1、BSI Standards PublicationPD CEN/TR 10345:2013Guideline for statisticaldata treatment of interlaboratory tests for validationof analytical methodsPD CEN/TR 10345:2013 PUBLISHED DOCUMENTNational forewordThis Published Document is the UK implementation of CEN/TR10345:2013.The UK participation in its pr
2、eparation was entrusted to TechnicalCommittee ISE/102, Methods of Chemical Analysis for Iron and Steel.A list of organizations represented on this committee can beobtained on request to its secretary.This publication does not purport to include all the necessaryprovisions of a contract. Users are re
3、sponsible for its correctapplication. The British Standards Institution 2013. Published by BSI StandardsLimited 2013ISBN 978 0 580 81663 5ICS 03.120.30; 17.020Compliance with a British Standard cannot confer immunity fromlegal obligations.This Published Document was published under the authority of
4、theStandards Policy and Strategy Committee on 31 December 2013.Amendments issued since publicationDate Text affectedPD CEN/TR 10345:2013TECHNICAL REPORT RAPPORT TECHNIQUE TECHNISCHER BERICHT CEN/TR 10345 December 2013 ICS 17.020; 03.120.30 Supersedes CEN/TR 10345:2008English Version Guideline for st
5、atistical data treatment of inter laboratory tests for validation of analytical methods Guide pour le traitement statistique des donnes de validation de mthodes danalyse, issues dessais interlaboratoires Richtlinien fr die Behandlung von statistischen Daten von verschiedenen Laboren fr die Validieru
6、ng von Analysenverfahren This Technical Report was approved by CEN on 29 July 2013. It has been drawn up by the Technical Committee ECISS/TC 102. CEN members are the national standards bodies of Austria, Belgium, Bulgaria, Croatia, Cyprus, Czech Republic, Denmark, Estonia, Finland, Former Yugoslav R
7、epublic of Macedonia, France, Germany, Greece, Hungary, Iceland, Ireland, Italy, Latvia, Lithuania, Luxembourg, Malta, Netherlands, Norway, Poland, Portugal, Romania, Slovakia, Slovenia, Spain, Sweden, Switzerland, Turkey and United Kingdom. EUROPEAN COMMITTEE FOR STANDARDIZATION COMIT EUROPEN DE NO
8、RMALISATION EUROPISCHES KOMITEE FR NORMUNG CEN-CENELEC Management Centre: Avenue Marnix 17, B-1000 Brussels 2013 CEN All rights of exploitation in any form and by any means reserved worldwide for CEN national Members. Ref. No. CEN/TR 10345:2013 EPD CEN/TR 10345:2013CEN/TR 10345:2013 (E) 2 Contents P
9、age Foreword 3 1 Scope 4 2 Normative references 4 3 Principle 4 4 Preliminary rules 4 5 Procedure .6 6 Report .9 7 General remarks.9 Annex A (normative) Steps for the validation of a draft Standard 10 Annex B (normative) Synoptic of the operations described in Annex A . 13 Annex C (informative) Exam
10、ples 18 Bibliography . 44 PD CEN/TR 10345:2013CEN/TR 10345:2013 (E) 3 Foreword This document (CEN/TR 10345:2013) has been prepared by Technical Committee ECISS/TC 102 “Methods of chemical analysis of iron and steel”, the secretariat of which is held by SIS. Attention is drawn to the possibility that
11、 some of the elements of this document may be the subject of patent rights. CEN and/or CENELEC shall not be held responsible for identifying any or all such patent rights. This document supersedes CEN/TR 10345:2008. In comparison with the previous version of CEN/TR 10345, the following significant t
12、echnical change was made in Annex A: correction of the error in the last sentence of A.2 concerning the appropriate number of significant figures. PD CEN/TR 10345:2013CEN/TR 10345:2013 (E) 4 1 Scope This Technical Report is a guideline to carry out the statistical evaluation of data from an inter la
13、boratory test for method validation. Its purpose is to detail the methodology of ISO 5725-1:1994, ISO 5725-2:1994 and ISO 5725-3:1994 for the treatment of the data collected under the conditions used within the ECISS/TC 102 working groups. NOTE The present document is not a simplification of the ISO
14、 5725 standard, which is the only reference document. 2 Normative references The following documents, in whole or in part, are normatively referenced in this document and are indispensable for its application. For dated references, only the edition cited applies. For undated references, the latest e
15、dition of the referenced document (including any amendments) applies. ISO 5725-1:1994, Accuracy (trueness and precision) of measurement methods and results Part 1: General principles and definitions ISO 5725-2:1994, Accuracy (trueness and precision) of measurement methods and results Part 2: Basic m
16、ethod for the determination of repeatability and reproducibility of a standard measurement method ISO 5725-3:1994, Accuracy (trueness and precision) of measurement methods and results Part 3: Intermediate measures of the precision of a standard measurement method 3 Principle An inter laboratory test
17、 for method validation is organized at each stage of the development of a standard draft. Changing economic conditions have led to the optimization of the work of the participating laboratories. The principle retained by ECISS/TC 102 is to have three values by participant laboratory: two values obta
