INTERNATIONAL Iso STANDARD 5725-1 First edition 1998512415 Accuracy (trueness and precision) of measurement methods and results — Part 1: General principles and definitions Exactitude (ustesse ct fidélité! des résultats et méthodes de mesure — Partie 1: Pancipes généraux et détintions Kotarence number (80 5725-1:1994(6)ISO 5725-1:1994(E) Contents 1 Scope 2 Normative references 3 Definitions 2 4 Practical implications of the definitions for accuracy experiments 4 41 Standard measurement method 4 4.2, Accuracy experiment 4 4.3 Identical test items = 5 4.4 Short intervals of time 5 4.5 Participating laboratories «8 4.6 Observation conditions 5 5 Statistical inoue! 6 6.1 Basic model 6 5.2. Relationship between the basic model and the precision 7 5.3 Altemative models anee T 6 Experimental design considerations when estimating accuracy 7 6.1 Planning of an accuracy experiment 7 8.2 Standard measurement method isin sien @ 6.3 Selection of laboratories for the accuracy experiment 8 6.4 Selection of materials to be used for an accuracy experiment 10 7° Utilization of accuracy data " 7.1 Publication of trueness and precision values 72 Prectical applications of trueness and precision values Annexes ‘A Symbols and abbreviations used in ISO 5725 nnn 18 © 180 jaa AU cighis reserved, Unloss otherwise specified, ro part of thi aublicatian may bo reproduced fr uttized in any form or by ay means, aectione Or mechancal, ncudng photocopying and Snetefi, without pormiozon writing from tho publisher. Intemational Organization for Standardization (Case Postae 66 = Cx-1217 Guravw 20 * Switveland Printed i Switzedande1so B Charts of uncertainties for precision measures © Bibliography 1SO 5725-1:1994(E} 5 ”eso Foreword ISO (the International Organization for Standardization) is a worldwide federation of national standards bodios {ISO member bodies). The work ‘of preparing International Standards is normally cared out through ISO technical committees. Each member body interested in a subject for which @ technics! committee has been established has the right to be represented on thet committos. International organizatione, governmental ‘and non-governmontal, in liaison with ISO, also take part in the work. ISO collaborates closely with the International Electrotechnical Commission (EC) on all matters of electrotechnical standardization. Draft Intemational Standards adopted by the technical committees are Citculated { the rember bodies for voting. Publcetion es an International Standerd requires approval by at loast 75 % of the member bodiee casting a vote, International Standard ISO 6725-1 was prepared by Technical Committee ISO/IC 68, Applications of staistical methods, Subcommittee SC 6. ‘Measurement methods and results. ISO 725 consists of the following parts, under the general title Accuracy {irueness and precision) of measurement methods and results: — Part 1: Gonoral principles and definitions — Pant 2: Basic method for the deterinination of repestability and re~ producibilty of a standard measurement method — Part 3; Intermediate measures of the precision of a standard ‘measurement method — Part 4: Basie methods for the determinstion of the trueness of 2 standard measurement method — Part &: Alternative methods for the determination of the precision of a standard measurement method — Part 6: Use in practice of accuracy values Parts 1 to 6 of ISO 5725 together cancel and replace ISO 5/7:19185, hich has heen extended to cover trueness (in addition to precision) and intermediate precision concitions (in audition to repeatability and repro~ ductility conditions) Annexes A and B form an integral part of this part of ISO 5725. Annex C is for information only.e180 Introduction 0.1 150.5725 uses two terms “trueness” and "precision" to describe the accuracy of a measiement method “Truenass” rafers to the close ness of agreement between the arthmetic mean of a larga number of test results and the true or accepted reference value, “Precision” refers to the closeness of agreement between test results 0.2 The need to consider “precision” arises becsuse tests performed on presumably identical materials in presumably identical circumstances do not, in general, yield identical results. This is attributed to unavoidable random errors inherent in every measurement procedure: the factors that influence the outcome ot a measurement cannot all be completely controlled. In the practical interpretation of measurement data, this vari ability has to be taken into account. For instance, the difference between a test result and some specitied value may be within the scope of un- avoidable random errors, in which case a real deviation from such a specified value has not been established. Similarly. comparing test results from two batches of material will not indicate a fundamental quality dit- ference if (he difference between them can be allribuled to the inherent vatiation in the measurement procedure 0.3 Many different tactors (apart trom variations between supposedly identical specimens) may contribute to the variability of results trom a measurement method, including 2) ie operator: bj the equipment used; ) the calibration of the equipment; 4d) the environment (temperature, humidity, air pollution, ete.): fe} the time elapsed between measurements, Ihe vanablity hetween measurements pertormed by ditterent operators andjor with different equipment will usually he greater than the vanabilty benvenn measurements carried out within a short interval of time ny a single operator using the same equipment. 0.4 Tho gonoral torm for variability between ropeated measurement: procicion. Two conditions of pracicion, tormad ropoatatilty and roprodue ibility conditions, have been found necessary and, for many practical cases, useful for describing the variability of a measurement method, Un- der repeatability conditions, factors 2] to e) listed above are consideredIso constants and do not contribute to the variability, while undar reproduc ibility conditions they vary and do contribute to the variahilty of the test regults, Thus repeatability and reproducibilty are the tw extremes of precision, the first descriting the minimum and the sacand the maximum Variability in results. Other intermediate conditions between these two extrame conditions of precision are alsa conceivable. when one or more of factors a) to 2) are allowed 10 vary, and are used in certain specified circumstances. Precision is normally expressed in terms of standard devi- ations. 0.5. The “trueness” of a measurement method is of interest when it is possipie ta concawa of a true value for the property being measured. Al- ‘houigh, for some measurement methods, the true value canrwo! be known exactly, it may be possible 10 have an acceplud reference value for the property being measured; for exarnple, if suitable reference materials are available. or if the accepted reference value can be established by refer- ence lo anolher measurement method or by preparetion of @ known sample, The tueness of the measurement method can be investigated by comparing the accepted reference value with the level of the results given by the measurement method. Trucnose io normally expressed in terms of bias. Bies can ariso, for example, in chemical analysis i the measuremont mothod fails to extract all of an element, or if the presence ‘of one oloment interferes with the determination of another. 