MSA_Reference_Manual_4th_Edition.pdf

i
MEASUREMENT SYSTEMS ANALYSIS
Reference Manual
Fourth Edition
First Edition, October 1990 • Second Editi
on, February 1995; Second Printing, June 1998
Third Edition, March 2002; Second Printing, May 2003; Fourth Edition, June 2010
Copyright © 1990, © 1995, © 2002, © 2010 Chrysler Group LLC, Ford Motor Company, General Motors Corporation
ISBN#: 978-1-60-534211-5

iii
FOREWORD
This Reference Manual was developed by a Measurement Systems Analysis (MSA) Work Group,
sanctioned by the Chrysler Group LLC, Ford Motor Company, and General Motors Corporation Supplier
Quality Requirements Task Force, and under the auspices of the Automotive Industry Action Group
(AIAG). The Work Group responsible for this Fourth Edition were Michael Down (General Motors
Corporation), Frederick Czubak (Chrysler Group LLC), Gregory Gruska (Omnex), Steve Stahley
(Cummins, Inc.) and David Benham.
The manual is an introduction to measurement system analysis. It is not intended to limit evolution of
analysis methods suited to particular processes or commodities. While these guidelines are intended to
cover normally occurring measurement system situations, there will be questions that arise. These
questions should be directed to your authorized customer representative.
This Manual is copyrighted by Chrysler Group LLC, Ford Motor Company, and General Motors
Corporation, with all rights reserved, 2010. Additional manuals can be ordered from AIAG at
U
www.aiag.orgT
. Permission to reproduce portions of this manual for use within supplier organizations may
be obtained from AIAG at U
www.aiag.orgT
June 2010

iv
MSA 4th
Edition Quick Guide
Type of
Measurement System
MSA Methods Chapter
Basic Variable
Range, Average & Range, ANOVA,
Bias, Linearity, Control Charts
III
Basic Attribute
Signal Detection,
Hypothesis Test Analyses
III
Non-Replicable
(e.g., Destructive Tests)
Alternate Approaches IV
Complex Variable
Range, Average & Range, ANOVA,
Bias Linearity Control Charts
III, IV
Multiple Systems, Gages
or Test Stands
Control Charts ANOVA Regression Analysis III, IV
Miscellaneous Alternate Approaches IV
Other
White Papers – available at
AIAG website (www.aiag.org)
NOTE: Regarding the use of the GRR standard deviation
Historically, by convention, a 99% spread has been used to represent the “full” spread of
measurement error, represented by a 5.15 multiplying factor (where GRR
σ is multiplied by 5.15
to represent a total spread of 99%).
A 99.73% spread is represented by a multiplier of 6.0, which is 3σ
 and represents the full
spread of a “normal” curve.
If the reader chooses to increase the coverage level, or spread, of the total measurement
variation to 99.73%, use 6.0 as a multiplier in place of 5.15 in the calculations.
Note: The approach used in the 4th
Edition is to compare standard deviations. This is equivalent
to using the multiplier of 6 in the historical approach.
Awareness of which multiplying factor is used is crucial to the integrity of the equations and
resultant calculations. This is especially important if a comparison is to be made between
measurement system variability and the tolerance. Consequently, if an approach other than that
described in this manual is used, a statement of such must be stated clearly in any results or
summaries (particularly those provided to the customer).

v
TABLE OF CONTENTS
MSA 4th
Edition Quick Guide .....................................................................................................................................iv
TABLE OF CONTENTS ............................................................................................................................................v
List of Tables............................................................................................................................................................. vii
List of Figures .......................................................................................................................................................... viii
CHAPTER I General Measurement System Guidelines.........................................................................................1
Section A Introduction, Purpose and Terminology ...................................................................................................3
Introduction ...........................................................................................................................................................3
Purpose ..................................................................................................................................................................4
Terminology ..........................................................................................................................................................4
Section B The Measurement Process P
.....................................................................................................................13
Measurement Systems .........................................................................................................................................13
The Effects of Measurement System Variability.................................................................................................18
Section C Measurement Strategy and Planning.......................................................................................................25
Section D Measurement Source Development .......................................................................................................29
Gage Source Selection Process............................................................................................................................31
Section E Measurement Issues ................................................................................................................................41
Section F Measurement Uncertainty........................................................................................................................63
Section G Measurement Problem Analysis .............................................................................................................65
CHAPTER II General Concepts for Assessing Measurement Systems...............................................................67
Section A Background.............................................................................................................................................69
Section B Selecting/Developing Test Procedures....................................................................................................71
Section C Preparation for a Measurement System Study ........................................................................................73
Section D Analysis of the Results............................................................................................................................77
CHAPTER III Recommended Practices for Replicable Measurement Systems ................................................81
Section A Example Test Procedures........................................................................................................................83
Section B Variable Measurement System Study Guidelines...................................................................................85
Guidelines for Determining Stability...................................................................................................................85
Guidelines for Determining Bias P
F – Independent Sample Method ...................................................................87
Guidelines for Determining Bias – Control Chart Method..................................................................................92
Guidelines for Determining Linearity P
................................................................................................................96
Guidelines for Determining Repeatability and Reproducibility P
......................................................................101
Range Method....................................................................................................................................................102
Average and Range Method ..............................................................................................................................103
Analysis of Variance (ANOVA) Method ..........................................................................................................123
Section C Attribute Measurement Systems Study.................................................................................................131
Risk Analysis Methods......................................................................................................................................131
Signal Detection Approach................................................................................................................................143
Analytic Method P
..............................................................................................................................................145
CHAPTER IV Other Measurement Concepts and Practices .............................................................................151
Section A Practices for Non-Replicable Measurement Systems ...........................................................................153
Destructive measurement systems.....................................................................................................................153
Systems where the part changes on use/test ......................................................................................................153
Section B Stability Studies ....................................................................................................................................155
Section C Variability Studies.................................................................................................................................161
Section D Recognizing the Effect of Excessive Within-Part Variation................................................................167
Section E Average and Range Method – Additional Treatment............................................................................169
Section F Gage Performance Curve P
.....................................................................................................................177
Section G Reducing Variation Through Multiple Readings..................................................................................183
Section H Pooled Standard Deviation Approach to GRR P
....................................................................................185
APPENDICES..........................................................................................................................................................193

vi
Appendix A...............................................................................................................................................................195
Analysis of Variance Concepts..............................................................................................................................195
Appendix B...............................................................................................................................................................199
Impact of GRR on the Capability Index Cp...........................................................................................................199
Formulas:...............................................................................................................................................................199
Analysis:................................................................................................................................................................199
Graphical Analysis ................................................................................................................................................201
Appendix C...............................................................................................................................................................203
Appendix D...............................................................................................................................................................205
Gage R Study.........................................................................................................................................................205
Appendix E...............................................................................................................................................................207
Alternate PV Calculation Using Error Correction Term........................................................................................207
Appendix F...............................................................................................................................................................209
P.I.S.M.O.E.A. Error Model..................................................................................................................................209
Glossary....................................................................................................................................................................213
Reference List ..........................................................................................................................................................219
Sample Forms ..........................................................................................................................................................223
Index .........................................................................................................................................................................227

vii
List of Tables
Table I-B1: Control Philosophy and Driving Interest..................................................................................18
Table II-D 1: GRR Criteria...........................................................................................................................78
Table III-B 1: Bias Study Data....................................................................................................................90
Table III-B 2: Bias Study – Analysis of Bias Study P
..................................................................................92
Table III-B 3: Bias Study – Analysis of Stability Study for Bias..................................................................95
Table III-B 4: Linearity Study Data.............................................................................................................99
Table III-B 5: Linearity Study – Intermediate Results ................................................................................99
Table III-B 6: Gage Study (Range Method) .............................................................................................103
Table III-B 6a: Gage Repeatability and Reproducibility Data Collection Sheet.......................................105
Table III-B 7: ANOVA Table.....................................................................................................................127
Table III-B 8: ANOVA Analysis % Variation & Contribution......................................................................127
Table III-B 9: Comparison of ANOVA and Average and Range Methods ...............................................129
Table III-B 10: GRR ANOVA Method Report..........................................................................................129
Table III-C 1: Attribute Study Data Set......................................................................................................134
Table III-C 2: Cross tabulation Study Results...........................................................................................136
Table III-C 3: Kappa Summary .................................................................................................................137
Table III-C 4: Comparisons of Appraisers to Reference...........................................................................138
Table III-C 5: Study Effectiveness Table...................................................................................................139
Table III-C 6: Example Effectiveness Criteria Guidelines.........................................................................140
Table III-C 7: Study Effectiveness Summary ............................................................................................140
Table III-C 8: Table III-C 1 sorted by Ref Value........................................................................................143
Table IV-A 1: Methods Based on Type of Measurement System............................................................154
Table IV-H 1: Pooled Standard Deviation Analysis Data Set ...................................................................189
Table A 1: Estimate of Variance Components.........................................................................................195
Table A 2: 6 Sigma Spread......................................................................................................................196
Table A 3: Analysis of Variance (ANOVA)...............................................................................................197
Table A 4: Tabulated ANOVA Results.....................................................................................................198
Table A 5: Tabulated ANOVA Results.....................................................................................................198
Table B 1: Comparison of Observed to Actual Cp...................................................................................201
Table C 1:
*
2
d Table..............................................................................................................................203
Table F 1: Examples of the PISMOEA Model..........................................................................................211

viii
List of Figures
U
Figure I-A 1: Example of a Traceability Chain for a Length MeasurementT ................................................10
U
Figure I-B 1: Measurement System Variability Cause and Effect DiagramT ................................................17
U
Figure I-E 2: DiscriminationT........................................................................................................................46
U
Figure I-E 3: Impact of Number of Distinct Categories (ndc) of the Process Distribution on Control and
Analysis ActivitiesT ................................................................................................................................47
U
Figure I-E 4: Process Control ChartsT .........................................................................................................49
U
Figure I-E 5: Characteristics of the Measurement Process VariationT .........................................................50
U
Figure I-E 6: Relationships between Bias and RepeatabilityT .....................................................................62
U
Figure III-B 1: Control Chart Analysis for StabilityT .....................................................................................86
U
Figure III-B 2: Bias Study – Histogram of Bias StudyT ................................................................................91
U
Figure III-B 3: Linearity Study – Graphical AnalysisT ................................................................................100
U
Figure III-B 4: Average Chart – “Stacked”T
................................................................................................107
U
Figure III-B 5: Average Chart – “Unstacked”T
............................................................................................107
U
Figure III-B 6: Range Chart – “Stacked”T ..................................................................................................108
U
Figure III-B 7: Range Chart – “Unstacked”T ..............................................................................................109
U
Figure III-B 8: Run Chart by PartT .............................................................................................................109
U
Figure III-B 9: Scatter PlotT........................................................................................................................110
U
Figure III-B 10: Whiskers ChartT................................................................................................................111
U
Figure III-B 11: Error ChartsT.....................................................................................................................112
U
Figure III-B 12: Normalized HistogramT
.....................................................................................................113
U
Figure III-B 13: X–Y Plot of Averages by SizeT .........................................................................................114
U
Figure III-B 14: Comparison X–Y PlotsT ....................................................................................................115
U
Figure III-B 15: Completed GR&R Data Collection SheetT .......................................................................118
U
Figure III-B 16: Gage Repeatability and Reproducibility ReportT ..............................................................119
U
Figure III-B 18: Residual PlotT ...................................................................................................................126
U
Figure III-C 1: Example Process with Pp = Ppk = 0.50T.............................................................................132
U
Figure III-C 2: The “Gray” Areas Associated with the Measurement SystemT ...........................................132
U
Figure III-C 3: Example Process with Pp = Ppk = 1.33T.............................................................................141
U
Figure III-C 4: Attribute Gage Performance Curve Plotted on Normal Probability PaperT ........................149
U
Figure III-C 5. Attribute Gage Performance CurveT ..................................................................................150
U
Figure IV-E 1: Measurement Evaluation Control Chart (T &
X R U
) - 1T......................................................172
U
Figure IV-E 2: Measurement Evaluation Control Chart (T &
X R U
) - 2T......................................................173
U
Figure IV-E 3: Alternate Computations for Evaluating a Measurement Process (Part 1 of 2).T
................174
U
Figure IV-E 4: Alternate Computations for Evaluating a Measurement Process (Part 2 of 2).T
................175
U
Figure IV-F 1: Gage Performance Curve Without ErrorT...........................................................................180
U
Figure IV-F 2: Gage Performance Curve –ExampleT ................................................................................181
U
Figure IV-F 3: Gage Performance Curve Plotted on Normal Probability PaperT ......................................182
U
Figure IV-H 1: Pooled Standard Deviation Study Graphical AnalysisT .....................................................188
U
Figure IV-H 2: Dot diagram of h values.T
...................................................................................................191
U
Figure IV-H 3: Dot diagram of k values.T ...................................................................................................192
U
Figure B 1: Observed vs. Actual Cp (process based)T..............................................................................201
U
Figure B 2: Observed vs. Actual Cp (tolerance based)T ...........................................................................202

Chapter I
General Measurement System Guidelines
1
CHAPTER I
General Measurement System
Guidelines

Chapter I – Section A
Introduction, Purpose and Terminology
2

3
Quality of
Measurement
Data
Section A
Introduction
Measurement data are used more often and in more ways than ever before.
For instance, the decision to adjust a manufacturing process is now
commonly based on measurement data. The data, or some statistic calculated
from them, are compared with statistical control limits for the process, and if
the comparison indicates that the process is out of statistical control, then an
adjustment of some kind is made. Otherwise, the process is allowed to run
without adjustment. Another use of measurement data is to determine if a
significant relationship exists between two or more variables. For example, it
may be suspected that a critical dimension on a molded plastic part is related
to the temperature of the feed material. That possible relationship could be
studied by using a statistical procedure called regression analysis to compare
measurements of the critical dimension with measurements of the
temperature of the feed material.
Studies that explore such relationships are examples of what Dr. W. E.
Deming called analytic studies. In general, an analytic study is one that
increases knowledge about the system of causes that affect the process.
Analytic studies are among the most important uses of measurement data
because they lead ultimately to better understanding of processes.
The benefit of using a data-based procedure is largely determined by the
quality of the measurement data used. If the data quality is low, the benefit of
the procedure is likely to be low. Similarly, if the quality of the data is high,
the benefit is likely to be high also.
To ensure that the benefit derived from using measurement data is great
enough to warrant the cost of obtaining it, attention needs to be focused on
the quality of the data.
The quality of measurement data is defined by the statistical properties of
multiple measurements obtained from a measurement system operating under
stable conditions. For instance, suppose that a measurement system,
operating under stable conditions, is used to obtain several measurements of
a certain characteristic. If the measurements are all “close” to the master
value for the characteristic, then the quality of the data is said to be “high”.
Similarly, if some, or all, of the measurements are “far away” from the
master value, then the quality of the data is said to be “low”.
The statistical properties most commonly used to characterize the quality of
data are the bias and variance of the measurement system. The property
called bias refers to the location of the data relative to a reference (master)
value, and the property called variance refers to the spread of the data.
One of the most common reasons for low-quality data is too much variation.
Much of the variation in a set of measurements may be due to the interaction
between the measurement system and its environment. For instance, a

4
measurement system used to measure the volume of liquid in a tank may be
sensitive to the ambient temperature of the environment in which it is used.
In that case, variation in the data may be due either to changes in the volume
or to changes in the ambient temperature. That makes interpreting the data
more difficult and the measurement system, therefore, less desirable.
If the interaction generates too much variation, then the quality of the data
may be so low that the data are not useful. For example, a measurement
system with a large amount of variation may not be appropriate for use in
analyzing a manufacturing process because the measurement system’s
variation may mask the variation in the manufacturing process. Much of the
work of managing a measurement system is directed at monitoring and
controlling variation. Among other things, this means that emphasis needs to
be placed on learning how the measurement system interacts with its
environment so that only data of acceptable quality are generated.
Purpose
The purpose of this document is to present guidelines for assessing the
quality of a measurement system. Although the guidelines are general
enough to be used for any measurement system, they are intended primarily
for the measurement systems used in the industrial world. This document is
not intended to be a compendium of analyses for all measurement systems.
Its primary focus is measurement systems where the readings can be
replicated on each part. Many of the analyses are useful with other types of
measurement systems and the manual does contain references and
suggestions. It is recommended that competent statistical resources be
consulted for more complex or unusual situations not discussed here.
Customer approval is required for measurement systems analysis methods
not covered in this manual.
Terminology
The discussion of the analysis of measurement system can become confusing
and misleading without an established set of terms to refer to the common
statistical properties and related elements of the measurement system. This
section provides a summary of such terms which are used in this manual.
In this document, the following terms are used:
• Measurement is defined as “the assignment of numbers [or values]
to material things to represent the relations among them with respect
to particular properties.” This definition was first given by C.
Eisenhart (1963). The process of assigning the numbers is defined as
the measurement process, and the value assigned is defined as the
measurement value.

5
• Gage is any device used to obtain measurements; frequently used to
refer specifically to the devices used on the shop floor; includes
go/no-go devices (also, see Reference List: ASTM E456-96).
• Measurement System is the collection of instruments or gages,
standards, operations, methods, fixtures, software, personnel,
environment and assumptions used to quantify a unit of measure or
fix assessment to the feature characteristic being measured; the
complete process used to obtain measurements.
From these definitions it follows that a measurement process may be viewed
as a manufacturing process that produces numbers (data) for its output.
Viewing a measurement system this way is useful because it allows us to
bring to bear all the concepts, philosophy, and tools that have already
demonstrated their usefulness in the area of statistical process control.
Summary of TermsP
1
Standard
• Accepted basis for comparison
• Criteria for acceptance
• Known value, within stated limits of uncertainty, accepted as a true
value
• Reference value
A standard should be an operational definition: a definition which will
yield the same results when applied by the supplier or customer, with
the same meaning yesterday, today, and tomorrow.
Basic equipment
• Discrimination, readability, resolution
 Alias: smallest readable unit, measurement resolution, scale
limit, or detection limit
 An inherent property fixed by design
 Smallest scale unit of measure or output for an instrument
 Always reported as a unit of measure
 10 to 1 rule of thumb
• Effective resolution
 The sensitivity of a measurement system to process variation for
a particular application
1
See Chapter I, Section E for terminology definitions and discussion.

6
 Smallest input that results in a usable output signal of
measurement
• Reference value
 Accepted value of an artifact
 Requires an operational definition
 Used as the surrogate for the true value
• True value
 Actual value of an artifact
 Unknown and unknowable
Location variation
• Accuracy
 “Closeness” to the true value, or to an accepted reference value
 ASTM includes the effect of location and width errors
• Bias
 Difference between the observed average of measurements and
the reference value
 A systematic error component of the measurement system
• Stability
 The change in bias over time
 A stable measurement process is in statistical control with
respect to location
 Alias: Drift
• Linearity
 The change in bias over the normal operating range
 The correlation of multiple and independent bias errors over the
operating range
 A systematic error component of the measurement system

7
Width variation
• PrecisionP
2
 “Closeness” of repeated readings to each other
 A random error component of the measurement system
• Repeatability
 Variation in measurements obtained with one measuring
instrument when used several times by an appraiser while
measuring the identical characteristic on the same part
 The variation in successive (short-term) trials under fixed and
defined conditions of measurement
 Commonly referred to as E.V. – Equipment Variation
 Instrument (gage) capability or potential
 Within-system variation
• Reproducibility
 Variation in the average of the measurements made by different
appraisers using the same gage when measuring a characteristic
on one part
 For product and process qualification, error may be appraiser,
environment (time), or method
 Commonly referred to as A.V. – Appraiser Variation
 Between-system (conditions) variation
 ASTM E456-96 includes repeatability, laboratory, and
environmental effects as well as appraiser effects
• GRR or Gage R&R
 Gage repeatability and reproducibility: the combined estimate of
measurement system repeatability and reproducibility
 Measurement system capability; depending on the method used,
may or may not include the effects of time
• Measurement System Capability
 Short-term estimate of measurement system variation (e.g.,
“GRR” including graphics)
2
In ASTM documents, there is no such thing as the precision of a measurement system; i.e., the precision cannot
be represented by a single number.
GRR
A C B
Reference Value

8
• Measurement System Performance
 Long-term estimate of measurement system variation (e.g., long-
term Control Chart Method)
• Sensitivity
 Smallest input that results in a detectable output signal
 Responsiveness of the measurement system to changes in
measured feature
 Determined by gage design (discrimination), inherent quality
(Original Equipment Manufacturer), in-service maintenance, and
operating condition of the instrument and standard
• Consistency
 The degree of change of repeatability over time
 A consistent measurement process is in statistical control with
respect to width (variability)
• Uniformity
 The change in repeatability over the normal operating range
 Homogeneity of repeatability
System variation
Measurement system variation can be characterized as:
• Capability
 Variability in readings taken over a short period of time
• Performance
 Variability in readings taken over a long period of time
 Based on total variation
• Uncertainty
 An estimated range of values about the measured value in which
the true value is believed to be contained
All characterizations of the total variation of the measurement system
assume that the system is stable and consistent. For example, the
components of variation can include any combination of the items
shown in I-B 1.
The measurement
system must be stable
and consistent.

9
National
Measurement
Institutes
Traceability
Standards and
Traceability
The National Institute of Standards and Technology (NIST) is the principal
National Measurements Institute (NMI) in the United States serving under
the U.S. Department of Commerce. NIST, formerly the National Bureau of
Standards (NBS), serves as the highest level authority for metrology in the
U.S. NIST’s primary responsibility is to provide measurement services and
maintain measurement standards that assist U.S. industry in making traceable
measurements which ultimately assist in trade of products and services.
NIST provides these services directly to many types of industries, but
primarily to those industries that require the highest level of accuracy for
their products and that incorporate state-of-the-art measurements in their
processes.
Most of the industrialized countries throughout the world maintain their own
NMIs and similar to NIST, they also provide a high level of metrology
standards or measurement services for their respective countries. NIST
works collaboratively with these other NMIs to assure measurements made
in one country do not differ from those made in another. This is
accomplished through Mutual Recognition Arrangements (MRAs) and by
performing interlaboratory comparisons between the NMIs. One thing to
note is that the capabilities of these NMIs will vary from country to country
and not all types of measurements are compared on a regular basis, so
differences can exist. This is why it is important to understand to whom
measurements are traceable and how traceable they are.
Traceability is an important concept in the trade of goods and services.
Measurements that are traceable to the same or similar standards will agree
more closely than those that are not traceable. This helps reduce the need for
re-test, rejection of good product, and acceptance of bad product.
Traceability is defined by the ISO International Vocabulary of Basic and
General Terms in Metrology (VIM) as:
“The property of a measurement or the value of a standard whereby it can be
related to stated references, usually national or international standards,
through an unbroken chain of comparisons all having stated uncertainties.”
The traceability of a measurement will typically be established through a
chain of comparisons back to the NMI. However, in many instances in
industry, the traceability of a measurement may be linked back to an agreed
upon reference value or “consensus standard” between a customer and a
supplier. The traceability linkage of these consensus standards to the NMI
may not always be clearly understood, so ultimately it is critical that the
measurements are traceable to the extent that satisfies customer needs. With
the advancement in measurement technologies and the usage of state-of-the-
art measurement systems in industry, the definition as to where and how a
measurement is traceable is an ever-evolving concept.

10
Figure I-A 1: Example of a Traceability Chain for a Length Measurement
NMIs work closely with various national labs, gage suppliers, state-of-the-art
manufacturing companies, etc. to assure that their reference standards are
properly calibrated and directly traceable to the standards maintained by the
NMI. These government and private industry organizations will then use
their standards to provide calibration and measurement services to their
customers’ metrology or gage laboratories, calibrating working or other
primary standards. This linkage or chain of events ultimately finds its way
onto the factory floor and then provides the basis for measurement
traceability. Measurements that can be connected back to NIST through this
unbroken chain of measurements are said to be traceable to NIST.
Not all organizations have metrology or gage laboratories within their
facilities therefore depend on outside commercial/independent laboratories to
provide traceability calibration and measurement services. This is an
acceptable and appropriate means of attaining traceability to NIST, provided
that the capability of the commercial/independent laboratory can be assured
through processes such as laboratory accreditation.
A calibration system is a set of operations that establish, under specified
conditions, the relationship between a measuring device and a traceable
standard of known reference value and uncertainty. Calibration may also
include steps to detect, correlate, report, or eliminate by adjustment any
discrepancy in accuracy of the measuring device being compared.
The calibration system determines measurement traceability to the
measurement systems through the use of calibration methods and standards.
Traceability is the chain of calibration events originating with the calibration
standards of appropriate metrological capability or measurement uncertainty.
Each calibration event includes all of the elements necessary including
standards, measurement and test equipment being verified, calibration
methods and procedures, records, and qualified personnel.
Wavelength
Standard
Laser
Interferometer
Reference Gage
Blocks/Comparator
CMM Gage Blocks
Micrometers
Interference
Comparator
Fixture Gage
Working Standard
Reference
Standard
National
Standard
Production Gage
Calibration
Systems

11
True Value
An organization may have an internal calibration laboratory or organization
which controls and maintains the elements of the calibration events. These
internal laboratories will maintain a laboratory scope which lists the specific
calibrations they are capable of performing as well as the equipment and
methods/procedures used to perform the calibrations.
The calibration system is part of an organization’s quality management
system and therefore should be included in any internal audit requirements.
Measurement Assurance Programs (MAPs) can be used to verify the
acceptability of the measurement processes used throughout the calibration
system. Generally MAPs will include verification of a measurement system’s
results through a secondary independent measurement of the same feature or
parameter. Independent measurements imply that the traceability of the
secondary measurement process is derived from a separate chain of
calibration events from those used for the initial measurement. MAPs may
also include the use of statistical process control (SPC) to track the long-term
stability of a measurement process.
Note: ANSI/NCSL Z540.3 and ISO 10012 each provide models for many of
the elements of a calibration system.
When the calibration event is performed by an external, commercial, or
independent calibration service supplier, the service supplier’s calibration
system can (or may) be verified through accreditation to ISO/IEC 17025.
When a qualified laboratory is not available for a given piece of equipment,
calibration services may be performed by the equipment manufacturer.
The measurement process TARGET is the “true” value of the part. It is
desired that any individual reading be as close to this value as (economically)
possible. Unfortunately, the true value can never be known with certainty.
However, uncertainty can be minimized by using a reference value based on
a well defined operational definition of the characteristic, and using the
results of a measurement system that has higher order discrimination and
traceable to NIST. Because the reference value is used as a surrogate for the
true value, these terms are commonly used interchangeably. This usage is not
recommended.

12

Chapter I – Section B
The Measurement Process
13
Section B
The Measurement ProcessP
3
Measurement Systems
In order to effectively manage variation of any process, there needs to be
knowledge of:
• What the process should be doing
• What can go wrong
• What the process is doing
Specifications and engineering requirements define what the process should
be doing.
The purpose of a Process Failure Mode Effects AnalysisP
4
F (PFMEA) is to
define the risk associated with potential process failures and to propose
corrective action before these failures can occur. The outcome of the
PFMEA is transferred to the control plan.
Knowledge is gained of what the process is doing by evaluating the
parameters or results of the process. This activity, often called inspection, is
the act of examining process parameters, in-process parts, assembled
subsystems, or complete end products with the aid of suitable standards and
measuring devices which enable the observer to confirm or deny the premise
that the process is operating in a stable manner with acceptable variation to a
customer designated target. But this examination activity is itself a process.
General Process
Measurement Process
Unfortunately, industry has traditionally viewed the measurement and
analysis activity as a “black box”. Equipment was the major focus – the
more "important" the characteristic, the more expensive the gage. The
3
Portions of this chapter adapted with permission from Measurement Systems Analysis - A Tutorial by G. F.
Gruska and M. S. Heaphy, The Third Generation, 1987, 1998.
4
See the Potential Failure Mode and Effects Analysis (FMEA) Reference Manual – 4th
Edition.
Operation
Input Output

14
usefulness of the instrument, its compatibility with the process and
environment, and its usability was rarely questioned. Consequently these
gages were often not used properly or simply not used.
The measurement and analysis activity is a process – a measurement process.
Any and all of the management, statistical, and logical techniques of process
control can be applied to it.
This means that the customers and their needs must first be identified. The
customer, the owner of the process, wants to make a correct decision with
minimum effort. Management must provide the resources to purchase
equipment which is necessary and sufficient to do this. But purchasing the
best or the latest measurement technology will not necessarily guarantee
correct production process control decisions.
Equipment is only one part of the measurement process. The owner of the
process must know how to correctly use this equipment and how to analyze
and interpret the results. Management must therefore also provide clear
operational definitions and standards as well as training and support. The
owner of the process has, in turn, the obligation to monitor and control the
measurement process to assure stable and correct results which includes a
total measurement systems analysis perspective – the study of the gage,
procedure, user, and environment; i.e., normal operating conditions.
An ideal measurement system would produce only “correct” measurements
each time it is used. Each measurement would always agree with a standard.P
5
F
A measurement system that could produce measurements like that would be
said to have the statistical properties of zero variance, zero bias, and zero
probability of misclassifying any product it measured. Unfortunately,
measurement systems with such desirable statistical properties seldom exist,
and so process managers are typically forced to use measurement systems
that have less desirable statistical properties. The quality of a measurement
system is usually determined solely by the statistical properties of the data it
produces over time. Other properties, such as cost, ease of use, etc., are also
important in that they contribute to the overall desirability of a measurement
system. But it is the statistical properties of the data produced that determine
the quality of the measurement system.
Statistical properties that are most important for one use are not necessarily
the most important properties for another use. For instance, for some uses of
a coordinate measuring machine (CMM), the most important statistical
properties are “small” bias and variance. A CMM with those properties will
generate measurements that are “close” to the certified values of standards
that are traceable. Data obtained from such a machine can be very useful for
analyzing a manufacturing process. But, no matter how “small” the bias and
variance of the CMM may be, the measurement system which uses the CMM
may be unable to do an acceptable job of discriminating between good and
bad product because of the additional sources of variation introduced by the
other elements of the measurement system.
5
For a fuller discussion on the matter of standards see Out of the Crisis, W. Edwards Deming, 1982, 1986, p.
279-281.
Statistical
Properties of
Measurement
Systems

15
Management has the responsibility for identifying the statistical properties
that are the most important for the ultimate use of the data. Management is
also responsible for ensuring that those properties are used as the basis for
selecting a measurement system. To accomplish this, operational definitions
of the statistical properties, as well as acceptable methods of measuring them,
are required. Although each measurement system may be required to have
different statistical properties, there are certain fundamental properties
that define a “good” measurement system. These include:
1) Adequate discrimination and sensitivity. The increments of measure
should be small relative to the process variation or specification
limits for the purpose of measurement. The commonly known Rule
of Tens, or 10-to-1 Rule, states that instrument discrimination should
divide the tolerance (or process variation) into ten parts or more.
This rule of thumb was intended as a practical minimum starting
point for gage selection.
2) The measurement system ought to be in statistical control.FP6PF
This means that under repeatable conditions, the variation in the
measurement system is due to common causes only and not due to
special causes. This can be referred to as statistical stability and is
best evaluated by graphical methods.
3) For product control, variability of the measurement system must be
small compared to the specification limits. Assess the measurement
system to the feature tolerance.
4) For process control, the variability of the measurement system ought
to demonstrate effective resolution and be small compared to
manufacturing process variation. Assess the measurement system to
the 6-sigma process variation and/or Total Variation from the MSA
study.
The statistical properties of the measurement system may change as
the items being measured vary. If so, then the largest (worst) variation
of the measurement system is small relative to the smaller of either the
process variation or the specification limits.
Similar to all processes, the measurement system is impacted by both random
and systematic sources of variation. These sources of variation are due to
common and special causes. In order to control the measurement system
variation:
1) Identify the potential sources of variation.
2) Eliminate (whenever possible) or monitor these sources of variation.
Although the specific causes will depend on the situation, some typical
sources of variation can be identified. There are various methods of
6
The measurement analyst must always consider practical and statistical significance.
Sources of
Variation

16
presenting and categorizing these sources of variation such as cause-effect
diagrams, fault tree diagrams, etc., but the guidelines presented here will
focus on the major elements of a measuring system.
The acronym S.W.I.P.E.P
7
F is used to represent the six essential elements of a
generalized measuring system to assure attainment of required objectives.
S.W.I.P.E. stands for Standard, Workpiece, Instrument, Person and
Procedure, and Environment. This may be thought of as an error model for a
complete measurement system.P
8
Factors affecting those six areas need to be understood so they can be
controlled or eliminated.
Figure I-B 1 displays a cause and effect diagram showing some of the
potential sources of variation. Since the actual sources of variation affecting
a specific measurement system will be unique to that system, this figure is
presented as a thought starter for developing a measurement system’s
sources of variation.
7
T
This acronym was originally developed by Ms. Mary Hoskins, a metrologist associated with Honeywell, Eli
Whitney Metrology Lab and the Bendix Corporation.T
8
See Appendix F for an alternate error model, P.I.S.M.O.E.A.
S Standard
W Workpiece (i.e., part)
I Instrument
P Person / Procedure
E Environment

Chapter
I
–
Section
B
The
Measurement
Process
17
Figure
I-B
1:
Measurement
System
Variability
Cause
and
Effect
Diagram
Measurement System
Variability
Standard
Workpiece
(Part)
Instrument
(Gage)
Environment Person
(Appraiser)
geometric
compatibility
coef of thermal
expansion
elastic properties
calibration
stability
elastic
deformation
supporting
features
elastic
properties
mass
cleanliness
interrelated
characteristics
hidden
geometry
operational
definition
adequate
datums
skill
limitations
experience
training
understanding
training
experience
attitude
physical
educational
vibration
temperature
standard
vs ambient
equalization --
system components
traceability
air pollution
ergonomics
lighting
stress
cycles
thermal
expansion
sun
components
air drafts
people
lights
artificial
design amplification
contact
geometry
deformation
effects
build
maintenance
bias
variability
stability
linearity
repeatability
reproducibility
sensitivity
consistency
uniformity
calibration
p.m.
procedures
visual
standards
operational
definition
build
variation
build
tolerances
design validation
- clamping
- locators
- measurement points
- measurement probes
robustness
use
assumptions

18
The Effects of Measurement System Variability
Because the measurement system can be affected by various sources of
variation, repeated readings on the same part do not yield the same, identical
result. Readings vary from each other due to common and special causes.
The effects of the various sources of variation on the measurement system
should be evaluated over a short and long period of time. The measurement
system capability is the measurement system (random) error over a short
period of time. It is the combination of errors quantified by linearity,
uniformity, repeatability and reproducibility. The measurement system
performance, as with process performance, is the effect of all sources of
variation over time. This is accomplished by determining whether our
process is in statistical control (i.e., stable and consistent; variation is due
only to common causes), on target (no bias), and has acceptable variation
(gage repeatability and reproducibility (GRR)) over the range of expected
results. This adds stability and consistency to the measurement system
capability.
Because the output of the measurement system is used in making a
decision about the product and the process, the cumulative effect of all
the sources of variation is often called measurement system error, or
sometimes just “error.”
After measuring a part, one of the actions that can be taken is to determine
the status of that part. Historically, it would be determined if the part were
acceptable (within specification) or unacceptable (outside specification).
Another common scenario is the classification of parts into specific
categories (e.g., piston sizes).
For the rest of the discussion, as an example, the two category
situation will be used: out of specification (“bad”) and in specification
(“good”). This does not restrict the application of the discussion to other
categorization activities.
Further classifications may be reworkable, salvageable or scrap. Under a
product control philosophy this classification activity would be the primary
reason for measuring a part. But, with a process control philosophy, interest
is focused on whether the part variation is due to common causes or special
causes in the process.
Philosophy Interest
Product control Is the part in a specific category?
Process control Is the process variation stable and acceptable?
Table I-B1: Control Philosophy and Driving Interest
Effect on
Decisions

19
The next section deals with the effect of the measurement error on the
product decision. Following that is a section which addresses its impact on
the process decision.
In order to better understand the effect of measurement system error on
product decisions, consider the case where all of the variability in multiple
readings of a single part is due to the gage repeatability and reproducibility.
That is, the measurement process is in statistical control and has zero bias.
A wrong decision will sometimes be made whenever any part of the above
measurement distribution overlaps a specification limit. For example, a good
part will sometimes be called “bad” (type I error, producer's risk or false
alarm) if:
And, a bad part will sometimes be called “good” (type II error, consumer’s
risk or miss rate) if:
NOTE: False Alarm Rate + Miss Rate = Error Rate.
RISK is the chance of making a decision which will be
detrimental to an individual or process
That is, with respect to the specification limits, the potential to make the
wrong decision about the part exists only when the measurement system
error intersects the specification limits. This gives three distinct areas:
Effect on
Product
Decisions

20
where:
I Bad parts will always be called bad
II Potential wrong decision can be made
III Good parts will always be called good
Since the goal is to maximize CORRECT decisions regarding product status,
there are two choices:
1) Improve the production process: reduce the variability of the
process so that no parts will be produced in the II or “shaded”
areas of the graphic above.
2) Improve the measurement system: reduce the measurement
system error to reduce the size of the II areas so that all parts
being produced will fall within area III and thus minimize the
risk of making a wrong decision.
This discussion assumes that the measurement process is in statistical control
and on target. If either of these assumptions is violated then there is little
confidence that any observed value would lead to a correct decision.
With process control, the following needs to be established:
• Statistical control
• On target
• Acceptable variability
As explained in the previous section, the measurement error can cause
incorrect decisions about the product. The impact on process decisions would
be as follows:
• Calling a common cause a special cause
• Calling a special cause a common cause
Measurement system variability can affect the decision regarding the
stability, target and variation of a process. The basic relationship between the
actual and the observed process variation is:
Effect on
Process
Decisions

21
2 2 2
obs actual msa
σ σ σ
= 
where
2
obs
σ = observed process variance
2
actual
σ = actual process variance
2
msa
σ = variance of the measurement system
The capability indexP
9
F Cp is defined as
6
ToleranceRange
Cp
σ
=
The relationship between the Cp index of the observed process and the
Cp indices of the actual process and the measurement system is
derived by substituting the equation for Cp into the observed variance
equation above:
     
2 2 2
obs actual msa
Cp Cp Cp
− − −
= 
Assuming the measurement system is in statistical control and on target, the
actual process Cp can be compared graphically to the observed Cp.P
10
F
Therefore the observed process capability is a combination of the actual
process capability plus the variation due to the measurement process. To
reach a specific process capability goal would require factoring in the
measurement variation.
For example, if the measurement system Cp index were 2, the actual process
would require a Cp index greater than or equal to 1.79 in order for the
calculated (observed) index to be 1.33. If the measurement system Cp index
were itself 1.33, the process would require no variation at all if the final
result were to be 1.33 – clearly an impossible situation.
9
Although this discussion is using Cp, the results hold also for the performance index Pp.
10
See Appendix B for formulas and graphs.