18、ined in repeatability conditions (day 1) and a third obtained in intra laboratory reproducibility conditions (day 2). The data evaluation requires a complex statistical analysis, which may be very confusing for a non-specialist, even if it is widely detailed in the ISO 5725 standard. Consequently, i
19、t seems useful to clarify the methodology of this standard for the above purpose and to underline that difficulties found should be discussed and solved with statisticians. Values that are identified as statistically abnormal at 99 % (outliers) using numerical Cochrans and Grubbs tests lead to the e
20、limination of the laboratory that produced them, at the stage at which they are detected: this principle is adopted even though we risk wrongly eliminating one result in one hundred. Nevertheless, it is essential to advise the laboratory concerned about the reasons for these eliminations and to pay
21、particularly attention to this laboratorys results. Furthermore, in the case of a laboratory which produces values that are determined as statistically significant at 95 % (stragglers) by numerical Cochrans and Grubbs tests, particular attention should be paid to all the other values produced by thi
22、s laboratory. 4 Preliminary rules 4.1 First rule (to be clear) The inter laboratory test should be adapted in order to meet the following requirements: PD CEN/TR 10345:2013CEN/TR 10345:2013 (E) 5 to estimate the variances linked to the tested method (repeatability, intra laboratory reproducibility,
23、inter laboratories reproducibility); to check that inter laboratories variance is compatible with defined criteria for referee or routine methods (Aim CVR or Max CVR) within their full range of application. The following shall be imposed on the working groups: a) minimal number of participating labo
24、ratories (8 / 10 / 15 - see ISO 5725-1:1994, Annex B); b) rigorous implementation of the working programme; c) appropriate number of significant figures to be given for each transmitted value, in order to allow an optimal statistical data evaluation. 4.2 Second rule (to be modest) The statistical tr
25、eatment shall be performed by application of the ISO 5725 standard, and by following the procedure described hereafter. In order to solve the statistical problems encountered, the help of an expert should be sought. It should be noted that ISO 5725-2 and ISO 5725-3 dont give an exhaustive descriptio
26、n of all situations and that they clearly indicate that statistical data treatment should be performed by a person experienced in work planning and in statistical analysis (ISO 5725-2:1994, 6.2). Various situations may arise in practice which require the application of the variance analysis general
27、modes (ANOVA), and these ISO 5725 standards only specify simplified procedures. 4.3 Minimal characteristics of data population At least two samples should be tested for each concentration range to be determined, (for example between 0,010 and 0,099 we shall have two samples, between 0,10 and 0,99 we
28、 shall have two samples and so on), and should never be less than 5 for the full range of values. It is useful, when possible, that the inter laboratory test be performed using certified reference materials (CRMs) representing at least 50 % of the total number of samples to be tested. The remaining
29、samples can be internal reference materials provided by laboratories on condition that their homogeneity has been tested and found to be compatible with inter laboratory test requirements. In the present economic situation, an inter laboratory test should be planned with at least 8 laboratories from
30、 at least 5 different countries. The values provided by the participating laboratories should have a sufficient number of significant figures in order to enable correct statistical data treatment; although the number of significant figures does not influence the precision of the result, the transmis
31、sion of rounded values containing fewer figures means that the work cannot be correctly evaluated. Expressed as a w/w percentage, values should generally be written under the form listed below, i.e. containing 3 or 4 significant figures: xx,xx x,xxx 0,xxx x 0,0xx x 0,00x xx 0,000 xxx PD CEN/TR 10345
32、:2013CEN/TR 10345:2013 (E) 6 5 Procedure 5.1 Bases Here we look at the only case where the data population is strictly obtained by the methods defined in Clause 3, that is to say, 3 determinations for each sample and from each laboratory: two determinations under repeatability conditions called Day
33、1,1 and Day 1,2; the third determination under reproducibility conditions called Day 2. Statistical data evaluation is performed for each content level, that is to say in the present case, sample by sample. 5.2 Raw data examination Raw data shall be typed into a table and then printed. The raw data