0.6 The general term accuracy is used in ISO 6726 to rotor to both ‘yueness and precision ‘The term accuracy was at one time used to cover only the one component now named trueness, but it became cles that (o many persons it should Imply the total cisplagement of @ result frum 2 reference value, due to random as welll as syslematic effects, The term bias has been in use for statistical matters for a vary long time, but because it caused certain philasophiral objections among members ‘of some protessions (such as merical and legal practitioners), the positive aspect has been emphasized by the invention of the term trueness. viINTERNATIONAL STANDARD © ISO ISO 5725-1:1994(E) Accuracy (trueness and precision) of measurement methods and results — Part 1: General principles and definitions 1 Scope 1.1. The purpose of ISO 6725 is as follows: @) to outline the general principles to be understood when assessing accuracy (trueness and precision) of measurement methods and results, and in ap- plications, and to establish practical estimations ot the vanous measures by experiment {iSO 8725-1) b) to provide a basic mothod for estimating the two extreme measures ot the precision of measure ment methods by expenment (ISO 5/757}: ©). to provide @ procedure for obtaining intermediate measures of precision, giving the circumstances in which they apply and methods for estimating them (ISO b/25x3} 4d) to provide basic methods for the determination of the trueness of a measurement method (SO 8725-4); @) to provide some alternatives to the basic meth ods, given in |SO 5725-2 and ISO 5726-4, for de- termining the precision and trueness of measurement methods for use under certain cit- cumstances SU 5/25-5|; f) to present some practical applications of these measures of trueness and precision (ISO 5725-6), 1.2. This part of ISO $725 is concemed exclusively with measurement methods which yield measure ments on a continuous scale and give a single value 18 the test result, elthough this single value mey be the outcome of a calculation from a set of observe- tions. It defines values which describe, in quantitative terms, the ability of a measurement metnod to give a correct result (trusness) or to replicate a given result (precision). Thus there is ai implication thet exactly the same thing is being measured, in exactly (he Some vray, and that the measurement process is ur der contral This part of ISC.5775 may be anplied to a very wide range of materials, including liquids, powders and solid objects, manufactured ar naturally occumng, Provided that due consideration is given 10 any heterogeneity of the material 2. Normative references ‘The following standards contain provisions which, through reference in this text, constitute provisions of this part of ISO 5725. AL the time of publication, the editions indicated wore valid. All standards are subject to revision, and parties to agreements based on this part of ISO 5725 are encouraged to investivele the Possibility of epplying the most recent editions of the otondarde indicated below. Members of ILC and |SO maintain registers of currently valid International Standards.ISO 5725-1:1994(E) ISO 3534-1:1993, Statistics — Vocabulary and sym- ols — Part 1: Probabulty and general statistical terms. ISO 8726-2:1994, Accuracy (rueness and precision) of measurement methods and results — Part 2: Basic methed for the determination of repeatability and re~ producibity of e standard measurement method. ISO 5726-3:1994, Accuracy (trueness and precision) of measurement methads and resuits — Part 3 Intermediate measures of the precision of a standard measurement method. ISO 5725-4:1994, Accurecy (trueness aie! precision) of measuserrient methods and results — Part 4. Basic ineihods for the determination of the trueness of 2 ‘standard measurement method. 3 Definitions For the purposes af ISO 5725, the following defi- tions anny Some definitions are taken from ISO 3594-1. The symbols used in ISO 5726 are given in annex A 3.1. observed value: The value of a characteristic ‘obtained as the roault of o singlo obsorvation. [so 3804-1) 3.2. test result: Tho value of a choractoristic ob tainod by carrying out a specified test mothod. NOIE1 The test mothod should specify that one or a umber of individusl observations be true, ard thelt ever ‘age or another appropriate function (ouch ae the modian or tho standard doviation) bo reported as the test result. it may aso require standard corrections 10 be applied, such as correction of gas volumes to standard tenperature and proceure. Thus a test recult can bo a rosult coloulated from sovoral absorved valuos. In the simple caso, the test result is the observed value isert [sO 3534-1] 3.3 level of the test in a precision experiment: The general average ot the test results from all Iab- oratores for one particular matenai or specimen tested. 2.4 cell in a precision experiment: The test results at a single level obtained by one laboratory. eso 3.5 accepted reference value: A value that serves, aS an agreechunon reference for comparison, and which is derived as: 2) a theoretical or established value, based on scientific principles; b) an assigned or certified value, based on exper imental work of some national or international o ganization; ©) a consensus or certified value. based on collabor- ative experimental work under the auspices of @ scientific or engineering group: ) when 0}, b) and ¢} are not available, the expec tation of the (measurable) quantity, i. the mean of specified population of measurements. [ISO 3534-1] 3.6 accuracy: The closeness of agreement between a test result and the accepted reference value. NOTE 2 The torm accuracy, when appliod to a eet of test results, involves a combination of random components and 8 common systematic error 0: bias component Uso 3534-1] 3.7 truenoss: Tho closonoas of agreement setwoon the average valuc obtained from a large scrics of toat results and an accepted reference valve. NoTes 3. The measure of trusness is usually expressed in terms ot bias 4 Truenes hos boon referred to es “acouecy of the mean”, This usage is not recommended, [ISO 9634-1] 3.8 bias: Tho difference between the expectation of the test results snd an accepted reference value. NOTE 5 Bias is the total systematic ertor as contrasted Ww random enor. Theie may be one or more systematic erior components contributing to the bias. A liger systematic difference from the accepted reference value is reflected by a larger bias value. USO 3634-11 3.9 laboratory bias: The difference between the ‘expectation of the test results from a particuler lab oratory and sn accepted reference value.e1so 3.10. blas of the measurement method: The uif- ference between the expectatior of test results ob- tained from all Iaboratories using thet method and an accepted reference value. NOTE6 One example of this in operation would ha where a method purmorting 10 measure the sulfur content of @ curpound consistently fails to extract all the sulfur, giving @ negative bias to tho moncuromont mothod. Tho bias cf the measurement method is measured hy the dis placement of the average of results trom a farge number Of diferent laboratorias all using the same method. The bias cof @ meaaurement method may be ifferent at ditterant levels. 