Statistical interpretation of data - Part 6: Determination of statistical tolerance intervals

ISO 16269-6:2013 describes procedures for establishing statistical tolerance intervals that include at least a specified proportion of the population with a specified confidence level. Both one-sided and two-sided statistical tolerance intervals are provided, a one‑sided interval having either an upper or a lower limit while a two-sided interval has both upper and lower limits. Two methods are provided, a parametric method for the case where the characteristic being studied has a normal distribution and a distribution‑free method for the case where nothing is known about the distribution except that it is continuous. There is also a procedure for the establishment of two‑sided statistical tolerance intervals for more than one normal sample with common unknown variance.

Interprétation statistique des données - Partie 6: Détermination des intervalles statistiques de dispersion

L'ISO 16269-6:2013 décrit des méthodes permettant d'établir les intervalles statistiques de dispersion qui comprennent au moins une proportion spécifiée de la population à un niveau de confiance spécifié. Les intervalles statistiques de dispersion unilatéraux et bilatéraux sont tous deux présentés l'intervalle statistique de dispersion unilatéral étant caractérisé par une limite supérieure ou par une limite inférieure, tandis que l'intervalle statistique de dispersion bilatéral possède à la fois une limite supérieure et une limite inférieure. Deux méthodes sont exposées: une méthode paramétrique, lorsque la caractéristique étudiée est distribuée selon une loi normale, et une méthode non paramétrique, lorsqu'on ne sait rien de la distribution si ce n'est qu'elle est continue. Il existe également une procédure permettant d'établir les intervalles statistiques de dispersion bilatéraux pour plus d'un échantillon normal avec une variance identique mais inconnue.

Statistično tolmačenje podatkov - 6. del: Ugotavljanje statističnih tolerančnih intervalov

Ta del standarda ISO 16269 opisuje postopke za ugotavljanje statističnih tolerančnih intervalov, ki zajemajo vsaj določen del populacije z določeno ravnjo zaupanja. Zajeti so tako enostranski kot dvostranski intervali statistične tolerance, pri čemer ima enostranski interval zgornjo ali spodnjo mejo, dvostranski pa zgornjo in spodnjo mejo. Na voljo sta dve metodi: parametrična metoda, pri kateri ima preiskovana lastnost običajno porazdelitev in metoda brez porazdelitve za primere, kjer ni podatkov o porazdelitvi, razen da je zvezna. Na voljo je tudi postopek za vzpostavitev dvostranskih intervalov statistične tolerance za več kot en normalen vzorec z običajno neznano varianco.

General Information

Status
Published
Publication Date
28-Jan-2015
Technical Committee
Current Stage
6060 - National Implementation/Publication (Adopted Project)
Start Date
16-Jan-2015
Due Date
23-Mar-2015
Completion Date
29-Jan-2015

Relations

Standard
SIST ISO 16269-6:2015
English language
52 pages
sale 10% off
Preview
sale 10% off
Preview
e-Library read for
1 day
Standard
ISO 16269-6:2014 - Statistical interpretation of data — Part 6: Determination of statistical tolerance intervals Released:1/23/2014
English language
46 pages
sale 15% off
Preview
sale 15% off
Preview
Standard
ISO 16269-6:2014 - Interprétation statistique des données — Partie 6: Détermination des intervalles statistiques de dispersion Released:1/23/2014
French language
47 pages
sale 15% off
Preview
sale 15% off
Preview

Standards Content (Sample)


SLOVENSKI STANDARD
01-marec-2015
1DGRPHãþD
SIST ISO 16269-6:2006
6WDWLVWLþQRWROPDþHQMHSRGDWNRYGHO8JRWDYOMDQMHVWDWLVWLþQLKWROHUDQþQLK
LQWHUYDORY
Statistical interpretation of data - Part 6: Determination of statistical tolerance intervals
Interprétation statistique des données - Partie 6: Détermination des intervalles
statistiques de dispersion
Ta slovenski standard je istoveten z: ISO 16269-6:2014
ICS:
03.120.30 8SRUDEDVWDWLVWLþQLKPHWRG Application of statistical
methods
2003-01.Slovenski inštitut za standardizacijo. Razmnoževanje celote ali delov tega standarda ni dovoljeno.

INTERNATIONAL ISO
STANDARD 16269-6
Second edition
2014-01-15
Statistical interpretation of data —
Part 6:
Determination of statistical tolerance
intervals
Interprétation statistique des données —
Partie 6: Détermination des intervalles statistiques de dispersion
Reference number
©
ISO 2014
© ISO 2014
All rights reserved. Unless otherwise specified, no part of this publication may be reproduced or utilized otherwise in any form
or by any means, electronic or mechanical, including photocopying, or posting on the internet or an intranet, without prior
written permission. Permission can be requested from either ISO at the address below or ISO’s member body in the country of
the requester.
ISO copyright office
Case postale 56 • CH-1211 Geneva 20
Tel. + 41 22 749 01 11
Fax + 41 22 749 09 47
E-mail copyright@iso.org
Web www.iso.org
Published in Switzerland
ii © ISO 2014 – All rights reserved

Contents Page
Foreword .iv
Introduction .v
1 Scope . 1
2 Normative references . 1
3 Terms, definitions and symbols . 1
3.1 Terms and definitions . 1
3.2 Symbols . 2
4 Procedures . 3
4.1 Normal population with known mean and known variance . 3
4.2 Normal population with unknown mean and known variance . 3
4.3 Normal population with unknown mean and unknown variance . 4
4.4 Normal populations with unknown means and unknown common variance . 4
4.5 Any continuous distribution of unknown type . 4
5 Examples . 4
5.1 Data for Examples 1 and 2 . 4
5.2 Example 1: One-sided statistical tolerance interval with unknown variance and
unknown mean . 5
5.3 Example 2: Two-sided statistical tolerance interval under unknown mean and
unknown variance . . 6
5.4 Data for Examples 3 and 4 . 6
5.5 Example 3: One-sided statistical tolerance intervals for separate populations with
unknown common variance . 7
5.6 Example 4: Two-sided statistical tolerance intervals for separate populations with
unknown common variance . 8
5.7 Example 5: Any distribution of unknown type .10
Annex A (informative) Exact k-factors for statistical tolerance intervals for the
normal distribution .12
Annex B (informative) Forms for statistical tolerance intervals .17
Annex C (normative) One-sided statistical tolerance limit factors, k (n; p; 1−α), for unknown σ .21
C
Annex D (normative) Two-sided statistical tolerance limit factors, k (n; m; p; 1−α), for unknown
D
common σ (m samples) .26
Annex E (normative) Distribution‑free statistical tolerance intervals .40
Annex F (informative) Computation of factors for two-sided parametric statistical
tolerance intervals .42
Annex G (informative) Construction of a distribution‑free statistical tolerance interval for any
type of distribution .44
Bibliography .46
Foreword
ISO (the International Organization for Standardization) is a worldwide federation of national standards
bodies (ISO member bodies). The work of preparing International Standards is normally carried out
through ISO technical committees. Each member body interested in a subject for which a technical
committee has been established 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.
ISO collaborates closely with the International Electrotechnical Commission (IEC) on all matters of
electrotechnical standardization.
The procedures used to develop this document and those intended for its further maintenance are
described in the ISO/IEC Directives, Part 1. In particular the different approval criteria needed for the
different types of ISO documents should be noted. This document was drafted in accordance with the
editorial rules of the ISO/IEC Directives, Part 2. www.iso.org/directives
Attention is drawn to the possibility that some of the elements of this document may be the subject of
patent rights. ISO shall not be held responsible for identifying any or all such patent rights. Details of any
patent rights identified during the development of the document will be in the Introduction and/or on
the ISO list of patent declarations received. www.iso.org/patents
Any trade name used in this document is information given for the convenience of users and does not
constitute an endorsement.
For an explanation on the meaning of ISO specific terms and expressions related to conformity
assessment, as well as information about ISO’s adherence to the WTO principles in the Technical
Barriers to Trade (TBT) see the following URL: Foreword - Supplementary information
The committee responsible for this document is ISO/TC 69, Applications of statistical methods.
This second edition cancels and replaces the first edition (ISO 16269:2005), which has been technically
revised.
ISO 16269 consists of the following parts, under the general title Statistical interpretation of data:
— Part 4: Detection and treatment of outliers
— Part 6: Determination of statistical tolerance intervals
— Part 7: Median — Estimation and confidence intervals
— Part 8: Determination of prediction intervals
iv © ISO 2014 – All rights reserved

