract
File Contract
Attorney
Review Contract
Review Contract
Dealer
Write Contract
File Contract
Who
PROCESS FLOW
This approach to process mapping emphasizes who performs which tasks and which steps are done concurrently. It provides a clear visual perspective of the hand-offs and relationships between groups involved in the process. (Note: Either of the mapping techniques described on the previous pages can be placed on this type of map.) Always include the process customer and key suppliers as bands on this type of map. (Because customers are visible at the top, some people refer to this as a Service Blueprint.)
D M A
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: 1. Identify the participants involved in the process, write their names on cards, and pin or tape the cards in a vertical column on the wall with the customer at the top of the column. 2. Identify the action that initiates the process, write it on a card, and place it next to the appropriate participant card. In many instances,the process is 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 customers action.
3. Next, identify who receives the output of the second step and what activity they perform. Write this activity on a card and place it to the right of the second step on the same row as the participant performing the activity. 4. Identify who receives the output of this activity and what activities that participant performs. 5. Continue to create cards and position them in the same manner until the process is complete (i.e., the customer receives the final process output).
D M A
C
13
Tips In Subprocess Mapping Involve people who know (focus on) the as is Clarify process boundaries Brainstorm stepswrite on Post-its
Use verb-noun format (e.g., Prepare contract, not contracting ) Dont include Who in step description
Combine, eliminate duplicates, clarify steps Organize steps into proper flow and add arrows

Respect the boundaries Dont start problem-solving Validate and refine before analyzing
Here are some guidelines on building a subprocess map. These are not absolutebut they should help you avoid some of the pitfalls of process mapping.
Combine and ClarifyMake sure brainstormed steps are clear and dont overlap. Organize in FlowCreating the map is last. With all steps visible, its typically much easier to create a meaningful map without getting stuck on one or two minor issues. Is the customer involved in your subprocess step? If so, how?
Focus on As IsTo find out why problems are occurring in a process, you need to concentrate on how its working now. Clarify BoundariesIf youre working from a welldone high-level map, this should be easy. If not, youll need to clarify start and stop points. Brainstorm StepsIts usually much easier to identify the steps before you try to build the map. Starting each step description with a verb (e.g., Collate Orders; Review Credit Data) helps you focus on action in the process. Who does the step is best left in parentheses (or left out) you want to avoid equating a person with the process step.
D M A
C
14
Verifying CTQ Elements For Process Mapping Using Process Mapping in the Measure phase allows us to:
Select an upstream step that is generating defects Determine where to collect the data Define the output measure (a good place to begin looking is a No Exit out of Decision Diamonds)
D M A
C
15
Subprocess MappingBreakout Activity (20 minutes)

Desired Outcome: Practice creating a subprocess map What Team Preparation How Choose a facilitator, timekeeper, scribe and/or note taker Prepare the subprocess map for your own project Brainstorm the challenges of subprocess mapping on your own projects Who All Timing 30 secs.
Prepare Subprocess Map
All
19 mins.
Close Exercise
Choose a subprocess to report your results
All
30 secs.
D M A
C
16
Cause & Effect DiagramsCTQ 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
People
Measurement
Mother Nature
Procedure: 1. Draw a blank diagram on a flip chart. 2. Define your problem statement. 3. Label branches with categories appropriate to your problem. The categories shown are often used, but any categories can be used. 4. Brainstorm possible causes and attach them to appropriate branch. 5. For each cause ask, Why does this happen? 6. Analyze results; do any causes repeat?
7. As a team, determine the three to five most likely causes. 8. Determine which likely causes you will need to verify with data. Alternative procedure:

Brainstorm causes independent of categories. Use the Affinity Diagram and brainstorm items into categories.
D M A
C
17
Cause & Effect DiagramsExample

Cause & Effect for Capital Logistics
Example Of A CTQ Measurements
Dock Clock Watches Computer Tracking
People
Capital Truck Capital Driver or 3rd Party Carrier or Capital Truck Lease Driver New employees
Weight of freight Temperature Distance Overnight Delivery Precipitation Unload Delivery Drop off & Hook up
52' trailers 48' trailers Volvo Trucks Mack Trucks Freightliner Trucks
Why are the deliveries not timely?
Environment
Methods
Machines
The C&E answers the question of What am I going to measure? I can now focus on weight of freight or distance or temperature, etc. (This is the output characteristic.)
D M A
C
18
FMEACTQ Tool #3 An FMEA is a structured approach to:

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

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

FMEAs will be covered in detail during the Improve phase of the CD Rom. This section is a brief introduction.
D M A
C
19
Potential Failure Modes and Effects Analysis

Worksheet
Process/Product: Invoicing FMEA Team: Invoice Process Mgnt. Team Black Belt: E. Jones FMEA Process Occurrence Item/Process Potential Step Failure Mode Enter Amt Owed Potential Effect(s) Of Failure Overbill Potential Cause(s) Of Failure Wrong Ct Wrong Price Wrong Ct Wrong Price Missing No Payment 6 Sale Error De Error De Error Sale Error De Error De Error Sale Too Busy System Down Sales Busy System Down Current Controls Responsibility Recommended And Target Completion Action Date E. Jones 5/15/97 Detection Severity RPN FMEA Date: (Original) (Revised) Page: 10/15/96 5/25/97 1 of 1 Action Results Occurrence Detection Severity Action Taken
Inaccurate
8 5 8 5 5 7 5 7 7 3 7 3
Rare Rare Reviewed Reviewed Measured Measured
9 9 9 9 3 3 3 3 4 4 4 4
Underbill
576 DMAIC Team 360 to Investigate Root Causes 360 of Count & 225 Price Accessories 90 126 45 63 84 36 168 72
All DMAIC Tasks Complete
8 8 5 5 6 6 3 3 3 3 6 6
4 3 1 1 5 7 5 7 7 3 7 3
2 64 2 48 2 10 3 15 3 90 3 126 3 45 3 63 4 84 4 36 4 168 4 72
Delay
Delayed
Late Bill
No Bill
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.
RPN
D M A
C
20
Failure Mode And Effects Analysis How it Works

Function Part/Process
Failure Mode
Effects Severity (1-10) Causes Occurrence (1-10)
Controls Detectability (1-10)
RPN RPN R isk P riority Risk Priority N Number umber RPN = S x O x D = RPN = S x O x D = 1 1 to to 1000 1000

Review the product, service, or process Determine failure modes List one or more potential effects for each failure mode. Answer the question: If the failure occurs, what are the consequences? Identify potential causes List current controls

Assign Severity, Occurrence, and Detection Factors Calculate RPN.Prioritize RPNs from high to low. Identify the top issues, those with high RPNs. Determine preventive or remedial actions, especially for high-priority issues. Recalculate RPN after actions have been implemented

D M A
C
21
FMEA: Calculating Risk Priority Number (RPN) RPN = Severity x Occurrence x Detection
Severity Scale
Rating Criteria A Failure Could: Injure a customer or employee Be illegal Render product or service unfit for use Cause extreme customer dissatisfaction Result in partial malfunction Cause a loss of performance which is likely to result in a complaint Cause minor performance loss Cause a minor nuisance, but be overcome with no performance loss Be unnoticed and have only minor effect on performance Be unnoticed and not affect the performance
Bad
10 9 8 7 6 5 4 3
Occurrence Scale
Rating 10 9 8 7 6 5 4 3 2 1 Time Period More than once per day Once every 3-4 days Once per week Once per month Once every 3 months Once every 6 months Once per year Once every 1-3 years Once every 3-6 years Once every 6-100 years Probability > 30% 30% 5% 1% .03% 1 Per 10,000 6 Per 100,000 6 Per Million 3 Per 10 Million 2 Per Billion
Good
2 1
Detection Scale
Rating Definition 10 Defect caused by failure is not detectable 9 8 7 6 5 4 3 2 1 Occasional units are checked for defect Units are systematically sampled and inspected All units are manually inspected Units are manually inspected with mistake-proofing modifications Process is monitored (SPC) and manually inspected SPC is used with an immediate reaction to out-ofcontrol conditions SPC as above with 100% inspection surrounding out of-control conditions All units are automatically inspected Defect is obvious and can be kept from affecting the customer
Total RPN is determined by multiplying all of the individual scores. Note: Occurrence, detection and severity are independent estimates of risk.
D M A
C
22
Using A FMEA In The Measure Phase Helps to prioritize Project Ys Helps to identify where and how a process may fail to meet a CTQ (defects)
In this process, you may uncover an issue where you need to contain a problem (e.g., regulatory, non compliance, legal). If this occurs, it raises the priority.
D M A
C
23
Pareto ChartsCTQ Tool #4 Is there a defect that occurs frequently?
Segment data to look for a significant factor that influences the process
Frequency
Category of Defect
The Pareto chart is a bar chart with the bars (segmentation levels) arranged in descending order. It is an essential tool to help prioritize improvement targets by identifying the 20% of the problems that cause 80% of the poor performance (Pareto principle).
D M A
C
24
Pareto Charts (continued) Example

200 180 160
f* of D+B+F *f = frequency April 1 June 30 Number Of Units Investigated: 5,000
100 90 80 70 60 Cumulative Percentage

Cumulative Summation Line (Cum Sum line)
140 120 Frequency 100 80 60 40 20 0

Defect Type # Defects % Defects Cum %
f* of D+B
f* of D
50 40 30 20 10
LEGEND A: Illegible B: Bank Info Incomplete C: Missing Signature D: Personal Information Incomplete E: Employment History Incomplete F: Address Incomplete
D 105 50.5 50.5
B 40 19.2 69.7
F 20 9.6 79,3
A 10 4.8 84.1
C 5 2.4 86.5
E 3 1.5 88.0
Other* 25 12.0 100.0
Total 208 100
Approximately 80% of Defects from Defects D + B + F

A fully documented Pareto chart will include the Cumulative Summation line, or Cum Sum line, which depicts the running total of the frequency of each subsequent bar (segmentation factor). The right-hand axis on the graph will show the cumulative percent of defects. By reading the Cum Sum line against the cumulative percentage, your team can determine which of the segmentation levels comprise 80% of the total for the problem, and direct their attention to those levels. In this example, I would likely focus on the D segment, and this would become my project Y. You should also create a Pareto showing $s. This may cause your focus to change. * Other- Try to keep this category as small as possible. Find out what is in other.
D M A
C
25
Using Pareto Charts In The Measure Phase Quantitative tool use to determine segmented areas of focus Graphically displays most frequent occurrence of outcomes (little ys) Helps you get from Big Y to little y
D M A
C
26
QFDCTQ Tool #5 Definition of QFD (Quality Function Deployment)
Structured Methodology to Identify and Translate Customer Needs and Wants Into Measurable Features and Characteristics of a Product or Service
D M A
C
27
QFD FlowdownProduct Application
Functional Requirements (HOWs) Customer Requirements House of Quality #1

(WHATs)
Part Characteristics (HOWs) Functional Requirements

(WHATs)
GE Processes (HOWs) Part Characteristics

(WHATs)
GE Processes
House of Quality #2
Process Variables (HOWs)

(WHATs)
Y Key Functional Requirements
House of Quality #3
Y Key Part Characteristics
House of Quality #4 X
Key Manufacturing Processes
Key Process Variables
Product application of QFD links functional requirements to key process variables.
D M A
C
28
QFD FlowdownServices Application
Service Requirements (HOWs)
Service Functions/Processes (HOWs) Process Controls (HOWs) Service Requirements (WHATs) Service Functions/Proceses (WHATs) House of Quality #2
Customer Wants (WHATs)
House of Quality #1
Y Critical-to-Quality Characteristics (CTQs)
House of Quality #3
Key Service Processes
X Key Process Variables
Services application of QFD links project deliverables to key process tasks.
D M A
C
29
House Of Quality Summary

1b. 1b. Customer Customer Importance Importance How How important important are arethe thecustomer customer wants wants to to the the customer? customer? Roof Roof Correlation Correlation Conflict Conflict Resolution Resolution 3.Characteristics/measures 3.Characteristics/measures
Target Target Direction Direction

Current Rating Competitor #1 Competitor #2
HOW'S HOW'S
1a. 1a. Customer Customer Needs Needs
How How do do you you satisfy satisfythe thewants wants? ?
WHAT'S WHAT'S
H M L L L
M
2. 2. Competitive Competitive Assessment Assessment
What Whatdoes doesthe the customer customer want? want? Voice Voiceof of the the Customer Customer 4. Relationships 4. Relationships H H Strong Strong 99 M Medium 3 M Medium 3 LL Weak Weak 11
H M
M H
L M M
Where Whereare arewe weand and competitors competitorsrelative relativeto to customer customerimportance? importance?
Importance ImportanceRatings Ratings 6. 6. Target Target values values of of HOW's HOW's (units) (units) 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 quality you build. Room 1a Customer Needs (in their language) High-level wants identified by the customer Room 1b Customer Importance Customer ranking of the wants. A typical score is 1-5. 1least important 5most important Room 2 Competitive Components Customers view of how GE compares with the competition Room 3 Characteristics/Measures Measurable attributes that can be used to indicate how well the customers needs are met. Room 4 Relationships Strengths of interrelation between the WHATS and the HOWs. HStrong (9) MMedium (3) LWeak (1)
DMAIC GB G TX PG
To what extent will a HOW impact a WHAT? Using a 9, 3, and 1 scale forces a wider spread between the most important and least important items. Room 5 Competitive Benchmarks How the competition performs relative to the measures/characteristics. Room 6 Target and Specifications Target values for the HOWS. Setting measurable targets allows the team to define what is required to achieve customer satisfaction. Roof Relationship Impact of the HOWS on each other. ++ Strong positive + Positive - Negative -- Strong Negative Importance Ratings (IR) IR = (Customer Importance X Strength Of Relationship)
D M A
House Of Quality ExampleLunch Selection

Relationship High Positive Positive High Negative Negative
30
O = HOLD CONSTANT
Lunch QFD
Number of dishes used
Weight of portion
PB&J Sandwich
Time to prepare
Customer Needs
Cost of ingredients
Number of measured ingredients
Percent carbohydrate provided
Percent nutrition provided
Product Requirement
Instant Soup 3 3 3 1 3 2 3 3
Importance
Direction for improvement

Fills us up Is nutritious Tastes good Is easy to make Is easy to clean up Sticks with us Is inexpensive Is clean Importance Rating Target (Obtained from VOC) PB&J Instant Soup Fast Food Units Of Measure (Obtained from VOC) 5 4 3 4 2 4 1 2 H M M H M L H H M H L L M L M M H H M H L L 2 3 2 2 2 2 4 2 4 2 4 4 5 4 1 4
94* 52 59 48 60 10 18 16 33 25 2 8 33 25 15 10 3 3 5 1 0.50 0 1 0.25 0 3 0.50 1 0 3.00 0 $ Key H=9 M=3 L=1
16 40 73 10 oz % % min
*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 CTQs 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
Fast Food
D M A
C
31
Analyzing A House Of Quality

Empty Empty columns columns in in Room Room 3 3 -- Perhaps Perhaps an an unnecessary unnecessary measure measure characteristic characteristic signaling signaling it it does does not not affect affect customer customer wants wants Strong Strong negative negative relationship relationship in Identify in roof roof Identify trade-offs trade-offs
Current Rating
Competitor #1
H M L L L
L M H M L
Highest Highest score score on on competitive competitive comparison comparison -- Able Able to to lead lead the the market market with with existing existing product/service product/service
Empty rows in Room 33 -Empty rows in Room Unaddressed customer Unaddressed customer want would be aa major want would be major problem problem
Low Low score score on on competitive competitive comparison, comparison, but but high high score score on on competitive competitive benchmarks benchmarks -Market Market technical technical advantages advantages to to improve improve customer customer perception perception Low Low competitive competitive benchmarks benchmarks -- Poor Poor long-term long-term market market performance performance
Competitor #2
D M A
C
Measure 1Select CTQ Characteristics
CTQ And Performance Standard Worksheet

Heres our next step: Output Output Unit Unit
A A Lunch LunchPortion Portion
32
Define A
Define A
Healthy HealthyLunch Lunch
High-Level Need (VOC) Prepared Prepared Lunch Lunch
Measure 1
Total TotalWeight Weight In InOunces OuncesPer PerServing Serving
Measure 1
CTQ
Specification Specification Limits* Limits*
Weight Weight
LSL=14 LSL=14oz. oz. USL=18 USL=18oz. oz.
Measure 2
Measure 2
Target* Target*
16 16oz. oz.
You may not know Spec, Target or Defect at this time (in Step 1). If you know it now, write it down. This information will come from the VOC.
Defect Defect**
Measure 2
< <14 14oz. oz.Or Or> >18 18oz. oz.
Measure 2
1 1
DMAIC GB G TX PG
V 4.2.0
D M A
C
Methodology Step: Measure 1Select CTQ Characteristics
33

A A Lunch LunchPortion Portion
Define A
Define A
Healthy Healthy Lunch Lunch
High-Level Need (VOC)
Measure 1
% %Carbohydrate Carbohydrate Per PerServing ServingPer PerUSDA USDA
Prepared Prepared Lunch Lunch
Measure 1
CTQ
% %Carbohydrate CarbohydrateContent Content
Measure 2
> >20% 20%
Measure 2
Target* Target*
25% 25%
Defect Defect**
< <20% 20%
Measure 2
Measure 2
1 1
DMAIC GB G TX PG
D M A
C
34
Another Example What are your needs for this class?

# Tools Used In Project Exam Pass Rate Importance
Customer Needs
Learn The Material
Pass The Exam
Apply This Material To Project
Importance Rating Key High = 9 Medium = 3 Low = 1
59
53
37
# Completed Project Per Quarter
Knowledge Assessment Score
15
D M A
C
35
Points To Remember The process may look simple, but requires effort Many entries look obviousafter theyre written down If there are NO tough spots the first time It probably isnt being done right! Focus on the end-user customer Charts are not the objective. Charts are the means for achieving the objective. QFD is a perishable document. It has a shelf life and must be updated. QFD is a valuable decision support tool, not a decision-maker
QFD is a valuable decision support tool (assigning numbers to H-M-L does not make the numbers and cutoffs absolute decision points).
D M A
C
Methodology Step: Measure 1Select CTQ Characteristics
36

Knowledge Knowledge Assessment Assessment
Define A
Define A
GB GBKnowledge Knowledge Level Level
High-Level Need (VOC)
% %Correct Correct= = Measure 1 Total Correct Total Correct Questions QuestionsDivided DividedBy By100 100
Learn Learn Six Six Sigma Sigma
Measure 1 Knowledge Knowledge Assessment Score Assessment Score (Percentage (PercentageCorrect) Correct)
CTQ
Measure 2
80% 80%
Measure 2
Target* Target*
85% 85%
Defect Defect**
Measure 2
< <80% 80%
Measure 2
1 1
DMAIC GB G TX PG
V 4.2.0
D M A
C
37
Common QFD Pitfalls QFD on everything

Set the right granularity Dont apply to every last project
Lack of teamwork
Wrong participants Lack of team skills Lack of support or commitment
Too much chart focus Hurry up and get done Failure to integrate and implement QFD
D M A
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 teams objective is to use QFD to identify the subprocesses that will be the focus of DMAIC improvement projects.
D M A
QFDBreakout Activity (20 minutes)

Desired Outcome: Practice constructing a House Of Quality What Preparation How Choose a facilitator, scribe and timekeeper Review the partially completed House Of Quality on the next page Complete the relationship room of the house Calculate How Importance Ratings Calculate the How importance ratings Which How is most important to satisfy the customer? All 5 mins. Who All Timing 5 mins.
39
Complete Relationships
All
10 mins.
D M A
C
40
QFD Activity
HOWS Service Characteristics
% problems resolved on first call Cycle time to resolve problems % customers rating Customer Service Rep (CSR) excellent or good on customer survey of (CSR). courteousness Number of hours in the day helpdesk is available
Quick problem resolution

WHATS Customer Wants
Available when needed Courteous Service Importance rating Units Target
Importance
5 4 3
% 100
Hrs. <3
Hrs. days 24x7
% 100
To what extent will the Hows impact the What? Use L=1 M=3 H=9
Add one more service characteristic (CTQ) and place rating on the CTQ.
D M A
C
41
CTQ Tools: Summary There are a variety of tools available to identify the CTQ characteristic for your project (Project Y) Select the tool that is most appropriate for your project
D M A
D M A
Measure 2Define Performance Standards
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, youll 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 youre meeting the customers 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
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
Operational Definitions Defining The Measure

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

Everyone has a consistent understanding
To provide a clear way to measure the characteristic

Identifies what to measure Identifies how to measure it Makes sure that no matter who does
Always Pilot Your Operational the measuring, the resultsDefinitions are consistent
Operational definitions guide what properties will be measured and how they will be measured. There is no single right way to write an operational definition. There is only what people agree to for a specific purpose. The critical factor is that any two people using the operational definition will be measuring the same thing.
D M A
Operational Definitions For Output Measures Features

What: Must have specific and concrete criteria How: Must have a method to measure criteria Must be useful to both you and the customer (the wing-to-wing concept) Example: Loan Application Cycle Time What: Loan application cycle time is the number of hours from receipt of a loan application, to successful notification of decision for the loan application How: The clock starts when the computer attaches the time of application receipt at data entry
The clock stops when the phone caller notes time of completed application decision notification in desk log
Must Have Specific And Concrete Criteria Everyone has the same understanding of the definition. Different people can use the definition and know that their data will be measured in the same way. Example: If a team were studying long wait lines at a bank, they would agree on what to measure (e.g., length of the lines, how long people wait) and agree on a way to measure it. Must Have A Method To Measure Criteria The definition tells you how to get a value (either a number or yes/no). The definition clearly states how to get a measurement. Example: If a team were measuring length of lines at a bank, they would need to know whom to include in a count of people in linedo children count, or a spouse or friends waiting with someone? Must Be Useful To Both You And The Customer The operational definition makes it possible to make process improvements that are important to the customer and measurable for you. The definition should relate to issues that are important to both you and your customer. 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 customers perspective when developing operational definitions. GE Capital, Inc., 2000
DMAIC GB H TX PG V 4.2.0
D M A
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 customers perspective A customer-focused Project Y measure allows us to assess the process from the customers perspective Changes in the values of a good Project Y measure will predict changes in the degree to which the process meets the customer CTQs A Project Y measure expresses the voice of the process changes in input or process conditions drive changes in true output measures
Project Y data is the first key to understanding process performance and the starting point for understanding how to control process output.
D M A
Define Performance Standards

CTQ Elements (Performance Standards)5 Major Elements ONLY
Timely Call Pick-up (Output characteristic) Hold Time (Project Y Measure) Customer Need Customer waits on hold too long
Established in Define, Step A
Established in Measure, Step 1
CTQ
10 seconds or less (Target) Not greater than 20 seconds (specification/ tolerance limit) Any response taking More than 20 seconds (Defect)
A CTQ applies To each individual output unit
CTQs Are The Bridge Between Our Process Output And Customer Satisfaction
In Measure, Step 2, we are completing our CTQ Element Tree. Use the template on the next page or the one in your GB Workbook. Definitions: Characteristic: a word or phrase that describe some aspect of the product or service. Project Y: a definition of how the product/services 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. Specification: the variation around the target the customer will accept. Defect: failure to meet specification. Anything that results in customer dissatisfaction.
D M A
C
6

Output OutputUnit Unit
Define A C-O-P-I-S C-O-P-I-S A ACall Call
Define A Timely TimelyCall CallPick-up Pick-up C-O-P-I-S C-O-P-I-S
High Level Need (VOC) Customer Customer waits waits on on Hold Hold too too Long Long
Project ProjectY Y Operational Operational Definition Definition
Process ProcessMap Map Measure 1 QFD QFD C&E Time C&E Time From FromCall Call C-O-P-I-S C-O-P-I-S Placement Placementto to Pareto Customer Service Pareto Customer Service Pick-Up Pick-UpIn InSecs. Secs.
CTQ CTQ
Project ProjectY Y Measure Measure
Measure 1 QFD QFD Process ProcessMap Map HOLD HOLDTIME TIME C&E C&E Pareto Pareto
Measure 2 VOC VOC QFD QFD USL=20 USL=20Seconds Seconds
Target Target
QFD Measure 2 QFD Pareto Pareto 10 10Seconds Seconds VOC VOC
Defect Defect
Measure 2 FMEA FMEA > >20 20Seconds Seconds C&E C&E
# #of ofDefect Defect Opportunities Opportunities Per PerUnit Unit
FMEA 11 FMEA Process ProcessMap Map
Measure 2
D M A
C
7
Establishing A Performance Standard

A performance standard translates customer needs into quantified requirements for our product or process
Specific Needs Statement
Quick Quick Answer Answer
Timely Resolution of a Call
Project Project Y Y
Total time to resolve inquiry minus initiation time of inquiry 5 Minutes or less
CTQ CTQ
Target Target
Specification/ Specification/ tolerance tolerance limit(s) limit(s) (USL/USL) (USL/USL)
Not Greater Than 60 Minutes (USL) Any Response Taking More Than 60 Minutes
Defect Defect
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/processs characteristic is to be quantified. State as a formula if possible (Operational Definition) and state the metric (Project Y Measure) Target Value: Where we will aim our product/process. If there were no variations in the product/process, this is the value we would always achieve. Specification Limits: How much variation is the customer willing to tolerate in the delivery of our product or process? Defect: An output which is unacceptable to the customer. Anything that results in customer dissatisfaction.
D M A
Identify Project CTQs

Translating Needs into CTQs: Example 1Pizza Characteristic: temperature of pizza Project Y metric: temperature (degrees F) of Pizza at customer door step Need Need = = Hot Hot Target: 160 degrees Pizza Pizza Specification: LSL=140 degrees and USL=180 degrees Defect: any pizza temperature falling outside of the specifications Translating Needs into CTQs: Example 2Customer Service Characteristic: CSA Knowledge level Need Project Y: training test score Need = = Problem Target: 100 points Problem Solved Solved on on First First Specification: LSL= 90 points Call Call Defect: any score less than 90 points Translating Needs into CTQs: Example 3Information Characteristic: Time to find information Project Y: maximum time to find any piece of data Need Need = = II want want Target: 1 minute to to find find all all info info Specification: USL=1.5 minutes without without effort effort Defect: any time greater than 1.5 minutes
LSL=Lower Specification Limit USL= Upper Specification Limit
D M A
Specify Target And Specification Limits For Project Y

Once the Project Y metric has been established, the target and specification limits can be established.
Output Output Characteristic Characteristic Timely response
Specific Needs Statement

Customer Customer receives receives approval approval on on customer customer request request date date
Project Project Y Y
Variance between customer request and actual delivery Customer request date
CTQ CTQ
Target Target
Specification/ Specification/ tolerance tolerance limit(s) limit(s)
Variance not greater than one day Any response greater than 1 day
Defect Defect
Target/Nominal Value

Optimal value for Y Specification/Tolerance Limit(s) Window or range of acceptable output values for Y Also called performance standards Information about where to set target and specification limits comes from the customer. This information can be collected not only from VOC but also from: Contractual agreements Industry standards Customer observations Process data
D M A
Project Y
Six Sigma Performance Definitions Unit: Defect: Defect The item produced or processed relative to the Project Y as reviewed by the customer Any Project Y measurement value that does not meet the Y performance standards Opportunity: Any Y measurement event which provides a chance of not meeting the performance standard
10
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 customers perspective so looking at the percent of units with at least one defect gives us a perspective of how the customer sees the overall process performance. Notice that if units contain a number of defects, it is possible to reduce the overall total number of defects without having much impact on the percent of defective units. We need to be aware of both the total number of defects and the percent of units containing defects.
DMAIC GB H TX PG
V 4.2.0
D M A
Project Y Alignment
Business Big Ys
Key output metrics that are aligned with strategic goals/objectives of the business. Big Ys provide a direct measure of business performance.
11
Process Ys
PROCESS PROCESS Management Management
Y Y Y Y
Key output metrics that summarize process performance
Project Y
Key project metric defined from the customers 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 CTQs that the customer will feel. Not all Ys are useful measures of business performance or customer impact. Your Project Y should be correlated to a higher level Business Big Y or CTQ. You should be able to describe the link between your Project Y and the related Business Big Y or CTQ in specific terms.
D M A
C
12
6 Performance Standards DefinitionsBreakout Activity (5 minutes)

Desired Outcome Practice defining units and defect for your CTQ
What Team Preparation
How Choose a facilitator
Who All
Timing
Define units and defect
For your process output, agree on the unit and defect for your Project Y. Use the worksheet on page 17 of your GB Workbook. Choose a spokesperson to report your results
Facilitator
5 mins.
Close exercise
All
D M A
Minitab and Graphical Analysis Module Objectives
Understand the structure of Minitab* Understand data entry and correct data structure for analysis in Minitab Review variation Be able to create and interpret basic graphs in Minitab
In this module, you will focus on learning how to use Minitab and how to perform basic graphical analysis of data. We are stepping away from the DMAIC Flow for a moment. If youve 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 Minitabs 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
Minitab Windows
Menu Menu Bar Bar
Session Session Window: Window: Analytical Analytical Output Output
Data Data Window: Window: A A Worksheet, Worksheet, not not a a Spreadsheet Spreadsheet Graph Graph Window Window
Data WindowStores data to be used for analysis. Discussed in more detail on next page. Session WindowStores numeric output and character graphs that result from executing a statistics command.
You can move through the windows using Ctrl-Tab, the windows menu, or the following shortcuts.
Data Window Crtl+A Session Window Crtl+M Info Window Crtl+I Graph WindowGenerated whenever you run a command History Window Crtl+H that results in a graphical output. Info WindowStores synopsis of your data (not shown) window. History WindowStores all commands initiated (not shown) from either the menus or the session window (not shown).
D M A
Data Window
Minimizes the Window Data Entry Arrow
Closes the Window
Maximizes the Window Column Names Are Entered Here Data is Entered Here
Scroll Bars
Data window is a worksheet, not a spreadsheet. Variables are usually entered in columns, observations in rows. Each column has a title area which is 31-characters or less and this title must be unique. Everything in a column is considered to be the same variable. Column names are above the column, not in the first row. Minitab cells contain values only, not formulas, like Excel. To perform mathematical functions, use Calc > Calculator. The orientation of the data arrow determines whether the enter key will take you to the next row or the next column.
If the value you enter into the first row is something other than a number or date, Minitab assumes that the entire column will be text (T).
D M A
Minitab MenusSummary
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 Similar to the edit menu in most standard Windows applications Sort, code or manipulate data Calculate or generate data Basic statistics and quality tools Most often used by Green Belts Contains the commands that you will use to do graphical analysis during your project 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 Allows you to manage multiple graphs on the screen Product & process report can be used for determining performance levels (Calculating Sigma)

Edit Menu Manip. Menu Calc Menu Stat Menu Graph Menu Help Menu

Window Menu Six Sigma* Menu

Notes on Saving: Save Project AsWhen you save a project, you save all the worksheet, session, info, history and graph windows that are currently open. Minitab will use the file extension of .mpj. Save Current Worksheet AsWhen you save as a worksheet, you save only the worksheet portion (you will lose session window and graphs). Minitab will use the file extension of .mtw.
Notes on Opening: If you use the icon, only projects will be shown. You must use the file > open worksheet to view all Minitab worksheets. *Six Sigma add-on in Minitab can be obtained via a link in support central.
D M A
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 ParametersMinitab sometimes calls these parameters: X Factor Independent Y Response Dependent Input variable variables graph variables
By variable
D M A
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
D M A
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.
D M A
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: To determine whether a file can be opened by Minitab, choose Open Worksheet from the File menu. Notice that the area labeled Files of Type contains a drop-down menu. This is indicated by the arrow (triangle) at the right 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.
When importing from Excel, an Excel file can contain only one worksheet and the information contained in Row 1 will become the titles in Minitab. Note: During import, Minitab will convert percentages, dollars, etc.
D M A
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.
D M A
Using Minitab: A Typical Session Tips for moving data back and forth:
10
Structure the data so that each variable is in a single column Each column must have a title The column title must have fewer than 31 characters and be on a single line All data must immediately follow the column names Do not put empty rows between rows of data Columns containing dollar signs or commas cannot be transferred to Minitab using Copy or Paste, but can be imported using the import command. Reformat these numbers to include only decimal points. After movement into Minitab, check column heading type (D vs. T.)
If you want to use a file in both Minitab and Excel, keep these points in mind.
D M A
Using Minitab: A Typical Session Tables vs. Variable Columns Typical Excel Format:
Sales Office January February March April 387,980 45,700 456,789 349,050 Central 578,990 600,987 456,789 456,798 Southwest 435,800 542,700 345,988 564,050 Northeast 497,050 827,900 456,789 687,050 Southeast 613,242 61,689 456,789 434,567 Northwest
11
Format Needed For Minitab:

Sales Office Revenue Central 387,980 Central 45,700 Central 456,789 Central 349,050 Southwest 578,990 Southwest 600,987 Southwest 456,789 Southwest 456,798 Northeast 435,800 Northeast 542,700 Northeast 345,988 Northeast 564,050 Southeast 497,050 Southeast 827,900 Southeast 456,789 Southeast 687,050 Northwest 613,242 Northwest 61,689 Northwest 456,789 Northwest 434,567 Month January February March April January February March April January February March April January February March April January February March April
The best format for analysis of data in Minitab is variable columns.
The format of data typically used in Excel is table format. The best format for analysis of data in Minitab is variable columns. There are multiple ways to manipulate data. The major point is that Minitab likes to have the data in columns.
D M A
Using Minitab: A Typical Session Tips For Importing Data Into MinitabLets practice
File name: sales_data.xls
Sales 1999 Sales Office Central Southwest Northwest Southeast Northwest Total January $387,980.00 $578,990.00 $435,800.00 $497,050.00 $613,242.00 $2,513,062.00 February $45,700.00 $600,987.00 $542,700.00 $827,900.00 $61,689.00 $2,078,976.00 March $456,789.00 $456,789.00 $345,988.00 $456,789.00 $456,789.00 $2,173,144.00 April $349,050.00 $456,798.00 $564,050.00 $687,050.00 $434,567.00 $2,318,658.00
12
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 into an Excel spreadsheet. 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!
Note: For additional practice, see the Green Belt Training Support Central Site.
D M A
Using Minitab: A Typical Session Step 2: Select menu command
13
Most of the commands you use in your Green Belt project are in the Stat and Graph menus.
D M A
Using Minitab: A Typical Session Step 3: Enter parameters in dialog box
14
D M A
Using Minitab: A Typical Session Step 4: View results in session or graph window
15
Note: Sometimes there is only a session window. Therefore, check both the session window & the graph (if one exists).
D M A
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
16
Copying session window output:

1. Highlight the text lines you want to copy 2. Use the right mouse button to copy the text 3. Open the application into which you want to copy the text 4. Paste the text using Paste from the toolbar or Ctrl-V
Note: You can also copy graphs from the window > manage graphs.
D M A
Using Minitab: A Typical Session Steps 6 and 7: Print and Save
17
When you save a project, you can save the worksheet or session window individually, or save an entire project.
D M A
Recruitment Cycle Time Case Study
18
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 expyears experience of the person who filled the position C2: agentspecifies the agent who processed the request and found the resource C3: sitespecifies the company site from which the request originated (there are three sites in the area who request temporary resources) C4: type of resourcespecifies the type of resource requested (DBA, Programmer, Systems Analyst) C5: cycle timetime to fill position (working) days
D M A
Graphical Analysis Of Data Key Questions: How is my data distributed (variation)? What relationships exist between the Y variable and X variables?
19
D M A
ReviewVariation All repetitive activities have variation (fluctuation) Variation is a primary source of customer dissatisfaction In order for our customers to feel the quality at GE, we must reduce variation
20
A fundamental principle of process improvement is the measurement, reduction, and control of variation. Variation is present in all processes, whether they are personal processes (getting to work, fixing dinner) or business processes (approval cycle time, order to delivery). The output of a process will vary as it is repeatedly performed. Although our customers may accept some variation, when variation is too extreme, our customers will be dissatisfied.
D M A
Using Data To Understand Variation Plot The Data Using Variation Tools
Study Variation For A Period Of Time
Histogram
21
Study Variation Over Time

Run Chart
Measurement
Frequency
Measurement
Time
For Continuous Data Histogram Box Plot
For Discrete Data Bar Chart Pie Chart Histogram
For Continuous Data Run Chart Control Chart
For Discrete Data Control Charts Run Chart
Variation is a fact of life. There will always be fluctuations in our processes. Graphically displaying measures provides a basic description of the variation and its sources. Following any data collection effort, the first step toward understanding variation is to plot the data.
D M A
ReviewContinuous vs. Discrete Data Reminder: Data Type Is Critical! Data type dictates how much variation we will see:
Continuous datathe most information about variation in the process Discrete dataless information about variation in the process
22
Application Cycle Time Upper Specification Limit = 30 Days
Continuous
Y = days to process
Discrete
Y = late/on-time No. Recd 30 No. Late 2
Actual Times 28 21 21 30 12 17 23 16 25 29 11 9 13 24 26 30 27 30 34 11 27 20 23 29 24 49 27 10 24 29 29 21 29 30 28 28
USL
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?
DMAIC GB I TX PG
V 4.2.0
D M A
5 Ms & 1 P Sources Of Variation
23
Machines Methods Materials Measurements Mother Nature People
P R O C E S S
MachinesThe various appliances used in the transformation from inputs into outputs. For example, a PC can turn various sources of information into an organized manual that then relates to a training service. MethodsThe procedures, formal or otherwise, that transform inputs into outputs. For example, there is a standard procedure for billing collections in the GE Capital businesses. MaterialsThe components, tangible or otherwise, that are transformed from inputs into outputs. For example, paper stock and ink quality will affect a product brochures quality. MeasurementsThe tools that monitor a processsperformance. For example, a doctors blood pressure reading of a patient would determine subsequent activity related to treatment. Mother NatureThe environmental elements within a process that influence a customer need or requirement.
In a training session, failure to regulate the thermostat can result in a non-conducive learning environment. PeopleThe staffing that influences customer needs and requirements. While often dominant in the service industry, this is an area still too often blamed for failure to meet or exceed customer requirements. Sometimes in a service environment, the following categories are used:
PoliciesHigher level decision rules or management practices. ProceduresThe way in which tasks are performed. PlantThe building, equipment, work space, and environmental factors that affect performance. PeopleThe human element. PartsSystems, documents, and other supplies that are needed to perform the service.
DMAIC GB I TX PG
V 4.2.0
D M A
Two Types Of Variation Common Versus Special Causes
24
Type of Variation
Common Cause
Characteristics Characteristics
Always Present Expected Predictable Normal Not Always Present Unexpected Unpredictable Not Normal
Special Cause
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:

Common causes Special causes
The distinction between common and special causes is important to determine the basic strategy for process improvement and control.
Common causes are characteristics of the process and exist as a result of the presence and interaction of different process variables. Common causes affect everyone working in the process, and affect all of the outcomes. Common causes are always present and thus are predictable within bounds. Special causes are those causes that occur due to occasional extraordinary circumstances. Special causes are not always present, do not affect everyone working in the process, do not affect all of the outcomes, and are not predictable.
D M A
Describing Variation For A Period Of Time: Data Distributions
25
Key Questions: What is the shape of the distributionsymmetrical, skewed, twin peaks, flat? What is the central tendency (center) of the distribution? What is the variation (spread) of the distributionwide or narrow?
When describing variation, we usually focus on two things: Center Spread
D M A
Statistics
26
Statistics is concerned with making inferences about general populations and about characteristics of general populations We study outcomes of random experiments If a particular outcome is not known in advance, then we do not know the exact value assigned to the variable of that outcome:
The number of invoices received weekly The cost in dollars of reworking each part The number of surfaces that are rough on a cast part The number of calls received every Monday between the hours of 8-9 a.m.
We call such a value a random variable
D M A
Some Distributions A random variable can be expressed in terms of a distribution Uniform Distribution
P(X) X
27
Single roll of a die All permissible values P(X) are equally likely
Triangular Distribution
P(X)
Sums of a pair of dice Rapidly descending P(X), no tails

