# ISO 2602-1980

International Standard 2602 INTERNATIONAL ORGANIZATION FOR STANDARDIZATION.MElKt(AYHAPOAHAR OPTAHM3ALU4R n0 CTAH,QAPTl43A~bIWORGANISATION INTERNATIONALE DE NORMALISATION Statistical interpretation of test results - Estimation of the mean - Confidence interval Interpretation statistique de rksultats d’essais - Estimation de Ia moyenne - Intervalle de confiance Second edition - 1980-02-15 UDC 519.25 : 620.113 Ref. No. ISO 2602-1980 (E) Descriptors : statistical analysis, statistical tests, estimation, test results, mean, variance (statistics). Price based on 5 pages SIS Multi User Licence: ABB TECHNOLOGY LTD 1/6/2009Foreword ISO (the international Organization for Standardization) is a worldwide federation of national Standards institutes (ISO member bodies). The work of developing Inter- national Standards is carried out through ISO technical committees. Every member body interested in a subject for which a technical committee has been set up has the right to be represented on that committee. International organizations, governmental and non-governmental, in liaison with ISO, also take part in the work. Draft International Standards adopted by the technical committees are circulated to the member bodies for approval before their acceptance as International Standards by the ISO Council. International Standard ISO 2602 was developed by Technical Committee ISO/TC 69, Applica tions of s ta tis ticai me thods. This second edition was submitted directly to the ISO Council, in accordance with clause 5.10.1 of part 1 of the Directives for the technical work of ISO. lt cancels and replaces the first edition (i.e. ISO 2602-1973), which has been approved by the member bodies of the following countries : Australia Austria Belgium Czechoslovakia Egypt, Arab Rep. of France Germany, F. R. Hungary India Ireland Israel I taly Japan Netherlands New Zealand Poland Portugal Romania South Africa, Rep. of Sweden Switzerland Thailand United Kingdom USSR No member body had expressed disapproval of the document. 0 International Organkation for Standardkation, 1980 Printed in Switzerland SIS Multi User Licence: ABB TECHNOLOGY LTD 1/6/2009INTERNATIONAL STANDARD ISO 2602-1980 (EI Statistical interpretation of test results - Estimation of the mean - Confidence interval Second edition 0 Introduction The scope of this International Standard is limited to a special question. lt concerns only the estimation of the mean of a nor- mal population on the basis of a series of tests applied to a ran- dom Sample of individuals drawn from this population, and deals only with the case where the variance of the population is unknown. lt is not concerned with the calculation of an interval containing, with a fixed probability, at least a given fraction of the population (statistical tolerante limits). lt is recalled that ISO 2854 relates to the following collection of Problems (including the Problem treated in this International Standard) : - estimation of a mean and of the differente between two means (the variances being either known or unknown); - comparison of a mean with a given value and of two means with one another (the variances being either known or unknown, but equal); - estimation of a variance and of the ratio of two variances; - comparison of a variance with a given value and of two variances with one another. The test methods generally provide for several determinations which are carried out : - on the same item (where the test is not destructive); - on distinct portions of a very homogeneous product (a liquid, for example); - on distinct items sampled from an aggregate with a cer- tain amount of variability. In the first two cases, the deviations between the results ob- tained depend only upon the repeatability of the method. In the third case, they depend also on the variability of the product itself. The statistical treatment of the results allows the calculation of an interval which contains, with a given probability, the mean of the population of results that would be obtained from a very large number of determinations, carried out under the same conditions. In the case of items with a variability, this Interna- tional Standard assumes that the individuals on which the determinations are carried out constitute a random Sample from the original population and may be considered as indepen- dent. The interval so calculated is called the confidence interval for the mean. Associated with it is a confidence Ievel (sometimes termed a confidence coefficient), which is the probability, usually expressed as a percentage, that the interval does con- tain the mean of the population. Only the 95 % and 99 % levels are provided for in this International Standard. 1 Scope This International Standard specifies the statistical treatment of test results needed to calculate a confidence interval for the mean of a population. 