Introduction
A statistical tolerance interval is an estimated interval, based on a sample, which can be asserted with
confidence level 1 − α, for example 0,95, to contain at least a specified proportion p of the items in the
population. The limits of a statistical tolerance interval are called statistical tolerance limits. The
confidence level 1 − α is the probability that a statistical tolerance interval constructed in the prescribed
manner will contain at least a proportion p of the population. Conversely, the probability that this
interval will contain less than the proportion p of the population is α. This part of ISO 16269 describes
both one-sided and two-sided statistical tolerance intervals; a one-sided interval is constructed with an
upper or a lower limit while a two-sided interval is constructed with both an upper and a lower limit.
A statistical tolerance interval depends on a confidence level 1 − α and a stated proportion p of the
population. The confidence level of a statistical tolerance interval is well understood from a confidence
interval for a parameter. The confidence statement of a confidence interval is that the confidence
interval contains the true value of the parameter a proportion 1 − α of the cases in a long series of
repeated random samples under identical conditions. Similarly the confidence statement of a statistical
tolerance interval states that at least a proportion p of the population is contained in the interval in a
proportion 1 − α of the cases of a long series of repeated random samples under identical conditions.
So if we think of the stated proportion of p of the population as a parameter, the idea behind statistical
tolerance intervals is similar to the idea behind confidence intervals.
Statistical tolerance intervals are functions of the observations of the sample, i.e. statistics, and they will
generally take different values for different samples. It is necessary that the observations be independent
for the procedures provided in this part of ISO 16269 to be valid.
Two types of statistical tolerance interval are provided in this part of ISO 16269, parametric and
distribution-free. The parametric approach is based on the assumption that the characteristic being
studied in the population has a normal distribution; hence the confidence that the calculated statistical
tolerance interval contains at least a proportion p of the population can only be taken to be 1 − α if the
normality assumption is true. For normally distributed characteristics, the statistical tolerance interval
is determined using one of the Forms A, B, or C given in Annex B.
Parametric methods for distributions other than the normal are not considered in this part of ISO 16269.
If departure from normality is suspected in the population, distribution-free statistical tolerance
intervals may be constructed. The procedure for the determination of a statistical tolerance interval for
any continuous distribution is provided in Form D of Annex B.
The statistical tolerance limits discussed in this part of ISO 16269 can be used to compare the natural
capability of a process with one or two given specification limits, either an upper one U or a lower one L
or both in statistical process management.
Above the upper specification limit U there is the upper fraction nonconforming p (ISO 3534-2:2006, 2.5.4)
U
and below the lower specification limit L there is the lower fraction nonconforming p (ISO 3534-2:2006,
L
2.5.5). The sum p + p = p is called the total fraction nonconforming. (ISO 3534-2:2006, 2.5.6). Between
U L t
the specification limits U and L there is the fraction conforming 1 − p .
t
The ideas behind statistical tolerance intervals are more widespread than is usually appreciated, for
example in acceptance sampling by variables and in statistical process management, as will be pointed
out in the next two paragraphs.
In acceptance sampling by variables, the limits U and/or L will be known, p , p or p will be specified as
U L t
an acceptable quality limit (AQL), α will be implied and the lot is accepted if there is at least an implicit
100(1−α)% confidence that the AQL is not exceeded.
In statistical process management the limits U and L are fixed in advance and the fractions p , p and
U L
p are either calculated, if the distribution is assumed to be known, or otherwise estimated. This is an
t
example of a quality control application, but there are many other applications of statistical tolerance
[13]
intervals given in textbooks such as Hahn and Meeker.
In contrast, for the statistical tolerance intervals considered in this part of ISO 16269, the confidence
level for the interval estimator and the proportion of the distribution within the interval (corresponding
to the fraction conforming mentioned above) are fixed in advance, and the limits are estimated. These
limits may be compared with U and L. Hence the appropriateness of the given specification limits U
and L can be compared with the actual properties of the process. The one-sided statistical tolerance
intervals are used when only either the upper specification limit U or the lower specification limit L is
relevant, while the two-sided intervals are used when both the upper and the lower specification limits
are considered simultaneously.
The terminology with regard to these different limits and intervals has been confusing, as the
“specification limits” were earlier also called “tolerance limits” (see the terminology standard
ISO 3534-2:1993, 1.4.3, where both these terms as well as the term “limiting values” were all used as
synonyms for this concept). In the latest revision of ISO 3534-2:2006, 3.1.3, only the term specification
limits have been kept for this concept. Furthermore, the Guide for the expression of uncertainty in
[5]
measurement uses the term “coverage factor” defined as a “numerical factor used as a multiplier of
the combined standard uncertainty in order to obtain an expanded uncertainty”. This use of “coverage”
differs from the use of the term in this part of ISO 16269.
The first edition of this standard gave extensive tables of the factor k for one-sided and two-sided
tolerance intervals when the mean is unknown but the standard deviation is known. In this second
edition of the standard those tables are omitted. Instead, exact k-factors are given in Annex A when one
of the parameters of the normal distribution is unknown and the other parameter is known.
The first edition of this standard considered statistical tolerance intervals based only on a single sample
of size n. This edition considers statistical tolerance intervals for m populations with the same standard
deviation, based on samples from each of the m populations, each sample being of the same size n.
vi © ISO 2014 – All rights reserved

INTERNATIONAL STANDARD ISO 16269-6:2014(E)
Statistical interpretation of data —
Part 6:
Determination of statistical tolerance intervals
1 Scope
This part of ISO 16269 describes procedures for establishing statistical tolerance intervals that include
at least a specified proportion of the population with a specified confidence level. Both one-sided and
two-sided statistical tolerance intervals are provided, a one-sided interval having either an upper or a
lower limit while a two-sided interval has both upper and lower limits. Two methods are provided, a
parametric method for the case where the characteristic being studied has a normal distribution and
a distribution-free method for the case where nothing is known about the distribution except that it is
continuous. There is also a procedure for the establishment of two-sided statistical tolerance intervals
for more than one normal sample with common unknown variance.
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 edition of the referenced document (including any amendments) applies.
ISO 3534-1:2006, Statistics — Vocabulary and symbols — Part 1: General statistical terms and terms used
in probability
ISO 3534-2:2006, Statistics — Vocabulary and symbols — Part 2: Applied statistics
3 Terms, definitions and symbols
For the purposes of this document, the terms and definition given in ISO 3534-1, ISO 3534-2 and the
following apply.
3.1 Terms and definitions
3.1.1
statistical tolerance interval
interval determined from a random sample in such a way that one may have a specified level of confidence
that the interval covers at least a specified proportion of the sampled population
[SOURCE: ISO 3534-1:2006, 1.26]
Note 1 to entry: The confidence level in this context is the long-run proportion of intervals constructed in this
manner that will include at least the specified proportion of the sampled population.
3.1.2
statistical tolerance limit
statistic representing an end point of a statistical tolerance interval
[SOURCE: ISO 3534-1:2006, 1.27]
Note 1 to entry: Statistical tolerance intervals may be either
— one-sided (with one of its limits fixed at the natural boundary of the random variable), in which case they
have either an upper or a lower statistical tolerance limit, or
— two-sided, in which case they have both.
3.1.3
coverage
proportion of items in a population lying within a statistical tolerance interval
Note 1 to entry: This concept is not to be confused with the concept coverage factor used in the Guide for the
[5]
expression of uncertainty in measurement (GUM ) .
3.1.4
normal population
normally distributed population
3.2 Symbols
For the purposes of this part of ISO 16269, the following symbols apply.
k (n; p; 1 − α) factor used to determine the limits of one-sided intervals i.e. x or x when μ is
1 L U
known and σ is unknown
k (n; p; 1 − α) factor used to determine the limits of two-sided intervals i.e. x and x when μ is
2 L U
known and σ is unknown
k (n; p; 1 − α) factor used to determine the limits of one-sided intervals i.e. x or x when μ is
3 L U
unknown and σ is known
k (n; p; 1 − α) factor used to determine the limits of two-sided intervals i.e. x and x when μ is
4 L U
unknown and σ is unknown
k (n; p; 1 − α) factor used to determine x or x when the values of μ and σ are unknown for
C L U
one-sided statistical tolerance interval. The suffix C is chosen because this k-factor
is tabulated in Annex C.
k (n; m; p; 1 − α) factor used to determine x and x (i = 1,2,.,m; m ≥ 2) when the values of the means
D Li Ui
μ and the value of the common σ are unknown for the m two-sided statistical toler-
i
ance intervals. The suffix D is chosen because this k-factor is tabulated in Annex D.
n number of observations in the sample
p minimum proportion of the population asserted to be lying in the statistical toler-
ance interval
u p-fractile of the standardized normal distribution
p
x jth observed value .
j
x jth observed value ( j = 1,2,.,n) of ith sample (i = 1,2,.,m)
ij
x maximum value of the observed values: x = max {x , x , …, x }
max max 1 2 n
x minimum value of the observed values: x = min {x , x , …, x }
min min 1 2 n
x lower limit of the statistical tolerance interval
L
x upper limit of the statistical tolerance interval
U
n
x x= x
sample mean,
∑ j
n
j= 1
2 © ISO 2014 – All rights reserved