22
When a new process such as machining, manufacturing, stamping, material
handling, heat treating, or assembly is purchased, there often is a series of
steps that are completed as part of the buy-off activity. Oftentimes this
involves some studies done on the equipment at the supplier's location and
then at the customer's location.
If the measurement system used at either location is not consistent with the
measurement system that will be used under normal circumstances then
confusion may ensue. The most common situation involving the use of
different instruments is the case where the instrument used at the supplier has
higher order discrimination than the production instrument (gage). For
example, parts measured with a coordinate measuring machine during buy-
off and then with a height gage during production; samples measured
(weighed) on an electronic scale or laboratory mechanical scale during buy-
off and then on a simple mechanical scale during production.
In the case where the (higher order) measurement system used during buy-off
has a GRR of 10% and the actual process Cp is 2.0 the observed process Cp
during buy-off will be 1.96.P
11
F
When this process is studied in production with the production gage, more
variation (i.e., a smaller Cp) will be observed. For example, if the GRR of the
production gage is 30% and the actual process Cp is still 2.0 then the
observed process Cp will be 1.71.
A worst case scenario would be if a production gage has not been qualified
but is used. If the measurement system GRR is actually 60% (but that fact is
not known), then the observed Cp would be 1.28. The difference in the
observed Cp of 1.96 versus 1.28 is due to the different measurement system.
Without this knowledge, efforts may be spent, in vain, looking to see what
went wrong with the new process.
11
For this discussion, assume there is no sampling variation. In reality 1.96 will be the expected value but actual
results will vary around it.
Runoff part variation
CMM variation
New Process
Acceptance

23
Often manufacturing operations use a single part at the beginning of the day
to verify that the process is targeted. If the part measured is off target, the
process is then adjusted. Later, in some cases another part is measured and
again the process may be adjusted. Dr. Deming referred to this type of
measurement and decision-making as tampering.
Consider a situation where the weight of a precious metal coating on a part is
being controlled to a target of 5.00 grams. Suppose that the results from the
scale used to determine the weight vary 0.20 grams but this is not known
since the measurement system analysis was never done. The operating
instructions require the operator to verify the weight at setup and every hour
based on one sample. If the results are beyond the interval 4.90 to 5.10 grams
then the operator is to setup the process again.
At setup, suppose the process is operating at 4.95 grams but due to
measurement error the operator observes 4.85 grams. According to
instructions the operator attempts to adjust the process up by .15 grams. Now
the process is running at 5.10 grams for a target. When the operator checks
the setup this time, 5.08 grams is observed so the process is allowed to run.
Over-adjustment of the process has added variation and will continue to do
so.
This is one example of the funnel experiment that Dr. Deming used to
describe the effects of tampering.P
12
F The measurement error just compounds
the problem.
Four rules of the funnel experiment are:
Rule 1: Make no adjustment or take no action unless the process is
unstable.
Rule 2: Adjust the process in an equal amount and in an opposite
direction from where the process was last measured to be.
Rule 3: Reset the process to the target. Then adjust the process in an
equal amount and in an opposite direction from the target.
Rule 4: Adjust the process to the point of the last measurement.
The setup instruction for the precious metal process is an example of Rule 3.
Rules 2, 3 and 4 add progressively more variation. Rule 1 is the best choice
to produce minimum variation.
12
Deming, W. Edwards, Out of the Crisis, Massachusetts Institute of Technology, 1982, 1986.
Actual process variation
Production gage variation
Observed process variation
Process Setup/
Control (Funnel
Experiment)

24
Other examples of the funnel experiment are:
• Recalibration of gages based on arbitrary limits – i.e., limits not
reflecting the measurement system’s variability. (Rule 3)
• (Re)mastering the process control measurement system after an
arbitrary number of uses without any indication or history of a
change (special cause). (Rule 3)
• Autocompensation adjusts the process based on the last part
produced. (Rule 2)
• On the job training (OJT) where worker A trains worker B who later
trains worker C... without standard training material. Similar to the
“post office” game. (Rule 4)
• Parts are measured, found to be off target, but when plotted on a
control chart the process is shown to be stable – therefore, no action
is taken. (Rule 1)

Chapter I – Section C
Measurement Strategy and Planning
25
Section C
Planning is key before designing and purchase of measurement equipment or
systems. Many decisions made during the planning stage could affect the
direction and selection of measurement equipment. What is the purpose and
how will the measurement result be used? The planning stage will set the
course and have a significant effect on how well the measurement process
operates and can reduce possible problems and measurement error in the
future.
In some cases due to the risk involved in the component being
measured or because of the cost and complexity of the measurement
device, the OEM customer may use the APQP process and committee
to decide on the measurement strategy at the supplier.
Not all product and process characteristics require measurement systems
whose development falls under this type of scrutiny. Simple standard
measurement tools like micrometers or calipers may not require this in-depth
strategy and planning. A basic rule of thumb is whether the characteristic
being measured on the component or sub-system has been identified in the
control plan or is important in determining the acceptance of the product or
process. Another guide would be the level of tolerance assigned to a specific
dimension. Common sense is the guide in any case.
The type, complexity, and purpose of a measurement system may drive
various levels of program management, strategic planning, measurement
systems analysis, or other special consideration for measurement selection,
assessment and control. Simple measuring tools and devices (i.e., scales,
measuring tapes, fixed-limit or attribute gages) may not require the level of
management, planning, or analysis that more complex or critical measuring
systems demand (i.e., master or reference, CMM, test stand, automated on-
line gaging, etc.). Any measurement system may require more or less
strategic planning and scrutiny depending on a given product or process
situation. The decision as to the appropriate level shall be left to the APQP
team assigned to the measurement process and customer. The actual degree
of involvement or implementation in many of the activities below should be
driven by the particular measurement system, consideration of the supporting
gage control and calibration system, profound process knowledge, and
common sense.
Complexity

26
The first step is to establish the purpose for the measurement and how the
measurement will be utilized. A cross-functional team organized early in the
development of the measurement process is critical in accomplishing this
task. Specific considerations are made in relation to audit, process control,
product and process development and analysis of the “Measurement Life
Cycle”.
The Measurement Life Cycle concept expresses the belief that the
measurement methods may change over time as one learns and improves the
process. For example, measurement may start on a product characteristic to
establish stability and capability of the process. This may lead to an
understanding of critical process control characteristics that directly affect
the part characteristics. Dependency on part characteristic information
becomes less and the sampling plan may be reduced to signify this
understanding (five parts per hour sample reduced to one part per shift).
Also, the method of measurement may change from a CMM measurement, to
some form of attribute gaging. Eventually it may be found that very little part
monitoring may be required as long as the process is maintained or
measuring and monitoring the maintenance and tooling may be all that is
needed. The level of measurement follows the level of process
understanding.
Most of the measuring and monitoring could eventually end up at suppliers
of incoming material. The same measurement, on the same characteristic, at
the same area of the process, over an extensive period of time is evidence of
a lack of learning or a stagnant measurement process.
Before a measurement system can be purchased, a detailed engineering
concept of the measurement process is developed. Using the purpose
developed above, a cross-functional team of individuals will develop a plan
and concept for the measurement system required by the design. Here are
some guidelines:
The team needs to evaluate the design of the subsystem or component and
identify important characteristics. These are based on customer requirements
and the functionality of the subsystem or component to the total system. If
the important dimensions have been identified already, evaluate the ability to
measure the characteristics. For example, if the important characteristic of a
plastic injection molded component was on the mold parting line, the
dimensional check would be difficult and measurement variation would be
high.
One method to capture issues similar to these would be to use a FMEA
process to analyze areas of risk in gage design both from an ability to
measure to the part to the functionality gage (Design and Process FMEA).
This would aid in the development of the maintenance and calibration plan.
Develop a flow chart showing critical process steps in the manufacturing or
assembly of the part or subsystem. Identify key inputs and outputs to each
step in the process. This will aid in the development of the measurement
equipment criteria and requirements affected by the location in the process.
Identify the
Purpose of the
Measurement
Process
Criteria for a
Measurement
Process Design
Selection
Measurement Life
Cycle

27
A measurement plan, a list of measurement types, comes out of this
investigation.P
13
For complex measurement systems, a flow chart is made of the measurement
process. This would include delivery of the part or sub-system being
measured, the measurement itself, and the return of the part or sub-system to
the process.
Next use some method of brainstorming with the group to develop general
criteria for each measurement required. One of the simple methods to use is a
cause and effect diagram.P
14
F See the example in Figure I-B 1 as a thought
starter.
A few additional questions to consider in relation
to measurement planning:
• Who ought to be involved in the “needs” analysis? The flow chart
and initial discussion will facilitate the identification of the key
individuals.
• Why will measurement be taken and how will it be used? Will the
data be used for control, sorting, qualification, etc? The way the
measurement will be used can change the sensitivity level of the
measurement system.
• What level of sensitivity will be required? What is the product
specification? What is the expected process variability? How much
of a difference between parts will the gage need to detect?
• What type of information will be provided with the gage (e.g.,
manuals – operating, maintenance, etc.) and what basic operator
skills are required? Who will do the training?
• How are measurements taken? Will it be done manually, on a
moving conveyor, off-line, automatically, etc? Are the part location
and fixturing possible sources of variation? Contact or non-contact?
• How will the measurement be calibrated and will it be compared
with other measurement processes? Who will be responsible for the
calibration masters?
• When and where will the measurement be taken? Will the part be
clean, oily, hot, etc.?
Remember to use data to substantiate common assumptions about the
measurement process. It is better to be safe and collect data on the
environment, rather than to make decisions based on the wrong
information and having a system developed that is not robust to
environmental issues.
13
This can be considered as a preliminary control plan.
14
See Guide to Quality Control, Kaoru Ishikawa, published by Asian Productivity Organization, 1986.

28
Current measurement methods should be researched prior to investing in new
equipment. Proven measurement methods may provide more reliable
operation. Where possible, use measurement equipment that has a proven
track record.
Refer to “Suggested Elements for a Measurement System Development
Checklist” at the end of Chapter I, Section D, when developing and
designing concepts and proposals.
During and after the fabrication of the measurement equipment and
development of the measurement process (methods, training, documentation,
etc.), experimental studies and data collection activities will be performed.
These studies and data will be used to understand this measurement process
so that this process and future processes may be improved.
Research Various
Measurement
Process Methods
Develop and Design
Concepts and
Proposals

Chapter I – Section D
Measurement Source Development
29
Section D
This section addresses the quotation/procurement timeframe of the life of a
measurement process. It has been constructed to be a self-contained
discussion about the process of developing a measurement process quotation
package, obtaining responses to that package, awarding the project,
completing final design, developing the measurement process, and, finally,
marrying that measurement process to the production process for which it
was created. It is strongly encouraged that this chapter not be used without
reading and understanding the entire discussion about a measurement
process. To obtain the most benefit from the measurement process, study
and address it as a process with inputs and outputs.P
15
This chapter was written with the team philosophy in mind. It is not a job
description for the buyer or purchasing agent. The activities described here
will require team involvement to be completed successfully and it should be
administered within the overall framework of an Advanced Product Quality
Planning (APQP) team. This can result in healthy interplay between various
team functions – concepts arising out of the planning process may be
modified before the gage supplier arrives at a final design that satisfies the
measurement system requirements.
Generally, the “acquisition process” begins with formal communication
between the customer and supplier for a given project. Up-front
communication is crucial to the success of the project, since the groundwork
necessary for an effective future customer/supplier relationship will be done
at this stage. The acquisition process begins with the customer’s formal
presentation of the intent of the project in the form of a Request For Quote
(RFQ) followed by the supplier’s formal explanation of their proposal to
meet this intent (the Quotation). The customer and supplier(s) need to
thoroughly understand the project requirements, what the deliverables will be
and the methods by which both are to be achieved. This understanding is
derived from accurate timely communication between the two parties.
Once a concept has been agreed upon and a customer/supplier relationship
has been established for the project at hand, the detailed design, fabrication
of the measurement process, and development activities can commence.
Communication between the customer and the supplier at this time is
especially important. Since there may be several levels of concept approvals
to be carried out, and possible environmental changes and the potential of
team members changing, the measurement process project could falter or
even fail. This risk will be reduced if frequent, detailed communication is
maintained and documented between the customer and supplier and formal
responsibility (an individual) for maintaining communication is designated
15
See Chapter I, Section B
OUTPUT
INPUT
MEASUREMENT
PROCESS

30
Datum Coordination
by both parties. The ideal forum and format for this activity is the Advanced
Product Quality Planning (APQP) process.
After the measurement process has been conceptually designed, the activities
surrounding the acquisition of the process/system can begin.
Ideally, with the current prevalence in the use of Geometric Dimensioning &
Tolerancing (GD&T), datums need to be coordinated (i.e., made identical)
throughout the manufacturing process and the measurement system and this
needs to be established very early in the APQP process. Initial responsibility
for this may lie with the product design engineer, dimensional control, etc.
depending on the specific organization. When datum schemes do not match
throughout a manufacturing process, particularly in the measurement
systems, this leads to a situation where the wrong things may be measured,
and there may be fit problems, etc., leading to ineffective control of the
manufacturing process.
There may be times when a datum scheme used in a final assembly cannot
possibly match that used in a sub-component manufacturing process. When
such is the case, it can be established as early as possible in the APQP
process so that all team members understand possible difficulties and
conflicts that may lie ahead and have every opportunity to do something
about it. During this process, different datum schemes may need to be
explored in order to understand the impact of these differences.
Certain commodities present features which can yield more problems than
others, such as camshaft centering, or other round, cylindrical or tubular
characteristics. For example, a camshaft must be manufactured on centers
but the important product features are in its lobes. One method or datum
scheme may be required for manufacturing whereas another scheme is
required for measurement of the final product measurement.

31
Detailed Engineering
Concept
Prerequisites and
Assumptions
Before discussing the development of a gage supplier, it will be assumed that
issues such as “correct” engineering product design (GD&T) and “correct”
process design (one which allows for measurement at the proper time and
location in the process) have been resolved. However this should not detract
from consideration of these issues with appropriate team members early in
the APQP process.
It is assumed that the gage supplier will be involved with the APQP process,
a team approach. The gage supplier will develop a clear appreciation of the
overall production process and product usage so that his role is understood
not only by him but by others on the team (manufacturing, quality,
engineering, etc.).
There may be slight overlap in some activities or the order of those activities
depending on the particular program/project or other constraints. For
instance, the APQP team without much input from a gage source may
develop certain gage concepts. Other concepts may require the expertise of
the gage source. This may be driven by the complexity of the measurement
system and a team decision as to what makes sense.
Gage Source Selection Process
Develop the Quotation Package
Before a measurement process request for quotation package can be supplied
to a potential supplier for formal proposals, a detailed engineering concept of
the measurement process needs to be developed. The team of individuals
that will employ and be responsible for the maintenance and continual
improvement of the measurement process have direct responsibility for
developing the detailed concept. This can be part of the APQP team. To
better develop this concept, several questions need to be answered.
The team may research various issues to help decide which direction or path
will be followed for designing the measurement process. Some may be
dictated or heavily implied by the product design. Examples of the multitude
of possible issues that need to be addressed by the team when developing this
detailed concept may be found in the “Suggested Elements for a
Measurement System Development Checklist” at the end of this section.

32
Preventive
Maintenance
Considerations
All too often, customers rely too heavily on suppliers for solutions.
Before a customer asks a supplier to suggest solutions to process
problems, the foundation and intent of the process needs to be
thoroughly understood and anticipated by the team that owns that
process. Then and only then will the process be properly used,
supported and improved upon.
What activities should be scheduled for preventive maintenance (e.g.,
lubrication, vibration analysis, probe integrity, parts replacement, etc.)?
Much of these activities will depend on the complexity of the measurement
system, device or apparatus. Simpler gages may require only an inspection
at regular intervals, whereas more complex systems may require ongoing
detailed statistical analyses and a team of engineers to maintain in a
predictive fashion.
Planning preventive maintenance activities should coincide with the
initiation of the measurement process planning. Many activities, such as
draining air filters daily, lubricating bearings after the designated number of
operating hours, etc., can be planned before the measurement system is
completely built, developed and implemented. In fact this is preferable and
improves advanced measurement planning and costs. Data collection
methods and maintenance recommendations related to these activities can be
obtained from the original manufacturer, or developed by plant engineering,
manufacturing and quality personnel. After the measurement process is
implemented and in use, data pertaining to the function of the measurement
process need to be collected and plotted over time. Simple analytical
methods (run charts, trend analysis) can be conducted to determine the
stability of the system. Eventually, as the judgment of system stability
dictates, preventive maintenance routines can be scheduled accordingly.
Conducting preventive maintenance on a stable system, based on time series
information, will be less wasteful than conducting preventive maintenance on
a system with traditional techniques.
Specifications serve as guidelines for both the customer and supplier in the
design and build process. These guidelines serve to communicate acceptable
standards. Acceptable standards may be considered in two categories:
• Design Standards
• Build Standards
Format of the design standards may be different depending on who is paying
for the project. Cost issues may affect the format. Generally, it is a good
idea to have sufficient documented design detail that the design may be built
or repaired to original intent by any qualified builder – however, this decision
Specifications

33
Q
U
O
T
E
C
O
N
C
E
P
T
APPROVAL
Evaluate the
Quotations
may be driven by cost and criticality. The required format of the final design
may be some form of computer assisted design (CAD) or hardcopy
engineering drawings. It may involve engineering standards chosen from
those of the OEM, SAE, ASTM, or other organization, and the gage supplier
must have access to the latest level and understand these standards. The
OEM may require the use of particular standards at either the design or build
phase and may even require formal approvals before the measurement
system may be released for use.
Design standards will detail the method of communicating the design (CAD
– e.g., CATIA, Unigraphics, IGES, manual hardcopy, etc.) to the builder. It
may also cover performance standards for a more complex measurement
system.
Build standards will cover the tolerances to which the measurement system
must be built. Build tolerance should be based on a combination of the
capabilities of the process used to produce the gage or gage component, and
the criticality of the intended measurement. Build tolerance should not be a
mere given percent of product tolerance alone.
If duplicate fixtures or systems are required, proper planning and
standardizing can lead to interchangeability and flexibility.
Use of standard(ized) components or subassemblies also leads to
interchangeability, flexibility, reduced cost and, generally, less long-term
measurement error.
As quotations are received, the team ought to assemble to review and
evaluate them. Certain items can be noted:
 Are the basic requirements met?
 Are there any outstanding concerns?
 Do any of the suppliers exhibit an exceptional condition and why?
(An exceptional condition could be a significant disparity with
regard to price or delivery – this would not necessarily be
discounted as a negative factor – one supplier may have discovered
an item that others overlooked.)
 Do the concepts promote simplicity and maintainability?
Documentation is sometimes overlooked when acquiring a measurement
process. The significance that documentation takes with any successful
project is often misunderstood. The usual strategy behind documentation is
to provide an original set of mechanical and electrical designs (CAD or
hardcopy drawings) for the measurement process hardware at the time of
delivery. This may satisfy initial implementation requirements, but this
documentation does nothing with regard to defining potential wear points,
suggesting possible trouble areas or describing how to use the process. Thus,

34
the required documentation for any process ought to include more than
assembly and detailed drawings of the measurement equipment.
Effective documentation for any system serves the same purpose as a good
map on a trip. For example, it will suggest to the user how to get from one
point to another (user instructions or gage instructions). It provides the user
with possible alternative routes to reach the desired destinations
(troubleshooting guides or diagnostic trees) if the main route is blocked or
closed.
A complete documentation package may include:
• Reproducible set of assembly and detailed mechanical drawings
(CAD or hardcopy) (including any required masters)
• Reproducible set of electrical hard-wiring, logic and software
• Suggested spare parts list of heavy use or wear items/details.
This list should include items that may require considerable lead-
time to acquire
• Maintenance manuals with machine drawing cutaways and steps
to properly assemble and disassemble machine components
• Manuals defining utility requirements for setup and operation
and machine transport requirements (e.g., load bearing members)
• Diagnostic trees and a troubleshooting guide
• Certification reports (traceable to NIST where applicable)
• Calibration instructions
• User manuals that can be used by the technical support
personnel, the system operator and maintenance personnel
The above list can be used as a checklist when organizing the quotation
package; however it is not necessarily all-inclusive.
The central theme here is communication. Since documentation is a form of
communication, the team and others ought to be involved at every level of
the development of the measurement process documentation package.

35
Shipment
Qualification
at the Supplier
The gage or measurement system should be given a full dimensional layout
and functional test, where applicable, at the measurement system supplier
before shipment. Obviously, the chosen supplier must have qualified
measurement equipment and personnel on site in order to accomplish this. If
not, pre-arrangements should have been made to have this work done at an
outside independent qualified laboratory. Results of such dimensional layout
and/or testing should be done in accordance with customer design and build
standards and be fully documented and available for customer review.
After successful dimensional layout, the supplier should perform a
preliminary but formal measurement systems analysis. This again pre-
requires that the supplier have the personnel, knowledge and experience to
accomplish the appropriate analysis. The customer should predetermine with
the supplier (and perhaps the OEM) exactly what sort of analysis is required
at this point and should be aware of any guidance the supplier might need.
Some issues that may need discussion, negotiation or common agreement
are:
• Objective of the preliminary MSA study :
 Gage repeatability (GRP
16
F) versus gage repeatability and
reproducibility (GRR)
 Assessment of bias and/or linearity
 Assessment of the customer purpose for measurement
• Quantity of pieces, trials and operators in study
 Acceptance criteria
• Use of supplier personnel vs. customer supplied personnel
• Necessary training for personnel
 Are they qualified?
 Do they understand intent?
 What software might be used?Whatever results are
achieved at this point in time, it should be realized that
these are merely preliminary and judgment may be
needed as to the acceptability of the results.
CHECKLIST
• When should the equipment be shipped?
• How should it be shipped?
• Who removes equipment from the truck or rail car?
• Is insurance required?
• Should documentation be shipped with the hardware?
• Does the customer have the proper equipment to unload the
hardware?
• Where will the system be stored until shipment?
16
See Appendix D.

36
Documentation
Delivery
Qualification at
the Customer
• Where will the system be stored until implementation?
• Is the shipping documentation complete and easily
understandable for the loader, transporter, unloader and
installation crew?
Generally, what was done to qualify the measurement system above at the
supplier before shipment should be repeated in some manner at the customer
once delivery is completed. Since this becomes the first real opportunity to
study the measurement system in its intended environment, acceptance
standards and analysis methods used here should be considered seriously.
Attention to detail on the part of all parties involved is paramount to the
eventual success of this measurement system and the use of the data it
generates.
Before any measurement analysis is begun after receipt, the measurement
system should undergo a full dimensional layout to confirm it meets build
requirements/standards. The extent of this layout may be balanced against
layout work done previously at the measurement system supplier before
shipment and confidence in the quality of the layout results done at the
supplier as well as the lack of potential shipping damage. When comparing
results before and after shipment, be aware that there will likely be some
differences in these measurements because of differences in these
measurement systems.
The information that is required, at a minimum, to aid implementation and
startup of any system is the following: (This information ought to be
delivered to the customer prior to delivery.)
• CAD or hardcopy drawings, if required by team
• Process flow diagram of the system, where applicable
• User manuals
 Maintenance/service manual
 Spare parts list
 Troubleshooting guide
• Calibration instructions
• Any special considerations
At the outset, the delivered documentation needs to be noted as preliminary.
Original or reproducible documentation does not need to be delivered at this
time because potential revision may be necessary after implementation. In
fact, it is a wise idea to not have the original documentation package
delivered until after the entire system is implemented – suppliers are
generally more efficient with updating documentation than customers.

37
Suggested Elements for a Measurement System
Development Checklist
This list should be modified based on the situation and type of measurement system.
The development of the final checklist should be the result of collaboration between the
customer and supplier.
Measurement System Design and Development Issues:
 What is to be measured? What type of characteristic is it? Is it a mechanical property? Is it dynamic or
stationary? Is it an electrical property? Is there significant within-part variation?
 For what purpose will the results (output) of the measurement process be used? Production improvement,
production monitoring, laboratory studies, process audits, shipping inspection, receiving inspection, responses
to a D.O.E.?
 Who will use the process? Operators, engineers, technicians, inspectors, auditors?
 Training required: Operator, maintenance personnel, engineers; classroom, practical application, OJT,
apprenticeship period.
 Have the sources of variation been identified? Build an error model (S.W.I.P.E. or P.I.S.M.O.E.A.) using
teams, brainstorming, profound process knowledge, cause & effect diagram or matrix.
 Has a FMEA been developed for the measurement system?
 Flexible vs. dedicated measurement systems: Measurement systems can either be permanent and dedicated
or they can be flexible and have the ability to measure different types of parts; e.g., doghouse gages, fixture
gaging, coordinate measurement machine, etc. Flexible gaging will be more expensive, but can save money in
the long run.
 Contact vs. non-contact: Reliability, type of feature, sample plan, cost, maintenance, calibration, personnel
skill required, compatibility, environment, pace, probe types, part deflection, image processing . This may be
determined by the control plan requirements and the frequency of the measurement ( Full contact gaging may
get excessive wear during continuous sampling). Full surface contact probes, probe type, air feedback jets,
image processing, CMM vs. optical comparator, etc.
 Environment: Dirt, moisture, humidity, temperature, vibration, noise, electro-magnetic interference (EMI),
ambient air movement, air contaminants, etc. Laboratory, shop floor, office, etc? Environment becomes a key
issue with low, tight tolerances in the micron level. Also, in cases that CMM, vision systems, ultrasonic, etc.
This could be a factor in auto-feedback in-process type measurements. Cutting oils, cutting debris, and extreme
temperatures could also become issues. Is a clean room required?
 Measurement and location points: Clearly define, using GD&T, the location of fixturing and clamping points
and where on the part the measurements will be taken.
 Fixturing method: Free state versus clamped part holding.
 Part orientation: Body position versus other.
 Part preparation: Should the part be clean, non-oily, temperature stabilized, etc. before measurement?
 Transducer location: Angular orientation, distance from primary locators or nets.

38
 Correlation issue #1 – duplicate gaging: Are duplicate (or more) gages required within or between plants to
support requirements? Building considerations, measurement error considerations, maintenance considerations.
Which is considered the standard? How will each be qualified?
 Correlations issue #2 – methods divergence: Measurement variation resulting from different measurement
system designs performing on the same product/process within accepted practice and operation limits (e.g.,
CMM versus manual or open-setup measurement results).
 Automated vs. manual: on-line, off-line, operator dependencies.
 Destructive versus nondestructive measurement (NDT): Examples: tensile test, salt spray testing,
plating/paint coating thickness, hardness, dimensional measurement, image processing, chemical analysis,
stress, durability, impact, torsion, torque, weld strength, electrical properties, etc.
 Potential measurement range: size and expected range of conceivable measurements.
 Effective resolution: Is measurement sensitive to physical change (ability to detect process or product
variation) for a particular application acceptable for the application?
 Sensitivity: Is the size of the smallest input signal that results in a detectable (discernable) output signal for this
measurement device acceptable for the application? Sensitivity is determined by inherent gage design and
quality (OEM), in-service maintenance, and operating condition.
Measurement System Build Issues (equipment, standard, instrument):
 Have the sources of variation identified in the system design been addressed? Design review; verify and
validate.
 Calibration and control system: Recommended calibration schedule and audit of equipment and
documentation. Frequency, internal or external, parameters, in-process verification checks.
 Input requirements: Mechanical, electrical, hydraulic, pneumatic, surge suppressors, dryers, filters, setup and
operation issues, isolation, discrimination and sensitivity.
 Output requirements: Analog or digital, documentation and records, file, storage, retrieval, backup.
 Cost: Budget factors for development, purchase, installation, operation and training.
 Preventive maintenance: Type, schedule, cost, personnel, training, documentation.
 Serviceability: Internal and external, location, support level, response time, availability of service parts,
standard parts list.
 Ergonomics: Ability to load and operate the machine without injuries over time. Measurement device
discussions need to focus on issues of how the measurement system is interdependent with the operator.
 Safety considerations: Personnel, operation, environmental, lock-out.
 Storage and location: Establish the requirements around the storage and location of the measurement
equipment. Enclosures, environment, security, availability (proximity) issues.
 Measurement cycle time: How long will it take to measure one part or characteristic? Measurement cycle
integrated to process and product control.
 Will there be any disruption to process flow, lot integrity, to capture, measure and return the part?

39
 Material handling: Are special racks, holding fixtures, transport equipment or other material handling
equipment needed to deal with parts to be measured or the measurement system itself?
 Environmental issues: Are there any special environmental requirements, conditions, limitations, either
affecting this measurement process or neighboring processes? Is special exhausting required? Is temperature or
humidity control necessary? Humidity, vibration, noise, EMI, cleanliness.
 Are there any special reliability requirements or considerations? Will the equipment hold up over time?
Does this need to be verified ahead of production use?
 Spare parts: Common list, adequate supply and ordering system in place, availability, lead-times understood
and accounted for. Is adequate and secure storage available? (bearings, hoses, belts, switches, solenoids, valves,
etc.)
 User instructions: Clamping sequence, cleaning procedures, data interpretation, graphics, visual aids,
comprehensive. Available, appropriately displayed.
 Documentation: Engineering drawings, diagnostic trees, user manuals, language, etc.
 Calibration: Comparison to acceptable standards. Availability and cost of acceptable standards.
Recommended frequency, training requirements. Down-time required?
 Storage: Are there any special requirements or considerations regarding the storage of the measurement
device? Enclosures, environment, security from damage/theft, etc.
 Error/Mistake proofing: Can known measurement procedure mistakes be corrected easily (too easily?) by the
user? Data entry, misuse of equipment, error proofing, mistake proofing.
Measurement System Implementation Issues (process):
 Support: Who will support the measurement process? Lab technicians, engineers, production, maintenance,
outside contracted service?
 Training: What training will be needed for operators/inspectors/technicians/engineers to use and maintain this
measurement process? Timing, resource and cost issues. Who will train? Where will training be held? Lead-
time requirements? Coordinated with actual use of measurement process.
 Data management: How will data output from this measurement process be managed? Manual,
computerized, summary methods, summary frequency, review methods, review frequency, customer
requirements, internal requirements. Availability, storage, retrieval, backup, security. Data interpretation.
 Personnel: Will personnel need to be hired to support this measurement process? Cost, timing, availability
issues. Current or new.
 Improvement methods: Who will improve the measurement process over time? Engineers, production,
maintenance, quality personnel? What evaluation methods will be used? Is there a system to identify needed
improvements?
 Long-term stability: Assessment methods, format, frequency, and need for long-term studies. Drift, wear,
contamination, operational integrity. Can this long-term error be measured, controlled, understood, predicted?
 Special considerations: Inspector attributes, physical limitations or health issues: colorblindness, vision,
strength, fatigue, stamina, ergonomics.

40

Chapter I – Section E
Measurement Issues
41
Section E
Measurement Issues
Three fundamental issues must be addressed when evaluating a measurement
system:
1) The measurement system must demonstrate adequate sensitivity.
 First, does the instrument (and standard) have adequate
discrimination? Discrimination (or class) is fixed by design and
serves as the basic starting point for selecting a measurement
system. Typically, the Rule of Tens has been applied, which
states that instrument discrimination should divide the tolerance
(or process variation) into ten parts or more.
 Second, does the measurement system demonstrate effective
resolution? Related to discrimination, determine if the
measurement system has the sensitivity to detect changes in
product or process variation for the application and conditions.
2) The measurement system must be stable.
 Under repeatability conditions, the measurement system
variation is due to common causes only and not special (chaotic)
causes.
 The measurement analyst must always consider practical and
statistical significance.
3) The statistical properties (errors) are consistent over the expected
range and adequate for the purpose of measurement (product control
or process control).
The long-standing tradition of reporting measurement error only as a percent
of tolerance is inadequate for the challenges of the marketplace that
emphasize strategic and continuous process improvement. As processes
change and improve, a measurement system must be re-evaluated for its
intended purpose. It is essential for the organization (management,
measurement planner, production operator, and quality analyst) to understand
the purpose of measurement and apply the appropriate evaluation.

Measurement Issues
42
It is often assumed that measurements are exact, and frequently the analysis
and conclusions are based upon this assumption. An individual may fail to
realize there is variation in the measurement system which affects the
individual measurements, and subsequently, the decisions based upon the
data. Measurement system error can be classified into five categories: bias,
repeatability, reproducibility, stability and linearity.
One of the objectives of a measurement system study is to obtain information
relative to the amount and types of measurement variation associated with a
measurement system when it interacts with its environment. This information
is valuable, since for the average production process, it is far more practical
to recognize repeatability and calibration bias and establish reasonable limits
for these, than to provide extremely accurate gages with very high
repeatability. Applications of such a study provide the following:
• A criterion to accept new measuring equipment
• A comparison of one measuring device against another
• A basis for evaluating a gage suspected of being deficient
• A comparison for measuring equipment before and after repair
• A required component for calculating process variation, and the
acceptability level for a production process
• Information necessary to develop a Gage Performance Curve
(GPC),P
17
F which indicates the probability of accepting a part of some
true value
The following definitions help describe the types of error or variation
associated with a measurement system, so that each term is clearly
understood for subsequent discussion. An illustration is given for each
definition which graphically displays the meaning of each term.
17
See Chapter V, Section C.
Types of
Measurement
System
Variation

Measurement Issues
43
Operational Definition
“An operational definition is one that people can do business with. An
operational definition of safe, round, reliable, or any other quality
[characteristic] must be communicable, with the same meaning to vendor as
to the purchaser, same meaning yesterday and today to the production
worker. Example:
1) A specific test of a piece of material or an assembly
2) A criterion (or criteria) for judgment
3) Decision: yes or no, the object or the material did or did not
meet the criterion (or criteria)” FP18
Standard
A standard is anything taken by general consent as a basis for comparison; an
accepted model. It can be an artifact or ensemble (instruments, procedures,
etc.) set up and established by an authority as a rule for the measure of
quantity, weight, extent, value or quality.
The concept of ensemble was formalized in ANSI/ASQC Standard M1-
1996.P
19
F This term was used to stress the fact that all of the influences
affecting the measurement uncertainty need to be taken into account; e.g.,
environment, procedures, personnel, etc. “An example of a simple ensemble
would be an ensemble for the calibration of gage blocks consisting of a
standard gage block, a comparator, an operator, environment, and the
calibration procedure.”
Reference Standards
A standard, generally of the highest metrological quality available at a given
location, from which measurements made at that location are derived.
Measurement and Test Equipment (M&TE)
All of the measurement instruments, measurement standards, reference
materials, and auxiliary apparatus that are necessary to perform a
measurement.
Calibration Standard
A standard that serves as a reference in the performance of routine
calibrations. Intended to act as a buffer between the calibration workload and
the laboratory’s reference standard(s).
18
W. E. Deming, Out of the Crisis (1982, 1986), p. 277.
19
This definition was later updated as Measurement and Test Equipment or M&TE by subsequent military
standards.
Definitions
and Potential
Sources of
Variation

Measurement Issues
44
Transfer Standard
A standard used to compare a separate standard of known value to the unit
being calibrated.
Master
A standard used as a reference in a calibration process. May also be termed
as reference or calibration standard.
Working Standard
A standard whose intended use is to perform routine measurements within
the laboratory, not intended as a calibration standard, but may be utilized as a
transfer standard.
Careful consideration needs to be given to the material(s) selected for
a standard. The materials employed ought to reflect the use and scope
of the measurement system, as well as time-based sources of variation
such as wear and environmental factors (temperature, humidity, etc.).
Figure I-E 1: Relationships among the Various Standards
Reference Standard
Calibration Standard
Working Standard
Transfer Standard
Transfer Standard
Master
Master
Measurement
&
Test
Equipment
Check Standard

Measurement Issues
45
Check Standard
A measurement artifact that closely resembles what the process is designed
to measure, but is inherently more stable than the measurement process being
evaluated.
Reference Value
A reference value, also known as the accepted reference value or master
value, is a value of an artifact or ensemble that serves as an agreed upon
reference for comparison. Accepted reference values are based upon the
following:
• Determined by averaging several measurements with a higher level
(e.g., metrology lab or layout equipment) of measuring equipment
• Legal values: defined and mandated by law
• Theoretical values: based on scientific principles
• Assigned values: based on experimental work (supported by sound
theory) of some national or international organization
• Consensus values: based on collaborative experimental work under
the auspices of a scientific or engineering group; defined by a
consensus of users such as professional and trade organizations
• Agreement values: values expressly agreed upon by the affected
parties
In all cases, the reference value needs to be based upon an operational
definition and the results of an acceptable measurement system. To achieve
this, the measuring system used to determine the reference value should
include:
• Instrument(s) with a higher order discrimination and a lower
measurement system error than the systems used for normal
evaluation
• Be calibrated with standards traceable to the NIST or other NMI
True Value
The true value is the “actual” measure of the part. Although this value is
unknown and unknowable, it is the target of the measurement process. Any
individual reading ought to be as close to this value as (economically)
possible. Unfortunately, the true value can never be known with certainty.
The reference value is used as the best approximation of the true value in all
analyses. Because the reference value is used as a surrogate for the true
value, these terms are commonly used interchangeably. This usage is not
recommended.P
20
F
20
See also ASTM E177-90a.

