This international standard was developed in accordance with internationally recognized principles on standardization established in the Decision on Principles

for the
Development of International Standards, Guides and Recommendations issued by the World Trade Organization Technical Barriers to Trade (TBT) Committee.

Designation: D6299 − 22 An American National Standard

Standard Practice for

Applying Statistical Quality Assurance and Control Charting
Techniques to Evaluate Analytical Measurement System
Performance1
This standard is issued under the fixed designation D6299; the number immediately following the designation indicates the year of
original adoption or, in the case of revision, the year of last revision. A number in parentheses indicates the year of last reapproval. A
superscript epsilon (´) indicates an editorial change since the last revision or reapproval.

1. Scope* NOTE 4—For non-Gaussian processes, transformations of test results

may permit proper application of these tools. Consult a statistician for
1.1 This practice covers information for the design and further guidance and information.
operation of a program to monitor and control ongoing stability 1.7 This international standard was developed in accor-
and precision and bias performance of selected analytical dance with internationally recognized principles on standard-
measurement systems using a collection of generally accepted ization established in the Decision on Principles for the
statistical quality control (SQC) procedures and tools. Development of International Standards, Guides and Recom-
NOTE 1—A complete list of criteria for selecting measurement systems mendations issued by the World Trade Organization Technical
to which this practice should be applied and for determining the frequency Barriers to Trade (TBT) Committee.
at which it should be applied is beyond the scope of this practice.
However, some factors to be considered include (1) frequency of use of 2. Referenced Documents
the analytical measurement system, (2) criticality of the parameter being
measured, (3) system stability and precision performance based on 2.1 ASTM Standards:2
historical data, (4) business economics, and (5) regulatory, contractual, or D3764 Practice for Validation of the Performance of Process
test method requirements. Stream Analyzer Systems
1.2 This practice is applicable to stable analytical measure- D4175 Terminology Relating to Petroleum Products, Liquid
ment systems that produce results on a continuous numerical Fuels, and Lubricants
scale. D5191 Test Method for Vapor Pressure of Petroleum Prod-
1.3 This practice is applicable to laboratory test methods. ucts and Liquid Fuels (Mini Method)
D6300 Practice for Determination of Precision and Bias
1.4 This practice is applicable to validated process stream Data for Use in Test Methods for Petroleum Products,
analyzers. Liquid Fuels, and Lubricants
1.5 This practice is applicable to monitoring the differences D6617 Practice for Laboratory Bias Detection Using Single
between two analytical measurement systems that purport to Test Result from Standard Material
measure the same property provided that both systems have D6708 Practice for Statistical Assessment and Improvement
been assessed in accordance with the statistical methodology in of Expected Agreement Between Two Test Methods that
Practice D6708 and the appropriate bias applied. Purport to Measure the Same Property of a Material
NOTE 2—For validation of univariate process stream analyzers, see also D6792 Practice for Quality Management Systems in Petro-
Practice D3764. leum Products, Liquid Fuels, and Lubricants Testing
NOTE 3—One or both of the analytical systems in 1.5 may be laboratory
Laboratories
test methods or validated process stream analyzers.
D7372 Guide for Analysis and Interpretation of Proficiency
1.6 This practice assumes that the normal (Gaussian) model Test Program Results
is adequate for the description and prediction of measurement D7915 Practice for Application of Generalized Extreme
system behavior when it is in a state of statistical control. Studentized Deviate (GESD) Technique to Simultane-
ously Identify Multiple Outliers in a Data Set
E177 Practice for Use of the Terms Precision and Bias in
ASTM Test Methods
E178 Practice for Dealing With Outlying Observations
1
This practice is under the jurisdiction of ASTM Committee D02 on Petroleum
Products, Liquid Fuels, and Lubricants and is the direct responsibility of Subcom-
2
mittee D02.94 on Coordinating Subcommittee on Quality Assurance and Statistics. For referenced ASTM standards, visit the ASTM website, www.astm.org, or
Current edition approved May 1, 2022. Published May 2022. Originally contact ASTM Customer Service at [email protected]. For Annual Book of ASTM
approved in 1998. Last previous edition approved in 2021 as D6299 – 21. DOI: Standards volume information, refer to the standard’s Document Summary page on
10.1520/D6299-22. the ASTM website.

*A Summary of Changes section appears at the end of this standard

Copyright © ASTM International, 100 Barr Harbor Drive, PO Box C700, West Conshohocken, PA 19428-2959. United States

1
 D6299 − 22
E456 Terminology Relating to Quality and Statistics results from multiple individual analytical systems or different
E691 Practice for Conducting an Interlaboratory Study to instruments executing the same test method.
Determine the Precision of a Test Method 3.2.5 assignable cause, n—a factor that contributes to varia-
tion and that is feasible to detect and identify.
3. Terminology
3.2.6 bias, n—a systematic error that contributes to the
3.1 Definitions: difference between a population mean of the measurements or
3.1.1 More extensive lists of terms related to quality and test results and an accepted reference or true value.
statistics are found in Terminology D4175, Practice D6300,
and Terminology E456. 3.2.7 blind submission, n—submission of a check standard
3.1.2 repeatability conditions, n—conditions where inde- or quality control (QC) sample for analysis without revealing
pendent test results are obtained with the same method on the expected value to the person performing the analysis.
identical test items in the same laboratory by the same operator 3.2.8 check standard, n—in QC testing, a material having an
using the same equipment within short intervals of time. accepted reference value used to determine the accuracy of a
D6300 measurement system.
3.1.3 reproducibility (R), n—a quantitative expression for 3.2.8.1 Discussion—A check standard is preferably a mate-
the random error associated with the difference between two rial that is either a certified reference material with traceability
independent results obtained under reproducibility conditions to a nationally recognized body or a material that has an
that would be exceeded with an approximate probability of 5 % accepted reference value established through interlaboratory
(one case in 20 in the long run) in the normal and correct testing. For some measurement systems, a pure, single com-
operation of the test method. D6300 ponent material having known value or a simple gravimetric or
volumetric mixture of pure components having calculable
3.1.4 reproducibility conditions, n—conditions where inde-
value may serve as a check standard. Users should be aware
pendent test results are obtained with the same method on
that for measurement systems that show matrix dependencies,
identical test items in different laboratories with different
accuracy determined from pure compounds or simple mixtures
operators using different equipment.
may not be representative of that achieved on actual samples.
3.1.4.1 Discussion—Different laboratory by necessity 3.2.9 common (chance, random) cause, n—for quality as-
means a different operator, different equipment, and different surance programs, one of generally numerous factors, individu-
location and under different supervisory control. D6300 ally of relatively small importance, that contributes to
3.2 Definitions of Terms Specific to This Standard: variation, and that is not feasible to detect and identify.
3.2.1 More extensive lists of terms related to quality and
statistics are found in Terminology D4175, Practice D6300, 3.2.10 control limits, n—limits on a control chart that are
and Terminology E456. used as criteria for signaling the need for action or for judging
whether a set of data does or does not indicate a state of
3.2.2 accepted reference value, n—a value that serves as an
statistical control.
agreed-upon reference for comparison and that is derived as (1)
a theoretical or established value, based on scientific principles, 3.2.11 double blind submission, n—submission of a check
(2) an assigned value, based on experimental work of some standard or QC sample for analysis without revealing the check
national or international organization, such as the U.S. Na- standard or QC sample status and expected value to the person
tional Institute of Standards and Technology (NIST), or (3) a performing the analysis.
consensus value, based on collaborative experimental work 3.2.12 in-statistical-control, adj—a process, analytical mea-
under the auspices of a scientific or engineering group. surement system, or function that exhibits variations that can
3.2.3 accuracy, n—the closeness of agreement between an only be attributable to common cause.
observed value and an accepted reference value. 3.2.13 lot, n—a definite quantity of a product or material
3.2.4 analytical measurement system, n—a collection of one accumulated under conditions that are considered uniform for
or more components or subsystems, such as samplers, test sampling purposes.
equipment, instrumentation, display devices, data handlers, 3.2.14 out-of-statistical-control, adj—a process, analytical
printouts or output transmitters, that is used to determine a measurement system, or function that exhibits variations in
quantitative value of a specific property for an unknown addition to those that can be attributable to common cause and
sample in accordance with a test method. the magnitude of these additional variations exceed specified
3.2.4.1 Discussion—A standard test method (for example, limits.
ASTM, ISO) executed at a single site using a specific instru-
ment may be an example of an analytical measurement system. 3.2.14.1 Discussion—For clarification, a transition from an
3.2.4.2 Discussion—The control chart methodology and in-statistical-control system to an out-of-statistical-control sys-
work processes described in this practice are intended to be tem does not automatically imply that there is a change in the
applied independently to the final results produced from each fit for use status of the system in terms of meeting the
individual measurement system, or, differences between results requirements for the intended application.
from two individual measurement systems for the same test 3.2.15 precision, n—the closeness of agreement between
sample. They are not intended to be applied to combined final test results obtained under prescribed conditions.

2
 D6299 − 22
3.2.16 proficiency testing, n—determination of a laborato- similar in composition and property level to the QC samples
ry’s testing capability by participation in an interlaboratory used to establish the standard deviation.
crosscheck program. 3.2.22 validation audit sample, n—a QC sample or check
3.2.16.1 Discussion—ASTM Committee D02 conducts pro- standard used to verify precision and bias estimated from
ficiency testing among hundreds of laboratories, using a wide routine quality assurance testing.
variety of petroleum products and lubricants.
3.3 Symbols:
3.2.17 quality control (QC) sample, n—for use in quality 3.3.1 ARV—accepted reference value.
assurance programs to determine and monitor the precision and 3.3.2 EWMA—exponentially weighted moving average.
stability of a measurement system, a stable and homogeneous
material having physical or chemical properties, or both, 3.3.3 I—individual observation (as in I-chart).
similar to those of typical samples tested by the analytical 3.3.4 MR—moving range.
measurement system; the material is properly stored to ensure ¯ —average of moving range.
3.3.5 MR
sample integrity, and is available in sufficient quantity for
repeated, long term testing. 3.3.6 QC—quality control.
3.3.7 R'—site precision.
3.2.18 system expected value (SEV), n—for a QC sample
this is an estimate of the theoretical limiting value towards 3.3.8 SEV—site expected value.
which the average of results collected from a single in- 3.3.9 σR'—site precision standard deviation.
statistical-control measurement system under site precision 3.3.10 VA—validation audit.
conditions tends as the number of results approaches infinity.
3.3.11 χ2—chi squared.
3.2.18.1 Discussion—The SEV is associated with a single
measurement system; for control charts that are plotted in 3.3.12 λ—lambda.
actual measured units, the SEV is required, since it is used as
a reference value from which upper and lower control limits for 4. Summary of Practice
the control chart specific to a batch of QC material are 4.1 QC samples and check standards are regularly analyzed
constructed. by the measurement system. Control charts and other statistical
3.2.19 site precision (R'), n—for a single analytical mea- techniques are presented to screen, plot, and interpret test
surement system (see 3.2.4), the value which the absolute results in accordance with industry-accepted practices to as-
difference between two individual test results obtained under certain the in-statistical-control status of the measurement
site precision conditions is expected to exceed about 5 % of the system.
time (one case in 20 in the long run) in the normal and correct 4.2 Statistical estimates of the measurement system preci-
operation of the test method. sion and bias are calculated and periodically updated using
3.2.19.1 Discussion—It is defined as 2.77 times σR', the accrued data.
standard deviation of results obtained under site precision 4.3 In addition, as part of a separate validation audit
conditions. procedure, QC samples and check standards may be submitted
3.2.20 site precision conditions, n—for a single analytical blind or double-blind and randomly to the measurement system
measurement system (see 3.2.4), conditions under which test for routine testing to verify that the calculated precision and
results are obtained by one or more operators in a single site bias are representative of routine measurement system perfor-
location practicing the same test method on a single measure- mance when there is no prior knowledge of the expected value
ment system using test specimens taken at random from the or sample status.
same sample of material, over an extended period of time
spanning at least a 20 day interval. 5. Significance and Use
3.2.20.1 Discussion—Site precision conditions should in- 5.1 This practice may be used to continuously demonstrate
clude all sources of variation that are typically encountered the proficiency of analytical measurement systems that are
during normal, long term operation of the measurement sys- used for establishing and ensuring the quality of petroleum and
tem. Thus, all operators who are involved in the routine use of petroleum products.
the measurement system should contribute results to the site
precision determination. In situations of high usage of a test 5.2 Data accrued, using the techniques included in this
method where multiple QC results are obtained within a 24 h practice, provide the ability to monitor analytical measurement
period, then only results separated by at least 4 h to 8 h, system precision and bias.
depending on the absence of auto-correlation in the data, the 5.3 These data are useful for updating test methods as well
nature of the test method/instrument, site requirements, or as for indicating areas of potential measurement system im-
regulations, should be used in site precision calculations to provement.
reflect the longer term variation in the system.
3.2.21 site precision standard deviation, n—the standard 6. Reference Materials
deviation of results obtained under site precision conditions for 6.1 QC samples are used to establish and monitor the
an individual measurement system and materials that are precision of the analytical measurement system.