X
D M A
Distributions Normal Distribution

P(X)
28
Process/repair times Error fluctuations about an operating point
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 distributionsor unmixed. Stable distributions can have many different shapes. Why does the process, or product, result in a particular shape (distribution) and not some other shape? The answers to these questions may provide a better understanding of the process and how to improve it. The particular shape is not as important as why this shape and why not some other shape? Why this distribution, why not some other distribution?
D M A
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 7 6 5 4 3 2 1 0
Center
29
Center Skewed Distribution
20 Frequency
Frequency
10
0 12 13 14 15 16 17 18 19
Spread
Bimodal Distribution
Center Center
0 10 20 30 40 50 60 70 80 90
Spread
6 5 Frequency 4 3 2 1 0 7
11 13 15 17 19 21 23
The shape of a data set can be determined by examining a histogram, or, if testing the distribution for normality, using a Normal Probability Plot. There are many different shapes a data set may assume. The plots above illustrate three common shapes. The top left is a normal distribution, the middle is a bimodal distribution, and the top right plot is a skewed distribution. Because of its predictive qualities, the normal distribution is of special interest.
D M A
The Normal Curve Is The Data Distribution Normal?
30
Definition: A probability distribution is where the most frequently occurring value is in the middle and other probabilities tail off symmetrically in both directions. Characteristics:
The curve does not reach zero The curve can be divided in half with equal pieces falling either side of the most frequently occurring value The peak of the curve represents the center of the process The area under the curve represents 100% of the product the process is capable of producing
The normal curve is a probability distribution that forms the basis for many decisions we will make about our processes. The curve is noted by its bell-shaped nature, where most of the values fall in the middle and fewer values fall in either direction. The curve has several important characteristics:

InfinityThe curve theoretically does not reach zero. It goes out to infinity in either direction. SymmetryRoughly half of the values will be above the average, and half below. CenteringThe peak of the curve tells us where the process is centered. Process TotalityThe area under the curve represents virtually 100% of whatever we are measuring.
D M A
The Normal Curve (continued) Specific Characteristics

34.13% 34.13%
31
2.14% 0.13% -3s -2s
13.60%
13.60%
2.14% 0.13%
-1s
X
68.26% 95.46% 99.73%
+1s
+2s
+3s
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 segments. Each segment is mathematically called a standard deviation from the mean. It is also noted by the small s. 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.
Most common statistical terms and analysis tools are based on a normal data distribution. If we use these tools with a data set that is not normal, the accuracy of the tools may be compromised.
D M A
Normal Probability Plot We use a normal Probability Plot to determine if our data is normal Normal Probability Plot
.999 .99 .95
32
Probability
.80 .50 .20 .05 .01 .001 2 12 22 32
Cycle Time
Average: 16.3921 StDev: 5.61675 N: 240 Anderson-Darling Normality Test A-Squared: 0.208 P-Value: 0.864
Use this method to determine if distribution is normal
If p>0.05, then data is normal
What Is A Normality Test? A normality test is a statistical process used to determine if a sample, or any group of data, fits a standard normal distribution. A normality test can be done mathematically or graphically. A normality test can be thought of as a litmus test for determining if a distribution is a normal distribution. The Y axis is the cumulative % of data points which fall below the value on the X axis. Why Is A Normality Test Useful? Many statistical tests (tests of means and tests of variances) assume that the data being tested is normally distributed. A normality test is used to determine if that assumption is valid.
DMAIC GB I TX PG
When Should I Use A Normality Test? There are two occasions when you should use a normality test:
When you are first trying to characterize raw data, normality testing is used in conjunction with graphical tools such as histograms and box plots. When you are analyzing your data, and you need to calculate basic statistics such as Z values or to employ statistical tests that assume normality, such as t-Test and ANOVA.
Interpreting A Normal Probability Plot
When plotted data follows a straight line, the Anderson-Darling p-value will exceed 0.05 and will increase, I have normally distributed data. Rule: p > .05 indicates data is normal.
D M A
Normal Probability Plot (continued)

Distribution Type:
Normal
Roughly Normal Distribution
7 6 6 5
33
Bimodal curve
Bimodal Distribution
20
Skewed
Skewed Distribution
6 5
Long-Tailed
Long-Tailed Distribution
Frequency
Frequency
5 4 3 2 1 0 12 13 14 15 16 17 18 19
Frequency
3 2 1 0 7 9 11 13 15 17 19 21 23
10
Frequency
0 10 20 30 40 50 60 70 80 90
4 3 2 1
0 7 9 11 13 15 17 19 21 23
Normal Probability Plot for a Normal Distribution

99 ML Estimates Mean: StDev: 95 90 80
Normal Probability Plot for a Bimodal Distribution

ML Estimates Mean: StDev: 14.6382 5.47084
Normal Probability Plot for a Skewed Distribution

ML Estimates Mean: StDev: 15.0790 12.6232 99 95 90 80
Normal Probability Plot for Long-Tailed Distribution
Percent
70 60 50 40 30 20 10 5 1 0 10 20 30
70 60 50 40 30 20 10 5 1 0 10 20 30
Percent
Percent
How Distribution Looks On Normality Curve: Straight line Zig-zag Two lines (Stable Operations) S curve
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?
Percent
D M A
What If Your Data Is Not Normal? If you conclude that Y is non-normally distributed, there are two general approaches:
Approach 1: Variance-based Thinking (VBT) Methodology possibly multiple processes embedded
segmentation and stratification span reduction
34
Approach 2: Non-parametric techniques (beyond the scope of this course)
Expectation: Green Belts Should Be Able To Do Approach 1
If your data is non-normal, you may be able to proceed. It is important that you consult with your MBB before proceeding.
D M A
MinitabHistogram Recruitment Cycle Time Case Study

Datafile: Question: Fill_time.mtw What does the distribution of cycle time to fill positions look like?
Frequency
30
35
Tool:
20
Histogram
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 bar or interval in the histogram. The Y axis is calculated for you. Data A single column of data for each Structure: histogram To Make A Graph > Histogram Histogram: Variable: Cycle Time Click OK
10
0 0 10 20 30
Cycle Time
D M A
MinitabNormal Probability Plot Recruitment Cycle Time Case Study

Datafile: Question: Fill_time.mtw
.999
36
Normal Probability Plot
Probability
Are the cycle time data normally distributed? Normality Test Continuous Y A single column of data
.99 .95 .80 .50 .20 .05 .01 .001 2 12 22 32
Tool: Data Type: Data Structure:
Cycle Time
Average: 16.3921 StDev: 5.61675 N: 240 Anderson-Darling Normality Test A-Squared: 0.208 P-Value: 0.864
To Make A Normal Stat > Basic Probability Plot: Statistics > Normality Test Variable: Cycle Time Test For Normality: Choose Anderson-Darling Click OK
Since p > 0.05, the data is normal. There are a few ways to determine normality. We have standardized on this Anderson-Darling normality test. We will discuss p-values later in the Analyze phase.
D M A
Center Or Central Tendency Descriptive Statistics: Represents the nominal value of the process.
Mean (X) Median (middle data point) Quartile Values (Q1, Q3)
37
Normal Distribution
Long-tailed Distribution
Skewed Distributions
Q1
Q3
A good guess of central tendency can be made by visually examining a histogram. A more precise estimate of central tendency can be made using descriptive statistics such as the mean, median, or quartile values. The distribution shape dictates which metrics should be used.
D M A
Center Or Central Tendency (continued)

The Mean, sometimes called the average, is the most likely or expected value. The formula for the mean is:
38
X=
X
n
1. The sum of all data values 2. Divide by number of data values
The Median is literally the middle of the data set where 50% of the data is greater than the median, and 50% of the data is less than the median. The most commonly used symbol ~ for the median is X. The procedure for calculating the median is :

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

Standard Deviation (s) Span Stability Factor (SF) = Q1/Q3
Normal Distribution
39
Long-tailed Distribution
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 Deviationmeasures average distance from the mean. Smaller is better. Spanmeasures unusually low & high distance from the median. Smaller is better. Stability Factorratio of first quartile and third quartile, 1.0 is best.
D M A
Variation (continued)
Normal Distributions:
Skewed Or Stable Ops Distributions:
40
The standard deviation is the average distance, or The stability factor (SF) is the ratio of the first quartile and third quartile values. The formula for deviation, that a given point is away from the Stability Factor is: mean. The formula for the standard deviation is:
Xi-X
SF = Q1/Q3
A- subtract each data value from the mean E- take the square root of the result
As Q1 and Q3 get closer together, their ratio decreases and the variation of the middle 50% decreases.
Xi X
Q 1 B- square each difference C- sum the squared differences

Q 1 Q 3 Q 1 Q 3
Q 3
(X X )
i
n 1
Q 1
Q 3
D- divide by 1 less than the number of data values
More Variatio n
Less Variatio n
Long-tailed Distributions:
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 dont 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 100-500 501-2000 2001-5000 5001+
DMAIC GB I TX PG
0 More Variation
Less Variation 1
Span 90%-10% 95%-5% 98%-2% 99%-1%
P P90, P10 P95, P5 P98, P2 P99, P1

D M A
The Computational Equations

Population Mean Population Variance Population Standard Deviation Xi
i =1 N
41
(Xi )
i =1
(Xi )
i =1
Sample Mean Sample Variance Sample Standard Deviation
=X=
X
i =1
2 = s2 =
(X
n i =1
X)
n1
X)
2
= s=
(X
n i =1
n1
The difference between population and sample:

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

Random data (1 to 99, n = 10 samples):
n 1 2 3 4 5 6 7 8 9 10 Xi 20 1 6 35 54 67 43 99 5 27 ----357 Xbar 35.7 35.7 35.7 35.7 35.7 35.7 35.7 35.7 35.7 35.7 -----Dev =(Xi-Xbar) -15.70 -34.70 -29.70 -0.70 18.30 31.30 7.30 63.30 -30.70 -8.70 ---------0.0 (Xi-Xbar)^2 246.49 1204.09 882.09 0.49 334.89 979.69 53.29 4006.89 942.49 75.69 -------------8726.10
42
Results & Take-Aways: Deviation = measure of Variation, measure of Spread SST = 8726.1 large Deviations = large SST [df = deg. of freedom = n-1]
Variance = Average SST = SST/df
s2 = Sample Variance = SST/9 = 969.6
2 = Population Variance = SST/10 = 872.6

Standard Deviation = Square Root of Variance: Sample StDev = s = Population SD = =
969.6 = 31.1
872.6 = 29.5
Its All Variation Deviation, SST, Variance, And Standard Deviation = Measures Of Spread
A key concept is that the variance is the average SST for a population and a near average for a sample = SST/(n-1) = SST/df. As the degrees of freedom (df) increase (which is the same as larger sample size (n)), the near average and average SST become essentially the same. Finally, the standard deviation is the square root of the variance.
D M A
SummaryMeasures Variation SummaryMeasures Of Of Variation:

Range = (Max - Min) Deviation = (Xi - Xbar) [Xbar = mean or mu] Sum-of-Squares (of the squared Deviations) = SST Variance = Average SST = SST/df (df = degrees of freedom) Std Deviation = SqrRoot (Variance) and Coeff. of Variation = CV = (StDev/Xbar)* 100 i.e., ratio of StDev to Meanexpressed 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.
43
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 numberswhich means that the spread around the mean will likely be changed.
D M A
Variation For A Period Of Time

Descriptive Statistics Summary
44
Shape
Normality Plot
Normal Probability Plot for a Normal Distribution
ML Estimates Mean:15.7224 StDev: 1.74183
Center (central tendency)
Spread (variation)
Mean X
( )
Standard Deviation (s)
normal
Normal Probability Plot for an Exponential Distribution
ML Estimates Mean: 15.0790 StDev:12.6232
Percent
Quartile Q1 or Q3
Stability Factor (SF)
skewed
Normal Probability Plot for a Long-Tailed Distribution
99 95 90 80 70 60 50 40 30 20 10 5 1 ML Estimates Mean: 14.6382 StDev:5.47084
Percent
Percent
~ Median X
( )
Span
long-tailed
0
99 95 90 80 70 60 50 40 30 20 10 5
10
20
30
ML Estimates Mean: 14.6382 StDev:5.47084
Normal Probability Plot for a Bimodal Distribution
The different processes must be stratified before descriptive statistics can be calculated.
bimodal
Percent
10
20
30
Q1= 25th Percentile Q2 = Median,, 50th Percentile Q3 = 75th Percentile Range = Maximum ValueMinimum Value Span = 99th Percentile1st Percentile SF = Q1/Q3 IQR = Inner Quartile Range = Q3-Q1 (middle 50th of data)
D M A
Displaying Variation For A Period Of Time Histogram

Time Estimates Round 1 16.48 18.89 13.18 11.11 14.67 16.53 14.79 18.06 14.48 14.89 15.63 13.95 13.74 17.67 10.23 13.67 11.35 15.03
45
Measurements
Time Estimates Round 2 13.84 13.50 15.41 14.35 14.37 14.63 13.58 14.75 11.95 14.36 16.17 15.15 12.48 14.12 19.00 13.81 12.97 14.19
Graphical Display
Histogram of Time Estimates
# of Occurrences
10
0 10 11 12 13 14 15 16 17 18 19
Time Estimates (in seconds)
Illustrates
Shape (pattern) of the data Central tendency (center) of the data

Variation (spread) of the data
After data is collected, it should be displayed graphically and described using basic statistical measures. The first graphical display used for describing variation is the histogram. Histograms combined with basic descriptive statistics provide a clear picture of the current performance of a process.
D M A
Displaying Variation For A Period Of Time (continued) Box Plots
46
Outlier Highest Value
Each segment represents 25% of the data points
Third Quartile (75%) value Q3 Median Q2 First Quartile (25%) value Q1
Lowest Value
Box Plots are graphical summaries of the patterns of variation in sets of data. The horizontal lines at the top and bottom of the plot represent the highest and lowest values. The horizontal line in the middle of the solid box represents the median, and the solid box represents the middle 50% of the data, with 25% of the data on the top side of the box and 25% of the data on the bottom side of the box. Points noted with an asterisk(*) represent extreme values or outliers. Outlier calculations High: Q3 + 1.5 (Q3-Q1) Low: Q1 - 1.5 (Q3-Q1)
D M A
SummaryVariation For A Period Of Time
47
10
Data
28 21 21 30 12 17
23 16 25 29 11 9
13 24 26 30 27 30
34 11 27 20 23 29
24 49 27 10 24 29
29 21 29 30 28 28
Histogram
5
0 10 11 12 13 14 15 16 17 18 19
Box Plots and histograms provide similar information but in different graphical formats.
D M A
MinitabGraphical Summary Recruitment Cycle Time Case Study

Datafile: Question: Fill_time.mtw How can I examine the shape, central tendency, and variation (spread) of my data using one Minitab command? Descriptive Statistics
4 10 16 22 28
48
Descriptive Statistics
Variable: Cycle Time
Anderson-Darling Normality Test A-Squared: P-Value: 0.208 0.864 16.3921 5.6167 31.5478 6.39E-02 -2.4E-01 240 3.0800 12.4067 16.6400 20.3067 32.2667 17.1063 6.1698 17.3441
Tool: Data Type: To Make A Histogram:
Mean StDev Variance Skewness Kurtosis N Minimum 1st Quartile Median 3rd Quartile Maximum 15.6779
Continuous Y
95% Confidence Interval for Mu
Stat > Basic Statistics > Display Descriptive Statistics: Variable: Cycle Time Click Graphs button Select Graphical Summary Click OK twiceonce in each dialog box

16.0 16.5 17.0 17.5
95% Confidence Interval for Sigma 5.1552 95% Confidence Interval for Median
95% Confidence Interval for Median
15.7274
This graphical summary is one of the most widely used charts in Minitab. Useful Items:
SkewnessA measure of asymmetry. A value more than or less than zero indicates skewness in the data. But, a zero does not necessarily indicate symmetry. KurtosisA measure of how different a distribution is from the normal distribution. A negative value typically indicates a distribution more peaked than the normal. A positive value typically indicates a distribution flatter than normal. Nsample size. Confidence Interval for MuTells us the interval in which the true population mean lies (with 95% confidence.)
Anderson-Darling normality TESTWe use the pvalue to determine whether our distribution is normal or non-normal. If p > 0.05, the distribution is normal. (we will discuss p-value & hypothesis testing in Analyze, Step 6.) MeanAverage of the samples. StDevStandard deviation of the samples. Variance(Standard Deviation Squared.) A measure of how far the data are spread about the mean.

D M A
Variation Over Time Run Chart

A graphical tool to monitor the stability of Project Y Allows observation of time order properties such as trend Should be used before any detailed data analysis
49
Example of a Run Chart

Median
Is the process stable over time?
An important basic tool for understanding variation is the Run Chart. Run Charts are simple time-ordered plots of data. On these plots one can perform tests for certain patterns in the data. Presence of these patterns indicate special causes. It is important to note that the data should be in time order for run charts to be valid.
D M A
Run ChartsSpecial Cause Patterns

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

Datafile: Question: Fill_time.mtw Does cycle time show any patterned variation over time? Run Chart Continuous Y
32
51
Run Chart for Cycle Time
Cycle Time
22
Tool: Data Type:
12
2 40 140 240
Observation
Data Structure: A single column of data is all that is necessary. It is assumed that the data is entered in the order in which it is collected (time order). To Make A Run Chart: Stat > Quality Tools > Run Chart Single Column: cycle time Subgroup size: 1
Number of runs about median: Expected number of runs: Longest run about median: Approx P-Value for Clustering: Approx P-Value for Mixtures:
124.000 121.000 9.000 0.651 0.349
Number of runs up or dow n: Expected number of runs: Longest run up or dow n: Approx P-Value for Trends: Approx P-Value for Oscillation:
151.000 159.667 4.000 0.091 0.909
The 4 p-values circled above indicate stability. Since p>0.05, there are no issues with Mixtures, Clusters, Oscillations or Trends. Therefore, our process appears stable (no special causes).
D M A
Two Types Of Variation Investigating Common vs. Special Causes For new process data, use a Run Chart to look for special causes Investigate special cause points for positive quick-fixes Common cause variation requires systematic improvement effort
52
For new process data, a Run Chart can be used to detect special cause situations. For processes that are already in statistical control, Control Charts are the preferred method. Control Charts are discussed in the Control phase of DMAIC training. But, here in the Measure and Analyze phase we suggest using the Run Chart to test for stability.
D M A
Two Types Of Variation (continued) Reacting To Common vs. Special Causes How you interpret variation . . .
Common Causes Special Causes
Mistake 1 Tampering (increases variation)
53
Common
True variation type...
Causes
Focus on systematic process change
Special Causes
Mistake 2 Under-reacting (missed prevention)
Investigate special causes for possible quick-fixes
Why is it important to know the source of variation and treat it according to the appropriate strategy? Because not reacting appropriately to the type of variation present in a process can seriously impact customer satisfaction and the amount of variation and defects, and it can increase costs. When interpreting variation, you can make two types of mistakes:

Treating special causes as common causes is essentially underreacting to individual data points. The problem with making this type of mistake is missing an opportunity to prevent defects and reduce process variation. An example of this type of mistake would be failing to correct the steering on a car heading for a ditch, and ending up in the ditch. Appropriately reacting to the source of variation in a process provides the correct economic balance between over-reacting and under-reacting to variation from a process.
Treating common causes as special causes Treating special causes as common causes
Treating common causes as special causes is essentially overreacting 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. Its like a novice driver who over-reacts to slop in the steering of a car. The result is more variation in the path of the car.
D M A
Graphical Analysis Tools Looking For Patterns In Data
54
Continuous Y Boxplot Scatterplot
Discrete Y Pareto Chart
D M A
Box Plots What differences do you see between the output from the different shifts?
60
55
Measure
30
10 Shift 1 Shift 2 Shift 3 Shift 4 Shift 5 Shift 6
Box plots can be used to identify differences in variation between subgroups that exist in your data. Look for 2 major items here: Centerare the medians similar? Spreaddo the boxes and the Whiskershave similar heights?
D M A
MinitabBox Plots Recruitment Cycle Time Case Study

Datafile: Question: Fill_time.mtw Is there a difference in variation in cycle time between agents?
30
56
Tool:
Box Plot
Cycle Time
20
Data Type: Continuous Y, Discrete X To Make: Graph > Box plot A Boxplot Graph variables: Y Cycle time Click OK X Agent
10
0 1 2 3 4
Agent
Additional Questions: Is there a difference in variation in cycle time between sites? Between types of resource?
D M A
Scatter DiagramsAnalyzing Relationships Use Scatter Diagrams To Study The Relationship Between Two Variables.
40
57
Cycle Time (Days) (Y)
35 30 25 20 15 10 5 1K 2K 3K 4K 5K 6K 7K 8K 9K 10K
Size Of Loan (X)
A Scatter Diagram is an important graphical tool for exploring the relationship between two continuous variables. Here, it appears that size of Loan and Cycle Time have a relationship (Larger Loans take longer to process positive relationship).
D M A
Warning! Correlation Does Not Imply Causation Correlation Between Number Of Storks And Human Population
100 200 300 80 80
58
70
70
Population (In Thousands)

60 60
50 100
50 200 300
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 correlationbut not causationbetween the occurrence of vaporlocks in automobiles and the use of public swimming pools.)
D M A
Common Construction Mistakes Examples

Problem
Bank Errors
59
Variable
Size Of Loan
Axis
Variable
# Of Errors In Closing Cost Estimate # Of Changes
Axis
Loan Delays
Length Of Time To Approve Loan Computers Response Time
Computers Are Too Slow
Computers System Load
Delays Unloading Airline Baggage
# Of People On Plane
Time To Unload Baggage
One of the most common mistakes that occurs with Scatter Diagrams is mixing up the X and Y variables. The X variable is the potential cause and is plotted on the horizontal axis. The Y variable is the effect and is plotted on the vertical axis.
D M A
Interpretation Of A Scatter Diagram Look For: Common patterns in the data Range of the predictor variable (X) Irregularities in the data pattern
60
Common Patterns In The Data See whether the potential cause variable and the effect variable are related to one another. Range Of The Predictor Variable (Xs) The range is the difference between the largest and smallest values. Check that the range of the potential cause variable is wide enough to show possible relationships with the effect variable. Irregularities In The Data Pattern Check whether the data pattern indicates possible problems in the data.
D M A
Interpreting A Scatter Diagram Look For Patterns

1 3 5
61
Strong Positive Correlation
Strong Negative Correlation
No Correlation
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: Positive Relationship: X Increases/Y Increases Graphs 1 and 2as X increases, Y increases Graph 1Strong, positive relationship between X and Y Not much scatter Points form a line with a positive slope Graph 2Weaker, positive relationship between X and Y Scatter is wider Points still form a line with a positive slope As trainer experience increases, participant satisfaction also increases. Negative Relationship: X Increases/Y Decreases Graphs 2 and 3as X increases, Y decreases Graph 3strong negative relationship between X and Y Not much scatter Points form a line with a negative slope Graph 4weaker negative relationship between X and Y Scatter is wider
DMAIC GB I TX PG
As trainer experience increases, participant satisfaction decreases No Relationship: For any value of X, there are many values of Y Graph 5there is no relationship between X and Y No one line best describes the data There is no relationship between trainer experience and participant satisfaction Non-Linear RelationshipThe relationship between X and Y is complex and cannot be summarized with a straight line. Graph 6Inverted u As trainer experience increases, participant satisfaction also increases until a critical threshold is reached, at which point participant satisfaction decreases. This is not a linear relationship.
D M A
MinitabScatter Plot Recruitment Cycle Time Case Study

Datafile: Question: Fill_time.mtw Is there an association between cycle time and years experience?
Cycle Time
62
30
Tool: Data Type:
Plot Continuous Y, Continuous X
20
10
To Make A Graph > Plot Scatter Plot: Y Cycle time X Years Experience Click OK
0 0 5 10 15 20 25
Yrs Exp
D M A
Pairs ExerciseAnalyze Patterns With A Partner:

Review scatter diagrams on following pages as assigned Determine the type of relationship you observe (e.g., negative, positive, strong, weak, etc.) Describe an example from your business of this type of relationship
63
D M A
Common Scatter Diagram Patterns (continued)
64
Plot
+/- Other
Strong, Weak, Other
Example
Effect
Potential Cause
Effect
Potential Cause
Effect
Potential Cause
D M A
65
Plot
+/- Other
Strong, Weak, Other
Example
Effect
Potential Cause
5
Effect
Potential Cause
Effect
Potential Cause
D M A
66
Plot
+/- Other
Strong, Weak, Other
Example
Effect
Potential Cause
Effect
Potential Cause
Effect
Potential Cause
D M A

Plot
67
+/- Other
Strong, Weak, Other
Example
10
Effect
Potential Cause
11
Effect
Potential Cause
12
Effect
Potential Cause
D M A
Pareto Charts Is There A Defect That Occurs Frequently?
68
Frequency
Category of Defect
We were introduced to Pareto charts in the CTQ tools module.
D M A
MinitabPareto Chart Recruitment Cycle Time Case Study

Datafile: Question: Tool: Data Structure:
69
Fill_time.mtw Which agent has the most cycle time defects? Pareto Chart A single column where the defect categories are recorded OR
Count
180 160 140 100 80 60 40 20 0 100 80 60 40 20 0
4 1 2 3
Pareto Chart for Agent

Y
A tally table in two columns. One column contains the Type of Defect, the second column contains the frequency of the defect.
To Make A Pareto Chart:
Stat>Quality Tools>Pareto Chart Chart Defects Data In: Agent By Variable: Cycle _Time_Defect Choose: One Chart Per Page, Independent Ordering of Bars Click: OK
Defect
Count Percent Cum %
35 56.5 56.5
12 19.4 75.8
8 12.9 88.7
7 11.3 100.0
Additional Questions:
Which site has the most cycle time defects? Which type of resource has the most cycle time defects?
To find out which site has the most defects: Follow steps above except, place Site in Chart Defects Data In Box. Alternate Method (Splitting A Worksheet) MANIP > SPLIT WORKSHEET By variable: Cycle_Time_Defect Click: OK Now you have 2 worksheets-one with N transactions and one with Y transactions Activate the correct worksheet and then follow the steps in the slide above
Alternate method (subsetting a worksheet) MANIP > SUBSET WORKSHEET Name: No Transactions Include or Exclude: CLICK ON Specify which rows to include Specify which rows to include: CLICK ON Rows that Match CLICK ON Condition Button Condition: Double Click on C6 then = N (cycle_time_defect) = N Click OK twice Now you have a worksheet with N transactions
Percent
120
D M A
D M A
C
1
Measure Data Collection And Collect Data Define 3Plan Review Objectives
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
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
Measure 3Plan data collection, validate measurement system, and collect data
2
Data Collection Plan Worksheet Define Review Objectives

Data collection purpose:
What to measure (output measure and segmentation factors)
Measure (Name)
Measure Type (Y, X)
Data Type (Discrete)/ (Continuous)
Operational Definition
Range Of Values
How to measure: Sampling Plan (Scheme, Frequency, Size)
Measurement Procedure
Refer to more detailed documentation, if necessary.
D M A
Measurement System Analysis (MSA) Worksheet Define Review Objectives

Type of Gage R&R Conducted (Check One): Gage R&R (Continuous Data) Gage R&R Results 1) Two-Way ANOVA Part p-value Oper p-value Oper & Part p-value
Date Conducted:
Significant? Y Y Y Pass? N N N
2) 3) 4) 5)
% Tolerance % Contribution % Study # Distinct Categories
[ <30% ] [ <8% ] [ <30% ] [ >4 ]
Y Y Y Y OK? Y Y Y Y
N N N N
Graphical Output 1) Effective Resolution [ >50% ] 2) Stability [R Chart] 3) Consistency Between Xbar consistency Between Oper 4) Systematic Shift [Oper/Part Inter. Plot] Attribute Gage R&R
N N N N
1) Repeatability [ >90% ] 2) % Reproducibility [ >90% ] 3) % Accuracy [ >90% ] _______________________________________________________________ 1) Gage R&R Pass? Y N, If NO: Plan for improvement: _______________________________________________________________ _______________________________________________________________
D M A
Define Review Objectives
D M A
C
1
Plan For Data Collection Define Review Objectives

Develop Operational Definitions And Procedures
Establish Data Collection Goals
Ensure Data Consistency And Stability
Collect Data And Monitor Consistency
Clarify purpose of data collection Identify what data to collect
Write and pilot operational definitions Develop and pilot data collection forms and procedures Establish a sampling plan
Test and validate measurement systems
Train data collectors Pilot process and make adjustments Collect data 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
Step 1: Establish Collection Goals Define Review Data Objectives In order to establish your data collection goals you must: State the purpose of the data collection Identify what data is required Asking these questions may help you clarify your goals: What do I need to know about my process? What data do I need? What is the plan for analysis once the data is collected? What data is already available?
D M A
Preparing For The Analyze Phase Define Review Objectives 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 Xs drive variation of the Project Y
Collect Project Y data to identify patterns and performance trends and establish current baseline defect rate Collect segmentation factor data to be used for later analysis
Segmentation Helps Us Understand Variation In Project Y

Think ahead about how you plan to analyze the data. Collect additional information corresponding to the Project Y that may be helpful in subsequent analysis. Include possible external factors that may be useful for segmenting the total Y data set. Consider how to collect continuous data instead of discrete data. Thinking about segmentation factors now allows you to gather these conditions along with data on your Project Y so they dont have to be reconstructed after the measurement is completed. Segmentation factors include external variables outside our control that may explain patterns in the Project Y measures.
For example, in deal businesses where the Project Y is the cycle time of closing the deal, a potential segmentation factor is client size (large, medium, or small). .
D M A
Common Segmentation Factors Define Review Objectives Segmenting by external factors will help us identify the drivers of variation in the process Possible categories: Customer Product Market Time Geography Size of account Degree of dissatisfaction
What Data Is Needed? Project Y data, collected over X time period Examples:
1.
2. Y = On-time delivery of services Objective: Measure on-time delivery performance for customer service center Identify possible sources of poor or irregular performance. Data Required: Deviations from target date/time for individual orders (minutes/hours, early or late) Data for segmenting factors [e.g., customer, market, geography, time (day, month, hour), request type]
Y = Application cycle time Understand cycle time performance of credit approvals process. Identify potential sources of poor or irregular performances.
Objective:
Data Required: Physical cycle time data (collect several business cycles) for individual credit applications. Data on possible segmentation factors external to the process (e.g., customer information, time/date applications received with product type, processing center).
D M A
Common Segmentation Factors Define Review Objectives Other Categories Factor What type When Where Who Example Complaints, defects, problems Year, month, week, day Country, region, city, work site Business, department, individual, customer type, market segment
Tip: Begin with factors outside the process boxoften these are factors that were not considered when the process was first designed
In service-related processes, When elements to pay special attention to include: relationship to deadline (e.g., end of quarter), time/number in queue, or timing versus systems changes. Note: Your team will need to segment the data in several different ways in order to uncover where the most significant differences occur.
D M A
How To Collect Data For Segmentation Define Review Objectives Identify the factors for segmentation before you start collecting data Make sure the segmentation factors can be measured reliably Record the segmentation factors for each Y data point collected Segmentation factors are typically easy to collect, so collect more segmentation factors rather than fewer
D M A
Step 1: Breakout (5 Minutes) Define Review Activity Objectives

Objective Determine Segmentation Method
Instructions
For your Project Y:

Brainstorm a list of segmentation factors Remember to also segment on unlikely parameters
Time
5 Minutes
D M A
Step 2: Develop Operation Definitions and Procedures Define Review Objectives Clearly specify variables to be collected: Operational definitions for all metrics Specific descriptions of how to take the measurement Specify the details of the data collection process: How to collect the data How to record the data The period of time for data collection The sampling plan to be followed
To help ensure consistency, clearly define each metric and the process for collecting it. Attention to these details will help ensure that the data you collect will give you an accurate picture of the variation in your process.
D M A
Operational Definitions Define Review Objectives Defining The Measure

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

Identifies what to measure Identifies how to measure it Makes sure that no matter who does the measuring, the results are consistent
Always Pilot Your Operational Definitions
D M A
Operational DefinitionsScale Define Review Objectives Of Scrutiny Choosing The Level Of Measurement Measure one scale or level smaller than what your customer measures For Example: If your customer measures cycle time in days, your scale of scrutiny would be hours If your customer measures cycle time in hours, your scale of scrutiny would be in minutes Scale of scrutiny may expose larger true variation
10
The scale of scrutiny is how finely you measure your process. Measuring one level smaller than your customer allows you to more fully understand and capture the variation in the process.
D M A
C
1
Sampling Define Review Objectives 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
x x x x x x x x x x x x x x x
N = 5,000 What is the average resolution time?
Sample
x x x
x x x x
n = 100 From the sample, we infer that the 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
When To Review Sample Objectives Define
When to sample Collecting all the data is impractical or too costly Data collection can be a destructive process When measuring a high-volume process When not to sample A subset of data may not accurately depict the process, leading to a wrong conclusion (every unit is unique e.g., structured deals)
Statistically Sound Conclusions Can Often Be Drawn From A Subset Of The Total Available Data
One of the first questions to ask is Do I need to sample? The major reason sampling is done is for efficiency reasons-it is often too costly or time consuming to measure all of the data. Sampling provides a good alternative to collect data in an effective and efficient manner. If the circumstances surrounding the data collection plan do not justify sampling, then sampling should not be done. This is often the case in low volume processes (e.g., deal processes).
D M A
Goal Of AReview Useful Sample Define Objectives Representative Samples Representative sample: All parts of the target population are represented (i.e., selected for measurement) equally The customers view is captured How to guarantee a representative sample: Design a sampling strategy Understand special characteristics of the population before sampling
Regardless of the situation, a sample must be representative of the population. For practical purposes a sample is representative if it accurately represents the target population. Consideration that may hinder collection of a representative sample include: The cost and ease of obtaining samples Time constraints Unknown characteristics of the population Samples that are not representative of a target population are called biased samples. Often, the biases are not recognized until the collected data has been analyzed.
D M A
Sample Size For Continuous Define Review ObjectivesData
How Do I Determine Sample Size? Sample size (n) depends on three things Level of confidence required for the result, How confident I am that the result represents the true population
Level of confidence increases as sample size increases
Precision or accuracy () required in the result, The error bars or uncertainty in my result
Precision increases as sample size increases
Standard deviation of the population (), How much variation is in the total data population
An estimate of standard deviation is needed to start
As standard deviation increases, a larger sample size is needed to obtain reliable results 2
1.96 n=
In this equation, 1.96 represents a 95% confidence level
Consider the following example: We want to estimate average call length in handling customer inquiries, and we want our estimate to be accurate to within + 1 minute. 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 minutes. We want to have 95% confidence that the estimate will be in the range of specified accuracy i.e., + 1 minute.
Therefore, from the statistical theory we can answer according to the formula:
1.96 n=
2
Where n = sample size, u = standard deviation and = degree of precision required. In our example, the required sample size is: n = [(1.96x5)/1] 2 = 96.04 or 96 samples
D M A
Common Segmentation Factors Define Review Objectives Other Categories Factor What type When Where Who Example Complaints, defects, problems Year, month, week, day Country, region, city, work site Business, department, individual, customer type, market segment
Tip: Begin with factors outside the process boxoften these are factors that were not considered when the process was first designed
In service-related processes, When elements to pay special attention to include: relationship to deadline (e.g., end of quarter), time/number in queue, or timing versus systems changes. Note: Your team will need to segment the data in several different ways in order to uncover where the most significant differences occur.
D M A
Calculating A Sample Size Define Review Objectives Continuous Data To Derive Sample Size Formula
CI X
Given Solve for n
X Z/2 * n X = Z /2 n Z /2 * n = n =
Standard error
( )
Z /2 *
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 Z/2 = Z score we usually set at 1.96 for 96% confidence. = estimated standard deviation
D M A
How To Estimate Deviation When It Is Unknown Define Review Standard Objectives 3 Ways To Estimate Use an existing Xbar or R chart = R /d2 where d2 is control chart factor
= UCL X /3
Collect a small pre-sample & calculate s (n = 30) Ask subject matter experts to take an educated guess at the plausible range of data
= (Highest known value - Lowest known value )/6
Control Chart Factor d2
n 2 3 4 5 6
d2 1.128 1.693 2.059 2.326 2.534
= R /d 2
Range or moving range chart should be in statistical control (stable) to use this estimate of
D M A
Sample Size For Continuous Define Review ObjectivesData (continued)

What Is The Performance For Delivery Time?
Population
Y = Delivery Time (Days)
1.96 n=
Calculate sample size (n) based on:

Precision () 95% confidence Level (1.96) Standard deviation ()
Sample
n Values
Calculate average ()
Conclusion: I know with 95% confidence that the population mean is +
Whenever samples are taken to estimate a population there will be differences between the true population values and the sample values. Through statistical theory we can determine the amount of variation we can expect in our estimates. This is known as a confidence interval. Confidence intervals are stated in terms of an interval and a confidence level. For example, when estimating average inquiry resolution time from samples, we may obtain a 95% 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.
Confidence intervals are important when precise estimates of populations are required, and the degree of precision in the estimate needs to be known. For those statistically inclined, an important statistical theory supporting confidence intervals is the central limits theorem. The central limit theorem states that regardless the shape of the population, sample averages will always be normally distributed and inversely proportioned to the square root of the sample size.
D M A
Finite Population Define Review Correction Objectives If You Have A Finite Population 1. Calculate sample size (n) 2. If n/N > .05 OR If n > N 3. Calculate n finite nfinite = n/(1+n/N)
Where n = sample size; N = population size
D M A
Sample Size For Discrete Data Define Review Objectives
10
How Do I Determine Sample Size? Sample size (n) depends on three things: Level of confidence required for the result, How confident I am that the result represents the true population
Level of confidence increases as sample size increases
Precision or accuracy () required in the result, The error bars or uncertainty in my result
Precision increases as sample size increases
Estimated proportion defective of the population (P)

An estimate at P is needed to start Sample size is maximized at P = 0.5
1.96 n= P(1 P )
2
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 .05 .50 .95 .02 .02 .02 n 456 2400 456
This calculation procedure is based on the binomial model and should only be used when nP > 5. As for continuous measurements, if the sample size (n) is more than 5% of the population size (N), the finite population correction should be used n (finite) = n/(1 + n/N) and n (finite) should be used as the sample size.
D M A
Sample Size For Discrete Data (continued) Define Review Objectives What Is The Defect Rate (P) Of A Process?
Population
11
Y = Proportion Defective
1.96 n= P(1 P )
Calculate sample size (n) based on:

Precision () 95% confidence Level (2) Estimated proportion defective (P)
Sample
n Values
Conclusion: I know with 95% confidence that the population proportion defective is P +
Calculate Proportion Defective (P)
Recalculate n* based on the calculated P. If the new required sample size (n*) is more than the number of samples taken, take (n*-n) samples and recalculate P based on the full sample size. If it is not practical to take more samples, then use the actual n and P to recalculate the actual precision ().
Minimum Sample Size

1.96 n= P(1 P )
2
Example We want to estimate the defect rate (P) within + 0.02 (I.e., = 0.02). We expect P to be approximately 0.05.
1.96 n= 0.05(1 0.05 ) = 456 0. 2
2
D M A
How To Estimate and Define Review P Objectives 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 ) n
12
= /2
Sample Size & P
n
0 .5 P 1
D M A
Sample Size Considerations Define Review Objectives Beyond The Formulas The formulas give an approximate sample size Dont 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
13
What Is The Impact On The Customer If Your Sample Size Is Not Representative Of The Process?
D M A
Sampling Exercise Define Review Objectives

Objective Instructions
Apply Sample-Size Formulas 1.
14
2. 3.
4.
5.
In your table-team, answer the assigned questions and be prepared to report your answers. Assume confidence level of 95%. Determine the sample size needed if the following is known. P = 0.20, = 0.0784 Give an estimated proportion defective guessed to be 5%, how many observations should we take to estimate the proportion defective within 2%? 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? We want to estimate average hold time at one of our cost centers within 2 seconds. We will assume hold time standard deviation is 10 seconds. How many calls do we need to sample?
Time
10 Minutes
D M A
Measurement System AnalysisObjectives Recognize that observed variation of a product/process includes the true variation of the product/process & the variation due to the measurement system Identify & describe possible sources of variation in a measurement process Describe the importance of a validated measurement system Describe the terms precision, accuracy & resolution in relation to MSA Use appropriate tools to validate measurement system, analyze, and interpret results
Gage R&R for continuous data Attribute R&R for discrete data
One of the objectives of the Measure Phase is to validate your measurement system. A Gage R&R Study will help us do this! The focus of this module is to review the methodology and tools to validate your measurement system. MSA = Measurement System Analysis Gage R&R = Gage Repeatability & Reproducubility
GE Capital, Inc., 2000 DMAIC GB M TX PG V 4.2.0
D M A
Possible Sources Of Variation
Inputs
Process Outputs
Inputs
Measureme nt Process
Observations Outputs Measuremen ts Data
Observed Process Variation
Actual Process Variation
Measurement Process Measurement Process Variation Variation Variation Variation due to due to gage gage Variation Variation due to due to operator operator Other Other sources sources
Long-term Process Variation
Shortterm Process Variation Accuracy Accuracy (Bias) (Bias)
(Environment (Environment , etc.) , etc.)
Precision Precision (Measureme (Measureme nt Error) nt Error)
Discriminatio Discriminatio nn (Resolution) (Resolution)
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 We will be evaluating the variation in our measurement systems. We will include the variation due to the operator and gage (Accuracy, Precision and Discrimination). When collecting data, we are seeing the process through the lens of our measurement system. We never really see the actual process variation-instead, we see the actual process variation and the measurement process variation. Therefore, the total observed variation can be broken down into 2 parts: the actual process variation and the variation (or measurement error) created by the measurement process. Why is MSA important?

