1、Designation: E 1488 02e1An American National StandardStandard Guide forStatistical Procedures to Use in Developing and ApplyingTest Methods1This standard is issued under the fixed designation E 1488; the number immediately following the designation indicates the year oforiginal adoption or, in the c
2、ase of revision, the year of last revision. A number in parentheses indicates the year of last reapproval. Asuperscript epsilon (e) indicates an editorial change since the last revision or reapproval.e1NOTEThe introduction was deleted editorially in March 2003.1. Scope1.1 This guide identifies stati
3、stical procedures for use indeveloping new test methods or revising or evaluating existingtest methods, or both.1.2 This guide also cites statistical procedures especiallyuseful in the application of test methods.2. Referenced Documents2.1 ASTM Standards:E 29 Practice for Using Significant Digits in
4、 Test Data toDetermine Conformance with Specifications2E 105 Practice for Probability Sampling of Materials2E 122 Practice for Choice of Sample Size to Estimate aMeasure of Quality for a Lot or Process2E 141 Practice for Acceptance of Evidence Based on theResults of Probability Sampling2E 177 Practi
5、ce for Use of the Terms Precision and Bias inASTM Test Methods2E 178 Practice for Dealing with Outlying Observations2E 456 Terminology Relating to Quality and Statistics2E 691 Practice for Conducting an Interlaboratory Study toDetermine the Precision of a Test Method2E 1169 Guide for Conducting Rugg
6、edness Tests2E 1301 Guide for Proficiency Testing by InterlaboratoryComparisons2E 1325 Terminology Relating to Design of Experiments2E 1402 Terminology Relating to Sampling22.2 ISO Standards:ISO 5725 Accuracy (Trueness and Precision) of Measure-ment Methods and Results3ISO 17025 General Requirements
7、 for the Competence ofTesting and Calibration Laboratories3ISO Guide to the Expression of Uncertainty in Measure-ment33. Terminology3.1 Definitions:3.1.1 statistical procedures, nthe organized techniquesand methods used to collect, analyze, and interpret data.3.1.1.1 DiscussionStatistical procedures
8、 include the sam-pling considerations or the experiment design for the collectionof data, or both, and the numerical and graphical approaches tosummarize and analyze the collected data.3.2 For all other formal definitions of statistical terms, seeTerminology E 456.4. Significance and Use4.1 All ASTM
9、 test methods are required to include state-ments on precision and bias.44.2 Since ASTM began to require all test methods to haveprecision and bias statements that are based on interlaboratorytest methods, there has been increased concern regarding whatstatistical experiments and procedures to use d
10、uring the devel-opment of the test methods. Although there exists a wide rangeof statistical procedures, there is a small group of generallyaccepted techniques that are very beneficial to follow. Thisdocument is designed to provide a brief overview of theseprocedures and to suggest an appropriate se
11、quence of carryingout these procedures.4.3 Statistical procedures often result in interpretations thatare not absolutes. Sometimes the information obtained may beinadequate or incomplete, which may lead to additional ques-tions and the need for further experimentation. Informationoutside the data is
12、 also important in establishing standards andin the interpretation of numerical results.1This guide is under the jurisdiction of ASTM Committee E11 on Quality andStatistics and is the direct responsibility of Subcommittee E11.20 on Test MethodEvaluation and Quality Control.Current edition approved O
13、ct. 10, 2002. Published December 2002. Originallypublished as E 1488 92. Last previous edition E 1488 96.2Annual Book of ASTM Standards, Vol 14.02.3Available from American National Standards Institute, 11 W. 42nd St., 13thFloor, New York, NY 10036.4See the Form and Style Manual for ASTM Standards th
14、at specifies, whenpossible, precision statements shall be estimated based on the results of aninterlaboratory test program.1Copyright ASTM International, 100 Barr Harbor Drive, PO Box C700, West Conshohocken, PA 19428-2959, United States.5. Summary of Guide5.1 Outlined below is a suggested sequence
15、of four phasesuseful in the development of a test method. A flowchart isprovided in Fig. 1. Such a sequence of analyses may need to bemodified in specific situations. The assistance of a qualifiedstatistician is recommended at each review phase.5.2 Design Phase:5.2.1 This phase includes the formaliz
16、ation of the scope andthe significance and use sections. It may include determiningthe purpose and describing a general approach to the testmethod but usually does not involve statistical studies.5.3 Development Phase5.3.1 Studies may be conducted to evaluate the basicperformance of the method. The
17、draft test method is preparedand sampling requirements and the test result are clearlydefined.5.3.2 A flow chart is extremely valuable to identify thesequence of operations involved in a test method, for example,the sampling steps required to obtain the test specimens,definition of the test determin