3.11 laboratory component of bias: The difference between the laboratory bias and the bias of the meaeurement method. NOTES. 7. The laboratory component of bias is spect to a oven lacoratory and the conoitions of measurement within the laboratory, and olso it may be different et diferent levels of the test 8 The laboratory component of bias is relative to the Coverall average result, not the true or reference valuo, 3.12 precision: The closeness of agreement be- ‘tween independent test results obtained under stipu- lated conditions. NOTES 9. Precision depends only on the distribution of random terrors and does not relate to the true value or the specified value 10. The meacure of precicion ie ususlly exproesed in terme fof imerecision and computed as a standard deviation at tha test results, Loss precision is rotlected by a laroer stencara deviation. 11 “Independent test resurts” means results obtained in ‘2 manner not influenced by any previous result on the sare, (oF similar test object. Quantitative measures of precision dopend efitically on the stipulated consitions. Repeatability ‘and reproduetbilty conditions ate particular sets of extreme conditions. [ISO 3634-1] 3.13 repeatability: Precision under repeatability concitions, USO 3534-1] 3.14 repeatability conditions: Conditions where independant test results are obtained with the same ‘method on identical test items in the same laboratory ISO 5725-1:1994(E) by the same operator using the same eguipment within short intervals of time. [iso 3694-1] 3.15 repeatability standard deviation: The stan dard deviation of test results obtsined under repeat ability conditions. NoTES 12. tis 2 measure of dispersion of the distribution of test results under repeatability conditions. 13. Similarly “repeatability variance” and “repeatability co- efficient of variation” could be defined and use as meas- ures of the dispersion of test results under repectabiity conditions. [ISO 3636-1] 2.16 repeatability limit: Tho value less than or ‘equal to which the absolute difference between two test results obtsined under repeatability conditions may be expected to be with a probability of 95 %. NOTE 14 The symbol used is r [1S0 9594-1) 3.17 reproducibility: Precision under reproducibility conitions. [ISO 3634-1] 3.18 reproducibility conditions: Concitions where test results are cbtained with the same method on identical test items in cifferent laboratories with ci ferent operators using different equipment. [SO 634-1] 3.19 reproducibility standard deviation: The stan dard deviation of test results obtained under repro- ducibility conditions. NOTES 15. itis @ measure of the cispersion of the dstiibution of test results under reproducibility conditions. 16 Similarly *roproducibilty varianco” and “roproducibilty coatficient of variation” could be defined and used 2s ‘measures of the dispersion of test resutts under reproduc iblity conditions [ISO 3634-1]ISO 5725-1:1994(E) 3.20 reproducibility limit: The value less than or ‘equol to which the absolute differonco botwoon two test results obtained under reproducibility conditions may be expected to be with a probability of 95 % NOTE 17 The symbol used is & USO 3534-1] 3.21. outlier: A member of a sot of values which is inconsiatont with the other mombere of that set. NOTE 18 1SO5726:2 speatias the statistical tests and the significance level to be used to identify outliers in ‘rueness and precision exporimenta, 3.22 collaborative assessment experiment: An interlaboratory experiment in which the performance Of each laboratory is assessed using the same stan dard measurement method on identical materiel. NOTES 18. The definitions given in 3.16 end 3.20 apply to results that vary on a continuous scale. If the test result is discrete Or raunded ot, the fepaatabilty mit and the reproducibility limit 2s defined above are each the mininiuin value equal 10 or below which the absolute difference betwaen two single test results is axpacted to lie with a probability of not less than 95 % 20 The definitions given in 3.8 to 3.11, 3.15, 3.18, 3.19 and 3.20 refer to theoretical values which in realty remain un- known, The values for reproducibility and reneatabilty stan- ard deviations and bias actually determined by experiment (as described in 1SO 67262 and ISO 6726-4) sro, in stat- istical terms, estimates of thase values, and as such are subject to errors. Consequently, for examaie, the probaailty lovels associated with the limits r and # wall not be exactly 5%. They will approxmato to 96 % whon many labors- tories heve taken part in tho precision experiment, but may ‘bo considerably different trom 96 % when fower than 30 laboratories have participated. This is unavoidable but does ‘ot seriously dotract from thoir practical utility a¢ they are primarily designed to sorve 2s tools for judging whether the ‘itterence between results could be asciived to random Uncertainties inherent in the measurement method or not. Difforeneoe larger than the repoatabilty limit r or the repro- ductility limit aro suspect. 21 The symbols rand 2 are alrendy in general use for ‘other purposas; in ISO 3634-1 r is recommended for the Correlation coeticient and B lor W) tor the range ot a single series of observations. However, there should be no con fusion if the full wordings repeatability limit r end reproduc ibiity limit R are used whenaver there is a possibilty of misunderstanding, partculary when they are quoted in standards, 4 Practical implications of the definitions for accuracy experiments 1 Standard measurement method 4.1.4 In order that the measurements are mode in the same way, the measurement method shall have been standardized. All measurements shall be cartiod out according to that standard method. This means that thore has to be a written document that lays down in full detail how the measurement shall be carried out, preferably including a description as to how the measurement specimen should be obtained and prepared 4.1.2 The existence of a documented measurement method implies the existence of an organization ro ‘sponsible for tho establishment of the mossurement method under study. NOTE 22 The stendord measurement method is cis- cussed nove (uly in 62 4.2 Accuracy experiment 4.2.1 The accuracy (trueness end precision) meas- ures should be determined from a series of test re oulto roported by tho participating laboratories, organized under a panel of experts established spe- ifically for that purpose. Such an interlaboratory experiment is called an “act curacy experiment”. The accuracy experiment may also be called 2 “precision” or “trueness exper- iment” according to its limited purpose. If the purpose. is to determine trueness, then a procision oxperiment shall oithor have boon completed proviously or shall occur simultaneously The estimates of accuracy derved trom such an ex- periment should always be quoted as being valid only for tests cattied out according to the standaid measurement method, 4.22 An accuracy experiment can often be consic- ered to be a practical test of the adequacy of the standard measurement method, Une ot the main purposes of standarrization \s to eliminate ehiterences between users (laboratories) as far as possible, and the data provided by an accuracy exneriment will re- veal how effectively this purpose has been achieved Pronounced differences in the withirdaboratory vari ances (see ciause 7) or between the laboratory means may indicale that the stendard measuremente1s0 method is not yot sutticiently detailed and can poss- ibly be improved. If so, this should be reported to the standardizing body with @ request for further investi- ation. 