n
x ()im=12,,., , x = x
i ii∑ j
sample mean of ith sample,
n
j=1
n n
 
 
n xx−
∑∑j j
n  
1 j ==11j
 
s                  sample standard deviation, s = xx− =
()
j

n−11nn−
()
j =1
n
s                 sample standard deviation of ith sample, im=12,,., , s = ()xx−
()
i
ii∑ ji
()n−1
j=1
m n m
1 1
2 2
s                 pooled sample standard deviation, s = ()xx−= s
P
P ∑∑∑ ij i i
mn()−1 m
i==1 j=1 i 1
1 − α confidence level for the assertion that the proportion of the population lying within
the tolerance interval is greater than or equal to the specified level p
μ population mean
μ population mean of the ith population (i = 1,2,.,m)
i
σ population standard deviation
4 Procedures
4.1 Normal population with known mean and known variance
When the values of the mean, μ, and the variance, σ , of a normally distributed population are known, the
distribution of the characteristic under investigation is fully determined. There is exactly a proportion
p of the population:
a) to the right of x = μ −μ × σ (one-sided interval);
L p
b) to the left of x = μ + μ × σ (one-sided interval);
U p
c) between x = μ −μ × σ and x = μ + μ × σ (two-sided interval).
L (1+p)/2 U (1+p)/2
In the above equations, μ is the p-fractile of the standardized normal distribution.
p
NOTE As such statements are known to be true, they are made with 100 % confidence.
4.2 Normal population with unknown mean and known variance
When one or both parameters of the normal distribution are unknown but estimated from a random
sample, intervals with similar properties to the ones in 4.1 can still be constructed. Suppose for example
that the mean is unknown but the variance is known. Then a constant k can be found such that the
interval between
xx=−kxσσ and =+xk
LU
contains at least a proportion p of the population with a specified confidence of 1−α. Note two important
distinctions from the situation in 4.1 where the parameters were assumed known. First, when one or
more parameters are estimated the interval contains at least a proportion p of the population, not exactly
a proportion p of the population. Secondly, when parameters are estimated, the statement is only true
with a pre-specified confidence of 1−α. The factor k in the expression of the limits above depends on the
unknown parameters of the normal distribution, on the proportion p, on the confidence coefficient 1−α,
and on the number of observations in the random sample. Exact k-factors are given in Annex A when one
of the parameters of the normal distribution is unknown and the other parameter is known.
4.3 Normal population with unknown mean and unknown variance
Forms A and B, given in Annex B, are applicable to the case where both the mean and the variance of
the normal population are unknown. Form A applies to the one-sided case, while Form B applies to the
two-sided case. Form A is used with the tables of k-factors in Annex C, or alternatively using the exact
formula for the k-factor given in clause A.5 in Annex A. Form B is used with the k-factors given in the
first column of the tables of Annex D. Details about the derivation of the k-factors of Annex D are given
in Annex F.
4.4 Normal populations with unknown means and unknown common variance
Form C, given in Annex B, is applicable to the case where both the means and the variances of the normal
populations are unknown. Furthermore, the variances are assumed to be identical for all populations
under consideration, in which case we talk of the common variance.
4.5 Any continuous distribution of unknown type
If the characteristic under investigation is a variable from a population of unknown form, then a
statistical tolerance interval can be determined from the sample order statistics x of a sample of n
(i)
independent random observations. The procedure given in Form D used in conjunction with Tables E.1
and E.2 provides the steps for the determination of the required sample size based on the order statistics
to be used, the desired confidence level, and the desired content.
NOTE 1 Statistical tolerance intervals where the choice of end points (based on order statistics) does not
depend on the sampled population are called distribution-free statistical tolerance intervals.
NOTE 2 This International Standard does not provide procedures for distributions of known type other
than the normal distribution. However, if the distribution is continuous, the distribution-free method may be
used. Selected references to scientific literature that may assist in determining tolerance intervals for other
distributions are also provided at the end of this document.
5 Examples
5.1 Data for Examples 1 and 2
Forms A to B, given in Annex B, are illustrated by Examples 1 and 2 using the numerical values of
[2]
ISO 2854:1976 , Clause 2, paragraph 1 of the introductory remarks, Table X, yarn 2: 12 measures of
the breaking load of cotton yarn. It should be noted that the number of observations, n = 12, given here
[1]
for these examples is considerably lower than the one recommended in ISO 2602 . The numerical data
and calculations in the different examples are expressed in centinewtons (see Table 1).
Table 1 — Data for Examples 1 and 2
Values in centinewtons
x 228,6 232,7 238,8 317,2 315,8 275,1 222,2 236,7 224,7 251,2 210,4 270,7

These measurements were obtained from a batch of 12000 bobbins, from one production job, packed
in 120 boxes each containing 100 bobbins. Twelve boxes have been drawn at random from the batch
and a bobbin has been drawn at random from each of these boxes. Test pieces of 50 cm length have been
4 © ISO 2014 – All rights reserved

cut from the yarn on these bobbins, at about 5 m distance from the free end. The tests themselves have
been carried out on the central parts of these test pieces. Previous information makes it reasonable to
assume that the breaking loads measured in these conditions have virtually a normal distribution. It is
demonstrated in ISO 2854; 1976 that the data do not contradict the assumption of a normal distribution.
By using the box plot graphical test of outliers given in ISO 16269-4, one can also conclude that none of
the data values can be declared as outlier with significance level α = 0,05.
The data in Table 1 give the following results:
Sample size: n = 12
n
Sample mean: x==x 3024,1/12=252,01
∑ j
n
j=1
n n
 
 
n xx−
∑∑j j
 
j ==11j 166772,27
 
Sample standard deviation: s = = = 1263,426333= 5,545
nn()−1 12×11
The formal presentation of the calculations will be given in Example 1 using Form A in Annex B (one-sided
interval, unknown variance and unknown mean).
5.2 Example 1: One-sided statistical tolerance interval with unknown variance and un-
known mean
A limit x is required such that it is possible to assert with confidence level 1 − α = 0,95 (95 %) that
L
at least 0,95 (95 %) of the breaking loads of the items in the batch, when measured under the same
conditions, are above x . The presentation of the results is given in detail below.
L
Determination of the statistical tolerance interval of proportion p:
a)  one-sided interval “to the right”
Determined values:
b)  proportion of the population selected for the statistical tolerance interval: p = 0,95
c)  chosen confidence level: 1 − α = 0,95
d)  sample size: n = 12
Value of tolerance factor from Table C.2 : kn(;p;)12−=α ,736 4
C
Calculations:
n
x==x
j