18、ation, how a test result is to becomputed, and running the tests on the specimens.5.4 Validation Phase5.4.1 The test method is examined for such concerns as itsstability, ruggedness, statistical control and the contributions tovariability. The completion of this phase should result inpreliminary est
19、imates of precision and the identification andsuggested ways to estimate potential contributors to uncer-tainty.5.4.2 Evaluation of Short Term Control of Test MethodAtest method must exhibit an ability to provide consistent resultsat least over short time periods. Preliminary studies or a pilottest
20、should be conducted to evaluate the short term stability ofthe test method. A small series of repeated tests should beconducted.5.4.3 Analysis of VariabilityStatistically designed experi-ments conducted in one or two laboratories can be used toassess the relative magnitudes of different sources or p
21、otentialcontributors to variability of the test results. Such studies canprovide estimates of intermediate measures of precision.5.4.4 Ruggedness TestA ruggedness test (see GuideE 1169) is a statistically designed experiment that helps iden-tify problems in running the test method, clarifies errors,
22、 andpoints out possible environmental conditions, which mayadversely affect the test method or point out need for tighteningrequirements. The ruggedness test can assist in locating waysof reducing variability in the test method.5.4.5 Preliminary Estimates of PrecisionFrom the variousstudies conducte
23、d in accordance with 5.4.25.4.4, preliminaryFIG. 1 Sequence of StepsE148802e12estimates of repeatability standard deviations should be devel-oped and published in this test method. Until an interlaboratorystudy is performed, these estimates generally are considered tobe provisional. Information on h
24、ow a lab should developuncertainty estimates should also be provided.5.4.6 Statistical ControlA test method must show capa-bility of performing in a consistent way over time. The use ofcontrol charts (see Manual 7)4to monitor a proposed, orexisting, test method over time is one recommended way toexa
25、mine the controllability or stability of a test method. Thisstatistical control should be demonstrated in one or twolaboratories using homogeneous material (test specimen).5.5 Evaluation Phase:5.5.1 The test method is subjected to interlaboratory studiesto provide estimates of within-laboratory repe
26、atability andbetween-laboratory reproducibility. Additional information issupplied from proficiency studies when conducted.5.5.2 Interlaboratory Study (ILS)In accordance withASTM Form and Style Manual, whenever feasible, an inter-laboratory study must be conducted. This procedure willprovide specifi
27、c estimates of variation anticipated when usingthe test method.5.5.3 Protocol for the ILS, Practice E 691 provides a guidefor developing the ILS for the test method. A first step is thewriting of an ILS Protocol, which will set out what needs to bedone before the test specimens (or test materials) a
28、re distrib-uted to the participating laboratories.5.5.4 Precision StatementsUsing the estimates of varia-tion obtained in the interlaboratory test, one may prepareprecision statements using Practices E 691 and E 177 orequivalent procedures.6. Development of Test Method Sampling and TestResult6.1 Pro
29、posed standards that are under development shouldbe treated in a formal manner following as many of thesuggested procedures as possible. Standards that are already inexistence as approved test methods or in general practicerequire periodic review that would include selected procedures.6.2 Under Deve
30、lopmentThe development stage involvestest methods that are in the preliminary stages during whichequipment may not have been fully tested, practices are notagreed upon, and operators have yet to be adequately trained.Often this stage also applies to standards that have not yet beenapproved.6.2.1 It
31、is essential that tests for statistical control, rugged-ness, and variability analyses be conducted prior to anyinterlaboratory test programs.6.2.2 After all major environmental contributors have beenidentified, controlled, and incorporated into the test method,and after adequate standardized equipm
32、ent is available, aninterlaboratory test can be conducted. The interlaboratory testprogram must be completed prior to the first 5-year review. Thecommittee should strive to have interlaboratory results as soonas possible.6.2.3 After evaluating data from ruggedness tests, variabil-ity analysis, or an
33、 interlaboratory test program, changes to thetest method may be suggested.6.2.4 If major changes are made to the test method, a repeatof the various steps is usually necessary. Precision and biasstatements should reflect the most current version of the testmethod.6.3 Existing StandardsThese standard