Data is only as good as the process that measures it MSA identifies how much variation is present in the measurement process Understanding measurement variation is necessary for identifying true process variation and maximizing true Y improvements Without MSA, you run the risk of making decisions based on an inaccurate picture of your process MSA helps direct efforts aimed at decreasing measurement variation Excessive measurement variation distorts our understanding of what the customer feels
D M A
Measurement Error vs. Process Variation Call Taking Y= Time to Respond (Response Time)
Process Variation: Time to answer a question varies for different operators, locations, etc. Measurement Error: Error in capturing the time measurement due to vague definition of when to stop.
Application Processing Y= Time to Decision

Process Variation: Different types of applications vary in time to complete. Associates vary in their productivity. Measurement Error: Error in measurement time. Time stamp inconsistencies or application arrival times not recorded accurately.
Deal Approval Y = Time to Complete

Process Variation: Differing cycle times. Clients are treated differently. Measurement Error: Inconsistencies in start-up or set-up definitions.
Deal Approval Y = Agreeable Terms and Conditions

Process Variation: Deal value variation (net income generated). Measurement Error: Errors in recording net income.
Measurement Error vs Process Variation When conducting a Gage R&R, you will be looking at the Measurement Error and determining if it is at an acceptable level. We want to minimize the measurement error (or measurement process variation). By conducting 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 make process improvements. The Minitab results obtained from your Gage R&R will indicate where the largest sources of measurement variation are hiding. Your goal will be to fix or reduce this variation and then to re-run the Gage R&R. You cannot collect and/or use your data until the Gage R&R passes.
D M A
Planning Your MSA Questions to ask: What are you measuring? Who is measuring? What do you use to measure (gage)? How are you measuring? Do you always use the same procedure? Under what conditions are you measuring (6 ms)? What is the resolution of your measurement system? Does the customer measure the same way? Which analysis will you use (continuous, Attribute and/or Destructive)?
D M A
Performing An MSA An MSA can be conducted with Continuous or Discrete Data (we will review each method). When conducting the MSA, the following guidelines are suggested:
Choose 2-3 operators who normally perform the measurements. For continuous data, choose 10 parts/samples (e.g., applications, calls, deals, trucks, etc) to measure. The parts should be as dissimilar as possible (within the normal measurement range). These same, ten parts will be used throughout the entire MSA study. For discrete data, choose 30-40 parts/samples (higher is better). The parts must be labeled with a number. The part numbers will remain constant throughout the MSA. Each operator will measure the parts 2-3 times. A blank data collection form should be used for each trial. The Gage should be calibrated as per the calibration procedure prior to conducting the MSA.
Choosing Parts for your Study: The Parts are NOT randomly chosen. They should be chosen specifically for the study. For example, you need to choose parts that vary from one another (in size, weight, length of time, etc.) because one output in the MSA tells you if your measurement system can distinguish between parts. Measuring the Parts: The operators should measure each of the 10 parts and record them on a data collection form. On their second round of measurement, they should have a new blank form. Never allow the operator to see previous measurements or measurements from other operators. Analyzing the Results: After all the data is collected, you will run the MSA in Minitab (for Continuous data) or in Excel (for Discrete data). The remaining pages in this section explain the MSA output.
D M A
Measurement System Analysis

Five Components Of An MSA: Accuracythe differences between observed average measurement and a standard Repeatabilityvariation when one person repeatedly measures the same unit with the same measuring equipment
Reproducibilityvariation when two or more people measure the same unit with the same measuring equipment Stabilityvariation obtained when the same person measures the same unit with the same equipment over an extended period of time Linearitythe consistency of the accuracy across the entire range of the measurement system
Think in terms of shooting at a target. Accuracy: How well do you hit the target? Do you hit where you are aiming? Repeatability: Can you keep hitting the target in the same place? Reproducibility: Can another person with the same gun hit the target in the same place? Stability: If you come back to shoot at other times are you still able to shoot as accurately (consistency over time)? Linearity: Are you as accurate at 50 meters as you are at 250 meters?
D M A
Process Flow For MSA (Measurement System Analysis)
Conduct MSA
Take Preventive And Corrective Actions
OK? No Yes Continue Process Improvement
For Attribute R&R: 90% matches for each Repeatability, Reproducibility, and Accuracy analysis.
Examples of Corrective Actions:

Note: Attribute = Discrete
Review data and decisions Institute stopgap measures
Examples of Preventive Actions:

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

Measure gage over time to address stability Study linearity of gage
D M A
Attribute Data AR&Rs With Discrete Data, we will look at these things:
Repeatability: Do repeated measures match within an operator? Reproducibility: Do repeated measures match between operators? Accuracy: Do the measures match the standard value?
An MSA helps us determine if our measurement system must be improved, and if so, gives guidance as to how
Although it has been emphasized that it is best to find a Y that is associated with continuous data, this is not always possible. In some cases, the response will be discrete. The measurement system must still be validated. Attribute = Discrete AR&R = Attribute Repeatability and Reproducibility is the method well use to analyze Discrete Data.
D M A
AR&R Guidelines For Evaluation Repeatability: 90% of repeated measures within an operator match Reproducibility: 90% of the repeated measures across operators match Accuracy: 90% of the individual measures match the standard
Guidelines for determining how to design an AR&R study: 1. For accuracy: The sample size is the total number 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.)
The examples that follow balance all three characteristics of an AR&R study.
D M A
How To Determine Sample Size In The AR&R How many samples should I take for an AR&R Study? Typically, 30-40 samples should give you a good indication of your measurement system Ten samples may be enough if running 30-40 samples would be too costly or time consuming. Use your Business Sense and learn from the study. Always ensure you choose samples that represent typical measures in your process.
10
For example, if your process requires a Customer Service Representative to code customer complaints as A-K categories, you would need to ensure your samples represented each of those codes. You also should choose samples that clearly fit that category and also ones that may fall within a gray area. (if such samples exist.) Therefore, the AR&R study will show whether your current measurement system is adequate.
D M A
AR&R Data Sheet

'True' Operator 1 Sample Answer Trial1 Trial2 N N 1 N N N 2 N N N 3 N D D 4 D D D 5 D N N 6 N N D 7 D N N 8 N N N 9 N N N 10 N D D 11 D N N 12 D D D 13 D N N 14 N D D 15 D D D 16 N N N 17 N N N 18 N N N 19 N N N 20 N D D 21 D N N 22 N N N 23 N N N 24 N N N 25 N D D 26 D N N 27 N N N 28 N N N 29 N N D 30 D D D 31 D N N 32 N D N 33 N N N 34 N N N 35 N D D 36 D N N 37 N N N 38 N N N 39 N N N 40 N Trial3 N N N D D N N N N N D N D N D D N N N N D N N N N D N N N N D N N N N D N N N N Operator 2 Trial1 Trial2 N N N N N N D D D D N N D D D N N N N N D D D D D D N N D D N N N N N N D D N N D D D D D D N N N N D D N N N N N N D D D D N N D D N N N N D D N N N N N N D D Trial3 N N N D D N D D N N D D D N D N N N D N D D D N N D N N N D D N N N N D N N N D Operator 3 Trial1 Trial2 N N N N D N D D D N N N D D N N N N D N D D D D D D N N D D N N N N N N N N D D D D N N N N D D N N D D N N N N N N D D D D N N N N N N N D D D D N N N N N N N Trial3 N N N D D N D N N D D D D N D N N N N N N N N D N D N N N D D N N N N D N D N N
11
This data sheet uses 3 operators, 3 trials, and 40 units to measure repeatability, reproducibility and accuracy.
Repeatability of the measurement system is assessed 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.
An extra column is added to define the true state of the standard according to a subject matter expert. In this example, N = Non-defective and D = Defective.
D M A
AR&R Example: Repeatability

Sample 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 Operator 1 Tr1 Tr2 TR3 N N N N N N N N N D D D D D D N N N N D N N N N N N N N N N D D D N N N D D D N N N D D D D D D N N N N N N N N N N N N D D D N N N N N N N N N N N N D D D N N N N N N N N N N D N D D D N N N D N N N N N N N N D D D N N N N N N N N N N N N Operator 1 Match? Y Y Y Y Y Y N Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y N Y Y N Y Y Y Y Y Y Y 0.925 Operator 2 Tr1 Tr2 Tr3 N N N N N N N N N D D D D D D N N N D D D D N D N N N N N N D D D D D D D D D N N N D D D N N N N N N N N N D D D N N N D D D D D D D D D N N N N N N D D D N N N N N N N N N D D D D D D N N N D D N N N N N N N D D D N N N N N N N N N D D D Operator 2 Match? Y Y Y Y Y Y Y N Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y N Y Y Y Y Y Y Y 0.950 Operator 3 Tr1 Tr2 Tr3 N N N N N N D N N D D D D N D N N N D D D N N N N N N D N D D D D D D D D D D N N N D D D N N N N N N N N N N N N D D N D D N N N N N N N D D D N N N D D D N N N N N N N N N D D D D D D N N N N N N N N N N D N D D D D N N N N D N N N N N N Operator 3 Match? Y Y N Y N Y Y Y Y N Y Y Y Y Y Y Y Y Y N N Y Y Y Y Y Y Y Y Y Y Y Y Y N Y N N Y Y 0.800
12
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 nonmatches/120) from 1 and multiply by 100% to get the percent match for repeatability. 13/120 = .1083 did not match. The percent matched = 100 (1- .1083) = 89.17% Since 89.17% matched is very close to 90% matched, gage repeatability technically fails but is practically acceptable.
DMAIC GB M TX PG
V 4.2.0
D M A
AR&R Example: Reproducibility

Sample 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 Operator 1 Tr1 Tr2 TR3 N N N N N N N N N D D D D D D N N N N D N N N N N N N N N N D D D N N N D D D N N N D D D D D D N N N N N N N N N N N N D D D N N N N N N N N N N N N D D D N N N N N N N N N N D N D D D N N N D N N N N N N N N D D D N N N N N N N N N N N N Operator 1 Operator 2 Tr1 Tr2 Tr3 N N N N N N N N N D D D D D D N N N D D D D N D N N N N N N D D D D D D D D D N N N D D D N N N N N N N N N D D D N N N D D D D D D D D D N N N N N N D D D N N N N N N N N N D D D D D D N N N D D N N N N N N N D D D N N N N N N N N N D D D Operator 2 Operator 3 Tr1 Tr2 Tr3 N N N N N N D N N D D D D N D N N N D D D N N N N N N D N D D D D D D D D D D N N N D D D N N N N N N N N N N N N D D N D D N N N N N N N D D D N N N D D D N N N N N N N N N D D D D D D N N N N N N N N N N D N D D D D N N N N D N N N N N N Operator 3 Match? Y Y N Y N Y N N Y N Y N Y Y Y N Y Y N N N N N N Y Y Y Y Y N Y Y N Y N Y N N Y N 0.525
13
Reproducibility (plus repeatability) of the measurement system is assessed by determining the number of times operators match on the same unit. In this example: A Y in the Match? column indicates a match, and an N indicates a non-match. On Unit 1, for example, all three operators matched, while on Unit 3, only Operators 1 and 2 matched. There are 40 samples or 40 opportunities for a defect in reproducibility. 19/40 = .475 did not match. The percent matched = 100 (1-.475) = 52.50% matched. Since 52.50% matched is lower than 90% matched, gage reproducibility is not acceptable.
D M A
AR&R Example: Accuracy

Sample 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 # of N Std Value N N N D D N D N N N D D D N D N N N N N D N N N N D N N N D D N N N N D N N N N Operator 1 Tr1 Tr2 N N N N N N D D D D N N N D N N N N N N D D N N D D N N D D D D N N N N N N N N D D N N N N N N N N D D N N N N N N N D D D N N D N N N N N D D N N N N N N N N 5 2 Tr3 N N N D D N N N N N D N D N D D N N N N D N N N N D N N N N D N N N N D N N N N 4 Operator 2 Tr1 Tr2 N N N N N N D D D D N N D D D N N N N N D D D D D D N N D D N N N N N N D D N N D D D D D D N N N N D D N N N N N N D D D D N N D D N N N N D D N N N N N N D D 6 5 Tr3 N N N D D N D D N N D D D N D N N N D N D D D N N D N N N D D N N N N D N N N D 5 Operator 3 Tr1 Tr2 N N N N D N D D D N N N D D N N N N D N D D D D D D N N D D N N N N N N N N D D D D N N N N D D N N D D N N N N N N D D D D N N N N N N N D D D D N N N N N N N 5 4 Tr3 N N N D D N D N N D D D D N D N N N N N N N N D N D N N N D D N N N N D N D N N 4 Not Match? 0 0 1 0 1 0 2 2 0 2 0 3 0 0 0 3 0 0 3 2 1 3 3 3 0 0 0 0 0 2 0 0 3 0 1 0 1 1 0 3 40
14
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 nonmatches 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.
DMAIC GB M TX PG
D M A
AR&R Analysis Example: Accuracy (continued) To summarize the AR&R MSA Repeatability: 13/120 non-matched = 100 (1-.1083) = 89.17% (FailBut Close!) Reproducibility: 19/40 non-matched = 100 (1-.475) = 52.50% matched (Fail) Accuracy: 40/360 non-matched = 100 (1-.1111) = 88.89% matched (FailBut Close!)
15
Note: For Repeatability, 89.17% (and Accuracy 88.89%) we considered both of these as Passing due to the closeness to 90%. Use your judgment. If an improvement can be made, feel free to improve the measurement system. The 90% rate is an approximation not a hard limit.
D M A
Measurement System Analysis (MSA) Worksheet

Type of Gage R&R Conducted (Check One): Gage R&R (Continuous Data) Gage R&R Results 1) Two-Way ANOVA Part p-value Oper p-value Oper & Part p-value % Tolerance % Contribution % Study # Distinct Categories [ <30% ] [ <8% ] [ <30% ] [ >4 ]
16
Date Conducted: 1/2/01
Significant? Y N Y N Y N Pass? Y Y Y Y OK? Y Y Y Y N N N N N N N N
2) 3) 4) 5)
Graphical Output 1) 2) 3) 4)

Effective Resolution [ >50% ] Stability [R Chart] Consistency Between Xbar consistency Between Oper Systematic Shift [Oper/Part Inter. Plot] Attribute Gage R&R
1) % Repeatability [ >90% ] 89.17% OK 2) % Reproducibility [ >90% ] 52.5% Need to improve. See plan below 3) % Accuracy [ >90% ] 88.89% OK Gage R&R Pass? Y N, If NO: Plan for improvement: Will work on Reproducibility problem by conducting a team meeting
with the 3 operators and discussing the differences in their measurements. Suspect that additional training is needed for less experienced operators because the data showed less experienced operators answering differently. Will also investigate if there is anything unique with each of the samples where reproducibility was an issue.
D M A
Causes Of Measurement System Variation If you fail the Gage R&R, here are some factors that could cause measurement process variation (measurement error). Repeatability Operational definitions Maintain stability Reproducibility Operational definitions Consistent use of gage Training Varying work environment AUDIT-follow-up on training Human/physical characteristics Performance measures Unclear requirements
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).
17
D M A
Causes Of Measurement System Variation (continued) Accuracy Error in master Instrument used improperly by appraiser Operational definition Standard not understood
18
D M A
Other Considerations
19
Temporal Effects Gage R&R depends upon our ability to measure things multiple times If the item is an event, it may not happen the same way twice. For this reason, Gage R&R may be impossible. If the event is recorded, it may be possible to conduct a Gage R&R study If the event cannot be recorded, but multiple judges can observe at once, reproducibility can be estimated, but repeatability cannot
Video cameras or audio voice recordings can allow a Gage R&R to be conducted. Use todays technology to help you in validating your measurement system.
D M A
Other Considerations Temporal Effects Example 1 Many Olympic sports are judged based on a numerical score sheet filled out by a judge. The difference among different judges readings of one live event is a measure of reproducibility.
20
For this same case, repeatability can be estimated only if the competition is taped. A random sample of 10 performances shown to five judges, two times each, would allow estimates of both repeatability and reproducibility.
D M A
Other Considerations Temporal Effects Example 2 Call centers field service calls from around the world at centralized locations. Call duration and call quality are both recorded and tracked very closely. Call quality is calculated based on a scoring sheet filled out by a quality monitor. Calls are scored on a continuous scale from 1-100 and can be assumed to be continuous for the purposes of a Gage R&R calculation.
21
D M A
Other Considerations
22
Temporal Effects Example 2 (continued) In one monitoring scenario, monitors listen in live and score the calls. For a live listen it is impossible to calculate repeatability, but if two monitors listen at the same time, it is possible to calculate reproducibility. In another monitoring scenario, calls are recorded at random and scored at a later date. For this recorded scenario, both repeatability and reproducibility may be calculated.
D M A
Continuous Data MSA, First Step
23
Test-Retest Study
D M A
TestRetest Study
Best Practice Hint: Do Test-Retest before Gage R&Ra quick look at the situation.
24
Why: Determine the precision of the system, instrument, device, or gagewhere Precision = Measurement Error = Repeatability How: Repeatedly measure the same item Same conditions, operator, device, and location on itemsame, same, same Completely mount and dismount item for each measurementexercise 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 ( ) and standard deviation (s) of the repeated measurements.
TestRetest studies are sometimes called Calibration Tests. This study is done prior to a Gage R&R. A TestRetest Study will indicate if you have either a precision or accuracy problem. A TestRetest Study is typically run using a certified or working Standard. The same item is measured repeatedly.
D M A
TestRetest Study Guidelines Device precision should be less than 1/10 of the tolerance: s < 1/10 x tolerance*
25
If s exceeds 1/10 x tolerance*, then the measurement system is unacceptable because it lacks precision. Action is required to find and remove the sources of this error, including the replacement of the device. Device accuracy may be estimated if you know the true value of the test unit: Inaccuracy = Bias = X - True Value
Precision is a term used to describe the amount of measurement error (repeatability error) in a measurement system. Precision is also used to describe the term (delta) in confidence intervals. There it indicates 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: () S 1/10 (USL X) or () S 1/10 ( X-LSL). (2) use the natural boundary (for example, use 0 for a cycle time measurement) as the other spec. (Tolerance = USL-0).
D M A
Process Flow For MSA (Measurement System Analysis)
26
Conduct MSA
Take Preventive And Corrective Actions
OK? No Yes Continue Process Improvement
For continuous data well look at these major items: %Tolerance, % Contribution, % Study and # Distinct Categories.
MSA is a set of methods for estimating the current amount of variation in the Measurement System. Examples of Corrective Actions:

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

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

Measure gage over time to address stability Study linearity of gage
D M A
Precision & Accuracy

Target Analogy
==
Xbar
27
True Value = Bull's Eye
I. Precise, not accurate

Xbar
==
True Value
II. Accurate, not precise

Xbar
==
True Value
III. Precise and accurate
Precision: What kind of repeatability does the whole system have? What is the scatter? (standard deviation)
Accuracy: Does the average reading agree with the actual size of the part? (mean)
D M A
Measurement System
28
Inputs
Process
Outputs
Observations Inputs Measurement Outputs Measurements Process Data
Product Variation
Measurement Variation
Total Variation
(Observed variation)
Example #1
(Actual variation)
Example #2
2Actual (Part) + 2Meas. System = 2Observed (Total)

Measurement System Variability - Investigated through R&R Study
The measurement system will also contribute some 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 were observing in our process. The key will be to minimize the measurement variation so that we are mainly seeing process variation.
= variance
D M A
Types Of Variation Estimated By The Gage R&R
29
Equipment Variation (EV)

(Repeatability) (Sources of variation from within the process) Within GageWithin OperatorWithin Part/ProcessEtc. The variation introduced into the measurement process from within one or more elements of the measurement processsuch as: within operator variationwithin gage variationwithin part variationwithin method variation.
Appraiser Variation (AV)

(Reproducibility) (Source of variation from across the process) Across GagesAcross Operators Across Parts/ProcessEtc. The variation introduced into the measurement process by effects going across the measurement processsuch as different appraisersdifferent part configurationsdifferent checking methods.
Equipment variationvariation within a sample group. Appraiser variationvariation between sample groups.
D M A
Relationship Between EV, AV And R&R R&R is the Reproducibility (AV) and Repeatability (EV) of the Measurement System. It represents the total variation in the Measurement System.
30
R&R
EV
AV
2Equipment + 2Appraiser = 2Total (R&R)

EV = Equipment Variation AV = Appraiser Variation The variation due to the equipment and appraiser do not directly add up to determine the total (R&R) variation. There is an overlap between EV and AV. You can use the Pythagorean theorem (right angles) to add to the influence of each EV and AV to calculate the total variation.
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.
2= variance
D M A
Relating R&R To Specification Window How much of the tolerance is used up by the Measurement System variation?
31
99% (5.15 )
R&R
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
Were interested in how this variation compares to the tolerance with which we are working. If the measurement system takes up lots of variation, we dont have any room for process variation. Minitab calls this variation % Tolerance. This is one or our 4 key metrics in the Gage R&R Study.
D M A
Gage R&R For Continuous Data: Example GB Case Study: On Time Delivery of Shipments to Customers
32
Background: The dispatchers for Capital Logistics keep a record of the time when truck drivers radio in to report delivery of the shipment was made to the customer (this is a requirement from the customers). Since there are 3 dispatchers recording the delivery time and 10 truck drivers calling in, the Green Belt needs to validate these delivery times for repeatability, reproducibility & accuracy.
First, we will walk through the process to conduct a Gage R&R Study in Minitab, analyze the results and draw conclusions. We will use Capital Logistics study of the delivery of shipments to customers. Then, you will do an activity to conduct a Gage R&R Study in Minitab with a new set of data from the same Capital Logistics study.
D M A
Gage Reproducibility & Repeatability (GR&R) Study: Steps 1. Set up a Minitab Worksheet 2. Perform studycollect & enter Data. (use file: Gage R&R-Continuous Data.mtw) 3. Perform calculations & prepare Charts.
Perform Gage R&R StudyANOVA method (ANOVA = Analysis of Variance)
33
4. Analyzeinterpret & draw conclusions. 5. Investigate variation in measurement system, take action, make recommendationskeep, 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, well need to fix our measurement system prior to collecting any data. We want to ensure the variation we are observing is mainly process variation.
D M A
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&Rcontinuous data.mtw).
or does it cloud what I see?
D M A
Gage R&R For Continuous Data: Example
35
Step 1: Set up Minitab data sheet. GB asked 3 dispatchers to record the truck driver call-in time for the 10 different truck drivers. Data sheet for recording the data was set up in Minitab to keep track of the 10 trucks (Part/Truck), 3 dispatchers (Oper), 2 runs (Trial) and the Y of Difference from Target (Meas). Open the File: Gage R&RContinuous Data.mtw
D M A
Gage R&R For Continuous Data: Example (continued)
36
Step 2: Perform StudyCollect & Enter Data. Truckers called in to report delivery and dispatchers recorded the difference from target time. The data was collected and recorded on a data sheet and was input into Minitab (see Data column).
Minitab File: Gage R&R Continuous Data.mtw
Use Minitab file: Gage R&RContinuous 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/Truckindicates which of the 10 trucks 2. Operindicates which one of the 3 dispatchers 3. Measindicates the difference, in minutes, from the target time 4. Trialindicates the trial number when the operator took the measurement.
D M A
Gage R&R For Continuous Data: Example (continued) Step 3: Perform Calculations & Prepare Charts Perform Gage R&R StudyANOVA method
37
Now lets perform the Gage R&R study on this dispatching process example. ANOVA is the best methodit breaks down the overall variation into three categories:

In Minitab: Click on STAT > Quality Tools > Gage R&R Study In the Gage R&R Study dialog box: 1. Select Columns for part, operator & measurement. 2. Select the ANOVA method 3. Select Options: Enter Tolerance width (note: Tolerance was set to 20 minutes here). The customer wanted the delivery within +/- 10 minutes of the target time. 4. Click OK, twice.
part-to-part repeatability reproducibility
and.it breaks down reproducibility further into its components:

operator operator by part The standard value for a 2-sided specification is 5.15. This is the number of standard deviations needed to capture 99% of your process measurements. If you only have a 1-sided spec, you would use 2.575 (half of 5.15).
D M A
Gage R&R For Continuous Data: Example (continued) Minitab Output will generate the following items which we will use to determine if our measurement system is acceptable:
1. 2. 3. 4. 5. 6. Graphical Summary Two-Way ANOVA Table with Interaction (p-values) % Comparisons Table (% Contribution) % Comparisons Table (% Tolerance) % Comparisons Table (% Study) Discrimination Index
38
We will cover each of these 6 outputs in this section.
D M A
The Central Question In A Gage R&R Study

There are 2 ways to answer the question Is my measurement variation too big? (1) FUTURE STATUS: Will my measurement variation be too large when I reach the 6-Sigma goal (i.e., with small process spread, good capability)? (2) PRESENT STATUS: Is my measurement variation too large right now, compared to a realistic estimate of current process spread (which may be narrow next year)?
39
R&R R&R %Tolerance %Tolerance
(Ratio (Ratioof ofVariances) Variances)
R&R R&R %Contribution %Contribution
(Discrimination (Discrimination Index) Index)
Number Numberof of Distinct Distinct Categories Categories
(Ratio (Ratioof ofStdDevs StdDevs---AIAG) AIAG)
R&R R&R% %Study Study Variation Variation
Red
30% 30%
Yellow
8% 8%
4 4
30% 30%
Green
10% 10%
2% 2%
10 10
15% 15%
CAUTION: The magnitude of these %s are exaggerations
If the Gage R&R (as a percent of tolerance) is less than 10%, the MS is green, or acceptable; if Gage R&R is 10-30% , the MS is yellow, or marginal; if Gage R&R is greater than 30%, the MS is red, or unacceptable. This estimate may be appropriate for evaluating how well the measurement system can perform with respect to specifications. We can use it to look into the future and ask Will my measurement system variation be too large when I reach the 6 Sigma goal (with small process spread, good capability)? If Gage R&R (as a percent contribution) is less than 2%, the MS is green, or acceptable; if the Gage R&R is 2-8%, the MS is yellow, or marginal; if Gage R&R is greater than 8%, the MS is red, or unacceptable.
If the discrimination index (number of distinct categories) is greater than 10, the measurement system (MS) is green, or acceptable; if the discrimination index is 4-10, the MS is yellow, or marginal; if the discrimination index is less than 4, the MS is red, or unacceptable. The discrimination index is just a transformation of %R&R contribution (for ease of interpretation.) It addresses the same issues and contains the same information. If Gage R&R (as a percent study) is less than 15%, the MS is green, or acceptable; if Gage R&R is 15-30%, the MS is yellow, or marginal; if Gage R&R is greater than 30%, the MS is red, or unacceptable.
Note: the limits shown above can be derived from one another. Some rounding was done for ease of interpretation.
D M A
Rules Of ThumbAcceptable Ranges 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 acceptablemake 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?
40
RULES OF THUMB: 1. Less than 10% = Good 2. 10% to 30% = Fairimprovements possible; look to find opportunities in process and training. 3. Over 30% = Problemmeasurement system not acceptable; fix problem with training, improved instruments, correct measuring process, and/or improved operational definitions for data collection. The gage is not adequate for Product/Process Acceptance decisions.
D M A
Rules Of Thumb (continued) B. % Contribution (or Gage R&R StdDev):
41
GR&R Variance should be small compared to Part-to-Part Variance applies in cases where Tolerance Width is not meaningful, and %Tolerance is unavailablesuch as one sided specs. 1. % Contribution < 2%Measurement System acceptable 2. % Contribution 2%-8%Measurement System marginal 3. % Contribution > 8%Measurement System unacceptable
Questions to ask for validating the measurement system:

Are we capturing the correct data? Does the data reflect what is happening in the process? How big is the measurement error? Can we detect process improvement if and when it happens? What are the sources of measurement error? Are the measurements being made with measurement units which are small enough to properly reflect the variation present? Is the Measurement System stable over time Is the Measurement System capable for this study? How much uncertainty should be attached to a measurement when interpreting it? How do we improve the measurement system?

D M A
Rules Of Thumb (continued) 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 groupshigh/low, good/bad = 3 =>can see three groupshigh/mid/low 4 =>acceptable measurement system (higher is better)
42
D. R&R % Study
1. % Study less than 15%measurement system acceptable. 2. % Study 15% to 30%may be acceptable (marginal) 3. % Study over 30%not acceptable
D M A
Minitab Gage R&RSession Window Analysis Here are the Results:

% Comparison TableFound in our Minitab Session Window
Source Total Gage R&R Repeatability Reproducibility Oper Oper*Part/Truck Part-To-Part Total Variation %Contribution 10.67 Fails! 3.10 7.56 2.19 5.37 89.33 100.00 %Study Var 32.66 Fails! 17.62 27.50 14.81 23.17 94.52 100.00 %Tolerance 171.53 Fails! 92.54 144.43 77.76 121.70 496.41 525.21
43
Number of distinct categories = 4 pass!
We fail the Gage R&R!
% Tolerance is 171.53% which fails, based on our criteria.This shows us the % of the Tolerance which is being taken up by the gage. In this case, 171.53% of our tolerance is taken up by the variation of the gage. 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 % Contribution is 10.67% which fails our criteria. % Contribution is based on overall measure of gage variation (regardless of the specification).
% Study, probably the least important, (because % contribution tell us roughly the same thing) also fails based on our criteria. This gives us a measure of gage variation compared to 99% of a normal distribution. Number of distinct categories passes! This tells us the number of divisions the measurement system can accurately measure across the seen process variation. For a more detailed explanation of output, refer to the end of this section.
.

D M A
Minitab Gage R&RGraphical Analysis Now Lets Review the Graphical Summary:
Gage R&R (ANOVA) for Meas
Xbar Chart by Oper
110 100 90 80 70 60 50 40 30 0
1 2 3
44
Gage name: Date of study : Reported by : Tolerance: Misc:
Oper*Part/Truck Interaction
Average
3.0SL=87.96 X=80.75 -3.0SL=73.54 110 100 90 80 70 60 50 40 1 2 3 4 5 6 7 8 9 10
Oper
1 2 3
Sample Mean
Part/Truck
R Chart by Oper
15
1 2 3
By Oper
3.0SL=12.52 110 100 90 80 70 60 50 40
Sample Range
10 5 0 0
R=3.833 -3.0SL=0.00E+00
5
1 2 3
Oper
Components of Variation
500 400
By Part/Truck
%Total Var %Study Var %Toler 110 100 90 80 70 60 50 40
Percent
300 200 100 0 Gage R&R Repeat Reprod Part-to-Part
6
1 2 3 4 5 6 7 8 9 10
Part/Truck
Chart 4: The chart shows the 10 parts and the average measurement of those parts broken-out by operator. Ideally, all 3 lines should be on top of each other. We can see that there is some measurement variation in parts 4, 8 and 10. We should investigate why.
Chart 5: This chart shows the average reading for all parts by operator. The red circle is the average. This could show us if one operator is measuring higher/lower (on average) than the others.
Note: For more detailed explanation, refer to the end of this section
D M A
Next Steps For The Green Belt
45
Sit down with your BB and/or Mentor to ensure proper analysis After an initial look at the Gage R&R Results for this example, the Green Belt will investigate the following:
Look at the feasibility of changing the gage to a digital clock Talk with the operators to find out why they are measuring differently Can a spreadsheet be developed to calculate time automatically? Can we implement a driver call-in procedure to reduce variation? Why was it harder to measure parts 4, 8 and 10?
Continue on to see the GBs Next Steps
D M A
Causes Of Measurement System Variation If you fail the Gage R&R, here are some factors that could cause measurement process variation (measurement error). Repeatability
Operational definitions Calibrating too often Granularity of the measure (e.g., nearest hour, minute, second) Maintain stability
46
Reproducibility
Operational definitions Consistent use of gage Training Varying work environment AUDITfollow-up on training Human/physical characteristics Performance measures Unclear requirements
If the Gage R&R fails, you need to improve the measurement system before moving on. Use these factors above as places to begin looking for areas of variation in your measurement system. By improving some of these factors, your measurement system variation should be reduced, (or brought to an acceptable level).
D M A
Causes Of Measurement System Variation (continued) Accuracy

Error in master Instrument not calibrated Worn components/trend or drift in master Instrument used improperly by appraiser Calibrating too often Operational definition Range of the master
47
D M A
Causes Of Measurement System Variation Linearity

Instrument calibrated incorrectly Error in master Worn instrument
48
Stability
Error in master Worn instrument Instrument measuring wrong characteristic Instrument not calibrated properly Instrument used improperly by appraiser
D M A
Gage R&R Action Plan
49
Step 5: Investigate sources of variation in the measurement system and make recommendationskeep, improve or replace the measurement system. After completing the analysis of the measurement system, the Green Belt investigated the sources of variation. By observing the process, the GB found:
The % Tolerance is the largest issue (171.53%). With the current gage and measurement process in place, the gage is not acceptable. To address this issue the GB has done the following:
Changed the Gage Implemented a standardized call-in procedure Implemented a new spreadsheet
Explanation of Bullets:
The Dispatchers are using a clock on the wall that is not digital. Therefore, the GB implemented a new procedure to use the computer clock. All computers were synchronized and a procedure was implemented to have the system automatically synchronize clocks 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.
The GB also noticed that all the data was positive, (indicating that the deliveries were always late). Upon investigation, the GB found that the dispatchers did not know that they should enter both positive and negative numbers. A new spreadsheet was created to fix this issue
D M A
Gage R&R Action Plan Step 5 (continued)
50
The % Contribution also failed (10.67%). From the Minitab data, the GB found Reproducibility to be the larger issue and the interaction between certain trucks and certain operators was the issue. To address these issues, the GB has done the following:
The data indicates that the measuring of the arrival time for trucks 4 & 10 seem to vary from operator to operator. By going back and reviewing the data and talking to the operators, the following was found:
There was a typo on the recording of time from Operator 1 on truck 4. He accidentally typed in the wrong number. To address this problem, a new spreadsheet was developed and now the Dispatcher is asked to verify his data prior to saving it in the spreadsheet. For Truck 10, the Operators had a very difficult time hearing the recording because the Truck Driver was using a cell phone with a poor connection. The GB changed the procedure so that each Dispatcher was now required to repeat-back the time given by the truck driver. The Truck Driver then confirms that the correct time is being recorded.
The data also indicated that Operator 2 seems to measure slightly lower than Operators 1 & 3. The GB found that Operator 2 was sitting on an angle and when he read the clock his time was distorted. By converting to the computer clock, this issue should be resolved.
D M A
C
51
Gage R&R For Continuous DataActivity (15 minutes)

Desired Outcome: Determine if the improved measurement system utilized by the Capital Logistics dispatchers is acceptable or not What Preparation How Choose a facilitator, scribe Who Timing
Conduct MSA
Complete 5 steps for a Gage R&R study and record results Use File: Gage R&RContinuous Data.mtw and use columns C6-C9
Team
10 mins.
Report Out
Complete Report Summary Present findings to class (see next page for report summary) Are there any additional items the GB should consider to improve the MSA even further?
Team
5 mins.
D M A

Type of Gage R&R Conducted (Check One): Gage R&R (Continuous Data) Gage R&R Results 1) Two-Way ANOVA Part p-value Oper p-value Oper & Part p-value % Tolerance % Contribution % Study # Distinct Categories [ 30%] [ 8%] [ 30%] [ 4]
Date Conducted:
52
Significant? Y N Y N Y N Pass? Y Y Y Y OK? Y Y Y Y N N N N N N N N
2) 3) 4) 5)
Graphical Output 1) 2) 3) 4) Effective Resolution [ >50% ] Stability [R Chart] Consistency Between Xbar consistency Between Oper Systematic Shift [Oper/Part Inter. Plot] Attribute Gage R&R
1) Repeatability [>90%] 2) % Reproducibility [>90%] 3) % Accuracy [>90%] Gage R&R Pass? Y N, IF NO: Plan for improvement:________________________________________ __________________________________________________________
D M A
Analyzing The Gage R&R Results Rules of Thumb 1. R&R (% Tolerance)
53
Less than 10%Measurement System is acceptable 10% to 30%maybe acceptablemake decision based on classification of characteristic, hardware application, customer input, etc. Over 30%Measurement System is not acceptable. Find problem, re-visit the fishbone diagram, remove root causes. Validate Measurement System again
2. % Contribution (or Gage R&R StdDev) : GR&R Variance should be small compared to Part-to-Part Varianceapplies in cases where Tolerance Width is not meaningful, and % Tolerance is unavailableSuch as one-sided specs.
1.% Contribution < 2%-Measurement System acceptable 2.% Contribution 2%-8%-Measurement System marginal 3.% Contribution > 8%-Measurement System unacceptable
D M A
Analyzing The Gage R&R Results (continued) 3. Number of Distinct Categories = a Signal-To-Noise Ratio = (StdDevparts/StdDevGR&R) X 1.41 and rounded
Guidelines: < 2no value for process control, parts all look the same = 2can see two groupshigh/low, good/bad = 3can see three groupshigh/mid/low > 4acceptable measurement system (higher is better)
54
4. R&R % Study
1. % Study less than 15%measurement system acceptable 2. % Study 15% to 30%may be acceptable (marginal) 3. % Study over 30%not acceptable
D M A
SummaryMeasurement System Analysis Variation in the measurement system will contribute to the observed variation in a process The resolution is the ability of the gage to see the variation in the process
The gage should be accurate: mean close to the true mean of the process, and precise: small variation.
55
Minimize the measurement process variation Use MSA (Gage R&R for Continuous Data or Attribute R&R for Discrete Data) to identify the amount of measurement system variation and process variation Understand how measurement error impacts your customer Measurement error is always a bigger deal than you think!
Make Sure Your MSA Is Examining The Actual Measurement System Itself
D M A
Measure Step 3Green Belt Project Activity (30 minutes)

Desired Outcome: Outline of a Data Collection Plan and MSA for your GB project What Your Individual GB Project How 1. Data Collection plan* Review and finalize your Data Collection Plan for your GB project Have you answered: What data will you need to collect? How will you collect it? From where? Who will collect the data? (can refer to the data collection worksheet) 2 . MSA* Develop an MSA for your project Y data List factors that might cause measurement system variation and how you would reduce the impact of those factors 15 mins. Timing 15 mins.
56
*Refer to the Self-Paced Workbook pp 43-49.
D M A
Gage R&R DetailsOptional Information
57
Pages 58-69 contain Optional/Additional information that will not be covered in the 3-Day Workshop
D M A
Full Gage R&R DetailsOptional Information Graphical Output

ANOVA simply stands for Analysis of Variance ANOVA simply stands for Analysis of Variance It is aa tool used to analyze the total variability among different sources. It is tool used to analyze the total variability among different sources. It partitions the total variation into buckets, and then allocates each bucket to each source It partitions the total variation into buckets, and then allocates each bucket to each source (part, operator, oper*part interaction). (part, operator, oper*part interaction).
Xbar Chart by Oper
Average
58
3
110 100 90 80 70 60 50 40 30 0
3.0SL=87.96 X=80.75 -3.0SL=73.54
110 100 90 80 70 60 50 40 1 2 3 4 5 6 7 8 9 10
Oper
1 2 3
Sample Mean
Part/Truck
R Chart by Oper
15
1 2 3
By Oper
3.0SL=12.52 110 100 90 80 70 60 50 40
Sample Range
10 5 0 0
R=3.833 -3.0SL=0.00E+00
5
1 2 3
Oper
500 400
By Part/Truck
Percent
6
1 2 3 4 5 6 7 8 9 10
Part/Truck
What do the graphs tell you about the measurement system? Is it acceptable? 1 > 50% of points should be outside control limits. 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.)
5 Scatter of individual measures should be equal by operator. 6 Scatter of individual measures should be equal by part (truck.)
D M A
Minitab Gage R&R ANOVA Graphical Output

Average of 2 measurements taken by operator 1.
59

Xbar Chart by Oper
110 100 90 80 70 60 50 40 30 0
1 2 3
Average
3.0SL=87.96 X=80.75 -3.0SL=73.54 110 100 90 80 70 60 50 40 1 2 3 4 5 6 7 8 9 10
Oper
1 2 3
Sample Mean
Oper 1
Oper 2
Oper 3
Part/Truck
110 100 90 80 70 60 50 40
R Chart by Oper
15
1 2 3
By Oper
3.0SL=12.52
Sample Range
10 5 0 0
R=3.833 -3.0SL=0.00E+00
Oper
500
By Part/Truck
Percent
Range of 2 measurements taken by operator 1.
400 300 200 100 0 Gage R&R Repeat Reprod Part-to-Part
Part/Truck
10
Xbar chart by Operator:

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

Xbar Chart by Oper
110 100 90 80 70 60 50 40 30 0
1 2 3
60
Average
3.0SL=87.96 X=80.75 -3.0SL=73.54 110 100 90 80 70 60 50 40 1 2 3 4 5 6 7 8 9 10
Oper
1 2 3
Sample Mean
Part/Truck
R Chart by Oper
15
1 2 3
By Oper
3.0SL=12.52 110 100 90 80 70 60 50 40
Sample Range
10 5 0 0
R=3.833 -3.0SL=0.00E+00
Oper
500 400
By Part/Truck
Percent
Part/Truck
10
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-toPart) 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). Well learn more later on % Contribution and % Tolerance.
D M A

Xbar Chart by Oper
110 100 90 80 70 60 50 40 30 0
1 2 3
61
Average
3.0SL=87.96 X=80.75 -3.0SL=73.54 110 100 90 80 70 60 50 40 1 2 3 4 5 6 7 8 9 10
Oper
1 2 3
Sample Mean
Part/Truck
R Chart by Oper
15
1 2 3
By Oper
3.0SL=12.52 110 100 90 80 70 60 50 40
Sample Range
10 5 0 0
R=3.833 -3.0SL=0.00E+00
Oper
500 400
By Part/Truck
Percent
Part/Truck
10
Operator by Part Interaction: Shows interaction between the operator and the part We want the lines to be on top of each other. Crossed lines could indicate the potential for interaction. Confirm suspected interactions with the p-value in ANOVA Table. We want all operators measuring the parts the very similarly (ie, getting the same measurement). Use this chart to help you look at particular parts which operators may be measuring differently. By reviewing this chart, we notice that the operators are measuring Part 4 & Part 10 differently. We would want to investigate why. This chart would also show us if one operator measures higher or lower than the others. It may indicate a Reproducibility issue. It appears that Operator 2 is measuring slightly lower than Operators 1 & 3. The Minitab Session Window can also tell us more.
By Operator: Main effect plot for Operator We want a flat horizontal line. This would indicate that it is unlikely the operator is having an effect...confirm with the p-value from ANOVA. This chart shows the measurements, by operator. Some dots may be on top of each other. Visually, you can see each operators measurements. By Part: Main effect plot for Part. We want to see lots of changes in slope. This would indicate that it is likely that the parts are having an effect. Confirm with the p-value from Anova. (Remember, we set up our study this way. We wanted differences in the parts.) More importantly, look for parts that have varying measurements. For instance, part 10 shows more variation in measurement than some of the other parts.
DMAIC GB M TX PG
V 4.2.0
D M A
Minitab Gage R&R ANOVA Table ANOVA TABLE

Two-Way ANOVA Table With Interaction Source DF SS 20587.1 480.0 036.7 387.5 22491.2 MS 2287.45 240.00 7.59 12.92 F 39.7178 4.1672 4.4588 P 0.00000 0.03256 0.00016
62
Sources of Variability
Part/Truck 9 Oper 2 Oper*Part/Truck 18 Repeatability 30 Total 59
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 summary. We will discuss each separately. Explanation of output for ANOVA Table: Part/Truck: p=0.00000; I would expect the parts to be significant because we chose truck travel routes that 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 its a repeatability or a reproducibility issue (Minitab does this for us in both the Comparisons Table and in the charts. Well show you later!)
Oper*Part/Truck: p=0.00016; This is the interaction between the operator and the parts (eg, when operators measure/record certain travel times). Since this is statistically significant, it is indicating that there may be some issues when some operators measure/record certain runs. We will look further into the Minitab output to find out which run and which operators.
D M A
Minitab Gage R&R % Tolerance % Comparisons Table

(Precision-to-Tolerance (Precision-to-ToleranceRatio) Ratio)
63
P/T P/TRatio Ratio
Source Total Gage R&R Repeatability Reproducibility Oper Oper*Part/Truck Part-To-Part Total Variation
%Contribution 10.67 3.10 7.56 2.19 5.37 89.33 100.00
%Study Var 32.66 17.62 27.50 14.81 23.17 94.52 100.00
%Tolerance 171.53 92.54 144.43 77.76 121.70 496.41 525.21
R&R R&Ras asaa% %of of Tolerance Tolerance
% Tolerance Acceptable Ranges

Not Acceptable Questionable Acceptable If you wanted to calculate %Tolerance by hand:
30% 10%
% T o le r a n c e =
5 .1 5
R & R ( M e a s . S y s t .)
T o le r a n c e
1 0 0 %
This shows us the % of the Tolerance which is being taken up by the gage. For instance, in this case, 171.53% of our tolerance is taken up by the variation of the gage. Refer to the next page for a visual explanation of % 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

If % Tolerance is unacceptable now, then once you reach a Six Sigma Process, the measurement variation will be too large compared to the process variation. In this case, the gage fails % Tolerance.
For this Gage: %Tolerance GRR = 171.53% This Gage Fails the Criteria for %Tolerance for the Total Gage RR Piece.
D M A
Minitab Gage R&R % Tolerance Graphically %Tolerance: Graphically

LSL USL
64
%Tolerance = 10%
%Tolerance = 30%
%Tolerance = 70%
Graphically, this shows the amount of the engineering tolerance used up just from the measurement system.
D M A
Minitab Gage R&R % Contribution % Comparisons Table

%R&R %R&RContribution Contribution(Total (Total Variation) Variation)
Variation due to the gage
65
Variation due to the parts
100%
Further Explanation:
%Contribution 10.67 3.10 7.56 2.19 5.37 89.33 100.00
%Study Var 32.66 17.62 27.50 14.81 23.17 94.52 100.00
%Tolerance 171.53 92.54 144.43 77.76 121.70 496.41 525.21
R&R R&Ras asaa% %of ofTotal TotalProcess Process Variation Variation
%Contribution 10.67 3.10 7.56 2.19 5.37 89.33 100.00
3.10 + 7.56 = 10.67
% Contribution Acceptable Ranges

Not Acceptable Questionable Acceptable
8.0% 2.0%
Now, we know that Reproducibility is our bigger issue (contributing more than Repeatability).
2.19 + 5.37 = 7.56
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

The measurement system (gage) is contributing to 10.67% of the total variation and the parts are contributing 89.33%. This Gage Fails the Criteria for % Contribution, but it is close to passing.
Note: If you dont have a tolerance, then, % Contribution becomes a very critical measurement for Gage R&R. It gives you a good indication of the measurement variation, now, compared to a realistic estimate of current process spread.
If you wanted to calculate %R&R Contribution by hand:
% R&R Contribution
2 R & R ( M e a s . S y s t .) 2 T o t a l
1 0 0 %
D M A
Minitab Gage R&R Discrimination Index Discrimination Index

Discrimination DiscriminationIndex Index
Number of Distinct Categories = 4
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 across the process variation. measure across the process variation. Indicates how aa gage can detect part-to-part variation ---- process shifts and Indicates how well well gage can detect part-to-part variation process shifts and improvement. improvement.
66
Number Numberof ofDistinct Distinct Categories Categories
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:

Std Dev (Parts) Std Dev (GRR) x 1.41
D is c rim =
2 2 2 T o ta l R &R ( 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 youve 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 youve designed and ran your study with similar parts (on purpose), this reading would be meaningless to you.
D M A
Minitab Gage R&R Key Indices

33Key KeyIndices Indicesfor forMeasurement MeasurementSystem SystemCapability Capability
11 22 33
67
(Precision-to-Tolerance (Precision-to-ToleranceRatio) Ratio)
P/T P/TRatio Ratio
%R&R %R&RContribution Contribution
Discrimination DiscriminationIndex Index
R&R R&Ras asaa% %of of Tolerance Tolerance

%Tolerance GRR = 171.53%
R&R R&Ras asaa% %of ofTotal Total Measure Variation Measure Variation
% R&R Contribution = 10.67%
Number Numberof ofDistinct Distinct Categories Categories

Discrimination Index = 4
The next question to be answered: What do I do now that the % Tolerance & % Contribution has failed? Well, this is what we have learned so far about the breakdown of % Tolerance and % Contribution:
For % Tolerance and % Contribution, we know the bigger issue is reproducibility. And, with Reproducibility, the bigger issue is the interaction between parts and operators. By looking at the graphical summary, it appears that trucks 4 and 10 may be the issue. Investigate why there is a difference. Is there something unique about those truck routes? Is it harder to measure the time for Trucks 4 & 10 due to time length? Should there be an Operational Definition or Procedure on how to measure? Was there a problem with calculating the time from target?
We also know that Operators and the Operator*Part/Truck Interaction is significant (from p-value). This is an indication that the operators are measuring differently. It also tells me there is some variation when certain operators measure certain trucks. Again, the graphs indicate which trucks and which operators.
I would investigate those areas mentioned above. Depending on your findings, you may need to implement a new training program with your operators on how to measure/record the time. You also may need to do something like having the computer measure the time when we measure parts similar to Parts 4 & 10. Once youve done those items, go back and re-run the Gage R&R using your implemented changes.
D M A
Gage R&R Acceptable Ranges
68
Acceptable Acceptable RangesANOVA RangesANOVA METHOD METHOD

This tells us how much variation the gage itself is contributing to the variation. This tells us how much of the tolerance is being taken up by the variation the gage. This tells us the ratio of each component of variation (the standard deviation for each component divided by the total standard deviation).
%Contribution Not Acceptable Questionable Acceptable
%Tolerance
% Discrimination Index 30% 15%
8.0% 2.0%
30% 10%
4 10
Provides the number of divisions that the Measurement System can accurately measure across the process variation. Indicates how well a gage can detect part-to-part variation--process shifts and improvement.
D M A

Type of Gage R&R Conducted (Check One): Gage R&R (Continuous Data) Gage R&R Results 1) Two-Way ANOVA Part p-value Oper p-value Oper & Part p-value 2) 3) 4) 5) % Tolerance % Contribution % Study # Distinct Categories
69
Date Conducted: 1/2/02
0.00000 0.03256 0.00016 171.53% 10.67% 32.66% 4
Significant? Y N Y N Y N Pass? [ 30%] [ 8%] [ 30%] [ 4 ] Y Y Y Y N N N N OK?
Graphical Output
1) Effective Resolution [ >50% ] Y N 2) Stability [R Chart] Y N 3) Consistency Between Xbar consistency Y N Between Oper 4) Systematic Shift [Oper/Part Inter. Plot] Y N _______________________________________________________________ Gage R&R Pass? Y N, If NO: Plan for improvement:
Look at the gage and possibly change to a digital clock. Review/Implement a driver call-in procedure. Suspect there is variation between drivers. Review/develop a new spreadsheet for calculating the time. Talk with operators to determine why Oper 2 measures differently than Operators 1 & 3. Talk with customers about the right specification, if possible. Train all operators and truck drives on new procedures. Re-run Gage R&R to verify improvements.
DMAIC GB M TX PG
D M A
D M A
Analyze Module Objectives Learn and apply the tools and concepts in the Analyze phase
You will learn the concepts and tools for the Analyze phase during this classroom session. After the classroom session, you can use the Six Sigma training CD to review how the Analyze phase is applied.
GE Capital, Inc., 2000 DMAIC GB N TX PG V 4.2.0
D M A
Analyze Phase Flowchart
A
ANALYZE PHASE OVERVIEW
Analyze 4: Establish Process Capability
Analyze 5: Define Performance Objectives
Analyze 6: Identify Variation Sources
D M A
C
3
The12-Step Process
Y Y X
X X
Y, X Simulation
D M A
C
4
Analyze Phase Overview

What is the Analyze phase? The Analyze phase is when your team:
Calculates baseline process capability for the process Defines the improvement goal for the project Analyzes historical data to identify the sources of variation
Why is the Analyze phase important? This phase is important because it clearly defines how well the process is currently performing and identifies how much the process will be improved. Steps involved in the Analyze phase Analyze 4: Establish process capability Analyze 5: Define performance objectives Analyze 6: Identify variation sources
D M A

Characterize the Root cause analysis process Critical Xs Stability Measure the Shape influence of the Center critical Xs on Variation Data Integrity the mean and variability MSA Capability Test Brainstorm ZBench ST & LT potential Xs Model Sampling plan Estimate
Collect data
D M A
D M A
Analyze 4Establish Process Capability
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 cant 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
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
Step 4.1: Graphically Analyze Data For Project Y
D M A
Review: Describing Variation Prior to Calculating Capability, we need to know: Key question #1StabilityVariation over time (Run Chart) How stable is the data? Key Question #2Shape, SpreadVariation for a period of time: Data Distributions (Graphical Analysis) What is the shape of the distributionsymmetrical, lopsided, twin peaks, long-tailed? (determination of normality) What is the central tendency (center or average) of the distribution? What is the variation (spread) of the distributionwide or narrow?
These attributes of your data will help determine which statistical calculations are better descriptors of your data set.
D M A
Graphically Analyze Data We can use the following example:

Data set: Holdtime.MPJ 1. First, look at Stability: STAT > Quality Tools > Run Chart
Approx p-value for clustering = 0.1139 Approx p-value for mixtures = 0.8861 Approx p-value for trends = 0.2883 Approx p-value for oscillator = 0.7117
p 0.05 indicates stability. Therefore, data is stable.
2. Secondly, look at Shape: STAT > Basic Statistics > Normality Test Anderson-Darling p-value = 0.885 p-value 0.05 indicates data is normal. Therefore, data is normal. 3. Third, look at Centering and Spread: STAT > Basic Statistics > Display Descriptive Statistics Again, p-value = 0.885. Data is normal. Mean = 8.5 and S = 0.10
p 0.05 indicates normal data. Therefore, will use mean and standard deviation to describe data.
D M A
Steps 4.2: Calculate Baseline Sigma
D M A
What Is Process Capability?
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 CTQs by looking at the probability that the process will produce a defect.
D M A
Common Metric For Comparison Process Purchase Order Generation Accounts Receivable Customer Service Supplier Delivery Performance 98% accuracy 33 days average aging 82% rated 4 or 5 on responsiveness 95% on-time delivery
Which process is performing best?
Its difficult to compare processes with very different kinds of measure. By expressing process capability as a sigma value we can compare different types of processes.
D M A
Data Analysis Roundup

Process Capability Tools and Terminology
e.g Light off
e.g Cycle Time, Length, Weight
e.g Light On
Discrete DiscreteData Data

Defects per Opportunity Defects per Million Opportunities Six Sigma Product Report
Continuous Continuous Data Data

Six Sigma Process Report: ZLong term ZShort term = ZBench =
reported yield
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 LT or SigmaLT. This is no longer standard usage to denote capability). ZST = ZBenchST = Sigma CapabilityST = ZBench ( or SigmaST is no longer recommended).
D M A
Process Capability Continuous Data Verify we have a normal distribution Calculate ZLSL and/or ZUSL Determine probability of a defective Determine ZBench
We will look at how to do each of these steps in detail!
D M A
Calculating ZContinuous 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.
Z=
1 2 3 4
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, youll need to use the Discrete, DPMO method.
D M A
Calculating Capability
X = 8.5 s = 0.1
xx
Probability Probability of ofaadefect defect less lessthan than LSL LSL Probability Probability of ofaadefect defect greater greater than thanUSL USL
11
LSL LSL
USL USL
ZLSL
ZUSL
Standard Deviations -4 Units of Measure 8.1
-3 8.2
-2 8.3
-1 8.4
0 8.5
3 8.8
4 8.9
8.6 8.7
ZUSL = USL - X = 8.7 - 8.5 = 0.2 = 2 s ZLSL = X - LSL = 8.5 - 8.2 = 0.3 = 3 s
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
0.1
0.1
0.1
0.1
D M A
Reading The Z Table

Yield
12
Probability Of A Defective
Example = .028066
Specification Limit
Z-score
0.20 .420740315 1.71 0.25 .401293634 1.76 .382088486 1.81 1.86 1.91 1.96 2.01 2.06 2.11 2.16 2.21 2.26 2.31 2.36 2.41 2.46 2.51 2.56 2.61 2.66 2.71 2.76 2.81 2.86 2.91 2.96 3.01 .043632958 .039203955 .035147973 .031442864 .028066724 .024998022 .022215724 .019699396 .017429293 .015386434 .013552660 .011910681 .010444106 .009137469 .007976235 .006946800 .006036485 .005233515 .004527002 .003906912 .003364033 .002889938 .002476947 .002118083 .001807032 .001538097 .001306156 3.22 3.27 3.32 3.37 3.42 3.47 3.52 3.57 3.62 3.67 3.72 3.77 3.82 3.87 3.92 3.97 4.02 4.07 4.12 4.17 4.22 4.27 4.32 4.37 4.42 4.47 4.52
Units of Measure
.000640954 .000537758 .000450127 .000375899 .000313179 .000260317 .000215873 .000178601 .000147419 .000121399 .000099739 .000081753 .000066855 .000054545 .000044399 .000036057 .000029215 .000023617 .000019047 4.73 4.78 4.83 4.88 4.93 4.98 5.03 5.08 5.13 5.18 5.23 5.28 5.33 5.38 5.43 5.48 5.53 5.58 5.63 .000001153 .000000903 .000000705 .000000550 .000000428 .000000332 .000000258 .000000199 .000000154 .000000118 .000000091 .000000070 .000000053 .000000041 .000000031 .000000024 .000000018 .000000014 .000000010
To the right of the Z-score you will note the tail area
Z = 1.91
1.35 1.40 1.45 1.50 .088507862 .080756531 .073529141 .066807100
Table Of Area Under 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 well need to eventually calculate Process Capability.
D M A
z
(Values of Z from 0.00 to 4.99)
SingleTail Z Table
13
Z
0.00 0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 0.90 1.00 1.10 1.20 1.30 1.40 1.50 1.60 1.70 1.80 1.90 2.00 2.10 2.20 2.30 2.40 2.50 2.60 2.70 2.80 2.90 3.00 3.10 3.20 3.30 3.40 3.50 3.60 3.70 3.80 3.90 4.00 4.10 4.20 4.30 4.40 4.50 4.60 4.70 4.80 4.90
0.00 5.00e-001 4.60e-001 4.21e-001 3.82e-001 3.45e-001 3.09e-001 2.74e-001 2.42e-001 2.12e-001 1.84e-001 1.59e-001 1.36e-001 1.15e-001 9.68e-002 8.08e-002 6.68e-002 5.48e-002 4.46e-002 3.59e-002 2.87e-002 2.28e-002 1.79e-002 1.39e-002 1.07e-002 8.20e-003 6.21e-003 4.66e-003 3.47e-003 2.56e-003 1.87e-003 1.35e-003 9.68e-004 6.87e-004 4.83e-004 3.37e-004 2.33e-004 1.59e-004 1.08e-004 7.23e-005 4.81e-005 3.17e-005 2.07e-005 1.33e-005 8.54e-006 5.41e-006 3.40e-006 2.11e-006 1.30e-006 7.93e-007 4.79e-007
D M A
z
(Values of Z from 5.00 to 9.99)
SingleTail Z Table
Z 5.00 5.10 5.20 5.30 5.40 5.50 5.60 5.70 5.80 5.90 6.00 6.10 6.20 6.30 6.40 6.50 6.60 6.70 6.80 6.90 7.00 7.10 7.20 7.30 7.40 7.50 7.60 7.70 7.80 7.90 8.00 8.10 8.20 8.30 8.40 8.50 8.60 8.70 8.80 8.90 9.00 9.10 9.20 9.30 9.40 9.50 9.60 9.70 9.80 9.90 0.00 2.87e-007 1.70e-007 9.96e-008 5.79e-008 3.33e-008 1.90e-008 1.07e-008 5.99e-009 3.32e-009 1.82e-009 9.87e-010 5.30e-010 2.82e-010 1.49e-010 7.77e-011 4.02e-011 2.06e-011 1.04e-011 5.23e-012 2.60e-012 1.28e-012 6.24e-013 3.01e-013 1.44e-013 6.81e-014 3.19e-014 1.48e-014 6.80e-015 3.10e-015 1.39e-015 6.22e-016 2.75e-016 1.20e-016 5.21e-017 2.23e-017 9.48e-018 3.99e-018 1.66e-018 6.84e-019 2.79e-019 1.13e-019 4.52e-020 1.79e-020 7.02e-021 2.73e-021 1.05e-021 4.00e-022 1.51e-022 5.63e-023 2.08e-023 0.01 2.72e-007 1.61e-007 9.44e-008 5.48e-008 3.15e-008 1.79e-008 1.01e-008 5.65e-009 3.12e-009 1.71e-009 9.28e-010 4.98e-010 2.65e-010 1.40e-010 7.28e-011 3.76e-011 1.92e-011 9.73e-012 4.88e-012 2.42e-012 1.19e-012 5.80e-013 2.80e-013 1.34e-013 6.31e-014 2.96e-014 1.37e-014 6.29e-015 2.86e-015 1.29e-015 5.74e-016 2.53e-016 1.11e-016 4.79e-017 2.05e-017 8.70e-018 3.65e-018 1.52e-018 6.26e-019 2.55e-019 1.03e-019 4.12e-020 1.63e-020 6.39e-021 2.48e-021 9.53e-022 3.63e-022 1.37e-022 5.10e-023 1.88e-023 0.02 2.58e-007 1.53e-007 8.95e-008 5.19e-008 2.98e-008 1.69e-008 9.55e-009 5.33e-009 2.94e-009 1.61e-009 8.72e-010 4.68e-010 2.49e-010 1.31e-010 6.81e-011 3.52e-011 1.80e-011 9.09e-012 4.55e-012 2.26e-012 1.11e-012 5.40e-013 2.60e-013 1.24e-013 5.86e-014 2.74e-014 1.27e-014 5.82e-015 2.64e-015 1.19e-015 5.29e-016 2.33e-016 1.02e-016 4.40e-017 1.88e-017 7.98e-018 3.35e-018 1.39e-018 5.72e-019 2.33e-019 9.40e-020 3.76e-020 1.49e-020 5.82e-021 2.26e-021 8.66e-022 3.29e-022 1.24e-022 4.62e-023 1.70e-023 0.03 2.45e-007 1.45e-007 8.48e-008 4.91e-008 2.82e-008 1.60e-008 9.01e-009 5.02e-009 2.77e-009 1.51e-009 8.20e-010 4.39e-010 2.33e-010 1.23e-010 6.38e-011 3.29e-011 1.68e-011 8.48e-012 4.25e-012 2.10e-012 1.03e-012 5.02e-013 2.41e-013 1.15e-013 5.43e-014 2.54e-014 1.17e-014 5.38e-015 2.44e-015 1.10e-015 4.87e-016 2.15e-016 9.36e-017 4.04e-017 1.73e-017 7.32e-018 3.07e-018 1.27e-018 5.23e-019 2.13e-019 8.58e-020 3.42e-020 1.35e-020 5.29e-021 2.05e-021 7.86e-022 2.99e-022 1.12e-022 4.18e-023 1.54e-023 0.04 2.33e-007 1.37e-007 8.03e-008 4.65e-008 2.66e-008 1.51e-008 8.50e-009 4.73e-009 2.61e-009 1.43e-009 7.71e-010 4.13e-010 2.19e-010 1.15e-010 5.97e-011 3.08e-011 1.57e-011 7.92e-012 3.96e-012 1.96e-012 9.61e-013 4.67e-013 2.24e-013 1.07e-013 5.03e-014 2.35e-014 1.09e-014 4.97e-015 2.25e-015 1.01e-015 4.49e-016 1.98e-016 8.61e-017 3.71e-017 1.59e-017 6.71e-018 2.81e-018 1.17e-018 4.79e-019 1.95e-019 7.83e-020 3.12e-020 1.23e-020 4.82e-021 1.86e-021 7.14e-022 2.71e-022 1.02e-022 3.79e-023 1.39e-023 0.05 2.21e-007 1.30e-007 7.60e-008 4.40e-008 2.52e-008 1.43e-008 8.02e-009 4.46e-009 2.46e-009 1.34e-009 7.24e-010 3.87e-010 2.05e-010 1.08e-010 5.59e-011 2.88e-011 1.47e-011 7.39e-012 3.69e-012 1.83e-012 8.95e-013 4.34e-013 2.08e-013 9.91e-014 4.67e-014 2.18e-014 1.00e-014 4.59e-015 2.08e-015 9.33e-016 4.14e-016 1.82e-016 7.92e-017 3.41e-017 1.46e-017 6.15e-018 2.57e-018 1.07e-018 4.38e-019 1.78e-019 7.15e-020 2.85e-020 1.12e-020 4.38e-021 1.69e-021 6.48e-022 2.46e-022 9.22e-023 3.43e-023 1.26e-023 0.06 2.10e-007 1.23e-007 7.20e-008 4.16e-008 2.38e-008 1.35e-008 7.57e-009 4.21e-009 2.31e-009 1.26e-009 6.81e-010 3.64e-010 1.92e-010 1.01e-010 5.24e-011 2.69e-011 1.37e-011 6.90e-012 3.44e-012 1.70e-012 8.33e-013 4.03e-013 1.94e-013 9.20e-014 4.33e-014 2.02e-014 9.30e-015 4.25e-015 1.92e-015 8.60e-016 3.81e-016 1.68e-016 7.28e-017 3.14e-017 1.34e-017 5.64e-018 2.36e-018 9.76e-019 4.00e-019 1.62e-019 6.52e-020 2.59e-020 1.02e-020 3.99e-021 1.54e-021 5.89e-022 2.23e-022 8.36e-023 3.10e-023 1.14e-023 0.07 1.99e-007 1.17e-007 6.82e-008 3.94e-008 2.25e-008 1.27e-008 7.14e-009 3.96e-009 2.18e-009 1.19e-009 6.40e-010 3.41e-010 1.81e-010 9.45e-011 4.90e-011 2.52e-011 1.28e-011 6.44e-012 3.21e-012 1.58e-012 7.75e-013 3.75e-013 1.80e-013 8.53e-014 4.01e-014 1.87e-014 8.60e-015 3.92e-015 1.77e-015 7.93e-016 3.51e-016 1.54e-016 6.70e-017 2.88e-017 1.23e-017 5.17e-018 2.16e-018 8.93e-019 3.66e-019 1.48e-019 5.95e-020 2.37e-020 9.31e-021 3.63e-021 1.40e-021 5.35e-022 2.02e-022 7.57e-023 2.81e-023 1.03e-023 0.08 1.89e-007 1.11e-007 6.46e-008 3.72e-008 2.13e-008 1.20e-008 6.73e-009 3.74e-009 2.05e-009 1.12e-009 6.01e-010 3.21e-010 1.69e-010 8.85e-011 4.59e-011 2.35e-011 1.19e-011 6.01e-012 2.99e-012 1.48e-012 7.21e-013 3.49e-013 1.67e-013 7.91e-014 3.72e-014 1.73e-014 7.95e-015 3.63e-015 1.64e-015 7.32e-016 3.24e-016 1.42e-016 6.16e-017 2.65e-017 1.13e-017 4.74e-018 1.98e-018 8.17e-019 3.34e-019 1.35e-019 5.43e-020 2.16e-020 8.47e-021 3.30e-021 1.27e-021 4.85e-022 1.83e-022 6.86e-023 2.54e-023 9.32e-024 0.09 1.79e-007 1.05e-007 6.12e-008 3.52e-008 2.01e-008 1.14e-008 6.35e-009 3.52e-009 1.93e-009 1.05e-009 5.65e-010 3.01e-010 1.59e-010 8.29e-011 4.29e-011 2.20e-011 1.12e-011 5.61e-012 2.79e-012 1.37e-012 6.71e-013 3.24e-013 1.55e-013 7.34e-014 3.44e-014 1.60e-014 7.36e-015 3.35e-015 1.51e-015 6.75e-016 2.98e-016 1.31e-016 5.66e-017 2.43e-017 1.03e-017 4.35e-018 1.81e-018 7.48e-019 3.06e-019 1.24e-019 4.95e-020 1.96e-020 7.71e-021 3.00e-021 1.16e-021 4.40e-022 1.66e-022 6.21e-023 2.30e-023 8.43e-024
14
D M A
Abridged Process Sigma Conversion Table

ST Process Sigma 6.0 5.9 5.8 5.7 5.6 5.5 5.4 5.3 5.2 5.1 5.0 4.9 4.8 4.7 4.6 4.5 4.4 4.3 4.2 4.1 4.0 3.9 3.8 3.7 3.6 3.5 3.4 3.3 3.2 3.1 3.0 2.9 2.8 2.7 2.6 2.5 2.4 2.3 2.2 2.1 2.0 1.9 1.8 1.7 1.6 1.5 1.4 1.3 1.2 1.1 1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 Defects Per 1,000,000 3.4 5 8 10 20 30 40 70 100 150 230 330 480 680 960 1,350 1,860 2,550 3,460 4,660 6,210 8,190 10,700 13,900 17,800 22,700 28,700 35,900 44,600 54,800 66,800 80,800 96,800 115,000 135,000 158,000 184,000 212,000 242,000 274,000 308,000 344,000 382,000 420,000 460,000 500,000 540,000 570,000 610,000 650,000 690,000 720,000 750,000 780,000 810,000 840,000 860,000 880,000 900,000 920,000 Defects Per 100,000 0.34 0.5 0.8 1 2 3 4 7 10 15 23 33 48 68 96 135 186 255 346 466 621 819 1,070 1,390 1,780 2,270 2,870 3,590 4,460 5,480 6,680 8,080 9,680 11,500 13,500 15,800 18,400 21,200 24,200 27,400 30,800 34,400 38,200 42,000 46,000 50,000 54,000 57,000 61,000 65,000 69,000 72,000 75,000 78,000 81,000 84,000 86,000 88,000 90,000 92,000 Defects Per 10,000 0.034 0.05 0.08 0.1 0.2 0.3 0.4 0.7 1.0 1.5 2.3 3.3 4.8 6.8 9.6 13.5 18.6 25.5 34.6 46.6 62.1 81.9 107 139 178 227 287 359 446 548 668 808 968 1,150 1,350 1,580 1,840 2,120 2,420 2,740 3,080 3,440 3,820 4,200 4,600 5,000 5,400 5,700 6,100 6,500 6,900 7,200 7,500 7,800 8,100 8,400 8,600 8,800 9,000 9,200 Defects Per 1,000 0.0034 0.005 0.008 0.01 0.02 0.03 0.04 0.07 0.1 0.15 0.23 0.33 0.48 0.68 0.96 1.35 1.86 2.55 3.46 4.66 6.21 8.19 10.7 13.9 17.8 22.7 28.7 35.9 44.6 54.8 66.8 80.8 96.8 115 135 158 184 212 242 274 308 344 382 420 460 500 540 570 610 650 690 720 750 780 810 840 860 880 900 920 Defects Per 100 0.00034 0.0005 0.0008 0.001 0.002 0.003 0.004 0.007 0.01 0.015 0.023 0.033 0.048 0.068 0.096 0.135 0.186 0.255 0.346 0.466 0.621 0.819 1.07 1.39 1.78 2.27 2.87 3.59 4.46 5.48 6.68 8.08 9.68 11.5 13.5 15.8 18.4 21.2 24.2 27.4 30.8 34.4 38.2 42 46 50 54 57 61 65 69 72 75 78 81 84 86 88 90 92
15
Long-Term Yield 99.99966% 99.9995% 99.9992% 99.9990% 99.9980% 99.9970% 99.9960% 99.9930% 99.9900% 99.9850% 99.9770% 99.9670% 99.9520% 99.9320% 99.9040% 99.8650% 99.8140% 99.7450% 99.6540% 99.5340% 99.3790% 99.1810% 98.930% 98.610% 98.220% 97.730% 97.130% 96.410% 95.540% 94.520% 93.320% 91.920% 90.320% 88.50% 86.50% 84.20% 81.60% 78.80% 75.80% 72.60% 69.20% 65.60% 61.80% 58.00% 54.00% 50% 46% 43% 39% 35% 31% 28% 25% 22% 19% 16% 14% 12% 10% 8%
Note: Subtract 1.5 to get longterm Sigma level

D M A
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 influenced by whether we are looking at long-term data or short-term data. Long-Term: Data collected over a long enough period of time and over diverse enough conditions such that it is likely to contain some process shifts and other special causes. Short-Term: Data collected over a short enough period of time so that shifts and other special causes are unlikely
For now, and as a default, treat all data as long-term data. The convention for Six Sigma is to report short-term, as you will see later. Why do we report Short-Term Sigma? We want to standardize reporting of the best a process can do, given its current capability. Therefore, Short-Term Sigma tells us this.
D M A
Reporting Sigma Values Short-Term Sigma = Long-Term Sigma + Sigma Shift

If Shift is unknown, then assume 1.5 Assume that sigma calculated from project data is long-term sigma A rational subgrouping sampling scheme for data collection (in the Measurement Phase) must have been used if you are calculating a shift (other than using 1.5.)
17
It is possible to calculate short-term, but for most Green Belt projects we do not attempt to do so; instead we use the 1.5 sigma shift rule. In order to truly calculate short-term sigma, you have to use several rational subgroups. If you did not use a rational sub-grouping sampling scheme for data collection in the Measure phase, it will not be possible to calculate actual short-term sigma. (Therefore, add 1.5 to long-term sigma.)
Special Cause = Between subgroup variation due to assignable causes, non-random influences. Common Cause = Within subgroup variation, inherent in a process, random influences.
To ensure you are sampling* properly, these 3 items must be true:

Representative Data Enough Data (at least 20 subgroups) Measured with something that passes the Gage R&R
*Refer to the workbook for further details on sampling (Tools > Measure phase)
D M A
Principles Of Rational Subgrouping 1. Never knowingly subgroup unlike things together 2. Minimize variation within each subgroup
18
Group homogeneous units, within a logic, within a reason The Xbar shows differences between subgroups that are bigger than that shown within subgroups Subgroup frequency should reflect the process Use individuals with limited data Use subgroups when logical
3. Maximize variation between subgroups
4. Treat the chart in accordance with the use of the data

Rational subgrouping refers to grouping the data for analysis in a meaningful way to understand variation. Rational subgrouping attempts to select groups of data such that mainly common cause variation is within groups, and mainly special cause variation is between groups.
D M A
Generalizing The Correction

Six Sigma Centered
Process Capability
19
.0005 ppm
T T
.0005 ppm
LSL LSL
USL USL USL
Six Sigma Shifted 1.5
TT LSL
3.4 ppm
4.5
USL
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 LongTerm Capability Zero Sigma Short-Term Capability = -1.5 Sigma LongTerm Capability
D M A
The Universal Equation For Z . . . so what are the possibilities? Z= SL -
20
SL
=USL LSL
T (Target) (Mean)
Z =
st (short-term) lt (long-term)
st lt
and how do we choose the right one?