252,01
n
j= 1
n n
 
 
n xx−
∑∑j j
 
j ==11j
 
s =
= 35,545
nn−1
()
kn(;ps;)19−×α = 7,2653
C
Results: one-sided interval “to the right”
The tolerance interval which will contain at least a proportion p of the population with confidence
level 1 − α has a lower limit:
xx=−kn(;ps;)1−×α =154,7
L C
5.3 Example 2: Two-sided statistical tolerance interval under unknown mean and un-
known variance
Suppose it is required to calculate the limits x and x such that it is possible to assert with a confidence
L U
level 1 − α = 0,95 that in a proportion of the batch at least equal to p = 0,90 (90 %) the breaking load falls
between x and x .
L U
The column with m = 1 and the row with n = 12 in Table D.4 gives
kn(;11;;p −=α),26703
D
whence
xx=−kn(;11;;ps−×α),=−25201 2,,6703×=35545 157,0
LD
xx=+kn(;11;;ps−×α),=+25201 2,,6703×=35545 347,0
UD
5.4 Data for Examples 3 and 4
Suppose the percentage of solids in each of four batches of wet brewer’s yeast, each from a different
supplier, is to be determined. The percentages of the four batches are normally distributed with
unknown means μ i = 1,2,3,4. From previous experience of these suppliers, it may be assumed that the
i
variances are the same. A test for the following data gives no reason to suppose otherwise. The data
are therefore assumed to have a common variance σ . The researcher wants to determine two-sided
statistical tolerance intervals for the percentages of solids in each batch.
[14]
The values of random samples of size n = 10 from four batches are given in Table 2:
6 © ISO 2014 – All rights reserved

Table 2 — Data for Examples 3 and 4
Values in percent
j
i
1 2 3 4 5 6 7 8 9 10
1 20 18 16 21 19 17 20 16 19 18
2 19 14 17 13 10 16 14 12 15 11
3 11 12 14 10 8 10 13 9 12 8
4 10 7 11 9 6 11 8 12 13 14
Notice that the jth value of the ith sample is denoted: x .
ij
These results yield the following:
Sample size:  n = 10
Number of samples:  m = 4
Sample means of each of the four batches:
x ==184/,10 18 4 ;   x ==141/,10 14 1 ;    x ==107/,10 107 ;   x ==101/,10 101
1 2 3 4
Sample variances of each of the four batches:
2 2
n n n n
   
2 2
   
nx − x nx − x
∑ 1jj∑ 1 ∑ 2jj∑ 2
   
j=1 j=1 264 j=1 j=1 689
2 2
   
s = = =2,9333 ;   s = = =7,6556
1 2
nn()−1 10×9 nn()−1 10×9
2 2
n n n n
   
2 2
   
nx − x nx − x
∑ 3jj∑ 3 ∑ 4jj∑ 4
   
j=1 j=1 381 j=1 j=1 609
2   2  
s = = =4,2333 ;   s = = =6,7667
3 4
nn()−1 10×9 nn()−1 10×9
Pooled sample standard deviation:
m
s ==s (,2,9333++7 6556 4,,2333+=67667),23232
P ∑ i
m 4
i=1
Degrees of freedom of the pooled standard deviation:
f = m(n − 1) = nm − m = 36
5.5 Example 3: One-sided statistical tolerance intervals for separate populations with
unknown common variance
Suppose it is desired to calculate lower statistical tolerance intervals for the four suppliers, i.e. it is
desired to calculate intervals that contain at least a proportion p for all suppliers. Table C cannot provide
the answer but the intervals are of the same form as was given in Example 1, namely a constant multiplied
by the estimated standard deviation and subtracted from the estimated mean
xx=−kn(;fp;;1−×α),s
LPii i
where the constant k(n ;f;p;1−α) depends on the size of the ith sample and the degrees of freedom of
i
the pooled standard deviation. The expression for the constant is derived in Clause A.5 in Annex A, see
Formula (A.14);
kn(;fp;;1−=α)(tfnu ;),
i 1−α ip
n
i
where tf(;nu ) denotes the 1−α quantile of the non-central t-distribution with non-centrality
1−α ip
parameter nu and f degrees of freedom. The non-central t-distribution and in particular its quantiles
ip
are available in statistical software packages. Suppose a proportion p = 0,95 and a confidence coefficient
1 − α = 0,95 is desired. In this case n = 10 and f = m(n − 1) = nm − m = 36, so the constant is
i
kt(;1036;;0,95 0,95)(==10×1,6449;36),2,3471
0,95
where the 0,95 quantile of the standardized normal distribution u = 1,6449 is inserted.
0,95
The values provided in the tables in Annex C are the special cases where the degrees of freedom are
equal to the sample size minus 1 which is the degrees of freedom of the standard deviation based on a
single sample of size n
kn(;pk;)11−=αα(;nn−−;;pt1 )(=−nu ;)n 1 ,
C 1−α p
n
i.e. the special case, where the degrees of freedom of the estimate of the variance is n − 1.
It follows that the one-sided statistical tolerance limits computed for all four batches are as follows.
First batch:   xx=−kn ;;;ναps11− ×= 84,,02−×3471 2,,3232=1294
()
LP11 1
Second batch: xx=−kn();;;ναps11− ×= 41,,02−×3471 2,,3232=864
LP22 2
Third batch:  xx=−kn;;;ναps11− ×= 07,,02−×3471 2,,3232=466
()
LP33
Fourth batch: xx=−kn;;;ναps11− ×= 01,,02−×3471 2,,3232=406
()
LP44
If the upper statistical tolerance limits had been required, the same quantities would be combined
except that the constant times the standard error would be added to the estimated mean.
5.6 Example 4: Two-sided statistical tolerance intervals for separate populations with
unknown common variance
Case 1 — Computation for all batches (m = 4)
Table D.5 in Annex D gives for n = 10, m = 4, f = m(n − 1) = 4(10 − 1) = 36, p = 0,95 and 1 − α = 0,95 and the
value of the two-sided statistical tolerance factor for unknown common variability σ as
kn(;mp;;12−=α),5964.
D
It follows that the two-sided statistical tolerance limits computed simultaneously for all batches are as
follows.
8 © ISO 2014 – All rights reserved

First batch:
xx=−kn;;mp;,11−α ×=s 8402−×,,5964 23232=12,36
()
LD11 P
xx=+kn;;mp;,11−α ×=s 8402+×,,5964 23232=24,44
()
UD11 P
Second batch:
xx=−kn;;mp;,11−α ×=s 4102−×,,5964 23232=80, 6
()
LD22 P
xx=+kn;;mp;,11−α ×=s 4102+×,,5964 23232=20,14
()
UD22 P
Third batch:
xx=−kn();;mp;,11−α ×=s 0702−×,,5964 23232=46, 6
LD33 P
xx=+kn;;mp;,11−α ×=s 0702+×,,5964 23232=16,74
()
UD33 P
Fourth batch:
xx=−41kn;;mp;,−α ×=s 1010−×2,,5964 23232=40, 6
()
LD4 P
xx=+kn;;mp;,11−α ×=s 0102+×,,5964 23232=16,14
()
UD44 P
NOTE The lower limits have been rounded down and the upper limits have been rounded up (in the second
decimal place) to maintain the integrity of the confidence statements.
Case 2 — Individual computation for each batch (m = 1)
It is possible to compute these tolerance limits separately for each batch. For n = 10, m = 1, f = m(n − 1) =
1(10 − 1) = 9, p = 0,95 and 1 − α = 0,95, the value of the two-sided statistical tolerance factor for unknown
common variability σ equals
k (;1010;,95;,0953),= 3935
D
and can be found in Annex D (Table D.4).
Sample standard deviations of four batches:
2 2
ss== 2,,9333=17127 ;   ss== 7,,6556 =27669
11 22
2 2
ss== 4,,2333=20575 ;   ss== 6,,7667 =26013
33 44
Hence the two-sided statistical tolerance limits are as follows:
First batch:
xx=−kn(;ms;,0950;,95)(×=xk−×10;;10,;95 09,)5 s
LD11 11 D 1
=18,40−−×3,,3935 17127 =12,58
xx=+kn(;mp;;11−×α)(sx=+ks01;;09,;50,)95 ×
UD11 11 D 1
=+18,,40 339935×=1,,7127 24 22