34、s comprise testmethods that are in common use for which standard equipmentmay exist and for which experienced operators have beentrained and are available.6.3.1 Control charting, ruggedness tests, and variabilityanalyses will be useful, especially if they have not previouslybeen conducted. Such test
35、s may provide better informationabout variation and necessary tolerances than has previouslybeen available.6.3.2 If precision estimates have not been establishedthrough an actual interlaboratory test program, then such aprogram should be initiated.7. Data and Sampling7.1 Sample Determination:7.1.1 T
36、he sampling section of a standard should indicateclearly what constitutes the primary sampling unit, how thatsampling unit is further subdivided, and how multiple testvalues are designated.7.1.2 In considering the implication of test results as theyrelate to the material, the test method should be c
37、lear as towhether the sampling method or the test is destructive ornondestructive.7.1.3 The user of the test method should be aware ofwhether the standard calls for a random sample. In somestandards, as for example in sampling from coils or rolls ofmaterial, samples may be taken only from certain po
38、rtions ofthe material.7.2 Test Result DeterminationThe procedure for deter-mining a test result must be clear and unambiguous.7.2.1 An observation leads to an observed value.7.2.2 Several observed values may lead to a test determina-tion. The observed values need not be the same type ofmeasurements
39、(for example, they may consist of three readingssuch as length, width, and mass).7.2.3 Several Test determinations may lead to a test result,as by averaging three test determinations.7.2.4 A test result is the consequence of a single executionof the entire test method.7.3 Type of DataThe kind of dat
40、a that results from theapplication of the test method determines the types of statisticalanalyses to be performed.7.3.1 Numerical versus Categorical/Attribute DataMostof the statistical procedures referred to in this standard dealwith numerical data. Control charts are available for all typesof data
41、, but all interlaboratory test procedures currently in usedepend on numerical data.7.3.2 “Normally” Distributed DataMost of the statisticalprocedures referred to in this guide consider that the unknowndistribution of the test results can be modeled by a normaldistribution.8. Preliminary Evaluation o
42、f Short Term Control8.1 A test method must be capable of providing consistentresults over short time periods. The first efforts at evaluating aE148802e13test method should include repeating the method on the sameor as close to the same materials under constant conditions overa short time period. Thi
43、s will provide some initial informationabout how close measurements can be repeated. This type ofexperiment should be repeated several times to determine howwell the test method can perform at different time periods.8.2 Since the tests may involve only a few sets of samplemeasurements, an experiment
44、al design model is the appropri-ate mode of evaluation of the results.NOTE 1We recommend that the Analysis of Means (ANOM) proce-dure be utilized to determine how well the mean level remains at the sametarget level. This also permits an easy graphical and conceptual transitionto a future control cha
45、rt (as recommended in Section 11).NOTE 2Each sample will consist of small number of repeats. Todetermine if the variability remains consistent from sample to sample anAnalysis of Ranges (ANOR) can be similarly conducted.NOTE 3New standards are being developed to provide further guid-ance.9. Analysis
46、 of Variation9.1 Important contributions to variability must be ascer-tained. These sources may involve applying the test method atdifferent laboratories, with different operators, over differentdays, with different apparatus, using different samples, and soon.9.2 A statistically designed experiment
47、 for estimating“Components of Variance” is usually conducted to identify therelative contribution to the variation due to each of the factorsunder consideration.9.3 A study of variability may be conducted in one or onlya few laboratories because of the difficulty of managing theexperiment (in contra
48、st to an ILS).9.4 A qualified statistician should be involved in organizingand working with the task group throughout the project.10. Ruggedness Testing10.1 The committee should attempt to identify all variablesthat are believed to have possible major influence on theprecision or bias of the test me
49、thod.10.1.1 The ruggedness test usually is conducted in one ortwo laboratories with each “treatment” set at two levels. Theselevels are based on the conditions specified in the test method,and the low and high levels for each treatment are derived fromthe reasonable extremes that might be encountered in use. Thistest often should be one of the first procedures carried out andmay need to be repeated when significant changes in the testmethod are made.10.2 The test should include each such variable at levels asreasonably extreme as possible and likely to be encou
