Precision of test methods; Determination of repeatability and reproducibility for a standard test method by interlaboratory tests
Some standard content:
1 Introduction
National Standard of the People's Republic of China
Precision of test methods
Determination of repeatability and reproducibility for a standard test method by interlaboratory tests
Precision of test methods Determination of repeatability and reproducibility for a standard test method by interlaboratory tests 1.1 Scope of application
LC 519.281
GR 6379B6
The standard applies to test methods that have been or are being standardized and are mastered and used by a large number of laboratories, with a single value as the final test result.
t.2 Purpose
The test conducted to determine the repeatability and reproducibility of a standard test method is called a precision test. This standard specifies the basic principles and methods for organizing precision tests and statistical analysis among laboratories to determine the values of repeatability r and reproducibility R of a standard test method. 1.3 Application conditions
This standard assumes that the test results of each laboratory on the same level of material are from the same normal population or an approximately normal population, and that the indoor variances of the laboratories participating in the test are consistent. When applying this standard, full attention must be paid to the above assumptions (see 3.23.3): 1.4 Terms and symbols
The statistical terms and symbols used in this standard are shown in GB3358-82 "Statistical terms and symbols": For ease of use, the main terms and symbols used in this standard are listed in Appendix A (Supplement). 1.5 Reference standards
This standard refers to and adopts Guangdong International Standard ISO5725-8] "Precision of test methods - Determination of repeatability and reproducibility of standard test methods through inter-laboratory tests"
2 Organization of precision test
2.1 Test institutions and personnel
2.1.1 Leading group
The technical unit responsible for the test method standard or the drafting unit shall be responsible for organizing the leading group, and relevant units are invited to participate. The members of the leading group must be familiar with the test method and its application. Among them, at least one member is allowed to have knowledge of mathematical statistics, test design and data analysis. 2.1.2 Executive Person in Charge The specific organization of the test should be entrusted to a laboratory. The laboratory should designate a member as the executive person in charge to be responsible for the overall organization of the inter-laboratory precision test. 2.1.3 Test Person in Charge Each unit participating in the test should designate at least one member as the test person in charge, who is responsible for the test work of the unit. 2.1.4 Operator National Bureau of Standards 1986-05-13 Issued 1987-05-01 GB 679-86 Each laboratory participating in the test should designate a member who can perform the test according to the regulations as an operator. 2.2 Tasks and Division of Work
2.2.1 Tasks of the Leading Group
The task of the leading group is to discuss and determine the following issues: a: Test benchmark. Is there an available benchmark that meets the requirements of the test method? For example, standard materials and standard equipment. b, Level range and number of levels. What is the range of levels that may be followed in practice? How many levels should be taken in the test? What are the appropriate samples equivalent to these levels?
C, Sample preparation and distribution, how to perform a certain degree of homogenization when preparing the samples? Otherwise. The inhomogeneity of the samples will be included in the values of repeatability and reproducibility R. How large should the number of repetitions n be, and how many samples will be distributed to each laboratory? Is one sample sufficient for n repetitions or two separate samples distributed to each laboratory at each level? Is it necessary to conduct a split level test? (See 2.6, 2) Is it necessary to distribute additional samples for practice before the formal test? d, Laboratories participating in the test, the number of laboratories participating in the inter-laboratory precision test, the conditions that should be met, and which laboratories will be selected to participate in the test in the end.
e: Implementation of the test work. Specify the format of test records and test reports, specify the decimal places that test results should be reported, and propose what other situations and information are needed in addition to the numerical values of test results? Make clear requirements for the work of the executive person in charge, and solve relevant problems in the implementation of the test work in a timely manner.
. Repeatability and reproducibility values. Statistical analysis of the test results, discuss the report on the statistical analysis, and finally determine the values of repeatability and reproducibility R, decide whether it is necessary to improve the standard test method, and report the above results to the technical unit. 2.2.2 Tasks of the executive person in charge
The task of the executive person in charge is to organize the test planned by the leadership group. The person in charge must: Ensure that the number of laboratories required to participate in the test is recruited, and ensure that each laboratory has determined the test person in charge and operator who meets the requirements,
b. Organize and supervise the preparation of samples, and issue the samples quickly. Special instructions should be given to unstable samples. A certain number of reserve samples must be reserved for each level.
c. Prepare work instructions and notify each test leader as soon as possible to collect opinions and questions. Design appropriate forms for the operator's work records and the test leader's test result report. Collect test results and compile them into forms that are convenient for statistical analysis. e.