Measurement Issues
46
Discrimination
Discrimination is the amount of change from a reference value that an
instrument can detect and faithfully indicate. This is also referred to as
readability or resolution.
The measure of this ability is typically the value of the smallest graduation
on the scale of the instrument. If the instrument has “coarse” graduations,
then a half-graduation can be used.
A general rule of thumb is the measuring instrument discrimination ought to
be at least one-tenth of the range to be measured. Traditionally this range has
been taken to be the product specification. Recently the 10 to 1 rule is being
interpreted to mean that the measuring equipment is able to discriminate to at
least one-tenth of the process variation. This is consistent with the
philosophy of continual improvement (i.e., the process focus is a customer
designated target).
Figure I-E 2: Discrimination
The above rule of thumb can be considered as a starting point to determine
the discrimination since it does not include any other element of the
measurement system’s variability.
Because of economic and physical limitations, the measurement system will
not perceive all parts of a process distribution as having separate or different
measured characteristics. Instead the measured characteristic will be grouped
by the measured values into data categories. All parts in the same data
category will have the same value for the measured characteristic.
If the measurement system lacks discrimination (sensitivity or effective
resolution), it may not be an appropriate system to identify the process
variation or quantify individual part characteristic values. If that is the case,
better measurement techniques should be used.
The discrimination is unacceptable for analysis if it cannot detect the
variation of the process, and unacceptable for control if it cannot detect the
special cause variation (See Figure I-E 3).
half
interval
full
interval

Measurement Issues
47
Number of Categories Control Analysis
Can be used for control only if:
• The process variation is small
when compared to the
specifications
• The loss function is flat over
the expected process
variation
• The main source of variation
causes a mean shift
• Unacceptable for estimating
process parameters and indices
• Only indicates whether the
process is producing
conforming or nonconforming
parts
• Can be used with semi-
variable control techniques
based on the process
distribution
• Can produce insensitive
variables control charts
• Generally unacceptable for
estimating process parameters
and indices since it only
provides coarse estimates
• Can be used with variables
control charts
• Recommended
Figure I-E 3: Impact of Number of Distinct Categories (ndc) of the Process Distribution on
Control and Analysis Activities
Symptoms of inadequate discrimination may appear in the range chart.
Figure I-E 4 contains two sets of control charts derived from the same data.
Control Chart (a) shows the original measurement to the nearest thousandth
of an inch. Control Chart (b) shows these data rounded off to the nearest
hundredth of an inch. Control Chart (b) appears to be out of control due to
the artificially tight limits. The zero ranges are more a product of the
rounding off than they are an indication of the subgroup variation.
A good indication of inadequate discrimination can be seen on the SPC range
chart for process variation. In particular, when the range chart shows only
one, two, or three possible values for the range within the control limits, the
measurements are being made with inadequate discrimination. Also, if the
range chart shows four possible values for the range within control limits and
more than one-fourth of the ranges are zero, then the measurements are being
made with inadequate discrimination. Another good indication of inadequate
discrimination is on a normal probability plot where the data will be stacked
into buckets instead of flowing along the 45 degree line.
5 or more Data Categories
2 - 4 Data Categories
1 Data Category

Measurement Issues
48
Returning to Figure I-E 4, Control Chart (b), there are only two possible
values for the range within the control limits (values of 0.00 and 0.01).
Therefore, the rule correctly identifies the reason for the lack of control as
inadequate discrimination (sensitivity or effective resolution).
This problem can be remedied, of course, by changing the ability to detect
the variation within the subgroups by increasing the discrimination of the
measurements. A measurement system will have adequate discrimination if
its apparent resolution is small relative to the process variation. Thus a
recommendation for adequate discrimination would be for the apparent
resolution to be at most one-tenth of total process six sigma standard
deviation instead of the traditional rule which is the apparent resolution be at
most one-tenth of the tolerance spread.
Eventually, there are situations that reach a stable, highly capable process
using a stable, “best-in-class” measurement system at the practical limits of
technology. Effective resolution may be inadequate and further improvement
of the measurement system becomes impractical. In these special cases,
measurement planning may require alternative process monitoring
techniques. Customer approval will typically be required for the alternative
process monitoring technique.

Measurement Issues
49
Figure I-E 4: Process Control ChartsP
21
21
Figure I-E 4 was developed using data from Evaluating The Measurement Process, by Wheeler and Lyday,
Copyright 1989, SPC Press, Inc., Knoxville, Tennessee.
25
20
15
10
5
Subgroup 0
0.145
0.140
0.135
Sample
Mean
Mean=0.1397
UCL=0.1444
LCL=0.1350
0.02
0.01
0.00
Sample
Range
R=0.00812
UCL=0.01717
LCL=0
Xbar/R Chart
Discrimination = .001
25
20
15
10
5
Subgroup 0
0.145
0.140
0.135
Sample
Mean
Mean=0.1398
UCL=0.1438
LCL=0.1359
0.02
0.01
0.00
Sample
Range
R=0.0068
UCL=0.01438
LCL=0
Xbar/R Chart
Discrimination = .01
B
A

Measurement Issues
50
For most measurement processes, the total measurement variation is usually
described as a normal distribution. Normal probability is an assumption of
the standard methods of measurement systems analysis. In fact, there are
measurement systems that are not normally distributed. When this happens,
and normality is assumed, the MSA method may overestimate the
measurement system error. The measurement analyst must recognize and
correct evaluations for non-normal measurement systems.
Figure I-E 5: Characteristics of the Measurement Process Variation
Accuracy
Accuracy is a generic concept of exactness related to the closeness of
agreement between the average of one or more measured results and a
reference value. The measurement process must be in a state of statistical
control, otherwise the accuracy of the process has no meaning.
In some organizations accuracy is used interchangeably with bias. The ISO
(International Organization for Standardization) and the ASTM (American
Society for Testing and Materials) use the term accuracy to embrace both
bias and repeatability. In order to avoid confusion which could result from
using the word accuracy, ASTM recommends that only the term bias be used
as the descriptor of location error. This policy will be followed in this text.
Bias
Bias is often referred to as “accuracy.” Because “accuracy” has several
meanings in literature, its use as an alternate for “bias” is not recommended.
Location
Variation
Measurement
Process
Variation
Location
Width

Measurement Issues
51
Bias is the difference between the true value (reference value) and the
observed average of measurements on the same characteristic on the same
part. Bias is the measure of the systematic error of the measurement system.
It is the contribution to the total error
comprised of the combined effects of all
sources of variation, known or unknown,
whose contributions to the total error tends to
offset consistently and predictably all results
of repeated applications of the same
measurement process at the time of the
measurements.
Possible causes for excessive bias are:
• Instrument needs calibration
• Worn instrument, equipment or fixture
• Worn or damaged master, error in master
• Improper calibration or use of the setting master
• Poor quality instrument – design or conformance
• Linearity error
• Wrong gage for the application
• Different measurement method – setup, loading, clamping, technique
• Measuring the wrong characteristic
• Distortion (gage or part)
• Environment – temperature, humidity, vibration, cleanliness
• Violation of an assumption, error in an applied constant
• Application – part size, position, operator skill, fatigue, observation
error (readability, parallax)
The measurement procedure employed in the calibration process (i.e., using
"masters") should be as identical as possible to the normal operation's
measurement procedure.

Measurement Issues
52
Stability
Stability (or drift) is the total variation in the measurements obtained with a
measurement system on the same master or parts when measuring a single
characteristic over an extended time period. That is, stability is the change
in bias over time.
Reference Value
Time
Possible causes for instability include:
• Instrument needs calibration, reduce the calibration interval
• Normal aging or obsolescence
• Poor maintenance – air, power, hydraulic, filters, corrosion, rust,
cleanliness
• Worn or damaged master, error in master
• Improper calibration or use of the setting master
• Instrument design or method lacks robustness
• Distortion (gage or part)
• Environmental drift – temperature, humidity, vibration, cleanliness
Linearity
The difference of bias throughout the expected operating (measurement)
range of the equipment is called linearity. Linearity can be thought of as a
change of bias with respect to size.

Measurement Issues
53
Note that unacceptable linearity can come in a variety of flavors. Do not
assume a constant bias.
Possible causes for linearity error include:
• Instrument needs calibration, reduce the calibration interval
• Poor maintenance – air, power, hydraulic, filters, corrosion, rust,
cleanliness
• Worn or damaged master(s), error in master(s) – minimum/
maximum
BIAS
BIAS
Value N
Value 1

Measurement Issues
54
Repeatability
Reference Value
• Improper calibration (not covering the operating range) or use of the
setting master(s)
• Distortion (gage or part) changes with part size
• Environment – temperature, humidity, vibration, cleanliness
Precision
Traditionally, precision describes the net effect of discrimination, sensitivity
and repeatability over the operating range (size, range and time) of the
measurement system. In some organizations precision is used
interchangeably with repeatability. In fact, precision is most often used to
describe the expected variation of repeated measurements over the range of
measurement; that range may be size or time (i.e., “a device is as precise at
the low range as high range of measurement”, or “as precise today as
yesterday”). One could say precision is to repeatability what linearity is to
bias (although the first is random and the other systematic errors). The
ASTM defines precision in a broader sense to include the variation from
different readings, gages, people, labs or conditions.
Repeatability
This is traditionally referred to as the "within appraiser" variability.
Repeatability is the variation in measurements obtained with one
measurement instrument when used several times by one appraiser while
measuring the identical characteristic on the same part. This is the inherent
variation or capability of the equipment itself. Repeatability is commonly
referred to as equipment variation (EV), although this is misleading. In fact,
repeatability is the common cause (random error) variation from successive
trials under defined conditions of measurement. The best term for
repeatability is within-system variation when the conditions of measurement
are fixed and defined – fixed part, instrument, standard, method, operator,
environment, and assumptions. In addition to within-equipment variation,
repeatability will include all within variation (see below) from any condition
in the error model.
Width
Variation

Measurement Issues
55
Possible causes for poor repeatability include:
• Within-part (sample): form, position, surface finish, taper, sample
consistency
• Within-instrument: repair; wear, equipment or fixture failure, poor
quality or maintenance
• Within-standard: quality, class, wear
• Within-method: variation in setup, technique, zeroing, holding,
clamping
• Within-appraiser: technique, position, lack of experience,
manipulation skill or training, feel, fatigue
• Within-environment: short-cycle fluctuations in temperature,
humidity, vibration, lighting, cleanliness
• Violation of an assumption – stable, proper operation
• Instrument design or method lacks robustness, poor uniformity
• Distortion (gage or part), lack of rigidity
• Application – part size, position, observation error (readability,
parallax)
Reproducibility
This is traditionally referred to as the "between appraisers" variability.
Reproducibility is typically defined as the variation in the average of the
measurements made by different appraisers using the same measuring
instrument when measuring the identical characteristic on the same part.
This is often true for manual instruments influenced by the skill of the
operator. It is not true, however, for measurement processes (i.e., automated
systems) where the operator is not a major source of variation. For this
reason, reproducibility is referred to as the average variation between-
systems or between-conditions of measurement.
The ASTM definition goes
beyond this to potentially
include not only different
appraisers but also different:
gages, labs and environment
(temperature, humidity) as well
as including repeatability in the
calculation of reproducibility.
Reproducibility
Appraiser A C B

Measurement Issues
56
Potential sources of reproducibility error include:
• Between-parts (samples): average difference when measuring types
of parts A, B, C, etc, using the same instrument, operators, and
method.
• Between-instruments: average difference using instruments A, B, C,
etc., for the same parts, operators and environment. Note: in this
study reproducibility error is often confounded with the method
and/or operator.
• Between-standards: average influence of different setting standards
in the measurement process.
• Between-methods: average difference caused by changing point
densities, manual versus automated systems, zeroing, holding or
clamping methods, etc.
• Between-appraisers (operators): average difference between
appraisers A, B, C, etc., caused by training, technique, skill and
experience. This is the recommended study for product and process
qualification and a manual measuring instrument.
• Between-environment: average difference in measurements over time
1, 2, 3, etc. caused by environmental cycles; this is the most common
study for highly automated systems in product and process
qualifications.
• Violation of an assumption in the study
• Operator training effectiveness
• Application – part size, position, observation error (readability,
parallax)
As mentioned in the two definitions above, there are differences in
the definitions used by ASTM and those used by this manual. The
ASTM literature focuses on interlaboratory evaluations with interest
on laboratory-to-laboratory differences including the potential for
different operators, gages and environment as well as within
laboratory repeatability. Therefore, ASTM definitions need to
encompass these differences. By ASTM standards, repeatability is
the best the equipment will be under current conditions (one
operator, one gage, short period of time) and reproducibility
represents more typical operating conditions where there is
variation from multiple sources.
Gage R&R or GRR
Gage R&R is an estimate of the combined variation of repeatability and
reproducibility. Stated another way, GRR is the variance equal to the sum of
within-system and between-system variances.

Measurement Issues
57
2 2 2
GRR reproducibility repeatability
= 
σ σ σ
Sensitivity
Sensitivity is the smallest input that results in a detectable (usable) output
signal. It is the responsiveness of the measurement system to changes in
measured feature. Sensitivity is determined by gage design (discrimination),
inherent quality (OEM), in-service maintenance, and the operating condition
of the instrument and standard. It is always reported as a unit of measure.
Factors that affect sensitivity include:
• Ability to dampen an instrument
• Skill of operator
• Repeatability of the measuring device
• Ability to provide drift free operation in the case of electronic or
pneumatic gages
• Conditions under which the instrument is being used such as ambient
air, dirt, humidity
Consistency
Consistency is the difference in the variation of the measurements taken
over time. It may be viewed as repeatability over time.
Factors impacting consistency are special causes of variation such as:
• Temperature of parts
• Warm up required for electronic equipment
• Worn equipment
Reference Value
GRR
A C B

Measurement Issues
58
Uniformity
Uniformity is the difference in variation throughout the operating range of
the gage. It may be considered to be the homogeneity (sameness) of the
repeatability over size.
Factors impacting uniformity include:
• Fixture allows smaller/larger sizes to position differently
• Poor readability on the scale
• Parallax in reading
Capability
The capability of a measurement system is an estimate of the combined
variation of measurement errors (random and systematic) based on a short-
term assessment. Simple capability includes the components of:
• Uncorrected bias or linearity
• Repeatability and reproducibility (GRR), including short-term
consistency
Refer to Chapter III for typical methods and examples to quantify each
component.
An estimate of measurement capability, therefore, is an expression of the
expected error for defined conditions, scope and range of the measurement
system (unlike measurement uncertainty, which is an expression of the
expected range of error or values associated with a measurement result). The
capability expression of combined variation (variance) when the
measurement errors are uncorrelated (random and independent) can be
quantified as:
GRR
linearity
bias
capability
2
)
(
2
2
σ
σ
σ 
=
There are two essential points to understand and correctly apply
measurement capability:
First, an estimate of capability is always associated with a defined scope of
measurement – conditions, range and time. For example, to say that the
capability of a 25 mm micrometer is 0.1 mm is incomplete without
qualifying the scope and range of measurement conditions. Again, this is
why an error model to define the measurement process is so important. The
scope for an estimate of measurement capability could be very specific or a
general statement of operation, over a limited portion or entire measurement
range. Short-term could mean: the capability over a series of measurement
cycles, the time to complete the GRR evaluation, a specified period of
Measurement
System
Variation

Measurement Issues
59
production, or time represented by the calibration frequency. A statement of
measurement capability need only be as complete as to reasonably replicate
the conditions and range of measurement. A documented Control Plan could
serve this purpose.
Second, short-term consistency and uniformity (repeatability errors) over the
range of measurement are included in a capability estimate. For a simple
instrument, such as a 25 mm micrometer, the repeatability over the entire
range of measurement using typical, skilled operators is expected to be
consistent and uniform. In this example, a capability estimate may include
the entire range of measurement for multiple types of features under general
conditions. Longer range or more complex measurement systems (i.e., a
CMM) may demonstrate measurement errors of (uncorrected) linearity,
uniformity, and short-term consistency over range or size. Because these
errors are correlated they cannot be combined using the simple linear formula
above. When (uncorrected) linearity, uniformity or consistency varies
significantly over range, the measurement planner and analyst has only two
practical choices:
1) Report the maximum (worst case) capability for the entire
defined conditions, scope and range of the measurement system,
or
2) Determine and report multiple capability assessments for defined
portions of the measurement range (i.e., low, mid, larger range).
Performance
As with process performance, measurement system performance is the net
effect of all significant and determinable sources of variation over time.
Performance quantifies the long-term assessment of combined measurement
errors (random and systematic). Therefore, performance includes the long-
term error components of:
• Capability (short-term errors)
• Stability and consistency
Refer to Chapter III for typical methods and examples to quantify each
component.
An estimate of measurement performance is an expression of the expected
error for defined conditions, scope and range of the measurement system
(unlike measurement uncertainty, which is an expression of the expected
range of error or values associated with a measurement result). The
performance expression of combined variation (variance) when the
measurement errors are uncorrelated (random and independent) can be
quantified as:
y
consistenc
stability
capability
e
performanc
2
2
2
2
σ
σ
σ
σ 

=

Measurement Issues
60
Again, just as short-term capability, long-term performance is always
associated with a defined scope of measurement – conditions, range and
time. The scope for an estimate of measurement performance could be very
specific or a general statement of operation, over a limited portion or entire
measurement range. Long-term could mean: the average of several capability
assessments over time, the long-term average error from a measurement
control chart, an assessment of calibration records or multiple linearity
studies, or average error from several GRR studies over the life and range of
the measurement system. A statement of measurement performance need
only be as complete as to reasonably represent the conditions and range of
measurement.
Long-term consistency and uniformity (repeatability errors) over the range of
measurement are included in a performance estimate. The measurement
analyst must be aware of potential correlation of errors so as to not
overestimate the performance estimate. This depends on how the component
errors were determined. When long-term (uncorrected) linearity, uniformity
or consistency vary significantly over the range, the measurement planner
and analyst has only two practical choices:
1) Report the maximum (worst case) performance for the entire
defined conditions, scope and range of the measurement system,
or
2) Determine and report multiple performance assessments for a
defined portion of the measurement range (i.e., low, mid, larger
range).
Uncertainty
Measurement uncertainty is defined by VIM as a “parameter, associated
with the result of a measurement, that characteristics the dispersion of the
values that could reasonably be attributed to the measurand.”P
22
F See
Chapter I, Section F, for more detail.
Of a measurement system’s parameters, accuracy and precision are most
familiar to operating personnel since they are used in everyday life as well as
technical and sales discussions. Unfortunately, these terms are also the most
fuzzy as they are often thought of interchangeably. For example, if the gage
is certified by an independent agency as accurate, or if the instrument is
guaranteed to have high precision by the vendor, then it is incorrectly thought
that all readings will fall very close to the actual values. This is not only
conceptually wrong but can lead to wrong decisions about the product and
process.
This ambiguity carries over to bias and repeatability (as measures of
accuracy and precision). It is important to realize that:
22
Measurand is then defined by the VIM as “the particular quantity subject to measurement”.
Comments

Measurement Issues
61
• Bias and repeatability are independent of each other (See Figure I-E
6).
• Controlling one of these sources of error does not guarantee the
control of the other. Consequently, measurement systems control
programs (traditionally referred to as Gage Control Programs) ought
to quantify and track all relevant sources of variation.P
23
23
See also Chapter I, Section B.

Measurement Issues
62
Figure I-E 6: Relationships between Bias and Repeatability

Chapter I – Section F
Measurement Uncertainty
63
Section F
Measurement Uncertainty is a term that is used internationally to describe the
quality of a measurement value. While this term has traditionally been
reserved for many of the high accuracy measurements performed in
metrology or gage laboratories, many customer and quality system standards
require that measurement uncertainty be known and consistent with required
measurement capability of any inspection, measuring or test equipment.
In essence, uncertainty is the value assigned to a measurement result that
describes, within a defined level of confidence, the range expected to contain
the true measurement result. Measurement uncertainty is normally reported
as a bilateral quantity. Uncertainty is a quantified expression of measurement
reliability. A simple expression of this concept is:
True measurement = observed measurement (result)  U
U is the term for “expanded uncertainty” of the measurand and measurement
result. Expanded uncertainty is the combined standard error (uc), or standard
deviation of the combined errors (random and systematic), in the
measurement process multiplied by a coverage factor (k) that represents the
area of the normal curve for a desired level of confidence. Remember, a
normal distribution is often applied as a principle assumption for
measurement systems. The ISO/IEC Guide to the Uncertainty in
Measurement establishes the coverage factor as sufficient to report
uncertainty at 95% of a normal distribution. This is often interpreted as k = 2.
c
ku
=
U
The combined standard error (uc) includes all significant components of
variation in the measurement process. In most cases, methods of
measurement systems analysis performed in accordance with this manual can
be used as a tool to quantify many of the sources of measurement
uncertainty. Often, the most significant error component can be quantified by
2
performance
σ . Other significant error sources may apply based on the
measurement application. An uncertainty statement must include an adequate
scope that identifies all significant errors and allows the measurement to be
replicated. Some uncertainty statements will build from long-term, others
short-term, measurement system error. However, the simple expression can
be quantified as:
2 2 2
c performance other
u σ σ
= 

Chapter I – Section F
64
It is important to remember that measurement uncertainty is simply an
estimate of how much a measurement may vary at the time of measurement.
It should consider all significant sources of measurement variation in the
measurement process plus significant errors of calibration, master standards,
method, environment and others not previously considered in the
measurement process. In many cases, this estimate will use methods of MSA
and GRR to quantify those significant standard errors. It is appropriate to
periodically reevaluate uncertainty related to a measurement process to
assure the continued accuracy of the estimate.
The major difference between uncertainty and the MSA is that the MSA
focus is on understanding the measurement process, determining the amount
of error in the process, and assessing the adequacy of the measurement
system for product and process control. MSA promotes understanding and
improvement (variation reduction). Uncertainty is the range of measurement
values, defined by a confidence interval, associated with a measurement
result and expected to include the true value of measurement.
Traceability is the property of a measurement or the value of a standard
whereby it can be related to stated references, usually national or
international standards, through an unbroken chain of comparisons all having
stated uncertainties. Therefore understanding the measurement uncertainty of
each link in the chain is essential. By including both the short-term and long-
term sources of measurement variation that are introduced by the
measurement process and the chain of traceability, the measurement system’s
measurement uncertainty can be evaluated assuring that all effects of
traceability are taken into account. This, in turn, may reduce measurement
correlation issues.
The Guide to the Expression of Uncertainty of Measurement (GUM) is a
guide to how the uncertainty of a measurement may be evaluated and
expressed. While it provides the user with an understanding of the theory
and sets guidelines as to how the sources of measurement uncertainty can be
classified and combined, it should be considered a high level reference
document, not a “how to” manual. It does provide guidance to the user in
some of the more advanced topics such as, statistical independence of the
sources of variation, sensitivity analysis, degrees of freedom, etc. that are
critical when evaluating more complex, multi-parameter measurement
systems.
Measurement
Uncertainty
and MSA
Measurement
Traceability
ISO Guide to
the
Expression of
Uncertainty in
Measurement

Chapter I – Section G
Measurement Problem Analysis
65
Section G
An understanding of measurement variation and the contribution that it
makes to total variation needs to be a fundamental step in basic problem
solving. When variation in the measurement system exceeds all other
variables, it will become necessary to analyze and resolve those issues before
working on the rest of the system. In some cases the variation contribution of
the measurement system is overlooked or ignored. This may cause loss of
time and resources as the focus is made on the process itself, when the
reported variation is actually caused by the measurement device.
In this section a review will be made on basic problem solving steps and will
show how they relate to understanding the issues in a measurement system.
Each company may use the problem resolution process which the customer
has approved.
If the measurement system was developed using the methods in this manual,
most of the initial steps will already exist. For example, a cause and effect
diagram may already exist giving valuable lessons learned about the
measurement process. These data ought to be collected and evaluated prior to
any formal problem solving.
Identify the Issues
When working with measurement systems, as with any process, it is
important to clearly define the problem or issue. In the case of measurement
issues, it may take the form of accuracy, variation, stability, etc. The
important thing to do is try to isolate the measurement variation and its
contribution, from the process variation (the decision may be to work on the
process, rather than work on the measurement device). The issue statement
needs to be an adequate operational definition that anyone would understand
and be able to act on the issue.
Identify the Team
The problem solving team, in this case, will be dependent on the complexity
of the measurement system and the issue. A simple measurement system may
only require a couple of people. But as the system and issue become more
complex, the team may grow in size (maximum team size ought to be limited
to 10 members). The team members and the function they represent need to
be identified on the problem solving sheet.
Flowchart of Measurement System and Process
The team would review any historical flowcharting of the measurement
system and the process. This would lead to discussion of known and
unknown information about the measurement and its interrelationship to the
process. The flowcharting process may identify additional members to add to
the team.
Step 1
Step 2
Step 3

Chapter II – Section G
66
Cause and Effect Diagram
The team would review any historical Cause and Effect Diagram on the
Measurement System. This could, in some cases, result in the solution or a
partial solution. This would also lead to a discussion on known and unknown
information. The team would use subject matter knowledge to initially
identify those variables with the largest contribution to the issue. Additional
studies can be done to substantiate the decisions.
Plan-Do-Study-Act (PDSA)P
24
This would lead to a Plan-Do-Study-Act, which is form of scientific study.
Experiments are planned, data are collected, stability is established,
hypotheses are made and proven until an appropriate solution is reached.
Possible Solution and Proof of the Correction
The steps and solution are documented to record the decision. A preliminary
study is performed to validate the solution. This can be done using some
form of design of experiment to validate the solution. Also, additional studies
can be performed over time including environmental and material variation.
Institutionalize the Change
The final solution is documented in the report; then the appropriate
department and functions change the process so that the problem won’t recur
in the future. This may require changes in procedures, standards, and training
materials. This is one of the most important steps in the process. Most issues
and problems have occurred at one time or another.
24
W. Edwards Deming, The New Economics for Industry, Government, Education, The MIT Press, 1994, 2000
Step 4
Step 5
Step 6
Step 7

Chapter II
General Concepts for Assessing Measurement Systems
67
CHAPTER II
General Concepts for Assessing
Measurement Systems

Chapter II
General Concepts for Assessing Measurement Systems
68

Chapter II – Section A
Background
69
Section A
Background
Two important areas need to be assessed:
1) Verify the correct variable is being measured at the proper
characteristic location. Verify fixturing and clamping if
applicable. Also identify any critical environmental issues that
are interdependent with the measurement. If the wrong variable
is being measured, then no matter how accurate or how precise
the measurement system is, it will simply consume resources
without providing benefit.
2) Determine what statistical properties the measurement system
needs to have in order to be acceptable. In order to make that
determination, it is important to know how the data are to be
used, for without that knowledge, the appropriate statistical
properties cannot be determined. After the statistical properties
have been determined, the measurement system must be assessed
to see if it actually possesses these properties or not.
Phase 1 testing is an assessment to verify the correct variable is being
measured at the proper characteristic location per measurement system
design specification. (Verify fixturing and clamping if applicable) Also if
there are any critical environmental issues that are interdependent with the
measurement. Phase 1 could use a statistically designed experiment to
evaluate the effect of the operating environment on the measurement
system’s parameters (e.g., bias, linearity, repeatability, and reproducibility).
Phase 1 test results can indicate that the operating environment does not
contribute significantly to the overall measurement system variation.
Additionally, the variation attributable to the bias and linearity of the
measurement device should be small compared with the repeatability and
reproducibility components.
The knowledge gained during Phase 1 testing should be used as input to the
development of the measurement system maintenance program as well as the
type of testing which should be used during Phase 2. Environmental issues
may drive a change in location or a controlled environment for the
measurement device.
For example, if there is a significant impact of repeatability and
reproducibility on the total measurement system variation, a simple two-
factor statistical experiment could be performed periodically as a Phase 2
test.
Phase 2 testing provides ongoing monitoring of the key sources of variation
for continued confidence in the measurement system (and the data being
generated) and/or a signal that the measurement system has degraded over
time.
Does the
measurement
process satisfy the
requirements over
time?
Phase 1 & 2
Understand the
measurement
process and does
it satisfy the
requirements?

Chapter II – Section A
Background
70

Chapter II – Section B
Selecting/Developing Test Procedures
71
Section B
“Any technique can be useful if its limitations are understood and observed.”P
25
Many appropriate procedures are available for assessing measurement
systems. The choice of which procedure to use depends on many factors,
most of which need to be determined on a case-by-case basis for each
measurement system to be assessed. In some cases, preliminary testing may
be required to determine if a procedure is appropriate for a particular
measurement system or not. Such preliminary testing ought to be an integral
part of the Phase 1 testing discussed in the previous section.
General issues to consider when selecting or developing an assessment
procedure include:
• Should standards, such as those traceable to NIST, be used in the
testing and, if so, what level of standard is appropriate? Standards are
frequently essential for assessing the accuracy of a measurement
system. If standards are not used, the variability of the measurement
system can still be assessed, but it may not be possible to assess its
accuracy with reasonable credibility. Lack of such credibility may be
an issue, for instance, if attempting to resolve an apparent difference
between a producer’s measurement system and a customer’s
measurement system.
• For the ongoing testing in Phase 2, the use of blind measurements
may be considered. Blind measurements are measurements obtained
in the actual measurement environment by an operator who does not
know that an assessment of the measurement system is being
conducted. Properly administered, tests based on blind measurements
are usually not contaminated by the well-known Hawthorne effect.P
26
• The cost of testing.
• The time required for the testing.
• Any term for which there is no commonly accepted definition should
be operationally defined. Examples of such terms include accuracy,
precision, repeatability, reproducibility, etc.
25
W. Edwards Deming, The Logic of Evaluation, The Handbook of Evaluation Research, Vol. 1, Elmer L.
Struening and Marcia Guttentag, Editors
26
The “Hawthorne Effect” refers to the outcomes of a series of industrial experiments performed at the
Hawthorne Works of Western Electric between November 1924 and August 1932. In the experiments, the
researchers systematically modified working conditions of five assemblers and monitored the results. As the
conditions improved, production rose. However, when working conditions were degraded, production continued
to improve. This was thought to be the results of the workers having developed a more positive attitude toward
the work solely as a result of them being part of the study, rather than as a result of the changed working
conditions. See A History of the Hawthorne Experiments, by Richard Gillespie, Cambridge University Press,
New York, 1991.

Chapter II – Section B
72
• Will the measurements made by the measurement system be
compared with measurements made by another system? If so, one
should consider using test procedures that rely on the use of
standards such as those discussed in Phase 1 above. If standards are
not used, it may still be possible to determine whether or not the two
measurement systems are working well together. However, if the
systems are not working well together, then it may not be possible,
without the use of standards, to determine which system needs
improvement.
• How often should Phase 2 testing be performed? This decision may
be based on the statistical properties of the individual measurement
system and the consequence to the facility, and the facility’s
customers of a manufacturing process that, in effect, is not monitored
due to a measurement system not performing properly.
In addition to these general issues, other issues that are specific to the
particular measurement system being tested may also be important. Finding
the specific issues that are important to a particular measurement system is
one of the two objectives of the Phase 1 testing.

Chapter II – Section C
Preparation for a Measurement System Study
73
Section C
As in any study or analysis, sufficient planning and preparation ought to be
done prior to conducting a measurement system study. Typical preparation
prior to conducting the study is as follows:
1) The approach to be used should be planned. For instance,
determine by using engineering judgment, visual observations,
or a gage study, if there is an appraiser influence in calibrating or
using the instrument. There are some measurement systems
where the effect of reproducibility can be considered negligible;
for example, when a button is pushed and a number is printed
out.
2) The number of appraisers, number of sample parts, and number
of repeat readings should be determined in advance. Some
factors to be considered in this selection are:
(a) Criticality of dimension – critical dimensions require more
parts and/or trials. The reason being the degree of confidence
desired for the gage study estimations.
(b) Part configuration – bulky or heavy parts may dictate fewer
samples and more trials.
(c) Customer requirements.
3) Since the purpose is to evaluate the total measurement system,
the appraisers chosen should be selected from those who
normally operate the instrument.
4) Selection of the sample parts is critical for proper analysis and
depends entirely upon the design of the MSA study, purpose of
the measurement system, and availability of part samples that
represent the production process.
For Product Control situations where the measurement result and
decision criteria determine, “conformance or nonconformance to
the feature specification” (i.e., 100% inspection or sampling),
samples (or standards) must be selected, but need not cover the
entire process range. The assessment of the measurement system
is based on the feature tolerance (i.e., %GRR to TOLERANCE).
For Process Control situations where the measurement result and
decision criteria determine, “process stability, direction and
compliance with the natural process variation” (i.e., SPC, process
monitoring, capability, and process improvement), the availability of
samples over the entire operating range becomes very important.
An independent estimate of process variation (process capability
study) is recommended when assessing the adequacy of the
measurement system for process control (i.e., %GRR to process
variation).

74
When an independent estimate of process variation is not
available, OR to determine process direction and continued
suitability of the measurement system for process control, the
sample parts must be selected from the process and
represent the entire production operating range. The
variation in sample parts (PV) selected for MSA study is used to
calculate the Total Variation (TV) of the study. The TV index
(i.e., %GRR to TV) is an indicator of process direction and
continued suitability of the measurement system for process
control. If the sample parts DO NOT represent the production
process, TV must be ignored in the assessment. Ignoring TV does
not affect assessments using tolerance (product control) or an
independent estimate of process variation (process control).
Samples can be selected by taking one sample per day for
several days. Again, this is necessary because the parts will be
treated in the analysis as if they represent the range of production
variation in the process. Since each part will be measured several
times, each part must be numbered for identification.
5) The instrument should have a discrimination that allows at least
one-tenth of the expected process variation of the characteristic
to be read directly. For example, if the characteristic’s variation
is 0.001, the equipment should be able to “read” a change of
0.0001.
6) Assure that the measuring method (i.e., appraiser and
instrument) is measuring the dimension of the characteristic and
is following the defined measurement procedure.
The manner in which a study is conducted is very important. All analyses
presented in this manual assume statistical independenceP
27
F of the individual
readings. To minimize the likelihood of misleading results, the following
steps need to be taken:
1) The measurements should be made in a random orderP
28
F to
ensure that any drift or changes that could occur will be spread
randomly throughout the study. The appraisers should be
unaware of which numbered part is being checked in order to
avoid any possible knowledge bias. However, the person
conducting the study should know which numbered part is being
checked and record the data accordingly, that is Appraiser A,
Part 1, first trial; Appraiser B, Part 4, second trial, etc.
2) In reading the equipment, measurement values should be
recorded to the practical limit of the instrument discrimination.
Mechanical devices must be read and recorded to the smallest
unit of scale discrimination. For electronic readouts, the
measurement plan must establish a common policy for recording
the right-most significant digit of display. Analog devices should
27
There is no correlation between readings.
28
See Chapter III, Section B, “Randomization and Statistical Independence”

75
be recorded to one-half the smallest graduation or limit of
sensitivity and resolution. For analog devices, if the smallest
scale graduation is 0.0001”, then the measurement results should
be recorded to 0.00005”.
3) The study should be managed and observed by a person who
understands the importance of conducting a reliable study.
When developing Phase 1 or Phase 2 test programs there are several factors
that need to be considered:
• What effect does the appraiser have on the measurement
process? If possible, the appraisers who normally use the
measurement device should be included in the study.
Each appraiser should use the procedure – including all steps
– they normally use to obtain readings. The effect of any
differences between methods the appraisers use will be
reflected in the Reproducibility of the measurement system.
• Is appraiser calibration of the measurement equipment likely to
be a significant cause of variation? If so, the appraisers should
recalibrate the equipment before each group of readings.
• How many sample parts and repeated readings are required?
The number of parts required will depend upon the significance
of the characteristic being measured and upon the level of
confidence required in the estimate of measurement system
variation.
Although the number of appraisers, trials and parts may be
varied when using the recommended practices discussed in
this manual, the number of appraisers, trials and parts should
remain constant between Phase 1 and Phase 2 test programs
or between sequential Phase 2 tests for common measurement
systems. Maintaining commonality between test programs and
sequential tests will improve comparisons between the various
test results.

76

Chapter II – Section D
Analysis of the Results
77
Section D
The results should be evaluated to determine if the measurement device is
acceptable for its intended application. A measurement system should be
stable before any additional analysis is valid.
Acceptability Criteria – Gage Assembly and Fixture Error
An improperly designed fixture or poorly assembled gage will increase
measurement error. This is normally found when the measurements indicate
or display process instability or out-of-control conditions. This may be due to
excessive gage variation or poor repeatability and poor GRR values.
In general, the first thing to do when an apparent measurement issue exists, is
to review the assembly and setup instructions to make sure the gage was
properly assembled, (NOTE: this will not be in the instructions) and, for
example, the clamps/probes are positioned properly and have the right load.
Also, for automated measurement, verify the program follows required or
expected protocol.
If problems are found in any of these areas, reset or repair the gage and
fixtures, then rerun the measurement evaluation.
Acceptability Criteria – Location Error
Location error is normally defined by analyzing bias and linearity.
In general, the bias or linearity error of a measurement system is
unacceptable if it is significantly different from zero or exceeds the
maximum permissible error established by the gage calibration procedure. In
such cases, the measurement system should be recalibrated or an offset
correction applied to minimize this error.
Acceptability Criteria – Width Error
The criteria as to whether a measurement system’s variability is satisfactory
are dependent upon the percentage of the manufacturing production process
variability or the part tolerance that is consumed by measurement system
variation. The final acceptance criteria for specific measurement systems
depend on the measurement system’s environment and purpose and should
be agreed to by the customer. (See Chapter I Section B- “The Effects of
Measurement System Variability”)
When beginning to evaluate an organization’s measurement
systems, it can be useful to set priorities on which measurement
systems to initially focus. Since the final (total) variation is based
on a combination of the process and measurement variation,
( Total Process MSA
σ σ σ
= 
2 2
), when SPC is being applied for
process control or collecting process data, and the control chart
indicates that the process is stable and the total variation is
acceptable, the measurement system can be considered
Width Error
Location Error
Assembly or
Fixture Error

78
acceptable for this use and does not require separate re-
evaluationP
29
F. If an out-of-control condition or nonconformance is
found in this situation, the first thing that should be done is to
evaluate the measurement system.
For measurement systems whose purpose is to analyze a process, a general
guidelines for measurement system acceptability is as follows:
GRR Decision Comments
Under 10
percent
Generally considered to be an
acceptable measurement system.
Recommended, especially useful when trying to sort or
classify parts or when tightened process control is
required.
10 percent to
30 percent
May be acceptable for some
applications
Decision should be based upon, for example, importance
of application measurement, cost of measurement device,
cost of rework or repair.
Should be approved by the customer.
Over 30
percent
Considered to be unacceptable Every effort should be made to improve the measurement
system.
This condition may be addressed by the use of an
appropriate measurement strategy; for example, using the
average result of several readings of the same part
characteristic in order to reduce final measurement
variation.
Table II-D 1: GRR Criteria
Additional Width Error Metric
Another statistic of the measurement system variability is the number of
distinct categories (ndc)P
30
F. This statistic indicates the number of categories
into which the measurement process can be divided. This value should be
greater than or equal to 5.
Caution:
The use of the GRR guidelines as threshold criteria alone is NOT an
acceptable practice for determining the acceptability of a measurement
system.
29
If the total process is in statistical control for both the mean and variation charts and the total variation is
acceptable then it can be assumed either (1) both the actual process and measurement variability are acceptable
or (2) the measurement variation is not acceptable with respect to the process variation (which is extremely
small) but both are in statistical control. In either case the process is producing acceptable product. In case (2)
the existence of a nonconformance or out of control condition could be a false alarm (i.e., the product is
acceptable but the evaluation says it is not) which would cause an evaluation of both the measurement system
and process.
30
See Chapter I, Section E, “Measurement Issues”.