34、table should be studied in order to detect potential typing errors in the data supplied by each participating laboratory and/or in the final input stage (ISO 5725-2:1994, 7.2.6). Performing a normality test of the data population to be tested is recommended. 5.3 Intra laboratory repeatability varian
35、ce (Cochrans test) A first graphical evaluation of the raw data may be performed in order to test the intra laboratory repeatability consistency by using Mandels k test (all data included): its only purpose is to get an overview of the data population. Further statistical treatments are carried out
36、under the hypothesis that the intra laboratory repeatability variances belong to the same normal population. Cochrans test should therefore be performed (ISO 5725-2:1994, 7.3.3) in order to detect unexpected values of intra laboratory variances, which shall then be discarded so that statistical anal
37、ysis may be pursued. Strictly speaking, Cochrans test should only be used to evaluate a population of measurements obtained in repeatability conditions. In practice, it is advisable to proceed as follows: perform Cochrans test with Day 1,1 and Day 1,2 values, after a normality test for this data pop
38、ulation; discard laboratories having an unexpected variance. It is advisable not to perform an iteration when using Cochrans test, except in the case of a large laboratory population (i.e. greater than 15), or when there is a particular statistical reason to justify it. A common rule sometimes used
39、is that the outliers shall not represent more than 10 % of the whole data. In practice, in order to have a clear view of the data after the Cochrans test has been performed, it is advisable either to print a new table of the remaining data or to clearly identify discarded data. 5.4 Intra laboratory
40、reproducibility variance (Grubbs test) Further statistical evaluations are carried out under the hypothesis that intermediate variances (intra laboratory reproducibility) belong to the same normal population. Grubbs test should be applied (ISO 5725-2:1994, 7.3.4) in order to detect unexpected means
41、that should then be discarded so that statistical analysis may be pursued. PD CEN/TR 10345:2013CEN/TR 10345:2013 (E) 7 In practice, it is advisable to proceed in the following order: apply Grubbs test to the daily means data, that is to say to the pair of values (Day 1,1 + Day 1,2)/2 and Day 2 for a
42、ll of the laboratories; firstly, test the highest mean after having confirmed that the data population is normal; if the test is positive, discard the laboratory concerned; secondly, test the lowest mean; if the test is positive, discard the laboratory concerned; if neither of the two former tests d
43、etect an unexpected mean, perform the test with the two highest mean values; if the test is positive discard the laboratory(ies) concerned; perform the test with the two lowest mean values; if the test is positive discard the laboratory(ies) concerned. NOTE 1 It is advisable not to perform iteration
44、s when using Grubbs test. NOTE 2 It is not necessary to perform the test whenever the value or the two values to be tested come from a population containing respectively a second or a third identical value. 5.5 Inter laboratory reproducibility variance (Grubbs test) A first graphical evaluation of t
45、he raw data may be performed in order to test the inter laboratory consistency using Mandels h test (all retained data remaining after the application of Cochran and Grubbs tests are included). Further statistical treatments are carried out under the hypothesis that the laboratories means belong to
46、the same normal population. Grubbs test should be performed (ISO 5725-2:1994, 7.3.4) in order to detect unexpected laboratories mean values which shall then be discarded so that statistical analysis may be pursued. In practice, it is advisable to proceed in the following order: firstly, test the hig
47、hest laboratory mean, (Day 1,1 + Day 1,2 + Day 2)/3, after having confirmed that the data population is normal; if the test is positive discard the laboratory concerned; secondly, test the lowest laboratory mean; if the test is positive discard the laboratory concerned; if neither of the two former
48、tests detect an unexpected laboratory mean, perform the test with the two highest laboratory mean values; if the test is positive discard the laboratories concerned; perform the test with the two lowest laboratory mean values; if the test is positive discard the laboratories concerned. PD CEN/TR 103
49、45:2013CEN/TR 10345:2013 (E) 8 NOTE 1 It is advisable not to perform iterations when using Grubbs test. NOTE 2 It is not necessary to perform the test whenever the value or the two values to be tested come from a population containing respectively a second or a third identical value. 5.6 Retained data examination The application of Cochran and Grubbs tests may lead to the elimination of raw data from some laboratories. These eliminations have been performed for e