4.3. Identical test items 4.3.1. In an accuracy experiment, samples of a spe- cifie material or specimens of a specific product are ent from a central point to a number af laharatories, in different places, ditterent countnes, or even in dit- ferent continents Ihe definition of repeatability con- citions (3.14) stating that the measurements in these laboratories shall be performed on Identical test items refers to the moment when these measurernents are actually carried out. To achieve this, iwo different conditions have to be satisfied. 6). the samples have to be identical whon diepatched to the laboratorics; 'b) thoy have to remain identical during transport and during the aifferant time intervals that may elapse before the measurements are actually performed, In organizing accuracy oxporiments, both conditions shall be carctully observed. NOTE 23 fully in 6.4 The selection of material Is discussed more 4.4 Short intervals of time AAA According to the defiition of repeatability conditions (3.14), measurements for the determ- nation of repeatability have to be made under con- stant operating concitions; ie during the time covered ny The measurements, factors such 2s those listed in 0.3 should be constant. In particular, the equipment should not be recalibrated between the measurements unless Unis is an essential pelt of every single measurement. In practice, tests under repeatability conditions shoukl be conducted in as short a time as possible in order to minimize changes in those factors, such as environmental, which cannot always ke guaranteed constant. 442 There is obso a second consideration which mey effect the interval elapsing between measure- ments, and that is that the test results are assumed to be independent. If itis foarod that provioue reculte may influence subsequent test results (and 90 reduce tho estimate of repeatability variance), it may be nocoscary to provide separate epocimens coded in euch @ way that an operator will not know which ore supposedly identical. Instructions would be given 2s 10 the order in which those specimens are to be 1SO 6725-1:1994(E) measured, and presumably that order will be ran- domized s0 that all the “identical” items are not measured together. This might mean that the time interval between repested measurements may appear 10 defeat the object of a short intorval of time unless. ‘the mosouromonte are of cuch 9 naturo that tho whole series of measurements could all be completed within 2 short interval of time. Common sense must prevail 45. Participating laboratori 45.1. A basic assumption underying this part of ($0 5725 is that, for a standard measurement method, repeatability will be, at least approximately, the same for all laboratories applying the stemlaid procedure, 0 that It is permissible to wstablish one common ‘average repeatability standard deviation which will be applicable to any laboratory. However, any laboratory can, by carrying out 3 series of measurements under repeatability conditions, arrive at an estimate of its ov repeatability standard dev: ation for the measurement method and check it againot tho common standard value. Such 9 pro- cedure is dealt with in ISO 6726-6. 45.2. The quentities defined in 2.8 to 9.20 in theory epply to all laboratories which are likely to perform the jeasuromont method. In practice, they are deter- minod from a sample of this population of labore. tories. Further details of the selection of this sample 2r@ given in 6.3. Provided the instructions given there regarding the number of laboretories to be included and the number of measurements that they carry out are fcllowed, then the resulting estimates of truenass and precision should suttice. It, however, at sama fu- ture date it should become evident that the labora- tories participating were not, or are no longer, truly representative of all those using the standard measurement method, then the measurement shall be repeated, 4.6 Observation conditions 46.1 The factors which contribute to the variability of the observed values obtained within 2 laboratory are listed in 0.9. They may be given as time, operator and equipment when observations at different times include the effects due to the change of environ- mental conditions and the rocaiibration of oquioment between observations. Under repeatability conditions, observations are carried out with all these factors constant, and under reproducibility conditions obcor- Vations are carried out ot different laboratories; Le. not only with all the other factors varying but also with additional etfects due to the difference between lab-ISO 5725-1:1994(E) oratories in management and maintenance of the leboratory, stability checking of the observations, etc 4.6.2 It may be useful on occasion to consider intermediate pracision conditions, in which observa- tlons are camed out in the same lanoratory but one ‘or more of the factors time, operator of equipment are allowed 10 vary. In establishing the precision of a measurement method. it is very important 10 define the anpropriate observation conditions, i.e. whether the above thige factors should be constant or not Furthermore, the size of the variability ansing trom a factor will depend on the measurement method. For example, in chemical analysis, the factors “operator” and “time” may dominate: likewise with microanaly- sis the factors “equipment” and “environment”. and with physical testing “equipment” and “calibration” may dominate, 5 Statistical model 5.1 Basic model For estimating the accuracy (trueness and precision) of ameasurement method, itis useful to assume that every test result, y, is the sum of three components: 4Bie Oy y whore, for the particular material tested, im is the general mean (expectation): Bis the laboratory component of bias under re- peatabilty conditions, € is the random error occurting in every ‘moaouromont undor repeatability conditions. 5.1.1 General mean, » 6.1.1.1 The general mean m is the level of the test; spocimene of different purities of a chemical, or dif- ferent materials (e.g. different types of steel), will correspond to different levels. In many technical siti ations the level of the test is exclusively defined by the measurement mathod, and tha notion of an inde pendent true value does not apply. However. in some situations the concept of a true value y of the test Property may hold good, such as the true concen- ‘ation of 2 solution that is being ttrated. The level m is not necessarily equal to the tue value pt. 5.1.1.2 When examining the difference between esl results obtained by the seme measurement method, the bias of the measurement method will have no influence and can be ignored. However, 180 when camparing test results with a valie specified in a contract or a Standard where the contract or speck fication refers to the tua value (x) and not to the “level of the test” (ml. or when comparing results, produced using different measurement methods, the bias of the measurement method will have to be teken into account. If @ Wue value exists and 2 satis- factory reference material is available, the bias of the measurement method should be determined as shown in ISO 5725-4. 