2.2.3 Tasks of the test leader
The task of the test leader is to ensure that the progress of the unit's test work is consistent with the instructions issued by the executive leader and report the test results, mainly:
a: Transfer the sample between operators according to the instructions of the executive leader. b. Supervise the progress of the test work. Ensure the strict implementation of the test method, but the person himself should not participate in the test and shall not change the test method.
c: Collect test results and all abnormal conditions or problems, and report to the executive leader. 2.2.4 Tasks of the operator
The operator's task is to strictly complete the test according to the standard test method and report all abnormal conditions and problems. 2.3 Number of test levels, number of laboratories and number of repetitions 2.3.1 Number of test levels
The number of levels selected in the precision test should take into account the applicable level range and the cost required to complete the test. If the level range is very wide, the repeatability and reproducibility R may be related to the level value n. In this case, at least 6 levels should be selected so that the relationship between repeatability and (or) reproducibility R and the level value m can be better determined. If the level range is narrow and the relationship between repeatability and (or) reproducibility and m needs to be determined, at least 4 levels should be selected. 2.3.2 Number of laboratories
The number of laboratories is related to the number of levels α. For single-level tests, the number of laboratories should be no less than 15. For multi-level tests, the number of laboratories should be no less than 8. 2.8.3 Number of repetitions GB6379--86 For the number of repetitions II, except for the case where multiple repetitions are required (such as the test of some simple objects), the recommended value is 2. 2.4 Selection and requirements for laboratories participating in the test 2.4.1 Laboratories participating in the precision test should be randomly selected from the experimental space where the test method is applied as much as possible, and the distribution of laboratories participating in the test in different climatic regions should be taken into consideration. 2.4.2 The requirements for laboratories participating in the test are: a. Prepare the ten necessary instruments, equipment, reagents and other necessary items used for the test; b. Organize the operation strictly in accordance with the requirements of the test procedures, handle the samples strictly in accordance with the instructions, and have qualified operators to conduct the test and ensure the test quality. c. Complete the test strictly in accordance with the time and steps specified in the plan. 2.5 Requirements for samples According to the definition of repeatability and reproducibility, the tests of each laboratory in the precision test must use the same samples. Therefore, the uniformity of the samples must be ensured in the preparation, distribution, transportation, storage and testing of the samples. For example, the following points should be noted: 2.5.1 The samples must be prepared and distributed in strict accordance with the provisions of the standard test method. For each level, samples should be prepared from batches of materials to ensure that the number of samples obtained is sufficient to complete the entire test. Some reserve materials should be kept. Liquid or micro-powdered materials should be stirred evenly. For mixtures of powders of different densities or different particle sizes, attention should be paid to eliminating segregation caused by shaking. Special storage arrangements must be made for unstable materials. Methods and handling methods, for example: samples that may react with the atmosphere should be sealed in a vacuum or inert gas-filled narrow-necked bottle; perishable samples (food or blood, etc.) must be sent to the laboratory in a refrigerated state; hygroscopic, oxidizable or volatile samples are prone to deterioration after the container is opened. Therefore, each level should be divided into two containers, and the test name and sample text mark should be written on the label of all samples. When distributing samples, the sample name, quantity range and detailed instructions on transportation, storage and extraction must be indicated. 2.5.2 For some samples that cannot be transported, operators and equipment from participating laboratories can be concentrated at the test site for testing. 2.5.3 When the measured parameters are short-term or variable (such as river water, etc.), care should be taken to always test under conditions as close to the phase as possible.
2.5.4 When testing samples related to uniformity (such as metal, rubber or textile fibers), and it is not possible to repeat the test on the same sample, the non-uniformity of the sample will be reflected in the repeatability? and reproducibility R: At this time, the values of repeatability and reproducibility R obtained by precision testing are only applicable to the specific materials used, and should be clearly stated. 2.6 Specific organization of test work
2.6.1 Each laboratory must retest each of the g levels n times, for a total of 4n tests. The test work must be organized and carried out according to the following requirements.