3
 D6299 − 22
6.1.1 Select a stable and homogeneous material having reproducibility standard deviation for the test method. An
physical or chemical properties, or both, similar to those of F-test should be applied to test acceptability.
typical samples tested by the analytical measurement system. NOTE 7—The uncertainty in the ARV is inversely proportional to the
NOTE 5—When the QC sample is to be utilized for monitoring a process square root of the number of values in the average. For example, use of 16
stream analyzer performance, it is often helpful to supplement the process non-outlier results in calculating the ARV reduces the uncertainty of the
analyzer system with a subsystem to automate the extraction, mixing, ARV by a factor of 4 relative to the single result precision. The bias tests
storage, and delivery functions associated with the QC sample. described in this practice assume that the uncertainty in the ARV is
negligible relative to the precision of the measurement system being
6.1.2 Estimate the quantity of the material needed for each evaluated. If less than 16 values are used in calculating the average, this
specific lot of QC sample to (1) accommodate the number of assumption may not be valid. It is also assumed that the property of
analytical measurement systems for which it is to be used interest of the check standard is stable over the period of its intended use,
(laboratory test apparatuses as well as process stream analyzer and stored in a manner meeting the requirement of 3.2.17 quality control
(QC) sample.
systems) and (2) provide determination of QC statistics for a NOTE 8—Examples of exchanges that may be acceptable are ASTM
useful and desirable period of time. D02.92 Proficiency Test Program; ASTM D02.01 N.E.G.; ASTM
6.1.3 Collect the material into a single container and isolate D02.01.A Regional Exchanges; International Quality Assurance Exchange
it. Program, administered by Innotech ALBERTA.
6.1.4 Thoroughly mix the material to ensure homogeneity. 6.2.3 For some measurement systems, single, pure compo-
6.1.5 Conduct any testing necessary to ensure that the QC nent materials with known value, or simple gravimetric or
sample meets the characteristics for its intended use. volumetric mixtures of pure components having calculable
6.1.6 Package or store QC samples, or both, as appropriate value may serve as a check standard. For example, pure
for the specific analytical measurement system to ensure that solvents, such as 2,2-dimethylbutane, are used as check stan-
all analyses of samples from a given lot are performed on dards for the measurement of Reid vapor pressure by Test
essentially identical material. If necessary, split the bulk Method D5191. Users should be aware that for measurement
material collected in 6.1.3 into separate and smaller containers systems that show matrix dependencies, accuracy determined
to help ensure integrity over time. (Warning—Treat the from pure compounds or simple mixtures may not be repre-
material appropriately to ensure its stability, integrity, and sentative of that achieved on actual samples.
homogeneity over the time period for which it is to stored and 6.3 Validation audit (VA) samples are QC samples and
used. For samples that are volatile, such as gasoline, storage in check standards, which may, at the option of the users, be
one large container that is repeatedly opened and closed may submitted to the measurement system in a blind, or double
result in loss of light ends. This problem can be avoided by blind, and random fashion to verify precision and bias esti-
chilling and splitting the bulk sample into smaller containers, mated from routine quality assurance testing.
each with a quantity sufficient to conduct the analysis.
Similarly, samples prone to oxidation may benefit from split- 7. Quality Assurance (QA) Program for Individual
ting the bulk sample into smaller containers that can be Measurement Systems
blanketed with an inert gas prior to being sealed and leaving
7.1 Overview—A QA program (1)4 may consist of five
them sealed until the sample is needed.)
primary activities: (1) monitoring stability and precision
6.2 Check standards are used to estimate the accuracy of the through QC sample testing, (2) monitoring accuracy, (3)
analytical measurement system. periodic evaluation of system performance in terms of preci-
6.2.1 A check standard may be a commercial standard sion or bias, or both, (4) proficiency testing through participa-
reference material when such material is available in appropri- tion in interlaboratory exchange programs where such pro-
ate quantity, quality and composition. grams are available, and (5) a periodic and independent system
NOTE 6—Commercial reference material of appropriate composition validation using VA samples may be conducted to provide
may not be available for all measurement systems. additional assurance of the system precision and bias metrics
6.2.2 Alternatively, a check standard may be prepared from established from the primary testing activities. At minimum,
a material that is analyzed under reproducibility conditions by the QA program must include at least item one and item two,
multiple measurement systems. The accepted reference value subject to check standard availability (see 7.1.1).
(ARV) for this check standard shall be the average after 7.1.1 For some measurement systems, suitable check stan-
statistical examination and outlier treatment has been applied.3 dard materials may not exist, and there may be no reasonably
6.2.2.1 Exchange samples circulated as part of an interlabo- available exchange programs to generate them. For such
ratory exchange program, or round robin, may be used as check systems, there is no means of verifying the accuracy of the
standards. For the average computed from an exchange sample system, and the QA program will only involve monitoring
to be usable as the Accepted Reference Value (ARV) of a check stability and precision through QC sample testing.
standard, the standard deviation computed from at least 16 NOTE 9—For guidance on the establishment and maintenance of the
non-rejected normally distributed results (single submission essentials of a quality system, see Practice D6792.
per participant) shall not be statistically greater than the NOTE 10—For guidance on the analysis and interpretation of profi-
ciency test (PT) program results, see Guide D7372.

3 4
For guidance in statistical and outlier treatment of data, refer to Practices The boldface numbers in parentheses refer to the list of references at the end of
D6300, D7915, E178, and E691. this standard.

4
 D6299 − 22
7.2 Monitoring System Stability and Precision Through QC 7.5.2 Periodically analyze results from control charts, ex-
Sample Testing—QC test specimen samples from a specific lot cluding those data points with assignable causes, to quantify
are introduced and tested in the analytical measurement system the bias and precision estimates for the measurement system.
on a regular basis to establish system performance history in 7.6 Proficiency Testing:
terms of both stability and precision. 7.6.1 Participation in regularly conducted interlaboratory
7.3 Monitoring Accuracy: exchanges where typical production samples are tested by
7.3.1 Check standards may be tested in the analytical multiple measurement systems, using a specified (ASTM) test
measurement system on a regular basis to establish system protocol, provide a cost-effective means of assessing measure-
performance history in terms of accuracy. ment system accuracy relative to average industry perfor-
mance. Such proficiency testing may be used instead of check
7.4 Test Program Conditions/Frequency: standard testing for systems where the timeliness of the
7.4.1 Conduct both QC sample and check standard testing accuracy check is not critical. Proficiency testing may be used
under site precision conditions. as a supplement to accuracy monitoring by way of check
NOTE 11—It is inappropriate to use test data collected under repeat- standard testing.
ability conditions to estimate the long term precision achievable by the site 7.6.2 Participants plot their signed deviations or statistics
because the majority of the long term measurement system variance is due from the consensus values (exchange averages) on control
to common cause variations associated with the combination of time, charts in the same fashion described below for check standards,
operator, reagents, instrumentation calibration factors, and so forth, which to ascertain if their measurement processes are non-biased
would not be observable in data obtained under repeatability conditions.
relative to industry average.
7.4.2 Test the QC and check standard samples on a regular 7.7 Independent System Validation—Periodically, at the dis-
schedule, as appropriate. Principal factors to be considered for cretion of users, VA samples may be submitted blind or double
determining the frequency of testing are (1) frequency of use of blind for analysis. Precision and bias estimates calculated using
the analytical measurement system, (2) criticality of the pa- VA samples test data may be used as an independent validation
rameter being measured, (3) established system stability and of the routine QA program performance statistics.
precision performance based on historical data, (4) business
economics, and (5) regulatory, contractual, or test method NOTE 14—For measurement systems susceptible to human influence,
the precision and bias estimates calculated from data where the analyst is
requirements. aware of the sample status (QC or check standard) or expected values, or
NOTE 12—At the discretion of the laboratory, check standards may be both, may underestimate the precision and bias achievable under routine
used as QC samples. In this case, the results for the check standards may operation. At the discretion of the users, and depending on the criticality
be used to monitor both stability (see 7.2) and accuracy (see 7.3) of these measurement systems, the QA program may include periodic
simultaneously. If check standards are expensive, or not available in blind or double-blind testing of VA samples.
sufficient quantity, then separate QC samples are employed. In this case, 7.7.1 The specific design and approach to the VA testing
the accuracy (see 7.3) is monitored less frequently, and the QC sample program will depend on features specific to the measurement
testing (see 7.2) is used to demonstrate the stability of the measurement
system between accuracy tests.
system and organizational requirements, and is beyond the
intended scope of this practice. Some possible approaches are
7.4.3 It is recommended that a QC sample be analyzed at the noted as follows.
beginning of any set of measurements and immediately after a 7.7.1.1 If all QC samples or check standards, or both, are
change is made to the measurement system. submitted blind or double blind and the results are promptly
7.4.4 Establish a protocol for testing so that all persons who evaluated, then additional VA sample testing may not be
routinely operate the system participate in generating QC test necessary.
data. 7.7.1.2 QC samples or check standards, or both, may be
7.4.5 Handle and test the QC and check standard samples in submitted as unknown samples at a specific frequency. Such
the same manner and under the same conditions as samples or submissions should not be so regular as to compromise their
materials routinely analyzed by the analytical measurement blind status.
system. 7.7.1.3 Retains of previously analyzed samples may be
7.4.6 When practical, randomize the time of check standard resubmitted as unknown samples under site precision condi-
and additional QC sample testing over the normal hours of tions. Generally, data from this approach may only yield
measurement system operation, unless otherwise prescribed in precision estimates as retain samples do not have ARVs.
the specific test method. Typically, the differences between the replicate analyses are
plotted on control charts to estimate the precision of the
NOTE 13—Avoid special treatment of QC samples designed to get a measurement system. If precision is level dependent, the
better result. Special treatment seriously undermines the integrity of
precision estimates. differences are scaled by the standard deviation of the mea-
surement system precision at the level of the average of the two
7.5 Evaluation of System Performance in Terms of Precision results.
and Bias:
7.5.1 Pretreat and screen results accumulated from QC and 8. Procedure for Pretreatment, Assessment, and
check standard testing. Apply statistical techniques to the Interpretation of Test Results
pretreated data to identify erroneous data. Plot appropriately 8.1 Overview—Results accumulated from QC, check
pretreated data on control charts. standard, and VA sample testing are pretreated and screened.

5
 D6299 − 22
Statistical techniques are applied to the pretreated data to rules) (3) or Strategy 2 (EWMA) (4, 5, 6) are prescribed
achieve the following objectives: techniques for (a) routine recording of QC sample and check
8.1.1 Identify erroneous data (outliers). standard test results, and (b) immediate assessment of the “in
8.1.2 Assess initial results to validate system stability and statistical control” (7) status of the system that generated the
assumptions associated with use of control chart technique (for data. The I chart is intended to detect occurrence of a sudden,
example, dataset normality, adequacy of variations in the unique event that causes a large deviation from the expected
dataset relative to measurement resolution). value for the QC material. Strategy 1 (additional Run Rules) or
8.1.3 Deploy, interpret, and maintain control charts. Strategy 2 (EWMA) is intended to detect small levels of
8.1.4 Quantify long term measurement precision and bias. sustained shifts or drifts of the complete analytical system. MR
chart is intended to detect changes in the analytical system
NOTE 15—Refer to the annex for examples of the application of the overall variability.
techniques that are discussed below and described in Section 9.
NOTE 17—The control charts and statistical techniques described in this
8.2 Pretreatment of Test Results—The purpose of pretreat- practice are chosen for their simplicity and ease of use. It is not the intent
ment is to standardize the control chart scales so as to allow for of this practice to preclude use of other statistically equivalent or more
data from multiple check standards or different batches of QC advanced techniques, or both.
materials with different property levels to be plotted on the 8.3.1 Control charting may be viewed as a two-staged work
same chart. process where:
8.2.1 For QC sample test results, no data pretreatment is Stage 1 comprises assessment of initial test results (for a
necessary if results for different QC samples are plotted in new batch of QC material) and construction of the control chart
actual measurement units on different control charts. with graphically represented assessed results and statistical
8.2.2 For check standard sample test results that are to be values that describes the location of where future test results
plotted on the same control chart, two cases apply, depending for this QC material from the measurement systems are
on the measurement system precision: expected to fall within, on the assumption that the measure-
8.2.2.1 Case 1—If either (1) all of the check standard test ment system and QC material remains unchanged.
results are from one or more lots of check standard material Stage 2 comprises regular assessment of future test results
having the same ARV(s), or (2) the precision of the measure- (for the QC material) as they arrive in chronological order
ment system is constant across levels, then pretreatment against the established expectations in Stage 1; as well as a
consists of calculating the difference between the test result and periodic reevaluation of the expectation statistics of all accrued
the ARV: results to update the expectations statistics established from
Pretreated result 5 test result 2 ARV~ for the sample! (1) Stage 1, if necessary. See Fig. 1.

8.2.2.2 Case 2—Test results are for multiple lots of check STAGE 1—Assessment and Chart Construction
standards with different ARVs, and the precision of the 8.4 Assessment of Initial Results—Assessment techniques
measurement system is known to vary with level, are applied to test results collected during the initial startup
Pretreated result5 (2) phase of or after significant modifications to a measurement
system (see Note 19). Perform the following assessment after
@ test result 2 check standard ARV# /sqrt @ ~ standard error of ARV! 2 1 at least 20 results (pretreated if appropriate) have become
available. The purpose of this assessment is to ensure that these
~ std dev of site test method at the ARV level! 2 # results are suitable for deployment of control charts (described
where the standard error of the ARV is the uncertainty asso-
ciated with the ARV as supplied by the check standard sup- in A1.4).
plier; the standard deviation of site test method at the ARV NOTE 18—These techniques may also be applied as diagnostic tools to
level is the established standard deviation of the site’s test investigate out-of-control situations.
method under site precision conditions at nominally the ARV NOTE 19—During the data collection phase in Stage 1, users may
level. In the event the ARV was established through round deploy the procedures described in 8.7.2.3 or 8.7.3 (Q–procedure) or 8.7.4
robin testing, standard deviations determined from outlier- to monitor measurement process performance.
free and normally distributed round robin test results may be 8.4.1 Screen for Suspicious Results—Results (pretreated if
used to calculate the standard error of the ARV in accor- appropriate) should first be visually screened for values that are
dance with statistical theory. (See Note 16.) inconsistent with the remainder of the data set, such as those
8.2.2.3 If the ARV was not arrived at by round robin testing, that could have been caused by transcription errors, followed
a standard error of the ARV should be determined by users in by an outlier assessment using GESD (see Practice D7915) or
a technically acceptable manner. other equivalent statistical technique. Those flagged as suspi-
cious should be investigated. Discarding data at this stage must
NOTE 16—It is recommended that the method used to determine the be supported by evidence gathered from the investigation. If,
standard error of the ARV be developed under the guidance of a
statistician. after discarding suspicious pretreated results there are less than
15 values remaining, collect additional data and start over.
8.2.3 Pretreatment of results for VA samples is done in the 8.4.2 Screen for Unusual Patterns—The next step is to
same manner as described in 8.2.1 and 8.2.2. examine the results (pretreated if appropriate) for non-random
8.3 Control Charts (1, 2)—Individual (I), moving range of patterns such as continuous trending in either direction, un-
two (MR) control charts, and either Strategy 1 (additional run usual clustering, and cycles. One way to do this is to plot the