2 Field of application The test results are expressed by measurements of a con- tinuous Character. This International Standard does not cover tests of a qualitative Character (for example presence or absence of a property, number of defectives, etc.). The probability distribution taken as a mathematical model for the total population is a normal distribution for which Parameters, mean m and Standard deviation a, are unknown. The normality assumption is very widely satisfied : the distribu- tion of the results obtained under test conditions is generally a normal or nearly normal distribution. lt may, however, be useful to check the validity of the assump- tion of normality by means of appropriate methods’). The calculations may be simplified by a Change of the origin or the unit of the test results but it is dangerous to round off these results. 1) This subject is in preparation. SIS Multi User Licence: ABB TECHNOLOGY LTD 1/6/2009ISO 2602-1980 (E) lt is not permissible to discard any observations or to apply any corrections to apparently doubtful observations without a justification based on experimental, technical or other evident grounds which should be clearly stated. The midpoint of class i is designa ted by yi. The mean m is then estimated by the weighted mean of all midpoin ts of classes : _ 1 k Y = y C niYi i= 1 The test method may be subject to systematic errors, the deter- mination of which is not taken into consideration here. lt should be noted, however, that the existente of such errors may invalidate the methods which follow. In particular, if there is an unsuspected bias the increase of the Sample size n has no influence on bias. The methods that are treated in ISO 2854 may be useful in certain cases for identifying systematic errors. 6 Confidence interval for the mean The confidence interval for the population mean is caiculated from the estimates of the mean and of the Standard devia tion . 3 References The use alternative method of of the range is given calculating the in the annex. confidence interval by ISO 2854, Statistical treatment of data - Problems of estima- tion and tests of means and variances. ISO 3534, Statistics - Vocabulary and symbols. 61 . Estimation of the Standard deviation 4 Definitions and Symbols 6.1.1 Case of ungrouped results The vocabulary and Symbols used in this International Standard are in conformity with ISO 3534. The estimate of the Standard deviation o, calculated from the squares of the deviations from the arithmetic mean, is given by the formula : 5 Estimation of the mean J 1 * S= n-l z (Xi - x7* i= 1 5.1 Case of ungrouped results After the discarding of any doubtful results, the series com- prises n measurements Xi (where i = 1, 2, 3, . . . . n), some of which may have the same value. Xi is the value of the ith measurement (i = 1, 2, 3, . . . . n); y1 is the total number of measurements; The mean m of the underlyi ng normal distribu by the arithmetic mean x of the n results : tion is estimated - x is the arithmetic mean of the n measurements, calculated as in clause 5.1. 1 * - x - = n t: xi i= 1 For ease of calculation, the use of the following formula is recommended : 5.2 Case of results grouped in classes When the number of results is sufficiently high (above 50 for example), it may be advantageous to group them into classes of the same width. In certain cases, the results may also have been directly obtained grouped into classes. 6.1.2 Case of grouped results The frequency of the ith class, i.e. the number of class i, is deno ted by ni. results In the case of grouping by classes, the formula for the estimate of the Standard deviation is written : The number of classes being denoted by k, we have : 1 k S= n-l c ni (yi - fl* i= 1 n = c ‘i i=l SIS Multi User Licence: ABB TECHNOLOGY LTD 1/6/2009ISO 2602-1980 (E) For ease of calculation, the use of the following formula is recommended : or b) *o 95 K?Z--LS Al- n at the confidence level 99 % : where Yj is the mid-Point of the ith class (i = 1, 2, 3, . . . . k); *o 99 m x- qoggw I The coefficients 4 0,975’ q)gg5’ qo,g5’ qo,gg1 arc men in tat& 3. n 2 3 4 5 6 7 8 9 IO 11 12 Table 3 Confidence level 1 Confidence level Two-sided case One-sided case qo,975 6,353 1,304 0 717 0,507 0,399 0,333 0,288 0,255 0,230 0,210 0,194 qo.995 31,828 3,008 1,316 0 843 0,628 0,507 0 429 0,374 0,333 0,302 0,277 qo,95 qo,99 3,157 15 910 0,885 2,111 0 529 1,023 0 388 0 685 0,312 0,523 0,263 0 429 0,230 0,366 0,205 0,322 0,186 0 288 0,170 0,262 0 158 0,241 From : E. Lord. The use of range in place of the Standard deviation in t-test (Biometrika, Vol. 34, 1947, pp.41-671, with entry 4095 in = 2) corrected. - lTl x - Yo,95 w SIS Multi User Licence: ABB TECHNOLOGY LTD 1/6/2009This page intentionally left blank SIS Multi User Licence: ABB TECHNOLOGY LTD 1/6/2009