All n tests shall be performed by the same operator using the same equipment. b. The same set of tests shall be performed under the conditions specified in the definition of repeatability, i.e. within a short time, by the same operator and without any intermediate recalibration of the equipment (except when such intermediate recalibration is an essential part of the test).
c. If it is necessary to change operators during the test, they shall not be changed between n tests on the same test set. They may be changed after n tests on a certain test set have been completed. The change shall be reported together with the test set. d. It shall be emphasized to the operator that the tests carried out under the conditions specified in the definition of repeatability shall be regarded as n independent tests on different samples of the test set. The purpose of the test is to determine whether the test results will be affected by multiple errors, so that the results of the previous tests will not affect the subsequent results and thus affect the repeatability variance. 2.6.2 If there is a concern that the operator's 1st test will be affected by the results of the 2nd test, a split test may be used when n=2. It is known to the operator that two interlaced samples or one sample are tested twice. Two series of samples with different sound in water, m, and m (1ma-m are very small), are prepared. Each laboratory in the laboratory tests the test groups of series A and series B once. The test results of series A and series B must be clearly distinguished. The two cannot be interchanged: the values of repeatability T and current R obtained from the split-level test correspond to the average level m = GB6379-86
-m, +mm). The statistical analysis of the split-level test is slightly different. See 3.3.2.2n2
2.6.3 The time allowed from the receipt of the sample to the end of the test must be limited. 2.6.4 The equipment should be calibrated in advance according to the provisions of the standard procedures. 2.6.5 The following requirements should be made for the operator: | |tt||Before conducting the test, the operator should not receive additional instructions beyond the test method standard. .
The operator should be encouraged to ask questions about the test method standard, especially whether the provisions in the standard are clear and concise. b.
When the operator conducts the test for the first time or after a period of time, the normal precision may not be achieved. At this time, according to the decision of the leadership team or the test person in charge, the operator may be required to conduct a small amount of practice to make him proficient in the test method before the formal test. This practice cannot be conducted on fixed specimens, but should be issued to specimens outside the execution person. d. The operator should be required to report any occasional failure to circumvent instructions or comply with instructions. 2.6.6 When reporting the test results, each laboratory participating in the test should use the same decimal places. In precision tests, it is advisable to use more than the usual test method standards. The decimal point specified in the standard shall be taken as the maximum value. When repeatability r or repeatability R is related to the level value m, appropriate rounding provisions may be made for different levels.
2.7 Report of test results
The test responsible person for each experimental case shall write a comprehensive report of the test. The report shall include the following: Final test results. In particular, to prevent errors in copying or printing, copies of the results obtained by the operator can be used. a.
The original observations used to calculate the final test results. The operator's work records should be copied as much as possible. The operator's opinions on the test method standard. c.
The anomalies and interferences that occurred during the test. A description of possible changes in operators and which tests were performed by which operators should be included. d.
The date the sample was received.
The time and date of the test. ||t t||Information about the equipment used in the test. h
Other relevant information,
3 Statistical analysis of precision test
3.1 Overview
3.1.1 Person in charge of statistical analysis
The statistical analysis of the test results obtained from the precision test shall be the responsibility of the member of the leadership team who has the knowledge of test design and data analysis. 3.1.2 Main tasks of statistical analysis
Statistical analysis includes the following five tasks:
Collate and verify the original test results;
b. Check and deal with abnormal values;
c. Calculate the total half mean m, repeatability and reproducibility R values; d. Establish the mathematical relationship between (or R) and m and determine the final value of repeatability and reproducibility R; e. Report to the leadership team for the leadership team to make a decision. [The above b:c, d tasks can be processed by computer system, and the calculation procedures are shown in Appendix E (reference). 3.1.3 Units and data
The combination of laboratories and levels is called a "unit" of precision test. Ideally, there are p9 units composed of P laboratories and g levels of tests, each of which contains data from n repeated tests. In practice, there may be redundant data, missing data and abnormal values, and different situations should be handled accordingly. 3.1.3.1 Redundant data
Sometimes more than n repeated tests may be made. In this case, the person in charge of the test should report the reasons for this situation and which data are correct. If all the data are valid, all the data can be used for calculation according to the provisions of 3.3.1.3. 3.1.3.2 Missing data
GB 6379-86
Sometimes several data may be missing, such as due to loss of specimens or negligence during testing. The person in charge of the test shall report the reasons for the missing data. For units with some or all missing data, the calculation shall be carried out according to the provisions of 3.3, 1.2. In the split level test, if one of the two data of a single core is missing, the other data must be discarded and the entire single core shall be regarded as a blank unit.