5.1.2 Terms 5.1.2.1 This term is considered to be constant du ing any series of tests performed under repeatability conditions, but to differ in value for tests carried out under other conditions. When test results are always ‘compared botwoon the camo two laboratorios, it ic necessary for them to determine their relative bias, either from their individual bias values as determined during an accuracy experiment, or by carrying out a private trial between themselves. However, in order to make general statements regarding differences between two unspecified laboratories, or when make ing comparisons batwean two laboratories that have hot determined their own bias, then a general distr bution of laboratory components of bias must be ‘considered, This was the reasoning behind the con cept of reproducibility. The procedures awen in ISO 5725-2 were develoved essuming thet the distr bution of laboratory components of bias is approx mately normal, but in practice they work for most distributions provided that they are unimodal 5.12.2 The variance of B is called the between- laboratory variance and is expressed es var (8) = 22) whore @% includes the betwesnoperator and between-equipment variabilitios. In the basic precision experiment described in ISO 5725-2, these components are not separated. Methods are aiven in ISO 5725-3 for measuring the size of some of the random components of B. 8.1.2.3 In general, B can be considered as the sum of both random and systematic components. No at- tempt is made to give here an exhaustive list of the factors that contibute to B, but they include different climatic conditions, variations of equipment within the Mmenufacturer's tolerences, and even differences in) the techniques in which operators are trained in dit- ferent places.©1s0 5.1.3 Error torm ¢ 5.1.3.1 This term represents a random error occur- ring in every test result and the procedures olen throughout this part af ISO 5725 were developed as- suming that the distribution of this error variable was approximately normal, but in practice they work for most distributions provided thet they are unimodal 8.1.3.2. Within a single laboratory, its variance under repeatability conditions is called the within-laboratory variance and is expressed as: vat (¢) = oy +8) 5.1.3.3. It may ba expected that ajy will have differ- ent values in different laboratories due to differences such as in the skills of the operators. but in this part of 1SO 5726 it is assumed that for # properly starr dardized measurement method such differences be- tween laboretories should be small and that tis Justfable wo establish a common value of within laboratory variance for all the laboratories using the measurement method. This common value, which is estimated by the arithmetic mesn of the within laboratory variances, is callod tho ropeatabilty vari ance and is designated by: of Var (e) - oy 2) This arithmetic moan ie taken over sil those labors tories taking port in the accuracy experiment which romain after outliers have been excluded 5.2. Relationship between the basic model and the precision 5.2.1 When the hasie model in 5.1 is adopted, tha repeatability variance is measured directly as the var- ance af the error term e, but the reproducibility vari= ance dapends on the sum of the repeatability variance and the between-boratory variance mentioned in 81.2.2, 5.2.2 Two quantities are required as measures of precision, the repeatability standard deviation «= var +8 ‘and the reproducibility standard deviation Votre © 6.3 Alternative modols Extensions to the basic model are used when apnro- priate and are described in the relevant parts of ISO 5795. 6 Experimental design considerations when estimating accuracy 6.1 Planning of an accuracy experiment 6.1.1 The actual planning of an expenment to est mate the precision andjor truaness of a standard measurement method should be the task of @ panel of experts familiar with the measurement method and Its application, At least one member of the panel should have experience in the statistical design and analysis of experiments. 6.1.2 Tho following questions should be considered when planning the experiment. a) Is a satisfactory standard av: measurement method? ble for the b) How many laboratories should be recruited to co- operate in the experiment? ©) How should tho laboratories be recruited, and what requirements should they satisfy? 1) What is the range of levels encountered in prac- tice? ¢) How many levels should be used in the exper iment? fi What aro suitable materiale to roprosont thoco levels and how should they be prepared? @) What number of replicates should be specified? h) What time-frame should be specified for the completion of all the measurements? i) Is the basic model of 5.1 appropriate, or should @ Modified one be considered? j] Are any special precautions needed to ensure that identical materials are measured in the same stato mall laboratories? ‘These questions are considered in 6.2 to 6.4.ISO 5725-1:1994(E) 6.2 Standard measurement method As stated in 4.1, the measurement method under in- vestigation shall be one that has been stendarvized, Such a method has to be robust, i.e. small variations: in the procedure shoukl not produce unexpectedly Jerge changes in the results. If this might happen, Uete shall be adequate precautions or warnings. It is alsu desireble thet in the process of developing a stenderd measurement method every effort has been made to remove or reduce bias. Similar experimental procedures may be used 10 measure the trueness and orecision of both estab lished measurement methods and recently standard zed Measurement methods. In the latter vase, the results obtained shoukl be regarded as preliminary estimates, because the trueness and precision could change as laboratories gain experience. The document setting out the measurement method shall be unambiguous and complete. All essential operations concerning the environment of the pro- cedure, the reagents and apparatus, preliminary checking of equipment, and te preparation of the test specimen should be included in the measure- ment method, possibly by references to other written procedures that are available to the operators, The manner of calculating end expressing the test resutt should be precisely specified, including the number of significant figures to be reported. 6.3 Selection of laboratories for the accuracy experiment 6.3.1 Choice of laboratories From a stetistical point of view, those laboratories participating in any experiment to estimate accuracy should have been chosen at random from all the lab- oratones using the measurement method. Volunteers might not represent a realistic cross-section. How- fever, other practical considerations, such as a re quirement that the participating laboratories be distributed over different continents or climatic re- gions, may affect the pattern of representation ‘The participating laboratories should nat consist ex: clusively of those that have gained special experience dunng the process of standardizing the method Neither should they consist of specialized “raference” laboratories in order to demonstrate the accuracy to which the method can perform in export hands. The number of laboratories to be recruited to partic pate in @ cooperative interlaboratory experiment and the number of test results required from each labore tory at each level of the test are interdependent. A guide to deciding how many there should be is given in 63.2 0 63.4. 