Z LT = SL LT
The larger the ZShift , the greater the control problem. 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.
ZST =
SL T ST
Short -term is the best the process can do. SL - T is the distance the specification is from the target. ZST is bigger than
ZShift = ZST Z LT
D M A
Z-Bench
Short-Term Long-Term
P(d)LSL P(d)USL
21
LSL
Z-Long-Term
Zlt = SL - lt
_ x
USL
SL - T st T - LSL st
Z-Short-Term
Zst =
Z-Bench-Long-Term
ZUSL= USL- lt ZLSL=
- LSL
Z-Bench-Short-Term
lt ZUSL= USL- T st ZLSL=
P(d)USL = from Z table
P(d)LSL = from Z table
P(d)USL = from Z table
P(d)LSL = from Z table
P(d)Total = P(d)USL + P(d)LSL ZB-lt = from Z table
P(d)Total = P(d)USL + P(d)LSL ZB-st = from Z table
In Six Sigma, it is common to report out process capability using ZBench. As shown previously, ZBench corresponds to yield, i.e., we look at the chance of being outside the specification limit on either side. ZBench is calculated for both short and long-term. The difference between the two is ZShift.
D M A
ActivityCalculating Process CapabilityContinuous Data What is the process capability for a process that has: Mean = 5 Standard Deviation = 2 Upper Spec. Limit = 9
22
You can use the hand calculation or you can open the spreadsheet: SIGMACAL.xls. Enter: Xbar: 5 S: 2 USL: 9 LSL: (leave blank) Note: This gives you ZST (the 1.5 shift was added to ZLT).
D M A
Graphically Analyze DataBreakout Activity (20 minutes)

Desired Outcome: Graphical Analysis of Capital Logistics Project Y Data Minitab File: GB Case Study.mtw What Run Chart How Use the Run Chart tool in Minitab to investigate the variation in the project Y data over time Use the Normal Probability Plot in Minitab to analyze the shape of the project Y data Use the Descriptive Statistics tool in Minitab to analyze the shape, normality, central tendency and spread of the project Y data Use the Minitab Six Sigma Process Report to calculate Process Sigma Solutions You can check your answers using the solution sheets on the following pages Who All Timing 5 mins.
23
Shape, Normality, Central Tendency And Spread
All
10 mins.
All
5 mins.
We always analyze the data this way: 1. Look at StabilityIs the process Stable? 2. Look at ShapeDo I have a normal distribution? 3. Look at the SpreadWhat measure of dispersion should I use? Recall from our Case Study: Time *(minutes early or late) = (Target Arrival Time) (Actual Arrival Time) Date 1 = Date of Shipment * Spec. for time = 60 min.
D M A
SolutionRun Chart Stability Calculations for Case study data.
24
First Step, Check Stability Stat > Quality Tools > Run Chart
D M A
SolutionRun Chart
25
1. Double click on C5
2. Double click on C1 Each Date is a subgroup 3. Click OK
D M A
SolutionRun Chart
Each Date is a column of data points.
26
Run Chart for time

100
time
Median
-100 2 7 12 17
Each column has its mean plotted as a red square
Subgroup Number
Number of runs about median: Expected number of runs: Longest run about median: Approx P-Value for Clustering: Approx P-Value for Mixtures: 10.0000 9.4706 4.0000 0.6050 0.3950 Number of runs up or dow n: Expected number of runs: Longest run up or dow n: Approx P-Value for Trends: Approx P-Value for Oscillation: 9.0000 11.0000 4.0000 0.1118 Spread 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, Im looking for stability. How do I analyze this chart?
Look at p-values. P > .05 indicates stability. If the process is unstable, look for the most recent stable period and analyze with Minitab. If there is not a most recent stable time period, investigate & remove special causes.
Note: To manipulate data in Minitab: MANIP > Subset worksheet > include/exclude row #s
D M A
SolutionNormality Testing
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.
27
Normal Probability Plot
.999 .99 .95
Probability
.80 .50 .20 .05 .01 .001 -100 0 100

Anderson-Darling Normality Test A-Squared: 0.130 P-Value: 0.983
time
Average: -4.40316 StDev: 43.1714 N: 506
The data is Normal by this test method if the P-value is > = .05
Second Step: Check Shape (Normality) Stat > Basic Statistics > Normality Test (Variable = Time) The results of the Anderson-Darling Normality Test tell you if the data is normal or not. A p-value of greater than or equal to 0.05 means the data is normal. A p-value of less than 0.05 means that the data is not normal. Step 6 will address p-values in more detail.
D M A
SolutionNormality Testing (continued)
28
Variable: time
Anderson-Darling Normality Test A-Squared: P-Value: Mean StDev Variance Skewness Kurtosis N Minimum 1st Quartile Median 3rd Quartile Maximum -8.174
-10 -5 0
0.130 0.983 -4.4032 43.1714 1863.77 7.65E-02 -1.2E-01 506 -114.021 -35.085 -4.956 23.421 120.046 -0.633 46.009 0.532
This data is normal .
-100
-60
-20
20
60
100
95% Confidence Interval for Mu 95% Confidence Interval for Sigma 40.665 95% Confidence Interval for Median -9.369
Third Step: Check Spread The graphical summary of descriptive statistics (Stat Basic StatisticsDescriptive.) 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 datas mean and standard deviation this may not be a fit to the actual distribution.
D M A
Minitab Six Sigma Process Report

MINITAB FILE: GB case study.mtw
29
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 lets look at how to calculate process capability for continuous data in Minitab, using the Six Sigma Process Report. The subgroup size can be entered in two different ways:

Size of each subgroup 10 (subgroup size must be constant) ID columnoper 50 (subgroup size may be constant or not)
Warning: If you choose to use a subgroup size of 1, the short-term values calculated in the Six Sigma Process Report will be invalid since there is not variation within the groups.
The Six Sigma Process Report is used to calculate the long-term and short-term z-values of your process. In order to be accurate in the calculation of the z-values, it is important that we enter all of the information available. If a target is available, then enter the target value for the calculation of the z-values. If the upper and lower specification limits are entered and the target is left blank, Minitab approximates the target value as the midpoint of the specific range. If only one specification limit is entered and the target is left blank, then Minitab approximates the target as the mean of the data. Hence, when the target is available, always input the given value to avoid approximations made by Minitab.
D M A

Report 1: Executive Summary
Process Performance
Actual (LT) Potential (ST)
30
Process Demographics
Date: Reported by: Project: Department: Process: Characteristic: Units: Upper Spec: Lower Spec: Nominal: Opportunity: 60 -60
LSL
USL
-100
1,000,000
100
100,000
10,000
Process Benchmarks
1000 100
Sigma
(Z.Bench)
0.97
0.98
10
PPM
1
166781 164350
10
15
Report 1 contains the essence:

Sigma potential (ST) = ZBench-ST: if you were to use just one number to indicate the capability of your process, it is this one. If a firm XY claims to be a 5 company, they talk about this value. If you only see ZST, it means ZBench-ST. It is a convention to benchmark other companies using this value. However, it is a good practice to report out both ZBench values, for the short and long-term. Sigma actual (LT) = ZBench-LT: this value corresponds to the actual performance of your process, and is easier to comprehend with PPM. 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.
Upper graph: show short (dotted) and long-term (full line) distributions. By definition, the short-term distribution is shown to be on target. In this case, the long-term distribution is slightly below the target. Lower graph: shows cumulative PPMs.
Note: When reporting ZST, determine if you will be using the shift calculated from your data or the standard 1.5 shift. If using the 1.5 shift, you will need to manually add the 1.5 to ZLT to get ZST. Minitab cannot accommodate a predetermined shift. (When in doubt, use the 1.5 shift and/or consult your BB/MBB).
D M A

Report 2: Process Capability for time
Xbar and S Chart
1. Between Subgroup Variation
40 30 20 10 0 -10 -20 -30 -40 -50
Subgroup
3.0SL=21.49
31
Capability Indices
ST Mean StDev Z.USL Z.LSL Z.Bench
3.0SL=61.28
LT
-4.4032 43.1714 1.4918 1.2878 0.9670 0.0098 0.067876 0.098905 0.166781 83.3219 166781
X=-4.403
-3.0SL=-30.29
0.0000 43.1471 1.3906 1.3906 0.9767 0.0098 0.082175 0.082175 0.164350 83.5650 164350 0.46 0.43
10
15
2. Within Subgroup Variation
80 70 60 50 40 30 20 10
Z.Shift P.USL P.LSL P.Total Yield PPM
S=42.70
-3.0SL=24.12
3. Because variation is not out of control, there is no LT Variation
Potential (ST) Capability

Process Tolerance
-129.508 129.508
I
I I I
Actual (LT) Capability

Process Tolerance
-133.981
I
I
Cp Cpk Pp Ppk
125.175
I
I I
0.46 0.43
4. This makes sense, we only have 1 months worth of data
-60
60
-60
60
Specifications
Specifications
Data Source: Time Span: Data Trace:
Report 2 is more detailed:

The table to the right side contains all the details of the calculation of ZBench for short and long-term. The graphs to the left side are called control charts. This subject will be covered in the control session of your training. In order to use Minitabs 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.
Note: When reporting ZST, determine if you will be using the shift calculated from your data or the standard 1.5 shift. If using the 1.5 shift, you will need to manually add the 1.5 to ZLT to get ZST. Minitab cannot accommodate a predetermined shift. (When in doubt, use the 1.5 shift and/or consult your BB/MBB).
D M A
Rules For Using The Six Sigma Process Report 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
32
Use the Decision Flow on the next page to help you!
D M A
Process Capability Decision Flow

Is your Process Stable (use Run Chart)? Yes Do you have Continuous Data? No Yes
33
Is the Distribution Normal?
Yes
Use Six Sigma> Process Report (Input Data, Specification Limits and Target as appropriate)
No Analyze for Special CausesShould not calculate process capability until the process is stable. Unstable process data can only be used to describe the capability of the current data set, not the future capability of the process.
No Do you have Rational Subgroups? No Use Z Bench-LT only and add 1.5 Shift to get Z Bench-ST
Six Sigma>Product Report, DPMO, L1 (Input D, U and O) Sigma Conversion Table gives ZST Z LT = Z ST - Z Shift (1.5) Product Report and L1 give Z Bench Z Bench = Z ST Z LT = Z Bench - Z Shift (1.5)
Yes Is the SChart incontrol? Yes No Use Z Bench-LT only and add 1.5 Shift to get Z Bench-ST
Note:
Z ST = Z LT + Zshift Z ST = Z Bench-ST Z LT = Z Bench-LT
Z Bench-LT , Z Bench- ST & Z Shift all valid on the Process Report
D M A
The Normal Curve And Capability Poor Design Capability

High Probability of Defective High Probability of Defective Good Design Capability
34
LSL
USL
Low Probability of Defective
Low Probability of Defective
LSL
USL
D M A
SummaryZ-Value
35
Basic statistical summaries, histograms, dotplots, boxplots, and run charts are used to visualize data and better understand a process The Z-score is a non-dimensional quantity that enables us to compare different processesit represents the process capability The Z-score is the number of standard deviations that will fit between the mean and the respective specification limit of a normal distribution The Z-score corresponds to yield, or the area under the curve inside the specification limits
Note on Capability:
Capability can be described as any of the following: ZBenchST = ZST = ST = Sigma CapabilityST = SigmaST = ZBench ZBenchLT = ZLT = LT = Sigma CapabilityLT = SigmaLT In the new BOK, ST and LT are not recommended ways to describe capability. ST = pooled standard deviation and LT = standard deviation.
D M A
SummaryContinuous Data Process Capability
36
The Minitab Six Sigma Process Report is used to describe capability with continuous data

Displays the actual capability relative to the target distribution By rationally subgrouping (subgroup size >1), long-term capability, short-term capability, and shift are calculated.
Rational Subgrouping refers to grouping the data for analysis in a meaningful way to understand variation. Rational Subgrouping attempts to select groups of data such that mainly common cause variation is within groups, and mainly special cause variation is between groups.

Special Cause = Between group variation, due to assignable causes, nonrandom influences Common Cause = Within group variation, inherent in a process, random influences
A subgroup that contains only common cause variation, or random variation, represents the short-term capability of the process or entitlement (ZBench)
D M A
Process Capability & YieldDiscrete Data
37
Calculate the distribution of defects, defects per million opportunity (DPMO) Utilize Z tables to convert DPMO to yield or sigma level Understand the difference between classical, Throughput Yield, and Rolled Throughput Yield Calculate submitted, observed, and escaping defect levels
Classical Yield (YC) = Final Yield Throughput Yield (YTP) = First Pass Yield Rolled Throughput Yield (YRT)
D M A
Definitions Unit (U) The number of parts, sub-assemblies, assemblies, or systems inspected or tested
38
Squares: 4 units
Opportunity (OP) A characteristic you inspect or test
Circles: 5 opportunities per unit
Defect (D) Anything that results in customer dissatisfaction. Anything that results in a non-conformance.
Black circles: 9 defects
Note: Black dots indicate a defect. Clarification DefectAny single non-conformance (defect=black dot) DefectiveA unit with 1 or more defects. (defective=any square with 1 or more black dots).
D M A
Formulas Defects per Unit

DPU = D/U 9/4 = 2.25
39
Total Opportunities
TOP = U*OP 4*5 = 20
Defects per Opportunity (Probability of a Defect)

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

DPMO = DPO*1,000,000 .45*1,000,000 = 450,000
Once we calculate DPMO, we go to the Abridged Sigma Table (next page) to get ZBenchST or ZBenchLT
D M A
ConvertingDPMO DPMOTo ToZ ZST Converting

ZST = Reported Sigma (Short-Term) Abridged Process Sigma Conversion Table
Long term D PMO 500,000 460,172 420,740 382,089 344,578 308,538 274,253 241,964 211,855 184,060 158,655 135,666 115,070 96,801 80,757 66,807 54,799 44,565 35,930 28,716 22,750 17,864 13,903 10,724 8,198 6,210 4,661 3,467 2,555 1,866 1,350 968 687 483 337 233 159 108 72 48 32 21 13 9 5 3.4 Actual Sigm a (long term ) 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2 2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9 3 3.1 3.2 3.3 3.4 3.5 3.6 3.7 3.8 3.9 4 4.1 4.2 4.3 4.4 4.5 Reported Sigm a (s hort term ) 1.5 1.6 1.7 1.8 1.9 2 2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9 3 3.1 3.2 3.3 3.4 3.5 3.6 3.7 3.8 3.9 4 4.1 4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 5 5.1 5.2 5.3 5.4 5.5 5.6 5.7 5.8 5.9 6
40
ZST = 1.6 for 450,000 DPMO
Z ST Z ST 2 3 4 5 6
DPMO DPMO 308,538 66,807 6,210 233 3.4
To Calculate Process Sigma:

Go to the Memory JoggerAbridged 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. Were always conservative. Therefore, report the lower value for ZST or ZLT. And, you cannot interpolate, so choose the lower sigma value.
D M A
DPMO Calculation Example

Errors Detected 28 Wrong amount 14 Wrong address 12 Improper accounting code
41
30 Invoices Mailed Late
Prepare Invoice
500 Preliminary Invoices
Review Invoice
Fix Errors
Mail Invoice
470 Invoices Mailed On-Time
446 Accurate Invoices
Customer CTQs

Invoice mailed on date specified Invoice is error free Correct address Correct amount
How would the customer measure the process performance? First, determine the unit, defect opportunities per unit, and the number of defects from each perspective. For this example, assume the process of reviewing and correcting invoices is 100% effective. Use the worksheets on the following pages.
D M A
Calculating Process Sigma Discrete Data Method
42
1. Number of Units processed 2. Number of Defect Opportunities Per Unit 3. Total number of Defects made (include defects made and later fixed)
500 U = __________
OP = __________ 3
72 D = __________
4. Solve for Defects Per Opportunity D DPO = U OP =
( 72 ) ( 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 units for this example. Based on the CTQs listed for this example, there are 3 defect opportunities per unitlate, 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 CTQs 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 CTQs. If we were to examine this from the perspective of our internal customer, the list of CTQs would be different. .
Now calculate the DPMO and determine the capability.
D M A
ExerciseCalculating Process Sigma Discrete Data
43
1. Number of Units processed 2. Number of Defect Opportunities Per Unit 3. Total number of Defects made
U = __________ 500
3 OP = __________ 30 D = __________
4. Solve for Defects Per Opportunity DPO = D 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 =
Lets verify our answer using Minitabs Product Report.

1.
First open a blank worksheet and enter 3 column headings and your data: Defects 30 Units 500 Opportunities 3
2.
Six Sigma > Product Report Defects = Defects Units = Units Opportunities = Opportunities Click OK Verify ZBench = ZST (from your calculations)
3. 4.
D M A
C
44
ActivityCalculating Process CapabilityDiscrete Data
Project: Billing Product Improvement CTQ: Complete Billing Information Defect: Information missing from bill Unit: A bill Opportunities for defect per unit: 3 (items of information that can be missing on each bill) Out of 50 bills, you find 20 defects What is the sigma level of the process? What is the DPMO value for this process?
Now lets look at how to calculate process capability for discrete data in Minitab, using the Six Sigma Product Report.
D M A
Six Sigma Product Report

45
D M A
Discrete Data Examples

1. Select Defects, Units 2, & Opportunities
46
2. Select OK.
D M A
Six Sigma Product Report Output
47
Report 7: Product Performance
Characteristic
Defs
Units
Opps
TotOpps
DPU
DPO
PPM
ZShift
ZBench
U D
1 89 506 1 506 0.176 0.175889 175889
DPMO
ZST
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:

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

Z = 2.4 ST
D M A
Yield
48
Four parts are made 1. After first inspection: 1 passed, 3 failed Rework 3 parts
2. After second inspection: 1 passed, 2 failed Rework 2 parts
3. After third inspection: 1 passed, 1 scrapped

What is the Yield of this Process?
D M A
Types Of Yield
49
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 = # parts making it through final inspections = 3 Total # Parts = 4 First-Time Yield YFT = # Parts that are acceptable after first inspection = 1 Total # Parts = 4 Probability of Single Unit Being Defect-Free (SDF) SDF = P(0) = e-DPU = e-2.25 = .1054 = 10.54% DPU = Defects per unit = 9/4 = 2.25
D M A
Extending The Concept A given process has two operations. Each operation has a throughput yield of 99 %. The rolled yield equals:
Process Centered Process Centered
50
Op 1 99%
Without Inspection or Test
Op 2 99%
Output 98%
. . . There is an 98% probability that any given unit of product could pass through both operations defect free.
Rolled Throughput Yield = YRT = 98%
When the defects of the subsequent operations are independent, rolled throughput yield (YRT) = .99 x .99 = .9808
D M A
Rolled Throughput Yield (YRT)
51
Receive parts from Supplier
Following Receiving Inspection and Line Fall-out...
99.53% Yield (YTP)
97.13% Yield (YTP)
From Machining Operations
At Test Stands on first attempt
94.52% Yield (YTP)
YRT = .9953*.9713 *.9452 = 91.38%
Right First Time
G. Reimer 12/21/94 - Charlotte NC
You can establish a yield for each step of your process. If you have multiple processes, each time you go through additional steps, you lose yield at the end of the process. This is called Rolled Throughput Yield. YRT is calculated by multiplying all the yields together. The more complex the process, the worse the YRT becomes. We waste a lot of time scooping up the spilled paint.
D M A
Quantification Of Defects
Knowing the observed defect level and the test effectiveness, we can estimate submitted defects and escaping defects. For example:
52
Inspection E = 0.8
Submitted DPU Level (DPUS ) Escaping DPU Level (DPUE )
DPUS = DPUO + DPUE DPUO = DPUS x E DPUE = DPUS x (1-E)
Observed DPU Level (DPUO )
Let n denote the number of defects going into an inspection operation. Let E denote the proportion likelihood of detecting a defect. Inspection catches E*n. The number escaping to the customer is (1-E)*n. If you submit parts with a DPUS level, inspection observes (catches) a DPUO level of E*DPUS = Observed DPU. The customer sees a product with an escaping DPUE = (1-E)*DPUS. Submitted DPUS = Observed DPUO + Escaping DPUE.
DMAIC GB O TX PG
Example If we have a process that is submitting 100 defects (DPUS =100) and our inspection process is 80% effective (E=.8) at finding these defects, then we could observe 80 (DPUO) and 20 would escape to the customer (DPUE).
E = 0.8
DPUS=100 DPUE=20
DPUO=80
D M A
Quantification Of Defects (continued)

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
53
Inspection E = 0.8 Submitted DPUS Level = 0.31 (calc) Escaping DPUE Level = 0.06 (calc)
Observed DPUO Level = 0.25 (given)

The only thing that is real is the DPU observedthat is what we caught at inspection. In this example DPUO = .25. A typical value for E = .9. Your operation will differ slightly; in this example E = .8. Review the calculations below: DPUS*.8 = DPUO DPUS = DPUO /.8 DPUE = DPUS- DPUO
D M A
Exercise (15 minutes)
54
Calculating Submitted, Observed and Escaping Defect Levels Given: The requirement is to ship the product with no more than one defect in 500 units shipped The final test is to be performed using automatic test equipment having an effectiveness of 0.95 Knowing the maximum level of escaping defects, we can estimate the maximum defect/unit level: Observed in the final test. (This sets a minimum throughput yield limit.) Submitted to the final test. (This sets a minimum on the combination of defects created in model assembly and from the escaping defect levels from prior tests.)
Use the worksheet on the following page to help you work through this exercise.
D M A
Exercise (continued)
Calculating Submitted, Observed and Escaping Defect Levels
55
Inspection E = 0.95
Submitted DPUS Level Escaping DPUE 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? ______
Observed DPUO Level
D M A
Answers To Exercise 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 0.05 1 - effectiveness 1 -E 2b. Maximum acceptable observed defects = DPUO = DPUS - DPUE = 0.040 - 0.002 = 0.038 DPUO 3. YTP = 1-.038 = .962 = 96.2%
56
D M A
SummaryDiscrete Data Process Capability
57
Define defects, units and opportunities with your team. Be sure the definitions make sense and are consistent with similar processes and customer definitions. Defects will be stated as defects per million opportunities. Discrete data is generally considered long-term data. For discrete data, Minitab Six Sigma product report is used to calculate capability from defects and opportunities Determine DPMO (which is long-term), then determine the corresponding Zvalue (ST capability)
D M A
SummaryZ-Shift Often, you must assume a shift value (default 1.5) to estimate short-term capability Our customers experience the long-term capability of the process To minimize shift, we need to reduce special cause variation
58
D M A
C
1
Analyze 5Define Performance Objectives
What does it mean to Define Performance Objectives? A performance objective is a statement of your project Ys 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.
5.1 Identify Performance Objectives.
GE Capital, Inc., 2000 DMAIC GB P TX PG V 4.2.0
D M A
C
2
How Do I Define My Performance Objectives? Use benchmarking Use other sources
D M A
C
3
Defining Performance Objectives With Benchmarking
If I benchmark, performance standards are based upon: Closing the gap with the competition Exceeded projected competitive performance Similar performance in dissimilar businesses Gathering best practices from multiple sources to become best in class Becoming as good or better than a substitute product/service
D M A
C
4
Defining Performance Objectives With Other Sources
If I dont benchmark, performance objectives are based upon: For a process with ZST 3 Sigma Capability, decrease % defects by 10x and for ZST > 3, decrease % defects by 2x If your process is in statistical control (Run Chart), the next improved performance objective comes from a capability investment as in facilities, equipment, digitization, etc. Corporate mandate Compliance/legal VOC data
D M A
C
5
Nature Of BenchmarkingReference Material
Benchmarking is the process of continually searching for the best methods, practices and processes, and either adopting or adapting their good features and implementing them to become the best of the best.
The remaining pages are for your reference on Benchmarking. There are different types of Benchmarking. These include: Competitive Benchmarking Product Benchmarking Process Benchmarking Best Practices Benchmarking
D M A
C
6
What It Is And What It Isnt

Benchmarking Is
A continuous process A process of investigation that provides valuable information A process of learning from others; a pragmatic search for ideas A time-consuming, laborintensive process requiring discipline A viable tool that provides useful information for improving virtually any business process
Benchmarking Isnt
A one-time event A process of investigation that provides simple answers Copying, imitating Quick and easy
D M A
C
7
Common Benchmarking Mistakes 1. Internal process(es) is unexamined 2. Site visits feel good, but dont 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 isnt assigned 8. The wrong benchmarking partner is selected 9. The effort fails to look outside the industry (outside the box) 10. No follow-up action is taken
D M A
C
8
Benchmarking Six-Step Process

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 6 leaders Post findings and/or visit report on local server/6 bulletin board Enter information on GE Intranet benchmarking project database
D M A
C
9
Sources Of Information Library Database Internal Reviews Internal Publications Professional Associations Industry Publications Special Industry Reports Functional Trade Publications Seminars Industry Data Firms Industry Experts University Sources Company Watches Newspapers Advertisements Newsletters Original Research Customer Feedback Supplier Feedback Telephone Surveys Inquiry Service Networks Worldwide Web
D M A
C
10
Capital Logistics Case Study Project Goal(s)

Reduction of deliveries that are too early or too late
Benchmarking Process
Identified 20 similar 3rd party Logistics providers from Logistician Magazine Selected 3PL (2) that provide supply chain solutions to the petroleum industry (Non-competitorsCapital Logistics strategically avoids supporting petroleum industry because of risks) Polled logistics managers (telephone polls) to determine delivery practices
Results
Identified one viable new delivery methodThe Dynamic Routing System
Computer system that selects routes based on individual driver performance for load types and regional deliveries
Invited Petro-hauz to Stamford, CT for product demo with Capital Logistics Engineering and Technology Teams
D M A
C
11
Benchmarking Example 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
60
25000
25000
0.002
0.002400
2400
1.500
4.320
D M A
C
12
SummaryAnalyze 5
A performance objective is determined by using Zshort-term, benchmarking, or defect reduction goals Benchmarking is a process of identifying best practices, measuring our own practices against those best practices, and adapting the appropriate best practices to our own processes Revenue & cost implications are also due for benefit analysis
D M A
C
1
Analyze 6Identify Variation Sources
What does it mean to Identify Variation Sources? In step 6 you develop a list of statistically significant Xs, chosen based on analysis of historical data. This list is then prioritized to identify those Xs 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 Xs before selecting the Critical (or Vital Few) Xs 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 (Xs). In other words, you can change the output of a process (Y) only by changing the input & process variables (Xs). Therefore, in order to improve products and processes, you must shift your focus from monitoring the outputs of a process (Ys) to optimizing the inputs to the process and correcting the root causes of defects (Xs). You should use data and process analysis to identify potential Xs, 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
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
C
2
Identify The Vital Few Transfer Function
(X)
Project Y
Relationship that explains Y in terms of X
Process variables
Understanding The Gives Insight Into The Vital Few Xs

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 (ftransfer function), we will be able to pinpoint the key Xs that drive the variation in Y.
D M A
C
3
Identify The Vital Few
Process
Input Measures (Xs)
Outputs (Ys)
Process Measures (Xs)
Definitions:

Process

Project Y (Dependent Variable) Product or service produced or delivered by the process Process Xs (Independent Variable) Those variables that influence the output and are generally controllable by those who operate the process Input Xs (Independent Variable) Materials and information used by the process to create the outputs. Inputs are often outside the control of the process operator.
Provides context for improvement Links outputs (Ys) to inputs (Xs) and process variables (Xs)

In Analyze 6 we will be analyzing upstream variables, or input and process variables (Xs), to determine how they affect output variables (Ys) and to what extent. Remember Y=f (X1, X2, ..., Xn)
or in words. The results we get (the Ys) are a function of the process and input variables (the Xs). In Analyze, we will study the process and the data in order to gain understanding of how it all fits together.
D M A
C
4
Analyze 6.1: Identify Possible Causes of Variation
D M A
Methods To Identify Possible Sources Of Variation

Methods To Identify Vital Xs
Graphical Analysis
Process Map 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 processgaining knowledge of what impacts the output. How do the various Xs impact the Ys? With this knowledge, we will pursue a solution that will improve the process. We have already learned about graphical analysis, process map analysis and cause and effect analysis in previous modules. In step 6, we will learn how to apply them in the Analyze phase.
D M A
Review: Graphical Analysis Looking For Patterns In Data Continuous Y Boxplot Scatterplot Histogram Discrete Y Pareto Chart
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 Xs. Lets do an activity to briefly review these tools.
D M A
Review ActivityWhich Graphical Analysis Tool Would You Use?

For each scenario described below, which tool would you use to investigate? For each scenario, what pattern would you look for to make your conclusions?
1. (a) A Six Sigma project is being conducted in the field to improve the cycle time for warranty repair returns. The warranty return cycle time was measured for a period of 6 weeks for 4 regions. The Green Belt wants to investigate whether there is any difference in average warranty repair cycle time between each of the regions. (b) The Green Belt would also like to know if there is any difference in the variance of the cycle time between each of the regions. 2. Checks Are Us is a payroll processing firm. Timecard errors are routinely monitored and recorded. A Black Belt investigating the errors wishes to determine if there are any differences in the number of errors between five of its major customers. The number of errors contained in a sample of 150 employees was recorded for five weeks. How would you proceed with stratifying the data to make this investigation? 3. Tungsten steel erosion shields are fitted to the low pressure blading in steam turbines. The most important feature of a shield is its resistance to wear. Resistance to wear can be measured by abrasion loss, which is thought to be associated with the hardness of steel. How would you investigate the relationship between hardness (the X) and abrasion loss (the Y)?
D M A
C
8
Graphical AnalysisBreakout Activity (15 minutes)

Desired Outcome: Identify possible causes of variation in the Capital Logistics case study data. File: Green Belt Case Study.mtw What Minitab Analysis How Use the appropriate graphical 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 Out Be prepared to share your results with the class All Who All Timing 15 mins.
Load types: D/H Drop & HookA driver unhooks their full trailer, and then hooks into an empty to take back. UnloadThe driver opens the trailer and allows the store to unload its freight, and brings back the trailer. To create Boxplots: Stat > Basic Statistics: > Display Descriptive Statistics variable: time click: by variable: load type or driver click Graph > Choose Boxplots To create Scatter Plots: Graph > Plot X: Distance Y: Time
D M A
Process Map Analysis Types Of Analysis Moments of truthWhat does the customer feel? Nature of workWhat is of value to the customer? Flow of workWhat makes the customer wait?
The purpose of Process Map Analysis is to:

View our process and its performance from the customers perspective Identify current process problems and opportunities Create a shared sense of urgency for improving the process clearly identify and document valueadded/nonvalue-added activities and process flow measures and characteristics.
The second, Nature of Work, is concerned with value analysis: understanding whether or not the work in your process is valued by the customer or by your internal culture. The third, Flow of Work, characterizes the work along the dimension of time. The two latter components are focused on the Profitability piece of the GE Capital Quality Vision.
There are three key components to the Process Map Analysis step: The first, Moments of Truth, ensures that we are Completely Satisfying Customer Needs through a customer-focused process improvement perspective.
D M A
Moments Of Truth A Typical Deployment Flow Chart

Initial Contact Customer Service/Produce
Moments of Truth
10
Delivery
Follow-Up
Intermediaries Or Front Line Backroom Activity Support Or Suppliers
A moment of truth is defined as anytime a customer draws a critical judgment, positive or negative, about the service, based upon a service experience (or lack thereof). Certainly every time we touch the customer is a moment of truth, but it isnt limited to these cases. From the customers view, the process is a set of service encounters. In this case, the customers believe the steps 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.
If youre a service person and you get it wrong at your point in the customers chain of experience, you are very likely erasing from the customers mind all the memories of the good treatment he or she may have had up until now. But if you get it right, you have a chance to undo all the wrongs that may have happened before the customer got to you. You really are the moment of truth.
D M A
Nature Of WorkValue Analysis

Value-Added Work
Steps That Are Essential Because They Physically Change The Product/Service, The Customer Is Willing To Pay For Them, And They Are Done Right The First Time.
11
Nonvalue-Added Work
Steps That Are Considered Non-Essential To Produce And Deliver The Product Or Service To Meet The Customers Needs And Requirements. The Customer Is Not Willing To Pay For Them.
Steps That Are Not Essential To The Customer, But That Allow The ValueAdding 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

REs are things in your process that are done more than one time (Redo, Recall, Reissue). REs usually cause you to loop back to an earlier point in your process. REs consume time, add to the complexity, and use additional resources.
OPTIONAL: Show the video: Time the dimension of Quality
D M A
Types Of NonvalueAdded Work
12
Internal Failure
Delay
External Failure
Preparation/Set-Up
Control/Inspection
Move
What Does The Customer Value?

Delay Steps where work is waiting to be processed at that step. Examples:

Internal Failure Steps that are related to correcting in-process errors due to failures in current or prior step in the process. Example:
Backlogs Queues Storage Bottlenecks
Rework External Failure Steps which relate to fixing errors in the product that the customer has found and has returned to you. Examples: Recall Control/Inspection Steps for internal process review often referred to as the checker-checking-the checker. Examples: Approval/review Inspection Bureaucracy
Customer Follow-up
Preparation/Set-Up Steps that prepare work for a subsequent activity in the process. Examples:

Entering into a computer Retrieve policies/pricing

Move Steps that entail the physical transport or transmit of outputs between activities in a process. Example:
Fax/mail
D M A
Flow Of Work
13
Process Time
Delay Time
Cycle Time
To analyze the flow of work, break the process flow down to its lowest component, and analyze the movement of the component through the process. For example, become an application and trace its progress through the process. It is similar to a process flowchart, but with the value-added and nonvalue-added steps clearly identified.
Terms Cycle Time The total time from the point in which a customer requests a good or service until the good or service is delivered to the customer. Cycle time includes process time and delay time. Process Time The total time that a unit of work is having something done to it other than time due to delays or waiting. It includes the time taken for value-added steps, internal failure, external failure, control, inspection, preparation/set-up, and more. Delay Time Total time in which a unit of work is waiting to have something done to it.
D M A
Flow Of WorkProcess Disconnects Gaps Redundancies Implicit or unclear requirements Inefficient hand-offs Conflicting objectives Common problem areas
14
Process disconnects can often be found where there are delays in the process. These are points in the process where work can be disrupted or delayed, or where defects can be created. They include:

Gaps: responsibility for a given step in the process is unclear, or the process seems to go off track. Redundancies: duplication of efforts such as when two people or groups approve a document. Redundancies occur when different groups take action that they are unaware is being done somewhere else in the process. Implicit or unclear requirements: operational definitions do not exist or clear differences exist in perspective or interpretation.
Inefficient hand-offs: no check is in place to assure the process is continuing without delays. For example, Department A sends something to Department B but neither has a way to know if it was properly received. Conflicting objectives: the goals of one group cause problems or errors for another. For example, one group is focused on process speed while another is oriented to error reductionthe result may be that neither group accomplishes its objectives. Common problem areas: occurs when steps are repeated in a variety of places in process. Noting these areas may provide insight into potential solutions.
Your team should encode these disconnects and highlight them directly on your process map.
D M A
C
15
Flow Of WorkBe The Unit

Unclear requirements 1. Receive application in mail and open envelope 2. Place application in mail slot Are we extending loan? No 19. Generate turndown letter
Redundancy
11. Make loan decision
Yes
Inefficient hand-off
10. Review for completeness and make decision 9. Queue application for credit review
12. Generate loan packet
3. Move application to Entry Dept. Unclear requirements
13. Place in out-box
18. Postman picks up outbound mail
4. Place application in in-box
8. Score application
Inefficient hand-off
14. Move to mailroom
17. Place in outbound mail basket
5. Retrieve application and review for completeness
Is application complete? No 6. Call to obtain necessary information
Yes
7. Enter application to computer system
15. Wait for postage
16. Post package or letter
Cycle time is the total time taken from the point at which the customer requests a good or service until the good or service is delivered to the customer. It is the sum of the process time, move time, inspect time, delay time, and storage time. Cycle time analysis often provides eye-opening insights. People seldom think of work in this context. When an Improvement Team presents cycle time data, management may become defensive. For example, when one manager at a commercial insurer found out it was taking 100 days, on average, to process the commercial line of a new business, she resisted the analysis. After heated discussion, she suggested that the period 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 averaged 103 days!
Constructing a time line helps identify areas of potential improvement by illustrating the time required for each step in a process. Remember: Time data may not be available for all steps in a process. Some possible sources include: Time, schedule, and activity logs Files, studies, and reports
If data is not available:

Estimate and ten track to determine actual average value
Use minimum of 30 data points to determine an average value when calculating averages, also record the highest and lowest time so consistency is understood.
DMAIC GB Q TX PG
V 4.2.0
D M A
C
16
Linking Value Analysis With Process Flow Summarized Analysis

Process Step Est. Avg. Time (Mins) Value-Added Nonvalue-Added Internal Failure External Failure Control/Inspection Delay Prep/Set-Up Move Value-Enabling Total
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 Total
1 120 5 10 15 90 15 120 2 120 5 8
% Total
% Steps
1 120 15 120 3 180 7
957 48
100% 5.0%

180
18.8%
8 690
.8% 72.1%

30 1 957
3.1% .1% 100%
Linking value analysis with cycle time analysis creates a compelling business case for change. Displaying this information on process maps emphasizes focus areas for improvement. Tips/Traps
Conclusion: Example 7 steps (-40% of total steps) provide 100% value and take 48 minutes or 7% of total cycle time. 11 steps (-61% of steps) are non-value-added, and consume 94.9% of cycle time.
Dont 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% valueadded, 80% nonvalue-added. Think of the process map from the point of view of the product or service.
DMAIC GB Q TX PG
V 4.2.0
D M A
C
17
Review: Cause & Effect Diagrams
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 Xs.
D M A
Cause & Effect DiagramsThe Five Whys The Five Whys drill deep into the process to identify potential Root Cause(s)
Ask why five times to identify deeper causes Use process data to answer each why question
18
Example A project team has verified that the X (complicated form) accounts for the difference in cycle times between small and medium loans. They use the Five Whys to drill deeper in the process. 1. Why do complicated forms cause delays in the underwriting steps? Because underwriters receive incomplete applications 2. Why do they receive incomplete applications? Because customers dont fill out the form accurately or completely 3. Why dont customers fill out the form accurately or completely? Because the format is confusing
4. Why is the format confusing? Because the directions are hard to read 5. Why are the directions hard to read? Because the font size is too small
D M A
C
19
Cause And EffectA Breakout Activity (25 minutes)

Desired Outcome: Identify possible causes of variation in the Capital Logistics case study data What Preparation How Select a facilitator, scribe and timekeeper Who All Timing 5 mins.
Construct Cause And Effect Diagram
As a result of the QFD constructed in Measure 1 for the Capital Logistics customer service desk, a project team elects a project Y of cycle time to resolve customer problems In the Analyze phase of the project, the team uses a cause and effect diagram to brainstorm potential Xs Construct a cause and effect diagram for the problem statement Why is it taking too long to resolve customer problems? Base cause and effect diagram on your general knowledge of customer service desks. What might be potential Xs?
All
15 mins.
Five Whys Close
Use the ask why approach to identify deeper causes as appropriate Choose a spokesperson to share the results with the class
All
All
5 mins.
D M A
C
20
Prioritization Of XsControl/Impact Matrix

IMPACT High Medium Low
C O N T R O L
In Our Control
Out Of Our Control
Always Verify With Data

Use your teams process knowledge and business experience to list possible Xs in a Control/Impact Matrix. Then use process data to verify or disprove placement of the Xs. Prioritization Steps 1. Using the Control/Impact Matrix shown above, examine each X in light of two questions: What is the impact of this X on our process? Is this X in our teams control or out of our teams 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 Xs. Begin with the High Impact/In Our Control category.
D M A
C
21
Prioritization Of XsControl/Impact Matrix (continued)

Example Machines Methods Materials
Measurements
Mother Nature
People
Why Is There Difference In The Variation In Cycle Time Between Small And Medium Loans?
High

IMPACT Medium
Low
Too many defects Complicated form Too much review Duplication of effort
C O N T R O L
In Our Control
Too long for customer number

Complexity Evaluation of risk worthiness
Out Of Our Control
Too long to get credit report

Not enough staff Not well trained
The project team completed the Cause & Effect diagram and the prioritization matrix. The Xs that are identified as in control/high impact appear to be the top priority. However, the team should discuss the other Xs to make sure they have reached consensus for those priority Xs.
D M A
Control/Impact MatrixBreakout Activity (15 minutes)

Desired Outcome: Practice using a Control/Impact Matrix to prioritize Xs What Preparation Construct Cause And Effect Diagram Prioritize Potential Xs How Choose a facilitator, scribe, timekeeper and/or note taker Review the potential Xs generated for the customer service desk project in the cause and effect diagram activity Who All All Timing 2 mins.
22
Use the Control/Impact Matrix on the next page to prioritize the Xs identified by your team Choose a spokesperson to report out on your high impact Xs. Are all of these Xs within the control of the team?
All
10 mins.
Close
All
3 mins.
D M A
C
23
Prioritization Of XsControl/impact Matrix
IMPACT High Medium Low
C O N T R O L
In Our Control
Out Of Our Control
D M A
C
24
Analyze 6.2 Narrow list of Potential Causes
D M A
The Idea Of Sampling

Based on the sample, we make decisions about the population.
25
Population (Universe): A set of characteristics that defines membership in the complete set. Sample: A subset of members that possesses the same characteristics as that of the population. Why should we take a sample? Should the sample be random? Is it possible to have sampling error? How many samples should be taken? What are some everyday examples of sampling?
In most cases under study, it is impossible to measure every element of a population. In particular, we want samples that consist of observations that are independent and random. These samples, called random samples, can be used to make decisions about the population. This involves a certain amount of risk. We can minimize this risk, and, more importantly, understand what the risks are if we use statistical tests. If you were interested in finding the average height of all of the people in the United States, what advantages would exist in taking a sample?
D M A
Hypothesis TestingIntroduction
26
Refers to the use of statistical analysis to determine if observed differences between two or more data samples are due to random chance or to be true differences in the samples Increase your confidence that probable Xs are statistically significant Used when you need to be confident that a statistical difference exists
An assertion or conjecture about one or more parameters of a population(s).