3. 1.3.3 Outliers
In the original test results or the values derived from the original data, there are values that deviate too much from other values. Such values may be outliers and must be inspected and handled in accordance with the provisions of 3.2. 3.1.4 Tables and symbols
3.1.4.1 Tables
Arrange the data according to the following format.
Arrange the raw data according to Table 1.
Non-split level test
Laboratory
Arrange the unit average according to Table 2.
Laboratory
Split level test
Laboratory
According to Table 3 Arrange the unit variance or unit difference.
Non-split level test
Laboratory
3.1.4.2 Symbols in the table
GB6379—86
Laboratory
The meanings of the symbols in Table 1, Table 2 and Table 3 are as follows: nij
The number of test results in the unit of the ith level of the ith laboratory. Yik-The kth data of the test result in the unit of the ith level of the ith laboratory. Split level test
(k=1, 2, **, n)
The test results of sub-level A and sub-level B of laboratory i at level. YiA +Yie—
P,—For the first water, after eliminating outliers, the number of laboratories that reported at least one test result; in split-level tests, only when there are test results in both sub-levels, the laboratory is included in the number of laboratories, and the unit mean, when calculated, should be one decimal place more than the original test result in Table 1. Si,——Unit variance, used in the case of non-split-level tests, the unit standard deviation S, should be calculated with one decimal place more than the original test result in Table 1. e
-[(Yik)*-
—Unit difference, in the case of split-level tests, the positive and negative signs should be taken into account. di= Yia- Yie
Note: After the outlier test, some data may be corrected or eliminated. When calculating the repeatability and reproducibility R after the domain, the values of Y, 口 and P, may be different from the values listed in Tables 1, 2 and 3 based on the original data. Therefore, when reporting the final values of A and R, it is necessary to indicate that the noise data have been corrected or eliminated.
3.1.4.3 Simplified symbols
GB 6a79-86
In 3.2 and 3.3. In the statistical tests and calculations of repeatability T and reproducibility R, since they are all performed separately at each level, j is fixed and does not need to be used as a subscript, so the subscript j is omitted for the symbols defined in 3.1.4.2 in the corresponding chapters. 3.2 Inspection and treatment of outliers
3.2.1 Judgment of outliers
This standard stipulates that the Grubbs method is used to inspect the outliers in the test results within each unit, and the Cochran method is used to inspect the outliers in the variance of each laboratory: the Grubbs method and the Dixon II method are used to inspect the outliers in the average value of each laboratory.
P represents the probability of occurrence of the observed value of the test statistic of the above three methods, and the test result is one of the following situations...a. P5%, that is, the Cochran, Grubbs or Dixon test statistic is less than its 5% critical value, then the tested value is judged to be a normal value:
b, %, P1%, that is, the test statistic is between the 5% and 1 critical values, then the tested value is judged to be an abnormal value and is marked with a single asterisk "
c, P1\i, that is, the test statistic is less than the critical value, then the tested value is judged to be a highly abnormal value and is marked with a double asterisk "\ The critical value tables of Cochran test, Grubbs test and Dixon test are listed in Appendix B (Supplement), Appendix C (Supplement) and Appendix D (Supplement) respectively.
3.2.2 Test for outliers
3.2.2.1 Cochran test
Cochran test is a test for homogeneity of variance and is only suitable for non-split water half test. Given P variances S, they are all variances of n test results with the same number of repetitions: The statistic of Cochran's test is C
where Smax represents the maximum value among all S. If the test is significant, then according to 3.2.1, Sma is judged to be an outlier or a highly abnormal value
If the number of test results in some units is not constant due to redundant, missing or excluded data, the number of test results in the majority of units is taken.
If the largest variance is judged to be an abnormal value or a highly abnormal value, after eliminating this value, the Cochran test is performed on the remaining data in pairs. The above process can be repeated until no abnormal value can be detected. 3.2.2.2 Grubbs test
The Grubbs test used in this standard is quoted from GB4883-85 "Statistical Processing and Interpretation of Data - Delimitation and Processing of Normal Samples".