Second batch:
xx=−kn(;mp;;11−×αα)(sx=−kp01;; ;)1−×s
LD22 22 D 2
=−14,,10 33935×22,,7669 =470
xx=+kn(;mp;;11−×αα)(sx=+kp01;; ;)1−×s
UD22 22 D 2
=+14,,10 33935×22,,7669 =2350
Third batch:
xx=−kn(;mp;;11−×αα)(sx=−kn;;mp;)−×s
LD33 33 D 3
=−10,,70 3394×2,00575 =37, 1
xx=+kn(;mp;;11−×αα)(sx=+kp01;; ;)1−×s
UD33 33 D 3
=+10,,70 33935×22,,0575 =17 69

Fourth batch:
xx=−kn(;mp;;11−×αα)(sx=−kp01;; ;)1−×s
LD44 44 D 4
=−10,,10 33935×22,,6013 =127
xx=+kn(;mp;;11−×αα)(sx=+kp01;; ;)1−×s
UD44 44 D 4
=+10,,10 33935×22,,6013 =18 93

When comparing the result of both cases it can be declared that the statistical tolerance intervals for
batches 2, 3 and 4 are substantially smaller in Case 1 than in Case 2. But the statistical tolerance interval
for the first batch is only a little larger in Case 2. The explanation is that the constant k is smaller in
D
Case 1 than in Case 2 because the degrees of freedom are larger in Case 1. Batch 1 has the smallest
estimated standard deviation and this compensates for the increase in the constant k .
D
We can conclude that the statistical tolerance intervals computed simultaneously for several populations
can yield intervals shorter than the statistical tolerance intervals computed for each random sample
separately, provided that the underlying normal populations have the same variance. This nice property
follows from the fact that on the average, the estimate of the variance computed from several random
samples is ’better’ than the estimate computed from one random sample, because the latter is based on
a smaller number of observations.
5.7 Example 5: Any distribution of unknown type
Assume we have a sample, x , x ,…, x , of independent random observations on a population (continuous,
1 2 n
discrete, or mixed) and let its order statistics be x ≤ x ≤ … ≤ x .
(1) (2) (n)
It is possible to determine the sample size necessary to achieve at least 100(1 −α) % confidence that at
least 100p % of the population are lying between the vth smallest observation (i.e., order statistic x )
(v)
and the wth largest observation (i.e., order statistic x ).
(n–w+1)
1)  Determine the sample size n necessary to achieve at least 95 % confidence that at least 99 % of the
population’s measured values lie between the minimum and maximum observations, i.e. between the
first (v = 1) and the nth (w = 1) sample order statistics.
Based on the above description v + w = 2, p = 0,99, and 1 − α = 0,95. The minimum sample size determined
from Table E.1 is 473 (the actual confidence level is 95,020 %). A few examples are given below.
2)  Determine the sample size n necessary to achieve at least 95 % confidence that at least 95 % of the
population’s measured values are greater than or equal to the minimum sample order statistic (v = 1
and w = 0).
10 © ISO 2014 – All rights reserved

Based on the above description, v + w = 1, p = 0,95, and 1 − α = 0,95. The minimum sample size determined
from Table E.1 is 59 (the actual confidence level is 95,151 %).
3)  Determine the sample size n necessary to achieve at least 95 % confidence that at least 99 % of the
population’s units are acceptable with at most one permissible nonconforming unit in the sample.
Based on the description in Annex G, v + w = 2 (v + w −1 = 1 because 1 is the maximum permissible number
of nonconforming items in the sample), p = 0,99, and 1 − α = 0,95. The minimum sample size determined
from Table E.1 is 473 (the actual confidence level is 95,020 %). Note that this result is identical to that of
the first example in this section.
4)  Suppose that the distribution of X is expected to have long tails (i.e., produces occasional extreme
positive and negative values) and extra measures are considered necessary to ensure the resulting
statistical tolerance interval is of a useful length. The experimenter decides to exclude lower and upper
order statistics such that the statistical tolerance interval is constructed between the fifth smallest (v
= 5) and fifth largest (w = 5) order statistics. Determine the sample size n necessary to achieve at least
90 % confidence that at least 99 % of the population’s measured values lie within this interval.
Based on the description in Annex G, v + w = 10, p = 0,99, and 1 − α = 0,90. The minimum sample size
determined from Table E.1 is 1418 (the actual confidence level is 90,000 %) and the associated order
statistics are x and x .
(5) (1414)
Annex A
(informative)
Exact k-factors for statistical tolerance intervals for the normal
distribution
Annex A gives the exact k-factors for calculating tolerance intervals based on a single normal sample. In
this annex, a sample of size n from the N(μ, σ) distribution is considered. Let x and s denote the sample
mean and the sample standard deviation, respectively. Initially, we assume that x and s are estimated
2 2 2
from the same sample, and in that case the x -distribution of (n − 1)s /σ has n − 1 degrees of freedom.
But we might have an independent estimate of the standard deviation with degrees of freedom f, where
typically f is greater than n − 1. For example, this would be the case if the estimate of the standard
deviation were based on several independent samples with a common standard deviation. The exact
formulas are easily modified to deal with this situation.
Type of interval Mean Standard deviation Symbol
One-sided Known Unknown k (n;p;1 − α)
Two-sided Known Unknown k (n;p;1 − α)
One-sided Unknown Known k (n;p;1 − α)
Two-sided Unknown Known k (n;p;1 − α)
One-sided Unknown Unknown k (n;p;1 − α)
C
A.1 One-sided statistical tolerance interval with known mean and unknown
standard deviation
The interval [,−∞ μσ+u ] contains a proportion p of the population, and if
p
μμ+>ks +u σ ,
p
then the interval [,−∞ μ+ks] will contain a proportion of the population that is larger than p. We want
to determine k such that this happens with the probability 1 − α, i.e.
u
 
s
p
Pk()μμ+>su+=σ P > =−1 α (A.1)
 
p
σ k
 
2 2 2
The distribution of s /σ is χ /(n − 1) with n − 1 degrees of freedom, so it follows from the last equality
in Formula (A.1) that
u
χ ()n−1
p
α
=
k n−1
so
n−1
ku= (A.2)
p
χ ()n−1
α
12 © ISO 2014 – All rights reserved

Here χ ()n−1 is the α fractile of the χ distribution with n − 1 degrees of freedom, so this is a value that
α
2 2
is exceeded with probability 1 − α by the random variable s (n − 1)/σ .
The variable k in Formula (A.2) is k (n;p;1 − α).
A.2 Two-sided statistical tolerance interval with known mean and unknown
standard deviation
The interval [,μσ++uuμσ] contains a proportion p of the population, and if
1−+pp1
2 2
μμ+>ks +u σ,
...