6.3.2 Number of laboratories required for an estimate of precision 6.3.2.1 The vetious quentities represented by the symbol a in equations (2) to (6) of clause 5 are true standard deviations whose values are not known, an object of @ precision experiment being to estimate them. When an estimate (s) of a true standard devi ation (a) is to be made, conclusions can be drawn as to the range about « within which the estimate (s) can bo oxpocted to lia, This is a well understood statistical problem which is solved by the use of the chi-squared distribution and the number of results fromm which the estimate of s was based. One formula frequently used is: Plas <+Al=P a Often A is quoted in percentage terms, enabling & statement (0 be made tial tie estinled standard Ueviations (s) can Le exuected to be within A either side of the true standard deviation (a) with a certain probability P. 6.3.2.2 Fora single level of the test, the uncertainty in the repeatability standard devietion will depend on the number of laboratories (p) and the number of test rogults within oach laboratory (x). For the reprodue iblity standard deviation, the procedure is more com plicated ss this is determined from two standard dovistions [ees equation (6)]. An extra factor 7 is needed. representing the ratio of the reproducibility standard deviation to the repeatability standard devi- ation, that is: yeni, ® 6.3.23 Assuming 4 probability level P of 95 %, ep- proximate equations for the velues of A heve been prepared end are givon below. The cavations are in tended for the purposes of planning how many lab oratories to recruit and deciding how many test results are to be required from cach laboratory at cach level of the test. These equotions do not give confi dence limits and so they should nat be used during the analysis stage to calculate confidence limits. The equations are a5 follows, For repeatability A-4,-1.96,/=—4 4 Ve i oIso For reproducibility Am Ag = 1.96, 5 27 p— 1p 0) NOTE 24 A semple vorionce which has » degross of freedom and expectation a may be assumed tn have, ane proximately, a normal clstrinution with variance 2a*fr Equations |9) end (10] were derived by making this os- ‘sumption about the variances involved in the estimation of 12, and 1 The adequacy of the approximation was checkers hy an exact calculation, 6.3.2.4 The value of y is not known, but often pre- liminary estimates are available of the within- laboratory standard — deviations andthe. betweentaboratory stendard devietions obtained dur ing the process of standardizing tho moacurement method. Exact values of tho uncertainty poreantagos for repestability and reproducibility standard devi- ations with different numbers of laboratories (p) and difforont numbers of rosults por laboratory (n) are given in table and are also plotted in chart torm in annex B. 6.3.3 Number of laboratories required for the estimate of bias 6.3.3.1 The bias of the measurement methou, 6, may be estimated from: b=F-u ay whore Fis the grand mean of all the test results ob- tained by all the laboratories at @ particular level of the experiment: Is the acceptad reference value pul tn? -1)F + (@-e-) ISO 5725-1:1994(E) Tho uncertainty of this estimate can be expressed by the equation: P Ld —Aag<3< 0+ Aap] = 0.95 a2 Which shows that the estimate will be within Acg uf the true measurement method bias with a probability of 0.95. In terms of the factor y [see euuetion (BY) nt) +1 Fon Values of A are given in table 2. A= 1,56, (13) 6.3.3.2 The laboratory bias, a, at the time of the ex periment may be estimated from: aay F is the arithmetic mean of all the results ob- tained by the laboratory at a particular level of the experiment; is the accepted reference velue. The uncertainty of thie estimate can be expressed by the equation: P[d—Ayo,<4< A+ Ays,] = 0.95 5) which shows that the estimate will be within Aya, of ‘the true laboratory bias with a probability of 0,95. Here: the within-laboratory uncertainty is 1,96 ve Values of ayy ate given in tables. Aw -(18) Table 1 — Values showing the uncertainty of estimates of the repeatability and reproducibility standard deviations No. of fe laboratories “ 2 15 2 na | n—4 | a2 | n-3 | a4 | w-2 | 2-3 | n-4 | w-2 | 2-3 5 oa | 036 | 04s | 027 | 022 | 61 | ose | 087 | 06s | 067 0 oa | oar | 02 | oz | 026 | oz | om | ose | oss | 04s | o4o | vas 15 036 | 028 | o21 | 026 | o21 | ots | 033 | 031 | 030 | 036 | 036 | 36 20 ost | 022 | 018 | 022 | o18 | ore | 028 | o27 | 025 | 031 | 031 | 031 28 oze | 020 | or6 | 020 | ors | o14 | 026 | 024 | 023 | 028 | oz8 | o27 20 025 | ote | 018 | ota | 075 | 013 | 023 | 022 | o21 | 025 | os | o26 35 073 | a17 | ara | 017 | ora | 012 | 021 | 020 | a19 | 023 | 023 | 023 40 ozz | a16 | o13 | 016 | 013 | o11 | 020 | o19 | ore | 022 | o22 | ureISO 5725-1:1994(E) Table 2 — Values of 4, the uncertainty of an estimate of the bias of the measurement method ava Value of A laboratories. yet p ® n=3 6 ae | 076 | o72 | 069 10 62 | osa | os | 040 8 as | oa | om | 040 20 oaa | 098 | 030 | 0,35 5 ogg | os | o32 | o37 30 oae | oz: | o20 | o28 35 03a | 029 | 027 | 0,26 40 oav_| oz | oz | uz Table 3 — Values of 4,,. the uncertainty of an estimate of the withintaboratory bias No. tat outs awe ot, A om 0 et i on 20 one Fa oes FA oe & oss “ ost 6.3.4 Implications in the choice of laboratories The choice of the number of laboratories will be 2 compromise between availability of resources and 6 desire to roduce tho uncortainty of tho cotimatos to a satisfactory level. From figures B.1 and B2 annex B it can be seen that estimates of the repeat- ability and reproducibility standard devietions could differ substantially from their true values if only @ ‘small number (p = 5) of laboratories take part in a precision experiment, and thet increasing the number of the laboratories by 7 or 3 yields only small re- ductions in the uncertainties of the estimates when pis greater than 20. It is common to choose a val of p between 8 and 15. When a is larger than o, (ie yis lager than 2), a5 is often the case, litle Is {0 be gained by obtaining more than n= 2 test results per laboratory per level. 10 ©1so 6.4 Selection of materials to be used for an accuracy experiment 6.4.1 The materials to be used in an experiment to determine the accuracy of a measurement method should represent fully those to which the measure- ment method is expected to be applied in normal use ‘As 8 general rule, five differant materials will usually provide a eufficiantly wide range of levels to sllow the accuracy to be established adequately. A smaller number might be appropriate in the first investigation of @ recently developed measurement method when itis suspected that modifications to the method may be necessary, followed by further accuracy exper iments. 