To determine whether it is true or false, we must examine the entire population, this is impossible!! Instead, use a random sample to provide evidence that either supports or does not support the hypothesis The conclusion is then based upon statistical significance It is important to remember that this conclusion is an inference about the population determined from the sample data

D M A
Hypothesis Testing For Equal Means 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?
27
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 Xs have an effect on the Y. We can compare two sets of data, with X set at different values, thereby determining if that X has an effect. Examples: Does a process perform better using machine/material/fixture/tool A or B? Does the purchased material conform to the desired specifications? Is there a difference in performance between vendor A or B? Is there a difference in your process after you make a change? Is the process on target?
D M A
C
28
Concepts Of Hypothesis Testing

1. All processes have variation. 2. Samples from one given process may vary.
3. How can we differentiate between samplebased chance variation and a true process difference?
To improve processes, we need to identify factors which 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.
This way everyone makes the same decisions.
D M A
C
29
Kinds Of Differences Continuous data: Differences in averages Differences in variation Differences in distribution shape of values Discrete data: Differences in proportions
There are a variety of different hypothesis tests. Each one tests for a different kind of difference.
D M A
Hypothesis Testing 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
30
D M A
C
31
Nature Of Hypothesis
Null Hypothesis (Ho): Usually describes a status quo The one you assume unless otherwise shown The one you reject or fail to reject based upon evidence Signs used in Minitab: = or > or <
Note that we are not proving the hypothesis to be true or false. We will reject or fail to reject the null hypothesis based on the evidence from our samples. Failing to reject the null hypothesis implies that the data does not provide sufficient evidence to conclude that a difference exists. On the other hand, rejection of the null hypothesis implies that the sample data provides sufficient evidence to say that a difference exists.
Alternative Hypothesis (Ha): Usually describes a difference Signs used in Minitab: or < or >
D M A
ActivityHypothesis Statements (10 minutes)

Write the null and alternate hypothesis testing statements for each Scenario 1: You have collected delivery time of supplier A and supplier B. You wish to test scenario below:
whether or not there is a difference in delivery time from supplier A and B.
32
Null hypothesis : collected delivery time of supplier A and supplier B. Scenario 1: statement You have You wish to test statement: whether or not there is a difference in delivery time from Alternate hypothesis 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 your hypothesis. Null hypothesis statement :
Alternate hypothesis statement:

Null hypothesis statement :
Scenario 2: You that there is a difference in the cycle time process Scenario 3: You have suspect implemented process improvements to reduce cycle time to to process purchase orders in your company. You have collected cycle time before the process purchase orders in site 1 of your company compared to site 2. You are improvements and after the process improvement was implemented. You are going to perform a going to perform a that hypothesis test to verifyhave your hypothesis. hypothesis test to verify the process improvements resulted in a reduction in cycle
time.
Null hypothesis statement : Null 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:
D M A
C
33
Hypothesis Testing Guilty vs. Innocent Example The only four possible outcomes: 1. An innocent person is set free. 2. An innocent person is jailed.
Correct decision Type I error =
The probability of this type of error occurring we represent as
3. A guilty person is set free.
Type II error =
The probability of this type of error occurring we represent as
4. A guilty person is jailed.
Correct decision
Why is it important to minimize the chance of making a Type I error in a six-sigma project?
D M A
C
34
Hypothesis TestingAnother View
Ho: Person is innocent. Ha: Person is guilty. Ho
Truth Truth
Ha
Innocent
Ho Innocent, Set Free
Guilty
Guilty, Set Free Type II
Set Free Verdict Verdict

Ha Innocent, Jailed Type I
Guilty, Jailed
Jailed
This is a visual way of looking at the four possible outcomes of hypothesis testing.
D M A
C
35
Hypothesis Testing The P-value is calculated by Minitab The probability of getting the observed difference or greater when the Ho is true. If p 0.05, then there is no statistical evidence of a difference existing. Ranges from 0.0 - 1.0 The alpha () 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 < : Reject Ho P > : 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 Different
0.00 0.05
Accept Ho Not Different

1.00
p-value
D M A
C
36
Statistical Tests In Minitab
Some basic statistical tests are shown below with the command for running each test in Minitab.
What The Tool Tests Mean of population data is different from an established target. Mean of population 1 is different from mean of population 2. The means of two or more populations is different. Statistical Test 1-Sample t-test
Stat > Basic Statistics > 1-Sample t
Graphical Test
Histogram
2-Sample t-test
Stat > Basic Statistics > 2-Sample t
Histogram
1-Way ANOVA
Stat > ANOVA > One-Way
Histogram
Variance among two or more populations is different.
Homogeneity of Variance
Stat > ANOVA > Homogeneity of Variance
Scatter Plots
Output (Y) changes as the input (X) changes.
Linear Regression
Stat > Regression >Fitted Line Plot
Box Plots
Output counts from two or more subgroups differ.
Frequency
Chi-Square Test of Independence

Stat > Tables > Cross Tabulation OR Chi-Square Test
Pareto C AB D E
Category
M NO
Data is normally distributed
Normality Test
Stat > Basic Statistics
D M A
C
37
Select A Statistical Test Hypothesis tests to find relationships between project Y and potential Xs
Y
Continuous Continuous Simple Linear Regression
2 Sample t-Test (Compare Means of two samples) ANOVA (Compare means of multiple samples) Homgeneity of Variance (Compare variances)
Discrete
X
Discrete
Chi-Square Test
D M A
C
38
Hypothesis Test Summary

Normal Data
Variance Tests (Continuous Y) X2 - Compares a sample variance to a known population variance. F-test- Compares two sample variances. Homogeneity of Variance LevinesCompares two or more sample variances Mean Tests (Continuous Y) T-test One-sampleTests if sample mean is equal to a known mean or target. T-test Two-sampleTests if two sample means are equal. ANOVA One-WayTests if two or more sample means are equal. ANOVA Two-WayTests if means from samples classified by two categories are equal. CorrelationTests linear relationship between two variables. RegressionDefines the linear relationship between a dependent and independent variable. (Here, Normality applies to the residuals of the regression.)
Non-Normal Data
Variance Tests (Continuous Y) Homogeneity of Variance LevinesCompares two or more sample variances. Median Tests (Continuous Y) Moods Median TestAnother test for two or more medians. More robust to outliers in data. CorrelationTests linear relationship between two variables. Proportion Tests (Continuous Y) P-TestTests if two population proportions are equal. Chi-Square TestTests if three or more relative counts are equal.
There are a number of hypothesis tests for both normal and non-normal data. You should consult a Black Belt or Master Black Belt if you are not sure which test to use for you project, or if your project involves non-normal data. The next page shows a summary of the tests that we will look at in detail during this training. We have already looked at the Normality Test in previous modules.
D M A
C
39
We Always Look At Our data In This Way 1. 2. 3. 4. Check StabilityRun Chart, p-values Check ShapeAnderson-Darling Normality Test Check SpreadAppropriate Hypothesis Test Check CenterAppropriate Hypothesis Test
The remainder of this module will teach you how to choose the correct hypothesis test. Not all material will be covered in class. The instructor will choose the tests to cover, based on time restrictions. Prior to taking the GB certification exam you should review this section and go through the Minitab commands.
D M A
C
40
Choosing The Correct Hypothesis Test

YES NO
Are Xs Discrete?
Is the data normal?
Moods Median HOV
NO
YES
Regression
Are Ys Continuous? NO
CHI SQUARE (2)
YES
Comparing Only 2 Groups? YES
NO
Multiple Groups ANOVA HOV
Can I Match Xs With Xs?
NO
Are We Comparing To A Standard? YES
NO
2 Sample tTest HOV
YES
Paired t
Note: In order to use this chart, we are assuming our Xs are discrete. Otherwise, use Regression. Moods Median Test Ho = Median1 = Median 2 = ...Median n Ha = At least one Median is Different Homogeniety of Variance (Hov) Levenes Test 2 2 2 Ho = 1 = 2 2 = 3 ... n
Ha = At least 2 is Different
1 Sample t
Paired t-Test Ho = Difference = 0 Ha = Difference 0 One Sample t-Test Ho = = Target Ha = Target Two Sample t-test Ho = 1 = 2
Ha = 1 2
Chi Square Ho = 2 = 0 Ha = 2 0 Analysis of Variance (ANOVA) Ho = 1 = 2 = ... n Ha = At least one is Different

D M A
C
41
Which Hypothesis Testing Tool Would You Use?
For each scenario described below, which hypothesis testing tool would you use? Assume normal distribution, where appropriate
1. A six-sigma project is being conducted in the field to improve the cycle time for warranty repair returns. The warranty return cycle time was measured for a period of 6 weeks for 4 regions. The Green Belt suspects that there is a difference in average warranty repair cycle time among each of the regions. How would you test whether there is a statistically significant difference in mean cycle time for the different regions? 2. Tungsten steel erosion shields are fitted to the low pressure blading in steam turbines. The most important feature of a shield is its resistance to wear. Resistance to wear can be measured by abrasion loss, which is thought to be associated with the hardness of steel. How would you test whether there is a statistically significant relationship between resistance to wear and abrasion hardness of steel? 3. Your business purchases sheet stock from two different suppliers. It has found an unacceptably large number of defects being caused by thickness beyond tolerance levels. Data for overall mean thickness data was analyzed and found to be on target. Data was collected that would identify a potential difference in the variation of the thickness of the material by supplier. 4. Checks Are Us is a payroll processing firm. Timecard errors are routinely monitored and recorded. A Black Belt investigating the errors wishes to determine if there are any differences in the number of errors among five of its major customers. The number of errors contained in a sample of 150 employees was recorded for five weeks. How would you test if there is a statistically significant difference in the number of errors among the customers?
D M A
C
42
Data Analysis
D M A
C
43
StabilityMinitab Run Chart Lets look at an example from our GB ANOVA.mpj. Step 1: Check the stability of the processing time for SOUTH Territory. Are there any trends in the data?
MINITAB FILE: GB ANOVA.mpj
File Description: Territory (C1): Sales Territory Zone (C2): city within the sales territory Total Time (C3): Processing Time- stacked Territories C4-C8 processing time per territory noted Use Minitab to check the stability of your processing time data by territory. The Run chart is used to look at the stability of data in time order or sequentially ordered. Is the process stable over time for the South Territory?
D M A
C
44 44
StabilityMinitab Chart StabilityMinitab Run Run Chart

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 order for run charts to be valid. A run is one or more consecutive points in the same direction. A new run begins each time there is a change in direction. Do not count points exactly on the median. Test for Clusters: Ho: No fewer runs observed than expected. Ha: Fewer runs observed than expected. Clusters result from sampling from one process for a period of time and then from another process with a different center for a period so there are fewer runs than expected by sampling error alone. Test for Mixtures: Ho: No more runs observed than 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.
DMAIC GB Q TX PG
Test for Trends: Ho: No fewer runs observed than expected. Ha: More runs observed than expected. Trends result from a gradual increase or decrease in the measure so that there are fewer runs than expected by sampling error alone. Test for Oscillations: Ho: No more runs observed than expected. Ha: More runs observed than expected. Oscillations are similar to mixtures in that they indicate too many runsmore 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 each of the other territories for stability.
D M A
C
45
The P-Value Review Alpha is the probability of making a Type I error
The p-value is the probability of getting the observed difference or greater when Ho is true Unless there is an exception based on engineering judgment, we will set an acceptance level of a Type I error at a = 0.05 Thus, any p-value less than 0.05 means we reject the null hypothesis
p < : Reject Ho p > : Accept Ho

The alpha is another decision criteria. It will give you the same conclusions as the confidence interval. In general, comparing p-value to alpha is the decision criteria we use most often in this course.
D M A
C
46
Data Analysis
D M A
C
47
ShapeMinitab Descriptive Statistics Step 2: Check the shape of the data. Is the data normally distributed?
MINITAB FILE: GB ANOVA.MPJ
Ho: The data is normally distributed Ha: The data is not normally distributed
D M A
C
48
Descriptive StatisticsInput and Output

1. Double Click
3. Click OK
2. Click on Graphs, Select Graphical Summary
Variable: South
Anderson-Darling Normality Test A-Squared: P-Value: Mean StDev Variance Skewness Kurtosis N Minimum 1st Quartile Median 3rd Quartile Maximum 8.6366
7 8 9 10 11 12 13 14 15 16
Non-normal distribution!
3.095 0.000 11.9348 11.1064 123.351 1.75318 2.69797 46 1.0000 4.0000 9.0000 14.2500 45.0000 15.2330 13.9887 12.0000
10
20
30
40
95% Confidence Interval for Mu 95% Confidence Interval for Sigma 9.2120 95% Confidence Interval for Median 6.9173
Data is Normal if P-Value is > 0.05 (Accept Ho)
If the p-value is > 0.05, then we Accept the Ho and the data is normally distributed. In this case, our p-value is 0.000, therefore, we have a non-normal distribution.
D M A
C
49
Non-Parametric Tests
What if I dont have normal data? Perform Non-parametric Tests Use HOV for spread and Moods Median for Center
Normal data allows a great variety of techniques to be used in analyzing it However, responding correctly is critical if the data is not normal (which is not at all uncommon)
D M A
C
50
Non-Parametric Tests
One of the advantages of nonparametric tests is that they assume no knowledge about the underlying distributions. They often use an analysis of the ordered ranks of the data. Nonparametric tests are more powerful for non-normal data than the equivalent t-tests and ANOVA tests The Moods Median and Homogeneity of Variance Test (HOV) are nonparametric tests which are very similar to the One-Way ANOVA test
Many times in commercial projects the data are not normally distributed. Cycle time is one example. Cycle time will often be skewed to the right. Another example is ordinal data. This is data that is not continuous but can be placed in ordered categories such as good, better, and best. This often occurs with survey data. In these cases a test that does not depend on assumptions of normality may be very useful
D M A
C
51
SpreadTest On Variances
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 Xs have an effect on the Y. We can compare two sets of data, with X set at different values, thereby determining if that X has an effect. Examples: Does a process perform better using machine/material/ fixture/tool...A or B? Does the purchased material conform to the desired specifications (m & s)? Is there a difference in performance between vendor A or B? Is there a difference in your process after you make a change? Has a variation source been removed?
D M A
C
52
SpreadHomogeneity Of Variance
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
D M A
C
53
Test On Variances, ExampleInput
2 2 = 2 = 2 = 2 = Ho : East West IFG Central South Ha : at least one is different = .05
D M A
C
54
Test On Variances, ExamplesOutput

2 2 = 2 = 2 = 2 = Ho : East West IFG South Central
Ha : at least one is different

95% Confidence Intervals for Sigmas
= .05
Factor Levels Central
Homogeneity of Variance Test for total time
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
15
25
35
45
55
We have standardized on The Levenes test because it is more robust to nonnormality of samples.
p < : Reject Ho p > : Accept Ho

If you want to compare the performance of two territories, you should test their variances independently of the other territories. This way the extra noise of the other territories will not interfere with the analysis. You may end up with different conclusions running them independently, versus running them with all the other territories. The confidence intervals that are shown are intervals for the standard deviation, even though we are comparing variances (variance = 2).
D M A
CenterNon-Normal Data Test
55
Moods Median Test A Moods Median Test is used to test two or more population medians. This test is robust to outliers or errors in data (shows statistical significance). This test will show statistical significance. It will not show practical significance.
Just as with our earlier example, a picture is worth a thousand words. Looking at a graphical depiction of the data can help to verify whether or not findings (even findings with statistical significance) have any practical significance. By way of an analogy, even though you could statistically prove that people are heavier before a haircut than they are afterwards, nobody seeks out a barber shop when they go on a diet. Even though a statistical difference might be able to be proven, the difference is negligible from a practical standpoint. Viewing the data graphically helps to confirm how much practical significance exists in what youve found.
D M A
C
56
CenterMoods Median Test Step 4: Check the centers. Are the centers different by region? MINITAB FILE: GB ANOVA.MPJ Stat > Nonparametrics > Moods Median Test
Response = Total Time Factor = Territory
Answer the following questions:

1.Are the medians different? (interpret the p-value) 2.Which medians are different? (interpret the confidence intervals)
~ ~ ~ ~ ~ H :X = X = X = X = X
o 1 2 3 4 a
H : At least one is different
Another type of analysis that may be investigated is whether or not multiple samples of data can be shown to be different from each other by a (statistically) significant amount. The example shown here is questioning whether similar territories are taking different amounts of time to process loan applications as measured in days. One test which can check for this condition in nonnormal data is called a Moods Median Test. It should be noted that, unless it can proven otherwise, the two groups are assumed to have the same median. In other words, the null hypothesis for a Moods Median Test is that the medians are equal.
D M A
C
57
Moods Median Test
D M A
Moods Median Test

Mood Median Test Mood median test for total ti Chi-Square = 108.88 DF = 4 P = 0.000 Individual 95.0% CIs Territor Central East IFG South West N<= 97 47 2 25 22 N> Median 15 1 11 21 91 3.0 10.0 28.0 9.0 15.0 Q3-Q1 4.7 8.8 56.5 10.3 17.5 (-+) (+-) ---------+---------+---------+------20 Overall median = 9.0 Recall, when two groups are equal, the confidence intervals overlap. Ho: All medians are equal Ha: At least one median is not equal 40 60 ---------+---------+---------+------(+ (+) ------+---------------------)
58
Notice that Minitab has converted the continuous data into attribute data. For each level of the X, Minitab supplies a count of how many values fall at or below and above the overall median. The Moods Median test then performs a simple Chisquare test (which well cover later) on this summary table
If the number above and below are proportional, Chisquare will be low and p will be high, resulting in the conclusion that the medians are equal. If the number above and below are not proportional, Chisquare will be high and p will be low, resulting in the conclusion that the medians are not equal
The results from manually performing a Chi-square test using this summary table are the same as the Moods Median test
D M A
C
59
Hypothesis Testing Procedure
1. Write the null hypothesis
Ho: There is no difference between Population A and B
pop1 = pop2
Team preparation 2. Write the alternate hypothesis
Ha: There is a difference between Samples A and B pop1 pop2 =.05 (typical for DMAIC projects)
3. Decide on the alpha level
4. Chose hypothesis test
Choose the correct test, given the type of X and Y data.
5. Gather evidence and test/conduct analysis
Collect data, run analysis, get p-value
6. Decide to Reject H0, or not reject H0, and draw conclusion
If p 0.05 conclude, no difference between populations If p < 0.05 conclude, the populations are different
The first step in hypothesis testing is to write a null hypothesis for the population being investigated. State that there is no difference between populations that you are studying. The next step is to write the alternate hypothesis (which is often what your team REALLY thinks is true), which indicates that a difference does exist between the population. The decision regarding a p-value is usually straightforward. By convention, a p-value of .05 is typical because that is sufficient to give us the confidence we need to go forward whether we reject or do not reject the null hypothesis. The team then collects whatever additional data is necessary. The correct hypothesis test can then be chosen and the analysis is conducted. The team now has the necessary information in order to know whether or not to reject or accept the null hypothesis.
D M A
1-Sample Hypothesis 1. Ho : = constant = T Ha : constant = T
60
Ho
Ha
2. Ho : 2 = constant = T Ha : 2 constant = T
There are 7 types of hypotheses that we will be covering. We are standardizing the method by always trying to accept, with a degree of confidence, the alternative hypothesis when evidence supports the decision. We are the prosecuting attorney. Set up the hypotheses according to this standard convention.
D M A
C
61
1-Sample t-Test
Ho : = constant = T Ha : constant = T
Minitab File: GB case study.mtw Null Hypothesis: There is no difference in Junes delivery time average with that of a target set by Mays delivery time average of 5 minutes late. Minitab Command: Stat > Basic Statistics > 1Sample ttest
D M A
C
62
1-Sample t-Test
D M A
C
63
1-Sample t-Test
1. Double click anywhere on the C5 line to select time as a variable
2. Click Test Mean and enter 5.0 as Mays average, the Alternative is that they are not equal 3. Click Graphs to show a pictorial
4. Click on Histogram of data 5. Click OK
D M A
C
64
1-Sample t-Test
Histogram of time
(with Ho and 95% t-confidence interval for the mean)
50 40
Frequency
30 20 10 0 _ X Ho [ ] -100 0 100
time
The session window has our answer. If the p-value is less than .05, we reject the Null Hypothesis and accept its alternate, that there is a statistical difference in averages between this months data and Mays data.
D M A
C
65
2-Sample Hypothesis 3. Ho :1 = 2 Ha : 1 2 4. Ho : 1 2 Ha : 1 > 2 5. Ho : 1 2 = 22 Ha : 1 2 22

1
D M A
C
66
2-Sample t-Test
Ho :1 = 2 Ha : 1 2
Minitab File: GB case study.mtw Null Hypothesis: There is no difference in the average delivery time between the two different load types. Minitab Command: Stat > Basic Statistics > 2Sample tTest
Lets test the hypothesis that there is no difference in the average delivery time between the two different load types. D/HDrop & HookA driver unhooks their full trailer, and then hooks into an empty trailer to take back. UnloadThe driver opens the trailer and allows the store to unload its freight, and brings back the trailer.
D M A
C
67
2-Sample t-Test
D M A
C
68
2-Sample t-Test
1. Double click anywhere on the C5 line to select time as a variable for samples. 2. Click on C3 to chose LoadType as the variable for subscripts.
3. Click Graphs to show a pictorial.
4. Click on Boxplots of data.
5. Click OK.
D M A
C
69
2-Sample t-test
Boxplots of time by LoadType
(means are indicated by solid circles)
100
time
-100
D/H Unload
LoadType
The session window has our answer. If the p-value is less than .05, we reject the Null Hypothesis and accept its alternate, that there is a statistical difference in our Drop & Hook and Unload averages.
D M A
C
70
Multi Sample Hypothesis 6. Ho : 1 = 2 = . . . = n Ha : at least one is not equal.
7. Ho : 12 = 22 = . . . = n2 Ha : at least one is not equal.
D M A
C
71
Analyze Of Variance (ANOVA)
Ho : 1 = 2 = . . . = n Ha : at least one is not equal.
Minitab File: GB case study.mtw Null Hypothesis: There is no difference in the average delivery time between Capital Logistics drivers. Minitab Command: Stat > ANOVA > One-Way
Lets test the hypothesis that there is no difference in the average delivery time between Capital Logistics drivers, Temporary drivers working for us, and our 3 Third party carriers we use in areas that are not supported by Capital Logistics.
D M A
C
72
Analysis Of Variance (ANOVA)

D M A
73
1. Double click anywhere on the C5 line to select time as a response variable.
2. Double click C2 for Driver as a factor.
3. Click Graphs to show a pictorial. 4. Click on Boxplots of data. 5. Click OK.
D M A
C
74

Boxplots of time by Driver
(means are indicated by solid circles)
100
time
-100
Temp Labor Trans, Inc.
Cap. Log.
1. Note the sample number for each group.
2. These 3 are alike.
3. These 2 are definitely different.
The session window has our answer. If the p-value is less than .05, we reject the Null Hypothesis and accept its alternate, that there is a difference in at least 1 average between the 5 different groups of drivers.
DNGY Inc.
JJ Truck
Driver
D M A
C
75
Homogeneity Of Variance (HOV)
Ho
2 2 1 = 2 = ... = n 2
Ha : at least one is not equal
Minitab File: GB case study.mtw Null Hypothesis: There is no difference in the variation in delivery time between Capital Logistics drivers. Minitab Command: Stat > ANOVA > Homogeneity of Variance
Lets test the hypothesis that there is no difference in the variation in delivery times between Capital Logistics drivers, Temporary drivers working for us, and our 3 Third party carriers we use in areas that are not supported by Capital Logistics.
D M A
C
76

D M A
C
77
2. Double click C2 for Driver as a factor.
3. Click OK.
D M A
C
78

1. We have standardized our Levenes Test because it is more robust for non-normal data.
2. Our p-value is less than 0.05.
The graphical output has our answer. If the p-value is less than .05, we reject the Null Hypothesis and accept its alternate, that there is a difference in standard deviation between at least two of the drivers. Always use the p-value for Levenes Test.
D M A
Review: Scatter Plots

R-value
79
r = 1.0
r = +1.0
r = +.7
r = .7
r=0
r=0
The correlation coefficient r measures the strength of linear relationships. 1 r 1 When a relationship exists, the variables are said to be correlated. Perfect negative relationship: r = 1.0 No linear correlation: r = 0 Perfect positive relationship: r = +1.0 If no correlation is apparent, dont stop. Use histograms, regression, rational sub-grouping, etc. to try to extract some meaning from the data.
D M A
C
80
Simple Linear Regression
We have shown/talked about positive and negative correlation of two data sets Regression analysis is a statistical technique used to build the Y = (x) relationship between two or more variables. The model is often used for prediction. Regression is a hypothesis test. Ha: The X is a significant predictor of the response. It may be used to analyze relationships between the Xs, or between Y and X Regression is a powerful tool, but can never replace process knowledge about trends
D M A
C
81

Y = b 0 + b 1X 1 Y
Ha: The model is a significant predictor of the response. b0 = Predicted value of Y when X1 = 0 b1 = Slope of line change in Y per unit change in X1
Minitab File: GB case study.mtw Null Hypothesis: There is no correlation between our continuous Y metric (time) and a continuous X metric (distance) Minitab Command: Stat > Regression > Fitted Line Plot
Lets test the hypothesis that there is no correlation between our continuous Y metrictime and a continuous X metricdistance.
D M A
C
82

D M A
C
83
2. Double click C6 for Distance as a predictor.
3. We believe there to be a linear relationship. 4. Click OK.
D M A
C
84

Regression Plot
Y = -29.1923 + 0.174971X R-Sq = 10.5 %
100
time
2. This value means that 10.5% of the variation in Y - time is explained by this X - distance.
-100
1. There is a very weak positive relationship between distance and timely delivery
100
200
300
400
Distance
Distance is a significant factor.
Entire model is significant.
The session window has our answer. If the p-value is less than .05, we reject the Null Hypothesis and accept its alternate, that there is a correlation between delivery time and distance, but that it is a very weak one (as seen by Low R-Sq value). Hypothesis test for the model (p-value given for Regression) Ho: Model = O Ha: Model O
D M A
C
85
Chi-Square Test
Y
Continuous Continuous Discrete
X
Discrete Chi-Square Test
D M A
C
86
Chi-Square Tests
The Chi-Square Tests

Used for: 1 - Goodness-of-Fit Test: To test if an observed set of data fits a model (an expected set of data) 2 - Test of Independence: To test hypothesis of several proportions (contingency table)
Its for discrete data, any number of categories

For all cases, Ho: no difference in data Ha: difference exists
2
The Chi-Square test provides a way to determine if a given set of data has the hypothesized distribution. For example, if we roll a die 600 times., the number 1 occurs 96 times, 288 times, 3104 times, 4107 times, 5112 times, and 693 times. In theory, each number should occur 100 times. The Chi-Square test enables us to determine if the observed set of data, the output of 600 rolls of a die, fits an expected model. The Chi-Square test will also let you know if any factor has an impact on the output. If there is no difference in the data (Ho), this means no factor(s) have an impact on the output. If a difference exists (Ha), one or more factors have an impact on the output.
D M A
C
87
The Nature And Use Of Chi-Square
=
2
j=1
fo - fe fe
Suppose we flip a coin N = 100 times and observe 63 heads and 37 tails. Could this ratio of heads to tails occur by chance or should we conclude that the coin is somehow biased ?
g = number of categories fo = observed frequency fe = expected frequency (f0-fe)2 fe 3.38 3.38
Penny Heads 63/50*
Tails
37/50*
*=Observed Expected (f0) (fe)
2 = 6.76
The Goodness-of-Fit test examines the difference between what is expected and what is observed. If this difference is large, you should be suspicious that something is influencing the output (one of the factors). The c 2 value represents a test statistic used for comparison to a critical value. In this example, you would expect an unbiased coin to have equal occurrences of heads and tails (50/50). We know that we wont always have this ratio of heads to tails because of random chance. If there is a large difference between what is expected and what is observed, the chi-square value will be large (see formula). If the observed value is close to the expected value, the chi-square value will be small. How large does chi-square have to be to say the coin is biased?
D M A
The Goodness Of Fit Test

To determine whether a sample comes from a population with a known distribution, calculate chi-square as shown on the previous page. H0 = good fit (observed distribution is consistent with random outcomes from the expected model) Ha = sampled population does not have the expected distribution If the observed frequencies are close to the expected frequencies, chisquare is small. So a low chi-square leads to rejecting Ha. If chi-square is greater than a critical value, the null hypothesis is rejected. The critical values of chi-square for various confidence levels are given in the table Chi-Square Distribution. Each of the expected frequencies must be at least 5 for this test to be valid.
88
D M A
C
89
Degrees Of Freedom The chi-square distribution has degrees of freedom as a parameter. For a goodness-of-fit test, the number of degrees of freedom is k1, where k is the number of possible outcomes. Thus for the penny example, df = 2 - 1 = 1
D M A
C
90
The Statistical Test For the penny example: Critical value ( = 0.05, dof = 1) = 3.841 2Calculated > 2Table => There is a difference (Im seeing something unusual). Since the statistic from the data, 6.76, is greater than the table critical value, the alternative hypothesis is accepted: we are confident that we saw unusual behavior with that penny.
Use the tables on the next two pages to determine the chisquare critical value.
D M A
C
91
The Chi-Square Distribution
df=(# columns-1) x (# Rows-1)

df 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 40 50 60 70 80 90 100 .250 1.323 2.773 4.108 5.385 6.626 7.841 9.037 10.219 11.389 12.549 13.701 14.845 15.984 17.117 18.245 19.369 20.489 21.605 22.718 23.828 24.935 26.039 27.141 28.241 29.339 30.434 31.528 32.620 33.711 34.800 45.616 56.334 66.981 77.577 88.130 98.650 109.141 .100 2.706 4.605 6.251 7.779 9.236 10.645 12.017 13.362 14.684 15.987 17.275 18.549 19.812 21.064 22.307 23.542 24.769 25.989 27.204 28.412 29.615 30.813 32.007 33.196 34.382 35.563 36.741 37.916 39.087 40.256 51.805 63.167 74.397 85.527 96.578 107.565 118.498 .050 3.841 5.991 7.815 9.488 11.070 12.592 14.067 15.507 16.919 18.307 19.675 21.026 22.362 23.685 24.996 26.296 27.587 28.869 30.144 31.410 32.671 33.924 35.172 36.415 37.652 38.885 40.113 41.337 42.557 43.773 55.758 67.505 79.082 90.531 101.879 113.145 124.342 .025 5.024 7.378 9.348 11.143 12.832 14.449 16.013 17.535 19.023 20.483 21.920 23.337 24.736 26.119 27.488 28.845 30.191 31.526 32.852 34.170 35.479 36.781 38.076 39.364 40.646 41.923 43.194 44.461 45.722 46.979 59.342 71.420 83.298 95.023 106.629 118.136 129.561 .010 6.635 9.210 11.345 13.277 15.086 16.812 18.475 20.090 21.666 23.209 24.725 26.217 27.688 29.141 30.578 32.000 33.409 34.805 36.191 37.566 38.932 40.289 41.638 42.980 44.314 45.642 46.963 48.278 49.588 50.892 63.691 76.154 88.379 100.425 112.329 124.116 135.807 .005 7.879 10.597 12.838 14.860 16.750 18.548 20.278 21.955 23.589 25.188 26.757 28.300 29.819 31.319 32.801 34.267 35.718 37.156 38.582 39.997 41.401 42.796 44.181 45.558 46.928 48.290 49.645 50.993 52.336 53.672 66.766 79.490 91.952 104.215 116.321 128.299 140.169 .001 10.828 13.816 16.266 18.467 20.515 22.458 24.322 26.125 27.877 29.588 31.264 32.909 34.528 36.123 37.697 39.252 40.790 43.312 43.820 45.315 46.797 48.268 49.728 51.179 52.620 54.052 55.476 56.892 58.302 59.703 73.402 86.661 99.607 112.317 124.839 137.208 149.449
Alpha
D M A
C
92
The Chi-Square Distribution
df 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 40 50 60 70 80 90 100
.995 .000039 0.010 0.072 0.207 0.412 0.676 0.989 1.344 1.735 2.156 2.603 3.074 3.565 4.075 4.601 5.142 5.697 6.265 6.844 7.434 8.034 8.643 9.260 9.886 10.520 11.160 11.808 12.461 13.121 13.787 20.707 27.991 35.535 43.275 51.172 59.196 67.328
.990 .000160 0.020 0.115 0.297 0.554 0.872 1.239 1.646 2.088 2.558 3.053 3.571 4.107 4.660 5.229 5.812 6.408 7.015 7.633 8.260 8.897 9.542 10.196 10.856 11.524 12.198 12.879 13.565 14.256 14.953 22.164 29.707 37.485 45.442 53.540 61.754 70.065
.975 .000980 0.051 0.216 0.484 0.831 1.237 1.690 2.180 2.700 3.247 3.816 4.404 5.009 5.629 6.262 6.908 7.564 8.231 8.907 9.591 10.283 10.982 11.688 12.401 13.120 13.844 14.573 15.308 16.047 16.791 24.433 32.357 40.482 48.758 57.153 65.647 74.222
.950 .003930 0.103 0.352 0.711 1.145 1.635 2.167 2.733 3.325 3.940 4.575 5.226 5.892 6.571 7.261 7.962 8.672 9.390 10.117 10.851 11.591 12.338 13.091 13.848 14.611 15.379 16.151 16.928 17.708 18.493 26.509 34.764 43.188 51.739 60.391 69.126 77.929
.900 .015800 0.211 0.584 1.064 1.610 2.204 2.833 3.490 4.168 4.865 5.578 6.304 7.042 7.790 8.547 9.312 10.085 10.865 11.651 12.443 13.240 14.041 14.848 15.659 16.473 17.292 18.114 18.939 19.768 20.599 29.051 37.689 46.459 55.329 64.278 73.291 82.358
.750 .101500 0.575 1.213 1.923 2.675 3.455 4.255 5.071 5.899 6.737 7.584 8.438 9.299 10.165 11.036 11.912 12.792 13.675 14.562 15.452 16.344 17.240 18.137 19.037 19.939 20.843 21.749 22.657 23.567 24.478 33.660 42.942 52.294 61.698 71.145 80.625 90.133
.500 .455000 1.386 2.366 3.357 4.351 5.348 6.346 7.344 8.343 9.342 10.341 11.340 12.340 13.339 14.339 15.338 16.338 17.338 18.338 19.337 20.337 21.337 22.337 23.337 24.337 25.336 26.336 27.336 28.336 29.336 39.335 49.335 59.335 69.334 79.334 89.334 99.334
Alpha
D M A
C
93
The Statistical Test (continued) The 5% risk column in the table already tells us there is a difference. Alternately, a p-value (the probability of being wrong if accepting the alternative hypothesis) can be computed in EXCEL as follows: = CHIDIST(statistic,dof) = CHIDIST(6.76,1) Excel will return the p-value, in this case 0.0093. Since the probability of being wrong is less than 0.05, we accept the alternative hypothesis: we are confident that we saw unusual behavior with that penny.
Minitab does not have a Macro to accommodate the Goodness-of-Fit test. It is most convenient to compute the chi-square value in Minitab and then evaluate the corresponding p-value in Excel. One can obtain the p-value in Minitab, but it takes a little work.
D M A
C
94
Test Of Independence
Suppose a bill has been introduced in the Government to raise the speed limit. A polling organization polls 45 members of Political Party 1 and 55 members of Political Party 2, and finds 30 Party 1 and 35 Party 2 members oppose the proposal. Is there a significant effect of party affiliation on preference for the proposal? Ho = no difference between parties Ha = one party likes the proposal significantly more than the other We organize the data in a contingency table
Party 1 Oppose fo = 30 Party 2 fo = 35 Total = 65
Favor
fo = 15 Total = 45
fo = 20 Total = 55
Total = 35
Use this example to demonstrate the test for independence and contingency tables. The test for independence asks the question: Does some factor have an effect on the output? For example, does car brand affect injury severity? To test independence, we put the data in a contingency table. We then test to see if the row and column method for classification are independent.
D M A
Test Of Independence (continued) Now calculate the expected distribution, given there is no party preference. For example, the expected frequency of Party 1 opposing the proposal would be:
95
fe
Row Total Grand Total
Column Total
= (65/100) x 45 = 29.25 Note that for each cell, this is the total frequency for the row, times the total frequency for the column, divided by the total population (see chart on following page).
You will be determining your expected frequency on the ratio of number of people who oppose the proposal to the number of people who favor the proposal (regardless of party affiliation). Out of the total 100 (party affiliation unknown) people, 65 oppose the proposal and 35 favor the proposal. Expected ratios: 65/100 oppose; 35/100 favor. When you multiply these ratios by the total in a party, you get the expected frequency. If the population of Party 2 (or Party1) has a statistically different (p < ) observed frequency than the expected frequency (total or average) accept Haone party likes the bill significantly more than the other.
D M A
Expected Frequency And Degrees Of Freedom
96
fe = frow x fcol N
Party 1 Oppose fe = 29.25 Party 2 fe = 35.75 Total = 65
Favor
fe = 15.75 Total = 45
fe = 19.25 Total = 55
Total = 35
When performing a chi-square test on a contingency table, the number of degrees of freedom equals the number of rows in the table minus one, times the number of columns in the table minus one. df = (2-1) x (2-1) = 1
Expected frequencies are calculated using the methodology from the previous page. We could use the expected and observed frequencies to calculate the Chi-Square statistic like we have seen before, and then compare it to the critical Chi-Square value from the table using (alpha) and df. But, as usual, we will have Minitab do this for us. See the next pages.
D M A
Minitab Example
MINITAB FILE: (Create Your Own)
97
H o: H a:
There is no difference between party affiliation and preference on the proposal. There is a difference between party affiliation and preference on the proposal.
For this example, Party 1 = Republicans Party 2 = Democrats
D M A
Minitab Input And Results

1. Double click on both columns.
98
2. Click OK.
Minitab calculates the expected frequency (fe)
Minitab calculates the expected frequencies. The calculated chi-square is less than the critical value of 3.841, so we accept Ho. The p-value is greater than .05, so we are within our 95% confidence zone of good fit. There is no apparent relationship between political party affiliation and preference on the speed limit proposal.
D M A
Chi-Square TestCase Study Example
99
Minitab File: GB case study.mtw Ho: There is no difference between Driver types and their on-time delivery performance. Ha: At least one driver type is better or worse at on-time delivery performance. Minitab Command: Stat > Tables > CrossTabulation
In our familiar case study example, with our data in rows, we will let Minitab build the contingency table.
D M A
Chi-Square Test
100
MINITAB FILE: GB case study.MTW
D M A
Chi-Square Test
1. Double click on both columns C2 & C7..
101
2. Check the ChiSquare Analysis and Show Count buttons. 3. Click OK.
Since our discrete data is in columns and not tabled, we will first cross-tabulate and then calculate Chi-Square. Since our p-value is less than .05, we reject the Ho and accept its alternate, Ha. Compare this p-value with the one done on the ANOVA test for driver type & time. Are they the same?
D M A
C
102
Chi-Square TestSummary Goodness of Fit Test

Does our data fit an expected distribution?
Test of Independence
Does some factor have an effect on the output?
statistic critical value Accept H o > critical value Reject Ho

2
2
p-value
p < 0.05 Reject Ho p 0.05 Accept Ho
D M A
Hypothesis TestingBreakout Activity (20 minutes)
103
Desired Outcome: Select and perform the appropriate statistical test for a given project situation What Review Scenarios How Review the hypothesis testing scenarios recruitment cycle time case study that was introduced in the Minitab and Graphical Analysis module. For the scenario(s) assigned to your team, complete the hypothesis testing worksheet to select the appropriate statistical test Who All Timing 5 mins.
Select Test
All
5 mins.
Perform Test In Minitab
Perform the test in Minitab Write the p-value from the test and your conclusions on the hypothesis testing worksheet Be prepared to share your analysis and results with the class
All
10 mins.
Close
All
D M A
Hypothesis Testing Scenarios

The following scenarios relate to the HR recruitment cycle time case. In each case, perform the appropriate statistical test. (Data File: fill_time.mtw)
104
1. The project team wants to determine if there is a statistical difference in mean cycle time to fill technical positions depending on the type of resource requested. 2. The project team wants to determine if there is a statistical difference in cycle time variation, depending on the type of resource requested. 3. The project team wants to determine if there is statistical difference between the number of cycle time defects for each type of resource. 4. The project team wants to determine if recruitment cycle time increases as the number of years of experience of the person hired increases.
D M A
C
105
Hypothesis Testing Worksheet
1. State the H0 2. State the Ha 3. State level 4. What test should we use? 5. Will you reject or accept H0? 1. H0: ___________________________________________________________________
2. Ha: ___________________________________________________________________
3. level: ________________________________________________________________
4. Test: _________________________________________________________________
5. Accept or Reject H0? _____________________________________________________
D M A
C
106
Hypothesis Testing Worksheet
1. State the H0 2. State the Ha 3. State level 4. What test should we use? 5. Will you reject or accept H0? 1. H0: ___________________________________________________________________
2. Ha: ___________________________________________________________________
3. level: ________________________________________________________________
4. Test: __________________________________________________________________
5. Accept or Reject H0? _____________________________________________________
D M A
C
1
1
Improve Phase Flowchart
A
IMPROVE PHASE OVERVIEW
Improve 7: Screen Potential Causes
Improve 8: Discover Variable Relationships & Propose Solution
Improve 9: Establish Operating Tolerance & Pilot Solutions
GE Capital, Inc., 1999 DMAIC GB R TX PG V 4.2.0
D M A
Improve Phase Overview

What is the Improve phase? The Improve phase is when your team:
Selects those product performance characteristics that must be improved to achieve the improvement goal by identifying the major sources of variation in the process Develops and pilots process improvements
Why is the Improve phase important? This phase is important because it specifically identifies how the process should be improved. Steps involved in the Improve phase: Improve 7: Screen Potential Causes Improve 8: Discover Variable Relationships & Propose Solution Improve 9: Establish Operating Tolerances & Pilot Solution
D M A
C
3
The12-Step Process
Y Y X
X X
Y, X Simulation
DMAIC GB R TX PG
V 4.2.0
D M A
C
4
Improve 7Screen Potential Causes

What does it mean to Screen Potential Causes? Potential Causes are the possible Vital Few (or Critical) Xs discovered during Analyze 6 that have a significant effect on the Project Y. The Potential Causes are identified by the improvement team. During this step, the Vital Xs discovered in Analyze 6 will be statistically verified by obtaining more data, if necessary. Why is it important to Screen Potential Causes? Potential Xs are important because they help Green Belt teams focus their identification of root causes. Once potential Xs are verified as having a significant effect on the Project Y, then Green Belt teams will use the identified Vital Xs to focus improvement efforts. What are the project tasks for completing Improve 7? 7.1 Verify Vital Few Xs
IMPROVE STEP OVERVIEW
Improve 9: Establish Operating Tolerances & Pilot Solution
7.1 Verify Vital Few Xs
D M A
C
5
Improve 8Discover Variable Relationships and Propose Solution

What does it mean to Discover Variable Relationships and Propose a Solution? A Final Solution is the improvement strategy that the team develops based upon the root causes verified in the previous step.The final solution will address the Vital Xs while maintaining customer focus. Why is it important to Discover Variable Relationships and Propose a Solution? A Final Solution will eliminate or reduce the effect of the Vital Few Xs, thus improving the process capability of the Project Y. The Final Solution must not only address the Vital Few Xs, but you must also consider whether the solution satisfies the customers CTQs. By using the FMEA tool, you can identify possible Failure Modes within the Final Solution, identify what effect the Failure Modes would have on either the customers CTQs or the new process, and create action plans to eliminate or reduce the likelihood of that Failure Mode occurring. What are the project tasks for completing Improve 8? 8.1 Generate a transfer function to optimize the changes in Vital Xs (if necessary) 8.2 Propose solution 8.3 Refine solution
8.1 Generate a transfer function to optimize the changes in vital Xs (if necessary) 8.2 Propose solution 8.3 Refine solution
D M A
C
6
Improve 9Operating Tolerances & Pilot Solution

What does it mean to establish Operating Tolerances and Pilot Solutions? An Operating Tolerance is established when the X-Y transfer function is known and the required specification for Y is known. The tolerance for X can be set by using the specification for Y in the transfer function. A pilot is a test of all or part of a proposed solution on a small scale in order to better understand its effects and to learn about how to make the full scale implementation more effective. Why is it important to establish Operating Tolerances & Pilot Solution? Operating tolerances allow you to control the settings of Vital Xs so that you can better predict/control the output of the process the Y. A pilot allows you to improve a proposed solution to better meet customer CTQs, lower the risk of failure of a proposed solution, confirm expected results of a proposed solution, and increase the opportunity for feedback and buy-in for a proposed solution. What are the project tasks for completing Improve 9? 9.1 Establish operating tolerances for Vital Xs 9.2 Pilot solution
9.1 Establish operating tolerances for Vital Xs 9.2 Pilot solution
D M A

Collect data
D M A
D M A
C
1
Design Of Experiment: Learning Objectives
Define the purpose & benefits of conducting a Design of Experiment Describe screening DOE and optimization DOE and how they are used in the Improve phase Review full factorial DOE design Describe the 5 steps in analyzing DOE data, the key components of each step and the use of Minitab to analyze the data
Plot the raw data Plot the residual* Examine the factor effects Confirm impression with statistical procedures Summarize conclusions
Describe the use of fractional factorial design and explain the benefits & the trade-off over full factorial design
* We dont discuss residuals here, but to fully analyze your DOE, seek assistance from your BB/MBB and/or the Book of Knowledge
GE Capital, Inc., 2000 DMAIC GB S TX PG V 4.2.0
D M A
Design Of Experiments
With the use of Design of Experiments (DOE), we will be able to: Verify the Vital Few Xs that impact the quality of the Project Y Identify the best combination Xvalues to optimize process performance and meet customer CTQs
DOE is powerful in isolating the Vital Few causes from the many factors affecting a process. In identifying and testing solutions, a team can apply DOE to compare the results of several possible solutions. The team can develop a more effective final solution by measuring effectiveness of the improvements in action.
D M A
DOE Terminology
Independent VariablesXs
Also called factors Factors or variables we select in advance The causes
Project Y Dependent Independent X (5 Ms and 1 P)
Dependent VariablesY
(x) (x) (x)
M M M Project Y
(x)
Also called responses The quantity (Y) that we measure to determine the impact of the Xs The effect
(x) (x)
M M
Remember that the dependent variable (Y) is measured from the customers perspective. DOE allows us to identify the Xs that are the key drivers of variation in the Y.
D M A
C
4
DOE Terminology (continued)

Levels The test settings for X
For example: In our Capital Logistics GB Case Study: Y = On-Time Delivery of Shipments X1 = Truck Size Levels = 62 ft. & 66 ft. truck X2 = Tire Type Levels = Brand X & Brand Y
Main Effects Differences between each factor level Interactions Differences between two or more factor level combinations
For example: Is the On-Time Delivery different for: a) 62 ft. trucks vs. 66 ft. trucks? b) Brand X or Brand Y tires?
For example: Is On-Time Delivery different when: a) 62 ft. truck, Brand X tire b) 62 ft. truck, Brand Y tire c) 66 ft. truck, Brand X tire d) 66 ft. truck, Brand Y tire
Deviation from required delivery time. Y
Deviation from required delivery time. Y
Main Effects
Interactions
62 ft. truck 66 ft. truck
X
62 66
Truck Size Deviation from required delivery time. Y
Tire Type
X X Tire Type Y
DMAIC GB S TX PG
V 4.2.0
D M A
C
5
Strategy For Choosing The Appropriate Design The Knowledge Line

Current State of Process Knowledge
Low High
Type of Design Usual # of Factors Purpose: Identify
Main Effect /Screening >5 Main effect critical factors - vital few
Fractional Factorials 4-10
Full Factorials 1-5
Response Surface 2-3 Optimal factor settings Curvature in response, empirical models
Some Relationships interactions among factor
Estimate
Crude direction Some All main effects for improvement interactions and interactions
Step 7 Step 8
Screening DOE
Optimization DOE
Types of Experimental Design Screening DOEexperiments which are used to screen out the vital few X variables out of many at a very low cost.