6
 D6299 − 22

FIG. 1 Control Chart Work Process Block Diagram

results on a run chart (see A1.3) and examine the plot. If any 8.4.4.3 EWMA Overlay—For strategy 2, calculate the
non-random pattern is detected, investigate for and eliminate EWMA values and plot them on the I chart. Calculate the
the root cause(s). Discard the data set and start the procedure EWMA control limits and overlay them on the I chart.
again.
8.4.3 Test “Normality” Assumption, Independence of Test STAGE 2—Deployment for Monitoring and Periodic
Results, and Adequacy of Measurement Resolution—For mea- Re-assessment
surement systems with no prior performance history, or as a 8.4.5 Control Chart Deployment—Put these control charts
diagnostic tool for initial data collected on a new batch of QC into operation by regularly plotting the test results (pretreated
material, it is useful to test that the results from the measure- if appropriate) on the charts and immediately interpreting the
ment system are reasonably independent, with adequate mea- charts.
surement resolution, and may be adequately modelled by a 8.5 Control Chart Interpretation:
normal distribution. One way to do this is to use a normal
8.5.1 Apply control chart rules (see A1.5) to determine if
probability plot and the Anderson-Darling Statistic (see A1.4).
the data supports the hypothesis that the measurement system
If the results show obvious deviation from normality or
is under the influence of common causes variation only (in
obvious measurement resolution inadequacy (see A1.4), follow
statistical control).
the guidance in A1.4.2.6, Case 2.
8.5.2 Investigate Out-of-Control Points in Detail—Exclude
NOTE 20—Transformations may lead to normally distributed data, but from further data analysis those associated with assignable
these techniques are outside the scope of this practice. causes, provided the assignable causes are deemed not to be
8.4.4 Construction of Control Charts—If no obvious un- part of the normal process.
usual patterns are detected from the run charts, and no obvious
deviation from normality is detected, proceed with construc- NOTE 21—All data, regardless of in-control or out-of-control status,
needs to be recorded.
tion of the control charts as follows (see A1.5.1 – A1.5.3):
8.4.4.1 I Chart—Calculate the center line, control limits and 8.6 Scenario 1 for Periodic Updating of Control Charts
overlay them on the “run chart” to produce the I chart. Parameters:
8.4.4.2 Construct an MR plot and examine it for unusual 8.6.1 Scenario 1 covers (1) control charts for a QC material
patterns. If no unusual patterns are found in the MR plot, where there had been no change in the system, but more data
calculate and overlay the center line and control limits on the of the same level has been accrued; or (2) control charts for
MR plot to complete the MR chart. check standard pretreated results.

7
 D6299 − 22
8.6.2 When a minimum of 20 new in-control data points 8.7.1 Scenario 2 covers control chart for QC materials
becomes available, perform an F-test (see A1.8) of sample where an assignable cause change in the system had occurred
variances for the new data set versus the sample variance used due to a change of QC material as the current QC material
to calculate the current control chart limits. If the outcome of supply is exhausted. Minor or major differences in measured
the F-test is not significant, and, if the sample variance used to property level may exist between QC material batches. Since
calculate the current control limits is based on less than 100 control limit calculations for the I chart require a center value
data points, statistically pool both sample variances and then established by the measurement system, a special transition
update the current control limits based on this new pooled procedure is required to ensure that the center value for a new
variance and I-chart center line (Ī in equations Eq batch of QC material is established using results produced by
A1.10-A1.13) if updated (see 8.6.2.2). a measurement system that is in statistical control. This
8.6.2.1 If the outcome of the F-test is not significant, and if practice presents two procedures to be selected at the users’
the sample variance used to calculate the current control limits discretion.
is based on more than 100 data points, the statistical pooling of 8.7.1.1 Use of Precision Statistics from Previous Control
both sample variances to be used for update of the current Charts—Control chart statistics achieved (Īachieved, σachieved,
¯
control limits is recommended, but may be at the discretion of MR achieved) from previous completed I, MR chart for similar QC
the user. material may be used for the new QC batch transition tech-
8.6.2.2 If the outcome of the F-test is not significant, niques described in this section if either of the following
compute the t value in Eq 3 using the average of the new conditions is met:
in-control data, the current center line of the I-chart, and the (1) test method published reproducibility (Rpub) is not
current chart standard deviation (σR’) used to compute the dependent on the measurement level
I-chart control limits. Re-compute and update the I-chart center (2) for Rpub expressed as a function of the measurement
line to reduce its statistical uncertainty is permissible if all of level, the ratio:
the following conditions are met: [Rpub@Ī_achieved / Rpub@ 1st new QC result] is between 0.85 and
(1) |t| ≤ 1.7 1.15.
(2) ewmanewdata on one side of center line < 75 % where:
NOTE 22—The value 1.7 is based on a one-sided t-test of a “difference Rpub@Ī_achieved = published method reproducibility
= 0” null hypothesis versus an alternate hypothesis of either greater than evaluated at Īachieved level, and
or less than zero as chosen by the user at 5 % significance level, 40 to 250 Rpub@1st new QC result = published method reproducibility
degrees of freedom rounded up to 1st decimal for simplicity. evaluated at the 1st new QC result
~ Ī 2 x̄ newdata ! level.
current
t5 (3)
σ R' Œ 1
1
1
n1 n2
8.7.2 Procedure 1, Concurrent Testing:
8.7.2.1 Collect and prepare a new batch of QC material
when the current QC material supply remaining can support no
where: more than 20 analyses.
Īcurrent = the current I-chart center line, which is the arith- 8.7.2.2 Concurrently test and record data for the new
metic average calculated using all in control results material each time a current QC sample is tested. The result for
without the new data in 8.6.2; n1 is the number of the new material is deemed valid if the measurement process
results used to calculate Īcurrent, and in-control status is validated by the current QC material and
x̄ newdata = the arithmetic average of new results in 8.6.2; n2 is control chart.
the number of results used to calculate x̄ newdata . 8.7.2.3 Optionally, to provide an early indication of the
As a safeguard against slow drift in one direction that is below status of the new batch of QC material, immediately start a run
the detection power of the control chart rules, four consecutive chart and an MR plot for the new material. After five valid
adjustment of the I-chart center line in the same direction shall results become available for the new material, convert the run
trigger an accuracy verification by Check Standard (CS). chart into an I chart with trial control limits by adding a center
Follow Practice D6617 to determine the acceptable tolerance line based on the average of the five results and control limits
zone for the difference between the result obtained versus the using σachieved from previous control charts in 8.7.1.1.
Accepted Reference Value (ARV) of the CS. Similarly, set trial control limits for the MR chart based on
NOTE 23—Sigma can be either pooled or un-pooled, depending on ¯
MR achieved.
whether it was performed in 8.6.2.1.
8.7.2.4 After a minimum of 20 in-control data points are
8.6.3 If the outcome of the F-test is significant, investigate
collected on the new material, perform an F-test of sample
for assignable causes. Update the current control limits based
variance for the new data set (σnewdata)2 versus (σachieved)2 in
on sample variance and average calculated using the new data
8.7.1.1. If the outcome of the F-test is not significant, for Rpub
if it is determined that this new variance and average is
expressed as a function of the measurement level, evaluate
representative of current system performance under common
Rpub using the average of new results to re-confirm the ratio
cause variation.
Rpub@Ī_achieved / Rpub@new QC results average is between 0.85 and
8.7 Scenario 2 for Periodic Updating of Control Charts 1.15. If confirmed, and if σachieved is based on less than 100 data
Parameters: points, statistically pool both sample variances (Eq A1.30) and

8
 D6299 − 22
¯ ’s (Eq A1.29). Use the square root of this new pooled
MR tion for the check standard material; for Case 2, the standard
¯ as σ and MR
variance and pooled MR R'
¯ for the construction of deviation is 1 since Eq 2 is a standardized normal deviate.
the new I and MR charts in 8.7.2.7. 8.7.4 Procedure 2-B: Dynamically Updated I / EWMA
Chart—This is essentially an I-chart and EWMA with varying
8.7.2.5 If the outcome of the F-test in 8.7.2.4 is not control chart limits that are updated with the arrival of each
significant, and the ratio Rpub@Ī_achieved / Rpub@ new 20 QC results new result, which is judged using limits computed from all
average is between 0.85 and 1.15 for σachieved based on more than
previous results. The dynamic update combines the σknown (see
100 data points, the statistical pooling in 8.7.2.4 is 8.7.3.1) for the individual result with the varying standard error
recommended, but may be at the discretion of the user. If associated with the center line computed with all previous
pooling is not performed, use σachieved and MR ¯
achieved as σR' and results. This standard error (for the I-chart) steadily decreases
¯
MR for the construction of the new I and MR charts in 8.7.2.7. as the number of results used for its computation increases,
If the outcome of the F-test in 8.7.2.4 is not significant, but whilst for the EWMA, the standard error typically decrease
the ratio Rpub@Ī_achieved / Rpub@new QC results average is not initially and then increases towards its asymptotic value. See
¯
between 0.85 and 1.15, use σnewdata and MR ¯ A1.10 for details.
newdata as σR' and MR
for the construction of the new I and MR charts in 8.7.2.7. NOTE 25—Procedure 2-B was formerly referred to as Q-chart Option 1.
8.7.2.6 If the outcome of the F-test in 8.7.2.4 is significant, 8.7.5 Operate Procedure 2-B in conjunction with an MR
investigate for assignable causes. If it is determined that this chart per 8.7.3.2. After a minimum of 20 in control data points
new variance is representative of current system performance have been accrued, execute the steps from 8.7.2.4 to 8.7.2.7.
under common cause variation, use σnewdata and MR ¯ Because Procedure 2-B is technically equivalent to the I chart
newdata as σR'
¯ for the construction of the new I and MR charts in
and MR procedure, the user may either construct a new I/MR control
8.7.2.7. chart for the new batch of QC material as instructed in 8.7.2.7,
or continue to operate Procedure 2-B and MR chart for
8.7.2.7 Complete the Stage 1 assessments as per Section 8 measurement process stability and precision monitoring using
to 8.4.3. Construct new I and MR charts (and EWMA overlay the new batch of QC material.
for strategy 2) for this new batch of QC material as per Section
8.8 Short Run Scenario—Procedures described in 8.7.3 and
8.4.4.
8.7.4 may also be used to address short run situations where a
8.7.2.8 Switch over to the new I and MR charts upon single batch of QC material may provide only a limited number
depletion of current QC material. (less than 20) of QC test results and replacement of exactly the
8.7.3 Procedure 2-A, Q-Procedure (see A1.9): same material is not feasible or possible. For these short run
8.7.3.1 This procedure is designed to alleviate the need for QC batches, since there is insufficient data to properly charac-
concurrent testing of two materials. A priori knowledge of the terize the mean of batch, these procedures can only be used to
measurement process standard deviation (σknown) is required. monitor stability and precision of the measurement process.
σachieved meeting the requirements in 8.7.1.1 can be used as 8.9 Instrument Replacement or Post Overhaul Scenario—
σknown for this purpose. A Qr statistic is computed with the The procedures described in 8.7.3 and 8.7.4 may be used to
arrival of each new QC result commensurate with the 2nd address situations where an instrument is taken out of service
result, and compared against its theoretical mean (0) and 3 and is replaced by another qualified instrument, or, when the
sigma limits (6 3). See A1.9 for details. primary instrument is returned to service after a major overhaul
NOTE 24—It is recommended that this standard deviation estimate be such as replacement of critical parts or factory re-calibration.
based on at least 50 data points. For these situations, the existing system precision parameters
may be used, in conjunction with the MR chart, to monitor
8.7.3.2 When the Q-procedure is operational (minimum of
stability and precision of the replacement or overhauled
two data points), it may be used in conjunction with a MR chart
¯ measurement process, respectively, based on the assumption
constructed using the observations and MR achieved per 8.7.1.1 to
that the existing system precision parameter is still valid. After
provide QA of the measurement process. sufficient data is accrued, a statistical assessment shall be
8.7.3.3 After a minimum of 20 data points have been performed to confirm this assumption, or update the system
accrued (by the Q-procedure), execute the steps from 8.7.2.4 to precision parameters accordingly. Use of the existing precision
8.7.2.7. Because the Q-procedure is technically equivalent to will enable the system to be immediately put into service,
the I chart procedure, the user may either construct a new I/MR while providing a safeguard against the situation where the
control chart for the new batch of QC material as instructed in new system performance with replacement or overhauled
8.7.2.7, or continue to operate the Q–chart and MR chart for instrument is statistically worse than the previous system
measurement process stability and precision monitoring, performance. Use of these procedures is in addition to any
respectively, using the new batch of QC material. steps such as calibration and running check standards needed to
8.7.3.4 It is necessary to start a new Q-chart with each new qualify replacement instruments.
batch of QC material.
8.7.3.5 A common Q-chart and MR chart may be used for 9. Evaluation of System Performance in Terms of
pre-treated results as per Case 1 and Case 2 in 8.2. For Case 1, Precision and Bias
the standard deviation shall be the applicable standard devia- 9.1 Site Precision Estimated from Testing of QC Samples:

9
 D6299 − 22
9.1.1 Estimate the site precision of the measurement system 9.3.1 Plot the pretreated results as per Section 8 versus their
for material types and levels using the current active I-chart corresponding ARVs. Examine the plot for patterns indicative
parameter estimate σR' (see 8.6, 8.7) based on the root-mean- of level-dependent bias.
square (rms) formula for standard deviation. 9.3.2 If there is no discernible pattern, perform the t-test as
R' 5 2.77 3 σ R’ (4) described in 9.2 to determine if the average of all the pretreated
differences plotted on the I chart is statistically different from
9.1.1.1 Alternatively, in the absence of auto-correlation in zero.
the data (see A1.4), R' may be estimated as 2.46 times the 9.3.2.1 If the outcome of the t-test is that the average is not
average of the moving range ~ MR ¯ ! from the MR chart statistically different from zero, then the bias in the measure-
corresponding to the I-chart in 9.1.1. ment process is negligible.
9.3.2.2 If the outcome of the t-test is that the average is
R' 5 2.46 3 ¯
MR (5) statistically different from zero, then there is evidence that the
NOTE 26—The site precision standard deviation (σR’) is estimated from
measurement system is biased. The bias may be level depen-
the MR chart as R'/2.775 ~ MR
¯ ! /1.128.
dent. However, the statistical methodology for estimating the
9.1.1.2 For estimate of site precision standard deviation bias/level relationship is beyond the scope of this practice.
(σR') using retain results, first obtain the standard deviation of 9.3.3 If there is a discernible pattern in the plot in 9.3.1, then
differences by applying the root-mean-square formula below to the measurement system may exhibit a level dependent bias.
the differences between the original and retest results for The statistical methodology for estimating the bias/level rela-
samples with same nominal property level. If measurement tionship is beyond the scope of this practice.
process precision is known to be level independent, retest 9.3.4 If a bias is detected in 9.3.2.2, or if the plot in 9.3.3
results from samples with different property levels may be exhibits discernible patterns, investigate for root cause(s).
used. Otherwise, sample pairs with nominally similar property 9.3.4.1 If there is evidence of a bias versus level
level (general rule is within 2R) should be used to estimate the relationship, or, if users wish to perform a more rigorous
site precision at the nominal property level. Divide the standard examination of the bias versus level relationship with multiple
deviation of differences by 1.414 to obtain the estimate for site check standards, it is recommended that the principles of
precision standard deviation. (σR'). Practice D6708 be employed under the guidance of qualified
standard deviation of differences5 (6) statistical expertise.
10. Validation of System Performance Estimates Using
Π( ~ individual difference 2 average difference!
total number of differences
2
VA Samples
10.1 If the users decide to include VA sample testing as part
σ R' 5 ~ standard deviation of differences! ÷1.414 (7) of their QA program, then they should periodically evaluate the
9.1.2 Compare R' to published reproducibility of the test results obtained on the VA samples. The purpose of the
method at the same level, if available. R' is expected to be less evaluation is to establish whether the system performance
than or equal to the published value. Use the χ2 test described estimates described in Section 9 are reasonably applicable to
in A1.7. routinely tested samples.
10.2 VA sample test results should be evaluated indepen-
9.2 Measurement System Bias Estimated from Multiple
dently through an internal or external audit system, or both. It
Measurements of a Single Check Standard—If a minimum of
is recommended that the internal audit team not be limited to
15 test results is obtained on a single check standard material
the operators of the measurement system and their immediate
under site precision conditions, then calculate the average of all
supervisors.
the in-control individual differences plotted on the I chart.
Perform a t-test (see A1.6) to determine if the average is 10.3 Insofar as possible, analyze the results obtained on the
statistically different from zero. VA samples separately and in the same manner as those from
9.2.1 If the outcome of the t-test is that the average is not the routine QC and check standard testing program.
statistically different from zero, then the bias in the measure- 10.4 Using F- or t- tests, or both (see A1.8 and A1.6),
ment process is negligible. statistically compare the system performance estimates ob-
9.2.2 If the outcome of the t-test is that the average is tained from the VA sample testing program to the measurement
statistically different from zero, then the best estimate of the system accuracy and precision estimates from the QC sample
measurement process bias at the level of the check standard is testing program.
the average. If bias is deemed to be of practical significance by 10.5 If the comparison reveals that the estimates of the
the user, investigate for root causes, and take corrective measurement system performance using VA samples are sta-
measures. tistically different than estimates using QC and check standards
9.3 Measurement System Bias Estimated from Measure- per Section 9, investigate thoroughly for the assignable
ments of Multiple Check Standards—When using multiple cause(s) of this inconsistency. Until the causes are identified
check standards, determine if there is a relationship between and eliminated, the estimates from Section 9 should be
the bias and the measurement level. considered suspect.

10
 D6299 − 22

ANNEX

(Mandatory Information)

A1. STATISTICAL QUALITY CONTROL TOOLS

A1.1 Purpose of this Annex fraction by R at the new ARV level with unknown R’ to arrive
A1.1.1 The purpose of this annex is to provide guidance to at the estimated R’ at the new ARV level. This approach is
practitioners, including worked examples, for the proper ex- based on the assumption that the fraction of R’ and R is
ecution of the statistical procedures described in this practice. constant among different ARV levels. Users are cautioned that
See Tables A1.1-A1.14 and Figs. A1.1–A1.16. this assumption may not be valid if the published precision has
different functional forms between r and R. Note that this
NOTE A1.1—For some examples in this annex, 15 data points are used fraction is the inverse of TPI (Test Performance Index) as
to illustrate calculation and plotting methodologies; it is not the intention
of this annex to override the mandatory requirement of 20 minimum data
defined in Practice D6792.
points (see 8.4). Work is underway to revise the annex examples to use 20 Example:
data points for all examples. R’ of site (calculated from actual QC data) at sulfur level 10
A1.2 Pretreatment of Test Results (8.1 to 8.2.3) ppm = 2 ppm (published R at sulfur level of 10 ppm = 3 ppm).
Fraction of R’/R at 10 ppm = 2/3
A1.2.1 Throughout this annex, {Yi:i=1. . .n} denotes a
sequence of as measured test results. {Ii:i=1. . .n} will signify Estimated R’ of site at sulfur level at 15 ppm is estimated as:
a sequence of test results after pretreatment, if necessary. (2/3)* (published R at sulfur level of 15 ppm).
A1.2.2 If {Yi:i=1. . .n} is a sequence of results from a single A1.3 The Run Chart
QC sample, then
A1.3.1 A run chart is a plot of results in chronological order
Ii 5 Yi (A1.1) that may be used to screen data for unusual patterns.
with no pretreatment being required. Preferably, pretreated results are plotted. Use a run chart to
A1.2.2.1 An example of a sequence of results, Yi, from a screen data for unusual patterns such as continuous trending in
single QC sample is given in Columns 2 and 4 of Table A1.3. either direction, unusual clustering, and cycles. Several non-
random patterns are described in control chart literature. When
A1.2.3 If {Yi:i=1. . .n} is a sequence of results from a single
control parameters have been added to a run chart, it becomes
check standard, from multiple check standards having nomi-
a control chart of individual values (I chart).
nally the same ARV, or from multiple check standards having
different ARVs where the precision of the measurement system A1.3.2 Plot results on the chart. Plot the first result at the
does not vary with level, and if { Xi:i=1. . .n} is the sequence left, and plot each subsequent point one increment to the right
of corresponding ARVs, then of its predecessor. The points may be connected in sequence to
Ii 5 Yi 2 Xi (A1.2) facilitate interpretation of the run chart.

The site precision (R’) of the measurement process must be A1.3.3 Allow sufficient space in the x-axis direction to
essentially the same for all values {Xi}. accommodate as many results as should be obtained from a
A1.2.3.1 An example of a sequence of results from a single consistent batch of material. Allow enough space in the y-axis
check standard is given in Table A1.4. The preprocessed result, direction to accommodate the expected minimum and maxi-
Ii, is given in Column 4 of Table A1.4. mum of the data.
A1.2.4 If {Yi} is a sequence of results from different check A1.3.4 Example of a Run Chart for QC Results—The first
standards, and if the reproducibility varies with the level of the 15 results from Column 2 of Table A1.3 are plotted in sequence
accepted reference values, {Xi}, then as they are collected as shown in Fig. A1.1. The data would be
examined for unusual patterns.
I i 5 ~ Y i 2 X i ! /σ i (A1.3)
A1.3.5 Example of a Run Chart for Multiple Results from a
where σi are estimates of the standard deviation under site
Single Check Standard—The first 15 preprocessed results
precision conditions of the measurement process at levels {Xi}.
(differences) from Column 4 of Table A1.4 are plotted in
A1.2.4.1 Table A1.5 shows an example of results for mul-
sequence as they are collected as shown in Fig. A1.2. The data
tiple check standards where the precision of the measurement
would be examined for unusual patterns.
system is level dependent.
A1.2.4.2 Discussion—Site precision (R’) estimates at ARV A1.3.6 Example of a Run Chart for Results from Multiple
values that are significantly different from those in the site’s Check Standards—The first 15 preprocessed results (differ-
historical database may also be estimated proportionally using ences scaled by σi) from Table A1.5 are plotted in sequence as
the published R at the ARV level. Calculate the fraction of R’ they are collected as shown in Fig. A1.3. The data would be
and R at the ARV level with known R’ and multiply this examined for unusual patterns.

11
 D6299 − 22
A1.4 Normality, Data Independence, and Resolution
Adequacy Checks S
A 2 * 5 A 2 11
0.75 2.25
n
1 2
n D (A1.7)

A1.4.1 A normal probability plot (a special case of a q-q The quantity A2* is referred to as the A-D statistic (A-D).
plot) is used to visually assess the validity of the assumption A1.4.2.6 Guidance on Interpretation of the Two A-D Statis-
that the observations are normally distributed. Since the control tics (A-Drms and A-DMR): CASE 1—Both A-Drms and A-DMR
chart and limits prescribed in this practice are based on the are << 1.0. This is to be interpreted as, “no compelling
assumption that the data behavior is adequately modeled by the evidence to reject the hypotheses that the data is normal,
normal distribution, it is recommended that a test of this independent, with adequate measurement resolution.” Proceed
normality assumption be conducted. to construct control chart with either the rms-based or the
A1.4.1.1 To construct a normal probability plot: MR-based standard deviation estimate.
(1) Create a column of the observations sorted in ascending CASE 2—Both A-Drms and A-DMR are >>> 1.0, and the q-q
order. plot shows a few distinct “staircases,” which really means the
(2) Select the appropriate column from Fig. A1.4, based on majority of the data is clustered into a few unique values. This
the number of observations (n). is strong evidence that there is inadequate variation in the
(3) Plot each observation in the sorted column (y-value) dataset due to inadequate numerical resolution. Under these
against its corresponding value from Fig. A1.4 (z-value). circumstances, if the total number of unique values in the data
A1.4.1.2 Visually inspect the plot for an approximately set is less than six, increase data resolution (carry an additional
linear relationship. If the results are normally distributed, the decimal) and reevaluate both A-D statistics for the purpose of
plot should be approximately linear. Major deviations from control charting. Note that because results are used for internal
linearity are an indication of nonnormal distributions of the QA purposes, this should not be considered as a deviation from
differences. test method reporting requirements. If additional data resolu-
tion is not possible, or, if the total number of unique values in
NOTE A1.2—The assessment methodology of the normal probability
plot advocated in this practice is strictly visual due to its simplicity. For the data set is six or greater, or, if after increase in data
statistically more rigorous assessment techniques, users are advised to use resolution, both A-D statistics are still >>1.0, users may still
the Anderson-Darling technique described below, and consult a statisti- use regular plotting of chronological QC data to monitor for
cian. occurrence of an abnormal event. For the purpose of the latter,
A1.4.2 Anderson-Darling Statistic—The Anderson-Darling it is recommended that the run-chart be used with a lower and
(A-D) statistic is used to objectively test for normality, data upper percentile-based action limits, provided that there is no
independence, and adequacy of measurement resolution rela- visual indication of process trending in the data set used to
tive to the overall variation in the dataset. Two A-D statistics determine the action limits. The suggested percentiles are 1st
(A-Drms, A-DMR) are calculated using the identical procedure and 99th, based on a data set of at least 75 results, collected
outlined as follows, where A-Drms, A-DMR are the A-D statistic under site precision conditions. It is not the intent of this
calculated using numerical estimates of the sample standard practice to exclude use of other percentiles, or, use of other
deviation(s) as per the rms (root-mean-square) and the MR- user-defined action limits, provided the limits meet the appli-
(moving range of 2) techniques, respectively. The calculation cation requirements. Users are advised to seek qualified statis-
steps are as follows: tical guidance on how to determine the appropriate action
A1.4.2.1 Order the non-outlying results such that x1 ≤ x2 ≤ . limits and associated implications.
. . . xn CASE 3—A-Drms is << 1.0, but A-DMR > 1.0. This is
A1.4.2.2 Obtain standardized variate from the xi values as indicative that the test results are serially correlated, or not
follows: independent. A direct consequence of this non-independence is
that the standard deviation estimate using the moving range
w i 5 ~ x i 2 x̄ ! /s (A1.4)
technique will underestimate the variation of the total dataset.
for (i= 1 . . . n), where s is sample standard deviation of the The root cause for this non-independence is typically cyclic
results using either the rms or MR technique, and x̄ is the data caused by diurnal effect of the environment, or moderate
average of the results. trending of data due to normal degradation of test equipment.
If this is judged to be normal behavior of the measurement
NOTE A1.3—One standard deviation estimate ~ 0.89 × [average MR] of
the dataset. data, proceed to construct control chart with the rms -based
standard deviation estimate.
A1.4.2.3 Convert the wi values to standard normal cumula-
tive probabilities pi values using the cumulative probability A1.4.3 Example of Normal Probability Plot for QC
table for the standardized normal variate z (see Fig. A1.5): Results—Once 20 results have been obtained (Table A1.3),
they are sorted in ascending order and paired with the
p i 5 Probability ~ z,w i ! (A1.5)
corresponding z-values from Fig. A1.4. The paired results (see
2
A1.4.2.4 Compute A as: Table A1.6) are plotted as (x,y) points (see Fig. A1.6). A line
n may be added to the plot to facilitate examination of the data
( ~ 2i 2 1 ! @ ln~ p ! 1ln~ 1 2 p
i51
i n112i !# for deviations from linearity.
A2 5 2 2n (A1.6) A1.4.3.1 For the above example, the wi and pi values used in
n
the calculation of the A-Drms statistic are shown in Table A1.6,
A1.4.2.5 Compute the quantity A 2* as: as is the individual terms in the summation for A2. The value