6.4.2 When the measurements have to be per formed on discrote objects that are not altered by ‘measuring, they could, in principle at least, be carried out using the same set of objects in different labora tories. This, however, would nocessitate circulating the same set of objects around many laboratories of- ten situated far apart, in different countries or conti- nents, with a considerable risk of loss or damage ‘uring transport. It difterent items are to be used in ‘iiferant laboratones, then they shall be selected in such a way as to ensure that they can be presumed to be Identical for practical purposes. 6.4.3 In oclecting the material to represent the dif ferent levels, it should be considered whether the material chould be spocially homogonizad before pre paring the samples for cispatch, or whether the effect Of the heterogeneity of the material should be cluded in the accuracy values. 6.44 When meesurerients have to be performed con solid materials that cannot be huirivgenized (such 8 metals, rubber or textile fabrics) and when the measurements cannot be repeated on the same test piace, inhomogeneity in the test material will form an essential component of the precision of the measurement and the idea of identical material no longer holds good. Precision experiments cen still be cartied out, but the values of precision may only be valid for the particular material used and should be quoted as such. A more universal use of tho procision a8 determined will be acceptable only if it can bo demonstrated that the values do not differ signit- icantly between materials produced at different times or by diferent producers. Ths would require a more elaborate experiment than has heen considered in 1SO 5725.© (sO 6.4.5 In general, whore destructive testing is in- volved, the contribution to the variability in the test results arising from differences between the spec mens on which the measurements are performed shall either be negligible compared to the variability of the measurement method itself, or else shell form an inherent part of the vatiability of the measurement method, and thus be truly a component of precision. 8.4.8 When the materials under meesurement might henge with time, the overall time-scale of the ex periment should be chosen to take this into account. It might be eppropriate in some casce to epecity the times et which the samples are to be messured. 6.4.7 In all the above, reference is made to measur- ing in difforont laboratories, with the implication of ‘transportation of the test specimens to the laboratory, but some test specimens are not transportable, such as an oil storage tank. In such cases, measuring hy Gifferent lsboretories means that aifferent operators are sent with their equinment to the test site. In other ceases, the quantity being measured may be transitory or variable, such 2s water flow in a river, when care shall be taken that the different measurements ere made under, as nest 2s possible, tie seme conditions ‘The guiding princivly must always be that the objec tive is to Uetermine the ability to repeat the same measurement 648 The establishment of precision values for a ‘measurement method presupposes that the precision either is independent of tho material boing tostod, or depends on the material in a predictable manner. With somo moaeurement methods it is possible to quote the precision only in relation to one or more definable classes of tast material. Such data will be only a rough ‘guide to tho procision in other applications. More of- ten it is found thet the precision is closely related to the level of the tast, anc catermination of the nre- cision then inclidlas the establishment of a relation. ship between precision and level. Therefore, when publishing precision values for a standard measure- ment method, it is recommended that the material used in the precision experiment should be clearly specified alona wilh the range of materials to which the values can be expected to apply 6.4.9 For the assessment of trueness, at least one of the materials used: should have an accepted refer- cence value. If it is Ikely thet trueness varies with level, materials with accepted reference values will be needed at several levels. ISO 6726-1:1994(E) 7 Utilization of accuracy data 7.1. Publication of trueness and precision values 7.4.1. When the aim of a precision experiment is to ‘obtain ostimates of the repostability and reproducibil- ity standard dovistions undor tho conditions defined in 3.14 and 3.18, then the basic model of 5.1 shall be used. ISO 5725-2 then provides an appropriate method of estimating these standard deviations, or an alternative may be found in ISO 5725-5 When the aim is to obtain estimates of intermediate measures fof precision, than the alternative model and the methods given in ISO 5725-3 shall be used. 7.1.2 Whenever the bias of the measurement method has been determined, il should be published with @ statement regarding the reference against which that kiss was determined. Where the bias varies with the level of the test, publication should bo in the form of a table giving the level, the bias a3 de termined, and the reference used in that determ:- ration, 7.1.3. When an interlaboratory experiment has been performed for estimating trusness or procision, oach participating laboratory should be intormed of its lab- oratory component of bias relative to the general mean as determined from the experiment. Ihis infor- mation could be of value in the future it similar ex- penments are performed, but should not be used for caliration purposes, 7.1.4 The repeatability and reproducibly standard devietions for any standard measurement method shall be determined as laid down in parts 2 lo 4 of 180 5725, and should be published as part of the standard measurement method under a section en- titled precision. This section may also show the re- peatability and reproducibility limits (r and R). When precision does not vary vrith level, single average fig- ures can be given in each cose. Where precision varies with the level of the test, publicetion should be in the form of a table, such as tablo4, and may also be oxpressed a5 9 mathematical relationship. Inter mediate measures of precision should be presented ina similar form, nISO 5725-1:1994(E) Table 4 — Example of method of reporting standard deviations Repectabity | Reprodvebitty Range orlevel | deviation | devotion é & Fann From to From to 7.1.5 The definitions of repeatability and reproduc- ibility conditions (3.14 and 3.18) shall be gwen in the Precision clause. When intermediate measures of precision are given, care should be taken to state which of the factors (time, operators, equisment) have been allowed 1o vary, When the repeatability and reproducibility limits are givens some statement should be added linking them to the difference be- tween two test results and the 96 % provabilty level. Suggested wordings are 9 follows. ‘The difference between two test results found on identical test material by one operator using the same apparatus within the shortest feasible time interval will exceed the repeatability limit (r) on average not more than once in 20 cases in the formal and correct oneration of the method Test resulls on identical test inetetial reported by two lboratories wil differ by more than Uie re producibility limit (R) on average not more than’ ‘once in 20 cases in the normal and correct oper ation of the method. Ensure that the definition of a test result is clear, either by quoting the clause numbers of the measurement method standard that have to be to lowed to obtain the test result or by oll: mean. 7.1.6 In general, a briet mention of the accuracy ex- periment should be added et the end of this precision section. Suggested wording is as follows. ‘The accuracy data were determined from an ex- periment organzed and analysed in accordance with ISO 5725- (part) in [year) involving (p) lab- oratories ani (q) levels. Data trom | | laboratones contained oufliers. The outliers ware not included in the calculation of the repeatability standard de- vation and the reproducibility standard deviation. 