79
Applying the guidelines as the thresholds assumes that the calculated
statistics are deterministic estimates of the measurement system’s variability
(which they are not). Specifying the guidelines as the threshold criteria can
drive the wrong behavior. For example, the supplier may be creative in
achieving a low GRR by eliminating real life sources of variation (e.g., part
to gage interaction) or manipulating the measurement study (e.g., producing
parts outside the expected process variation).
Comments on the Application and Gage Acceptability
When looking at GRR and measurement variation it is important to look at
each application individually, to see what is required and how the
measurement is going to be used. For example: the required precision of
temperature measurement may be different for dissimilar applications. A
room thermostat can regulate the temperature for human comfort and is
economically priced, but may have a GRR upwards to 30%. It is acceptable
for this application. But in a laboratory, where small variations in
temperature can impact test results, a more sophisticated temperature
measurement and control is required. This thermostat will be more expensive
and is required to have less variability (i.e., to have a lower GRR).
The final acceptance of a measurement system should not come
down to a single set of indices. The long-term performance of the
measurement system should also be reviewed, for example, using
graphical analysis over time.

80

Chapter III
Recommended Practices for Replicable Measurement Systems
81
CHAPTER III
Recommended Practices for Replicable
Measurement Systems

Chapter III – Section A
Example Test Procedures
82

83
Section A
Examples of specific test procedures are presented in this chapter. The
procedures are simple to use and can be readily applied in a production
environment. As discussed previously, the test procedure which should be
used to understand a measurement system and to quantify its variability
depends on the sources of variation which may affect the measurement
system. In many situations the major sources of variation are due to the
instrument (gage/equipment), person (appraiser), and method (measurement
procedure). The test procedures in this chapter are sufficient for this type of
measurement system analysis.
The procedures are appropriate to use when:
 Only two factors or conditions of measurement (i.e., appraisers
and parts) plus measurement system repeatability are being
studied
 The effect of the variability within each part is negligible
 There is no statistical interaction between appraisers and parts
 The parts do not change functionally or dimensionally during the
study, i.e., are replicable
A statistical design of experiment can be conducted and/or subject matter
knowledge used to determine if these procedures are appropriate for any
specific measurement system.

84

Chapter III – Section B
Variable Measurement System Study – Guidelines
85
Section B
Variable Measurement System Study Guidelines
This section contains implementation guidelines for the measurement system
techniques described in Chapter I, Section E. A thorough review of Chapter
I, Section E is recommended to ensure proper application of these guidelines.
Guidelines for Determining Stability
Conducting the Study
1) Obtain a sample and establish its reference value(s) relative to a
traceable standard. If one is not available, select a production
partP
31
F that falls in the mid-range of the production measurements
and designate it as the master sample for stability analysis. The
known reference value is not required for tracking measurement
system stability.
It may be desirable to have master samples for the low end, the
high end, and the mid-range of the expected measurements.
Separate measurements and control charts are recommended for
each.
2) On a periodic basis (daily, weekly), measure the master sample
three to five times. The sample size and frequency should be
based on knowledge of the measurement system. Factors could
include how often recalibration or repair has been required, how
frequently the measurement system is used, and how stressful the
operating conditions are. The readings need to be taken at
differing times to represent when the measurement system is
actually being used. This will account for warm-up, ambient or
other factors that may change during the day.
3) Plot the data on an &
X R or &
X s control chart in time order.
Analysis of Results – Graphical
4) Establish control limits and evaluate for out-of-control or
unstable conditions using standard control chart analysis.
31
Caution should be taken where a production master could experience excessive wear due to use, material, and
handling. This may require modifying the production part, such as plating, to extend the life of the master.
Reference Value
Time

86
Analysis of Results – Numerical
Other than normal control chart analyses, there is no specific numerical
analysis or index for stability.P
32
F
If the measurement process is stable, the data can be used to determine the
bias of the measurement system.
Also, the standard deviation of the measurements can be used as an
approximation for the measurement system’s repeatability. This can be
compared with that of the process to determine if the measurement system
repeatability is suitable for the application.
Design of Experiments or other analytical problem solving
techniques may be required to determine the prime contributors to
the lack of measurement system stability.
Example – Stability
To determine if the stability of a new measurement instrument was
acceptable, the process team selected a part near the middle of the range of
the production process. This part was sent to the measurement lab to
determine the reference value which is 6.01. The team measured this part 5
times once a shift for four weeks (20 subgroups). After all the data were
collected, X & R charts were developed (see Figure III-B 1).
Figure III-B 1: Control Chart Analysis for Stability
Analysis of the control charts indicates that the measurement process is
stable since there are no obvious special cause effects visible.
32
See SPC Reference Manual.
0
Subgroup 10 20
5.7
5.8
5.9
6.0
6.1
6.2
6.3
Sample
Mean
6.021
UCL=6.297
LCL=5.746
0.0
0.5
1.0
Sample
Range
0.4779
UCL=1.010
LCL=0
Xbar/R Chart for Stability

87
Guidelines for Determining BiasP
33F – Independent Sample Method
The independent sample method for determining whether the bias is
acceptable uses the Test of Hypothesis:
0
1
0
0
H bias
H bias
=

The calculated average bias is evaluated to determine if the bias
could be due to random (sampling) variation.
In general, the bias or linearity error of a measurement system is
acceptable if it is not statistically significantly different from zero
when compared to the repeatability. Consequently, the repeatability
must be acceptable when compared to the process variation in order
for this analysis to be useful.
1) Obtain a sample and establish its reference value relative to a
traceable standard. If one is not available, select a production
part that falls in the mid-range of the production measurements
and designate it as the master sample for bias analysis. Measure
the part n  10 times in the gage or tool room, and compute the
average of the n readings. Use this average as the “reference
value.”
It may be desirable to have master samples for the low end of
the expected measurements, the high end, and the mid-range. If
this is done, analyze the data using a linearity study.
2) Have a single appraiser measure the sample n  10 times in the
normal manner.
3) Determine the bias of each reading:
i i
bias x reference
= − value
4) Plot the bias data as a histogram relative to the reference value.
Review the histogram, using subject matter knowledge, to
determine if any special causes or anomalies are present. If not,
continue with the analysis. Special caution ought to be exercised
for any interpretation or analysis when n < 30.
33
See Chapter I, Section E, for an operational definition and discussion of potential causes.

88
5) Compute the average bias of the n readings.
1
n
i
i
bias
avg bias
n
=
=
∑
6) Compute the repeatability standard deviation (see also Gage
Study, Range Method, below):
2
1
( )
1
n
i
i
repeatability r
X X
n
σ σ =
−
= =
−
∑
If a GRR study is available (and valid), the repeatability standard
deviation calculation should be based on the study results.
7) Determine if the repeatability is acceptable by calculating the
%EV = 100 [EV/TV] = 100 [ repeatability
σ /TV]
Where the total variation (TV) is based on the expected process
variation (preferred) or the specification range divided by 6 (see
also GRR study below).
If the %EV is large (see Chapter II, section D), then the
measurement system variation may be unacceptable. Since the
bias analysis assumes that the repeatability is acceptable,
continuing the analysis with a measurement system with a large
%EV will lead to misleading and confusing results. TF Note:
What specifically are we supposed to look at in Section D that
links to EV?
8) Determine the t statistic for the bias: F
34
r
b
n
= σ
σ
bias
b
average bias
t statistic t
σ
= =
9) Bias is acceptable (statistically zero) at the α level if
• the p-value associated with bias
t is less than α; or
34
The uncertainty for bias is given by b
σ .

89
• zero falls within the 1-α confidence bounds based on the
bias value:
   
b b
,1 , 1
2 2
σ σ
Bias t zero Bias t
α α
 
− −
   
−   
   
   
where  = n – 1 and
,1
2
t α
 −
is found using the standard t tables.
The α level which is used depends on the level of sensitivity
which is needed to evaluate/control the process and is
associated with the loss function (sensitivity curve) of the
product/process. Customer agreement should be obtained if an
α level other than the default value of .05 (95% confidence) is
used.
Example – Bias
A manufacturing engineer was evaluating a new measurement system for
monitoring a process. An analysis of the measurement equipment indicated
that there should be no linearity concerns, so the engineer had only the bias
of the measurement system evaluated. A single part was chosen within the
operating range of the measurement system based upon documented process
variation. The part was measured by layout inspection to determine its
reference value. The part was then measured fifteen times by the lead
operator.

90
Reference
Value = 6.00
Bias
1 5.8 -0.2
2 5.7 -0.3
3 5.9 -0.1
T 4 5.9 -0.1
R 5 6.0 0.0
I 6 6.1 0.1
A 7 6.0 0.0
L 8 6.1 0.1
S 9 6.4 0.4
10 6.3 0.3
11 6.0 0.0
12 6.1 0.1
13 6.2 0.2
14 5.6 -0.4
15 6.0 0.0
Table III-B 1: Bias Study Data
Using a spreadsheet and statistical software, the supervisor generated the
histogram and numerical analysis (see Figure III-B 2 & Table III-B 2).
5.6 5.7 5.8 5.9 6.0 6.1 6.2 6.3 6.4
0
1
2
3
4
Frequency
Measured value

91
0.4
0.2
0.0
-0.2
-0.4
4
3
2
1
0
X
_
Ho
Study-bias
Frequency
Histogram of Study-bias
(with Ho and 95% t-confidence interval for the mean)
Figure III-B 2: Bias Study – Histogram of Bias Study
The histogram did not show any anomalies or outliers requiring additional
analysis and review.
The repeatability of 0.2120 was compared to an expected process variation
(standard deviation) of 2.5. Since the %EV = 100(.2120/2.5) = 8.5%, the
repeatability is acceptable and the bias analysis can continue.
Since zero falls within the confidence interval of the bias (– 0.1107, 0.1241),
the engineer can assume that the measurement bias is acceptable assuming
that the actual use will not introduce additional sources of variation.

92
Table III-B 2: Bias Study – Analysis of Bias Study P
35
Guidelines for Determining Bias – Control Chart Method
If an X & R chart is used to measure stability, the data can also be used to
evaluate bias. The control chart analysis should indicate that the
measurement system is stable before the bias is evaluated.
1) Obtain a sample and establish its reference value relative to a
traceable standard. If one is not available, select a production part
that falls in the mid-range of the production measurements and
designate it as the master sample for bias analysis. Measure the part
n  10 times in the gage or tool room, and compute the average of
the n readings. Use this average as the “reference value.”
2) Conduct the stability study with g (subgroups) ≥ 20 subgroups of
size m.
35
Even though data is given with one digit after the decimal point, the results are shown as provided by a typical
statistical program using double precision; i.e., with multiple decimal digits.
Measured Value 0.12 14
Significant t value
(2-tailed)
Average
Bias
2.14479 -0.1107
df
t
statistic
0.1241
.0067
95% Confidence Interval
of the Bias
Lower Upper
Reference Value = 6.00, α =.05
Measured
Value
15 6.0067
Standard
Deviation,
Standard Error
of Mean,
0.2120 0.0547
n
b
σ
r
σ
Average

93
3) If the control chart indicates that the process is stable and m = 1, use
the analysis described for the independent sample method (see
above).
4) If m ≥ 2, plot the data as a histogram relative to the reference value.
Review the histogram, using subject matter knowledge, to determine
if any special causes or anomalies are present. If not, continue with
the analysis.
5) Obtain the X from the control chart
6) Compute the bias by subtracting the reference value from X .
bias = X – reference value
7) Compute the repeatability standard deviation using the Average
Range
*
2
R
d
repeatability
σ =
where *
2
d is based on the subgroup size (m) and the
number of subgroups in the chart (g). (see Appendix C)
σ /TV]
%EV will lead to misleading and confusing results; i.e., the
analysis can indicate that the bias is statistically zero while the
absolute magnitude of the bias exceeds acceptable equipment
values.

94
9) Determine the t statistic for the bias:P
36
r
b
gm
σ
σ =
bias
b
bias
t statistic t
σ
= =
10) Bias is acceptable (statistically zero) at the α level if zero falls
within the 1-α confidence bounds around the bias value:
   
b b
, 1 , 1
2 2
σ σ
Bias t zero Bias t
α α
 
− −
   
−   
   
   
where  is found in Appendix C
and ,1
2
t α
 −
is found using the standard t tables.
The α level which is used depends on the level of sensitivity
which is needed to evaluate/control the process and is
associated with the loss function (sensitivity curve) of the
product/process. Customer agreement should be obtained if an
α level other than the default value of .05 (95% confidence) is
used.
Example – Bias
Referencing Figure III-B 1, the stability study was performed on a part which
had a reference value of 6.01. The overall average of all the samples (20
subgroups of size 5 for n=100 samples) was 6.021. The calculated bias is
therefore 0.011.
numerical analysis (Table III-B 3).
Since zero falls within the confidence interval of the bias (-0.0299, 0.0519),
the process team can assume that the measurement bias is acceptable
assuming that the actual use will not introduce additional sources of
variation.
36
The uncertainty for bias is given by b
σ .

95
Table III-B 3: Bias Study – Analysis of Stability Study for Bias
Analysis of Bias Studies
If the bias is statistically non-zero, look for these possible causes:
• Error in master or reference value. Check mastering procedure.
• Worn instrument. This can show up in the stability analysis and will
suggest the maintenance or refurbishment schedule.
• Instrument made to wrong dimension.
• Instrument measuring the wrong characteristic.
• Instrument not calibrated properly. Review calibration procedure.
• Instrument used improperly by appraiser. Review measurement
instructions.
• Instrument correction algorithm incorrect.
If the measurement system has non-zero bias, where possible it should be
recalibrated to achieve zero bias through the modification of the hardware,
software or both. If the bias cannot be adjusted to zero, it still can be used
through a change in procedure (e.g., adjusting each reading by the bias).
Since this has a high risk of appraiser error, it should be used only with the
concurrence of the customer.
Measured Value .5371 72.7
Significant t
value
(2-tailed)
Bias
1.993 -.0299
df
t
statistic
.0519
.011
95% Confidence Interval
of the Bias
Lower Upper
Reference Value = 6.01, α =.05, m = 5, g = 20, d2P
*
= 2.334, d2 = 2.326
Measured
Value
100 6.021
Standard
Deviation,
Standard Error
of Mean,
.2048 .02048
n
b
σ
r
σ
Mean, X

96
Guidelines for Determining LinearityP
37
Linearity can be evaluated using the following guidelines:
1) Select g  5 parts whose measurements, due to process variation,
cover the operating range of the gage.
2) Have each part measured by layout inspection to determine its
reference value and to confirm that the operating range of the subject
gage is encompassed.
3) Have each part measured m  10 times on the subject gage by one of
the operators who normally use the gage.
 Select the parts at random to minimize appraiser “recall” bias in
the measurements.
4) Calculate the part bias for each measurement and the bias average for
each part.
 
, ,
i j i j i
bias x reference value
= −
,
1
m
i j
j
i
bias
bias
m
=
=
∑
5) Plot the individual biases and the bias averages with respect to the
reference values on a linear graph. (see Figure III-B 3.)
6) Calculate and plot the best fit line and the confidence band of the line
using the following equations.
For the best fit line, use: i i
y a x b
= 
where
i
x = reference value
i
y = bias average
and
37
Value 1 Value N

97
 
2
2
1
1
xy x y
gm
a slope
x x
gm
 
−  
 
= =
−
∑ ∑ ∑
∑ ∑
b y ax intercept
= − =
For a given x0 , the α level confidence bandsP
38
F are:
where
2
2
i i i i
y b y a x y
s
gm
− −
=
−
∑ ∑ ∑
lower:
 
 
2
0
2
1
2
0 2,1 2
1
i
gm
x x
x x
b a x t s
gm
α
− −
−
−
 
 
 
 
 − 
 
 
 
 
 
∑
upper:
 
 
2
0
2
1
2
0 2,1 2
1
i
gm
x x
x x
b a x t s
gm
α
− −
−
−
 
 
 
 
  
 
 
 
 
 
∑
7) The standard deviation of the variability of repeatability.
σ =
repeatability
s
Determine if the repeatability is acceptable by calculating the
σ /TV]
%EV will lead to misleading and confusing results.
8) Plot the “bias = 0” line and review the graph for indications of
special causes and the acceptability of the linearity. (see example
Figure III-B 3.)
38
See the note on selecting the α level in the “Guidelines for Determining Bias” area in Chapter III, Section B.

98
For the measurement system linearity to be acceptable, the “bias =
0” line must lie entirely within the confidence bands of the fitted
line.
9) If the graphical analysis indicates that the measurement system
linearity is acceptable then the following hypothesis should be true:
H0: a = 0 slope = 0
do not reject if
 
2, 1
2
2
gm
j
a
t t
s
x x
α
− −
= 
 
 
 
−
 
 
∑
If the above hypothesis is true then the measurement
system has the same bias for all reference values. For the
linearity to be acceptable this bias must be zero.
Ho: b = 0 intercept (bias) = 0
do not reject if
 
2, 1
2
2
2
1
gm
i
b
t t
x
s
gm x x
α
− −
= 
 
 

 
−
 
∑
Example – Linearity
A plant supervisor was introducing a new measurement system to the
process. As part of the PPAPP
39
F the linearity of the measurement system
needed to be evaluated. Five parts were chosen throughout the operating
range of the measurement system based upon documented process variation.
Each part was measured by layout inspection to determine its reference
value. Each part was then measured twelve times by the lead operator. The
parts were selected at random during the study.
39
Production Parts Approval Process manual, 4th
Edition, 2006

99
Part 1 2 3 4 5
Reference
Value 2.00 4.00 6.00 8.00 10.00
1 2.70 5.10 5.80 7.60 9.10
2 2.50 3.90 5.70 7.70 9.30
3 2.40 4.20 5.90 7.80 9.50
T 4 2.50 5.00 5.90 7.70 9.30
R 5 2.70 3.80 6.00 7.80 9.40
I 6 2.30 3.90 6.10 7.80 9.50
A 7 2.50 3.90 6.00 7.80 9.50
L 8 2.50 3.90 6.10 7.70 9.50
S 9 2.40 3.90 6.40 7.80 9.60
10 2.40 4.00 6.30 7.50 9.20
11 2.60 4.10 6.00 7.60 9.30
12 2.40 3.80 6.10 7.70 9.40
Table III-B 4: Linearity Study Data
linearity plot (Figure III-B 3).
Part 1 2 3 4 5
Reference
Value 2.00 4.00 6.00 8.00 10.00
1 0.7 1.1 -0.2 -0.4 -0.9
2 0.5 -0.1 -0.3 -0.3 -0.7
3 0.4 0.2 -0.1 -0.2 -0.5
B 4 0.5 1.0 -0.1 -0.3 -0.7
I 5 0.7 -0.2 0.0 -0.2 -0.6
A 6 0.3 -0.1 0.1 -0.2 -0.5
S 7 0.5 -0.1 0.0 -0.2 -0.5
8 0.5 -0.1 0.1 -0.3 -0.5
9 0.4 -0.1 0.4 -0.2 -0.4
10 0.4 0.0 0.3 -0.5 -0.8
11 0.6 0.1 0.0 -0.4 -0.7
12 0.4 -0.2 0.1 -0.3 -0.6
BIAS Avg. 0.491667 0.125 0.025 -0.29167 -0.61667
Table III-B 5: Linearity Study – Intermediate Results

100
Figure III-B 3: Linearity Study – Graphical Analysis
The graphical analysis indicates that special causes may be influencing the
measurements system. The data for reference value 4 appear to be bimodal.
Even if the data for reference value 4 were not considered, the graphical
analysis clearly shows that this measurement system has a linearity problem.
The R2
value indicates that a linear model may not be an appropriate model
for these data.P
40
F Even if the linear model is accepted, the “bias = 0” line
intersects the confidence bounds rather than being contained by them.
At this point, the supervisor ought to begin problem analysis and resolution
on the measurement system, since the numerical analysis will not provide
any additional insights. However, wanting to make sure no paperwork is left
unmarked, the supervisor calculates the t-statistic for the slope and intercept:
12.043
a
t = −
10.158
b
t =
Taking the default α = .05 and going to the t-tables with (gm – 2) = 58
degrees of freedom and a proportion of .975, the supervisor comes up with
the critical value of:
40
See standard statistical texts on analysis of the appropriateness of using a linear model to describe the
relationship between two variables.
10
9
8
7
6
5
4
3
2
1
0
-1
R-Sq = 71.4 %
Y = 0.736667 - 0.131667X
95% CI
Regression
Linearity Example
Bias Average
Reference Values
Bias
Bias = 0

101
58, .975 2.00172
t =
Since 58, .975
a
t t
 , the result obtained from the graphical analysis is
reinforced by the numerical analysis – there is a linearity problem with this
measurement system.
In this case, it does not matter what relation b
t has to 58, .975
t since there is a
linearity problem. Possible causes for linearity problems can be found in
Chapter I, Section E, “Location Variation.”
If the measurement system has a linearity problem, it needs to be recalibrated
to achieve zero bias through the modification of the hardware, software or
both.
If the bias cannot be adjusted to zero bias throughout the measurement
system range, it still can be used for product/process control but not analysis
as long as the measurement system remains stable.
Since this has a high risk of appraiser error, it should be used only
with the concurrence of the customer.
Guidelines for Determining Repeatability and ReproducibilityP
41
The Variable Gage Study can be performed using a number of differing
techniques. Three acceptable methods will be discussed in detail in this
section. These are:
• Range method
• Average and Range method (including the Control Chart method)
• ANOVA method
Except for the Range method, the study data design is very similar for each
of these methods. The ANOVA method is preferred because it measures the
operator to part interaction gauge error, whereas the Range and the Average
and Range methods does not include this variation. As presented, all
methods ignore within-part variation (such as roundness, diametric taper,
flatness, etc., as discussed in Chapter IV, Section D) in their analyses.
The ANOVA approach can identify appraiser-part interaction but it
can also evaluate other sources of variation which is the reason
why it was included. Historically, the assumption is made that the
interaction is zero, in which case the results of both approaches are
equivalent. With that said, the ANOVA approach is preferred
because of its flexibility if the user has access to a appropriate
computer program. If not, the X bar and R approach is appropriate
and can be done manually or via a computer program.
41

102
However, the total measurement system includes not only the gage itself and
its related bias, repeatability, etc., but also could include the variation of the
parts being checked. The determination of how to handle within-part
variation needs to be based on a rational understanding of the intended use of
the part and the purpose of the measurement.
Finally, all of the techniques in this section are subject to the prerequisite of
statistical stability.
Although reproducibility is usually interpreted as appraiser
variation, there are situations when this variation is due to other
sources of variation. For example, with some in-process
measurement systems there are no human appraisers. If all the
parts are handled, fixtured and measured by the same equipment,
then reproducibility is zero; i.e., only a repeatability study is
needed. If, however, multiple fixtures are used, then the
reproducibility is the between-fixture variation.
Range Method
The Range method is a modified variable gage study which will provide a
quick approximation of measurement variability. This method will provide
only the overall picture of the measurement system. It does not decompose
the variability into repeatability and reproducibility. It is typically used as a
quick check to verify that the GRR has not changed.
This approach has the potential to detect an unacceptable
measurement systemP
42
F 80% of the time with a sample size of 5
and 90% of the time with a sample size of 10.
The Range method typically uses two appraisers and five parts for the study.
In this study, both appraisers measure each part once. The range for each part
is the absolute difference between the measurement obtained by appraiser A
and the measurement obtained by appraiser B. The sum of the ranges is
found and the average range ( R ) is calculated. The total measurement
variability is found by multiplying the average range by *
2
1
d
where *
2
d is
found in Appendix C, with m = 2 and g = number of parts.
Parts Appraiser A Appraiser B Range (A, B)
1 0.85 0.80 0.05
2 0.75 0.70 0.05
3 1.00 0.95 0.05
4 0.45 0.55 0.10
5 0.50 0.60 0.10
42
i.e., %GRR > 30%

103
0.35
( ) 0.07
5 5
i
R
AverageRange R = = =
∑
*
2
0.07
0.0588
1.19 1.19
R R
GRR
d
     
= = = =
   
 
 
 
 
(Process Standard Deviation = 0.0777 from previous study)
% 100 * 75.7%
GRR
GRR
Process Standard Deviation
 
= =
 
 
Table III-B 6: Gage Study (Range Method)
To determine what percentage of the process standard deviation the
measurement variation consumes, convert the GRR to a percentage by
multiplying by 100 and dividing by the process standard deviation. In the
example (see Table III-B 6), the process standard deviation for this
characteristic is 0.0777, therefore,
% 100 * 75.7%
GRR
GRR
Process Standard Deviation
 
= =
 
 
Now that the %GRR for the measurement system is determined, an
interpretation of the results should be made. In Table III-B 6, the %GRR is
determined to be 75.7% and the conclusion is that the measurement system is
in need of improvement.
Average and Range Method
The Average and Range method ( X & R) is an approach which will provide
an estimate of both repeatability and reproducibility for a measurement
system. Unlike the Range method, this approach will allow the measurement
system’s variation to be decomposed into two separate components,
repeatability and reproducibility.P
43
F However, variation due to the interaction
between the appraiser and the part/gage is not accounted for in the analysis.
43
The ANOVA method can be used to determine the interaction between the gage and appraisers, if such exists.

104
Although the number of appraisers, trials and parts may be varied, the
subsequent discussion represents the optimum conditions for conducting the
study. Refer to the GRR data sheet in III-B 6a. The detailed procedure is as
follows:
1) Obtain a sample of n > 10 partsP
44
F that represent the actual or
expected range of process variation.
2) Refer to the appraisers as A, B, C, etc. and number the parts 1
through n so that the numbers are not visible to the appraisers.
See Chapter II, Section C.
3) Calibrate the gage if this is part of the normal measurement system
procedures. Let appraiser A measure n parts in a random orderP
45
F and
enter the results in row 1.
4) Let appraisers B and C measure the same n parts without seeing each
other’s readings; then enter the results in rows 6 and 11, respectively.
5) Repeat the cycle using a different random order of measurement.
Enter data in rows 2, 7 and 12. Record the data in the appropriate
column. For example if the first piece measured is part 7 then record
the result in the column labeled part 7. If three trials are needed,
repeat the cycle and enter data in rows 3, 8 and 13.
6) Steps 4 and 5 may be changed to the following when large part size
or simultaneous unavailability of parts makes it necessary:
 Let appraiser A measure the first part and record the reading in
row 1. Let appraiser B measure the first part and record the
reading in row 6. Let appraiser C measure the first part and
record the reading in row 11.
 Let appraiser A repeat reading on the first part and record the
reading in row 2, appraiser B record the repeat reading in row 7,
and appraiser C record the repeat reading in row 12. Repeat this
cycle and enter the results in rows 3, 8, and 13, if three trials are
to be used.
7) An alternative method may be used if the appraisers are on different
shifts. Let appraiser A measure all 10 parts and enter the reading in
row 1. Then have appraiser A repeat the reading in a different order
and enter the results in rows 2 and 3. Do the same with appraisers B
and C.
44
The total number of “ranges” generated ought to be > 15 for a minimal level of confidence in the results.
Although the form was designed with a maximum of 10 parts, this approach is not limited by that number. As
with any statistical technique, the larger the sample size, the less sampling variation and less resultant risk will
be present.
45
See Chapter III, Section B, “Randomization and Statistical Independence”

105
Table III-B 6a: Gage Repeatability and Reproducibility Data Collection Sheet
Gage Repeatability and Reproducibility Data Collection Sheet
PART
Appraiser
/Trial # 1 2 3 4 5 6 7 8 9 10
AVERAGE
1 A 1 0.29 -0.56 1.34 0.47 -0.80 0.02 0.59 -0.31 2.26 -1.36
2 2 0.41 -0.68 1.17 0.50 -0.92 -0.11 0.75 -0.20 1.99 -1.25
3 3 0.64 -0.58 1.27 0.64 -0.84 -0.21 0.66 -0.17 2.01 -1.31
4 Average a
X =
5 Range a
R =
6 B 1 0.08 -0.47 1.19 0.01 -0.56 -0.20 0.47 -0.63 1.80 -1.68
7 2 0.25 -1.22 0.94 1.03 -1.20 0.22 0.55 0.08 2.12 -1.62
8 3 0.07 -0.68 1.34 0.20 -1.28 0.06 0.83 -0.34 2.19 -1.50
9 Average b
X =
10 Range b
R =
11 C 1 0.04 -1.38 0.88 0.14 -1.46 -0.29 0.02 -0.46 1.77 -1.49
12 2 -0.11 -1.13 1.09 0.20 -1.07 -0.67 0.01 -0.56 1.45 -1.77
13 3 -0.15 -0.96 0.67 0.11 -1.45 -0.49 0.21 -0.49 1.87 -2.16
14 Average c
X =
15 Range c
R =
16
Part Average X =
p
R =
17 ([ ] [ ] [ ]) /[# ]
a b c
R R R R OF APPRAISERS
= =  =  = = = R =
18
DIFF
X [Max X = ] - [Min X = ] =
=
19 * R 4
UCL [R = ] [D ] =
=  =
*D4 = 3.27 for 2 trials and 2.58 for 3 trials. UCLR represents the limit of individual R’s. Circle those that are
beyond this limit. Identify the cause and correct. Repeat these readings using the same appraiser and unit as originally used or
discard values and re-average and recompute R and the limiting value from the remaining observations.
Notes:_____________________________________________________________________________________________________________

106
Analysis of Results – GraphicalP
46
The use of graphical tools is very important. The specific graphical tools
used depend on the experimental design employed to collect the data. A
systematic screening of the data for apparent special causes of variations by
using graphical tools should precede any other statistical analysis.
The following are some of the techniques which have proven to be useful.
(See also the Analysis of Variance Method).
The data from the measurement system analysis can be displayed graphically
by control charts. The idea of using control charts to answer questions
concerning the measurement system has been used by Western Electric (see
Reference List: AT&T Statistical Quality Control Handbook).
The averages of the multiple readings by each appraiser on each part are
plotted by appraiser with part number as an index. This can assist in
determining consistency between appraisers.
The grand average and control limits determined by using the average range
are also plotted. The resulting Average Chart provides an indication of
“usability” of the measurement system.
The area within the control limits represents the measurement sensitivity
(“noise”). Since the group of parts used in the study represents the process
variation, approximately one half or more of the averages should fall outside
the control limits. If the data show this pattern, then the measurement system
should be adequate to detect part-to-part variation and the measurement
system can provide useful information for analyzing and controlling the
process. If less than half fall outside the control limits then either the
measurement system lacks adequate effective resolution or the sample does
not represent the expected process variation.
46
Detailed descriptions of these analyses are beyond the scope of this document. For more information, refer to
the references and seek assistance from competent statistical resources
Average
Chart

107
Figure III-B 4: Average Chart – “Stacked”P
47
Review of the charts indicates that the measurement system appears to have
sufficient discrimination for processes with variation described by the sample
parts. No appraiser-to-appraiser differences are readily apparent.
Figure III-B 5: Average Chart – “Unstacked”
47
With the ANOVA approach, this is also referred to as appraiser-by-part interaction chart
-2
-1
0
1
2
Average
Appr A Appr B Appr C

108
The range control chart is used to determine whether the process is in
control. The reason being that no matter how large the measurement error
may be, the control limits will allow for that error. That is why the special
causes need to be identified and removed before a measurement study can be
relevant.
The ranges of the multiple readings by each appraiser on each part are plotted
on a standard range chart including the average range and control limit(s).
From the analysis of the data that are being plotted, several useful
interpretations can be made. If all ranges are in control, all appraisers are
doing the same job.
If one appraiser is out-of-control, the method used differs from the others.
If all appraisers have some out of control ranges, the measurement system is
sensitive to appraiser technique and needs improvement to obtain useful data.
Neither chart should display patterns in the data relative to the
appraisers or parts.
The ranges are not ordered data. Normal control chart trend
analysis must not be used even if the plotted points are connected
by lines.
Stability is determined by a point or points beyond the control limit;
within-appraiser or within-part patterns. Analysis for stability ought
to consider practical and statistical significance.
The range chart can assist in determining:
• Statistical control with respect to repeatability
• Consistency of the measurement process between appraisers for
each part
Figure III-B 6: Range Chart – “Stacked”
Range
Chart

109
Figure III-B 7: Range Chart – “Unstacked”
Review of the above charts indicates that there are differences between the
variability of the appraisers.
The individual readings are plotted by part for all appraisers (see Figure III-B
8) to gain insight into:
• The effect of individual parts on variation consistency
• Indication of outlier readings (i.e., abnormal readings)
Figure III-B 8: Run Chart by Part
Review of the above chart does not indicate any outliers or inconsistent parts.
Run Chart
1 2 3 4 5 6 7 8 9 10
-2
-1
0
1
2
part
Value
0.0
0.5
1.0
Range
Appr A Appr B Appr C

110
The individual readings are plotted by part-by-appraiser (see Figure III-B 9)
to gain insight into:
• Consistency between appraisers
• Indication of possible outliers
• Part-appraiser interactions
Review of the Figure III-B 9 does not indicate any significant outliers but
does indicate that appraiser C may have lower readings than the other
appraisers.
Figure III-B 9: Scatter Plot
Scatter Plot
-2
-1
0
1
2
1 2 3 4 5
-2
-1
0
1
2
6 7 8 9 10
Value
Value
A B C
Appraiser
Part
Part

111
In a Whiskers Chart, the high and low data values and the average by part-
by-appraiser are plotted (see Figure III-B 10). This provides insight into:
• Consistency between appraisers
• Indication of outliers
• Part-appraiser interactions
Part
Appr
10
9
8
7
6
5
4
3
2
1
3
2
1
3
2
1
3
2
1
3
2
1
3
2
1
3
2
1
3
2
1
3
2
1
3
2
1
3
2
1
2
1
0
-1
-2
Value
Figure III-B 10: Whiskers Chart
Review of Figure III-B 10 does not indicate any significant outliers but
indicates that appraiser B may have the most variability.
Whiskers
Chart

112
The data from the measurement system analysis can be analyzed by running
“Error Charts” (see Figure III-B 11) of the individual deviations from the
accepted reference values. The individual deviation or error for each part is
calculated as follows:
Error = Observed Value – Reference Value
or
Error = Observed Value – Average Measurement of the Part
This depends upon whether or not reference values of the data being
measured are available.
Figure III-B 11: Error Charts
Review of the above charts indicates:
• Appraiser A has an overall positive bias
• Appraiser B has the most variability but no apparent bias.
• Appraiser C has an overall negative bias
Error chart
0.5
0.0
-0.5
5
4
3
2
1
part
0.5
0.0
-0.5
10
9
8
7
6
part
A B C
Appraiser
error
error

113
The histogram plot (Figure III-B 12) is a graph that displays the frequency
distribution of the gage error of appraisers who participated in the study. It
also shows their combined frequency distribution.
If the reference values are available:
Error = Observed Value – Reference Value
Otherwise:
Normalized Value = Observed Value – Part Average
The histogram plot provides a quick visual overview of how the error is
distributed. Issues such as whether
bias or lack of consistency exists in
the measurements taken by the
appraisers, can be identified even
before the data are analyzed.
Analysis of the histograms (Figure
III-B 12) reinforces that of the error
charts. They also indicate that only
appraiser B has a symmetric form.
This may indicate that appraisers A
and C are introducing a systematic
source of variation which is resulting
in the biases.
Figure III-B 12: Normalized HistogramP
48
48
Note that the “0.0” of each of the histograms are aligned with respect to each other.
Normalized
Histogram
-0.7 -0.6 -0.5 -0.4 -0.3 -0.2 -0.1 0.0
0
1
2
3
4
5
6
7
Frequency
-0.4 -0.3 -0.2 -0.1 0.0 0.1 0.2 0.3 0.4 0.5 0.6
0
1
2
3
4
5
Frequency
0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45
0
1
2
3
4
5
6
7
Frequency
Appr A
Appr B
Appr C
Error = 0 line

114
plotted with the reference value or overall part averages as the index (see
Figure III-B 13). This plot can assist in determining:
• Linearity (if the reference value is used)
• Consistency in linearity between appraisers
Figure III-B 13: X–Y Plot of Averages by Size
X-Y Plot of
Averages by
Size
3
2
1
0
-1
-2
-3
3
2
1
0
-1
-2
-3
reference
3
2
1
0
-1
-2
-3
3
2
1
0
-1
-2
-3
reference
3
2
1
0
-1
-2
-3
3
2
1
0
-1
-2
-3
reference
Appr
A
Appr
B
Appr
C

115
plotted against each other with the appraisers as indices. This plot compares
the values obtained by one appraiser to those of another (see Figure III-B
14). If there were perfect agreement between appraisers, the plotted points
would describe a straight line through the origin and 45° to the axis.
Figure III-B 14: Comparison X–Y Plots
Comparison
X-Y Plots
1.25
-1.25
1.25
-1.25
1.25
-1.25
1.25
-1.25
Appr A
Appr B
Appr C

116
The Gage Repeatability and Reproducibility calculations are shown in
Figures III-B 15 and III-B 16. Figure III-B 15 shows the data collection sheet
on which all study results are recorded. Figure III-B 16 displays a report
sheet on which all identifying information is to be recorded and all
calculations made according to the prescribed formula.
Reproducible blank forms are available in the Sample Forms
section. The procedure for doing the calculations after data have
been collected is as follows:
(The following refers to Figure III-B 15)
1) Subtract the smallest reading from the largest reading in rows 1, 2
and 3; enter the result in row 5. Do the same for rows 6, 7, and 8;
and 11, 12, and 13 and enter results in rows 10 and 15,
respectively.
2) Entries in rows 5, 10 and 15 are ranges and therefore always
positive values.
3) Total row 5 and divide the total by the number of parts sampled to
obtain the average range for the first appraisers trials a
R . Do the
same for rows 10 and 15 to obtain b
R and c
R .
4) Transfer the averages of rows 5, 10, and 15 ( a
R , b
R , c
R ) to row
17. Add them together and divide by the number of appraisers and
enter results R (average of all ranges).
5) Enter R (average value) in rows 19 and multiply by D4 P
49
F to get
the upper control limit. Note: D4 is 3.27 if two trials are used. The
value of the Upper Control Limit (UCLR ) of the individual ranges
is entered in row 19. Note: The value of Lower Control Limit
(LCLR) for less than seven trials is equal to zero.
6) Repeat any readings that produced a range greater than the
calculated UCLR using the same appraiser and part as originally
used, or discard those values and re-average and recompute R and
the limiting value UCLR based upon the revised sample size.
Correct the special cause that produced the out-of-control
condition. If the data were plotted and analyzed using a control
chart as discussed previously, this condition would have already
been corrected and would not occur here.
7) Sum the rows (rows 1, 2, 3, 6, 7, 8, 11, 12, and 13). Divide the
sum in each row by the number of parts sampled and enter these
values in the right-most column labeled “Average”.
49
See Statistical Process Control (SPC) Reference Manual, 2005, or other statistical reference source for a table
of factors.
Numerical
Calculations

117
8) Add the averages in rows 1, 2 and 3 and divide the total by the
number of trials and enter the value in row 4 in the a
X block.
Repeat this for rows 6, 7 and 8; and 11, 12 and 13, and enter the
results in the blocks for b
X and c
X in rows 9 and 14,
respectively.
9) Enter the maximum and minimum averages of rows 4, 9 and 14 in
the appropriate space in row 18 and determine the differences.
Enter this difference in the space labeled DIFF
X in row 18.
10) Sum the measurements for each trial, for each part, and divide the
total by the number of measurements (number of trials times the
number of appraisers). Enter the results in row 16 in the spaces
provided for part average.
11) Subtract the smallest part average from the largest part average
and enter the result in the space labeled p
R in row 16. p
R is the
range of part averages.
(The following refers to Figure III-B 16)
12) Transfer the calculated values of R , DIFF
X and p
R to the blanks
provided on the report side of the form.
13) Perform the calculations under the column entitled “Measurement
Unit Analysis” on the left side of the form.
14) Perform the calculations under the column entitled “% Total
Variation” on the right side of the form.
15) Check the results to make sure no errors have been made.