12
 D6299 − 22
for A2 is 0.415, and the value for A2* (A-Drms) is 0.44. Similar n21
( ?I
i51 i11 2 Ii ?
¯
MR (A1.8)
calculation using MR technique yields an A-Drms value of 0.60. stage 1 5
n21
Since this is a CASE 1 outcome, the hypothesis of normality,
data independence, and adequate measurement resolution is A1.5.1.1 A horizontal center line is added at the level of the
accepted. mean of all the non-rejected results, (Ī stage1) at the conclusion of
Stage 1, calculated using Eq A1.9:
A1.4.4 Example of Normal Probability Plot for Multiple
n
Results from a Single Check Standard—The first 15 prepro-
Ī stage1 5
( I
i51 i
(A1.9)
cessed results (Table A1.4, Column 4) are sorted in ascending n
order and paired with the corresponding z-values from Fig.
A1.4. The paired results (Table A1.7) are plotted as x,y points A1.5.1.2 Upper control limits (UCL) and lower control
(Fig. A1.7). A line may be added to the plot to facilitate limits (LCL) are added, indicating the limits within which
examination of the data for deviations from linearity. about 99.7 % of all normally distributed measurement data are
expected to fall if variability of the measurement system is due
A1.4.5 Example of Normal Probability Plot for Results from to random error only.
Multiple Check Standards—The first 15 preprocessed results For standard deviation estimated from the moving range
(Table A1.5, Column 6) are sorted in ascending order and technique, calculate UCL and LCL using Eq A1.10 and A1.11:
paired with the corresponding z-values from Fig. A1.4. The
paired results (Table A1.8) are plotted as x,y points (Fig. A1.8). UCL 5 Ī12.66 ¯
MR (A1.10)
A line may be added to the plot to facilitate examination of the LCL 5 Ī 2 2.66 MR¯ (A1.11)
data for deviations from linearity. NOTE A1.4—Explanation of the factor 2.66 in Eq A1.10 and Eq A1.11:
A1.4.5.1 For this example, the wi, and pi values used in the since (MR-bar/1.128) = σ, therefore, 3* σR' = 3* (MR-bar/1.128) =
calculation of the Anderson-Darling statistic are shown in (3/1.128) * MR-bar = 2.66* MR-bar.
Table A1.8, as are the individual terms in the summation for A2. For standard deviation estimated from the root-mean-square
The value for A2 is 0.673, and the value for A2* is 0.713. Since technique, calculate UCL and LCL using Eq A1.12 and A1.13:
this value is less than 0.752, the hypothesis of normality is
UCL 5 Ī13 3 σ R' (A1.12)
accepted at the 95 % confidence level.
LCL 5 Ī 2 3 3 σ R' (A1.13)
A1.5 The Control Chart
A1.5.1.3 At the discretion of the users, upper warning limits
A1.5.1 I Chart—The I chart is a run chart to which control (UWL) and lower warning limits (LWL) may be added, and
limits and center line have been added. To establish placement these indicate the limits within which about 95 % of all
positions of the control limits for the I chart, an estimate of the normally distributed data are expected to fall.
variability of the measurement system under site precision
conditions (σR') is required. While there are several statistical UWL 5 Ī12 3 σ R' (A1.14)
techniques that may be used for this purpose, this practice
LWL 5 Ī 2 2 3 σ R' (A1.15)
advocates use of the rms (root-mean-square) technique to NOTE A1.5—Referring to Note A1.4, the UWL and LWL for sigma
estimate sigma, or, alternatively, in the absence of auto- calculated using (MR-bar/1.128) becomes I-bar 6 1.77 MR-bar.
correlation, the mr (moving range of two) technique for its
A1.5.2 EWMA Overlay—A EWMA overlay is a trend line
simplicity and robustness to outliers. After a minimum of 20
constructed from EWMA values calculated using the I-values.
results have been obtained for a new batch of QC material,
The EWMA trend line is typically overlaid on the I chart to
carry out the data assessment and control chart construction
enhance its sensitivity in detecting mean shifts that are small
activities at the conclusion of Stage 1 (section 8.4 to 8.4.4) in
relative to the measurement system precision. Each EWMA
accordance with this Annex. For Ī in equations Eq
value is a weighted average of the current result and previous
A1.10-A1.13, use Īstage per equation Eq A1.9. If applicable or
¯ in equations Eq A1.12, results, with the weights decreasing exponentially with the age
available (see 8.7.1.1), the σR’ and MR
of the reading.
Eq A1.13; Eq A1.10, Eq A1.11 should be based on statistically
¯ A1.5.2.1 A sequence of values, EWMAi, are calculated, and
pooled values using statistics (σstage1, MR stage1) calculated from
overlaid on the I chart and connected. Use the following
at least 20 non-rejected results at the conclusion of Stage 1 and
¯
recursion equation:
control chart statistics achieved (σachieved, MR achieved) from pre-
vious completed I, MR charts for similar QC material in EWMA1 5 I 1 (A1.16)
accordance with 8.7.2.4 or 8.7.2.5. Otherwise use the statistics EWMAi 5 ~ 1 2 λ ! EWMAi21 1λI i (A1.17)
¯
(σstage1, MR stage1) computed from at least 20 non-rejected results
¯ in equations Eq A1.12, Eq A1.13;
where λ is the exponential weighting factor. For application
from Stage 1 as σR’ and MR of this practice, a λ value between 0.2 to 0.4 is recommended.
Eq A1.10, Eq A1.11 respectively. σstage1, MR ¯
stage1 are calculated
as follows: NOTE A1.6—For the EWMA trend, a λ value of 0.4 closely emulates the
run rule effects of conventional control charts, while a value of 0.2 has
n optimal prediction properties for the next expected value. In addition,

σ stage 1 5 ! (
i51

n21
~I i 2 Ī !
2
these λ values also conveniently places the control limits (3-sigma) for the
EWMA trend at the 1 (for λ=0.2) to 1.5-sigma (for λ=0.4) values for I
chart.

13
 D6299 − 22
A1.5.2.2 The control limits for the EWMA chart are calcu- system has likely occurred and hence the system shall be
lated using a weight (λ) as follows: declared out-of-statistical-control:
(1) Two out of three consecutive results on the I-chart that
UCLλ 5 I13σ R’ Œ λ
22λ
(A1.18) are more than 2σR’ from the center line in the same direction.
(2) Five consecutive results on the I-chart that are more
LCLλ 5 I 2 3σ R’ Œ λ
22λ
(A1.19)
than 1σR’from the centerline in the same direction.
(3) Nine or more consecutive results above or below the
centerline on the I-chart.
A1.5.3 MR Chart—A MR of two chart is obtained by
(4) Seven consecutive results steadily increasing or de-
plotting the sequential range of two values given by:
creasing.
?
MRi 5 I i 2 I i21 ? (A1.20)
NOTE A1.7—Run rules are applicable to control charts generated under
and connecting each point. site precision conditions (see 3.2.20.1). If multiple QC sample results are
A1.5.3.1 The upper control limit for the MR chart is given obtained within a 24 h period, then only results separated by at least 4 h
by: should be used in site precision calculations and plotted on a control chart
in order to reflect the longer term variation in the system.
UCLMR 5 3.27 ¯
MR (A1.21)
Strategy 2: EWMA Strategy
A1.5.3.2 There is no lower control limit for an MR chart.
Use of the EWMA overlay and its associated control limits
A1.5.4 Declaration of an Out-of-Statistical-Control Sys- are described in this section. When either the EWMA exceeds
tem: its control limits, or, run rule (3) in Strategy 1 is violated, it
A1.5.4.1 I-chart—Individual values that are outside the shall be interpreted as a strong signal that a change in state of
upper (UCL) or lower (LCL) control limits are strong indica- the measurement system has occurred and hence the system
tions of an out-of-statistical-control condition. Therefore, the shall be declared out-of-statistical-control.
system shall be declared out-of-statistical-control, followed by
an immediate retest of a new QC sample to confirm the A1.5.4.2 MR Chart:
out-of-statistical-control event. In situations when the QC (1) Any moving range (MR) value that exceeds the upper
sample is immediately retested and there is no change to the control limit (UCL) should be investigated as a cautionary
measurement system between the original test and the retest, if event and any root cause identified documented.
the retest result is not more than 2σR’ from the center line in (2) Five (5) or more out of the past twenty (20) MR values
either direction, then the system is not considered to be that exceed the UCL are strong indications that the system
out-of-statistical-control since the out-of-statistical-control variability has increased and hence the system shall be declared
event is not confirmed. The original and retest results shall be out-of-statistical-control.
documented. In addition, one of the following strategies shall A1.5.5 Response to Confirmed Out-of-Statistical-Control
be used to detect out-of-statistical-control conditions and Declarations:
trigger an out-of-statistical-control declaration. If the I-chart A1.5.5.1 When a system is declared and confirmed to be
limit exceedance by the original result is ≤ 0.25σchart, and there out-of-statistical-control, efforts shall be made to investigate
is no MR limit exceedance associated with either the original for assignable causes, including consideration of QC sample
or the re-test result, for the purpose of minimizing rejection of integrity.
legitimate data, it is permissible to treat the I-chart limit
A1.5.5.2 The specific actions to be taken to address an
exceedance of the original result as a possible statistical
out-of-statistical-control declaration is beyond the scope of this
aberration, and use it instead of the re-test result for site
practice. Each situation should be addressed by users and
precision calculation and run rule consideration. Otherwise,
producers of the measurement system results, taking into
document the original out-of-control result as not confirmed
account the application requirement and risk tolerance levels.
and use the re-test result for site precision calculation and run
The factors suggested in Note 1 are also applicable to consider
rule consideration.
in formulating action plans to address each type of out-of-
Strategy 1: Run Rule Strategy statistical-control declaration.
Any one of the following occurrences shall be interpreted as The following interpretations may be used as a guide in
a strong signal that a change in state of the measurement formulating action plans:

14
 D6299 − 22

Out-of-Statistical-Control Event Interpretation Eq A1.10 and Eq A1.11 as 54.25 and 57.21 and added to the
1. Single result outside the UCL or There are assignable cause(s) run chart to produce the I chart (Fig. A1.10). EWMA values
LCL and is verified by an immediate associated with both QC test results
second QC sample result that is also as the deviation from the current
(Table A1.9, Column 4) and EWMA control limits, 54.99 and
outside the UCL or LCL. control chart center line (SEV) is of a 56.47, are overlaid on the I chart. Additional results and
magnitude that is not plausible due to calculated EWMA values are added as they are determined.
common causes.
A1.5.7.3 Example of a MR Chart for Multiple Results from
2. Single result outside the UCL or There is strong evidence to support a Single Check Standard—MRi values are calculated and
LCL and the immediate second QC the notion that there are assignable
sample result is in the zone between cause(s) associated with both the
plotted in sequence. After 15 results are obtained (Table A1.4),
¯ value is calculated and added to the plot. A UCL
the MR
the 2σR’ and the UCL or between initial out-of-statistical-control and the MR is
–2σR’ and the LCL. (See Strategy 1 second QC results. added to produce the MR chart (see Fig. A1.11).
run rule (1) for comparison)
A1.5.7.4 Example of I Chart and EWMA Overlay for
3. Out-of-statistical-control declaration The process mean has moved away
due to any of the rules in A1.5.4.1 from the current control chart center Multiple Results from a Single Check Standard—The average
Strategy 1 or Strategy 2. line (SEV). of the first 20 QC results (see Table A1.4, Column 4) is
4. Out-of-statistical-control declaration The variation (precision) as exhibited
calculated and plotted on the run chart as Ī. The upper and
due to A1.5.4.2 (2) (5 out of 20 MR by the results has increased beyond lower control limits are calculated from Eq A1.8 through Eq
violations). the expected variation as determined A1.11 and added to the run chart to produce the I chart. EWMA
by the data used to set the MR
control limits.
values and EWMA control limits may be overlaid on the I chart
(Fig. A1.12). Additional results and calculated EWMA values
A1.5.5.3 If investigation leads to a significant change in the are added as they are determined.
measurement system, for example, a recalibration or other
A1.5.7.5 Example of a MR Chart for Results from Multiple
major service to the measurement system, reset the applicable
Check Standards—MRi values are calculated and plotted in
run rule counts, or restart the EWMA prior to running the next
sequence. After 15 results are obtained (Table A1.5, Column 6,
QC sample. ¯ value is calculated and
displayed again in Table A1.10), the MR
A1.5.5.4 Frequent out-of-statistical-control declarations are
added to the plot. A UCLMR is added to produce the MR chart
a strong indicator that either the measurement system is not
(see Fig. A1.13).
properly validated and hence lacks robustness, or the control
limits are not determined correctly, or both. In this case, A1.5.7.6 Example of I Chart and EWMA Overlay for
validate the measurement system against its requirements and Results from Multiple Check Standards—The average of the
confirm that the control limits have been assessed in a correct first 15 QC results (see Table A1.5, Column 6) is calculated and
manner. plotted on the run chart as Ī. The upper and lower control limits
A1.5.5.5 Refer to Practice D6792, subsection 10.3 on Qual- are calculated from Eq A1.10 and Eq A1.11 and added to the
ity Control Charts, for further guidance. run chart to produce the I chart. EWMA values and EWMA
A1.5.6 If the control chart data exhibit a strong signal of control limits may be overlaid on the I chart (Fig. A1.14).
change in the state of the measurement system, investigate for Additional results and calculated EWMA values are added as
root causes. If this investigation leads to a significant change in they are determined.
the measurement system, for example, a recalibration or other A1.6 t-Test
major service to the measurement system, reset the run rule
counts or restart the EWMA. If frequent violation of run rules A1.6.1 A two sided t-test is used to check if a sample of
or EWMA control limits is encountered, this may signal that values comes from a population with a mean different from an
the measurement system is not properly validated and hence hypothesized value, µ0. In this practice, a t-test may be
lacks robustness. In this case, validate the measurement system performed on pretreated check standard test results to check for
against its requirements. If the investigation into the root bias relative to the ARVs. Since during pretreatment, accepted
causes does not lead to a significant change in the measurement reference value(s) have been subtracted from the raw results,
system, continue with the current control chart but treat results the hypothesized mean value is zero.
as suspect and use them with great caution. A1.6.1.1 For the purpose of performing the t-test, two
A1.5.7 Examples of Control Charts for QC and Check methods for calculating the t value are presented:
Standard Results: (1) By the root-mean square method, the standard devia-
tion of the pretreated results is calculated as:
A1.5.7.1 Example of a MR Chart for QC Results—MRi
values for the data from Table A1.3 are calculated and plotted n

!
2
¯ 50.500 value
in sequence. After 15 results are obtained, the MR ( ~I
i51
i 2 Ī !
is calculated and added to the plot. Computations are shown in SI 5 (A1.22)
n21
Table A1.9. A UCLMR=1.64 is added to produce the MR chart (2) The t value is calculated as:
(Fig. A1.9).
A1.5.7.2 Example of I Chart and EWMA Overlay for QC
?
t 5 =n Ī 2 µ 0 /S I ?
where µ0 is the hypothesized mean, which is zero (see
(A1.23)

Results—The average of the first 15 QC results (Table A1.9, A1.6.1).