INTERNATIONAL ISO
STANDARD 16269-6
Second edition
2014-01-15
Statistical interpretation of data —
Part 6:
Determination of statistical tolerance
intervals
Interprétation statistique des données —
Partie 6: Détermination des intervalles statistiques de dispersion
Reference number
©
ISO 2014
© ISO 2014
All rights reserved. Unless otherwise specified, no part of this publication may be reproduced or utilized otherwise in any form
or by any means, electronic or mechanical, including photocopying, or posting on the internet or an intranet, without prior
written permission. Permission can be requested from either ISO at the address below or ISO’s member body in the country of
the requester.
ISO copyright office
Case postale 56 • CH-1211 Geneva 20
Tel. + 41 22 749 01 11
Fax + 41 22 749 09 47
E-mail copyright@iso.org
Web www.iso.org
Published in Switzerland
ii © ISO 2014 – All rights reserved

Contents Page
Foreword .iv
Introduction .v
1 Scope . 1
2 Normative references . 1
3 Terms, definitions and symbols . 1
3.1 Terms and definitions . 1
3.2 Symbols . 2
4 Procedures . 3
4.1 Normal population with known mean and known variance . 3
4.2 Normal population with unknown mean and known variance . 3
4.3 Normal population with unknown mean and unknown variance . 4
4.4 Normal populations with unknown means and unknown common variance . 4
4.5 Any continuous distribution of unknown type . 4
5 Examples . 4
5.1 Data for Examples 1 and 2 . 4
5.2 Example 1: One-sided statistical tolerance interval with unknown variance and
unknown mean . 5
5.3 Example 2: Two-sided statistical tolerance interval under unknown mean and
unknown variance . . 6
5.4 Data for Examples 3 and 4 . 6
5.5 Example 3: One-sided statistical tolerance intervals for separate populations with
unknown common variance . 7
5.6 Example 4: Two-sided statistical tolerance intervals for separate populations with
unknown common variance . 8
5.7 Example 5: Any distribution of unknown type .10
Annex A (informative) Exact k-factors for statistical tolerance intervals for the
normal distribution .12
Annex B (informative) Forms for statistical tolerance intervals .17
Annex C (normative) One-sided statistical tolerance limit factors, k (n; p; 1−α), for unknown σ .21
C
Annex D (normative) Two-sided statistical tolerance limit factors, k (n; m; p; 1−α), for unknown
D
common σ (m samples) .26
Annex E (normative) Distribution‑free statistical tolerance intervals .40
Annex F (informative) Computation of factors for two-sided parametric statistical
tolerance intervals .42
Annex G (informative) Construction of a distribution‑free statistical tolerance interval for any
type of distribution .44
Bibliography .46
Foreword
ISO (the International Organization for Standardization) is a worldwide federation of national standards
bodies (ISO member bodies). The work of preparing International Standards is normally carried out
through ISO technical committees. Each member body interested in a subject for which a technical
committee has been established 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.
ISO collaborates closely with the International Electrotechnical Commission (IEC) on all matters of
electrotechnical standardization.
The procedures used to develop this document and those intended for its further maintenance are
described in the ISO/IEC Directives, Part 1. In particular the different approval criteria needed for the
different types of ISO documents should be noted. This document was drafted in accordance with the
editorial rules of the ISO/IEC Directives, Part 2. www.iso.org/directives
Attention is drawn to the possibility that some of the elements of this document may be the subject of
patent rights. ISO shall not be held responsible for identifying any or all such patent rights. Details of any
patent rights identified during the development of the document will be in the Introduction and/or on
the ISO list of patent declarations received. www.iso.org/patents
Any trade name used in this document is information given for the convenience of users and does not
constitute an endorsement.
For an explanation on the meaning of ISO specific terms and expressions related to conformity
assessment, as well as information about ISO’s adherence to the WTO principles in the Technical
Barriers to Trade (TBT) see the following URL: Foreword - Supplementary information
The committee responsible for this document is ISO/TC 69, Applications of statistical methods.
This second edition cancels and replaces the first edition (ISO 16269:2005), which has been technically
revised.
ISO 16269 consists of the following parts, under the general title Statistical interpretation of data:
— Part 4: Detection and treatment of outliers
— Part 6: Determination of statistical tolerance intervals
— Part 7: Median — Estimation and confidence intervals
— Part 8: Determination of prediction intervals
iv © ISO 2014 – All rights reserved

Introduction
A statistical tolerance interval is an estimated interval, based on a sample, which can be asserted with
confidence level 1 − α, for example 0,95, to contain at least a specified proportion p of the items in the
population. The limits of a statistical tolerance interval are called statistical tolerance limits. The
confidence level 1 − α is the probability that a statistical tolerance interval constructed in the prescribed
manner will contain at least a proportion p of the population. Conversely, the probability that this
interval will contain less than the proportion p of the population is α. This part of ISO 16269 describes
both one-sided and two-sided statistical tolerance intervals; a one-sided interval is constructed with an
upper or a lower limit while a two-sided interval is constructed with both an upper and a lower limit.
A statistical tolerance interval depends on a confidence level 1 − α and a stated proportion p of the
population. The confidence level of a statistical tolerance interval is well understood from a confidence
interval for a parameter. The confidence statement of a confidence interval is that the confidence
interval contains the true value of the parameter a proportion 1 − α of the cases in a long series of
repeated random samples under identical conditions. Similarly the confidence statement of a statistical
tolerance interval states that at least a proportion p of the population is contained in the interval in a
proportion 1 − α of the cases of a long series of repeated random samples under identical conditions.
So if we think of the stated proportion of p of the population as a parameter, the idea behind statistical
tolerance intervals is similar to the idea behind confidence intervals.
Statistical tolerance intervals are functions of the observations of the sample, i.e. statistics, and they will
generally take different values for different samples. It is necessary that the observations be independent
for the procedures provided in this part of ISO 16269 to be valid.
Two types of statistical tolerance interval are provided in this part of ISO 16269, parametric and
distribution-free. The parametric approach is based on the assumption that the characteristic being
studied in the population has a normal distribution; hence the confidence that the calculated statistical
tolerance interval contains at least a proportion p of the population can only be taken to be 1 − α if the
normality assumption is true. For normally distributed characteristics, the statistical tolerance interval
is determined using one of the Forms A, B, or C given in Annex B.
Parametric methods for distributions other than the normal are not considered in this part of ISO 16269.
If departure from normality is suspected in the population, distribution-free statistical tolerance
intervals may be constructed. The procedure for the determination of a statistical tolerance interval for
any continuous distribution is provided in Form D of Annex B.
The statistical tolerance limits discussed in this part of ISO 16269 can be used to compare the natural
capability of a process with one or two given specification limits, either an upper one U or a lower one L
or both in statistical process management.
Above the upper specification limit U there is the upper fraction nonconforming p (ISO 3534-2:2006, 2.5.4)
U
and below the lower specification limit L there is the lower fraction nonconforming p (ISO 3534-2:2006,
L
2.5.5). The sum p + p = p is called the total fraction nonconforming. (ISO 3534-2:2006, 2.5.6). Between
U L t
the specification limits U and L there is the fraction conforming 1 − p .
t
The ideas behind statistical tolerance intervals are more widespread than is usually appreciated, for
example in acceptance sampling by variables and in statistical process management, as will be pointed
out in the next two paragraphs.
In acceptance sampling by variables, the limits U and/or L will be known, p , p or p will be specified as
U L t
an acceptable quality limit (AQL), α will be implied and the lot is accepted if there is at least an implicit
100(1−α)% confidence that the AQL is not exceeded.
In statistical process management the limits U and L are fixed in advance and the fractions p , p and
U L
p are either calculated, if the distribution is assumed to be known, or otherwise estimated. This is an
t
example of a quality control application, but there are many other applications of statistical tolerance
[13]
intervals given in textbooks such as Hahn and Meeker.
In contrast, for the statistical tolerance intervals considered in this part of ISO 16269, the confidence
level for the interval estimator and the proportion of the distribution within the interval (corresponding
to the fraction conforming mentioned above) are fixed in advance, and the limits are estimated. These
limits may be compared with U and L. Hence the appropriateness of the given specification limits U
and L can be compared with the actual properties of the process. The one-sided statistical tolerance
intervals are used when only either the upper specification limit U or the lower specification limit L is
relevant, while the two-sided intervals are used when both the upper and the lower specification limits
are considered simultaneously.
The terminology with regard to these different limits and intervals has been confusing, as the
“specification limits” were earlier also called “tolerance limits” (see the terminology standard
ISO 3534-2:1993, 1.4.3, where both these terms as well as the term “limiting values” were all used as
synonyms for this concept). In the latest revision of ISO 3534-2:2006, 3.1.3, only the term specification
limits have been kept for this concept. Furthermore, the Guide for the expression of uncertainty in
[5]
measurement uses the term “coverage factor” defined as a “numerical factor used as a multiplier of
the combined standard uncertainty in order to obtain an expanded uncertainty”. This use of “coverage”
differs from the use of the term in this part of ISO 16269.
The first edition of this standard gave extensive tables of the factor k for one-sided and two-sided
tolerance intervals when the mean is unknown but the standard deviation is known. In this second
edition of the standard those tables are omitted. Instead, exact k-factors are given in Annex A when one
of the parameters of the normal distribution is unknown and the other parameter is known.
The first edition of this standard considered statistical tolerance intervals based only on a single sample
of size n. This edition considers statistical tolerance intervals for m populations with the same standard
deviation, based on samples from each of the m populations, each sample being of the same size n.
vi © ISO 2014 – All rights reserved