2 ©1so A description of the materials used in the accuracy experiment should be added, especially when the trusness or precision depend on the materials, 7.2 Practical applications of trueness and precision values Practical applications of trueness and precision values are covered in detail in |SO 5725-6, Some examples are as follows. 7.2.1. Checking the accoptability of test results A product specification could require repeated measurements to be obtained under repeatability conditions. A repeatability standard devistion may be used in these circumstances to check the arcentapil- ity of the test results and to decide what action should bbe taken if they are not acceptable. When both a supplier and @ purchaser measure the same materia and their results differ, repeatability and reproducibility standard deviations may be used (0 decide if the dif- ference is of a size that is to be expected with the meesurement method. 7.22 Stability of test results withis a laboratory By carrying out regular measurements on raterence materials, @ laboratory can check the stanility ot its results and produce evidence ta demonstrate its competence, with respect to both the bias and the repeatability of its testing. 7.2.2 Assessing the porformaneo of a laboratory Laboratory accrecitation schemes are becoming in- creasingly widespread. Knowledge of the truenoss and precision ot a measurement method allows the bias and reneatahilty of a candidate laboratary to be assessed, either using reterence materials or an interlaboratory experiment 7.2.4 Comparing altemative measurement methods Two measurement methods may be available for measuring the eame property, onc being simpler end less expensive than the other but less generally ap: plicable. Trusness and precision values may bo usod to justly the use of the less expensive method for some restricted range of materials,2 1s0 By Buy Boy etc. ee Cae Coie Coe cD, CR, Annex A (normative) ISO 5725-1:1994(E) Symbols and abbreviations used in ISO 5725 Intercept in the relationship smatom Factor used to calculate the uncer tainty of an estimate Slope in the relationship soatbm Component in a test result repre- senting tne deviation of a laboratory from the general averaga (iaboratory component of bias} Component of B representing all factors that do not change in inter mediate precision conditions Components of B representing fac tors that vary in intormodiate pro- cision conditions Interoent in the relationship Igs=c+digm Test statistics Critical values for statistical tests Critical difference for probability P Critical range for probability ? Slope in the relationship los=etdlom Component in a test result repre senting the random error occurring in every test result Critical range tactor p-quantilo of the F distribution with v1 and vz dogroce of froodom Grubbs’ test statistic Mandel's betweenaboratory con- sistency test statistic Leu. AM uct Mandel's statistic withindaboratory consistency test Lower control mit (ether action limit ar warning limit) Gonoral mean of the test property; level Number of factors considered in intermediate precision conditions Number of iterations Number of test results obtained in one labora tory at one level (ie. per cell) Number of laboratories participating in the inter- laboratory experiment Probability Number of levels of the test property in the interlaboratory experiment Repeatability limit Reproducibility limit Reference material Estimate of a standard deviation Predicted standard deviation Total ar sum of same expression Numbor of tost objects or groups Upper control limit (either action limit or warning lint) Weighting factor used in calculating a weighted regression Range of @ sel of test results Datum used for Grubbs’ test Test result 1B1SO 6725-1:1994(E) 4 ooh 0) 4 Acithmetic mean of test results Grand mean of test results Significance level Type | error probability Ratio of the reproducibility standard deviation to the repestability standerd deviation (eg/o,) Laborotory bios Estimate of 4 Bias of the measurement method Estimate ot 6 Detectable difference between two laboratory biases or the biases of two measurement methods True value or accepted reference value of a test property Number of degrees of freedom. Hetaetadle ratio between the repeatability stan dard deviations of method B and method A True value of a standard deviation Component ina test result representing Ue variation due to time since last calibration Detectaole ratio between the square roots ot the hetweendaboratary mean squares of method B and method A poquantile of the ;2distribution with » degrees of Freedom e1sa ‘Symbols used as subscripts © so m3 4,2,3. (11, (2), (9). Calibration-difterent Equipmentdifferent Idemifiar for a particular laboratory Identifier for intermediate measures of precision; in brackete, identification of the type of intermediate situation Identifier for 2 (SO 5725-2) Identifier for a group of tests or for a factor (ISO 5725-3) perticular level Identifier for a particular test result in a laboratory i at level j Between-laboratory (interlaboratory) Identifier for detectable bes Between-test-samole Operaturuifferent Probebility Repeatability Reproducibility Time-different Within-taboratory (intralaboratory) For test results, numbering in the order of obtaining them For test results, numbering in the order of ineroaeing magnitudoe1so ISO 5725-1:1994(E) Annex B (normative) Charts of uncertainties for precision measures Uncertainty ns (6 oC umner of laboratories Figure B.1 — The amount by which , ean bo oxpoctod to differ from the true value within a probability level of 95 % 5ISO 5725-1:1994(E) ©1S0 409) 0] Uncet tainty in #2 6! é 6 50 re 30 Numer of laboratories Figus B.2 — The amount by which s, can be expected to differ from the true value with level of 95% a probability 161S0 5725-1:1994(E) e1so Annex C (informative) ibliography [1] 1S0.2694-2:1908, Statistics — Vocabulary and [a] ISO.5725-6:1994, Accuracy (trueness and pre- cision) ot measurement methods and results — symbols — Part 2: Statistical quality control, [2] ISO 3830.3-1985, Statistics — Vocabulary and symhols — Part 3: Design of experiments. [3] ISO 5725-5—", Accuracy ftrueness and pre- cision) of measurement method's end results — Part 5: Alternative methods for the determi nation of the precision of @ standard moasure ment method. 1) To be published Part 6° Use in prachee of accuracy values. [5] ISO Guide 33.1989, Use of certified reference materials, [6] SO Guide 35:1989, Certification of reference materials — General and statistical principles. vISO 5725-1:1994(E) © 180 ICS 03,120.30 Descriptors: messurement, tests, test resuts, accuracy, reproductiliy, statistical aalyss, definitions, generaitias ee based on 11 pages