Column 2) is calculated and plotted on the run chart as Ī (3) Alternatively, by the MR approach, compute the alter-
=55.73. The upper and lower control limits are calculated from nate t value as:

15
 D6299 − 22

? ?
t MR 5 =n Ī 2 µ 0 / ~ MR
¯ /1.128! (A1.24) A1.7.3.2 If the computed χ2 value is less than or equal to the
where µ0 is the hypothesized mean, which is zero (see tabled value, then the site precision is either less than or
A1.6.1). statistically indistinguishable from the published reproducibil-
A1.6.1.2 Compare the computed t value from Eq A1.23 ity of the test method.
with the critical t values in Table A1.1 for (n–1) degrees of A1.7.4 Example—The site precision calculated from R' =
freedom. If tMR from Eq A1.24 is used, the appropriate degrees 2.77 × σR’ for the first 20 QC results in Table A1.3 is 1.24. The
of freedom are (n–1)/2. published reproducibility for the measurement method at the
A1.6.1.3 If the absolute value of the calculated t (or tMR) 58.88 level is 1.05. χ2 is therefore 19 × 1.242 / 1.052 = 26.50.
value is less than or equal to the critical t value, then µ0 is This value is less than the critical value of 30.1 for 19 degrees
statistically indistinguishable from the mean of the distribution. of freedom, so the site precision is not statistically greater than
For the case of check standard testing, this would indicate that the published reproducibility of the test method.
there is no statistically identifiable bias.
A1.6.1.4 If the absolute value of t is greater than the critical A1.8 Approximate F-Test
t value, then µ0 is statistically distinguishable from the mean of A1.8.1 In this practice, an approximate F-test is used to
the distribution, with 95 % confidence. For the case of check compare the variation exhibited by a measurement system over
standard testing, this would indicate a statistically identifiable two different time periods. It may also be used to compare the
bias in the measurement system. site precision estimated from a series of results from one QC
A1.6.2 Example of t-Test Applied to Multiple Results from a sample with that estimated using a different QC sample (see
Single Check Standard—For the first 15 preprocessed results in 8.6.1).
Column 4 of Table A1.4, Ī is –0.153. Since the results being A1.8.2 Compute the F value.
analyzed are the difference relative to the ARV, µ0 is zero. The
For σ estimated using moving range:
standard deviation of the first 15 preprocessed results is 0.493,
and the t value is 1.2034. The t value is less than the critical F5¯ ¯ 2
MR 1 2 /MR 2 (A1.27)
value for 14 degrees of freedom (t14 = 2.1448), so the average
¯ is the larger of the two average moving ranges,
where MR
difference between the check standard results and the accepted 1
¯
and MR2 is the smaller.
reference value is statistically indistinguishable from zero.
A1.6.3 Example of t-Test Applied to Results from Multiple For σ estimated using root-mean-square:
Check Standards—For the first 15 preprocessed results in σ 12
Column 6 of Table A1.5, Ī is –0.0719. Since the results being F5 (A1.28)
σ 22
analyzed are the difference relative to the ARV, µ0 is zero. The where precision 1 (σ1) is larger than (or equal to) precision 2
standard deviation of the first 15 preprocessed results is 0.550, (σ2). So F ≥ 1.
and the t value is 0.506. The t value is less than the critical A1.8.3 Compare the computed F value to the critical F
value for 14 degrees of freedom (t14 = 2.1448), so the average value read from Table A1.12, with (n1-1) degrees of freedom
difference between the check standard results and the accepted for the numerator and (n2-1) degrees of freedom for the
reference value is statistically indistinguishable from zero. denominator if using Eq A1.28, or 0.62(n1-1) degrees of
A1.7 Approximate Chi-Square Test freedom for the numerator and 0.62(n2-1) degrees of freedom
for the denominator if using Eq A1.27.
A1.7.1 The chi-square (χ2) test is used to compare the
A1.8.3.1 If the computed F value exceeds the tabled value,
estimated site precision to a published reproducibility value, as
then the two precisions are statistically distinguishable. We can
instructed in 9.1.2.
be 95 % confident that the process that gave rise to precision 1
A1.7.2 Compute the chi-square statistic. (σ1) is less precise (has larger site precision) than the process
For R’ estimated using moving range: that produced precision 2 (σ2).
~ n 2 1 ! R' 2 A1.8.3.2 If the computed F value is smaller than the tabled
χ2 5 (A1.25)
2R 2 value, then the precisions of the two samplings of the mea-
For R’ estimated using root-mean-square: surement process are statistically indistinguishable.

~ n 2 1 ! R' 2 NOTE A1.8—Although the approximate F-test is conducted at the 95 %

χ2 5 (A1.26) probability level, the critical F values against which the calculated F is
R2
compared come from the 97.5 percentiles of the F-statistic. If the ratio
where R' is the estimated site precision (R'=2.77 × σR' ) and ¯ 2 /MR
MR a
¯ 2 is calculated without requiring that the larger variance is in
b
R is the published reproducibility of the method. the numerator, the ratio would have to be compared against both the lower
2.5 percentile point and the upper 97.5 percentile point of the
A1.7.3 Compare the computed χ2 value to the critical χ2 F-distribution to determine if the two variances were statistically distin-
value in Table A1.11, with (n–1)/2 degrees of freedom for χ2 guishable. Because of the nature of the F-distribution, comparing
using moving range-based R’, with (n-1) degrees of freedom ¯ 2 /MR
MR ¯ 2 to the 2.5 percentile point when MR ¯ 2 /MR
¯ 2 is equivalent to
a b a b
for χ2 using root-mean-square based R’. ¯ 2 ¯ 2
comparing MRb /MRa to the 97.5 percentile point. Requiring that larger
A1.7.3.1 If the computed χ2 value exceeds the tabled value, variance is always in the numerator allows the “two-tailed” test to be
then the site precision exceeds the published reproducibility of accomplished in one step. If the variance of the two populations were
the method, with 95 % confidence. ¯ 2 .MR
equal, then there would be only a 2.5 % chance that MR ¯ 2 by more
1 2

16
 D6299 − 22

than the tabulated amount, and a 2.5 % chance that ¯ ¯ 2 by more

MR2 2 .MR 1 r = the result number in chronological sequence as it
than the tabulated amount with degrees of freedom reversed. arrives,
A1.8.4 If two precision estimates are statistically X̄r–1 = is the average calculated using all past results up to
indistinguishable, they may be pooled into a single estimate. r–1, and
For example, if MR¯ or precision 1 (σ ) was obtained from σ0 = the site historical standard deviation for the test
1 1
measurements on a single lot of QC sample material, while method as per 8.7.3.1.
¯ precision 2 (σ ) was obtained from measurements on a
MR A1.9.1.3 Plot and interpret the Q-statistic using the I/MR/
2 1
different lot of material, and, if they are not statistically EWMA methods without data pre-treatment as described in
distinguishable, they may be pooled. The appropriate pooled A1.5, with the exception that the following fixed values are to
precision estimate is: be used as the center and control limits:
(1) For the I-chart with EWMA:
For moving range based precision:
center line = 0.0; UCL = 3.0; LCL = –3.0; UCLewma = 1.5; LCLewma = –1.5
~ n 1 2 1 ! ~ MR
¯ ! 1 ~ n 2 1 ! ~ MR
1 2
¯!
2
(2) For the MR-chart:
MR pooled 5 (A1.29)
~n1 1 n2 2 2!
center line = 1.128; UCL = 3.86
For root-mean-square based precision:
A1.10 Procedure 2-B: Dynamically Updated I Chart
σ pooled 5 Œ ~ n 1 2 1 !~ σ 1 ! 2 1 ~ n 2 2 1 ! ~ σ 2 ! 2
n 1 1n 2 2 2
(A1.30) A1.10.1 Collect and prepare a new batch of QC material.
Whilst this can be performed at any time, it should be
A1.8.5 Example—Table A1.13 contains QC results for a completed when the current QC material supply remaining can
second QC sample measured by the same measurement system support no more than two analyses.
used to generate the results in Table A1.3. The standard A1.10.2 Validate the first result obtained on the new QC
deviation (σ) for the 25 results from the original QC sample material either by a concurrent test of the soon-to-be-depleted
(Table A1.3) is 0.439. The standard deviation (σ) for the 23 QC material, or by concurrently testing a check standard. If no
results for the new QC sample is 0.883. The F value is 4.05, special-cause signals are noted, then the result for the new
which is larger than the critical value of 2.30 for 22 and 24 material is considered to be valid.
degrees of freedom in the numerator and denominator, respec-
tively. Based on standard deviation, the precision of the A1.10.3 Plot the first result from the new material as the first
measurements for the two QC batches would be statistically point on the dynamically updated I / EWMA chart.
different, and hence the standard deviations should not be NOTE A1.10—One way to better understand the difference between the
pooled. dynamically updated I / EWMA and the traditional I / EWMA techniques
is as follows: The traditional I / EWMA uses a “forward looking” strategy
NOTE A1.9—For the example in A1.8.5 using Table A1.13 data, the where the in-statistical-control decision limits are fixed, and all future data
conclusion using root-mean-square based standard deviation is the correct is judged against these fixed limits. For the dynamically updated I /
conclusion, and this is different than the conclusion reached using the EWMA techniques, each new result is first judged against limits computed
moving ranges based standard deviation. Visual examination with confir- using all previous results. If it is in control, the control chart center line
mation from the Q–procedure showed a downward trend that induced and limits are re-computed by adding this latest in-control result into the
autocorrelation in the data. The moving ranges based technique did not dataset. This re-computed set of limits are then applied to the current and
capture the overall variation in the dataset, and therefore, it did not provide all past data to confirm the process is in control (because if it is, then
the correct conclusion. current and all past data should be inside the decision limits).

A1.9 Procedure 2-A: Q-Procedure A1.10.4 Center this value on the y-axis of the new chart.
Scale the y-axis to allow room for the initial result plus and
A1.9.1 Collect and prepare a new batch of QC material. minus five historical standard deviations, where the standard
Whilst this can be performed at any time, it should be deviations are appropriate to the level of the first result.
completed when the current QC material supply remaining can
support no more than two analyses. A1.10.5 Set n = 1. No center line, nor upper or lower control
limits, are plotted at this time.
A1.9.1.1 Validate the first result obtained on the new QC
material either by a concurrent test of the soon-to-be-depleted A1.10.6 Increment n by 1 to n = 2 with the arrival of the
QC material, or by concurrently testing a check standard. If no second new result. Compute and plot the center value, the
special-cause signals are noted, then the result for the new upper and lower limits using equations Eq A1.32, Eq A1.33,
material is considered to be valid. and Eq A1.34 to be used to judge the 2nd result. If both (1st and
A1.9.1.2 Beginning with the second result, calculate the Q 2nd) results are within the computed limits, declare the system
statistic as follows: is in control. Increment n by 1 to n = 3. Re-compute and re-plot
the center line and control limits using Eq A1.32, Eq A1.33,
S D
1
r21 2 ~X r 2 X̄ r21 ! and Eq A1.34. Set the re-computed control limits as the current
Qr 5 r 5 2,3,··· (A1.31)
r σ0 active control limits. Herein after, as each new result arrives,
where: plot and compare the new result to the current active control
limits. If the new result is within the limits, increment n by 1;
Qr = the Q statistic calculated using the current (rth) test
re-compute and re-plot the center line and control limits using
result Xr,
Eq A1.32, Eq A1.33, and Eq A1.34. Declare the system is in

17
 D6299 − 22
control if all results are within the re-computed limits. Set the A1.10.8.1 If the standard deviation for the new QC material
re-computed control limits as the current active control limits is the same as for the old material, continue the old MR chart
to be used to judge the next new result. All previous center and beginning with MR2, that is, the second result from the new
control limits can be optionally plotted and connected with material.
broken lines to show the trajectory of the statistics with past A1.10.8.2 If the standard deviation appropriate to the level
data. of the new material is different from the old, begin a new MR
A1.10.6.1 Center Value: chart, starting with MR2. The upper control limit for the new
n21 chart should be placed at 3.69σ.
Cn 5 ( I /~n 2 1!
i51
i (A1.32) A1.10.8.3 After at least 20 results have been obtained with
the new material, use a chi-square (see A1.7) or F-test (see
A1.10.6.2 Upper Control Limit: A1.8) to check that σ is appropriate for the new material.

UCLn 5 C n 13 σ Œ
~n 2 1!
n
(A1.33)
Strategy 2: EWMA (Exponentially Weighted Moving
Average) Strategy
where σ is the historical standard deviation appropriate for A1.10.9 EWMA Overlay on Procedure 2-B: Dynamically
test level Cn.
Update I Chart—An EWMA chart may be overlaid on the
A1.10.6.3 Lower Control Limit: dynamically updated I-chart, although it will not be meaningful
until n > 5.
LCLn 5 C n 2 3 σ Œ n
~n 2 1!
(A1.34) A1.10.9.1 The sequence of EWMA values, EWMAi, are
calculated, and overlaid on the I chart and connected. Use the
A1.10.7 Individual values, current or earlier, which are following recursion:
outside the current upper or lower control limits, are indica- EWMA1 5 I 1 (A1.35)
tions of an unstable system, and efforts should be made to
EWMAi 5 ~ 1 2 λ ! EWMAi21 1λI i (A1.36)
determine the cause. In a similar fashion to the I chart (as
described in A1.5.1), one of the following strategies shall be where λ is the exponential weighting factor, typically set to
used to detect changes in state of the measurement system that 0.4.
are considered to constitute an out-of-control situation. A1.10.9.2 The upper control limit for the EWMA chart is

Strategy 1: Run Rule Strategy

Any one of the following occurrences shall be interpreted as
UCL EWMA 5 C n 13σ ŒF G1
22λ F1 2 2λ
@ λ 1 2 ~ 1 2 λ ! 2n21 # 2 n 2 1 G
a strong signal that a change in state of the measurement (A1.37)
system has likely occurred: A1.10.9.3 The lower control limit for the EWMA chart is
A1.10.7.1 Two consecutive results on the Q–chart that are
more than 2 σ Œ n
~ n21 !
distant from the current expected value,
LCL EWMA 5 C n 2 3σ ŒF G
1
22λ F1 2 2λ
@ λ 1 2 ~ 1 2 λ ! 2n21 # 2 n 2 1 G
Cn, in the same direction; (A1.38)

A1.10.7.2 Five consecutive results on the Q–chart that are A1.10.9.4 The upper and lower EWMA (UCLEWMA and

more than σ Œ n
~ n21 !
distant from the current expected value in
LCLEWMA ) control limits associated with Cn are plotted in a
similar fashion as the UCLn and LCLn described in A1.10.6.2
and A1.10.6.3. Individual EMWA values, current or earlier,
the same direction.
which are outside the current EWMA upper or lower control
A1.10.7.3 Nine or more consecutive results on the Q–chart limits, are indications of an unstable system, and efforts should
that are on the same side of the current expected value. be made to determine the cause.
A1.10.7.4 Seven points in a row steadily increasing or
A1.10.10 Procedure 2-A, B Examples—Tables A1.13 and
decreasing.
A1.14 and Figs. A1.15 and A1.16 provide illustrative examples
A1.10.8 Continue or replace the MR chart, as appropriate. of procedures 2-A and 2-B.