INTERNATIONAL STANDARD ISO 16269-6:2014(E)
Statistical interpretation of data —
Part 6:
Determination of statistical tolerance intervals
1 Scope
This part of ISO 16269 describes procedures for establishing statistical tolerance intervals that include
at least a specified proportion of the population with a specified confidence level. Both one-sided and
two-sided statistical tolerance intervals are provided, a one-sided interval having either an upper or a
lower limit while a two-sided interval has both upper and lower limits. Two methods are provided, a
parametric method for the case where the characteristic being studied has a normal distribution and
a distribution-free method for the case where nothing is known about the distribution except that it is
continuous. There is also a procedure for the establishment of two-sided statistical tolerance intervals
for more than one normal sample with common unknown variance.
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 edition of the referenced document (including any amendments) applies.
ISO 3534-1:2006, Statistics — Vocabulary and symbols — Part 1: General statistical terms and terms used
in probability
ISO 3534-2:2006, Statistics — Vocabulary and symbols — Part 2: Applied statistics
3 Terms, definitions and symbols
For the purposes of this document, the terms and definition given in ISO 3534-1, ISO 3534-2 and the
following apply.
3.1 Terms and definitions
3.1.1
statistical tolerance interval
interval determined from a random sample in such a way that one may have a specified level of confidence
that the interval covers at least a specified proportion of the sampled population
[SOURCE: ISO 3534-1:2006, 1.26]
Note 1 to entry: The confidence level in this context is the long-run proportion of intervals constructed in this
manner that will include at least the specified proportion of the sampled population.
3.1.2
statistical tolerance limit
statistic representing an end point of a statistical tolerance interval
[SOURCE: ISO 3534-1:2006, 1.27]
Note 1 to entry: Statistical tolerance intervals may be either
— one-sided (with one of its limits fixed at the natural boundary of the random variable), in which case they
have either an upper or a lower statistical tolerance limit, or
— two-sided, in which case they have both.
3.1.3
coverage
proportion of items in a population lying within a statistical tolerance interval
Note 1 to entry: This concept is not to be confused with the concept coverage factor used in the Guide for the
[5]
expression of uncertainty in measurement (GUM ) .
3.1.4
normal population
normally distributed population
3.2 Symbols
For the purposes of this part of ISO 16269, the following symbols apply.
k (n; p; 1 − α) factor used to determine the limits of one-sided intervals i.e. x or x when μ is
1 L U
known and σ is unknown
k (n; p; 1 − α) factor used to determine the limits of two-sided intervals i.e. x and x when μ is
2 L U
known and σ is unknown
k (n; p; 1 − α) factor used to determine the limits of one-sided intervals i.e. x or x when μ is
3 L U
unknown and σ is known
k (n; p; 1 − α) factor used to determine the limits of two-sided intervals i.e. x and x when μ is
4 L U
unknown and σ is unknown
k (n; p; 1 − α) factor used to determine x or x when the values of μ and σ are unknown for
C L U
one-sided statistical tolerance interval. The suffix C is chosen because this k-factor
is tabulated in Annex C.
k (n; m; p; 1 − α) factor used to determine x and x (i = 1,2,.,m; m ≥ 2) when the values of the means
D Li Ui
μ and the value of the common σ are unknown for the m two-sided statistical toler-
i
ance intervals. The suffix D is chosen because this k-factor is tabulated in Annex D.
n number of observations in the sample
p minimum proportion of the population asserted to be lying in the statistical toler-
ance interval
u p-fractile of the standardized normal distribution
p
x jth observed value .
j
x jth observed value ( j = 1,2,.,n) of ith sample (i = 1,2,.,m)
ij
x maximum value of the observed values: x = max {x , x , …, x }
max max 1 2 n
x minimum value of the observed values: x = min {x , x , …, x }
min min 1 2 n
x lower limit of the statistical tolerance interval
L
x upper limit of the statistical tolerance interval
U
n
x x= x
sample mean,
∑ j
n
j= 1
2 © ISO 2014 – All rights reserved

n
x ()im=12,,., , x = x
i ii∑ j
sample mean of ith sample,
n
j=1
n n
 
 
n xx−
∑∑j j
n  
1 j ==11j
 
s                  sample standard deviation, s = xx− =
()
j

n−11nn−
()
j =1
n
s                 sample standard deviation of ith sample, im=12,,., , s = ()xx−
()
i
ii∑ ji
()n−1
j=1
m n m
1 1
2 2
s                 pooled sample standard deviation, s = ()xx−= s
P
P ∑∑∑ ij i i
mn()−1 m
i==1 j=1 i 1
1 − α confidence level for the assertion that the proportion of the population lying within
the tolerance interval is greater than or equal to the specified level p
μ population mean
μ population mean of the ith population (i = 1,2,.,m)
i
σ population standard deviation
4 Procedures
4.1 Normal population with known mean and known variance
When the values of the mean, μ, and the variance, σ , of a normally distributed population are known, the
distribution of the characteristic under investigation is fully determined. There is exactly a proportion
p of the population:
a) to the right of x = μ −μ × σ (one-sided interval);
L p
b) to the left of x = μ + μ × σ (one-sided interval);
U p
c) between x = μ −μ × σ and x = μ + μ × σ (two-sided interval).
L (1+p)/2 U (1+p)/2
In the above equations, μ is the p-fractile of the standardized normal distribution.
p
NOTE As such statements are known to be true, they are made with 100 % confidence.
4.2 Normal population with unknown mean and known variance
When one or both parameters of the normal distribution are unknown but estimated from a random
sample, intervals with similar properties to the ones in 4.1 can still be constructed. Suppose for example
that the mean is unknown but the variance is known. Then a constant k can be found such that the
interval between
xx=−kxσσ and =+xk
LU
contains at least a proportion p of the population with a specified confidence of 1−α. Note two important
distinctions from the situation in 4.1 where the parameters were assumed known. First, when one or
more parameters are estimated the interval contains at least a proportion p of the population, not exactly
a proportion p of the population. Secondly, when parameters are estimated, the statement is only true
with a pre-specified confidence of 1−α. The factor k in the expression of the limits above depends on the
unknown parameters of the normal distribution, on the proportion p, on the confidence coefficient 1−α,
and on the number of observations in the random sample. Exact k-factors are given in Annex A when one
of the parameters of the normal distribution is unknown and the other parameter is known.
4.3 Normal population with unknown mean and unknown variance
Forms A and B, given in Annex B, are applicable to the case where both the mean and the variance of
the normal population are unknown. Form A applies to the one-sided case, while Form B applies to the
two-sided case. Form A is used with the tables of k-factors in Annex C, or alternatively using the exact
formula for the k-factor given in clause A.5 in Annex A. Form B is used with the k-factors given in the
first column of the tables of Annex D. Details about the derivation of the k-factors of Annex D are given
in Annex F.
4.4 Normal populations with unknown means and unknown common variance
Form C, given in Annex B, is applicable to the case where both the means and the variances of the normal
populations are unknown. Furthermore, the variances are assumed to be identical for all populations
under consideration, in which case we talk of the common variance.
4.5 Any continuous distribution of unknown type
If the characteristic under investigation is a variable from a population of unknown form, then a
statistical tolerance interval can be determined from the sample order statistics x of a sample of n
(i)
independent random observations. The procedure given in Form D used in conjunction with Tables E.1
and E.2 provides the steps for the determination of the required sample size based on the order statistics
to be used, the desired confidence level, and the desired content.
NOTE 1 Statistical tolerance intervals where the choice of end points (based on order statistics) does not
depend on the sampled population are called distribution-free statistical tolerance intervals.
NOTE 2 This International Standard does not provide procedures for distributions of known type other
than the normal distribution. However, if the distribution is continuous, the distribution-free method may be
used. Selected references to scientific literature that may assist in determining tolerance intervals for other
distributions are also provided at the end of this document.
5 Examples
5.1 Data for Examples 1 and 2
Forms A to B, given in Annex B, are illustrated by Examples 1 and 2 using the numerical values of
[2]
ISO 2854:1976 , Clause 2, paragraph 1 of the introductory remarks, Table X, yarn 2: 12 measures of
the breaking load of cotton yarn. It should be noted that the number of observations, n = 12, given here
[1]
for these examples is considerably lower than the one recommended in ISO 2602 . The numerical data
and calculations in the different examples are expressed in centinewtons (see Table 1).
Table 1 — Data for Examples 1 and 2
Values in centinewtons
x 228,6 232,7 238,8 317,2 315,8 275,1 222,2 236,7 224,7 251,2 210,4 270,7