118
Figure III-B 15: Completed GR&R Data Collection Sheet
PART
Appraiser
/Trial # 1 2 3 4 5 6 7 8 9 10
AVERAGE
1 A 1 0.29 -0.56 1.34 0.47 -0.80 0.02 0.59 -0.31 2.26 -1.36 0.194
2 2 0.41 -0.68 1.17 0.50 -0.92 -0.11 0.75 -0.20 1.99 -1.25 0.166
3 3 0.64 -0.58 1.27 0.64 -0.84 -0.21 0.66 -0.17 2.01 -1.31 0.211
4 Average 0.447 -0.607 1.260 0.537 -0.853 -0.100 0.667 -0.227 2.087 -1.307 a
X = 0.1903
5 Range 0.35 0.12 0.17 0.17 0.12 0.23 0.16 0.14 0.27 0.11 a
R = 0.184
6 B 1 0.08 -0.47 1.19 0.01 -0.56 -0.20 0.47 -0.63 1.80 -1.68 0.001
7 2 0.25 -1.22 0.94 1.03 -1.20 0.22 0.55 0.08 2.12 -1.62 0.115
8 3 0.07 -0.68 1.34 0.20 -1.28 0.06 0.83 -0.34 2.19 -1.50 0.089
9 Average 0.133 -0.790 1.157 0.413 -1.013 0.027 0.617 -0.297 2.037 -1.600 b
X = 0.0683
10 Range 0.18 0.75 0.40 1.02 0.72 0.42 0.36 0.71 0.39 0.18 b
R = 0.513
11 C 1 0.04 -1.38 0.88 0.14 -1.46 -0.29 0.02 -0.46 1.77 -1.49 -0.223
12 2 -0.11 -1.13 1.09 0.20 -1.07 -0.67 0.01 -0.56 1.45 -1.77 -0.256
13 3 -0.15 -0.96 0.67 0.11 -1.45 -0.49 0.21 -0.49 1.87 -2.16 -0.284
14 Average -0.073 -1.157 0.880 0.150 -1.327 -0.483 0.080 -0.503 1.697 -1.807 c
X = -0.2543
15 Range 0.19 0.42 0.42 0.09 0.39 0.38 0.20 0.10 0.42 0.67 c
R = 0.328
16
Part
Average
0.169 -0.851 1.099 0.367 -1.064 -0.186 0.454 -0.342 1.940 -1.571 X =
p
R =
.0014
3.511
17 ([ a
R = 0.184] + [ b
R = 0.513] + [ c
R = 0.328]) / [ # OF APPRAISERS = 3] = 0.3417 R = 0.3417
18 [Max X = 0.1903] – [Min X = -0.2543] = DIFF
X = 0.4446
19
* [ R = 0.3417] x [D4 = 2.58] = UCLR = 0.8816
beyond this limit. Identify the cause and correct. Repeat these readings using the same appraiser and unit as originally used or
Notes:_____________________________________________________________________________________________________________

119
Figure III-B 16: Gage Repeatability and Reproducibility Report
Gage Repeatability and Reproducibility Report
Part No. & Name:
Characteristics:
Specifications:
From data sheet: R = 0.3417
Gage Name:
Gage No:
Gage Type:
DIFF
X = 0.4446
Date:
Performed by:
p
R = 3.511
Measurement Unit Analysis % Total Variation (TV)
Repeatability – Equipment Variation (EV)
EV = R  K1
%EV = 100 [EV/TV]
= 0.3417  0.5908 Trials K1 = 100 [0.20188/1.14610]
= 0.20188 2 0.8862 = 17.62%
3 0.5908
Reproducibility – Appraiser Variation (AV)
AV =  2 2
2
( ( ) )
-

DIFF
X K EV nr %AV = 100 [AV/TV]
=  
 
2 2
0.4446 0.5231 ( 0.20188 (10 3))
- = 100 [0.22963/1.14610]
= 0.22963 Appraisers 2 3 = 20.04%
n = parts r = trials K2 0.7071 0.5231
Repeatability & Reproducibility (GRR)
GRR =
2 2

EV AV %GRR = 100 [GRR/TV]
=  

2 2
0.20188 0.22963 Parts K3
= 100 [= 0.30575/1.14610]
= 0.30575 2 0.7071 = 26.68%
3 0.5231
Part Variation (PV) 4 0.4467 % PV = 100 [PV/TV]
PV = p
R  K3 5 0.4030 = 100 [1.10456/1.14610]
= 1.10456 6 0.3742 = 96.38%
Total Variation (TV) 7 0.3534
TV =
2 2

GRR PV 8 0.3375
=  

2 2
0.30575 1.10456 9 0.3249 ndc =  
1.41 PV
GRR
= 1.14610 10 0.3146
= 1.41(1.10456/0.30575)
= 5.094 ~ 5
For information on the theory and constants used in the form see MSA Reference Manual, Fourth edition.

120
The Gage Repeatability and Reproducibility Data Collection Sheet and
Report forms, Figures III-B 15 and III-B 16, will provide a method for the
numerical analysis of the study dataP
50
F. The analysis will estimate the
variation and percent of process variation for the total measurement system
and its components repeatability, reproducibility, and part variation. This
information needs to be compared to and complement the results of the
graphical analysis.
On the left side of the form (Figure III-B 16) under Measurement Unit
Analysis, the standard deviation is calculated for each component of
variation.
The repeatability or equipment variation (EV or E
σ ) is determined by
multiplying the average range ( R ) by a constant (K1). K1 depends upon the
number of trials used in the gage study and is equal to the inverse of *
2
d
which is obtained from Appendix C. *
2
d is dependent on the number of trials
(m) and the number of parts times the number of appraisers (g) (assumed to
be greater than 15 for calculating the value of K1)
The reproducibility or appraiser variation (AV or A
σ ) is determined by
multiplying the maximum average appraiser difference ( X DIFF ) by a
constant (K2 ). K2 depends upon the number of appraisers used in the gage
study and is the inverse of *
2
d which is obtained from Appendix C. *
2
d is
dependent on the number of appraisers (m) and g = 1, since there is only one
range calculation. Since the appraiser variation is contaminated by the
equipment variation, it must be adjusted by subtracting a fraction of the
equipment variation. Therefore, the appraiser variation (AV) is calculated by
 
 2
2
2
DIFF
EV
AV X K
nr
=  −
where n = number of parts and r = number of trials.
If a negative value is calculated under the square root sign, the appraiser
variation (AV) defaults to zero.
The measurement system variation for repeatability and reproducibility (GRR
or M
σ ) is calculated by adding the square of the equipment variation and the
square of the appraiser variation, and taking the square root as follows:
2 2
( ) ( )
GRR EV AV
= 
There are generally four different approaches to determine the process
variation which is used to analyze the acceptability of the measurement
variation:
50
The numerical results in the example were developed as if they were computed manually; i.e., the results were
carried and rounded to one additional decimal digit. Analysis by computer programs should maintain the
intermediate values to the maximum precision of the computer/programming language. The results from a valid
computer program may differ from the example results in the second or greater decimal place but the final
analysis will remain the same.
Analysis of Results
— Numerical

121
1) using process variation
o process variation, taken from the parts in the GRR study
itself
o use when the selected sample represents the expected
process variation (preferred option)
2) surrogate process variation
o use when sufficient samples to represent the process are not
available but an existing process with similar process
variation is available
3) Pp (or Ppk) target value
o use when sufficient samples to represent the process are not
available and an existing process with similar process
variation is not available or the new process is expected to
have less variability then an existing process
4) specification tolerance
o When the measurement system is to be used to sort the
process, and the process has a Pp <1.0
The part variation (part-to-part; part variation without measurement
variation) (PV or P
σ ) is determined by multiplying the range of part
averages ( p
R ) by a constant (K3). K3 depends upon the number of parts used
in the gage study and is the inverse of *
2
d which is obtained from Appendix
C. *
2
d is dependent on the number of parts (m) and (g). In this situation g = 1
since there is only one range calculation.
The total variation (TV or T
σ ) from the study is then calculated by summing
the square of both the repeatability and reproducibility variation and the part
variation (PV) and taking the square root as follows:
2 2
( ) ( )
TV GRR PV
= 
Using Historical Variation Information
To use this approach the information must be from a process that is in
statistical control. If the process variation is known and its value is based on
6σ , then it can be used in place of the total study variation (TV) calculated
from the gage study data. This is accomplished by performing the following
two calculations:
1)
6.00
process variation
TV =
2) 2 2
( ) - ( )
PV TV GRR
=

122
Using a Pp (or Ppk) target value
To use the Pp option, use the following TV in the GRR analysis:
6 6 6
6
σ
− − −
= = =
−
=
p
P
USL LSL USL LSL USL LSL
since P
s TV
USL LSL
then TV
Pp
and 2 2
( ) - ( )
PV TV GRR
=
Using the Tolerance (Specification Range).
When comparing measurement error from a GRR study to a tolerance, this is
the same as comparing it to a production process with a Pp of 1.00. OEM
customers rarely expect process variation to have as low a Pp(k) as 1.00, nor
do they accept a process at that low of a performance level. It may make
more sense to compare the measurement variation to a target production
process performance level which meets the customer requirement.P
51
To use this option, use the following TV in the GRR analysis:
6
−
=
USL LSL
TV
and 2 2
( ) - ( )
PV TV GRR
=
Indices
Once the variability for each factor in the gage study is determined, it can be
compared to the total variation (TV). This is accomplished by performing the
calculations on the right side of the gage report form (Figure III-B 16) under
“% Total Variation.”
The percent the equipment variation (%EV) consumes of the total variation
(TV) is calculated by 100[EV/TV]. The percent that the other factors consume
of the total variation can be similarly calculated as follows:
%AV = 100 [AV/TV]
%GRR = 100 [GRR/TV]
%PV = 100 [PV/TV]
THE SUM OF THE PERCENT CONSUMED BY EACH FACTOR
WILL NOT EQUAL 100%.
The results of this percent total variation need to be evaluated to determine if
the measurement system is acceptable for its intended application.
51
for example, see Chrysler, Ford, and GM, PPAP Manual.

123
If the analysis is based on the tolerance instead of the process
variation, then the Gage Repeatability and Reproducibility Report
form (Figure III-B 16) can be modified so that the right-hand side of
the form represents the percent of tolerance instead of percent of
total variation. In that case, %EV, %AV, %GRR and %PV are
calculated by substituting the value of tolerance divided by six in
the denominator of the calculations in place of the total variation
(TV). Either or both approaches can be taken depending on
the intended use of the measurement system and the desires
of the customer.
The final step in the numerical analysis is to determine the number of distinct
categories that can be reliably distinguished by the measurement system.
This is the number of non-overlapping 97% confidence intervals that will
span the expected product variation.P
52
F
 
1.41 PV
ndc
GRR
=
Given that the graphical analysis has not indicated any special cause
variation, the rule of thumb for gage repeatability and reproducibility
(%GRR) may be found in Chapter II, Section D.
For analysis, the ndc is the maximum of one or the calculated value truncated
to the integer. This result should be greater than or equal to 5.
To avoid a ndc = 0, which is possible with truncation alone, some computer
programs will round up the calculated result. This can result in differences in
final reports when the same data is evaluated by different programs
When using the Pp approach to TV, the calculation for ndc is:
. .
TV PV GRR
or PV TV GRR
then
TV GRR
PV
ndc
GRR GRR
= 
= −
−
= =
2 2 2
2 2 2
2 2
1 41 1 41
Analysis of Variance (ANOVA) Method
Analysis of variance (ANOVA) is a standard statistical technique and can be
used to analyze the measurement error and other sources of variability of
data in a measurement systems study. In the analysis of variance, the
variance can be decomposed into four categories: parts, appraisers,
interaction between parts and appraisers, and replication error due to the
gage.
52
The importance of the number of distinct categories (ndc) in control and analysis activities is discussed in
Chapter I, Section E, "Measurement Issues" (especially Figure I-E 3). Calculation of ndc is reviewed in Chapter
II, Section B, "Analysis of Results - Numerical".

124
The advantages of ANOVA techniques as compared with Average and
Range methods are:
• They are capable of handling any experimental set-up
• Can estimate the variances more accurately
• Extract more information (such as interaction between parts and
appraisers effect) from the experimental data.
The disadvantages are that the numerical computations are more complex
and users require a certain degree of statistical knowledge to interpret the
results. The ANOVA method as described in the following sections is
advised, especially if a computer is available.
The method of collecting the data is important in an ANOVA method. If the
data are not collected in a random manner, this can lead to a source of bias
values. A simple way to assure a balanced design for (n) parts, (k) appraisers,
and (r) trials is through randomization. One common approach to
randomization is to write A1 on a slip of paper to denote the measurement for
the first appraiser on the first part. Do this up to A(n) for the measurement by
the first appraiser on the nth
part. Follow the same procedure for the next
appraiser up to and including the kth
appraiser. The similar notation will be
used where B1, C1 denotes the measurement for second and third appraiser
on the first part. Once all nk combinations are written, then the slips of paper
can be put in a hat or bowl. One at a time, a slip of paper is selected. These
combinations (A1, B2, ...) are the measuring order in which the gage study
will be performed. Once all nk combinations are selected, they are put back
into the hat and the procedure is followed again. This is done for a total of r
times to determine the order of experiments for each repeat.
There are alternate approaches to generate a random sample. Care should be
exercised to differentiate among random, haphazard and convenience
sampling.P
53
F P
54
In general, all efforts need to be taken to assure statistical independence
within the study.
The data can be collected in a random manner using a form similar to Table
III-B 6a. For our example, there are ten parts and three appraisers, and the
experiment has been performed in random order three times for each part and
appraiser combination.
53
See Wheeler and Lyday, Evaluating the Measurement Process, Second Edition, 1989, p. 27.
54
Robert R. Coveyou (1915 - 20 Feb 1996) was an American mathematician who was a health physicist with the
Manhattan Project from 1943 during WW II. He became a recognized expert in pseudo-random number
generators.
Randomization
and Statistical
Independence
“The generation of
random numbers is
too important to be
left to chance.”
Robert R. Coveyou56

125
Graphical Analysis
Any of the graphical methods given in the discussion of Graphical Analysis
above can be used in the graphical analysis of the data collected as part of an
ANOVA study. These methods can be used to confirm and provide further
insight to the data (i.e., trends, cycles, etc.).
One graphical method that is suggested is called an interaction plot. This plot
confirms the results of the F test on whether or not the interaction is
significant. In this particular interaction plot, the average measurement per
appraiser per part vs. part number (1, 2, … etc.) is graphed in Figure III-B
17. The points for each appraiser average measurement per part are
connected to form k (number of appraisers) lines. The way to interpret the
graph is if the k lines are parallel there is no interaction term. When the lines
are nonparallel, the interaction can be significant. The larger the angle of
intersection is, the greater is the interaction. Appropriate measures should be
taken to eliminate the causes for the interaction. In the example in Figure III-
B 17, the lines are nearly parallel, indicating no significant interaction.
Figure III-B 17: Interaction Plot
Another graph sometimes of interest is the residuals plot. This graph is more
a check for the validity of the assumptions. This assumption is that the gage
(error) is a random variable from a normal distribution. The residuals, which
are the differences between the observed readings and predicted values, are
plotted. Predicted value is the average of the repeated readings for each
appraiser for each part. If the residuals are not randomly scattered above and
below zero (horizontal reference line), it could be because the assumptions
are incorrect and further investigation of the data is suggested.

126
Figure III-B 18: Residual Plot
Numerical Calculations
Although the values can be calculated manually, most people will use a
computer program to generate what is called the Analysis of Variance
(ANOVA) table (see Appendix A).
The ANOVA table here is composed of five columns (see Table III-B 7).
• Source column is the cause of variation.
• DF column is the degree of freedom associated with the source.
• SS or sum of squares column is the deviation around the mean of
the source.
• MS or mean square column is the sum of squares divided by
degrees of freedom.
• F-ratio column, calculated to determine the statistical
significance of the source value.
The ANOVA table is used to decompose the total variation into four
components: parts, appraisers, interaction of appraisers and parts, and
repeatability due to the instrument.
For analysis purposes, negative variance components are set to zero.
This information is used to determine the measurement system characteristics
as in the Average and Range Method
2
1
0
-1
-2
0.5
0.0
-0.5
Fitted Value
Residual
Residuals Versus the Fitted Values
(response is value)

127
Source DF SS MS F
Appraiser 2 3.1673 1.58363 34.44*
Parts 9 88.3619 9.81799 213.52*
Appraiser by Part 18 0.3590 0.01994 0.434
Equipment 60 2.7589 0.04598
Total 89 94.6471
* Significant at α = 0.05 level
Table III-B 7: ANOVA Table
Table III-B 7 shows the ANOVA calculations for the example data from
Figure III-B 15 assuming a balanced two-factor factorial design. Both factors
are considered to be random. Table III-B 9 shows the comparison of the
ANOVA method with the Average and Range method. Table III-B 10 shows
the GRR report for the ANOVA method.
Estimate of
Variance
Std.
Dev. (σ)
% Total Variation % Contribution
2
 = 0.039973
(Repeatability)
EV = 0.199933 18.4 3.4
2
 = 0.051455
(Appraiser)
AV = 0.226838 20.9 4.4
2
 = 0
(Interaction)
INT = 0 0 0
System =
0.09143
(
2 2 2
  
  )
GRR = 0.302373 27.9 7.8
2
σ = 1.086446
(Part)
PV = 1.042327 96.0 92.2
Total Variation TV =1.085 100.0
Table III-B 8: ANOVA Analysis % Variation & Contribution
(Estimate of variance is based on model without interaction)

128
 
1.04233
1.41 4.861 5
.30237
= = 
ndc
2 2
( )
Total Variation TV GRR PV
= 
( )
( )
% 100
components
total
of TotalVariation
σ
σ
 
=  
 
 
2
( )
2
( )
% ( ) 100
components
total
Contribution to Total Variance
σ
σ
 
=  
 
Analysis of GRR Studies
Both the Average and Range method and ANOVA method will provide
information concerning the causes of measurement system or gage variation.
For example, if repeatability is large compared to reproducibility, the reasons
may be:
• The instrument needs maintenance.
• The gage may need to be redesigned to be more rigid.
• The clamping or location for gaging needs to be improved.
• There is excessive within-part variation.
If reproducibility is large compared to repeatability, then possible causes
could be:
• The appraiser needs to be better trained in how to use and read
the gage instrument.
• Calibrations on the gage dial are not clear.
A fixture of some sort may be needed to help the appraiser use the gage more
consistently.

129
Method
Lower
90% CLP
55
Std. Dev. Upper
90% CL
% of Total
Variation
GRR*
EV 0.175 .202 0.240 17.6
AV 0.133 .230 1.016 20.1
INTERACTION -- na -- na
GRR 0.266 .306 0.363 26.7
PV 1.104 96.4
U
ANOVA
EV 0.177 0.200 0.231 18.4
AV 0.129 0.227 1.001 20.9
INTERACTION -- 0 -- 0
GRR 0.237 0.302 1.033 27.9
PV 1.042 96.0
* In the average and range method, the interaction component cannot be estimated.
Table III-B 9: Comparison of ANOVA and Average and Range Methods
Part No. & Name:
Characteristics:
Specifications:
Gage Name:
Gage No:
Gage Type:
Date:
Performed by:
STD. DEV. % TOTAL
VARIATION
PERCENT
CONTRIBUTION
Repeatability (EV) 0.200 18.4 3.4
Reproducibility (AV) 0.227 20.9 4.4
Appraiser by Part (INT) 0 0 0
GRR 0.302 27.9 7.9
Part (PV) 1.042 96.0 92.2
Measurement System is acceptable for Process Control and Analysis.
Note:
Tolerance = N.A.
Number of distinct data categories (ndc) = 4
Total variation (TV) = 1.085
Table III-B 10: GRR ANOVA Method Report
55
CL = Confidence Limit

130

Chapter III – Section C
Attribute Measurement Systems Study
131
Section C
Attribute measurement systems are the class of measurement systems where
the measurement value is one of a finite number of categories. This is
contrasted to the variables measurement system which can result in a
continuum of values. The most common of these is a go/no-go gage which
has only two possible results. Other attribute systems, for example visual
standards, may result in five to seven classifications, such as very good,
good, fair, poor, very poor. The analyses described in the preceding chapters
cannot be used to evaluate such systems.
As discussed in Chapter I, Section B, there is a quantifiable risk when using
any measurement systems in making decisions. Since the largest risk is at the
category boundaries, the most appropriate analysis would be the
quantification of the measurement system variation with a gage performance
curve. (See Chapter IV Section F)
Risk Analysis Methods
In some attribute situations it is not feasible to get sufficient parts with
variable reference values. In such cases, the risks of making wrong or
inconsistentP
56
F decisions can be evaluated by using:
• Hypothesis Test Analyses
• Signal Detection Theory
Since these methods do not quantify the measurement system variability,
they should be used only with the consent of the customer. Selection and use
of such techniques should be based on good statistical practices, an
understanding of the potential sources of variation which can affect the
product and measurement processes, and the effect of an incorrect decision
on the remaining processes and the final customer.
The sources of variation of attribute systems should be minimized by using
the results of human factors and ergonomic research.
Scenario
The production process is in statistical control and has the performance
indices of Pp = Ppk = 0.5 which is unacceptable. Because the process is
producing nonconforming product, a containment action is required to cull
the unacceptable parts from the production stream.
56
This includes the comparison of multiple appraisers.
Possible
Approaches

132
Figure III-C 1: Example Process with Pp = Ppk = 0.50
For the containment activity the process team selected an attribute gage that
compares each part to a specific set of limits and accepts the part if the limits
are satisfied; otherwise it rejects the part (known as a go/no-go gage). Most
gages of this type are set up to accept and reject based on a set of master
parts. Unlike a variable gage, this attribute gage cannot indicate how good or
how bad a part is, but only that the part is accepted or rejected (i.e., 2
categories). As for all gages, this attribute gage will have “Gray” areas where
wrong decisions can be made (see Figure III-C 2 below and Chapter II,
Section B).
Figure III-C 2: The “Gray” Areas Associated with the Measurement System
Since this has not yet been documented by the team, it needs to study the
measurement system. However, to address the areas of risk around the
specification limits, the team chose approximately 25% of the parts at or
close to the lower specification limit and 25% of the parts at or close to the
upper specification limit. In some cases where it is difficult to make such
parts the team may decide to use a lower percentage recognizing that this
may increase the variability of the results. If it is not possible to make parts
close to the specification limits the team should reconsider the use of
attribute gaging for this process. As appropriate for each characteristic, the
parts should be independently measured with a variable gage with acceptable
variation (e.g., a CMM). When measuring a true attribute that cannot be
LSL=.45 USL=.55
0.50 0.60
0.40

133
measured with a variable gauge use other means such as experts to pre-
determine which samples are good or defective.
Three appraisers are used, with each appraiser making three decisions on the
each part.
An acceptable decision was designated with a one (1) and an unacceptable
decision with zero (0). The reference decision and variable reference values
shown in Table III-C 1 were not initially determined. The table also indicates
in the “Coded” column whether the part is area I, area II, or area III part
indicated by “– ”, “x”, and “+” respectively.
Hypothesis Test Analyses – Cross-Tab Method
Since the team did not know the reference decisions for the parts, they
developed cross-tabulations comparing each of the appraisers to the other.
The cross-tabulation process analyzes distribution data for two or more
categorical variables. The results – presented in a matrix format – form a
contingency table that illustrates the interdependency between variables.P
57
F
57
Cross-tabulation is available in many statistical analysis software packages and is used in spreadsheet pivot
table functions.

134
Part A - 1 A - 2 A - 3 B - 1 B - 2 B - 3 C - 1 C - 2 C - 3 Reference Ref Value Code
1 1 1 1 1 1 1 1 1 1 1 0.476901 +
2 1 1 1 1 1 1 1 1 1 1 0.509015 +
3 0 0 0 0 0 0 0 0 0 0 0.576459 –
4 0 0 0 0 0 0 0 0 0 0 0.566152 –
5 0 0 0 0 0 0 0 0 0 0 0.570360 –
6 1 1 0 1 1 0 1 0 0 1 0.544951 x
7 1 1 1 1 1 1 1 0 1 1 0.465454 x
8 1 1 1 1 1 1 1 1 1 1 0.502295 +
9 0 0 0 0 0 0 0 0 0 0 0.437817 –
10 1 1 1 1 1 1 1 1 1 1 0.515573 +
11 1 1 1 1 1 1 1 1 1 1 0.488905 +
12 0 0 0 0 0 0 0 1 0 0 0.559918 x
13 1 1 1 1 1 1 1 1 1 1 0.542704 +
14 1 1 0 1 1 1 1 0 0 1 0.454518 x
15 1 1 1 1 1 1 1 1 1 1 0.517377 +
16 1 1 1 1 1 1 1 1 1 1 0.531939 +
17 1 1 1 1 1 1 1 1 1 1 0.519694 +
18 1 1 1 1 1 1 1 1 1 1 0.484167 +
19 1 1 1 1 1 1 1 1 1 1 0.520496 +
20 1 1 1 1 1 1 1 1 1 1 0.477236 +
21 1 1 0 1 0 1 0 1 0 1 0.452310 x
22 0 0 1 0 1 0 1 1 0 0 0.545604 x
23 1 1 1 1 1 1 1 1 1 1 0.529065 +
24 1 1 1 1 1 1 1 1 1 1 0.514192 +
25 0 0 0 0 0 0 0 0 0 0 0.599581 –
26 0 1 0 0 0 0 0 0 1 0 0.547204 x
27 1 1 1 1 1 1 1 1 1 1 0.502436 +
28 1 1 1 1 1 1 1 1 1 1 0.521642 +
29 1 1 1 1 1 1 1 1 1 1 0.523754 +
30 0 0 0 0 0 1 0 0 0 0 0.561457 x
31 1 1 1 1 1 1 1 1 1 1 0.503091 +
32 1 1 1 1 1 1 1 1 1 1 0.505850 +
33 1 1 1 1 1 1 1 1 1 1 0.487613 +
34 0 0 1 0 0 1 0 1 1 0 0.449696 x
35 1 1 1 1 1 1 1 1 1 1 0.498698 +
36 1 1 0 1 1 1 1 0 1 1 0.543077 x
37 0 0 0 0 0 0 0 0 0 0 0.409238 –
38 1 1 1 1 1 1 1 1 1 1 0.488184 +
39 0 0 0 0 0 0 0 0 0 0 0.427687 –
40 1 1 1 1 1 1 1 1 1 1 0.501132 +
41 1 1 1 1 1 1 1 1 1 1 0.513779 +
42 0 0 0 0 0 0 0 0 0 0 0.566575 –
43 1 0 1 1 1 1 1 1 0 1 0.462410 x
44 1 1 1 1 1 1 1 1 1 1 0.470832 +
45 0 0 0 0 0 0 0 0 0 0 0.412453 –
46 1 1 1 1 1 1 1 1 1 1 0.493441 +
47 1 1 1 1 1 1 1 1 1 1 0.486379 +
48 0 0 0 0 0 0 0 0 0 0 0.587893 –
49 1 1 1 1 1 1 1 1 1 1 0.483803 +
50 0 0 0 0 0 0 0 0 0 0 0.446697 –
Table III-C 1: Attribute Study Data Set

135
The first step is to summarize the observed data. Reviewing Table III-C 1,
the team examined the data for pairs of observers, counting when they agreed
and when they disagreed for each set of evaluations. That is, for the
evaluations, there are 34 times where A-1 = 1 and B-1 = 1; there are 32 times
where A-2 = 1 and B-2 = 1; and there are 31 times where A-3 = 1 and B-3 =
1for a total agreement of 97. The table below was constructed to summarize
the data distribution for the observer pair A*B. Similar tables were prepared
for observer pairs B*C and A*C.
B
.00 1.00 Total
A .00 44 6 50
(agree) (disagree)
1.00 3 97 100
(disagree) (agree)
Total 47 103 150
The second step is to estimate the expected data distribution. What is the
probability that an observer pair will agree or disagree on an observation
purely by chance? In 150 observations Observer A rejected the part 50 times
and Observer B rejected the part 47 times:
pA0 = 47/150 = 0.313
pB0 = 50/150 = 0.333
Since the two observers are independent, the probability that they agree that
the part is bad is given by:
p(A0 ∩ B0) = pA0 pB0 = 0.104
The expected number of times for Observer A and Observer B agree the part
is bad is estimated by multiplying the combined probability by the number of
observations:
150 x (pA0 pB0) = 150 x (47/150) x (50/150) = 15.7
The team made similar estimations of each category pair for each observer
pair to complete the following tables:

136
Table III-C 2: Cross tabulation Study Results
To determine the level of this agreement the team uses the (Cohen's) kappa
which measures the agreement between the evaluations of two raters when
both are rating the same object. A value of 1 indicates perfect agreement. A
value of 0 indicates that agreement is no better than chance. Kappa is only
available for tables in which both variables use the same category values and
both variables have the same number of categories.P
58
58
There are a number of statistics which can be used to determine inter-rater agreement. Different statistics are
appropriate for different types of measurement. See Reference List including: Bland, J. M., and Altman, D. G.
(1986); Cohen, J. (1960); Everitt, B. (1996); Fleiss, J. L. (1971); Krippendorff, K. (2004); Saal, F.E., Downey,
R.G. and Lahey, M.A. (1980); Shrout, P. and Fleiss, J. L. (1979); and Uebersax, John S. (1987)
A * B Crosstabulation
44 6 50
15.7 34.3 50.0
3 97 100
31.3 68.7 100.0
47 103 150
47.0 103.0 150.0
Count
Expected Count
Count
Expected Count
Count
Expected Count
.00
1.00
A
Total
.00 1.00
B
Total
B * C Crosstabulation
42 5 47
16.0 31.0 47.0
9 94 103
35.0 68.0 103.0
51 99 150
51.0 99.0 150.0
Count
Expected Count
Count
Expected Count
Count
Expected Count
.00
1.00
B
Total
.00 1.00
C
Total
A * C Crosstabulation
43 7 50
17.0 33.0 50.0
8 92 100
34.0 66.0 100.0
51 99 150
51.0 99.0 150.0
Count
Expected Count
Count
Expected Count
Count
Expected Count
.00
1.00
A
Total
.00 1.00
C
Total

137
Kappa is a measure of interrater agreement that tests if the counts in the
diagonal cells (the parts that receive the same rating) differ from those
expected by chance alone.
Let po = the sum of the observed proportions in the diagonal cells
pe = the sum of the expected proportion in the diagonal cells
then
1
o e
e
p p
kappa
p
−
=
−
Kappa is a measure rather than a test.P
59
F Its size is judged by using an
asymptotic standard error to construct a t statistic. A general rule of thumb is
that values of kappa greater than 0.75 indicate good to excellent agreement
(with a maximum kappa = 1); values less than 0.40 indicate poor agreement.
Kappa takes no account of the size of disagreement between the
raters, but only whether they agree or not.P
60
F
Upon calculating the kappa measures for the appraisers, the team came up
with the following:
Kappa A B C
A – .86 .78
B .86 – .79
C .78 .79 –
Table III-C 3: Kappa Summary
This analysis indicates that all the appraisers show good agreement between
each other.
This analysis is necessary to determine if there are any differences among the
appraisers but it does not tell us how well the measurement system sorts good
parts from bad. For this analysis the team had the parts evaluated using a
variable measurement system and used the results to determine the reference
decision.
With this new information another group of cross-tabulations was developed
comparing each appraiser to the reference decision.
59
As in all such categorical evaluations, a large number of parts covering the entire spectrum of possibilities is
necessary.
60
When the observations are measured on an ordinal categorical scale a weighted kappa can be used to better
measure agreement. Agreement between the two raters is treated as for kappa but disagreements are measured
by the number of categories by which the raters differ.

138
Table III-C 4: Comparisons of Appraisers to Reference
The team also calculated the kappa measure to determine the agreement of
each appraiser to the reference decision:
A B C
kappa .88 .92 .77
These values can be interpreted as each of the appraisers has good agreement
with the standard.
A * REF Crosstabulation
45 5 50
16.0 34.0 50.0
3 97 100
32.0 68.0 100.0
48 102 150
48.0 102.0 150.0
Count
Expected Count
Count
Expected Count
Count
Expected Count
.00
1.00
A
Total
.00 1.00
REF
Total
B * REF Crosstabulation
45 2 47
15.0 32.0 47.0
3 100 103
33.0 70.0 103.0
48 102 150
48.0 102.0 150.0
Count
Expected Count
Count
Expected Count
Count
Expected Count
.00
1.00
B
Total
.00 1.00
REF
Total
C * REF Crosstabulation
42 9 51
16.3 34.7 51.0
6 93 99
31.7 67.3 99.0
48 102 150
48.0 102.0 150.0
Count
Expected Count
Count
Expected Count
Count
Expected Count
.00
1.00
C
Total
.00 1.00
REF
Total

139
The process team then calculated the effectiveness of the measurement
system.
number of correct decisions
Effectiveness
total opportunities for a decision
=
% Appraiser1
% Score vs. Attribute2
Source Appraiser A Appraiser B Appraiser C Appraiser A Appraiser B Appraiser C
Total Inspected 50 50 50 50 50 50
# Matched 42 45 40 42 45 40
False Negative (appraiser biased toward rejection) 0 0 0
False Positive (appraiser biased toward acceptance) 0 0 0
Mixed 8 5 10
95% UCI 93% 97% 90% 93% 97% 90%
Calculated Score 84% 90% 80% 84% 90% 80%
95% LCI 71% 78% 66% 71% 78% 66%
System % Effective Score3
System % Effective Score vs.
Reference4
Total Inspected 50 50
# in Agreement 39 39
95% UCI 89% 89%
Calculated Score 78% 78%
95% LCI 64% 64%
Notes
(1)Appraiser agrees with him/herself on all trials
(2)Appraiser agrees on all trials with the known standard
(3)All appraisers agreed within and between themselves
(4)
(5)
All appraisers agreed within & between themselves AND agreed with the reference
UCI and LCI are the upper and lower confidence interval bounds, respectively
Table III-C 5: Study Effectiveness Table
Multiple tests of hypothesis between each pair of appraisers can be
conducted with the null hypothesis:
H0: The effectiveness of both appraisers is the same
Since the calculated score of each appraiser falls within the confidence
interval of the other, the team concludes that they cannot reject the null
hypotheses. This reinforces the conclusions from the kappa measures.
For further analysis, one of the team members brought out the following
table which provides guidelines for each appraiser’s results:

140
Decision
Measurement system
Effectiveness Miss Rate False Alarm
Rate
Acceptable for the appraiser  90%  2%  5%
Marginally acceptable for the
appraiser – may need
improvement
80%  5%  10%
Unacceptable for the appraiser
– needs improvement < 80% > 5% > 10%
Table III-C 6: Example Effectiveness Criteria Guidelines
Summarizing all the information they already had, the team came up with
this table:
Effectiveness Miss Rate False Alarm
Rate
A 84% 6.3% 4.9%
B 90% 6.3% 2.0%
C 80% 12.5% 8.8%
Table III-C 7: Study Effectiveness Summary
These results showed that the measurement system had different levels of
performance in Effectiveness, Miss Rate and False Alarm Rate depending on
the appraiser. No single appraiser had acceptable results in all three
categories. No single appraiser had unacceptable results in all three
categories. Do the acceptance guidelines need to be changed for this
process? Are these risks acceptable? Do the appraisers need better training?
Could the testing environment be improved? Most importantly, what did the
customer think about the measurement system and these results – what were
their expectations?
Does the customer accept these risk levels?
Sample Size
The question always arises: “How many samples should be used in the
study?” To the dismay of most people, the answer is “enough”. The purpose
of any measurement study (variables or attribute) is to understand the
properties of the measurement system.. A sufficient number of samples
should be selected to cover the expected operating range (see also Chapter II
Section C). With attribute measurement systems, the area of interest are the

141
Type II areas (see Chapter I, Section B). If the inherent process capability is
good (i.e., large Cp, Cpk or Pp, Ppk) then a small random sample may not
have many (or any) samples in this area. This means that as the process
capability improves , the required random sample for the attribute study
should become larger).
In the example above, the indices were Pp, Ppk = 0.5 (i.e. an expected process
performance of approximately 13% nonconformance), the sample selected
was 50.
An alternate approach to large samples is a “salted sample” where parts are
selected specifically from the Type II areas to augment a random sample to
ensure that the effect of appraiser variability is seen.
Concerns
1) There are no theory-based decision criteria on acceptable risk. The
above guidelines are heuristic and developed based on individual
“beliefs” about what will pass as “acceptable”. The final decision
criteria should be based on the impact (i.e., risk) to the remaining
process and final customer. This is a subject matter decision – not a
statistical one.
2) The analysis above is data dependent. For example, if the process
indices were Pp = Ppk = 1.33, then all the decisions would be correct
since no parts would fall in the region II (the “gray” areas) of the
measurement system.
Figure III-C 3: Example Process with Pp = Ppk = 1.33
With this new situation, it would be concluded that all the
appraisers were acceptable since there would be no decision
errors.