Grubbs test is only applicable to the case where an outlier is found in the test results. Arrange n observations X, XX, in order of numerical value as XX.. The statistic of Grubbs test is the larger of
GnXnX and -||tt| ... Dixon test statistics are the most
sample size
n :1113
n ::1~3u
the larger of
the larger of
the larger of
the larger of
if the test is significant, according to 3.2.1. n. is judged as an outlier or a highly outlier. If the minimum or maximum value in the data is judged as an outlier or a highly outlier, then after eliminating this value, the remaining n-1 data should be subjected to Dixon test: the above process is repeated until no normal value can be detected. 3.2.3 Handling of abnormal values
3.2.3.1 For abnormal values or highly abnormal values confirmed after testing, the cause should be checked from the technical error aspect first, for example, whether it is due to negligence during testing, calculation error, clerical error during writing or wrong sample, etc. If it is a calculation error or clerical error, it should be corrected. For the wrong sample, its test result should be filled in the corresponding unit. After the above inspection and correction, the abnormal value test should be carried out again. If it still cannot change the suspected test result, the highly abnormal value should be eliminated and the abnormal value should be retained. If there is sufficient reason, the leadership team can also decide to retain the highly abnormal value. 3.2.3.2 The Grubbs test and the Tsukkinson test specified in this standard must be carried out separately and independently. The Grubbs test is only carried out once, and the results of the two tests are recorded separately. If only one outlier is found, the result obtained by the Grubbs test is used; if multiple outliers are found, the result obtained by the Dixon test is used. 3.2.3.3 The Cochran test and Dixon test specified in this standard can be used repeatedly to test outliers one by one. However, when the approximation to the normality assumption is not sufficient, it may lead to continuous exclusion, so special caution should be exercised in making the final decision. The reasons should be carefully examined from the aspects of the test method and the organization and implementation of the test work. In particular, when multiple outliers are found, which ones to keep and which ones to exclude should be carefully studied before deciding.
3.2.3.4 If a laboratory has outliers and (or) highly outliers at several different levels, it means that the laboratory has a large number of indoor errors and (or) systematic errors. The laboratory should be regarded as an outlier laboratory and all its data should be excluded. However, if the number of outlier laboratories excluded on this basis is more than 15%, the leading group should study the applicability of the standard test method. Members of the leadership team who have knowledge of data analysis should give their opinions on the situation of outlier laboratories and report to the leadership team. 3.2.3.5 The Cochran test specified in this standard only tests the maximum value in the variance. It is possible that the variance of some laboratories is lower than that of other laboratories at all or most levels. This may be due to advanced technology and equipment, or improved or incorrect use of standard test methods. Such situations should be reported to the leadership team so that further investigation and research can be considered. 8.a Calculation of total mean m, repeatability r and reproducibility R After the original test results have been tested and processed for outliers, the total mean m, repeatability and reproducibility R values should be calculated for each level. When there are g levels, the calculation results are expressed as mi, and R, (j=1, 2,,). 3.3.1 Data, units and number of repetitions used in calculations 3.3.1.1 Basic data
The basic data used for calculations are listed in Tables 1, 2 and 3 of 3.1.4.1. However, it should be noted that some of the data in the above three tables may have been corrected or eliminated after the outlier test. The corrected data should be used, and the eliminated data cannot be used for calculation. 3.3.1.2 Non-blank cells
The cells used in the calculation should be non-blank cells. The number of non-blank cells is generally the same in Table 2 and Table 3 for a certain level. However, due to missing data, sometimes there is a cell containing only a single test result in Table 1. In this case, the cell is not blank in Table 2, but a blank cell in Table 3. In this case, you can: remove this single data, so that both Table 2 and Table 3 become blank cells. a.
b. If you think that the data is information that should not be lost, you can add a nominal value "zero" in Table 3, which will not affect the final result.