18
 D6299 − 22
TABLE A1.1 95th Percentile of Student’s t Distribution (1 through
100)
Degrees of Freedom t
1 12.7062
2 4.3027
3 3.1824
4 2.7764
5 2.5706
6 2.4469
7 2.3646
8 2.3060
9 2.2622
10 2.2281
11 2.2010
12 2.1788
13 2.1604
14 2.1448
15 2.1314
16 2.1199
17 2.1098
18 2.1009
19 2.0930
20 2.0860
21 2.0796
22 2.0739
23 2.0687
24 2.0639
25 2.0595
26 2.0555
27 2.0518
28 2.0484
29 2.0452
30 2.0423
31 2.0395
32 2.0369
33 2.0345
34 2.0322
35 2.0301
36 2.0281
37 2.0262
38 2.0244
39 2.0227
40 2.0211
41 2.0195
42 2.0181
43 2.0167
44 2.0154
45 2.0141
46 2.0129
47 2.0117
48 2.0106
49 2.0096
50 2.0086
55 2.0040
60 2.0003
65 1.9971
70 1.9944
75 1.9921
80 1.99006
85 1.98827
90 1.98667
95 1.98525
100 1.98397

19
 D6299 − 22
TABLE A1.2 95th Percentile of Student’s t Distribution (105
through 200)
Degrees of Freedom t
105 1.98282
110 1.98177
115 1.98081
120 1.97993
125 1.97912
130 1.97838
135 1.97769
140 1.97705
145 1.97646
150 1.97591
155 1.97539
160 1.97490
165 1.97445
170 1.97402
175 1.97361
180 1.97323
185 1.97287
190 1.97253
195 1.97220
200 1.97190

TABLE A1.3 Example of a Sequence of Results from a Single QC

Sample
QC/Check Standard QC/Check Standard
Sequence Number Sequence Number
Result Result
i i
Yi = Ii Yi = Ii
1 55.3 14 55.2
2 55.8 15 56.5
3 56.3 16 55.7
4 56.1 17 55.6
5 55.8 18 55.2
6 55.5 19 55.7
7 55.3 20 56.1
8 55.4 21 56.3
9 56.6 22 55.2
10 56.1 23 55.4
11 55.0 24 55.4
12 55.5 25 55.6
13 55.5

20
 D6299 − 22
TABLE A1.4 Example of a Sequence of Results from a Single
Check Standard
Check Standard Accepted Difference
Sequence Number
Result Reference Value Result - ARV
(Yi) (ARV = Xi) Ii
1 55.3 55.88 -0.58
2 55.8 55.88 -0.08
3 56.3 55.88 0.42
4 56.1 55.88 0.22
5 55.8 55.88 -0.08
6 55.5 55.88 -0.38
7 55.3 55.88 -0.58
8 55.4 55.88 -0.48
9 56.6 55.88 0.72
10 56.1 55.88 0.22
11 55.0 55.88 -0.88
12 55.5 55.88 -0.38
13 55.5 55.88 -0.38
14 55.2 55.88 -0.68
15 56.5 55.88 0.62
16 55.7 55.88 -0.18
17 55.6 55.88 -0.28
18 55.2 55.88 -0.68
19 55.7 55.88 -0.18
20 56.1 55.88 0.22
21 56.3 55.88 0.42
22 55.2 55.88 -0.68
23 55.4 55.88 -0.48
24 55.4 55.88 -0.48
25 55.6 55.88 -0.28

TABLE A1.5 Example of Results for Multiple Check Standards

Where the Precision of the Measurement System Is Level
Dependent
Result Preprocessed
Raw ARV Raw
Sequence σi Result
Result Yi Xi Difference
Number, i Ii
1 71.0 71.4 -0.40 1.14 -0.35
2 65.8 64.9 0.90 1.10 0.82
3 70.3 70.2 0.10 1.13 0.09
4 66.2 67.7 -1.50 1.11 -1.35
5 93.8 93.4 0.40 1.26 0.32
6 102.9 104.0 -1.10 1.33 -0.83
7 102.2 101.8 0.40 1.31 0.30
8 103.2 103.9 -0.70 1.32 -0.53
9 100 99.8 0.20 1.30 0.15
10 71.6 71.5 0.10 1.14 0.09
11 76.7 76.4 0.30 1.16 0.26
12 61.2 61.8 -0.60 1.08 -0.56
13 44.1 43.9 0.20 0.98 0.20
14 69.71 69.7 0.01 1.13 0.01
15 59.5 59.19 0.31 1.06 0.29
16 99.63 98.87 0.76 1.30 0.59
17 93.7 95.21 -1.51 1.27 -1.19
18 103.77 103.94 -0.17 1.32 -0.13
19 96.18 96.7 -0.52 1.28 -0.41
20 99.7 100.65 -0.95 1.31 -0.73
21 84.32 84.15 0.17 1.21 0.14
22 83.29 83.75 -0.46 1.21 -0.38
23 65.16 65.93 -0.77 1.10 -0.70
24 68.19 68.0 0.19 1.12 0.17

21
 D6299 − 22
TABLE A1.6 Example Data for a Normal Probability Plot for QC
Results
Original
ith Term in Eq
Sequence z-value Sorted Result wi pi
A1.6
No., I
11 -1.83 55.0 -1.47 0.07 -5.91
14 -1.28 55.2 -1.07 0.14 -14.35
1 -0.97 55.3 -0.86 0.19 -18.70
7 -0.73 55.3 -0.86 0.19 -21.94
8 -0.52 55.4 -0.66 0.25 -25.77
6 -0.34 55.5 -0.46 0.32 -21.44
12 -0.17 55.5 -0.46 0.32 -25.34
13 0.00 55.5 -0.46 0.32 -22.80
2 0.17 55.8 0.15 0.56 -16.52
5 0.34 55.8 0.15 0.56 -18.46
10 0.52 56.1 0.76 0.78 -11.50
4 0.73 56.1 0.76 0.78 -10.80
3 0.97 56.3 1.16 0.88 -8.65
15 1.28 56.5 1.57 0.94 -5.79
9 1.83 56.6 1.77 0.96 -3.25

TABLE A1.7 Example Data for a Normal Probability Plot for

Multiple Results from a Single Check Standard
Original
Sorted ith Term in
Sort No. Sequence z-value wi pi
Result Eq A1.6
No.
1 11 -0.88 -1.83 -1.47 0.07 -5.91
2 14 -0.68 -1.28 -1.07 0.14 -14.35
3 1 -0.58 -0.97 -0.86 0.19 -18.70
4 7 -0.58 -0.73 -0.86 0.19 -21.94
5 8 -0.48 -0.52 -0.66 0.25 -25.77
6 6 -0.38 -0.34 -0.46 0.32 -21.44
7 12 -0.38 -0.17 -0.46 0.32 -25.34
8 13 -0.38 0 -0.46 0.32 -22.80
9 2 -0.08 0.17 0.15 0.56 -16.52
10 5 -0.08 0.34 0.15 0.56 -18.46
11 10 0.22 0.52 0.76 0.78 -11.50
12 4 0.22 0.73 0.76 0.78 -10.80
13 3 0.42 0.97 1.16 0.88 -8.65
14 15 0.62 1.28 1.57 0.94 -5.79
15 9 0.72 1.83 1.77 0.96 -3.25

TABLE A1.8 Example Data for a Normal Probability Plot for

Results from Multiple Check Standards
Original
Sorted ith Term in
Sort No. Sequence z-value wi pi
Result Eq A1.6
No.
1 4 -1.35 -1.83 -2.320 0.010 -7.535
2 6 -0.83 -1.28 -1.375 0.084 -11.721
3 12 -0.56 -0.97 -0.885 0.188 -15.301
4 8 -0.53 -0.73 -0.831 0.203 -20.722
5 1 -0.35 -0.52 -0.504 0.307 -22.310
6 14 0.01 -0.34 0.150 0.560 -19.259
7 3 0.09 -0.17 0.295 0.616 -20.207
8 10 0.09 0 0.295 0.616 -21.627
9 9 0.15 0.17 0.404 0.657 -23.418
10 13 0.2 0.34 0.495 0.690 -22.641
11 11 0.26 0.52 0.604 0.727 -14.400
12 15 0.29 0.73 0.659 0.745 -11.994
13 7 0.3 0.97 0.677 0.751 -12.376
14 5 0.32 1.28 0.713 0.762 -9.718
15 2 0.82 1.83 1.621 0.948 -1.860
A2* = 0.713

22
 D6299 − 22
TABLE A1.9 Example Data for I Chart and EWMA Overlay for QC
Results
Sequence Number,
QC Result (Yi=Ii) Moving Range MRi EWMAi
I
1 55.3 55.3
2 55.8 0.5 55.50
3 56.3 0.5 55.82
4 56.1 0.2 55.93
5 55.8 0.3 55.88
6 55.5 0.3 55.73
7 55.3 0.2 55.56
8 55.4 0.1 55.49
9 56.6 1.2 55.94
10 56.1 0.5 56.00
11 55 1.1 55.60
12 55.5 0.5 55.56
13 55.5 0.0 55.54
14 55.2 0.3 55.40
15 56.5 1.3 55.84

Average (1 to 15) 55.73 0.500

16 55.7 0.8 55.78

17 55.6 0.1 55.71
18 55.2 0.4 55.51
19 55.7 0.5 55.58
20 56.1 0.4 55.79
21 56.3 0.2 55.99
22 55.2 1.1 55.68
23 55.4 0.2 55.57
24 55.4 0.0 55.50
25 55.6 0.2 55.54

TABLE A1.10 Example Data for a MR Chart for Results from

Multiple Check Standards
Result Sequence Preprocessed
Moving Range, MRi EWMAi
Number, i Result, Ii
1 -0.35 -0.35
2 0.82 1.17 0.12
3 0.09 0.73 0.11
4 -1.35 1.44 -0.48
5 0.32 1.67 -0.16
6 -0.83 1.15 -0.43
7 0.30 1.13 -0.14
8 -0.53 0.83 -0.29
9 0.15 0.68 -0.12
10 0.09 0.06 -0.03
11 0.26 0.17 0.08
12 -0.56 0.82 -0.17
13 0.20 0.76 -0.02
14 0.01 0.19 -0.01
15 0.29 0.28 0.11

Average -0.073 0.791

16 0.59 0.3 0.30

17 -1.19 1.78 -0.29
18 -0.13 1.06 -0.23
19 -0.41 0.28 -0.30
20 -0.73 0.32 -0.47
21 0.14 0.87 -0.23
22 -0.38 0.52 -0.29
23 -0.7 0.32 -0.45
24 0.17 0.87 -0.20

23
 D6299 − 22
TABLE A1.11 95th Percentiles of the Chi Square Distribution
Degrees
X
Freedom
7 14.1
8 15.5
9 16.9
10 18.3
11 19.7
12 21.0
13 22.4
14 23.7
15 25.0
16 26.3
17 27.6
18 28.9
19 30.1
20 31.4
21 32.7
22 33.9
23 35.2
24 36.4
25 37.7
26 38.9
27 40.1
28 41.3
30 43.8
35 49.8
40 55.8
45 61.7
50 67.5
60 79.1
70 90.5
80 101.9