142
3) There is usually a misconception of what the cross-tabulation results
really mean. For example, appraiser B results from Table III-C 4 are:
Since the purpose of the inspection is to find all the nonconforming
parts, most people look at the upper left corner as a measure of the
effectiveness of finding bad parts. This percent is the probability of
saying a part is bad given that it is bad:
Pr( )
calling the part bad abad part
Assuming that the process is improved to Pp = Ppk = 1.00, the
probability of interest to the producer is:
 
Pr the part isbad it iscalled bad
To determine this from the above data, Bayes’ Theorem must be
used.
 
   
       
Pr *Pr
Pr
Pr *Pr Pr *Pr
called bad bad bad
bad called bad
called bad bad bad called bad good good
=

 
 
   
.938* .0027
Pr
.938* .0027 .020* .9973
bad called bad =

 
Pr .11
bad called bad =
That is, these results indicate that if the part is called bad there is only
a 1 out of 10 chance that it is truly bad.
4) The analysis does not use the variable data information or even the
relative order information which was available when the reference
decision values were determined.
B * REF Crosstabulation
45 2 47
93.8% 2.0% 31.3%
3 100 103
6.3% 98.0% 68.7%
48 102 150
100.0% 100.0% 100.0%
Count
% within REF
Count
% within REF
Count
% within REF
.00
1.00
B
Total
.00 1.00
REF
Total

143
Signal Detection Approach
An alternate approach is to use Signal Detection
theoryP
61
F to determine an approximation of the width of
the region II area and from this, the measurement
system GRR. This requires that each of the sample
parts can be evaluated offline by a variables
measurement system. This reference value is shown in
the column Ref Value in Table III-C 1.
Steps:
1. Determine the Tolerance (specification range); from
Figure III-C 1:
USL = .550
LSL = .450
Then the Tolerance = USL – LSL = .100 . This value
will be used to calculate the GRR.
Guidelines:
• If Ppk>1, compare measurement system to process.
• If Ppk<1, compare measurement system to tolerance.
This “rule” amounts to comparing the measurement system
to whichever is most restrictive, the process or the tolerance.
In this example, the Ppk = .5 (see Figure III-C 1), so the
process is greater than the tolerance and this measurement
system should therefore be compared to tolerance.
Table III-C 8: Table III-C 1 sorted by Ref Value
For the data in Table III-C 1.
2) Rank order the data (and adjacent cells) from highest to lowest based
on the individual Reference Values (see Table III-C 8; Note: this table
has been divided into two columns to save space.).
3) Identify the beginning and end points of the two area IIs. In Table
III-C 8 this is shown by the column Code:
+ = accepted with total agreement
– = rejected with total agreement (in Table III-C 1)
x = disagreement
The width of these zones is what we are trying to determine, and the
average width of these zones will be used to compare the
61
See Reference List: Baker 1975
Ref Value Code Ref Value Code
0.599581 - 0.503091 +
0.587893 - 0.502436 +
0.576459 - 0.502295 +
0.570360 - 0.501132 +
0.566575 - 0.498698 +
0.566152 - 0.493441 +
0.561457 x 0.488905 +
0.559918 x 0.488184 +
0.547204 x 0.487613 +
0.545604 x 0.486379 +
0.544951 x 0.484167 +
0.543077 x 0.483803 +
0.542704 + 0.477236 +
0.531939 + 0.476901 +
0.529065 + 0.470832 +
0.523754 + 0.465454 x
0.521642 + 0.462410 x
0.520496 + 0.454518 x
0.519694 + 0.452310 x
0.517377 + 0.449696 x
0.515573 + 0.446697 -
0.514192 + 0.437817 -
0.513779 + 0.427687 -
0.509015 + 0.412453 -
0.505850 + 0.409238 -
dUSL
dLSL

144
measurement system to the specification tolerance, or to the process
6 sigma width (depending on the situation).
4) Referring to the above graphic, let dUSL = distance between the last
part accepted by all appraisers to the first part rejected by all. This
distance is equivalent to the gray zone II surrounding the USL above,
bounded by the upper zone I to the right, and Zone III to the left.
Note the values circled in Table III-C 8 above.
5) Let dLSL = distance between the last part accepted by all
appraisers in Zone III to the first part rejected by all appraisers
in Zone I.
Let di = distance between the last part accepted by all appraisers to the first
part rejected by all (for each specification).
Then,
d = average (di)
is an estimateP
62
F of the width of region II areas and, thus, an estimate of the
6*σ
= GRR
GRR
In this example (p. 126) where the tolerance is 0.100,
dLSL = 0.470832 – 0.446697 = 0.024135
dUSL = 0.566152 – 0.542704 = 0.023448
d = 0.0237915
or the estimated %GRR is,
%GRR = 24%
Since this example was generated by a measurement system with an actual
%GRR = 25%, this estimate will lead to the same evaluation of the
measurement system.
If only ordered data information is available, this technique can still be used
but requires subject matter knowledge to quantify the d’s.
62
The “goodness” of this estimate depends on the sample size and how close the sample represents the process.
The larger the sample, the better the estimate.
dUSL
dLSL
Zone I = parts agreed by all
appraisers to be rejected.
Zone III = parts agreed by all
appraisers to be accepted.
Zone II = questionable parts
without 100% agreement,
surrounding each specification
limit.

145
Analytic MethodP
63
As with any measurement system, the stability of the process should be
verified and, if necessary, monitored. For attribute measurement systems,
attribute control charting of a constant sample over time is a common way of
verifying stabilityP
64
F.
For an attribute measurement system, the concept of the Gage Performance
Curve (see Chapter IV, Section F) is used for developing a measurement
system study, which is used to assess the amount of repeatability and bias of
the measurement system. This analysis can be used on both single and double
limit measurement systems. For a double limit measurement system, only
one limit need be examined with the assumptions of linearity and uniformity
of error. For convenience, the lower limit will be used for discussion.
In general, the attribute measurement system study consists of obtaining the
reference values for several selected parts. These parts are evaluated a
number of times, (m), with the total number of accepts (a), for each part
being recorded. From the results, repeatability and bias can be assessed.
The first stage of the attribute study is the part selection. It is essential that
the reference value be known for each part used in the study. Eight parts
should be selected at as nearly equidistant intervals as practical. The
maximum and minimum values should represent the process range. Although
this selection does not affect confidence in the results, it does affect the total
number of parts needed to complete the gage study. The eight parts must be
run through the gage, m = 20 times, and the number of accepts, (a), recorded.
For the total study, the smallest part must have the value a = 0; the largest
part, a = 20; and the six other parts, a 19. If these criteria are not
satisfied, more parts with known reference values, (X), must be run through
the gage until the above conditions are met. If, for the smallest value a 0,
then smaller and smaller parts are taken and evaluated until a = 0. If, for the
largest value a 20, then larger and larger parts are taken until a = 20. If six
of the parts do not have 1 a 19, additional parts can be taken at selected
points throughout the range. These points are taken at the midpoints of the
part measurements already measured in the study. The first interval at the a =
0 end starts from the largest measurement where a = 0. For the a = 20 end,
the first interval starts at the smallest measurement where a = 20. For best
results, samples should be taken at both the a = 0 and a = 20 ends and
worked toward the middle of the part range. If necessary, the procedure can
be repeated until the criteria are met.
Once the data collection criteria have been satisfied, the probabilities of
acceptance must be calculated for each part using the following equations:
63
Adapted with permission from “Analysis of Attribute Gage Systems” by J. McCaslin & G. Gruska, ASQC,
1976.
64
Caveat: np > 4

146









=


−



=
5
.
0
5
.
0
20
5
.
0
5
.
0
0
,
5
.
0
5
.
0
'
m
a
if
a
m
a
if
m
a
a
m
a
if
m
a
Pa
The adjustments cover the instances where 1 a 19. For the instances
where a = 0 set a
P = 0 except for the largest reference value with a = 0, in
which a
P = 0.025. For the instances where a = 20 then a
P = 1 except for the
smallest reference value with a = 20 in which a
P = 0.975.
Once the a
P has been calculated for each XT , the Gage Performance Curve
(GPC) can be developed. Although the GPC can be presented graphically
(see Figure III-C 5), use of normal probability paper (see Figure III-C 4)
yields more accurate estimates of the repeatability and bias.
The calculated probabilities are plotted on normal probability paper and a
line of best fit is drawn through these points. The bias is equal to the lower
limit minus the reference value measurement that corresponds to a
P = 0.5,
or
bias = Lower Specification Limit – XT (at a
P = 0.5)
The repeatability is determined by finding the differences of the reference
value measurements corresponding a
P = 0.995 and a
P =0.005 and dividing
by an adjustment factor of 1.08P
65
F.
1.08
' '
T a T a
X (at P 0.995) X ( at P 0.005)
Repeatability
= − =
=
To determine if the bias is significantly different from zero, the following
statistic is used:
σ

=
repeatability
6.078 Bias
t
If this calculated value is greater than 2.093 (t025,19), then the bias is
significantly different from zero.
An example will clarify the attribute study data collection and calculation of
repeatability and bias.
65
The 1.08 adjustment (unbiasing) factor is specific for a sample size of 20 and was determined through a
simulation of this approach with a 99% repeatability range. To convert the result to a 6 sigma range, divide by
5.15 and multiply by 6.

147
Example:
An attribute gage is being used to measure a dimension that has a tolerance
of 0.010. The gage is an end-of-line 100% automatic inspection gage that is
affected by repeatability and bias. To perform the attribute study, eight parts
with reference values at intervals of 0.002 from –0.016 to –0.002 are run
through the gage 20 times each. The number of accepts for each part are:
XT a
–0.016 0
–0.014 3
–0.012 8
–0.010 20
–0.008 20
–0.006 20
–0.004 20
-0.002 20
Since there are two reference values with 1 a 19, at least four more parts
must be found. Therefore, it is necessary to run parts with reference values at
the midpoints of the existing intervals. These reference values and the
number of accepts are:
–0.015 1
–0.013 5
-0.011 16
Now there are five reference values with 1 a 19. The procedure requires
that one more part be found with 1 a 19. Therefore, the following part is
evaluated:
–0.0105 18
Now that the data collection criteria have been satisfied, the probabilities of
acceptance can be calculated using the binomial adjustments shown below.
XT a
a
P
-0.016 0 0.025
-0.015 1 0.075
-0.014 3 0.175
-0.013 5 0.275
-0.012 8 0.425
-0.011 16 0.775
-0.0105 18 0.875
-0.010 20 0.975
-0.008 20 1.000

148
These probabilities are plotted on normal probability paper as shown in
Figure III-C 4. By plotting a best-fit line through these points, the bias and
repeatability can be determined. The bias is equal to the lower limit minus
the reference value measurement that corresponds to Pa’ = 0.5.
From Figure III-C 4:
-0.010 - (-0.0123) 0.0023
bias = =
The repeatability is determined by finding the differences R of the reference
value measurements corresponding to a
P = 0.995 and a
P = 0.005 and
dividing by 1.08 From Figure III-C 4:
0.0084 ( 0.0163 )
1.08
R =
0.0079
0.0073
1.08
= =
Then
5.15
σ =
repeatability
R
= 0.00142 and the associated GRR range is 0.0085
To determine if the bias is significantly different from zero, calculate:
6.078 Bias
σ

=
repeatability
t
=
6.078 0.0023
9.84
0.00142

=
Since t0.025,19 = 2.093, the bias is significantly different from zero.
Like the variable Gage Performance Curve shown in Chapter IV, Section F,
the attribute GPC can also be plotted on plain graph paper (see Figure III-C
5). This can be accomplished in one of two ways. One approach would be to
run the sample study at the other limit of the specification. In the example,
the long method for attribute study would also have to be conducted at the
high limit of the specification, and the calculated values plotted accordingly.
However, because of the previously mentioned assumptions, it would not be
necessary to run the study again. Since the shape of the curve at the high
limit should be a “mirror image” of the curve at the low limit, the only
consideration necessary is the location of the curve with respect to XT values.
This location is determined by the bias.
The correct position of the curve would be defined at the point where the a
P
= 0.5, and the XT value is equal to the specification limit minus the bias. In
the example, this point would be XT = 0.010 – 0.0023 = 0.0077. The GPC
plotted in this manner is shown in Figure III-C 4.

149
Figure III-C 4: Attribute Gage Performance Curve Plotted on Normal Probability Paper

150
Figure III-C 5. Attribute Gage Performance Curve

Chapter IV
Other Measurement Concepts and Practices
151
CHAPTER IV

Chapter IV
152

Chapter IV – Section A
Practices for Complex or Non-Replicable Measurement Systems
153
Section A
Practices for Non-Replicable Measurement Systems
The focus of this reference manual is measurement systems where the
readings can be replicated for each part. Not all measurement systems have
this characteristic; for example:
• Destructive measurement systems
• Systems where the part changes on use/test.
The following are examples of approaches to the analysis of
measurement systems, including those not previously discussed in
this manual. This is not intended to be a complete listing covering
every type of measurement system but only examples of various
approaches. If a measurement system does not fit the manual’s
focus, it is recommended that assistance from competent statistical
resources be consulted.
Destructive measurement systems
When the part (characteristic) being measured is destroyed by the act of
measuring the process is known as destructive measurement. This includes
the whole class of measurement systems known as “destructive measurement
systems”; for example destructive weld testing, destructive plating testing,
salt spray/humidity booth testing, impact testing (gravelometer) or mass
spectroscopy and other material characteristic testing processes.
These are the “classic” examples of a non-replicable measurement system
since repeated readings cannot be taken on any single part.
Systems where the part changes on use/test
However, there are other measurement systems which are non-replicable
where the part, itself, is not harmed by the measurement process but the
characteristic being measure will change. Examples of this are: leak tests
with qualitative data, testing using engine test stands, transmission test
stands, vehicle dynamometers, etc.
Analysis of these systems will depend on whether
1) A homogeneous set of parts (small between part variation) can be
found to represent a single part;
2) The shelf life of the characteristic (property) is known and
extends beyond the expected duration of the study – i.e., the
measured characteristic does not change over the expected period
of use; or
3) The dynamic (changing) properties can be stabilized

Chapter IV – Section A
Practices for Complex or Non-Replicable Measurement Systems
154
The mapping of the studies described in this chapter and the various
scenarios are as follows:
Stability Studies
Scenario S1 S2 S3 S4 S5
The part is not changed by the measurement process; i.e.,
measurement systems that are non-destructive (replicable) and will
be used with parts (specimens) with:
• Static properties, or
• Dynamic (changing) properties which have been stabilized.
 
The shelf life of the characteristic (property) is known and extends
beyond the expected duration of the study; i.e., the measured
characteristic does not change over the expected period of use.
 
Destructive measurement systems  
Other non-replicable measurement systems.  
Measurement systems with dynamic characteristics: e.g., test
stands

Variability Studies
Scenario V1 V2 V3 V4 V5 V6 V7 V8 V9
The part is not changed by the measurement process; i.e.,
measurement systems that are non-destructive and will be
used with parts (specimens) with:
• Static properties, or
• Dynamic (changing) properties which have been
stabilized.

Above with p  2 instruments 
Destructive measurement systems  
Other non-replicable measurement systems.  
Measurement systems with dynamic characteristics: e.g.,
test stands
     
Measurement systems with dynamic characteristics:
with p  3 instruments

Table IV-A 1: Methods Based on Type of Measurement System

Chapter IV – Section B
Stability Studies
155
Section B
Stability Studies
S1: Single PartP
66F, Single Measurement Per Cycle
Application:
a) Measurement systems in which the part is not changed by the
measurement process; i.e., measurement systems that are non-
destructive and will be used with parts (specimens) with:
 Static properties, or
 Dynamic (changing) properties which have been stabilized.
b) The shelf life of the characteristic (property) is known and extends
Assumptions:
• The measurement system is known (documented) to have a linear
response over the expected range of the characteristic (property).
• Parts (specimens) cover the expected range of the process variation
of the characteristic.
Analyze using X & mR charts:
• Determine measurement system stability:
 Compare the plotted points to the control limits
 Look for trends (x chart only)
• Compare *
2
e
R
d
σ = (total measurement error) with the
repeatability estimate E
σ from a variability study (see next section).
• Determine the bias if reference value is known:
= −
66
A reference standard can be used if it is appropriate for the process.

Stability Studies
156
S2: n 3 PartsP
67F, Single Measurement Per Cycle Per Part
Application:
 Static properties, or
 Dynamic (changing) properties which have been stabilized.
b) The shelf life of the characteristic (property) is known and extends
Assumptions:
• The measurement system is known (documented) to have a linear
response over the expected range of the characteristic (property).
Analyze using a [z, R ] chart: where -
i i i
z x 
=
and i
 is the (reference) standard value or determined by the
average of a large number of successive readings of the part
(specimen).
• Determine measurement system stability:
 Compare the plotted points to the control limits
 Look for trends (z chart only)
• CompareP
68
F
e *
2
R
d
σ = with the repeatability estimate E
σ from a
variability study.
• Determine the bias if reference values are known:
= −
• Determine the linearity if n 3 parts were used:
 The parts (specimens) must cover the expected range of the
property
 Each part (specimen) should be analyzed separately for bias
and repeatability
 Quantify the linearity using the linearity analysis discussed in
Chapter 3, Section B
67
A reference standard can be used if it is appropriate for the process.
68
If more than one appraiser is involved in the data collection, then E
σ (repeatability estimate) is affected also by
the reproducibility of the measurement system. Quantify reproducibility by scatter and whisker plots indexed by
appraiser (see Chapter III, Section B).

Stability Studies
157
If more than one instrument is use in this study, determine consistency
(homogeneity of variation) among the instruments; e.g., use F test, Bartlett’s
test, Levene’s testP
69
F, etc.
S3: Large Sample from a Stable Process
Application:
The measurement system must be evaluating a homogeneous independent
identically distributed (“iid”) sample (collected and isolated). The
measurements of individual parts (specimens) are not replicated so this study
can be used with destructive and non-replicable measurement systems.
Assumptions:
• The shelf life of the characteristic (property) is known and extends
beyond the expected duration of the study; i.e., measured
characteristic does not change over the expected period of use and/or
storage.
of the characteristic (property).
• The measurement system’s linearity is known (documented) over the
expected range of the characteristic (property). (If the response is
non-linear, the readings must be adjusted accordingly.)
Analyze by:
• Determining the total variability via a capability study with n 30
parts. (This preliminary study should also be used to verify the
consistency of the sample: i.e., all parts (specimens) come from a
unimodal distribution.)
• 2 2 2
total process measurement system
σ σ σ
= 
• Measuring one or more individuals from the isolated sample per time
period, use x & R or x & mR charts with the control limits
determined by the capability study.
• Compare the plotted points to the control limits.
• Look for trends.
• Since the parts (specimens) do not change (an isolated sample), any
indication of instability would be attributed to a change in the
measurement system.
69
Dataplot, National Institute of Standards and Technology, Statistical Engineering Division (www.itl.nist.gov).

Stability Studies
158
S4: Split Specimens (General), Single Specimen Per Cycle
Application:
The measurements of individual parts (specimens) portions are not
replicated so this study can be used with destructive and non-replicable
measurement systems.
Assumptions:
storage.
• Parts (specimens) cover the extended range of the process variation
• Specimens are split into m portions. With m=2 portions, this is often
called a test-retest study.
Analyze using:
• Range chart to track the consistency of the measurements
(confounded with the “within-lot” consistency).
• Compare *
2
=
e
R
d
σ with the repeatability estimate E
σ from a
variability study.
• This is an upper bound study: 2 2 2
= +
e E btwn
σ σ σ
• Chart to track the consistency of the production process.
S4 with Pairs of Consecutive (Homogeneous) Parts from Different Lots – S4a
This study is the same as S4 with homogeneous parts from different lots. It is
an upper bound study, since 2 2 2 2
e E btwn lots
σ σ σ σ
=  

Stability Studies
159
S5: Test Stands
In this situation, multiple measurement instruments (test stands) evaluate the
same characteristics of a continual stream of product. The product stream is
randomly assigned to individual stands.
S5a: Attribute Responses
Analyze using p charts:
• Determine consistency (of decisions) among the stands: a single
chart including the results from all the stands.
• Determine stability within individual stands: separate chart for each
stand.
Analyze the total system stability with a p & mR chart where p is the
average over all the test stands in a given day.
S5b: Variable Data Responses
Analyze using ANOVA and graphical techniques:P
70
• Calculate x & s for each test stand (by characteristic), by time
period.
• Determine consistency among the stands: a single x & s chart
including the results from all the stands.
• Determine stability within individual stands: separate x & s chart
for each stand.
• Quantify the consistency (homogeneity of variation) among the
stands; e.g., use F test, Bartlett’s test, Levene’s test, etc.
• Determine if all stands are on the same target by comparing stand
averages; e.g., using a one-way ANOVA analysis. If any difference
exists, isolate different stands by using, for example, Tukey’s T test.
70
see also James, P. D., “Graphical Displays of Gage R&R Data,” AQC Transaction, ASQC, 1991

Stability Studies
160

Chapter IV – Section C
Variability Studies
161
Section C
Variability Studies
All descriptive studies are enumerative in nature in that they describe the
measurement system (including the effects of the environment) during the
study. Since measurement systems are to be used in making future decisions
about products, processes, or services, an analytic conclusion about the
measurement system is necessary. The transition from enumerative to
analytic results requires subject matter knowledge and expertise to:
• Assure that all expected measurement sources of variation are
considered in the design and execution of the study.
• Analyze the results (data) in light of the expected uses, environment,
control, maintenance, etc.
V1: Standard GRR Studies
These studies are those contained within this reference manual. These
studies include graphical analysis as well as numerical analysis.
V1a – Range Method (R&R)
V1b – Range Method (R&R and Within-Part)
V1c – ANOVA Method
V1d – Modified ANOVA/Range Method
V2: Multiple Readings With p 2 Instruments
This allows the comparison of multiple instruments.
Application:
1) Static properties, or
2) Dynamic (changing) properties which have been stabilized.
Assumptions:

Variability Studies
162
Analyze using Grubbs’P
71
F (or Thompson’s)P
72
F estimates:
• Process variability
• Instrument variability = repeatability
• Confidence region calculations are available
V3: Split specimens (m = 2)
Application:
The measurements of individual parts (specimens) portions are not replicated
so this study can be used with destructive and non-replicable measurement
systems and can be used to analyze measurement systems with dynamic
characteristics.
Assumptions:
storage.
• Specimens are split into m portions. With m=2 portions, this is often
called a test-retest study.
Analyze using regression techniques:
• Estimate repeatability with the error term: e
E
σ σ
=
• Linearity (by comparing estimated line with 45° line)
V3a – V3 with Pairs of Consecutive Parts
This study is the same as V3 using consecutive pairs of part rather than split
specimens. This study is used in situations where the part cannot be split
without destroying the property to be measured.
This is an upper bound study: E e btwn
σ σ σ
 
71
See Reference List, Grubbs, F. E., 1973.
72
See Reference List, Thompson, W. A., Jr., 1963.

Variability Studies
163
V4: Split Specimens (General)
Application:
The measurements of individual parts (specimens) portions are not replicated
so this study can be used with destructive and non-replicable measurement
systems and can be used to analyze measurement systems with dynamic
characteristics.
Assumptions:
storage.
• Split specimens into m portions where m = 0 mod 2 or 3; m 2 (e.g.,
m =3, 4, 6, 9,.....).
Analyze using:
• Standard GRR techniques including graphics
• ANOVA – Randomized Block Design (two-way ANOVA)
V4a – V4 with Pairs of Consecutive (homogeneous) Parts from
Different Lots
This study is the same as V4 using consecutive pairs of part rather than split
specimens. This study is used in situations where the part cannot be split
without destroying the property to be measured.
This is an upper bound study: E e parts lots
σ σ σ σ
  
The Following Studies Assume the Part (specimen) Characteristic
(property) is Dynamic.
V5: Same as V1 with Stabilized Parts
The parts used in the study are stabilized using a process based on
engineering knowledge and expertise; e.g., engines which are ‘broken-in’
versus ‘green’ engines.

Variability Studies
164
V6: Time Series Analysis
Assumptions:
• Repeated readings are taken over specified time intervals .
Analyze by determining the degradation model for each sample part:
• e
E
σ σ
=
• Consistency of degradation (if n 2)
V7: Linear Analysis
Assumptions:
• Repeated readings are taken over specified time intervals.
• The degradation in the measurement system is known (documented)
to have a linear response over the specified time intervals.
Analyze by linear regression:
• e
E
σ σ
=
• Consistency of degradation (if n 2)
V7a – V7 with a Homogeneous Sample
Analyze by linear regression:
This is an upper bound study: E e btwn
σ σ σ
 
V8: Time versus Characteristic (Property) Degradation
V6 & V7 can be modified to determine if the degradation is time (i.e., shelf
life) or activity dependent.

Variability Studies
165
V9: V2 with Simultaneous Multiple Readings and p 3 Instruments
Analyze same as V2 (see also Lipson & Sheth, sec 13.2).

Variability Studies
166

Chapter IV – Section D
Quantifying the Effect of Excessive Within-Part Variation
167
Section D
Recognizing the Effect of Excessive
Within-Part Variation
Understanding the sources of variation of a measurement system is important
for all measurement applications but becomes even more critical where there
is significant within-part variation. Within-part variation, such as taper or
out-of-round, can cause the measurement system evaluation to provide
misleading results. This is because unaccounted within-part variation affects
the estimate of repeatability, reproducibility, or both. That is, the within-part
variation can appear as a significant component of the measurement system
variation. Understanding the within-part variation present in the product will
result in a more meaningful understanding of the suitability of the
measurement system for the task at hand.
Examples of within-part variation which may be encountered are: roundness
(circular runout), concentricity, taper, flatness, profile, cylindricity, etc.P
73
F It
is possible that more than one of these characteristics may be present at the
same time within the same part (composite error). The strength of each
characteristic and their interdependencies may compound the data and the
resultant understanding of the measurement system. Recognition and
analysis of these additional sources of variation during a measurement
system study is paramount to understanding the actual measurement system
variation. A D.O.E., ANOVA or other more sophisticated statistical
technique may be required to analyze this situation. Whatever methodology
is chosen should be approved by the customer supplier quality representative.
Also, decisions that product design engineers make can unwittingly affect
how a part is measured, how a fixture may be designed, and the result can
affect the outcome of measurement error. An example might be a plastic part
that has a critical feature on the parting line (a parting line typically has
excess plastic material where the two halves of the mold join, and thus is an
uncontrolled surface). These factors ought to be considered during a Design
FMEA.
73
Many of these features are controlled in the design by way of Geometric Dimensioning and Tolerancing
(GD&T). GD&T provides an operationally defined method to check parts in a functional manner. Generally, a
functional check is an attribute check. Where variable data is required, issues can arise as to using a gage designed
for a functional check to yield variable data. This may sometimes be done by using the functional gage as a holding
fixture for a CMM study. However, when this is done it is critical that the fixture hold the part firmly and
repeatably in the same location (if it does not, the resultant MSA study should generate this error).

Chapter IV – Section D
Quantifying the Effect of Excessive Within-Part Variation
168
Once the within-part variation components are understood, it may be possible
to control these factors within the measurement system (e.g., by redesigning
the gage or using different fixturing methods/equipment) so that future data
is not confounded.

Chapter IV – Section E
Average and Range Method – Additional Treatment
169
Section E
There is an additional consideration worthy of mention relative to the
Average and Range method of measurement system assessment.P
74
F
The primary purpose of this graphical approach is the same as other well
designed measurement system analyses: to determine if the measurement
process is adequate to measure the manufacturing process variation and/or
evaluate product conformance
• Are all gages doing the same job?
• Are all appraisers doing the same job?
• Is the measurement system variation acceptable with respect to the
process variation?
• How good are the data obtained from the measurement process or
into how many non-overlapping groups or categories can the data be
divided?
1) Care should be taken to follow the “Preparation for Measurement
Systems Study”, Chapter II, Section C.
2) Have each appraiser check each sample for the characteristic being
studied. Record the first checks on the top data row of a control chart
(see Figure IV-E 1 & 2).
3) Repeat the checks and record the data on the second data row of the
control chart. (Note: Do not allow the appraisers to see their original
reading while making this second check.) The data should now show
two checks on the same part by each appraiser.
4) Analyze the data by calculating the average ( X ) and range (R) for
each subgroup.
5) Plot the range values on the range chart and calculate the average
range ( R ) (include all sub-group ranges (R) for all appraisers). Draw
this average range on the chart. Use the D4 factor for n = 2 to
calculate the control limit for the range chart. Draw in this limit and
determine if all values are in control.
 If all ranges are in control, all appraisers are doing the same
job.If one appraiser is out of control, his method differs from
the others.
 If all appraisers have some out-of-control ranges, the
measurement system is sensitive to appraiser technique and
needs improvement to obtain useful data.
74
The control chart example is taken with permission from “Evaluating the Measurement Process,” by Wheeler &
Lyday (see Reference List).
Procedural Steps

170
6) Next, plot the average for each subgroup ( X ) for all appraisers on
the average chart (see Figure IV-E 1 & 2). The average values
represent both variation and measurement variation.
Calculate the grand average ( X ) (include all subgroup averages ( X ) for all
appraisers). Draw this grand average ( X ) on the chart.
Now calculate the control limits for this chart using the A2 factor for n = 2
and average range ( R ) from the range chart and draw these limits on the
average chart. Note in this study, the range chart contains only measurement
variation. Thus, the area between the control limits of the averages represents
the amount of measurement variation in the system.
If all averages fall inside the control limits, the measurement variation
overshadows the process variation. In other words, the measurement process
has more variation than the manufacturing process and is of no value in
monitoring or controlling that process.
If less than half of the averages are outside the limits, the measurement
system is inadequate for process control.
On the other hand, if a majority of the averages fall outside the control limits,
it indicates that the signals from the manufacturing process are greater than
measurement variation. This measurement system can provide useful data for
controlling the process.
Worksheet Example
The question, “How good are the data collected by this measurement
system?” can be answered by completing the worksheet example, Figure IV-
E 3 & 4. All data needed for the worksheet can be found on the average and
range charts described above.
Following are the steps used in completing the worksheet example (Figure
IV-E 3 & 4):
1) Identify the measurement and characteristic being evaluated, who is
doing the study and the date of the study.
2) The average subgroup range ( R ) is obtained directly from the
control chart.
3) The number of replications (r) is the number of times each appraiser
checked the same part.
4) Calculate the estimated repeatability standard deviation ( ˆe
σ ), as
shown in Figure IV-E 3 by using the 2
d value for the corresponding
r.
5) Insert the number of appraisers (nA ) in the space provided.
6) Insert the number of samples (n) in the space provided.

171
7) Compute each appraiser average by averaging all the samples
obtained by each appraiser and enter these averages in the space
provided for each appraiser (A, B, C).
8) Examine the appraiser averages (A, B, C) and determine the range of
the appraiser averages by subtracting the lowest from the highest and
insert in the space for (RA).
9) Compute the estimated appraiser standard deviation ( ˆA
σ ) as shown
by using the *
2
d value for the corresponding nA value.
10) Compute the sample averages by averaging the value obtained by all
appraisers for each sample. For example, (sample 1 avg. of appraiser
A + sample 1 avg. of appraiser B + sample 1 avg. of the last
appraiser and divide this sum by the number of appraisers). This is
the best estimate of that sample’s true value. Place the value for each
sample in the space provided (1, 2, 3.....9, 10) in Figure IV-E 4.
11) Observe the sample averages (1, 2, 3......9, 10) and calculate the
range of the sample averages ( P
R ) by subtracting the lowest from
the highest. Insert this value in the space provided.
12) Estimate the sample-to-sample standard deviation ˆ
( )
P
σ as shown by
using the *
2
d value for the corresponding n value.
13) Compute the “Signal-to-Noise Ratio” (SN) by dividing the sample
standard deviation by the measurement standard deviation and insert
in the space provided.
P
GRR
SN =
σ
σ
14) Determine the number of distinct product categories that can be
distinguished by these measurements. Multiply SN by 1.41 and insert
in the space provided (Figure IV-E 4).
Only the integer portion of this number need be considered since it
is defining distinct categories. (Drop all decimal fractions.) (See
Figure IV-E 4.)
If the number of categories is less than two (2), the measurement system is of
no value in controlling the process. It is all noise and one part cannot be said
to be different from another.
If the number of categories is two (2), it means the data can be divided into
only high and low groups; however this is only the equivalent of attribute
data.
If the number of categories is three (3), the data can be divided into high,
medium and low groups. This is a slightly better measurement system.
A system containing four (4) or more categories would be much better than
the first three examples.