Types of Experimental Design Fractional Factorialsprovide a middle ground. They are used to expand on the study of the vital few X variables at a low cost.
Can study the main effects Cant study the interactions
Can study main effects Can study selected (and few in number)
interactions Full Factorialsare used to study the vital few X variables and all possible interactions. They can be quite costly when the number of vital few Xs are large.
Optimization DOEExperiments which are used to determine the proper settings for each vital X.
Can study main effects Can study all interactions
D M A
DOE Benefits Benefits of Design of Experiment (DOE)

Can be used to identify Vital Few sources of variation Defines the relationship between the inputs & outputs Allows you to measure the influence of the Vital Few variables on a response variable
Also allows you to measure the interactions between the Vital Few variables
Is more effective & efficient than testing one factor at a time Minimizes the number of test runs you have to make to draw valid conclusions about X & Y linkages Provides knowledge of best set-up conditions of Xs for improved Y performance
Statistical Design Of Experiment Raises Your Batting Average
Design of Experiment is an approach for effectively & efficiently exploring the cause and effect relationship between numerous process variables (Xs) and a process performance variable (Y).
D M A
C
7
The Process Of Experimentation 1. Define Project

Identify responses
2. Establish Current Situation 3. Perform Analysis

Identify factors Choose factor levels Select design Randomize runs Collect data Analyze data Draw conclusions Verify results
4. Determine Solutions 5. Record Results 6. Standardize 7. Determine Future Plans
D M A
C
8
The Process Of Experimentation 1. Define Project (problems)

The mileage on my new car is not up to advertised standards. I want to improve my cars mileage.
2. Establish Current Situation (& state hypothesis)

I currently do not get the manufacturers advertised gas mileage on my car Some combination of speed, gas octane, and tire pressure will provide me with the optimum gas mileage
D M A
C
9
The Process Of Experimentation (continued) 3. Perform Analysis

Dependent VariableGas Mileage (Y) Identify Factors (independent variables)
Tire Pressure (X) Octane (X) Speed (X)
Choose Factor Levels
Independent Variables (Xs) (Factors) Tire Pressure (psig) Octane Speed (mph)
Level 30 87 55
Level + 35 92 65
D M A
C
10
The Process Of Experimentation (continued) 3. Perform Analysis (continued)

Select Design
Number of test levels
(Number of factors)
= 23 = trials
Tire Pressure + + + +
Octane + + + +
Speed + + + +
Gas Mileage
How many trials would be necessary if there were 4 variables each at 2 levels? How many variables would there be before the cost became prohibitive? What would be a different type of experiment that would not require all possible runs? The experimental trials tables represent all possible combinations of the 3 factor levels. Note: The table above shows the standard order for a 2-level, 3-factor design.
D M A
C
11

Randomize Runs Tire Pressure - (30) + (35) - (30) + (35) - (30) + (35) - (30) + (35) Octane - (87) - (87) + (92) + (92) - (87) - (87) + (92) + (92) Speed - (55) - (55) - (55) - (55) + (65) + (65) + (65) + (65) Gas Mileage 26 27 30 33 18 21 19 22
What effects (main and interaction) seem to be significant?

Main effect is the effect of one X on the Y Interaction effects are the combined effects of two or more Xs on the Y
In this experiment, only one run of each experimental trial was conducted. Additional runs would increase the confidence level for the results. Replication is a repeat of all experimental trials to obtain additional data to increase the degree of belief in the experimental results. Replications are used when:

The interactions are of critical importance Data is tricky to get. Replications will supply an extra data point when data from an experimental trial is lost You need to increase the degree of confidence in the experimental results You need to reduce the risk when implementing solutions
D M A

Collect & Analyze Data
Pareto Chart of the Effects
(Response is Mileage, Alpha = .05)
12
C B A BC AC ABC AB 0 1 2 3 4 5 6 7 8 9
A: Pressure B: Octane C: Speed
Effect Size
What factor has the biggest effect on gas mileage?
The Pareto Diagram bars show the relative importance of each main effect and interaction in the experiment. The length of each bar shows relative impact. Bars that cross the dashed line have statistically significant impact on the Y. The dashed line is called the reference line. Any effects that extend past this reference line are statistically significant within our chosen confidence level.
D M A
C
13

Collect & Analyze Data (continued)
Main Effects Plot (data means for Mileage)

28 26
Mileage
24 22 20 30 Pressure 35 87 Octane 92 55 Speed 65
What is the best speed to get the best gas mileage?
The Main Effects Plot show the average value for each of the levels of the three independent variables. Each average is based on 4 data points. Extra credit: Can you look at the experimental trials table and say why each level has a total of 4 data points?
D M A
C
14

Collect & Analyze Data (Continued)
Cube Plot (data means) for gas mileage
19 -, +, + Pressure, Octane, Speed -, +, 30 + -, -, + +, -, + 18 21 + +, +, 33 +, +, + 22
Octane
Speed
27 26

-, -, -
Tire Pressure
+, -, -
Each corner of the cube represents an experimental trial from the trial table For example, when tire pressure is negative (-), octane is negative (-) and speed is negative (-), the gas mileage is 26 (x,y,z is the convention) Can you tell which combination of tire pressure, octane and speed is the best by looking at the cube?
The cube shows the result of the interaction between tire pressure, octane and speed.
D M A
C
15

Draw Conclusions - Speed has an important effect on gas mileage - Tire Pressure doesnt significantly affect gas mileage - Octane doesnt significantly affect gas mileage Verify Results - Run additional trials at optimum settings - Conduct MSA on data results
4. Determine Solutions
Drive at 55 mph to get the best gas mileage Set tire pressure for best tire wear results Buy the octane that optimizes cost or engine cleanliness
5. Record Results 6. Standardize 7. Determine Future Plans

Any ideas on how the results from this experiment could be used for further experimentation?

Replicate the experiment to get a better indication of the effects and to quantify variability Test beyond the original limits of the X Test at midpoint for nonlinear Y responses Try additional potential Xs Block by season, tire type, fuel type, etc.
D M A
C
16
Full Factorial DOE
D M A
Full Factorial DOE Full Factorial Design of Experiment Is used to study the Vital Few X variables and all possible interactions Can be quite costly when the number of Vital Few Xs is large Can study main effects Can study all interactions
17
D M A
C
18
Full Factorial Layout: 3 Factors, 2 Levels
Std. Order 1 2 3 4 5 6 7 8
Factor 1 + + + +
Factor 2 + + + +
Factor 3 + + + +
For 3 factors, each at 2 levels, there are 23 = 2 x 2 x 2 = 8 combinations of factor settings. (Levels Factors = Combinations) Notice the pattern of factor settings in the standard order.
This is a 3factor, 2level full factorial layout arranged in standard order. Each row of the full factorial layout represents a set of experimental conditions. There are 8 different experimental conditions in this design layout.
Does this full factorial layout look familiar? Where did you see this before?
D M A
C
19
Visualizing The Experimental Space

2 Factors = A SQUARE Factor 2

3 Factors = A CUBE Factor 1
Factor 2
+
Factor 1
Factor 3
Review how the square is constructed. To add a third factor, you need a threedimensional cube plot. A cube helps us visualize the experimental space covered by the 3 factors. Each corner represents 1 set of experimental conditions. 23 = (Two Levels)(Three Factors) = 8 experimental conditions.
D M A
C
20
The Factorial Pattern Of Experimentation

k Factors Order X1
k=1 k=2
X2 + + + + + + + + + + + + + + + +
X3 + + + + + + + + + + + + + + + +
X4 + + + + + + + + + + + + + + + +
k=3
k=4
k=5
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 2k
+ + + + + + + + + + + + + + + +
X5 + + + + + + + + + + + + + + + +
Xk
The number of trials = (2 levels)(k factors)s = 2k .

Full factorial layouts can be generated for any number of factors. This table displays factorial designs arranged in standard order. Notice the pattern. A handy way to remember is by thinking how the levels alternate in the columns. For the first column, the levels alternate every 20 = 1 trial. For the second column, the levels alternate every 21 = 2 trials. For the Kth, the levels alternate every 2K-1 trials.
D M A
C
21
Replication vs. Repetition

Replication: Multiple execution of all or part of the experimental process with the same factor settings
Build different experimental units In our car mileage example, replicate different experiments at a different time
Repeat tests: 2 or more observations that have the same levels for all factors
Performed on the same experimental unit
DESIGN LAYOUT EXAMPLE 2 design with 2 replications Std. Order 1 2 3 4 1 2 3 4 + + + + + + + + Response y1 y2 y3 y4 y5 y6 y7 y8
2
2 design with 2 repetitions Std. Order 1 2 3 4 + + + + Obs 1 y11 y12 y13 y14 Obs 2 y21 y22 y23 y24 Avg y1 y2 y3 y4
Replication has 2 important properties:

Why Replication? To measure experimental variability. So we can decide whether the difference between responses is due to the change in factor levels (an induced special cause) or to common cause variability. To see more clearly whether or not a factor is important. Replication provides the opportunity for factors that are unknown or uncontrollable to balance out. Along with randomization, replication acts as a biasdecreasing effect.
It allows an experimenter to obtain an estimate of experimental error. This estimate becomes the basic unit of measurement for determining whether observed differences are statistically significant. It yields more precise estimates of factor effects
Repetition
May affect the precision of the measured response, but it does not affect the logical structure of the experiment
D M A
C
22
Randomization: The Experimenters Insurance Definition of Randomization: To assign the order in which the experimental trials will be run using a random mechanism.
It is not the standard order It is not running in an order that is convenient Minitab will randomize for us
Why Randomization?
Averages the effect of any lurking variables over all of the factors in the experiment
What is a Lurking variable?

A variable that has an important effect and yet is not included among the factors under consideration
How and why should we randomize the order in which we conduct an experiment? Randomization reduces the introduction of a systematic bias into an experiment. We call this bias a lurking variable. Lurking Variables: Definition
Safeguard: Randomize the order of the experimental trials to protect against the effect of lurking variables Action: If the lurking variable creates a trend, it can be compensated for in the numerical analysis Conclusions can then be drawn from the original factors that are not affected by such lurking variables
A variable that has an important effect and yet is not included among the factors under consideration because: Its existence is unknown Its influence is thought to be negligible Data on it is unavailable
DMAIC GB S TX PG
V 4.2.0
D M A
C
23
Design Of Experiment Summary

Provides process performance data under different types of operating parameters Establishes cause and effect relationships between a process Y and possible Xs:
Verify the Vital Few Xs Determine X values for optimal performance Predict future process performance
Can be used in both incremental and exponential process improvement studies
Caution: Prior To Implementing Solutions, Understand The Difference Between Experimental And Actual Business/Process Conditions
D M A
C
24
Reference Material Analyzing DOE Data
The following pages are reference only and will become helpful if you are contemplating the use of a DOE in your project. Sit down with your mentor after you review this section and determine feasibility.
D M A
C
25
The Process Of Experimentation 1. Define Project

Identify responses
2. Establish Current Situation 3. Perform Analysis

3.1 3.2 3.3 3.4 3.5 3.6 3.7 3.8
4. Determine Solutions 5. Record Results 6. Standardize 7. Determine Future Plans
Identify factors Choose factor levels Select design Randomize runs Collect data Focus of this module Analyze data Draw conclusions Verify results
For review: The steps in Designing an Experiment. In this module, we will focus on Step #3Perform Analysis. We will investigate how you analyze the data collected from the experiment. You saw in the car gas mileage example how we analyzed the data from that full factorial experiment. Now we will use the Capital Logistics Case Study on On-Time Delivery to see how they used Design of Experiment to identify the vital few Xs.
D M A
C
26
Steps in Analysis: Full Factorial, Replicated Designs 3.6 Analyze Data: Diagnostics: Is data OK? 3.6.2. Plot the residuals*. 3.6.1. Plot the raw data.
Minitab will provide this for you.
3.6.3. Examine factor effects. Analysis: Make inferences 3.6.4. Confirm impressions with statistical procedures. 3.6.5. Summarize conclusions.
This is the Analyze Data step, which is one step in the process of experimentation on the previous page. Steps in analyzing the data: In steps 1 & 2, we begin by checking for problems with the data before doing the DOE analysis on the data. Minitab will plot the data for you. This will help to:

After using Minitab to check the data, the next step is to plot the residuals. Residuals are the leftover variation in the data after you have accounted for the main cause of variation. Plotting the residuals should show a normal distribution. If the residuals are not normally distributed, it is a sign that there is another variable present that has not been accounted for (a lurking variable). * We dont discuss residuals here, but to continue to Step 3, seek assistance from someone who knows how. Once the data has been validated and looks acceptable, you are ready to examine the factor effects. Lets look at our Capital Logistics GB Case Study.
Find defects, outlier or patterns in the data Get a feel for the actual data Get impressions of what the data is going to tell us, to guide our analysis
D M A
C
27
Background For DOE: Capital Logistics GB Case Study
In Step 6 of the Analyze Phase, the Green Belt and the project team at Capital Logistics determined, based on the results of their Hypothesis Testing, that there were 2 Xs that had an impact on the ontime delivery of shipments to customers. The 2 Xs were: Drivers Distance The team determined that there wasnt anything that they could do to change the distance that each truck traveled to make their deliveriesthe routes were automated by the system and were determined to be the shortest routes. They decided to focus on Driver as a factor, as this was under their immediate control. The team used the Cause & Effect diagram to brainstorm potential root causes as to why Driver/Trucking company influenced Time. The resulting list included: size of the vehicle type of tire dispatch method fuel capacity type of engine After reviewing their data further, the team decided to conduct a Design of Experiment. They would use Size of Vehicle and Tire Type as possible Vital Xs in their DOE. Each of these factors has 2 levels: Size of Vehicle is either 66 or 62 feet Tire Type is either X or Y The DOE was set-up and conducted to determine the Vital X that would have the greatest impact on the timeliness of shipment deliveries to their customers.
D M A
C
28
Examine Factor Effects The 3rd Step in Analyze Data: Y = f (X1, X2, X3, , Xn) Y (Response) = On-Time Delivery of Shipment Xi (Factors) = Size of Truck, Tire Type
How do the factors affect the response? How do the combinations (interactions) of factors affect the response?
We can write the equation that answers these questions
This is your Y = f(X) Transfer function: On-Time Delivery = Constant +(-) Size of Truck + Tire Type +(-) Size of Truck *Tire Type Interaction Effect.
We can simplify the factor names with abbreviations in our DOE (see next page).
D M A
C
29
Factor Names And Interaction Symbols
It is useful to abbreviate factor names. There are several ways to do this:

Choose the first letter or first two letters of each factor Assign factors a letter: A, B, C, D, . . .
Denote the interaction between Factor A and Factor B as:

A x B or AB
For two factors (A and B) there is one interaction: AB For three factors there are
Three two-way interactions: AB, AC, BC One three-way interaction: ABC
How many interactions are there for four factors (A,B,C,D)?
Interactions for 4 factors: 2-way: AB, AC, AD, BC, BD, CD, (6) 3-way: ABC, ACD, BCD, ABD (4) 4-way: ABCD (1) Total (11) Higher level interactions (3-way or more) are hard to plot and explain. They are also not very likely to occur.
D M A
C
30
Examine Factor EffectsIn Minitab Create Factorial Designin Minitab
Data File: GB Case Study.mtw
Type of design = full factorial
D M A
Examine Factor EffectsIn Minitab Select Two Replicates
31
D M A
C
32
Examine Factor EffectsIn Minitab Name Factors
Stat > DOE > Create Factorial Design Click on Factors. Enter Name. Click OK twice.
D M A
C
33
Examine Factor EffectsIn Minitab Inputting Responses into Minitab
Note: Ensure that when you enter this data into your Minitab Worksheet, you put the times in the correct rows, based on the correct truck & tire combination of (+) and (-).
Note: Ensure that when you enter this data into your Minitab Worksheet, you put the times in the correct rows. Your experiment may have run in a different order and you may have coded your factors differently. Truck Size 66 66 66 66 62 62 62 62 Tire Size X (+) X Y (-) Y X (+) X Y (-) Y Time 126 140 30 60 -50 -80 -30 -50
(+)
(-)
D M A
C
34
Examine Factor EffectsIn Minitab Analyze Factorial Design In Minitab: Stat > DOE > Analyze Factorial Design
D M A
C
35
Examine Factor EffectsIn Minitab Analyze Factorial Design
D M A
C
36
Examine Factor EffectsIn Minitab What does the Pareto Chart tell you about the factors & interactions?
Is Truck Size significant? Is Tire Type significant? Is the interaction of Truck Size & Tire Type significant?
D M A
C
37
Examine Factor EffectIn Minitab Analyze Factorial Plots
D M A
Examine Factor EffectsIn Minitab Set-up Main Effects
38
Stat > DOE > Factorial Plots Click OK.
D M A
Examine Factor EffectsIn Minitab Set-up Main Effects
39
Stat > DOE > Factorial Plots Click OK twice.
D M A
C
40
Main Effect Plots In our GB Capital Logistics On-Time Delivery Case Study:
Main Effects Plot (data means) for Time
-1
-1
90
60
Time
30
-30
Truck Size
Tire Type
How do Truck Size & Tire Type affect Time?
A useful way to graphically show factor effects is by using Main Effects Plot. Simply plot the average response for each level (+ or -). The steeper the line, the more significant the effect. The slope of the line determines whether the effect is positive or negative. A positive slope is a positive effect and vice versa.
D M A
Interaction Effect
Differences between 2 or more factor level combinations
In our Capital Logistics GB Case Study: On-Time Delivery If we are looking at 2 factors: Truck Size Tire Type Is On-Time Performance different when: Truck Size Tire Type 62 ft. X 62 ft. Y 66 ft. X 66 ft. Y Question: Which of the following indicates Interactions between factors? Factor A + +
Response
41
Response
+
Factor B
Response
Factor A
+ Factor A
Factor B
Factor B
Adapted from Lawson, John and John Erjavac, Basic Experimental Strategies and Data Analysis. Provo, UT: Brigham Young University, p. 104.
We use the same calculation for Interaction Effects as we did for Main Effects. The only difference is we first need to generate the interaction column. We then use this column for the calculation of the Interaction Effect. Why are interactions important? Why is it important to determine the effect of interactions?
If the lines are not parallel, there is interaction between the factors (variables).
D M A
Interaction Effect In our GB Capital Logistics On-Time Delivery Case Study:

Interaction Plot (data means) for Time
Truck Size -1 1 100
42
Late
On-Time
Mean
Early
50
-50
-1
Tire Type
How do Truck Size & Tire Type interact? How do the interactions affect time?
What does this tell us? In this case, our goal was On-Time-Delivery. None of the points give us OTD for mean=0. Therefore, we should go with the (-) Tire Type and the (-) Truck Size (X Tire Type and Truck Size of 62.) This will get us closest to a mean of zero.
D M A
C
43
Steps In Analysis: Full Factorial, Replicated Designs
Diagnostics: Is data OK?
3.6.1. Plot the raw data. 3.6.2. Plot the residuals. 3.6.3. Examine factor effects.
Analysis: Make inferences
3.6.4. Confirm impressions with statistical procedures. 3.6.5. Summarize conclusions.
D M A
Confirm Impressions
44
The 4th Step in Analyze Data: By now you have a good idea about the influence of the factors in the experiment. You can confirm these impressions statistically by performing a Hypothesis Test. Ho: Factor has no effect on the results Ha: Factor has an effect on the results p : Accept Ho p : Reject Ho
Statistical Analysis Confirms The Results Of The Graphical Analysis
Now we are determining the statistical significance of the effects.
D M A
C
45
Confirm Impressions With Minitab In our GB Case Study: What does the Minitab Session Window tell you ?
What do the p-values for:

Truck Size Tire Type Truck Size * Tire Type
Indicate? Does this statistical analysis confirm the graphical analysis?
D M A
Steps In Analysis: Full Factorial, Replicated Designs
46
Diagnostics: Is data OK?
3.6.1. Plot the raw data. 3.6.2. Plot the residuals. 3.6.3. Examine factor effects.
3.6.4. Confirm impressions with Analysis: statistical procedures. Make inferences 3.6.5. Summarize conclusions.
D M A
Summarize Conclusions The 5th Step in Analyze Data:

List all the conclusions you have made during the analysis Interpret the meaning of these results
For example, relate them to known physical properties or an experts personal knowledge of the process
47

Make recommendations Formulate and write conclusions in simple language
What conclusions would you make from the Capital Logistics DOE study?
D M A
C
48
Summarize Conclusions (continued)
From our Capital Logistics GB Case Study on On-Time Delivery: The Green Belt and the project team reviewed the results of the DOE. They saw that there seemed to be a link between on-time performance and the size of the truck unit, as well as an interaction effect between Tire Type and Size of Truck on on-time performance. They returned to their Fishbone (Cause & Effect) Diagram with greater granularity, and asked why these factors would influence on-time performance. They validated these reasons through field observations. The team validated that the larger size trucks took longer at some locations. In addition to the time involved with maneuvering in a cramped environment, on certain routes these trucks had a higher incidence of tire failure (lateral movement caused by tight turns increased the stress on the tires, increasing the number of flat tires). By using the data & results from the DOE, the team had determined the root causes of the on-time performance problem and were ready to start to brainstorm solutions to address/control these Vital Xs.
D M A
C
49
Reducing the Size of Experiments Fractional Factorial DOE
D M A
C
50
Reducing The Size Of Factorial Experiments

Many factors potentially impact
the quality of any process/product The factorial strategy is an efficient approach to experimentation, compared to one-at-a-time When factors are investigated at 2 levels, the number of experimental runs is 2k (note: the k needs to be raised, because it is 2 to the power of k - k is the exponent) This can result in a large number of runs, even with a relatively small number of factors Number of Factors 1 2 3 4 5 6 7 8 9 10 15 20 Number of Runs
For a 2Level Factorial with k Factors.. Number of runs = 2k
Limited resources allow you time to run only 4 trials in the allotted time. Which 4 trials will you choose?
Full factorial is time-consuming and costly as the number of factors increase. Also consider the number of 2-way and higher order interactions as the number of factors increases. For 4 factors, lets look at every possible effect (main & interaction) and the overall average. Evaluating the average is what the extra run is used for.
2 4 8 16 32 64 128 256 512 1024 32,768 1,048,576
From a Full Factorial (4 factors): Number Overall Average Main Effects: A B C D 2-way interactions (AB, AC, AD, BC, BD, CD) 3-way interactions (ABC, ABD, ACD, BCD) 4-way interactions (ABCD)
1 4 6
DMAIC GB S TX PG
V 4.2.0
D M A
C
51
Fractional Factorials
Fractional Factorials*: Provide a middle ground for design of experiment Are used to expand on the study of the Vital Few X variables at a low cost Can study main effects Can study selected (and few in number) interactions
Fractional Factorials Allow You To Reduce The Size Of An Experiment And Select Certain Runs
*Fractional Factorials are beyond the scope of this course. Consult your BB/MBB or refer to the Book of Knowledge for further information.
D M A
C
52
Strategy For Choosing The Appropriate Design The Knowledge Line

Current State of Process Knowledge
Low High
Type of Design Usual # of Factors Purpose: Identify
Main Effect /Screening >5 Main effect critical factors - vital few
Fractional Factorials 4-10
Full Factorials 1-5
Response Surface 2-3 Optimal factor settings Curvature in response, empirical models
Some Relationships interactions among factor
Estimate
Crude direction Some All main effects for improvement interactions and interactions
Step 7 Step 8
Screening DOE
Optimization DOE
Types of Experimental Design Screening DOEexperiments which are used to screen out the vital few X variables out of many at a very low cost.

Types of Experimental Design Fractional Factorialsprovide a middle ground. They are used to expand on the study of the vital few X variables at a low cost.
Can study the main effects Cant study the interactions
Can study main effects Can study selected (and few in number)
interactions Full Factorialsare used to study the vital few X variables and all possible interactions. They can be quite costly when the number of vital few Xs are large.
Optimization DOEExperiments which are used to determine the proper settings for each vital X.
Can study main effects Can study all interactions
D M A
Optimization Design Of Experiment Activity (40 minutes)

Desired Outcome: Practice how to set-up a DOE What Team Preparation How Choose a facilitator, scribe, timekeeper, and/or note taker Determine the timing for each activity below Agree on the example you will use for the exercise Your own project How to get a great lawn How to make the best chocolate chip cookies How to lose weight Set-up the DOE For the example you selected, identify 3 factors at 2 levels (+/-) each Using the worksheet on the following pages, design the experiment up through Step 6 Facilitator Who Team Timing 2 mins.
53
15 mins.
Set-up Optimization DOE
Discuss what the experiment might demonstrate, were you to execute it Brainstorm your list of questions for how to run this experiment Choose a spokesperson to report out on the design of your experiment, and the questions you have about running it
Facilitator
15 mins.
Report Out
All
8 mins.
D M A
Design Of ExperimentsBreakout Activity Worksheet
54
1. Define the problem 2. State the hypothesis 3. Identify the dependent (Y) and independent (X) variables Dependent Variable (Y) Independent Variables (Xs) = = 4. Determine the test levels for each independent variable (X). = =
Independent Variables (Xs) (Factors)
Level -
Level +
5. Calculate the number of trials to test all combinations of test levels. # of trials = # of test levels
# of independent variables
D M A
C
55
Design Of ExperimentsBreakout Activity Worksheet (Continued)

6. Construct the experimental trials table
Independent Variable (X) #1

Dependent Variable (Y)
Dependent Variable (Y)
D M A
Appendix: Example Of Designed Experiment: GE Card Services
56
D M A
C
57
DOE Example: Defining The CTQ

Overall Quality EX/VG - 57% .26 Authorizations EX/VG - 60% .25 Financial & Marketing Info EX/VG - 30% .11 Marketing Rep EX/VG - 54% .12 New Account Approval EX/VG - 34% .16 Card Production EX/VG - 47% * Billing Statements EX/VG - 40% * Payment Processing EX/VG - 54% .15 Customer Service EX/VG - 44%
Impact
Performance Overall 18% 16%
Impact Overall 13% 0.38 0.28 0.24 0.17 1 0.8 0.6 0.4 0.2 0
Performance
34%
0.59 0.23 0.2 1 0.8 0.6 0.4 0.2
Speed Of 24% 25% Approval Amount/ 18% 26% Type Of Info Credit Lines 16% 25% Granted 0 0% 20% 40% 60% 80% 100% Excellent Very Good
Timeliness 10% 33% Of Delivery Accuracy 20% Insert 12% Capability Card Quality
23% 41% 47% 53%
0% 20% 40% 60% 80% 100% Excellent Very Good
#1 Speed
Inseparable CTQs One Experience To The Customer
#1 Speed #2 Accuracy
Speed Is The #1 CTQ Of The New Account And Card Experience
A project team was working to improve customer satisfaction with the process of approval for and issuing of new credit cards. One of the key drivers of satisfaction was identified as time to issue cards, i.e., the cycle time between a consumers approval for a credit card, and the time he/she received it. Through segmentation, the team found that applications mailed or faxed in took considerably longer than other applications where the application information was entered electronically at the store.
When examining the process map, it became clear that many faxed and mailed applications needed additional information before they could be processed. This often involved phone or written correspondence with the applicant, which slowed the process considerably.
D M A
C
58
DOE Example: Causes Of Time Consuming Delays 79% of queuing for non-fraud! Why?
Non-Fraud Queue Pareto
50%
40% Frequency %
30%
20%
10%
0% Msg Home #/AC Wrng Promo Cd Msg/Wrng Brnch/Dlr/ Str # Msg/ Wrng SS# Unknown CBR Issues Msg/ Wrng Zip Duplicate
Missing and illegible information is a major cause of delays. What can be done to reduce missing and illegible information? Many felt that little could be done to improve the situation for handwritten applications.
The team discovered that 79% if the applications which were in queuei.e., on hold before being processed were in queue for reasons other than potential fraud issues. The team analyzed the reasons for non-fraud queuing and created a Pareto chart.
D M A
C
59
DOE Example: The Designed Experiment
Factor Name App. Type Instructions Flags
Explanation Terms of the credit agreement printed on the application Written instructions on how to complete the form included Critical information without which the application cant be processed is specially noted Delineated boxes to hold numeric information, e.g., phone
Standard Level Present (Full) No No
New Level Absent (Naked) Yes Yes
Boxes
No
Yes
(
Insurance Page Layout
)-
Not Body* Portrait Body* Landscape
Location on the form for the purchase of the optional credit card insurance Orientation of the form
The only influence that the business has on legibility and completeness is through the application itself. The team then decided to study what they could do to the application to make it easier for their customers to fill out. The factors shown above were chosen for study. *Body/Not Bodyrefers to location of the insurance text on the application form. On a landscape layout, not body applications had text to the right of the application. On a portrait layout, not body applications had text at the bottom of the application.
D M A
C
60
DOE Example: Experimental Complications

Complications
The difficulty in tracking many (in this case, 16) different applications through the process Large amount of person-to-person variation with respect to legibility and completeness, making it difficult to assess the effect of changes Test the applications outside the normal process Have each tester fill out two applications
Remedies
The applications were paired (eight pairs of applications) and customers were asked to fill out a pair of applications and provide feedback regarding the two application versions. The applications were paired so that customers experienced the factors at both levels. Eighty customers filled out a pair of applications in return for a $10.00 gift certificate. The order which each paired application was presented to the customer was randomized throughout the test to balance the effect of order on the results. Customers were asked whether they preferred one application version over another and why. Applicants signing for credit insurance were asked whether they understood that they had requested insurance coverage.
The completed applications were sent for imaging, and the imaged applications were reviewed by an associate to record the legibility and completeness of each field on each application. A legibility score was assigned using a 5-point scale; missing information was given a 0 and fields not relevant to the applicant were recorded as not being applicable. Preferred applications were given a score of 1, not preferred 0, and no preference was indicated by a 0.5.
D M A
C
61
DOE Example: Responses
Two Key Responses Were Studied: A combined completeness and legibility score, which was based on the percent of necessary information that was both present and legible A preference score, which told which of the two applications the customer preferred
D M A
C
62
DOE Example: Analysis

Main Effects Plot Means For Legibility And Completeness
0.849 Legibility and Completeness 0.843 0.837 0.831 0.825
Naked
App. Type
Full No Portrait
Yes
No
Yes
No
Yes
Body
Bottom
Landscape
Instruct
Flags
Boxes
Insurance
Pg. Layout
*
0.67 0.59 Preference 0.51 0.43 0.35
Main Effects Plot Means For Preference
Naked
Full
No
Yes
No
Yes
No
Yes
Body
Bottom
Landscape Portrait
App. Type * Effect was significant
Instruct
Flags
Boxes
Insurance
Pg. Layout
D M A
C
63
DOE Example: Results

Effect On Legibility And Completeness Full 82.8% H No 82.9% H Not Body 85.3% H Naked 84.8% H Yes 84.8% H Body 82.4% H Effect On Preference Full 34% No 57% H No 59% Not Body 56% Portrait 36% Naked 66% Yes 43% H Yes 41% Body 44% Landscape 64%
Factor Application Type Instructions Flags Boxes Insurance Page Layout
Recommendation Naked
Yes1 Yes No Not Body Landscape
The data does not show that instructions improve legibility and completeness, but it is felt that instructions should be provided for clarification purposes. Also, consideration should be given to improving the effectiveness of the instructions provided on the application.
Effect not significant
D M A
D M A
Step 9 Step 8 provided the experimental techniques to establish the relationship between the measurable Y characteristics and the controlling X factors. In step 9, those relationships (transfer functions) wil be used to define the key operating parameters and tolerances to achieve the desired performance of the CTQs/ One should be able to set the tolerance of X factors if the X-Y relationship and the specifications of Y are given.
GE Capital, Inc., 2000 DMAIC GB T TX PG V 4.2.0
D M A
Principle Of Tolerancing Tolerance: the allowable range of variation of X while still meeting the requirements of Y. Establish tolerance of X based on the requirements of Y, or often the specification limits via the Transfer function The tolerance should allow the project to reach the project objectives established in step 5 When more than one CTQ is involved, be aware of the tradeoffs among CTQs Be aware of the variations due to the measurement of Xs and Y Ys
USL T LSL Y = f (X)
Tolerance: the allowable range of X

X
Lower Limit of Tolerance
Upper Limit of Tolerance
It is complicated to calculate tolerance for (X) with multiple Xs. Use:

Crystal Ball Calculus
It will be covered in detail in the GE-DFSS course
D M A
Consideration Of Measurement Variations

Y
USL
The measurement variation of Y
T LSL Y = f (X)
New tolerance considering the variation of Y measurement X
Lower Limit of Tolerance
Upper Limit of Tolerance
USL
T LSL Y = f (X)
New tolerance considering the variation of X + Y measurement X
The measurement variation of X
Lower Limit Upper Limit of Tolerance of Tolerance
Estimate the gage error for Y using the e from the MSA study in Step 3. The width of the measurement error distribution is 5.15e Establish the gage error for X in Step 10. Use the 5.15e from the MSA of X.
D M A
Example: Weight Loss Wellness Program

lbs. week 3 USL
Weight 2 Loss
1 0
Target
Transfer Function
LSL
1.4
2.7
3 hrs./week
CTQ requirements: 4 Target to lose 2 lbs/week At least need to lose 1 lb/week No more than 3 lbs/week due to health concerns
These goals do not take into consideration any error in the Y or X due to measurement error (MSerror) Since we have performed a Gage R&R on both the Y and X measurement system we know the GRR % Tolerance for Y and X Lets use the information to develop the operating limit for our actual X
D M A
C
5
Example: Weight Loss Wellness Program (continued)
lbs. week 3 USL
Weight 2 Loss
1 0
Target
LSL
New operating limits on X due to variability in the Y and X measurements
1.55
Exercise
hr/week
2.55
To get a more conservative tolerance for X, taking into account measurement error (MSerror) on Y, we center the width of the 5.15 MSerror distribution on both the USL and LSL for Y. Q: Why center the measurement error distribution on the spec? A: The error in measurement system will effect us most at the ends of the tolerance Q: Why start from within the tolerance? A: This results in a more conservative tolerance for the X factor operating within. The new operating limits of X will ensure you stay in spec for Y.
D M A
Example: Weight Loss Wellness Program (continued)

lbs. week 3 USL
Weight 2 Loss
1 0
Target
LSL
1.5
Exercise
2.6
3 hrs./week
New operating limits on X due to variability in the Y measurements
How to Determine the values of the new Y start points Given Y Tolerance = 2; GR&R = R&R % Tolerance =10% 1. Calculate MSerror in Y units MSerror = (R&R % Tolerance) ( Tolerance) MSerror = (10%) ( 2) = .2 2. Take MSerror () ( MSerror) = (.2) = .1 3. Calculate new start points for Y axis New Y USL start point
Y USL- ( MSerror) = 3 - .1 = 2.9
New Y LSL start point

Y LSL + ( MSerror) = 1 + .1 = 1.1
Our purpose is to get an operating tolerance on the X but we first should adjust the Y tolerance due to measurement error.
D M A
Example: Weight Loss Wellness Program (Continued)

lbs. week 3 USL
Weight 2 Loss
1 0
Target
LSL
1.5
Exercise
2.6
3 hrs./week
New operating limits on X due to variability in the Y measurements
Given: X is R&R % Tolerance = 20% New X Tolerance = USL LSL (from last page) New X Tolerance = 2.6 1.5 = 1.1 Calculate MSerror in X units 1. MSerror = (20%) (1.1) = .22 2. Take () (MSerror) = () (.22) = .11 3. New X USL = USLx - ( (MSerror)) = 2.6 - .11 = 2.49 New X LSL = LSLx - ( (MSerror)) = 1.5 + .11 = 1.61
Correct X USL and LSL for the MSerror of the MSA system used to measure X using same methods as with Y. This has the effect of further tightening the X Tolerance If you operate in these new operating limits of X you should be ensured of staying in spec on Y. That means no defects.
D M A
Summary
Set the process tolerances based on product specifications (exercise time needed to achieve desired weight loss goal). Adjusted our process tolerances to account for product measurement variation The tolerance principle is relatively simple and the application is relatively straightforward when: 1) The measurement variation of of Y and X are sufficiently small to be ignored 2) Only one X factor involved However, when measurement variations of Y and X needs to be considered and multiple X factors are involved, the math techniques could be more complicated. Courses on DFSS Statistical Design Method or DFSS Statistical Tolerance are offered in various GE businesses.
D M A
Control Phase Flowchart
A
CONTROL PHASE OVERVIEW
Control 10: Validate Measurement System
Control 11: Determine Process Capability
Control 12: Implement Process Control System & Project Closure
GE Capital, Inc., 1999 DMAIC GB U TX PG V 4.2.0
D M A
Control Phase Overview
What is the Control phase? The Control phase is when your team: Ensures that the new process conditions are documented and monitored via statistical process control methods After a settlingin period, reassesses the process capability Closes the project Why is the Control phase important? This phase is important because it ensures that process improvements are implemented together with appropriate process controls to ensure that the process changes are sustained into the future. It also is an opportunity for the team to prove that the project process performance goal has been met. Steps involved in the Control phase Control 10: Validate Measurement System Control 11: Determine Process Capability Control 12: Implement Process Control System & Project Closure
D M A
C
3
The12-Step Process
Y Y X
X X
Y, X Simulation
D M A
Control 10Validate Measurement System