TABLE A1.12 97.5 Percentiles of the F-Statistic

Denominator, Numerator
degrees of
7 8 9 10 12 14 16 18 20 25 30 40 50 100
freedom
7 4.99 4.90 4.82 4.76 4.67 4.60 4.54 4.50 4.47 4.40 4.36 4.31 4.28 4.21
8 4.53 4.43 4.36 4.30 4.20 4.13 4.08 4.03 4.00 3.94 3.89 3.84 3.81 3.74
9 4.20 4.10 4.03 3.96 3.87 3.80 3.74 3.70 3.67 3.60 3.56 3.51 3.47 3.40
10 3.95 3.85 3.78 3.72 3.62 3.55 3.50 3.45 3.42 3.35 3.31 3.26 3.22 3.15
11 3.76 3.66 3.59 3.53 3.43 3.36 3.30 3.26 3.23 3.16 3.12 3.06 3.03 2.96
12 3.61 3.51 3.44 3.37 3.28 3.21 3.15 3.11 3.07 3.01 2.96 2.91 2.87 2.80
13 3.48 3.39 3.31 3.25 3.15 3.08 3.03 2.98 2.95 2.88 2.84 2.78 2.74 2.67
14 3.38 3.29 3.21 3.15 3.05 2.98 2.92 2.88 2.84 2.78 2.73 2.67 2.64 2.56
15 3.29 3.20 3.12 3.06 2.96 2.89 2.84 2.79 2.76 2.69 2.64 2.59 2.55 2.47
16 3.22 3.12 3.05 2.99 2.89 2.82 2.76 2.72 2.68 2.61 2.57 2.51 2.47 2.40
17 3.16 3.06 2.98 2.92 2.82 2.75 2.70 2.65 2.62 2.55 2.50 2.44 2.41 2.33
18 3.10 3.01 2.93 2.87 2.77 2.70 2.64 2.60 2.56 2.49 2.44 2.38 2.35 2.27
19 3.05 2.96 2.88 2.82 2.72 2.65 2.59 2.55 2.51 2.44 2.39 2.33 2.30 2.22
20 3.01 2.91 2.84 2.77 2.68 2.60 2.55 2.50 2.46 2.40 2.35 2.29 2.25 2.17
25 2.85 2.75 2.68 2.61 2.51 2.44 2.38 2.34 2.30 2.23 2.18 2.12 2.08 2.00
30 2.75 2.65 2.57 2.51 2.41 2.34 2.28 2.23 2.20 2.12 2.07 2.01 1.97 1.88
35 2.68 2.58 2.50 2.44 2.34 2.27 2.21 2.16 2.12 2.05 2.00 1.93 1.89 1.80
40 2.62 2.53 2.45 2.39 2.29 2.21 2.15 2.11 2.07 1.99 1.94 1.88 1.83 1.74
45 2.58 2.49 2.41 2.35 2.25 2.17 2.11 2.07 2.03 1.95 1.90 1.83 1.79 1.69
50 2.55 2.46 2.38 2.32 2.22 2.14 2.08 2.03 1.99 1.92 1.87 1.80 1.75 1.66
60 2.51 2.41 2.33 2.27 2.17 2.09 2.03 1.98 1.94 1.87 1.82 1.74 1.70 1.60
70 2.47 2.38 2.30 2.24 2.14 2.06 2.00 1.95 1.91 1.83 1.78 1.71 1.66 1.56
80 2.45 2.35 2.28 2.21 2.11 2.03 1.97 1.92 1.88 1.81 1.75 1.68 1.63 1.53
90 2.43 2.34 2.26 2.19 2.09 2.02 1.95 1.91 1.86 1.79 1.73 1.66 1.61 1.50
100 2.42 2.32 2.24 2.18 2.08 2.00 1.94 1.89 1.85 1.77 1.71 1.64 1.59 1.48

24
 D6299 − 22
TABLE A1.13 Example Data for Procedure 2-A – Q Procedure
σ0 = 0.51; λ = 0.4
Obs. No. Test Result UCL LCL UCL LCL
Q Q_ewma Q̄
(r) (X) Q Q Q_ewma Q_ewma
1 7.8 — 0.00 –3.00 3.00
2 7.1 –0.97 –0.39 0.00 –3.00 3.00
3 8.0 0.88 0.12 0.00 –3.00 3.00
4 7.7 0.11 0.12 0.00 –3.00 3.00
5 8.0 0.61 0.32 0.00 –3.00 3.00
6 7.9 0.32 0.32 0.00 –3.00 3.00 –1.50 1.50
7 7.6 –0.27 0.08 0.00 –3.00 3.00 –1.50 1.50
8 7.8 0.13 0.10 0.00 –3.00 3.00 –1.50 1.50
9 8.1 0.67 0.33 0.00 –3.00 3.00 –1.50 1.50
10 6.4 –2.56 –0.83 0.00 –3.00 3.00 –1.50 1.50
11 7.1 –1.01 –0.90 0.00 –3.00 3.00 –1.50 1.50
12 7.8 0.39 –0.38 0.00 –3.00 3.00 –1.50 1.50
13 6.8 –1.52 –0.84 0.00 –3.00 3.00 –1.50 1.50
14 7.4 –0.28 –0.61 0.00 –3.00 3.00 –1.50 1.50
15 8.2 1.26 0.13 0.00 –3.00 3.00 –1.50 1.50
16 7.8 0.42 0.25 0.00 –3.00 3.00 –1.50 1.50
17 7.0 –1.13 –0.30 0.00 –3.00 3.00 –1.50 1.50
18 6.4 –2.21 –1.07 0.00 –3.00 3.00 –1.50 1.50
19 8.2 1.35 –0.10 0.00 –3.00 3.00 –1.50 1.50
20 7.9 0.70 0.22 0.00 –3.00 3.00 –1.50 1.50

TABLE A1.14 Example Data for Procedure 2-B – Dynamically Updated I Chart
σ0 = 0.51; λ = 0.4
Obs. No. Test Result UCL LCL UCL LCL
ewma_r Cn = x̄_r-1
(r) (X_r) X_r X_r EWMA_r EWMA_r
1 7.8 7.8 —
2 7.1 7.52 7.80 9.97 5.63 8.67 6.93
3 8.0 7.71 7.45 9.33 5.57 8.21 6.69
4 7.7 7.71 7.63 9.40 5.86 8.35 6.92
5 8.0 7.82 7.65 9.36 5.94 8.36 6.94
6 7.9 7.85 7.72 9.40 6.04 8.43 7.01
7 7.6 7.75 7.75 9.41 6.09 8.47 7.03
8 7.8 7.77 7.73 9.37 6.09 8.45 7.01
9 8.1 7.90 7.74 9.36 6.11 8.47 7.01
10 6.4 7.30 7.78 9.39 6.16 8.51 7.05
11 7.1 7.22 7.64 9.25 6.03 8.38 6.90
12 7.8 7.45 7.59 9.19 5.99 8.33 6.85
13 6.8 7.19 7.61 9.20 6.01 8.35 6.87
14 7.4 7.27 7.55 9.14 5.96 8.29 6.80
15 8.2 7.64 7.54 9.12 5.95 8.28 6.79
16 7.8 7.71 7.58 9.16 6.00 8.33 6.83
17 7.0 7.42 7.59 9.17 6.01 8.34 6.85
18 6.4 7.01 7.56 9.14 5.98 8.31 6.81
19 8.2 7.49 7.49 9.07 5.92 8.24 6.75
20 7.9 7.65 7.53 9.10 5.96 8.28 6.78

FIG. A1.1 Example of a Run Chart for QC Results

25
 D6299 − 22

FIG. A1.2 Run Chart for Multiple Results from a Single Check
Standard

FIG. A1.3 Run Chart for Results from Multiple Check Standards

26
 D6299 − 22

FIG. A1.4 z–Values

27
 D6299 − 22

FIG. A1.4 z–Values (continued)

28
 D6299 − 22

NOTE 1—Probability (z < wi), where wi is the sum of the number in the left column and top row.
FIG. A1.5 pi Values

29
 D6299 − 22

FIG. A1.5 pi Values (continued)

30
 D6299 − 22

FIG. A1.9 Example of a MR Chart for QC Results

FIG. A1.6 Example of a Normal Probability Plot for QC Results

FIG. A1.10 Example of an I-Chart with EWMA Overlay for QC Re-

FIG. A1.7 Example of a Normal Probability Plot for Multiple Re- sults
sults from a Single Check Standard

FIG. A1.11 Example of a MR Chart for Multiple Results from a

FIG. A1.8 Example of a Normal Probability Plot for Results from Single Check Standard
Multiple Check Standards

31
 D6299 − 22

FIG. A1.12 Example of an I-Chart with EWMA Overlay for Multiple

Results from a Single Check Standard

FIG. A1.13 Example of a MR Chart for Results from Multiple

Check Standards

FIG. A1.14 Example of an I-Chart with EWMA Overlay for Results

from Multiple Check Standards

32
 D6299 − 22

FIG. A1.15 Procedure 2-A Q Chart for QC Material Transition

FIG. A1.16 Procedure 1-B Dynamically Updated I/EWMA Chart for QC Material Transition

33
 D6299 − 22
APPENDIX

(Nonmandatory Information)

X1. DERIVATION FOR CONTROL LIMITS FOR PROCEDURE 2-B: DYNAMICALLY UPDATED I / EWMA CONTROL
CHART

X1.1 Let {Xi: i = 1,2,...} are independent normal random X1.3.2 The variance of
variables with the same, known standard deviation, σ. If there
are no special causes, then they share a common, but unknown
F
Y n 5σ 2 ~ 1 2 λ ! 2 ~ n 2 1 ! 1 λ 2
n22

( ~1 2 λ!
, the second term of
i50
2i
G
mean µ. which is proportional to a partial geometric series, easily
summed and not-so-easily simplified to
X1.2 I-Chart (for Individual Observations)
X1.2.1 At the time of the nth observation, the monitored
variable is Xn. Its standard deviation is σ. If µ were known, then
var~ Y n ! 5 σ 2 F 1
22λ G
@ λ 1 2 ~ 1 2 λ ! 2n21 # . (X1.2)

the control limits would be µ 6 3σ. But, as µ is unknown, we X1.3.3 This variance very quickly approaches its asymp-
have to compare a newly arrived Xn to control limits con- σ 2λ
totic value of 22λ as n increases, and, if µ is known, the
structed using an estimate of µ, which, is X̄n−1, the mean of the
first n – 1 observations. control limits for Yn would be
X1.2.2 As Xn is independent of X̄n–1, the variance of
Xn – X̄n–1 is the sum of the variances, namely σ 2 1 n21 , so the
σ 2
µ63σ Œ 1
22λ
@ λ 1 2 ~ 1 2 λ ! 2n21 # . Again, when µ is unknown, it
must be estimated using X̄n–1. For the EWMA, both the

limits are X̄ n21 63σ Œ n

n21
.
variance of the estimate X̄n–1 and the correlation between X̄n–1
and with Yn has to be taken into consideration.
X1.3 EWMA Chart X1.3.4 The variance of Yn – X̄n–1 comes out to be
X1.3.1 Now the monitored variable is not Xn but the ~1 2 2 λ!σ2
var~ Y n ! 2 . Therefore, control limits for the latest
exponentially-weighted moving average Yn, where Y1 = X1 and, ~n 2 1!
EWMA Yn are:
for all n > 1, Yn = λXn + (1 – λ)Yn–1, and 0 < λ < 1 is a constant
to be chosen by the user. Yn may also be written as:
n X̄ n21 63σ ŒF G 1
22λ
1 2 2λ
@ λ 1 2 ~ 1 2 λ ! 2n21 # 2 n 2 1 .F G
Yn 5 ~1 2 λ! n21
X 11 (
i52
λ~1 2 λ! n2i
Xi (X1.1)
(X1.3)

REFERENCES

(1) Quality and Statistics: Total Quality Management, STP 1209, (5) Hunter, J. S., “The Exponentially Weighted Moving Average,” Jour-
Kowalewski, Jr., Milton J., editor, ASTM International, 1994. nal of Quality Technology, Vol 18, No. 4, October 1986, pp. 203–210.
(2) Manual on Presentation of Data and Control Chart Analysis, Manual (6) Albin, S. L., Kang, L., and Shea, G., “An X and EWMA Chart for
7A, ASTM International, 2002. Individual Observations,” Journal of Quality Technology, Vol 29, No.
(3) Champ, C. W., Woodall, W. H., “Exact Results for Shewhart Control 1, January 1997, pp. 41–48.
Charts with Supplementary Runs Rules,” Technometrics, Vol 29, No. (7) Glossary and Tables for Statistical Quality Control, ASQ Statistics
4, November 1987. Division, 4th ed., American Society for Quality, 2004.
(4) Hunter, J. S., “A One-Point Plot Equivalent to the Shewhart Chart (8) Quesenberry, C. P., “SPC Q-Charts for Start-up Processes and Short
with Western Electric Rules,” Quality Engineering, Vol 2, No. 1, or Long Runs,” Journal of Quality Technology, Vol 23, No. 3, July
1989-1990, pp. 13–19. 1991, pp. 213–224.

34
 D6299 − 22
SUMMARY OF CHANGES

Subcommittee D02.94 has identified the location of selected changes to this standard since the last issue
(D6299 – 21) that may impact the use of this standard. (Approved May 1, 2022.)

(1) Revised definition for site precision conditions in subsec- (3) Revised 8.6.2.2(2) and Note 22.
tion 3.2.20. (4) Revised subsection A1.10.8.3.
(2) Revised Note 7. (5) Revised Fig. A1.4.

Subcommittee D02.94 has identified the location of selected changes to this standard since the last issue
(D6299 – 20a that may impact the use of this standard. (Approved May 1, 2021.)

(1) Added D6617 to Section 2. (3) Added subsections 8.6.2.2, 8.7.1.1, 8.7.5
(2) Revised subsections 3.2.18, 6.2.2.1, 7.6.2, 8.3, 8.3.1, 8.4, (4) Revised Note 7 and Note 8.
8.4.3, 8.4.4, 8.4.4.1, 8.4.4.2, 8.4.4.3, 8.6.2, 8.6.2.1, 8.6.3, (5) Added Fig. 1.
8.7.2.3, 8.7.2.4, 8.7.2.5, 8.7.2.6, 8.7.2.7, 8.7.3.1, 8.7.3.2, (6) Revised Table A1.14.
8.7.3.3, 8.7.4, 8.8, 8.9, 9.1.1, 9.1.1.1, 10.5, A1.5.1, A1.5.1.1, (7) Revised Fig. A1.16.
A1.5.1.3, A1.5.4.1, A1.5.5, and A1.5.5.1.

ASTM International takes no position respecting the validity of any patent rights asserted in connection with any item mentioned
in this standard. Users of this standard are expressly advised that determination of the validity of any such patent rights, and the risk
of infringement of such rights, are entirely their own responsibility.

This standard is subject to revision at any time by the responsible technical committee and must be reviewed every five years and
if not revised, either reapproved or withdrawn. Your comments are invited either for revision of this standard or for additional standards
and should be addressed to ASTM International Headquarters. Your comments will receive careful consideration at a meeting of the
responsible technical committee, which you may attend. If you feel that your comments have not received a fair hearing you should
make your views known to the ASTM Committee on Standards, at the address shown below.

This standard is copyrighted by ASTM International, 100 Barr Harbor Drive, PO Box C700, West Conshohocken, PA 19428-2959,
United States. Individual reprints (single or multiple copies) of this standard may be obtained by contacting ASTM at the above
address or at 610-832-9585 (phone), 610-832-9555 (fax), or [email protected] (e-mail); or through the ASTM website
(www.astm.org). Permission rights to photocopy the standard may also be secured from the Copyright Clearance Center, 222
Rosewood Drive, Danvers, MA 01923, Tel: (978) 646-2600; http://www.copyright.com/

35