172
Figure IV-E 1: Measurement Evaluation Control Chart ( &
X R ) - 1

173
Figure IV-E 2: Measurement Evaluation Control Chart ( &
X R ) - 2

174
Measurement Done
Evaluated______FITTING LENGTH____ By_____R.W.L._____ Date__MM-DD-YY____
REPLICATION ERROR:Average Subgroup Range = = 7.6
R
Number of Replications = Subgroup Size = r = 2
Estimate Replication Standard Deviation
2
ˆ 6.738
e
R
d
σ
= =
CALCULATIONS FOR APPRAISER EFFECT: Write appraiser averages below:
Appraiser Average
Number of Appraisers = = 3
A
n A 103.1
B – lo 102.8
Number of Samples = = 5
n C – hi 108.1
=
108.1-102.8 A
R D ____
Range of Appraiser Averages = = 5.3
A
R E ____
F ____
σ
=
5.3
ˆ
1.906
A
σ
= =
*
2
ˆ 2.781
A
R
d
Figure IV-E 3: Alternate Computations for Evaluating a Measurement Process (Part 1 of 2).
2 1.410
3 1.906
4 2.237
5 2.477
6 2.669
7 2.827
8 2.961
9 3.076
10 3.178
nA
*
2
d
r
2 1.128
3 1.693
4 2.059
5 2.326
6 2.534
7 2.704
8 2.847
9 2.970
10 3.078
2
d

175
CALCULATIONS FOR MEASUREMENT ERROR STANDARD DEVIATION:
Estimate the measurement error standard deviation to be:
2 2
2 2
ˆ ˆ ˆ
= (6.7376) + (2.7806)
= 7.289
m e A
σ σ σ
= 
Write the appropriate value for ˆm
σ here: ˆ 7.289
m
σ =
CALCULATIONS FOR SIGNAL TO NOISE RATIO:
Write the average for each sample piece or batch below:
Sample Average
=
113.67 - 83.0 p
R
Range for these Sample Averages = = 30.67
p
R
Estimate Sample to Sample Standard Deviation:
30.67
12.38
2.477
= σ
= =
*
2
ˆ 12.38
P
P
d
R
Signal to Noise Ratio:
=
12.38
1.698
7.289
ˆ
1.698
ˆ
P
m
σ
σ
=
Thus the number of distinct product categories that can be
Reliably distinguished by these measurements is 1.41 x 1.698
ˆ
1.41 2.395
ˆ
P
m
σ
σ
=
 
 
 
or 2
This is the number of non-overlapping 97% confidence intervals that will span the range of product
variation. (A 97% confidence interval centered on a single measurement would contain the actual product
value that is represented by that measurement 97% of the time.)
Figure IV-E 4: Alternate Computations for Evaluating a Measurement Process (Part 2 of 2).
2 1.410
3 1.906
4 2.237
5 2.477
6 2.669
7 2.827
8 2.961
9 3.076
10 3.178
n
*
2
d
2 113.33
3 - Lo 83.00
4 102.17
5 - Hi 113.67
6
7
8
9
10
1 111.17

176

Chapter IV – Section F
Gage Performance Curve
177
Section F
Gage Performance CurveP
75
The purpose of developing a Gage Performance Curve (GPC) is to determine
the probability of either accepting or rejecting a part of some reference value.
Ideally, the GPC for a measurement without error is shown in Figure IV-F 1.
However, this is the ideal for measurement systems, rather than what
normally occurs.
Once the amount of error has been determined, it is possible to calculate the
probability of accepting a part of some reference value when using that
system.
To accomplish this, the assumption is made that the measurement system
error consists primarily of lack of repeatability, reproducibility and bias.
Further, repeatability and reproducibility are taken to be normally distributed
with some variance, 2
σ . Consequently, gage error is normally distributed
with a mean XT, the reference value, plus the bias, and has some variance,
2
σ . In other words:
2
, )
( T b
Actual Value from the Gage N X σ

=
The probability of accepting a part of some reference value is given by the
relationship
2
( , )
UL
a T
LL
P N X b dx
= 
 σ
Using the standard normal table
- ( ) - ( )
-
T T
a
UL X b LL X b
P  
σ σ
 
   
=    
   
where
2
- ( )
( , )
UL
T
T
UL X b
N X b dx
 σ
σ −

 
= 
 
 
 
2
- ( )
( , )
T
T
LL
LL X b
N X b dx
 σ
σ


 
= 
 
 

75
Adapted with permission from “Analysis of Attribute Gage Systems” by J. McCaslin & G. Gruska, ASQC,
1976.

178
Example:
Determine the probability of accepting a part when the reference torque value
is 0.5 Nm, 0.7 Nm, 0.9 Nm.
Using data from a previously conducted study:
Upper specification = USL = 1.0 Nm
Lower specification = LSL = 0.6 Nm
bias = b = 0.05 Nm
GRR
σ = 0.05 Nm
Applying the above to the formulas on the previous page:
- ( ) - ( )
-
T T
a
UL X b LL X b
P  
σ σ
 
   
=    
   
1.0 - (0.5 0.05) 0.6 - (0.5 0.05)
-
0.05 0.05
a
P  
 
   
=    
   
(9.0) (1.0)
a
P  
= −
= 1.0 – 0.84
= 0.16
That is, when the part has a reference value of 0.5 Nm it will be rejected
approximately 84% of the time.
Gage Performance Curve Example:
For XT = 0.7 Nm
1.0 - (0.7 0.05) 0.6 - (0.7 0.05)
-
0.05 0.05
a
P  
 
   
=    
   
(5.0) ( 3.0)
a
P  
= − −
= 0.999
If the reference value of the part is 0.7 Nm then it will be rejected less than
approximately 0.1% of the time.
For XT = 0.9 Nm
1.0 - (0.9 0.05) 0.6 - (0.9 0.05)
-
0.05 0.05
a
P  
 
   
=    
   
(1.0) ( 7.0)
a
P  
= − −
= 0.84

179
If the reference value of the part is 0.9 Nm then it will be rejected
approximately less than 16% of the time.
If the probability of acceptance is calculated for all values of XT and plotted,
then the Gage Performance Curve, as shown in Figure IV-F 2, would be
obtained.
This same curve can be more easily plotted on normal probability paper, as
shown in Figure IV-F 3. Once plotted, the GPC gives the probability of
accepting a part of any part size.
In addition, once GPC is developed, it can be used to calculate the
repeatability and reproducibility error, and the bias errorP
76
F.
The 6 GRR range can be determined by finding the XT value that corresponds
to Pa = 0.99865 (z = 3), and the XT value that corresponds to Pa = 0.00135 (z
= -3) for either of the two limits. The GRR is the difference between the two
XT values, as shown graphically in Figure IV-F 3.
An estimate of the bias is determined by finding the XT, for either the upper
or lower limit, that corresponds to Pa = 0.5, and calculating:
B = XT – LSL or BB = XT – USL
depending on which limit XT is chosen.P
77
F
76
See “Attribute Gage Study,” Chapter III, Section C.
77
This assumes that the measurement system is linear over the operating range.

180
Figure IV-F 1: Gage Performance Curve Without Error

181
Figure IV-F 2: Gage Performance Curve –Example

182
Figure IV-F 3: Gage Performance Curve Plotted on Normal Probability Paper

Chapter IV – Section G
Reducing Variation Through Multiple Readings
183
Section G
If the variation of the present measurement system is not acceptable (over
30%), there is a method that can be used to reduce the variation to an
acceptable level until proper measurement system improvements can be
made. The unacceptable variation may be reduced by taking multiple
statistically independent (non-correlated) measurement readings of the part
characteristic being evaluated, determining the average of these
measurements and letting the numeric value of the result be substituted for
the individual measurement. This method will, of course, be time consuming,
but is an alternative (with customer approval), until improvements are made
to the measurement system (i.e., redesigning or buying a new system). The
procedure for this alternative method is as follows:
1) Determine the number of multiple readings required to meet an
acceptable level of variation.
2) Follow the gage study procedure discussed earlier in this
guideline.
Example:
In the XYZ Corporate example, the percentage of tolerance GRR variation is
25.5%, with a 6 sigma spread equal to 0.24. The customer wants to reduce
this figure to at least 15%, which would have an equivalent 6 sigma spread of
0.14.P
78
F
To determine the number of multiple readings required to meet the desired
15% criteria, it must first be understood that the distribution of individual and
average measurements has the same mean numeric value. Secondly, the
variance of the distribution of averages is equal to the variance of the
distribution of individuals divided by the sample size. Understanding this
relationship as shown below, the number of multiple readings required can
then be determined.
2
2 (6 )
(6 )
x
n
σ
σ =
This can be approximated by
(6 )
(6 )
x
s
s
n
=
and
0.24
0.14
n
=
78
See note on page vi.

Chapter IV – Section G
184
so
1.714
n =
and
n = 3 (rounded to nearest integer)
Therefore, 3 multiple readings on the part characteristic will reduce the total
measurement system variation to approximately 0.14 and the %GRR to 15%.
This method should be considered as a temporary step until other
improvements are made on the measurement system. This should be used
only with the concurrence of the customer.

Chapter IV – Section H
Pooled Standard Deviation Approach to GRR
185
Section H
Pooled Standard Deviation Approach to GRRP
79
Analysis of a measurement system usually assumes that replicated data from
all the part/specimens can be obtained from all the appraisers in a random
fashion. This may not always be possible. If the measurement system
analysis includes multiple locations it can be logistically unfeasible to
demand a random sampling. Also some tests, notably chemical and
metallurgical analyses (inter- and intra- laboratory studies), may require a
cross section of diverse samples which are not part of a homogeneous
process and may not be available at the same time.
These situations may be handled by using a nested DOE. An alternate
approach is the pooled standard deviation GRR study which follows the
methodology discussed in ASTM E691.
This approach views each part as a separate material and then computes the
repeatability and reproducibility standard deviations as in E691. This will
yield multiple separate values of repeatability and reproducibility. Since the
parts are considered to be essentially identical, these separate estimates are
assumed to be effectively identical. Of course they will never be exactly the
same, but their average will give a good estimate of the true level of
repeatability and similar reproducibility.
If this approach is used to evaluate a group of laboratories there is
a concern of just what the nature of the “reproducibility” is. If this is
greater than zero most of the time (that is for most of the materials)
it should be interpreted as implying that there are differences
between operators; i.e., in an interlaboratory program this would
suggest that there are real differences between labs.
Although the E691 approach is typically used with a complete sample, it
lends itself to a sequential approach. This is useful when all the samples are
not available at the same time. It also can be used as part of the calibration
process to maintain information on the measurement system’s variability.
The following study description assumes that the study will be applied in a
sequential manner.
Care should be taken to follow the “Preparation for Measurement System
Study” shown in Chapter II, Section C.
Continue from step 6 on page 74:
79
Portions of this section including all of “Consistency Statistics” contributed by Neil Ullman of the American
Society for Testing and Materials (ASTM International).
Sequential
Application

186
7) Have each of the 2
m  appraisers evaluate the part for 3
r 
readings. Record the data on the appropriate rows of a data collection
sheet (see Sample Forms). (Note: Do not allow the appraisers to see
their original reading while making these multiple checks.)
8) Calculate the average ( X ) and standard deviation (s) for the new
part for each appraiser.
9) Plot the standard deviation values on the standard deviation chart and
calculate the average standard deviation ( s ) (include all sub-group
standard deviation for all appraisers). Draw this average standard
deviation on the chart. Use the B4 factor for r samples to calculate
the upper control limit for the standard deviation chart. Draw in this
limit and determine if all values are in control (see Figure IV-H 1).
10) Plot the average ( X ) for each subgroup for all appraisers on the
average chart (see Figure IV-H 1). The average values represent both
process variation and measurement variation.
11) Calculate the grand average ( X ) (include all subgroup averages
( X ) for all appraisers). Draw this grand average ( X ) line on the
chart.
12) Calculate the control limits for this chart using the A2 factor for r and
average standard deviation ( s ) from the standard deviation chart;
draw these limits on the average chart.
13) Analyze the data using the control charts and other graphical
techniques as discussed in the Average and Range Method (see
Chapter III).
14) Evaluate the measurement system’s parameters for each part by
pooling the appraisers’ results.
2
1
i
g
m
x
i
x
s
s
m
=
=
∑
2
1
i
g
m
s
i
g E
s
Repeatability s
m
=
= =
∑
2
2 g
g g
E
g A x
s
Reproducibility s s
m
= = −
2 2
g g g
g GRR E A
GRR s s s
= = 

187
E691 follows the convention wherein the MSA’s reproducibility is
referred to as the appraiser variation and the MSA’s GRR is called
reproducibility. In this case,
sappr =
2
2
3
r
x
s
s −
sR =
2 2
r appr
s s

where sr = sE = Repeatability and sR = GRR = ASTM
Reproducibility
15) Evaluate the overall measurement system’s parameters by pooling
the part results where g = number of parts.
2
1
i
g
E
i
E
s
Repeatability s
g
=
= =
∑
2
1
i
g
A
i
A
s
Reproducibility s
g
=
= =
∑
2
1
i
g
GRR
i
GRR
s
GRR s
g
=
= =
∑
When calculating the percent of total variation, the historic standard
deviation for the process should be used.
If the parts cover a wide variety of processes, for example different
metallurgical or chemical specimens, the evaluation of percent total
variation should be based on the process variation of specific
specimens and not the total variation over all specimens.
Care should be given in the interpretation of the measurement
system’s parameters in the case where appraisers are located in
different facilities (e.g., laboratories).
The Repeatability will include between-equipment variation as well
as within-equipment variation. This can be evaluated by calculating
and comparing the repeatability within each location.
The Reproducibility will include between-location variation as well
as between-appraiser variation. These components cannot be
separated with this study.

188
28
25
22
19
16
13
10
7
4
1
2
1
0
-1
-2
Sample
Sa
m
ple
M
e
a
n
_
_
X= 0.001
UCL= 0.374
LCL= -0.372
28
25
22
19
16
13
10
7
4
1
0.60
0.45
0.30
0.15
0.00
Sample
Sa
m
ple
St
De
v
_
S= 0.1908
UCL= 0.4901
LCL= 0
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
2
2
2
Xbar-S Chart of Response
Figure IV-H 1: Pooled Standard Deviation Study Graphical AnalysisP
80
80
Control limits based on pooled standard deviation adjusted by an unbiasing constant

Chapter
IV
–
Section
H
Pooled
Standard
Deviation
Approach
to
GRR
189
Table
IV-H
1:
Pooled
Standard
Deviation
Analysis
Data
Set
Part Across Parts
Trial 1 2 3 4 5 6 7 8 9 10
A 1 0.29 -0.56 1.34 0.47 -0.8 0.02 0.59 -0.31 2.26 -1.36
2 0.41 -0.68 1.17 0.5 -0.92 -0.11 0.75 -0.2 1.99 -1.25
3 0.64 -0.58 1.27 0.64 -0.84 -0.21 0.66 -0.17 2.01 -1.31
xbar 0.4467 -0.6067 1.2600 0.5367 -0.8533 -0.1000 0.6667 -0.2267 2.0867 -1.3067 xdbbar 0.19033333
sd 0.1779 0.0643 0.0854 0.0907 0.0611 0.1153 0.0802 0.0737 0.1504 0.0551 appr sd 0.10289153
B 1 0.08 -0.47 1.19 0.01 -0.56 -0.2 0.47 -0.63 1.8 -1.68
2 0.25 -1.22 0.94 1.03 -1.2 0.22 0.55 0.08 2.12 -1.62
3 0.07 -0.68 1.34 0.2 -1.28 0.06 0.83 -0.34 2.19 -1.5
xbar 0.1333 -0.7900 1.1567 0.4133 -1.0133 0.0267 0.6167 -0.2967 2.0367 -1.6000 xdbbar 0.06833333
sd 0.1012 0.3869 0.2021 0.5424 0.3946 0.2120 0.1890 0.3570 0.2079 0.0917 appr sd 0.3017394
C 1 0.04 -1.38 0.88 0.14 -1.46 -0.29 0.02 -0.46 1.77 -1.49
2 -0.11 -1.13 1.09 0.2 -1.07 -0.67 0.01 -0.56 1.45 -1.77
3 -0.15 -0.96 0.67 0.11 -1.45 -0.49 0.21 -0.49 1.87 -2.16
xbar -0.0733 -1.1567 0.8800 0.1500 -1.3267 -0.4833 0.0800 -0.5033 1.6967 -1.8067 xdbbar -0.25433333
sd 0.1002 0.2113 0.2100 0.0458 0.2223 0.1901 0.1127 0.0513 0.2194 0.3365 appr sd 0.19056058
sd xbar 0.26182 0.28005 0.19648 0.19751 0.24077 0.26555 0.32524 0.14385 0.21221 0.25125 Pooled sd's
Repeatability 0.13153 0.25721 0.17534 0.31863 0.26388 0.17736 0.13524 0.21252 0.19494 0.20385
pooled 0.204274 0.195107 0.23223 0.238896 0.229787 0.218795 0.218021 0.215578 0.21443466
Reproducibility 0.25056 0.23743 0.16839 0.07191 0.18644 0.24501 0.31573 0.07508 0.17991 0.22198
pooled 0.244086 0.221744 0.195372 0.193619 0.203089 0.222696 0.209998 0.206871 0.20843064
GRR 0.28299 0.35004 0.24310 0.32664 0.32310 0.30247 0.34347 0.22540 0.26527 0.30138
pooled 0.318285 0.295359 0.303481 0.307505 0.306671 0.312194 0.302708 0.29878 0.29904106
By
Part
results
Measurement
System

190
Consistency Statistics
The ASTM and ISO methodsP
81
F suggest that two “consistency” statistics, h
and k, be computed. The h values are calculated as:
h =
appr part
x
x x
s
−
For appraiser A and part 1 the average ( appr
x above) is 0.447 and the part
average ( part
x above) is 0.169. The standard deviation among appraisers ( x
s
above) is 0.262. Then
0.447 - 0.169 0.278
= = 1.06
0.262 0.262
=
h
The value of k is the ratio of the standard deviation for each part for each
appraiser to the repeatability standard deviation. In this case (appraiser A and
part 1 ) it is:
standard deviation (appr A, part 1) 0.178
1.35
repeatability 0.132
= =
k =
One reason these are computed is to allow comparisons among very different
materials.
Although in this example there is not a collection of widely different
materials which have different levels and possibly very different standard
deviations, the h and k calculations of E691 can still use to compare the
repeatability standard deviation and the response values by appraisers. In the
following table the h and k’s are listed by appraiser.
h Part
Appr 1 2 3 4 5 6 7 8 9 10 avg h "z"
A 1.06 0.87 0.82 0.86 0.88 0.32 0.65 0.80 0.69 1.05 0.80 2.53
B -0.14 0.22 0.29 0.24 0.21 0.80 0.50 0.32 0.46 -0.11 0.28 0.88
C -0.93 -1.09 -1.11 -1.10 -1.09 -1.12 -1.15 -1.12 -1.15 -0.94 -1.08 -3.41
k median k "z"
A 1.35 0.25 0.49 0.28 0.23 0.65 0.59 0.35 0.77 0.27 0.42 -3.20
B 0.77 1.50 1.15 1.70 1.50 1.20 1.40 1.68 1.07 0.45 1.30 3.14
C 0.76 0.82 1.20 0.14 0.84 1.07 0.83 0.24 1.13 1.65 0.84 -0.17
81
See ISO 5725

191
In the last two columns are the averages and a value of “z-value” to see if the
appraisers are significantly different. The h values indicate that appraiser A is
significantly high and appraiser C is significantly low in their readings of the
size of the parts. It is also this significant difference which creates the GRR
standard deviation.
The repeatability standard deviations can be also evaluate by looking at the k
values. To do this, compute the median k and then an approximate “z score”.
With this study, the expected median is about 0.861 with a standard deviation
of approximately 0.439. The median k for appraiser A is then about –3.2
standard deviations below the expected level and appraiser B is as
significantly high. So we see very great differences in the performance of just
these three operators.
The graphs of h (Figure IV-H 2) and k (Figure IV-H 3) also help to illustrate
these differences. Appraiser C has much lower results than the others.
Similarly, the k values show how much lower appraiser A variation is with
respect to repeatability. These are certainly issues to be examined about the
performance of the measurement method as carried out by these appraisers.
h values
-1.5
-1
-0.5
0
0.5
1
1.5
0 1 2 3 4
Appraiser
A B C
Figure IV-H 2: Dot diagram of h values.

192
k values
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
0 1 2 3 4
Appraiser
A B C
Figure IV-H 3: Dot diagram of k values.

Appendix A
Analysis of Variance Concepts
195
Appendix A
The GRR numerical analysis can be done following the formulas in Table A
3. This is what is called the Analysis of Variance (ANOVA) table. The
ANOVA table is composed of six columns:
• Source column is the cause of variation.
• DF column is the degree of freedom associated with the source.
• SS or sum of squares column is the deviation around the mean of the
source.
• MS or mean square column is the sum of squares divided by degrees
of freedom.
• EMS or expected mean square column determines the linear
combination of variance components for each MS. The ANOVA
table decomposes the total source of variation into four components:
parts, appraisers, interaction of appraisers and parts, and replication
error due to the gage/equipment repeatability.
• The F-ratio column is calculated only for the interaction in a MSA
ANOVA; it is determined by the mean square of interaction divided
by mean square error.
The estimate of variance components for each source is given in Table A 1. F
82
Variance Estimate
Equipment (EV) 2
e
MS
 = P
83
Interaction (INT) 2 AP e
MS MS
r
−
=

Appraiser (AV) 2 A AP
MS MS
nr
−
=

Part (PV) 2 P AP
MS MS
kr
−
=
σ
Table A 1: Estimate of Variance Components
Since each mean square is a sample quantity subject to sampling variation
and the computations involve differences of mean squares, then negative
variance component estimates are possible. This is a bit of a problem since
the “master” variance components are equal or close to zero or have a small
82
In this table, all components of variation are assumed to be random effects.
83
In this application of ANOVA to measurement systems analysis, the ANOVA error term equates to the MSA
equipment variation, MSE.

Appendix A
196
sample size. For analysis purposes, the negative variance component is set to
zero.
Standard deviation is easier to interpret than variance because it has the same
unit of measurement as the original observation. In practice, the basic
measure of spread is given by 6 times the standard deviation.P
84
F Table A 2
shows the 6 sigma spread for a measure of repeatability called equipment
variation (EV) and measure of reproducibility called appraiser variation (AV).
If the interaction of part and appraiser is significant, then there exists a non-
additive model and therefore an estimate of its variance components is given.
The GRR in Table A 2 is the total of measurement system variation.
6
= e
EV MS Equipment Variation = Repeatability
-
6
= A AP
MS MS
AV
nr
Appraiser Variation = Reproducibility
-
6
= AP e
AP
MS MS
I
r
Interaction of Appraiser by Part
2 2 2
( ) ( ) ( )
AP
GRR EV AV I
=  
Gage R&R
-
6
= P AP
MS MS
PV
kr
Part Variation
Table A 2: 6 Sigma Spread
In the additive model, the interaction is not significant and the variance
components for each source is determined as follow: First, the sum of square
of gage error (SSe from Table A 3) is added to the sum of square of appraiser
by part interaction (SSAP from Table A 3) and which is equal to the sum of
squares pooled (SSpool ) with (nkr – n – k +1)P
85
F degrees of freedom. Then the
SSpool will be divided by the (nkr – n – k +1) to calculate MSpool. The 6 sigma
spread limit then will be:
6
= pool
EV MS
-
6
=
A pool
MS MS
AV
nr
2 2
( ) ( )
GRR EV AV
= 
-
6
=
P pool
MS MS
PV
kr
84
This is the 99.73% range. See note on page vi.
85
Where n = number of parts, k = number of appraisers and r = number of trials.

Appendix A
197
In order to determine if the interaction is significant, compute the F
statistic of appraiser by part interaction (see Table A 3). Compare
this F statistic to an upper percentage point of an F distribution with
numerator and denominator degrees of freedom taken from the
ANOVA (Table A 3).
In order to decrease the risk of falsely concluding that there is no interaction
effect, choose a high significance level. Once the GRR has been determined
then the %GRR can be calculated in relation to process performance.
2 2
.. ...
1
n
i
P
i
x x
SS
kr nkr
=
 
= −
 
 
∑
2 2
. . ...
1
k
j
A
j
x x
SS
nr nkr
=
 
= −
 
 
 
∑
 
2
2 ...
1 1 1
n k r
ijm
i j m
x
TSS x
nkr
= = =
= −
∑∑∑
2 2
2 2
. . .
.. ...
1 1 1 1
n k n k
ij j
i
AP
i j i j
x x
x x
SS
r kr nr nkr
= = = =
   
 
= − − 
   
 
   
 
   
∑∑ ∑ ∑
[ ]
e A P AP
SS TSS SS SS SS
= −  
Source DF SS MS F EMS
Appraiser k – 1 SSA
MSA =
( 1)
A
SS
k −
2 2 2
r nr
  
 
Parts n – 1 SSP
MSP =
( - 1)
P
SS
n
2 2 2
r kr
  σ
 
Appraiser-by-Part (n – 1) (k – 1) SSAP
MSAP =
( - 1) ( - 1)
AP
SS
n k
AP
e
MS
MS
2 2
r
 

Equipment nk (r – 1) SSe
MSe =
( - 1)
e
SS
nk r
2

Total nkr – 1 TSS Appraiser ~ N(
2
0, )
Parts ~ N(
2
0,σ )
Appraiser x Part ~ N(
2
0, )
Equipment ~ N(
2
0, )
Table A 3: Analysis of Variance (ANOVA)
Tables A 4 and A 5 show the ANOVA calculations for our example data
from Figure III-B 15.

Appendix A
198
Source DF SS MS F EMS
Appraiser 2 3.1673 1.58363 79.41* 2 2 2
3 30
  
 
Parts 9 88.3619 9.81799 492.29* 2 2 2
3 9
  σ
 
Appraiser by Part 18 0.3590 0.01994 0.434 2 2
3
 

Equipment 60 2.7589 0.04598 2

Total 89 94.6471
* Significant at α = 0.05 level
Table A 4: Tabulated ANOVA Results
Since the calculated F value for the interaction (0.434) is less than the critical value of , ,
Fα 18 60 , the interaction term
is pooled with the equipment (error) term. That is, the estimate of variance is based on the model without
interaction.
Estimate of
Variance
Std.
Dev. (σ)
6 (σ) % Total Variation % Contribution
2
 = 0.039973
(Equipment)
0.199933 EV = 1.199598 18.4 3.4
2
 = 0.051455
(Appraiser)
0.226838 AV = 1.361028 20.9 4.4
2
 = 0
(Interaction)
INT = 0 0 0
GRR = 0.09143
(
2 2 2
  
  )
0.302373 GRR = 1.814238 27.9 7.8
2
σ = 1.086447
(Part)
1.042327 PV = 6.253962 96.0 92.2
Total Variation 1.085 TV = 6.51 100.0
Table A 5: Tabulated ANOVA Results
1.41( / ) 1.41(6.25/1.81) 4.87 4
= = = 
ndc PV GRR
2 2
( )
Total Variation TV GRR PV
= 
( ) ( )
( ) ( )
6
% 100 100
6
σ σ
σ σ
   
= =
   
   
   
components components
total total
of Total Variation
2 2
( ) ( )
2 2
( ) ( )
6
% ( ) 100 100
6
σ σ
σ σ
   
= =
   
   
   
components components
total total
Contribution Total Variance

Appendix B
Impact of GRR on the Capability Index Cp
199
Appendix B
Formulas:
2 2 2
Obs Act Meas
σ σ σ
=  (1)
where Obs = the observed process variation
Act = the actual process variation
Meas = the measurement system variation
6
x
x
U L
Cp
σ
−
= (2)
where U, L are the upper and lower specification values
x = Obs or Act as defined in (1)
% *100%
p
GRR GRR
= (3)
based on process variation:
6
Meas
p
Obs
k
GRR
σ
σ
=
(4)
Note: 1
p
GRR  since 2 2
Obs Meas
σ σ
 by equation (1)
based on the tolerance range:
Meas
p
k
GRR
U L
σ
=
−
(5)
In (4) and (5), k is normally taken to be 6.
Analysis:
* Act
Obs Act
Obs
Cp Cp
σ
σ
=
2 2
* Obs Meas
Act
Obs
Cp
σ σ
σ
−
= using (1)
with GRR based on the process variation
2
1
* Obs
Obs Act
Obs
GRR
Cp Cp
σ
σ
−
= using (4)
2
* 1
Act
Cp GRR
= − (6)
or

Appendix B
200
2
1
Obs
Act
Cp
Cp
GRR
=
−
(6’)
with GRR based on the tolerance range
1
* Meas
Obs Obs
GRR
Cp
σ
σ
= using (2) and (5)
consequently
 
2
* 1 *
Obs Act Obs
Cp Cp Cp GRR
= − (7)
and
 
2
1 *
Obs
Act
Obs
Cp
Cp
Cp GRR
=
−
(7’)

Appendix B
201
Graphical Analysis
Based on (6), the family of lines for CpObs with respect to CpAct is:
Figure B 1: Observed vs. Actual Cp (process based)
Actual GRR
10% 20% 30% 40% 50% 60% 70% 90%
Observed Cp with Cp based on Process Range
1.3 1.29 1.27 1.24 1.19 1.13 1.04 0.93 0.57
Observed Cp with Cp based on the Tolerance
1.3 1.29 1.26 1.20 1.11 0.99 0.81 0.54 never
Table B 1: Comparison of Observed to Actual Cp
Observed
C
p
0
0.5
1
1.5
2
2.5
3
3.5
0.5 1 1.5 2 2.5 3
Actual Cp
10 %
30 %
50%
70%
90%

Appendix B
202
Based on (7’), the family of lines for CpAct with respect to CpObs is:
Figure B 2: Observed vs. Actual Cp (tolerance based)
0.0
1.0
2.0
3.0
4.0
5.0
6.0
0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2.0
Observed Cp
Actual
C
p
%GRR 70% 60% 50%
30%
10%

Appendix
C
*
2
d
Table
203
Appendix
C
Table
C
1:
*
2
d
Table
Values associated with the Distribution of the Average Range
Subgroup Size ( )
m
2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
1 1.0
1.41421
2.0
1.91155
2.9
2.23887
3.8
2.48124
4.7
2.67253
5.5
2.82981
6.3
2.96288
7.0
3.07794
7.7
3.17905
8.3
3.26909
9.0
3.35016
9.6
3.42378
10.2
3.49116
10.8
3.55333
11.3
3.61071
11.9
3.66422
12.4
3.71424
12.9
3.76118
13.4
3.80537
2 1.9
1.27931
3.8
1.80538
5.7
2.15069
7.5
2.40484
9.2
2.60438
10.8
2.76779
12.3
2.90562
13.8
3.02446
15.1
3.12869
16.5
3.22134
17.8
3.30463
19.0
3.38017
20.2
3.44922
21.3
3.51287
22.4
3.57156
23.5
3.62625
24.5
3.67734
25.5
3.72524
26.5
3.77032
3 2.8
1.23105
5.7
1.76858
8.4
2.12049
11.1
2.37883
13.6
2.58127
16.0
2.74681
18.3
2.88628
20.5
3.00643
22.6
3.11173
24.6
3.20526
26.5
3.28931
28.4
3.36550
30.1
3.43512
31.9
3.49927
33.5
3.55842
35.1
3.61351
36.7
3.66495
38.2
3.71319
39.7
3.75857
4 3.7
1.20621
7.5
1.74989
11.2
2.10522
14.7
2.36571
18.1
2.56964
21.3
2.73626
24.4
2.87656
27.3
2.99737
30.1
3.10321
32.7
3.19720
35.3
3.28163
37.7
3.35815
40.1
3.42805
42.4
3.49246
44.6
3.55183
46.7
3.60712
48.8
3.65875
50.8
3.70715
52.8
3.75268
5 4.6
1.19105
9.3
1.73857
13.9
2.09601
18.4
2.35781
22.6
2.56263
26.6
2.72991
30.4
2.87071
34.0
2.99192
37.5
3.09808
40.8
3.19235
44.0
3.27701
47.1
3.35372
50.1
3.42381
52.9
3.48836
55.7
3.54787
58.4
3.60328
61.0
3.65502
63.5
3.70352
65.9
3.74914
6 5.5
1.18083
11.1
1.73099
16.7
2.08985
22.0
2.35253
27.0
2.55795
31.8
2.72567
36.4
2.86680
40.8
2.98829
45.0
3.09467
49.0
3.18911
52.8
3.27392
56.5
3.35077
60.1
3.42097
63.5
3.48563
66.8
3.54522
70.0
3.60072
73.1
3.65253
76.1
3.70109
79.1
3.74678
7 6.4
1.17348
12.9
1.72555
19.4
2.08543
25.6
2.34875
31.5
2.55460
37.1
2.72263
42.5
2.86401
47.6
2.98568
52.4
3.09222
57.1
3.18679
61.6
3.27172
65.9
3.34866
70.0
3.41894
74.0
3.48368
77.9
3.54333
81.6
3.59888
85.3
3.65075
88.8
3.69936
92.2
3.74509
8 7.2
1.16794
14.8
1.72147
22.1
2.08212
29.2
2.34591
36.0
2.55208
42.4
2.72036
48.5
2.86192
54.3
2.98373
59.9
3.09039
65.2
3.18506
70.3
3.27006
75.2
3.34708
80.0
3.41742
84.6
3.48221
89.0
3.54192
93.3
3.59751
97.4
3.64941
101.4
3.69806
105.3
3.74382
9 8.1
1.16361
16.6
1.71828
24.9
2.07953
32.9
2.34370
40.4
2.55013
47.7
2.71858
54.5
2.86028
61.1
2.98221
67.3
3.08896
73.3
3.18370
79.1
3.26878
84.6
3.34585
90.0
3.41624
95.1
3.48107
100.1
3.54081
104.9
3.59644
109.5
3.64838
114.1
3.69705
118.5
3.74284
10 9.0
1.16014
18.4
1.71573
27.6
2.07746
36.5
2.34192
44.9
2.54856
52.9
2.71717
60.6
2.85898
67.8
2.98100
74.8
3.08781
81.5
3.18262
87.9
3.26775
94.0
3.34486
99.9
3.41529
105.6
3.48016
111.2
3.53993
116.5
3.59559
121.7
3.64755
126.7
3.69625
131.6
3.74205
11 9.9
1.15729
20.2
1.71363
30.4
2.07577
40.1
2.34048
49.4
2.54728
58.2
2.71600
66.6
2.85791
74.6
2.98000
82.2
3.08688
89.6
3.18174
96.6
3.26690
103.4
3.34406
109.9
3.41452
116.2
3.47941
122.3
3.53921
128.1
3.59489
133.8
3.64687
139.4
3.69558
144.7
3.74141
12 10.7
1.15490
22.0
1.71189
33.1
2.07436
43.7
2.33927
53.8
2.54621
63.5
2.71504
72.6
2.85702
81.3
2.97917
89.7
3.08610
97.7
3.18100
105.4
3.26620
112.7
3.34339
119.9
3.41387
126.7
3.47879
133.3
3.53861
139.8
3.59430
146.0
3.64630
152.0
3.69503
157.9
3.74087
13 11.6
1.15289
23.8
1.71041
35.8
2.07316
47.3
2.33824
58.3
2.54530
68.7
2.71422
78.6
2.85627
88.1
2.97847
97.1
3.08544
105.8
3.18037
114.1
3.26561
122.1
3.34282
129.8
3.41333
137.3
3.47826
144.4
3.53810
151.4
3.59381
158.1
3.64582
164.7
3.69457
171.0
3.74041
14 12.5
1.15115
25.7
1.70914
38.6
2.07213
51.0
2.33737
62.8
2.54452
74.0
2.71351
84.7
2.85562
94.9
2.97787
104.6
3.08487
113.9
3.17984
122.9
3.26510
131.5
3.34233
139.8
3.41286
147.8
3.47781
155.5
3.53766
163.0
3.59339
170.3
3.64541
177.3
3.69417
184.2
3.74002
15 13.4
1.14965
27.5
1.70804
41.3
2.07125
54.6
2.33661
67.2
2.54385
79.3
2.71290
90.7
2.85506
101.6
2.97735
112.1
3.08438
122.1
3.17938
131.7
3.26465
140.9
3.34191
149.8
3.41245
158.3
3.47742
166.6
3.53728
174.6
3.59302
182.4
3.64505
190.0
3.69382
197.3
3.73969
16 14.3
1.14833
29.3
1.70708
44.1
2.07047
58.2
2.33594
71.7
2.54326
84.5
2.71237
96.7
2.85457
108.4
2.97689
119.5
3.08395
130.2
3.17897
140.4
3.26427
150.2
3.34154
159.7
3.41210
168.9
3.47707
177.7
3.53695
186.3
3.59270
194.6
3.64474
202.6
3.69351
210.4
3.73939
17 15.1
1.14717
31.1
1.70623
46.8
2.06978
61.8
2.33535
76.2
2.54274
89.8
2.71190
102.8
2.85413
115.1
2.97649
127.0
3.08358
138.3
3.17861
149.2
3.26393
159.6
3.34121
169.7
3.41178
179.4
3.47677
188.8
3.53666
197.9
3.59242
206.7
3.64447
215.2
3.69325
223.6
3.73913
18 16.0
1.14613
32.9
1.70547
49.5
2.06917
65.5
2.33483
80.6
2.54228
95.1
2.71148
108.8
2.85375
121.9
2.97613
134.4
3.08324
146.4
3.17829
157.9
3.26362
169.0
3.34092
179.7
3.41150
190.0
3.47650
199.9
3.53640
209.5
3.59216
218.8
3.64422
227.9
3.69301
236.7
3.73890
19 16.9
1.14520
34.7
1.70480
52.3
2.06862
69.1
2.33436
85.1
2.54187
100.3
2.71111
114.8
2.85341
128.7
2.97581
141.9
3.08294
154.5
3.17801
166.7
3.26335
178.4
3.34066
189.6
3.41125
200.5
3.47626
211.0
3.53617
221.1
3.59194
231.0
3.64400
240.5
3.69280
249.8
3.73869
Num
b
e
r
of
S
ubg
ro
u
ps
(
g)
20 17.8
1.14437
36.5
1.70419
55.0
2.06813
72.7
2.33394
89.6
2.54149
105.6
2.71077
120.9
2.85310
135.4
2.97552
149.3
3.08267
162.7
3.17775
175.5
3.26311
187.8
3.34042
199.6
3.41103
211.0
3.47605
222.1
3.53596
232.8
3.59174
243.1
3.64380
253.2
3.69260
263.0
3.73850
d2 1.12838 1.69257 2.05875 2.32593 2.53441 2.70436 2.8472 2.97003 3.07751 3.17287 3.25846 3.33598 3.40676 3.47193 3.53198 3.58788 3.64006 3.68896 3.735
cd 0.876 1.815 2.7378 3.623 4.4658 5.2673 6.0305 6.7582 7.4539 8.1207 8.7602 9.3751 9.9679 10.5396 11.0913 11.6259 12.144 12.6468 13.1362
Table entries: 1st
line of each cell is the degrees of freedom () and the 2nd
line of each cell is ; is the infinity value of : additional values of  can be built up from the constant difference, .
cd
Note: The notation used in this table follows that of Acheson Duncan, Quality Control and Industrial Statistics,5th
edition, McGraw-Hill, 1986.
 
2
2
R
 σ  is distributed approximately as a
2
 distribution with  degrees of freedom where R is the average range of g subgroups of size m
2
d
2
d*
2
d*
2
d*

Appendix D
Gage R Study
205
Appendix D
Gage R Study
Application:
• Offers a preliminary determination of short-term gage repeatability
only.
• May be used for screening before shipment from gage supplier.
• May be used in pre-production when part availability is minimal.
• May be used during gage development – e.g., for quickly comparing
different clamping locations; comparing different methodologies.
• This method CANNOT be used for final gage acceptance without
other more complete and detailed MSA methods.
Assumptions:
• Process stability and variation cannot be known at this point in time.
• Linearity and bias are not issues.
• Reproducibility is not considered here. The purpose is to focus on
gage repeatability alone.
• I/MR (Individual/Moving Range) chart will provide minimal
assessment of stability.
Analyze by:
Take one part, one operator; place part in fixture, measure; take part out of
fixture; repeat this 9 more times with the same part and same operator.
Plot data on I/MR chart; assess stability. If data appears unstable, perform
corrective action.P
86
If stable, calculate sigma of individuals by using either s for all readings or
*
2
/
MR d ; multiply by 6; divide this value by the characteristic tolerance;
multiply by 100%. Review %Repeatability against previously established
gage acceptability standards, or use for comparative purposes during gage
development.
86
Some judgment may be needed here, as 10 subgroups of individual data is insufficient to establish stability;
however, obvious instabilities may still be assessed and provide value to the analysis.