The number of non-blank cells at different levels may be different, so the subscript i should be used. 3.3.1.3 Number of repetitions for each unit
Due to redundant data, missing data or elimination of some original data, the number of repetitions for each unit in Table 1 may not be the same. For level i of the i-th laboratory, the number of repetitions is recorded as tt. 3.3.2 Examples of calculation formulas and calculation steps for two types of tests The following are examples of calculation formulas and calculation steps for two types of tests. Each example only discusses one level. The subscript j is omitted for simplicity. The highly abnormal values found have been eliminated. Since Table 1 is irrelevant to the calculation, the accepted data in Tables 2 and 3 are only listed. If S is a negative value in the calculation, S-0 is substituted in the calculation formula for reproducibility R. 3.3.2.1 Non-divided level test
a. Basic data of a certain level in Table 2 and Table 3 Laboratory
Number of repetitions
Calculation formula and numerical results
Number of laboratories p
Ti=zn,
T,=(n,-I) S
T (p- 1)
S= StI S?
R =- 2.8VSK
GB 6379 -86
T, - 508.30
7: = 10 767.765 0
Te - 24
T = 58
T 0.632 5
(24 :11)
S1 - [24× 10 767.765 0-508, 3024 (11-1)
- 0. 048 6 7× 1
24 (11 -1)
24- 58
S = 0. 088 4 + 0. 048 6 = 0, 137 0 bottle
r = 2.8 /0. 648 6 =0.62
R-2.8VU.1370-1.01
-,=…nThen the calculation formula can be simplified to
Note If the number of repetitions is equal, that is, n,= n,=…T-Ey
Sh - St- S?
r - 2.8/s#
St - [-, T]- S.
p(p-1)
R = 2.8 /ST
3.3.2.2 Split level test
a. Basic data of the levels in Tables 2 and 3
Calculation formulas and numerical results
Laboratory number P
T, =Edi
2p(p-1)
St=[ -
p(p-1)
St - st + $?
GB 8T986
18-615
18 +895
- 0, 47
T=169.390
T,=3189.327850
T, = -4. 52
T,=2.2838
9 ×2. 283 8 -(- 4.52) *
2×9×8
= 0. 000 860
9 × 3 189. 327 850 169. 39019×8
0.000 860
S.= 0.152 050
S1 = 0.152 050 - 0. 000 860
= 0. 152 910
7 =2.8 st
R= 2. 8SK
GB6379—86
F= 2.8 /0.000 860 =0.082
R = 2, 8 /0. 152 910 = 1. 0953.4 Establish the regression equation between r (or R) and m and determine the final value of repeatability and reproducibility R For single-level or multi-level tests with less than four levels, the repeatability and reproducibility R values of each level obtained by 3.3 are the final values of the corresponding level.
For multi-level tests with no less than four levels, it should be considered whether there is a functional relationship between (or R) and m. If there is an obvious functional relationship, the regression equation is obtained based on m at different levels and the corresponding (or R;) obtained by 3.3. The values of repeatability r and reproducibility R of each level obtained by the regression equation are taken as the final values. If there is no clear functional relationship, the repeatability and reproducibility R values at each level obtained from 3.3 are the final values. 3.4.1 Function type
Generally, one of the following two functions can be used for fitting! a.
r=a+bm (linear relationship)
gr = igc + d Igm
(or the equivalent exponential relationship r cm, d 1) If the person in charge of the statistical analysis has reason to believe that it is another functional relationship, other functions can also be used for fitting. 3.4.2 Determination of functional relationship
3.4.2.1 Obtain the regression equations of the two functions respectively for the case of T =α + m
Use weighted quadratic alternative regression to find the values of α and b. The steps are as follows: First, find ru: Let Waj=
T=ZWo, mi
T =ZWujmit
ruy--a + br m
Secondly, find Tai, let Wi,=
Here r=
a,=[
b,=[
T TT T,
TT - T
T T-TT
TT - T
T,= m,Waj
T =ZWajr,
α=[
TT-TTT
(2)=[Www.bzxZ.net
b,=[
T TT T,
TT - T
T T-TT
TT - T
T,= m,Waj
T =ZWajr,
α=[
TT-TTT
(2)=[
b,=[
T TT T,
TT - T
T T-TT
TT - T
T,= m,Waj
T =ZWajr,
α=[
TT-TTT
(2)
Tip: This standard content only shows part of the intercepted content of the complete standard. If you need the complete standard, please go to the top to download the complete standard document for free.