What does it mean to Validate Measurement System? In the Measure phase, you focused on validating the measurement system for the Project Y. In this step, you should now validate the measurement system for the Vital Xs. A Measurement System Analysis (MSA) is the overall process to test your data for its validity. Before actually performing the MSA, you will have to develop a data collection plan for the Vital Xs. You should focus on the Vital Xs at this point because you want to develop Control Plans for each Vital X. Why is it important to Validate Measurement System? Validating the Measurement System is important because you need to have data that you can trust in order to make decisions with that data. If your data is not telling you the truth, you are likely to make a decision with that data that will add more variation to the process. As part of your Process Control System, you will monitor the Vital Xs to make sure they stay in control. Therefore, you must validate the Measurement system used to collect the Vital X data. What are the project tasks for completing Control 10? 10.1 Validate Measurement System 10.2 Determine whether Measurement System is adequate to measure Vital Few Xs
CONTROL STEP OVERVIEW
10.1 Validate Measurement System 10.2 Determine whether Measurement System is adequate to measure Vital Few Xs
D M A
C
4
Control 11Determine Process Capability

What does it mean to Determine Process Capability? Process Capability is the measure of process performance relative to the Customers CTQ. It is measured again at this point in the process in order to see if there was a process improvement. If a pilot is conducted, the improvement team should measure process capability from the pilot and then, once full implementation takes place (after Step 12), measure process capability again. Why is it important to Determine Process Capability? Process Capability is important at this stage of the project because you want to be able to compare baseline performance before improvements to the new Process Capability once improvements have been implemented. It is also important to statistically validate the improvement in the process through the use of a Hypothesis Test, such as the 2-Sample t-Test. What are the project tasks for completing Control 11? 11.1 Evaluate Process Capability of the improved process 11.2 Confirm statistically that the improvement goal has been met
11.1 Evaluate Process Capability of the improved process 11.2 Confirm statistically that the improvement goal has been met
D M A
Control 12Implement Process Control Systems

What does it mean to Implement Process Control System? A Process Control System is a system of activities whose purpose is to maintain process performance at a level that satisfies customers needs and drives the ongoing improvement of process performance. A Process Control System consists of three parts: Documentation, a Monitoring Plan and a Response Plan. During this step of the project, you are also ready for project implementation. Once the improvement is implemented and the process improvement is verified, the final step of the project is Project Closure. During Project Closure, you will finalize project documentation, leverage learning to other areas of the business and celebrate project completion with your team. Why is it important to Implement Process Control? A Process Control System will help maintain the gain that occurred with the process improvement. Without control of an improved process, the process tends to revert to its old performance. What are the project tasks for completing Control 12? 12.1 Implement a Control strategy for each Vital X 12.2 Prepare Process Control plan 12.3 Implement a solution 12.4 Close the project
12.1 Implement a Control strategy for each Vital X 12.2 Prepare Process Control plan 12.3 Implement a solution 12.4 Close the project
D M A

Collect data
Process Control Plan Worksheet

8
Documentation Response Plan

Checking CheckingThe TheWork Work
Monitoring
Plan PlanFor ForDoing DoingThe TheWork Work
Response ResponseTo ToSignals Signalsof ofChange Change
Process Step (From Process Map Attached) Monitoring Standards

Identifies Who Should Do What With The Defective Output. Describes What Should Be Done For Those IllServed By These Defects.
Procedure (From SOPs)
Key Process & Output Measures
Method For Corrective Action Recording For Containment Data
Preventive Action Procedure For Process Improvement

Describes What Must Be Done When We Fail Process Sigma Standard. Describes What Must Be Done To Prevent Special Causes That Worsen Performance Or To Incorporate These Special Causes That Improve Performance.
Refer To Diagram That Illustrates Process Steps Separated By Function. Shows Transfers Between Functions, And Which Function Is Responsible At Each Step.
Root Cause Xs Refers To Document That Project Ys Describes How The Task Should Process Sigma For Each Be Done, Or Project Y Refers To A Document That Describes The Step.
For Each Measure, Describes Any Target, Numeric Limits, Or Tolerances To Which A Process Should Conform
For Each Measure, Describes How The Monitored Data Should Be Recorded, Who Should Record The Data And How.
D M A
Forms Of Intellectual Property
Patent Best Protection against independent development by others Gives right to exclude others from making, using selling, the invention or 20 years from filing Can wait 2-3 years for patent to issue Right granted by the government for new, useful, nonobvious inventions Trade Secret Rights exist only so long as actually kept secret Protects against theft or misappropriation, not independent development Does not require inventiveness in the patent sense Copyright Protects original works of authorship fixed in a tangible medium Does not protect against independent development Trademark Defines source of goods or services Arbitrary, fanciful marks get stronger protection
D M A
Intellectual Property Infringement Avoidance Clearance And Assessment
10
It is critical to assure that improvement or new designs developed in the course of the project do not infringe on the Intellectual Property of others, external to GE. GE has a clear policy on infringement avoidance. Intellectual Property Infringement Avoidance Process. The team must conduct the necessary reviews to assure no infringement (consult the IP Business Champion for guidance). In general, if the answer is yes to any of the questions below, consult with the IP Business Champion and IP Designee for your business to obtain clearance before proceeding.
1) Is the team aware of any non-GE businesses who use substantially the same process or sell substantially the same process? 2) Will the project result in a product or process that is significantly different than a current GE product or process, and will it be commercially significant? 3) Have or will any services or products been obtained from an outside party I connection with this project, and is the project result commercially significant? 4) Will the product or process be exposed to non-GE parties, and be commercially significant? 5) Will the product or process be exposed to the public, and be commercially significant?
If the project obtained clearance from the Infringement Avoidance screen, then consider whether there is Intellectual Property to Capture and Protect, via Patent, Trade Secret, Copyright or Trademark. If the project results in a process or product that is new, non-obvious and useful and will be of significant commercial value. It may be patentable. Consult the IP Business Champion and IP Designee for your business and submit the new process or product for evaluation through the IDEAS website.
D M A
Statistical Process Control Overview
GE Capital, Inc., 2000 DMAIC GB V TX PG V 4.2.0
D M A
Statistical Process Control: Learning Objectives
Describe the two types of Control Charts Describe and use Variable and Attribute Control Charts Select the appropriate Control Chart for a given process based on data type Define the control limits and how they differ from specification limits Explain the decision errors in SPC Explain the rules (Minitab and Western Electric) to determine whether a process is out of control Describe and practice the use of Minitab to generate I&MR, X & R, U, C, P and NP chart
D M A
Statistical Process Control
The Application of Statistical Techniques for Measuring and Analyzing the Variation in Processes
Control Charts Are The Primary Tools In SPC
First developed by Dr. Walter Shewhart, control charts serve as the best way to determine if a process is in control. By in control, we mean that the process is consistent and predictable. When a process is out of control, something about the process has changed. A cause for the change in the process should be sought.
D M A
Five Main Uses Of Control Charts
To reduce scrap and rework and for improving productivity Defect prevention. In control means less chance of nonconforming units produced. Prevents unnecessary process adjustments by distinguishing between common cause variation and special or assignable cause variation Provides diagnostic information so that an experienced operator can determine the state of the process by looking at patterns within the data. The operator can then make the necessary changes to improve the process performance. Provides information about important process parameters over time
D M A
Process States The Four States of a Process 1. Chaos

Process out-of-control, producing non-conforming product Even the level of nonconformance is unstable Assignable causes dominate the output Fixes dont work for very long
How To Begin To Sort Out The Problems To Be Solved?
D M A
Process States (continued) The Four States of a Process 2. The Brink of Chaos
Process unstable, some nonconforming product is being produced Instability will continually change product characteristics Process output is influenced by assignable causes No assurance the next piece produced will be conforming
How To Determine Existence Of Assignable Causes?
D M A
Process States (continued) The Four States of a Process 3. The Threshold State
Process inherently stable over time, but producing some nonconforming product (process sigma is less than process sigma goal/standard Proportion nonconforming predictable Some non-conformances will be shipped If process natural spread is greater than the tolerance, common causes must be reduced/removed Process must be monitored to assure desired effect is achieved
How To Be Certain That The Process Has Improved?
D M A
Process States (continued) The Four States of a Process 4. The Ideal State
The process is inherently stable over time (process sigma > process goal/standard) Operating conditions are not changed arbitrarily (follow the process plan) The process average is set and kept at the proper level The natural spread of the process is less than the specified tolerance
How To Be Certain That None Of These Conditions Change Or Degrade?
D M A
Variation And Control Charts
40
Upper Control Limit* (UCL)
30
Measurement
20
Average
10
0 0 10 20
Lower Control Limit* (LCL)
Time Order of Sample
Control charts are used to distinguish between common and special causes of variation and use that understanding to control and improve processes. Control charts are characterized by two things: 1. The average (mean), or centerline, which represents the middle point about which plotted measures are expected to vary randomly. 2. Control limits, both upper and lower, which represent the performance boundaries you can expect for the process. Although measures vary, one would not expect to see plotted measures outside of these boundaries if the process operated predictably. *Note: Minitab displays UCL as 3.0SL (Three sigma limits) Minitab displays LCL as3.0SL
D M A
Determining If Your Process Is Out Of Control
10
Zone A Zone B Zone C Zone C Zone B Zone A
Upper Control Limit (UCL)
Average
Lower Control Limit (LCL)
Source: Memory Jogger Plus, 1994 GOAL/QPC
Dividing the control chart into zones can aid in detecting special cause variation. Each of the zones represent standard deviations from the mean. Zone C, for example, is + and - one standard deviation from the average.
D M A
Minitab Rules
1. One point beyond zone A (Western Electric) 2. Nine points in a row in zone C or beyond (All on one side.) (Western Electric) 3. Six points in a row, all increasing or decreasing 4. Fourteen points in a row, alternating up and down 5. Two out of three points in a row in zone A or beyond (Western Electric) 6. Four out of five points in a row in zone B or beyond (Western Electric) 7. Fifteen points in a row in zone C, above or below center 8. Eight points in a row beyond zone C, above or below center
11
+3 +2 +1 -1 -2 -3 C C B B
These are the rules available in Minitab. You can find these rules under the TESTS button inside the Control Chart screen. GE has standardized on the abovementioned rules to identify special cause variation.. You can select the rules you want Minitab to use. One thing to be careful with all these rules is that you may not want to react too quickly. Several factors need to be considered: Are consequences high (i.e. cost, safety, etc) if we do not react to a potentially unstable system using the 8 rules? Can we potentially make a process unstable if we react too quickly to the control chart, when in reality the process is not out of control (i.e., Do not fix it, if it is not broken)?
D M A
Control Limits vs. Specification Limits Control Limits

Defined based on process performance (+/- 3 estimated standard deviations from the mean) Help determine if your process is in control (without special cause variation) Plotted on control charts Change when there is a verified, significant change to your process Represent the voice of the process
12
Customer Specification Limits

Defined based on feedback from the customer(s) Help determine if your process is producing defects Plotted on histograms (not control charts) Change when your customers say they do! Represent the voice of the customer
Control limits are not the same as tolerances.

Control limits are what you have Tolerances are what you want
While the data collected for the control charts demonstrates: Mean = 40 days UCL= 46 days LCL= 39 days The UCL and LCL are calculated based on the data collectedthe UCL and LCL are NOT related at all to spec. limits.
Example: Customer may define the following for shipment delivery from time of order: Target= 20 days USL= 45 days
D M A
Control Limits vs. Specification Limits (continued)

It is possible to have a stable (in control) process that has unacceptable variation Assume both process A and B are statistically performing in control
13
PROCESS A
Lower Spec. Limit Upper Spec. Limit
PROCESS B
Lower Spec. Limit Upper Spec. Limit
Process A has acceptable variation when evaluated against customer specification limits
Process B has unacceptable variation when evaluated against customer specification limits
When a process is in statistical control and has unacceptable variation, work on the reduction of variation due to common causes To reduce common cause variation, make improvements to the Vital Few X variables
D M A
Selecting Measures For Control Charts What Do I Monitor?

Process
14
Input Variables (Xs)
Output (Ys)
Key input and process measures (X) that track variables identified in your project as key drivers of Project (Y) variables
Key output measures (Y) from the customers perspective
Process Variables (Xs)
Measure The Process And Not The People
The measures that you choose to monitor in the control plan are the critical measures for your process. These are the process management measures that the process owner will rely on to track process performance over time.
D M A
Selecting Measures For Control Charts (continued) How Do I Monitor The Control Plan?
15
Establish Data Collection Goals
Develop Operational Definitions & Procedures
Ensure Data Consistency & Stability
Collect Data & Monitor Consistency
Use the 4-step data collection plan in order to develop a process that will protect the integrity of the data.
Following the 4-step data collection plan will ensure that the data collected on an ongoing basis will accurately reflect the variation in the process. Since this data will be collected over a long period of time, measurement systems analysis to ensure consistency should be a major concern for both the project team as well as the process owner after the hand-off. Make sure that periodic audits of the data collection are an integral part of the overall control plan.
D M A
Types Of Errors In Control Charts 3 -level Control Limits

Created by Dr. Walter Shewhart to minimize two types of mistakes Placed empirically because they minimize the two types of mistakes Are not probability limits
16
Two types of Mistakes:

Calling a special cause of variation a common cause of variation (Missing a chance to identify a change in the process) Calling a common cause of variation a special cause of variation (Interfering with a stable process, wasting resources looking for special causes of variation that do not exist)
Control limits vs. Specification limits

Control limits are determined by process data Specification limits are determined by customer needs
D M A
Two Types Of Control Charts

VARIABLE CHART
17
Uses Measured Values Cycle Time, Lengths, Diameters, etc. Generally One Characteristic Per Chart More Expensive, But More Information
ATTRIBUTE CHART
Pass/Fail, Good/Bad, Go/No-Go Information Can Be Many Characteristics Per Chart Less Expensive, But Less Information
Variable = continuous data Attribute = discrete data
D M A
Selecting the Appropriate Control Chart
18
Variable or Attribute Data?*
Variable Variable
Attribute Attribute
No
Rational Subgrouping Possible?
Yes
No
Constant Lot Size?
Yes
Defects or Defective
Individuals Individuals & & Moving Moving Range Range
Xbar Xbar & & Range Range
u u
p p
c c
np np
There are many different types of control charts. The chart pictured here is a helpful guide to decide which type of chart to use and when. The primary determinant of which type of control chart to use is the type of data being analyzed. If the data being analyzed is continuous (variable), you would use either:
The Individuals & Moving Range (I-MR): use if Rational Subgrouping is not possible (you want to see each transaction on the control chart) The Xbar & Range (X - R) use if Rational Subgrouping is possible. Typically subgroup size will be between 3 - 5. Use (X - S) if subgroup size is 7 or greater
If the data being analyzed is discrete (attribute), you would need to determine: If you have a constant lot (sample) size or if the lot (sample) size is not constant If you are counting defects, you may have multiple defects per unit. If you are counting defectives, the unit either is or is not defective. In this module we will study the different types of control charts and when to use each one. What you want the Control Chart to tell you is: Is my level of non-conformance consistent? Am I controlling my significant Xs?
* Refer to Memory Jogger
TM
for further details.
D M A
Interpreting Control ChartsActivity (20 minutes)

Desired Outcome: Practice deciding which control chart is the appropriate choice for given data examples What Preparation Review and Feedback How Choose a partner for the exercise For each of the examples listed on the following page Determine the appropriate type of control chart Determine how you would label the X and Y axes Close Keep notes on your responses for report-out Partners 10 mins. Who All Partners Timing 1 min. 9 mins.
19
D M A
Selecting Control Charts-Partner Activity (continued)
20
1. A team from Marketing is tracking response rates (yes, no) on telemarketing calls. Each day, they randomly sample 100 responses from their outgoing calls. The number of positive responses is plotted on the chart. 2. The technical support area is monitoring average response time on help requests. Each hour, 5 calls are sampled. 3. A team is monitoring the number of applications received each day with incomplete fields. All applications are being audited. Each day, 200-250 applications are received. 4. A team is monitoring the cycle time for the underwriters loan decisions. The data comes in slowly; that is, one loan is completed daily. 5. A team is tracking the number of fields left blank on an application. Each day, a sample of 100 applications is audited. 6. Another team is doing the same thing as the team in example 5 (counting the number of fields left blank on an application), only they are looking at all the applications that come in daily. Each day, 50-100 applications are turned in. 7. The team is tracking the number of abandoned calls to a customer call center in a shift. Each shift takes from 200-400 calls.
D M A
21
Variable Control Charts Xbar R-Chart
In class we will show you how to construct in Minitab two continuous Control Charts (Xbar and R Chart and I/MR Chart) and one Discrete Chart (p Chart). The remainder will be reference material. Note: All Control Chart interpretation is identical.
D M A
Selecting The Appropriate Control Chart
22
Variable Variable
Attribute Attribute
No
Yes
No
Constant Lot Size?
Yes
u u
p p
c c
np np
We will now focus on the Variable Control Chart with Rational Subgrouping: the Xbar & Range Chart.
D M A
A Valid Variable Control Chart Has Data in time or production sequence

To show stability, time-to-time variation
23
A measure of central tendency

To portray behavior of process center
A measure of variability Control limits

To allow separating common cause from assignable (special) cause
Xbar & Range Charts Xbar Chart: a plot of the sample means over time R-Chart: a plot of the range (difference between highest and lowest values) of a sample over time
General Comments on Xbar & Range Charts:

In Minitab: Stat > Control Charts > Xbar-R
Cluster or periodic measurements of characteristic Frequency depends on speed and stability of process. Subgroup averages are plotted on Xbar chart Xbar chart monitors central tendency of a process over time Subgroup ranges plotted on Range Chart R-Chart monitors the variability of a process over time Provides data-smoothing effect
D M A
24
Variable Control Charts I & MR Chart
D M A
Selecting The Appropriate Control Chart
25
Variable Variable
Attribute Attribute
No
Yes
No
Constant Lot Size?
Yes
u u
p p
c c
np np
D M A
Individuals & Moving Range Charts Can use in intermittent operations Similar to Xbar & R-Charts, Except...
Single Values, Not Subgroups Range Values must be artificially constructed Somewhat Noisier, since you track each individual value (transaction,etc.)
26
Individuals Chart: a plot of the individual values over time Moving Range Chart: a plot of the moving range (for two samples |Xi ( Xi-1)|over time)
General Comments on I & MR Charts:

Range value artificially constructed from successive readings Subgroup size for X is n = 1; for MR is usually 2 Some correlation between charts is possible More noise in charttougher to spot true process shift Displays the variability between individual observations over time Assumes that past and present data is equally important
In Minitab: Stat > Control Charts > I-MR
D M A
Building An Individuals And Moving Range Chart

Individual Data Moving Range 55 56 59 55 N/A ABS(55-56) = 1 ABS(56-59) = 3 ABS(59-55) = 4 Individuals
27
4 3 2 1 0
Moving Range
An Individuals chart plots actual dataeach transaction. A Moving Range chart calculates ranges between consecutive data points (absolute value). You lose one data point, since the range is not calculated for the first point.
D M A
Take AwaysVariable Control Charts
28
Variable control charts can be used with continuous data to tell when a process is:
Experiencing only common cause variation and working at its intended best When the process is disturbed and needs corrective action
Control charts:
Time ordered plot of data Reflect the expected range of variation of the data Identify when a special cause appears to be influencing the data
Xbar & R-Charts are used for plotting means and ranges of subgroups over time I & MR charts are used for plotting individual values and moving ranges over time
D M A
Take AwaysVariable Control Charts
29
Control limits are typically calculated as + 3 standard deviations away from the mean of the process Control limits and specification limits are not the same
Control limits are calculated from the sample data; they are internal to the process Specification limits are determined by your performance standard; they are external to the process (generally come from the customer)
Know when a process is out of control: Western Electric Rules Control charts are only as good as the actions that you take to keep the process in control
D M A
Variable Control Charts: Activity (10 minutes)

Objective
30
Practice using Minitab to generate an Xbar and R Chart. Think about the correct Data File: Variable Control Charts.mtw rational subgroup. The quality team has 25 weeks of historical data, contained in the file listed above. You have been asked to examine their data and believe an Xbar and R Chart, or an I/MR Chart would be the appropriate tool to use. 1. Plot the data using an I and MR chart. 2. Plot the data as Xbar/R-charts, forming subgroups by shift, day, and week. 3. For each chart created in 1 and 2 above, answer the following: A. Is the chart for spread stable? B. Is the chart for center stable?
Instructions
Time
10 minutes
Stat > Control Charts > Xbar-R.
D M A
Minitab Output
Stat > Control Chart > I-MR Variable: Response Time Click on TESTS: Perform All 8 Tests
31
I and MR Chart for Response

600
Response Time
Individual Value
500 400 300 200 100 0 -100 0 1 2 1 1 1 11 5 4 4 4 1 1 1 5 6 2 11 1 11 1 5 5 5 1 1 7 7 1 1 1 1 1 1 4 4 4 4 3.0SL=205.2 X=84.00 7 7 7 2 1000 -3.0SL=-37.19
Subgroup
500
Moving Range
500 400 300 200 100 0 2 2 2 22 2 1 1 1 1 2 1 1 1 1 1
1 1
1 2
11 11 1
1 1
1 1 1
1 1 1 11
1 1
3.0SL=148.9 2 2 2 2 2 2 2 R=45.57 -3.0SL=0.00E+00
D M A
Minitab Output
(continued)
32
Stat > Control Charts > Xbar-R Single Cause: Response Time Subgroup Size: Day Click on > Tests: Perform All 8 Tests
SUBGRP.MTW: Xbar/R for Response Time
Xbar/R Chart for Response Time: Day
D M A
33
Attribute Control Charts
D M A
Selecting The Appropriate Control Charts Attribute Data
34
Variable Variable
Attribute Attribute
No
Yes
No
Constant Lot Size?
Yes
u u
p p
c c
np np
If the data being analyzed is discrete (attribute), you would need to determine:

If you have a constant lot (sample) size or if the lot (sample) size is not constant
If you are counting defects you may have multiple defects per unit. If you are counting defectives, the unit either is or is not defective. In this module, we will study the different types of control charts and when to use each one. What you want the Control Chart to tell you is:

Is my level of non-conformance consistent?
Am I controlling my significant Xs? TM * Refer to Memory Jogger for further details.
D M A
Important Definitions A Defect

A single characteristic that does not meet requirements
35
A Defective
A unit that contains one or more defects
Attribute Charts Can Consider Either Case, Depending On The Chart Type Chosen
Other terminology may also be used:

A defect is also known as a nonconformance A defective is known as a nonconforming
D M A
Classification Of Attribute Chart Types
36
Constant Sample Size
Variable Sample Size
Defects
Can not count non-occurrences
Defective
np
Can count occurrences
To choose the appropriate attribute control chart, use the following factors: 1. Is it defects or defectives that you are investigating in the data you are collecting? For example:

Is it pass/fail for the part or process? Or is it several sub-components or sub-processes passed/failed for the part or process)?
2. Is it a constant lot (sample) size or variable lot (sample) size?
D M A
37
Attribute Control Charts P-Chart
D M A
Selecting The Appropriate Control Chart P-Chart
38
Variable Variable
Attribute Attribute
No
Yes
No
Constant Lot Size?
Yes
u u
p p
c c
np np
D M A
P-Chart Chart of proportion defective Variable subgroup/lot size (n)
39

Pchart shows the Proportion Defectivethe proportion that do not conform The units may have one or more defects
D M A

Desired Outcome: Practice interpreting P-Chart What Preparation How In your team, read background information on the next page Who Team Timing 5 mins.
40
Interpret Control Charts
Interpret P-Chart results and answer questions
Team
5 mins.
Close Exercise
Choose a spokesperson to report out on your responses to the questions
All
5 mins.
D M A
Interpreting Control ChartsActivity
41
The loan department has made a change to the loan application process. They think the change will help weed out the bad loan risks before the applications are actually submitted. If they are right, the approval rate for submitted applications should increase. They want to check to see if this change in the process has had the desired effect. 1. Is there evidence that the new process made a difference? 2. What is the number of opportunities for defect in this process measure? 3. Is this an input, process, or output variable? Why?

3.0SL = 97.71
May 1 To November 5
% Applications Approved
P-Chart Of Loan Approval Rate
90
New Process Implemented
93.79 89.86
P = 85.94
82.02
80
78.09
-3.0SL = 74.18 70 0 10 20 Week (Time Order) 30
Note: SL is the standard Minitab notation for Sigma Limits which are the same as Control Limits.
To create a p-chart in Minitab: STAT > CONTROL CHARTS > P
D M A
Reference MaterialsAttribute Control Charts C-Chart Reference Materials Attribute Control Charts C-Chart
42
Pages 43-60 cover c, np and u charts. All Attribute Charts are interpreted as discussed in the previous pages for p-charts. Review these charts with your mentor for additional exercises. Proceed to page 61
Stat > Control Charts: All 4 Attribute Charts: (p,np,c,u) can be generated from this menu.
D M A
Selecting The Appropriate Control Chart Attribute Data
43
Variable Variable
Attribute Attribute
No
Yes
No
Constant Lot Size?
Yes
u u
p p
c c
np np
D M A
C-Chart Chart for counts Based on Poisson distribution

High probability of finding defect of some type. Large samples are needed if defect probabilities are low. Lower probability of a defect of a given type
44
Works best on complex unit of product Constant subgroup/lot size
C bar = DPU = number of defects detected/subgroup A subgroup may contain one or more physical units. You need C bar up around 5. If it is less than 5, your distribution will be skewed because it will be truncated by 0. Increase your subgroup size if you need to increase C bar or collapse existing categories into meaningful groups. More specifically, combine two or more categories so that the frequencies are greater than 5.
D M A
C
45

Desired Outcome: Practice interpreting C-Chart What Preparation How In your team, read background information. Review chart on next 2 pages Interpret C-Chart by answering the questions on the page Choose a spokesperson to report out on your responses to the questions Who All Timing 5 mins.
Interpret Control Charts Close Exercise
Facilitator
(Sub-group)
5 mins.
All
5 mins.
D M A
Interpreting Control ChartsActivity (continued)
46
An applications group of a large insurer recently made some changes to the design of their application. The old applications had 10 fields for the customer to fill out; the new application has 5 fields for the customer to fill out. They are now monitoring the incoming applications to ensure the new design has had an impact on completeness. A random sample of 100 applications are reviewed each day and the total number of incomplete fields are plotted. The chart for this data is on the following page. 1. Why is the control chart a C-Chart? 2. Is the number of errors in a sample stable over the 20day sample? 3. What questions does the chart raise? Extra Credit!! Calculate baseline sigma for the first 20 days of service for the new application.
Day 1 2 3 4 5 Data 5 5 11 5 8 Day 6 7 8 9 10 Data 9 9 8 8 7 Day 11 12 13 14 15 Data 6 6 7 5 6 Day 16 17 18 19 20 Data 7 7 8 12 13
D M A
47
3.0SL=15.87 15
Number of Incomplete Fields
10
C=7.600
-3.0SL=0.00E+00
10
20
Day of Sample
D M A
48
Attribute Control Charts U-Chart
D M A
Selecting The Appropriate Control Chart Attribute Data
49
Variable Variable
Attribute Attribute
No
Yes
No
Constant Lot Size?
Yes
u u
p p
c c
np np
D M A
U-Chart Chart for defects per unit, with variable lot (subgroup) size Same logic as C-Chart, except variable lot (subgroup) size (n)
50
Compare to C-Chart
In Minitab: Stat > Control Charts > I
D M A
U-Chart Activity (20 minutes)

Desired Outcome: Practice constructing and interpreting U-Chart What Preparation How In your team, prepare to create a U-Chart in Minitab In the U-Chart.mtw, column errors contains data (in time sequence) of the defects found each day on customer orders. A defect is defined to be inaccurate information found on a customer order. Both the number of defects and the daily number of orders are recorded. Using Minitab, construct a U-Chart of the data What are your observations? Close Choose a spokesperson to report out your conclusions All 9 mins. Who All Timing 1 min.
51
Construct And Interpret The U-Chart
Team
10 mins.
Note: Do not turn the page until you have completed the activity.
D M A
U-Chart Example Minitab Menu Commands

MINITAB FILE: U_Chart.mtw
52
D M A
U-Chart Minitab Input & Output
53
U Chart for errors
Sample Count
3.0SL=2.114 U=1.764 -3.0SL=1.415
1 0 10 20 30
Sample Number
CONTROL: The process does not look stablethere are indications pf special cause variations. Why do the control limits vary? Lot (Sample) size varies.
D M A
54
Attribute Control Charts NP-Chart
D M A
Selecting The Appropriate Control Chart NP-Chart
55
Variable Variable
Attribute Attribute
No
Yes
No
Constant Lot Size?
Yes
u u
p p
c c
np np
D M A
NP-Chart NP-chart: number of defectives (non-conforming items) in subgroup Same logic as the P-Chart, except constant subgroup/lot size (n)
56
D M A

Desired Outcome: Practice interpreting NP-Chart What Preparation How In your team, read background information on next 2 pages Interpret NP-Chart results and answer questions Choose a spokesperson to report out on your responses to the questions Who Team Timing 5 mins.
57
Interpret Control Charts Close Exercise
Team
5 mins.
All
5 mins.
D M A

An applications group of a large insurer recently made some changes to the design of their application. They are now monitoring the incoming applications to ensure the new design has had an impact on completeness. 150 applications are chosen randomly and reviewed each day. The number that are incomplete are plotted each day The chart for this data is on the following page. 1. Why is the chart an NPchart? 2. What questions does the chart raise about the completeness of the insurance application? 3. What is the number of opportunities for defect in this measure? 4. What is the difference between the value plotted on a Pand an NPChart?
Day 1 2 3 4 5 Data 12 22 14 15 12 Day 6 7 8 9 10 Data 14 16 19 11 1 Day 11 12 13 14 15 Data 2 15 8 12 12 Day 16 17 18 19 20 Data 11 17 12 11 21
58
D M A
59
25 3.0SL=23.13
20
# of Incomplete Applications
15 NP=12.85 10
Day of Sample
5 -3.0SL=2.567 0 0 10 20
D M A
Attribute Chart Subgroup Size Rule of Thumb:

Select a subgroup size that will provide an average defect/defective of approximately
C, U, NP > 5.0
60
To Make UCL & LCL Nearly Symmetrical Around the Mean

For NP-charts, to select the appropriate sample size such that 95% of the subgroups will have at least one defective, use the relationship
n= 3
D M A
Summary Of Attribute Charts
61
Useful when variable data not available Use count/classification datapass/fail, good/bad Same general rules for interpretation as variable charts Useful as end-to-end overview; use variable charts for further study of problems Can use data gathered for other purposes Generally less expensive to administer, but tell you less Shortcomings Including too many variables makes interpretation difficult Must fit the parameters you are evaluating to theoretical distribution [Poisson (C, U-Charts), Binomial (P, NP-Charts)] Need to evaluate whether constant/non-constant lot size will help you with root cause analysis Sensitivity is dependent on magnitude of defect level

Variable Variable
Attribute Attribute
No
Yes
No
Constant Yes Lot Size?
Individuals Individuals& & Moving Moving Range Range

DMAIC GB V TX PG
Xbar Xbar& & Range Range u u p p c c np np

D M A
Take Aways Attribute Control Charts Attribute control charts are used to monitor the level of nonconformance of a process Select the appropriate attribute control chart based upon
Constant vs. variable lot (sample) size Defects vs. defectives
62
Defect
A single characteristic that does not meet requirements
Defective
A unit that contains one or more DEFECTS
D M A
SummaryControl Charts All control charts have the same form

Time plot of data Statistical limits placed +/- 3s from center line
63
Choose the control chart according to:

Data type Sample size
Interpret and act on the control chart

Investigate special causes of variation Assess baseline sigma in order to understand if we are performing to customer CTQs Shrink common cause variation by making fundamental changes in the Vital Few X variables
Maintain the control chart

Make notes on the control chart to indicate problems, changes, or important events
D M A
64
Calculating Control Limits

(Optional)
D M A
Control Limit Calculations The formulas for calculating control limits are shown on the following pages There are different tables for continuous and discrete data Both continuous and discrete data control charts have control limits that are placed +/- 3 estimated sigma from the average line
65
D M A
Control Limit CalculationsFormulas
66
Continuous Data Control limits are calculated using control chart factors and the Range-Bar (an estimate of short-term sigma) Control chart factors were invented by Shewart in the 1920s to avoid long-hand calculation Control chart factors are shown in the table according to the sample size for each subgroup Individual control charts are considered to have a sample of size 2, the number of data points make up the moving range Discrete Data Control limits are calculated using a formula that estimates sigma without the necessity to transform the data The normal estimate for sigma (under radical) is then multiplied by three
D M A
Control Limit CalculationsContinuous Data Table

Type Control Chart Sample Size n Central Line* Control Limits
67
Average and Range < 10, but usually 3 to 5 X and R
X=
(X1 + X2 +Xk) k
UCLx = X + A2R LCLx = X - A2R UCLR = D4R LCLR = D3R UCLx = X + A3s LCLx = X - A3s UCLs = B4s LCLs = B3s
R=
(R1 + R2 +Rk) k (X1 + X2 + Xk) k (s1 + s2 +sk) k
Average and Standard Deviation X and s Median and Range
X= Usually 10 s=
X and R
< 10, but usually 3 to 5
X =
1 + X 2 + X k) (X k
(R1 + R2 +Rk) k (X1 + X2 + Xk) k I(Xi+1 - Xi)I (R1 + R2 +Rk-1) k-1
+ A2R UCLx = X LCLx = X - A2R

UCLR = D4R LCLR = D3R UCLx = X + E2Rm LCLx = X - E2Rm
R=
Individuals and Moving Range 1 X and Rm
X= Rm = Rm =
UCLRm = D4Rm LCLRm = D3Rm
k = # of subgroups, X = median value within each subgroup *X =

Xi n
D M A
Control Limit CalculationsTable Of Contents

Sample Size n X and R Chart X and s Chart
68
A2 1.880 1.023 0.729 0.577 0.483 0.419 0.373 0.337 0.308
D3 0 0 0 0 0 0.076 0.136 0.184 0.223
D4 3.267 2.574 2.282 2.114 2.004 1.924 1.864 1.816 1.777
A3 2.659 1.954 1.628 1.427 1.287 1.182 1.099 1.032 0.975
B3 0 0 0 0 0.030 0.118 0.185 0.239 0.284
B4 3.267 2.568 2.266 2.089 1.970 1.882 1.815 1.761 1.716
c4* .7979 .8862 .9213 .9400 .9515 .9594 .9650 .9693 .9727
2 3 4 5 6 7 8 9 10
Sample Size n
and R Chart X 2 A
1.187 0.691 0.509 0.412 D3 0 0 0 0 0 0.076 0.136 0.184 0.223 D4 3.267 2.574 2.282 2.114 2.004 1.924 1.864 1.816 1.777 E2 2.659 1.772 1.457 1.290 1.184 1.109 1.054 1.010 0.975
X and Rm Chart
D3 0 0 0 0 0 0.076 0.136 0.184 0.223
D4 3.267 2.574 2.282 2.114 2.004 1.924 1.864 1.816 1.777
d2* 1.128 1.693 2.059 2.326 2.534 2.704 2.847 2.970 3.078
2 3 4 5 6 7 8 9 10
D M A
Control Limit CalculationsDiscrete Data Table
69
Type Control Chart
Sample Size
Central Line
Control Limits
Fraction Defective p-Chart
Variable, usually 50
For each subgroup: p = np/n For all subgroups: np/ n p=
*UCLp = p + 3 *LCLp = p - 3 UCLnp = np + 3 LCLnp = np - 3 UCLc = c + 3 LCLc = c - 3 *UCLu = u + 3 *LCLu = u - 3
p(1 - p) n p(1 - p) n np(1 - p)
Number Defective np-Chart Number of Defects c-Chart Number of Defects Per Unit u-Chart
Constant, usually 50
For each subgroup: np = # defectives For all subgroups: np/k np = For each subgroup: c = # defects For all subgroups: c/k c= For each subgroup: u = c/n For all subgroups: u= c/ n
np(1 - p) c c u n u n
Constant
Variable
np = # defectives c = # of defects n = sample size within each subgroup k = # of subgroups
* This formula creates changing control limits. To avoid this, use average sample sizes n for those samples that are within 20% of the average sample size. Calculate individual limits for the samples exceeding 20%. If the Lower Control Limit (LCL) is a negative number, set the LCL to zero.
D M A

Six Sigma Manual

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Six Sigma Manual

Uploaded by

Copyright:

Available Formats

D M A

GE Capital, Inc., 2000 DMAIC GB A TX PG V4.2.0

* Elevator Speeches will be shared throughout the day

GE Capital, Inc., 2000 DMAIC GB A TX PG V4.2.0

Analyze 5Define Performance Objective 5

* Elevator Speeches will be shared throughout the day

GE Capital, Inc., 2000 DMAIC GB A TX PG V4.2.0

GE Capital, Inc., 2000 DMAIC GB B TX PG V 4.2.0

OverviewClassroom Learning Objectives

GE Capital, Inc., 2000 DMAIC GB B TX PG V 4.2.0

GE Capital, Inc., 2000 DMAIC GB B TX PG V 4.2.0

GE Capital, Inc., 2000 DMAIC GB B TX PG V 4.2.0

Compile and evaluate customer CTQ data.

GE Capital, Inc., 2000 DMAIC GB B TX PG V 4.2.0

GE Capital, Inc., 2000 DMAIC GB B TX PG V 4.2.0

DefineProject Learning Objectives (continued)

GE Capital, Inc., 2000 DMAIC GB B TX PG V 4.2.0

GE Capital, Inc., 2000 DMAIC GB B TX PG V 4.2.0

GE Capital, Inc., 2000 DMAIC GB B TX PG V 4.2.0

DefineTest Learning Objectives

GE Capital, Inc., 2000 DMAIC GB B TX PG V 4.2.0

Measure1Classroom Learning Objectives 1 Describe and define the deliverables of Step 1.

GE Capital, Inc., 2000 DMAIC GB B TX PG V 4.2.0

GE Capital, Inc., 2000 DMAIC GB B TX PG V 4.2.0

GE Capital, Inc., 2000 DMAIC GB B TX PG V 4.2.0

Measure 1Project Learning Objectives (continued)

GE Capital, Inc., 2000 DMAIC GB B TX PG V 4.2.0

Measure 1Project Learning Objectives (continued)

GE Capital, Inc., 2000 DMAIC GB B TX PG V 4.2.0

GE Capital, Inc., 2000 DMAIC GB B TX PG V 4.2.0

MeasureTest Learning Objectives

GE Capital, Inc., 2000 DMAIC GB B TX PG V 4.2.0

GE Capital, Inc., 2000 DMAIC GB B TX PG V 4.2.0

MeasureTest Learning Objectives

GE Capital, Inc., 2000 DMAIC GB B TX PG V 4.2.0

MeasureTest Learning Objectives

GE Capital, Inc., 2000 DMAIC GB B TX PG V 4.2.0

MeasureTest Learning Objectives

GE Capital, Inc., 2000 DMAIC GB B TX PG V 4.2.0

MeasureTest Learning Objectives

GE Capital, Inc., 2000 DMAIC GB B TX PG V 4.2.0

MeasureTest Learning Objectives

GE Capital, Inc., 2000 DMAIC GB B TX PG V 4.2.0

GE Capital, Inc., 2000 DMAIC GB B TX PG V 4.2.0

AnalyzeClassroom Learning Objective (continued)

GE Capital, Inc., 2000 DMAIC GB B TX PG V 4.2.0

AnalyzeProject Learning Objectives

GE Capital, Inc., 2000 DMAIC GB B TX PG V 4.2.0

AnalyzeTest Learning Objectives

GE Capital, Inc., 2000 DMAIC GB B TX PG V 4.2.0

AnalyzeTest Learning Objectives

GE Capital, Inc., 2000 DMAIC GB B TX PG V 4.2.0

AnalyzeTest Learning Objectives

GE Capital, Inc., 2000 DMAIC GB B TX PG V 4.2.0

AnalyzeTest Learning Objectives

GE Capital, Inc., 2000 DMAIC GB B TX PG V 4.2.0

GE Capital, Inc., 2000 DMAIC GB B TX PG V 4.2.0

AnalyzeTest Learning Objectives (continued)

GE Capital, Inc., 2000 DMAIC GB B TX PG V 4.2.0

AnalyzeTest Learning Objectives (continued)

GE Capital, Inc., 2000 DMAIC GB B TX PG V 4.2.0

GE Capital, Inc., 2000 DMAIC GB B TX PG V 4.2.0

GE Capital, Inc., 2000 DMAIC GB B TX PG V 4.2.0

GE Capital, Inc., 2000 DMAIC GB B TX PG V 4.2.0