Appendix E
Alternate PV Calculation Using Error Correction Term
207
Appendix E
PV has been defined here as 2 2
TV GRR
− . Since this definition of part
variation can (does) include EV, there may be times when it is important to
extract the EV influence on PV. This may be done by the following formula
(note the similarity to the AV formula where EV influences are extracted).
2
2
3
( )
P
EV
PV R K
k r
 
 
 
 
=  −

where RP = range of the part averages, k = # of appraisers, r = # trials.
Recognition of this method of calculating PV was published in 1997.P
87
F This
method is presented here as a more statistically correct alternative to the
commonly accepted definition of PV historically used in this manual.
Generally, when EV contaminates PV, it does so to the extent of only a
percentage point or two.
87
“Reliable Data is an Important Commodity,” Donald S. Ermer and Robin Yang E-Hok, University of
Wisconsin, Madison, published in The Standard, ASQ Newsletter of the Measurement Quality Division, Vol
97-1, Winter, 1997.

Appendix E
208

Appendix F
P.I.S.M.O.E.A. Error Model
209
Appendix F
Similar to all processes, a measurement system is impacted by random and
systematic sources of variation. These sources of variation are due to
common and special (chaotic) causes. In order to understand, control and
improve a measurement system, the potential sources of variation ought to
first be identified. Although the specific causes will depend on the situation,
a general error model can be used to categorize sources of variation for any
measurement system. There are various methods of presenting and
categorizing these sources of variation using simple cause & effect, matrix,
or tree diagrams.
The acronym P.I.S.M.O.E.A.P
88
F represents another useful model for defining
a measurement system by its basic sources of variation. It is not the only
model, but does support universal application.
Error Source Alias or Component Factor or Parameter
P Part
Production part, sample, measurand, Unit
Under Test (UUT), artifact, check standard
Unknown
I Instrument
Gage, unit of M&TE, master gage, measuring
machine, test stand
Means of comparison
S Standard
Scale, reference, artifact, check standard,
intrinsic standard, consensus, Standard
Reference Materials (SRM), class, acceptance
criteria
Known value accepted as
truth *, reference value, or
acceptance criteria
M Method
On-the-job training, verbal, work instruction,
control plan, inspection plan, test program, part
program
How
O Operator
Appraiser, calibration or test technician,
assessor, inspector Who
E Environment
Temperature, humidity, contamination,
housekeeping, lighting, position, vibration,
power, Electromagnetic Interference (EMI),
noise, time, air
Conditions of measurement,
noise
A Assumptions
Statistical, operational, calibration, constants,
handbook values, thermal stability, modulus of
elasticity, laws of science
Criteria, constant, or
supposition for reliable
measurement
* Actual or physical true values are unknown
The alias error source varies with the measurement application and industry.
These are common examples. However, the parameter and characteristic
effect is always the same.
88
P.I.S.M.O.E.A. was originally developed by Mr. Gordon Skattum, Senior ASQ CQE, metrologist and Director
of Integrated Manufacturing for Rock Valley College Technology Center.

Appendix F
210
All methods of Measurement Systems Analysis are quality tools. All quality
tools are based on assumptions. If you violate the assumptions, the tool
becomes unpredictable at best and could lead to a false conclusion at worst.
The typical statistical assumptions of a Gage R&R study, include: normal
process, random and independent trials, stable, and test-retest criteria. When
one or more assumption is violated (e.g., non-normal measurement process,
operator bias) the tool and analysis ultimately become unstable, confusing,
and misleading. %GRR evaluations for product and process control can be
overestimated. There are also non-statistical assumptions related to
measurement systems (e.g., calibration, operational, coefficients and rates of
expansion, physical laws and constants). The measurement planner should be
able to identify, control, correct, or modify the MSA method to
accommodate significant violations for the assumptions in the measurement
process.
Violation of the test-retest criteria is always a consideration for destructive
test or in-process measurement systems where the feature is changing. The
measurement planner ought to consider appropriate hybrids, modifications,
or alternative techniques to the standard measurement system study.
Assumptions tend to contribute greater to total measurement variation when:
1) using more powerful methods or tools (e.g., ANOVA, regression analysis,
designed experiments, probability predictions, and control charts), and 2) as
measurement precision increases. High precision measurement applications
must often plan for, and sometimes apply, correction for coefficients and
thermal expansion, deformation, creep or other assumptions in the
measurement process.
The greatest danger for a quality analyst or measurement planner regarding
assumptions is that it is assumed they are insignificant or constant and
therefore too often ignored.
Table F 1 demonstrates examples of the PISMOEA model for three different
measurement situations.

Appendix F
211
Error
Source
Typical Production,
Automotive MSA
Automated In-Process
or Test Stand
Calibration
P
Random production parts,
entire process range
Production units, test samples,
check standards, artifacts
Gage, UUT, test sample,
measurand
I Single type of production gage DCC CMM, test stand Master gage and equipment
S
Scale, master standard, or class;
meets the “10 to 1 rule”
Scale and geometry, reference test
standards
Master, reference, intrinsic or
consensus, artifact
M
Standard Operating Procedures
(S.O.P.), often verbal, may be
documented; control plan
Documented S.O.P., DCC program
or automated test cycle
Documented, formal calibration
procedure
O
(2-3) typical, trained, who
normally operate
Restricted test operator,
specialized training and skill
Qualified technician, ISO 17025
proficiency evidence
E
Stable production and operating
conditions
Often controlled
Control limits, optimized, a
principle error source
A Statistical, often ignored Statistical, application specific
Cannot be assumed, a principle
error source
Purpose Process control (SPC)
Product control,
100% inspection
Product control,
calibration tolerance
Table F 1: Examples of the PISMOEA Model
The degree of influence and contribution to measurement variation for
specific error sources will depend on the situation. A matrix, cause and
effect, or fault tree diagram will be a useful tool to identify and understand
the dominant sources of measurement variation for control and improvement.
Measurement systems analysis starts by understanding the purpose and
process of measurement. All sources of chaotic and illegitimate errors ought
to be removed. A measurement study is a planned experiment that follows
simple concepts: DEFINE significant error sources, FIX some, allow one or
more CONTROL factors to change, MEASURE multiple trials, ANALYZE
results, and TAKE ACTION.

Glossary
213
Glossary
See the Statistical Process Control (SPC) Reference Manual for additional glossary definitions.
5.15 vs. 6 GRR
σ Multiplying Factor See note on page iv.
Accuracy The closeness of agreement between an observed value and the
accepted reference value.
Analysis of Variance A statistical method (ANOVA) often used in designed experiments
(DOE), to analyze variable data from multiple groups in order to
compare means and analyze sources of variation.
Apparent Resolution The size of the least increment on the measurement instrument is the
apparent resolution. This value is typically used in literature as
advertisement to classify the measurement instrument. The number of
data categories can be determined by dividing the size into the expected
process distribution spread ( 6σ ).
NOTE: The number of digits displayed or reported does not always
indicate the resolution of the instrument. For example, parts measured
as 29.075, 29.080, 29.095, etc., are recorded as five (5) digit
measurements. However, the instrument may not have a resolution of
.001 but rather .005.
Appraiser Variation The variation in average measurements of the same part (measurand)
between different appraisers (operators) using the same measuring
instrument and method in a stable environment. Appraiser variation
(AV) is one of the common sources of measurement system variation
(error) that results from differences in operator skill or technique using
the same measurement system. Appraiser variation is commonly
assumed to be the “reproducibility error” associated with a
measurement system; this is not always true (see Reproducibility).
Bayes’ Theorem A mathematical formula used for calculating conditional probabilities.
The probability of a hypothesis G conditional on a given body of data
M is the ratio of the unconditional probability of the conjunction of the
hypothesis with the data to the unconditional probability of the data
alone
The probability of G conditional on M is defined as PR
MT
(G) = P(G &
M)/P(G), provided that both terms of this ratio exist and P(G) > 0
Bias The difference between the observed average of measurements (trials
under repeatability conditions) and a reference value; historically
referred to as accuracy. Bias is evaluated and expressed at a single
point within the operating range of the measurement system.
Calibration A set of operations that establish, under specified conditions, the
relationship between a measuring device and a traceable standard of
known reference value and uncertainty. Calibration may also include
steps to detect, correlate, report, or eliminate by adjustment any
discrepancy in accuracy of the measuring device being compared.

Glossary
214
Calibration Interval A specified amount of time or set of conditions between calibrations
during which the calibration parameters of a measuring device are
considered valid.
Capability An estimate of the combined variation of measurement errors (random
and systematic) based on a short-term assessment of the measurement
system.
Confidence Interval An interval or range of values, calculated from sample data, that
contains, with a (100 - α) degree of certainty, the population parameter
of interest, e.g., the true population average.
α, called the Level of Significance, is the probability of committing a
Type I error.
See Montgomery (1997) or Juran and Godfrey (1999) for calculation
methods.
Control Chart A graph of a process characteristic, based on sample measurements in
time order, used to display the behavior of a process, identify patterns
of process variation, assess stability, and indicate process direction.
• It displays the plotted values of some statistic gathered from that
characteristic, a centerline, and one or two control limits.
• It minimizes the net economic loss from Type I and Type II errors.
• It has two basic uses: as a judgment to determine if a process has
been operating in statistical control, and to aid in maintaining
statistical control.
Data A collection of observations under a set of conditions that may be
either variable (a quantified value and unit of measure) or discrete
(attribute or count data such as pass/fail, good/bad, go/no-go, etc.).
Designed Experiment A planned study involving statistical analysis of a series tests in which
purposeful changes are made to process factors, and the effects
observed, in order to determine the relationship of process variables
and improve the process.
Discrimination Alias smallest readable unit, discrimination is the measurement
resolution, scale limit, or smallest detectable unit of the measurement
device and standard. It is an inherent property of gage design and
reported as a unit of measurement or classification. The number of data
categories is often referred to as the discrimination ratio since it
describes how many classifications can be reliably distinguished given
the observed process variation.
Distinct Data Categories The number of data classifications or categories that can be reliably
distinguished determined by the effective resolution of the
measurement system and part variation from the observed process for a
given application. See ndc.
Effective Resolution The size of the data category when the total measurement system
variation is considered is the effective resolution. This size is
determined by the length of the confidence interval based on the
measurement system variation. The number of distinct categories, ndc,
can be determined by dividing the size into the expected process
distribution spread. For the effective resolution, a standard estimate of
this ndc (at the 97% confidence level) is 1.41[PV/GRR]. (See Wheeler,
1989, for an alternate interpretation.)

Glossary
215
F ratio A statistic representing the mathematical ratio of the between-group
mean square error to the within-group mean square error for a set of
data used to assess the probability of random occurrence at a selected
level of confidence.
Gage R&R (GRR) An estimate of the combined variation of repeatability and
reproducibility for a measurement system. The GRR variance is equal
to the sum of within-system and between-system variances.
Histogram A graphical representation (bar chart) of the frequency of grouped data
to provide a visual evaluation of the data distribution.
In Control State of a process when it exhibits only random, common cause
variation (as opposed to chaotic, assignable, or special cause variation).
A process operating with only random variation is statistically stable.
Independent The occurrence of one event or variable has no effect on the probability
that another event or variable will occur.
Independent and Identically Distributed Commonly referred to as “iid”. A homogeneous group of data which
are independent and randomly distributed in one common distribution.
Interaction A combined effect or outcome resulting from two or more variables
that is significant. Non-additivity between appraiser and part. Appraiser
differences depend on the part being measured.
Inter-rater agreement (Also inter-rater reliability, or concordance) The degree of agreement
among raters. It gives a score of how much homogeneity, or consensus,
there is in the ratings given by the appraisers.
There are a number of statistics which can be used to determine inter-
rater reliability which are appropriate for different types of
measurement. Some options are: joint-probability of agreement,
Cohen's kappa and the related Fleiss' kappa, inter-rater correlation,
concordance correlation coefficient and intra-class correlation.
Kappa (Cohen’s) A statistical measure of inter-rater agreement for qualitative
(categorical) items. It takes into account the agreement occurring by
chance.
Linearity The difference in bias errors over the expected operating range of the
measurement system. In other terms, linearity expresses the correlation
of multiple and independent bias errors over the operating range.
Long-Term Capability Statistical measure of the within-subgroup variation exhibited by a
process over a long period of time. This differs from performance
because it does not include the between-subgroup variation.
Measurand The particular quantity or subject to be measured under specified
conditions; a defined set of specifications for a measurement
application.
Measurement system A collection of instruments or gages, standards, operations, methods,
fixtures, software, personnel, environment, and assumptions used to
quantify a unit of measure or fix assessment to the feature characteristic
being measured; the complete process used to obtain measurements.

Glossary
216
Measurement System Error The combined variation due to gage bias, repeatability, reproducibility,
stability and linearity.
Metrology The science of measurement.
ndc Number of distinct categories.  
1.41 PV
GRR
Non-replicable The inability to make repeated measurements on the same sample or
component due to the dynamic nature of the measurand.
Number of Distinct Categories See ndc
Out-of-Control State of a process when it exhibits chaotic, assignable , or special cause
variation. A process that is out of control is statistically unstable.
Part Variation Related to measurement systems analysis, part variation (PV)
represents the expected part-to-part and time-to-time variation for a
stable process.
Part-to-Part Variation Piece-to-piece variation due to measuring different parts.
Performance An estimate of the combined variation of measurement errors (random
and systematic) based on a long-term assessment of the measurement
system; includes all significant and determinable sources of variation
over time.
Precision The net effect of discrimination, sensitivity and repeatability over the
operating range (size, range and time) of the measurement system. In
some organizations precision is used interchangeability with
repeatability. In fact, precision is most often used to describe the
expected variation of repeated measurements over the range of
measurement; that range may be size or time. The use of the more
descriptive component terms is generally preferred over the term
“precision”.
Probability An estimate (in proportion or fraction), based on a particular
distribution of collected data, describing the chance a specific event
will occur. Probability estimates range between 0 (impossible event) to
1(sure thing). Set of conditions or causes working together to produce
an outcome.
Process Control Operational state when the purpose of measurement and decision
criteria applies to real-time production to assess process stability and
the measurand or feature to the natural process variation; the
measurement result indicates the process is either stable and “in-
control” or “out-of-control.”
Product Control Operational state when the purpose of measurement and decision
criteria is to assess the measurand or feature for conformance to a
specification; the measurement result is either “in-tolerance” or “out-
of-tolerance.”

Glossary
217
Reference Value A measurand value that is recognized and serves as an agreed upon
reference or master value for comparison:
• A theoretical or established value based on scientific
principles;
• An assigned value based on some national or international
organization;
• A consensus value based on collaborative experimental work
under the auspices of a scientific or engineering group; or
• For a specific application, an agreed upon value obtained
using an accepted reference method.
A value consistent with the definition of a specific quantity and
accepted, sometimes by convention, as appropriate for a given purpose.
NOTE: Other terms used synonymously with reference value
accepted reference value
accepted value
conventional value
conventional true value
assigned value
best estimate of the value
master value
master measurement
Regression Analysis A statistical study of the relationship between two or more variables. A
calculation to define the mathematical relationship between two or
more variables.
Repeatability The common cause, random variation resulting from successive trials
under defined conditions of measurement. Often referred to as
equipment variation (EV), although this is misleading. The best term
for repeatability is within-system variation when the conditions of
measurement are fixed and defined – fixed part, instrument, standard,
method, operator, environment, and assumptions. In addition to within-
equipment variation, repeatability will include all within variation from
the conditions in the measurement error model.
Replicable The ability to make repeated measurements on the same sample or
component where there is no significant physical change to the
measurand or measurement environment.
Replication Multiple test trials under repeatable (identical) conditions.
Reproducibility The variation in the average of measurements caused by a normal
condition(s) of change in the measurement process. Typically, it has
been defined as the variation in average measurements of the same part
(measurand) between different appraisers (operators) using the same
measuring instrument and method in a stable environment. This is often
true for manual instruments influenced by the skill of the operator. It is
not true, however, for measurement processes (i.e., automated systems)
where the operator is not a major source of variation. For this reason,
reproducibility is referred to as the average variation between-systems
or between-conditions of measurement.

Glossary
218
Resolution May apply to measurement resolution or effective resolution
The capability of the measurement system to detect and faithfully
indicate even small changes of the measured characteristic. (See also
discrimination.)
The resolution of a measurement system is if there is an equal
probability that the indicated value of any part which differs from a
reference part by less than  will be the same as the indicated value of
the reference part. The resolution of a measurement system is impacted
by the measurement instrument as well as other sources of variation of
the total measurement system.
Scatter Diagram A X-Y plot of data to assess the relationship between two variables.
Sensitivity Smallest input signal that results in a detectable (discernable) output
signal for a measurement device. An instrument should be at least as
sensitive as its unit of discrimination. Sensitivity is determined by
inherent gage design and quality, in-service maintenance, and operating
condition. Sensitivity is reported in units of measurement.
Significance level A statistical level selected to test the probability of random outcomes;
also associated with the risk, expressed as the alpha (α ) risk, that
represents the probability of a decision error.
Stability The absence of special causes of variation; the property of being in
statistical control.
Refers to both statistical stability of a measurement process and
measurement stability over time. Both are vital for a measurement
system to be adequate for its intended purpose. Statistical stability
implies a predictable, underlying measurement process operating
within common cause variation (in-control). Measurement stability
(alias drift) addresses the necessary conformance to the measurement
standard or reference over the operating life (time) of the measurement
system.
Tolerance Allowable deviation from a standard or nominal value that maintains
fit, form, and function.
See also Specification
Uncertainty A parameter associated with the result of a measurement that
characterizes the dispersion of the values that could reasonably be
attributed to the measurand (VIM); the range assigned to a
measurement result that describes, within a defined level of confidence,
the limits expected to contain the true measurement result. Uncertainty
is a quantified expression of measurement reliability.
Unimodal A contiguous group of data (distribution) that has one mode

Reference List
219
Reference List
1. ASTM E691-99, Standard Practice for Conducting an Interlaboratory Study to Determine the
Precision of a Test Method, (www.astm.org).
2. ASTM E177-90a, Standard Practice for Use of the Terms Precision and Bias in ASTM Test
Method, (www.astm.org).
3. ASTM E456-96, Standard Terminology for Relating to Quality and Statistics, (www.astm.org).
4. AT&T Statistical Quality Control Handbook, Delmar Printing Company, Charlotte, NC, 1984.
5. Baker, E. “Signal Detection Theory of Quality Control Inspector Performance”, Journal of
Quality Technology, Vol. 7, No.2, April 1975.
6. Bland, J. M., and Altman, D. G. (1986). Statistical methods for assessing agreement between two
methods of clinical measurement. Lancet i, pp. 307–310.
7. Burdick, Richard K., and Larsen, Greg A., “Confidence Intervals on Measures of Variability in
R&R Studies,” Journal of Quality Technology, Vol. 29, No. 3, July 1997.
8. Cohen, J. (1960) "A coefficient for agreement for nominal scales" in Education and
Psychological Measurement. Vol. 20, pp. 37–46.
9. Dataplot, National Institute of Standards and Technology, Statistical Engineering Division,
(www.itl.nist.gov).
10. Deming, W. E., Out of the Crisis, Massachusetts Institute of Technology, 1982, 1986.
11. Deming, W. E., The New Economics for Industry, Government, Education, The MIT Press, 1994,
2000.
12. Duncan, A. J., Quality Control and Industrial Statistics, 5th ed., Richard D. Irwin, Inc.,
Homewood, Illinois, 1986.
13. Eagle, A. R., “A Method For Handling Errors in Testing and Measuring,” Industrial Quality
Control, March, 1954, pp. 10-14.
14. Ermer, Donald S., and E-Hok, Robin Yang, “Reliable Data is an Important Commodity,”
University of Wisconsin, Madison; published in The Standard, ASQ Newsletter of the
Measurement Quality Division, Vol. 97-1, Winter 1997.
15. Everitt, B. (1996) Making Sense of Statistics in Psychology (Oxford University Press).
16. Fleiss, J. L. (1971) "Measuring nominal scale agreement among many raters" in Psychological
Bulletin. Vol. 76, No. 5, pp. 378–382.
17. Foster, Lowell W., GEO-METRICS: The Metric Application of Geometric Tolerancing, Addison-
Wesley Publishing Company, Reading, Massachusetts, 1993.

Reference List
220
18. Gillespie, Richard, A History of the Hawthorne Experiments, Cambridge University Press, New
York, 1991.
19. Grubbs, F. E., “On Estimating Precision of Measuring Instruments and Product Variability,”
Journal of the American Statistical Association, Vol. 43, 1948, pp.243-264.
20. Grubbs, F. E., “Errors of Measurement, Precision, Accuracy and the Statistical Comparison of
Measuring Instruments,” Technometrics, Vol. 15, February 1973, pp. 53-66.
21. Gruska, G. F., and Heaphy, M. S., “Measurement Systems Analysis,” TMI Conference Workshop,
Detroit, Michigan, 1987.
22. Guide to the Expression of Uncertainty in Measurement, ISO/IEC Guide 98:1995.
23. Hahn, J. H. and Nelson, W., “A Problem in the Statistical Comparison of Measuring Devices,”
Technometrics, Vol. 12., No. 1, February 1970, pp. 95-102.
24. Hamada, Michael, and Weerahandi, Sam, “Measurement System Assessment Via Generalized
Inference,” Journal of Quality Technology, Vol. 32, No. 3, July 2000.
25. Hayes, A. F. & Krippendorff, K. (2007). Answering the call for a standard reliability measure for
coding data. Communication Methods and Measures, 1, 77-89.
26. Heaphy, M. S., et. al., “Measurement System Parameters,” Society of Manufacturing Engineering
-IQ81-154, Detroit, Michigan, 1981.
27. Hicks, Charles R.and Turner K.V. Jr. Fundamental Concepts in the Design of Experiments,
Oxford University Press, New York, 1999.
28. International Vocabulary of Basic and General Terms in Metrology (abbr. VIM), ISO/IEC Guide
99:2007.
29. Ishikawa, Kaoru, Guide to Quality Control, Asian Productivity Organization, 1986.
30. Jaech, J. L., “Further Tests of Significance for Grubbs’ Estimators,” Biometrics, 27, December,
1971, pp. 1097-1101.
31. James, P. D., “Graphical Displays of Gage R&R Data,” AQC Transaction, ASQC, 1991.
32. Krippendorff, K. (2004). Content analysis: An introduction to its methodology. Thousand Oaks,
CA: Sage. pp. 219-250.
33. Lipson, C., and Sheth, N. J., Statistical Design & Analysis of Engineering Experiment, McGraw-
Hill, New York, 1973.
34. Maloney, C. J., and Rostogi, S. C., “Significance Test for Grubbs’ Estimators,” Biometrics, 26,
December, 1970. pp. 671-676.
35. Mandel, J., “Repeatability and Reproducibility,” Journal of Quality Technology, Vol.4, No.2,
April, 1972, pp.74-85.

Reference List
221
36. McCaslin, J. A., and Gruska, G. F., “Analysis of Attribute Gage Systems,” ASQC Technical
Conference Transactions, 1976, pp. 392-399.
37. Measurement Systems Analysis Reference Manual, Third Edition, Chrysler, Ford, GM, Third
Printing, 2002.
38. Minitab User’s Guide: Guide to Minitab: Release 15 - Minitab , Minitab, Inc., 2007.
39. Montgomery, Douglas C., and Runger, George C., “Gauge Capability and Designed Experiments.
Part I: Basic Methods; Part II: Experimental Design Models and Variance Component
Estimation,” Quality Engineering, Marcel Dekker, Inc., 1993.
40. Nelson, L., “Use of the Range to Estimate Variability,” Journal of Quality Technology, Vol. 7,
No.1, January, 1975, pp.46-48.
41. Potential Failure Mode and Effects Analysis (FMEA) Reference Manual, Fourth Edition,
DaimlerChrysler, Ford, GM, 2008.
42. Production Part Approval Process (PPAP), Fourth Edition, DaimlerChrysler, Ford, GM, 2008.
43. Saal, F.E., Downey, R.G. and Lahey, M.A (1980) "Rating the Ratings: Assessing the
Psychometric Quality of Rating Data" in Psychological Bulletin. Vol. 88, No. 2, pp. 413–428.
44. Shrout, P. and Fleiss, J. L. (1979) "Intraclass correlation: uses in assessing rater reliability" in
Psychological Bulletin. Vol. 86, No. 2, pp. 420–428.
45. Spiers, Beryl, “Analysis of Destructive Measuring Systems,” ASQC Quality Congress
Transactions, Toronto, 1989.
46. Statistical Process Control Reference Manual (SPC), Second Edition, Chrysler, Ford, GM, 2005.
47. Thompson, W. A., Jr., “The Problem of Negative Estimates of Variance Components,” Annals of
Mathematical Statistics, No. 33, 1962, pp. 273-289.
48. Thompson, W. A., Jr., “Precision of Simultaneous Measurement Procedures”, American
Statistical Association Journal, June, 1963, pp. 474-479.
49. Traver, R. W., “Measuring Equipment Repeatability – The Rubber Ruler,” 1962 ASQC
Convention Transactions, pp. 25-32.
50. Tsai, Pingfang, “Variable Gauge Repeatability and Reproducibility Study Using the Analysis of
Variance Method,” Quality Engineering, Marcel Dekker, Inc., 1988.
51. Uebersax, John S. (1987). "Diversity of decision making models and the measurement of
interrater agreement" in Psychological Bulletin. Vol 101, pp. 140–146.
52. Vardeman, Stephen B., and VanValkenburg, Enid S., “Two-Way Random-Effects Analyses and
Gauge R&R Studies,” Technometrics, Vol. 41, No. 3, August 1999.
53. Wheeler, D. J., “Evaluating the Measurement Process When the Testing is Destructive,” TAPPI,
1990 Polymers Lamination & Coatings Conference Proceedings, (www.tappi.org search for
document plc90805).

Reference List
222
54. Wheeler, D. J., and Lyday, R. W., Evaluating the Measurement Process, SPC Press, Inc.,
Knoxville, Tennessee, 2006 .

223
Sample Forms
The reader has permission to reproduce the forms in this section for internal use only, not for
commercial use or resale.
The forms in this section each represent a possible format for GRR data collection and reporting.
They are not mutually exclusive to other types of formats which may contain the same
information and accomplish the same results.

224
PART
Appraiser
/Trial # 1 2 3 4 5 6 7 8 9 10
AVERAGE
A 1
2
3
Average a
X =
Range a
R =
B 1
2
3
Average b
X =
Range b
R =
C 1
2
3
Average
c
X =
Range c
R =
Part Average X =
p
R =
([ a
R = ] + [ b
R = ] + [ c
R = ]) / [ # OF APPRAISERS = ] = R =
DIFF
X [Max X = ] - [Min X = ] =
= DIFF
X =
* R 4
UCL [R = ] [D ] =
=  =
beyond this limit. Identify the cause and correct. Repeat these readings using the same appraiser and unit as originally used
or
Notes:__________________________________________________________________________________________________________
___

225
Gage Repeatability and Reproducibility Report
Part No. & Name:
Characteristics:
Specifications:
From data sheet: R =
Gage Name:
Gage No:
Gage Type:
DIFF
X =
Date:
Performed by:
p
R =
Measurement Unit Analysis % Total Variation (TV)
Repeatability – Equipment Variation (EV)
EV = R  K1
Trials K1 %EV = 100 [EV/TV]
= _______  _______ 2 0.8862 = 100 [_______/_______]
= _______ 3 0.5908 = _______%
Reproducibility – Appraiser Variation (AV)
AV =  2 2
2
( ( ))
-

DIFF
X K EV nr %AV = 100 [AV/TV]
=  2 2
_____ _____ ( _____ (__ __))
-
  = 100 [_______/_______]
= _______ Appraisers 2 3 = _______%
n = parts r = trials K2 0.7071 0.5231
Repeatability & Reproducibility (GRR)
GRR =
2 2

EV AV %GRR= 100 [GRR/TV]
=  
2 2
_____ _____
 Parts K3
= 100 [_______/_______]
= _______ 2 0.7071 = _______%
Part Variation (PV) 3 0.5231
PV = p
R  K3 4 0.4467 % PV = 100 [PV/TV]
= _______  _______ 5 0.4030 = 100 [_______/_______]
= _______ 6 0.3742 = _______%
Total Variation (TV) 7 0.3534
TV =
2 2

GRR PV 8 0.3375 ndc =  
1.41 PV
GRR
=  
2 2
_____ _____
 9 0.3249 = 1.41(_______/_______)
= _______ 10 0.3146 = _______
For information on the theory and constants used in the form see MSA Reference Manual, Fourth edition.

Index
228
Acceptability Criteria, 79
Accuracy (also see Bias), 6, 50, 215, 222
Analysis Of Variance (ANOVA), 103, 105, 109, 125,
126, 127, 128, 129, 130, 131, 161, 163, 165, 169,
197, 199, 200, 212, 215
Analytic Method, 147
Apparent, 215
Appraiser Variation (AV), 7, 122, 124, 125, 129, 131,
197, 198, 200, 209, 215, 227
ASTM (American Society For Testing And Materials),
5, 6, 7, 33, 45, 50, 54, 55, 56, 187, 189, 192, 221
Attribute, 133, 136, 147, 151, 152, 161, 179, 181,
223
Average, 95, 103, 105, 108, 109, 114, 115, 119, 126,
128, 129, 130, 131, 171, 176, 177, 188, 226
Bias, 6, 50, 51, 61, 62, 89, 90, 91, 92, 93, 94, 96, 97,
99, 215, 221
Calibration, 10, 34, 36, 38, 39, 43, 213, 215, 216
Calibration Standard, 43
Capability Index, 201
Cause and Effect, 17, 68
Cause and Effect Diagram, 17, 68
Charts, Error, 114
Check Standard, 45
Consistency, 8, 57, 110, 112, 113, 116, 166, 187, 192
Control Chart Analysis, 88
Control Charts, 49
Coordinate Measuring Machine (CMM), 14, 25, 26,
37, 38, 59, 134, 169, 213
Cp (see Capability Index), 21, 22, 143, 201, 203, 204
Cross‐tabulation, 135
Customer, 4, 48, 91, 96, 233
Designed Experiment, 216, 223
Destructive measurement systems, 155, 156
impact testing (gravelometer), 155
mass spectroscopy, 155
material characteristic testing, 155
plating testing, 155
salt spray/humidity booth testing, 155
weld testing, 155
Discrimination, 5, 41, 46, 216
Drift (see Stability), 6, 39
Effective Resolution, 38, 216
Effectiveness, 141, 142
Environment, 16, 37, 51, 54, 211
Equipment Variation (EV) (see also Repeatability),
227
Ergonomics, 38
Error, 19, 39, 79, 80, 97, 114, 115, 182, 209, 211,
213, 218
Error Charts, 114
Error Rate, 19
EV (see Equipment Variation), 54, 90, 93, 95, 99,
122, 124, 125, 129, 131, 197, 198, 200, 209, 219,
227
False Alarm Rate, 19, 142
Fixturing, 37
FMEA, 13, 26, 37, 169, 223
Gage, 5, 7, 31, 35, 42, 56, 61, 79, 81, 90, 103, 105,
107, 118, 121, 122, 125, 131, 147, 148, 150, 151,
152, 161, 179, 180, 181, 182, 183, 184, 198, 207,
211, 212, 213, 217, 222, 223, 226, 227
Gage Performance Curve (GPC), 42, 147, 148, 150,
151, 152, 179, 180, 181, 182, 183, 184
Gage Repeatability (GR), 35, 107, 118, 120, 121, 122,
125, 226, 227
Gage Repeatability and Reproducibility (GRR), 7, 56,
107, 118, 121, 122, 125, 161, 198, 212, 217, 222,
226, 227
Geometric Dimensioning and Tolerancing (GD&T),
169
GPC (see Gage Performance Curve), 42, 148, 150,
179, 181
GRR (see Gage Repeatability and Reproducibility), 7,
18, 22, 35, 56, 58, 60, 64, 75, 76, 79, 80, 81, 90,
95, 99, 104, 105, 106, 122, 123, 124, 125, 129,
130, 131, 145, 146, 150, 163, 165, 181, 185, 186,
187, 189, 193, 197, 198, 199, 200, 201, 202, 203,
212, 216, 217, 225, 227
Guide to the Expression of Uncertainty in
Measurement, 222
Histogram, 93, 115, 217
Hypothesis Test Analyses, 133, 135
Kappa, 138, 139, 217
Linearity, 6, 51, 52, 98, 100, 101, 102, 116, 164, 207,
217
Location Error, 79
Maintenance, 34, 36, 38
Master, 44, 213
Measurand, 60, 217
Measurement Life Cycle, 26
Measurement System, 1, 7, 8, 13, 17, 18, 28, 31, 37,
38, 39, 67, 68, 69, 75, 79, 83, 87, 131, 133, 134,
155, 156, 171, 187, 212, 218, 222, 223
Measurement System Performance, 8
Measurement Uncertainty, 63
Miss Rate, 19, 142
NIST (National Institute of Standards and
Technology), 9, 159, 221

Index
229
NIST (see National Institute of Standards and
Technology), 9, 10, 11, 34, 45, 73
NMI (National Measurements Institute), 9, 10, 45
Non‐replicable, 218
Number of Distinct Categories (ndc), 47, 80, 125,
131, 216, 218, 227
Observed Value, 114, 115
Operational Definition, 43
P.I.S.M.O.E.A., 16, 37, 211
P.I.S.M.O.E.A. Error Model, 211
Performance Curve, 42, 147, 148, 150, 151, 152,
179, 180, 181, 182, 183, 184
Pooled Standard Deviation, 187, 190, 191
Precision, 7, 54, 218, 221, 222, 223
Process Control, 49, 75, 131, 215, 218, 223
Process Failure Mode Effects Analysis (PFMEA), 13
Product Control, 75, 218
R&R (see Gage Repeatability and Reproducibility or
GRR), 7, 56, 163, 198, 212, 217, 221, 222, 223
Random, 213, 223
Randomization, 76, 106
Range, 38, 90, 95, 103, 104, 105, 110, 111, 124, 126,
128, 129, 130, 131, 160, 163, 171, 176, 177, 188,
203, 207, 223, 226
Range Chart, 110, 111
Range Method, 90, 104, 105, 128, 131, 163, 171, 188
Reference Value, 45, 92, 101, 114, 115, 145, 219
Replicable, 83, 155, 219
Reproducibility, 7, 55, 77, 103, 107, 118, 121, 122,
125, 131, 189, 198, 207, 215, 219, 222, 223, 226,
227
Resolution, 38, 215, 216, 220
Resolution, Apparent, 215
Resolution, Effective, 38, 216
Risk Analysis Methods, 133
Rule of Tens, 15, 41
Run Chart, 111
S.W.I.P.E., 16, 37
Scatter Plot, 112
Sensitivity, 8, 38, 57, 220
Signal Detection Approach, 145
Standard, 5, 16, 43, 44, 45, 105, 163, 165, 176, 177,
187, 190, 191, 198, 209, 211, 213, 221
Standard, Calibration, 43
Standard, Check, 45
Standard, Reference, 43
Standard, Transfer, 44
Standard, Working, 44
Statistical Process Control (SPC), 11, 47, 49, 75, 79,
88, 118, 213, 215, 223, 224
Test Stands, 155, 156, 161
Tolerance, 124, 131, 145, 203, 220
Total Variation (TV), 15, 76, 90, 95, 99, 119, 123,
124, 125, 129, 131, 200, 227
Traceability, 9, 10, 64
Transfer Standard, 44
True Value, 45
Uncertainty (see Measurement Uncertainty), 8, 60,
63, 64, 65, 220, 222
Uniformity, 8, 58
Variable Gage Study, 103
Variation, 7, 15, 50, 76, 103, 119, 123, 124, 129, 131,
169, 185, 198, 200, 215, 218, 227
Whiskers Chart, 113
Width Error, 79, 80
Working Standard, 44

Feedback
231
M.S.A. Manual User Feedback Process
Consistent with the concept of continual improvement, this automotive industry measurement systems analysis
(MSA) manual is being subjected to a formal periodic review/revision process. In line with the concept of customer
satisfaction, this review will entail consideration of not only any applicable vehicle manufacturer requirement
changes from year to year but also of feedback from users of the manual for the purpose of making it more value-
added and effective to the automotive industry and user communities. Accordingly, please feel free to offer, in
writing, your feedback comments, both pro and con, relative to the manual’s understandability, “user-friendliness,”
etc., in the area indicated below. Please indicate specific manual page numbers where appropriate. Forward your
feedback to the address indicated below:
Your Name _____________________________________________________
Representing ________________________________________________
Company/Division Name
Address _____________________________________________________
_____________________________________________________
_____________________________________________________
Phone ___(____)________________
Please list your top three automotive customers and their locations.
________________________ ________________________
Customer Location
________________________ ________________________
Customer Location
________________________ ________________________
Customer Location
Feedback Comments (attach additional sheets if needed)
_____________________________________________________
_____________________________________________________
_____________________________________________________
_____________________________________________________
_____________________________________________________
Send Comments To:
Automotive Industry Action Group
Suite 200
26200 Lahser Road
Southfield, Michigan 48033
MSA, 4th
Please access www.aiag.org to submit your feedback electronically.

MSA (Measurement Systems Analysis) 4th
Edition Errata Sheet
5) Compute the average bias of the n readings.
1
n
i
i
bias
avg bias
n
=
=
∑
6) Compute the repeatability standard deviation (see also Gage Study,
Range Method, below):
2
1
( )
1
n
i
i
repeatability r
X X
n
σ σ =
−
= =
−
∑
If a GRR study is available (and valid), the repeatability standard
deviation calculation should be based on the study results.
σ /TV]
%EV will lead to misleading and confusing results.
8) Determine the t statistic for the bias: 34
r
b
n
=
σ
σ
bias
b
average bias
t statistic t
σ
= =
9) Bias is acceptable (statistically zero) at the α level if
• the p-value associated with bias
t is more than α; or
b
34
The uncertainty for bias is given byσ .

MSA_Reference_Manual_4th_Edition.pdf

More Related Content

What's hot (20)

Similar to MSA_Reference_Manual_4th_Edition.pdf (20)

Recently uploaded (20)

MSA_Reference_Manual_4th_Edition.pdf