These measurements were obtained from a batch of 12000 bobbins, from one production job, packed
in 120 boxes each containing 100 bobbins. Twelve boxes have been drawn at random from the batch
and a bobbin has been drawn at random from each of these boxes. Test pieces of 50 cm length have been
4 © ISO 2014 – All rights reserved

cut from the yarn on these bobbins, at about 5 m distance from the free end. The tests themselves have
been carried out on the central parts of these test pieces. Previous information makes it reasonable to
assume that the breaking loads measured in these conditions have virtually a normal distribution. It is
demonstrated in ISO 2854; 1976 that the data do not contradict the assumption of a normal distribution.
By using the box plot graphical test of outliers given in ISO 16269-4, one can also conclude that none of
the data values can be declared as outlier with significance level α = 0,05.
The data in Table 1 give the following results:
Sample size: n = 12
n
Sample mean: x==x 3024,1/12=252,01
∑ j
n
j=1
n n
 
 
n xx−
∑∑j j
 
j ==11j 166772,27
 
Sample standard deviation: s = = = 1263,426333= 5,545
nn()−1 12×11
The formal presentation of the calculations will be given in Example 1 using Form A in Annex B (one-sided
interval, unknown variance and unknown mean).
5.2 Example 1: One-sided statistical tolerance interval with unknown variance and un-
known mean
A limit x is required such that it is possible to assert with confidence level 1 − α = 0,95 (95 %) that
L
at least 0,95 (95 %) of the breaking loads of the items in the batch, when measured under the same
conditions, are above x . The presentation of the results is given in detail below.
L
Determination of the statistical tolerance interval of proportion p:
a)  one-sided interval “to the right”
Determined values:
b)  proportion of the population selected for the statistical tolerance interval: p = 0,95
c)  chosen confidence level: 1 − α = 0,95
d)  sample size: n = 12
Value of tolerance factor from Table C.2 : kn(;p;)12−=α ,736 4
C
Calculations:
n
x==x
j

252,01
n
j= 1
n n
 
 
n xx−
∑∑j j
 
j ==11j
 
s =
= 35,545
nn−1
()
kn(;ps;)19−×α = 7,2653
C
Results: one-sided interval “to the right”
The tolerance interval which will contain at least a proportion p of the population with confidence
level 1 − α has a lower limit:
xx=−kn(;ps;)1−×α =154,7
L C
5.3 Example 2: Two-sided statistical tolerance interval under unknown mean and un-
known variance
Suppose it is required to calculate the limits x and x such that it is possible to assert with a confidence
L U
level 1 − α = 0,95 that in a proportion of the batch at least equal to p = 0,90 (90 %) the breaking load falls
between x and x .
L U
The column with m = 1 and the row with n = 12 in Table D.4 gives
kn(;11;;p −=α),26703
D
whence
xx=−kn(;11;;ps−×α),=−25201 2,,6703×=35545 157,0
LD
xx=+kn(;11;;ps−×α),=+25201 2,,6703×=35545 347,0
UD
5.4 Data for Examples 3 and 4
Suppose the percentage of solids in each of four batches of wet brewer’s yeast, each from a different
supplier, is to be determined. The percentages of the four batches are normally distributed with
unknown means μ i = 1,2,3,4. From previous experience of these suppliers, it may be assumed that the
i
variances are the same. A test for the following data gives no reason to suppose otherwise. The data
are therefore assumed to have a common variance σ . The researcher wants to determine two-sided
statistical tolerance intervals for the percentages of solids in each batch.
[14]
The values of random samples of size n = 10 from four batches are given in Table 2:
6 © ISO 2014 – All rights reserved

Table 2 — Data for Examples 3 and 4
Values in percent
j
i
1 2 3 4 5 6 7 8 9 10
1 20 18 16 21 19 17 20 16 19 18
2 19 14 17 13 10 16 14 12 15 11
3 11 12 14 10 8 10 13 9 12 8
4 10 7 11 9 6 11 8 12 13 14
Notice that the jth value of the ith sample is denoted: x .
ij
These results yield the following:
Sample size:  n = 10
Number of samples:  m = 4
Sample means of each of the four batches:
x ==184/,10 18 4 ;   x ==141/,10 14 1 ;    x ==107/,10 107 ;   x ==101/,10 101
1 2 3 4
Sample variances of each of the four batches:
2 2
n n n n
   
2 2
   
nx − x nx − x
∑ 1jj∑ 1 ∑ 2jj∑ 2
   
j=1 j=1 264 j=1 j=1 689
2 2
   
s = = =2,9333 ;   s = = =7,6556
1 2
nn()−1 10×9 nn()−1 10×9
2 2
n n n n
   
2 2
   
nx − x nx − x
∑ 3jj∑ 3 ∑ 4jj∑ 4
   
j=1 j=1 381 j=1 j=1 609
2   2  
s = = =4,2333 ;   s = = =6,7667
3 4
nn()−1 10×9 nn()−1 10×9
Pooled sample standard deviation:
m
s ==s (,2,9333++7 6556 4,,2333+=67667),23232
P ∑ i
m 4
i=1
Degrees of freedom of the pooled standard deviation:
f = m(n − 1) = nm − m = 36
5.5 Example 3: One-sided statistical tolerance intervals for separate populations with
unknown common variance
Suppose it is desired to calculate lower statistical tolerance intervals for the four suppliers, i.e. it is
desired to calculate intervals that contain at least a proportion p for all suppliers. Table C cannot provide
the answer but the intervals are of the same form as was given in Example 1, namely a constant multiplied
by the estimated standard deviation and subtracted from the estimated mean
xx=−kn(;fp;;1−×α),s
LPii i
where the constant k(n ;f;p;1−α) depends on the size of the ith sample and the degrees of freedom of
i
the pooled standard deviation. The expression for the constant is derived in Clause A.5 in Annex A, see
Formula (A.14);
kn(;fp;;1−=α)(tfnu ;),
i 1−α ip
n
i
where tf(;nu ) denotes the 1−α quantile of the non-central t-distribution with non-centrality
1−α ip
parameter nu and f degrees of freedom. The non-central t-distribution and in particular its quantiles
ip
are available in statistical software packages. Suppose a proportion p = 0,95 and a confidence coefficient
1 − α = 0,95 is desired. In this case n = 10 and f = m(n − 1) = nm − m = 36, so the constant is
i
kt(;1036;;0,95 0,95)(==10×1,6449;36),2,3471
0,95
where the 0,95 quantile of the standardized normal distribution u = 1,6449 is inserted.
0,95
The values provided in the tables in Annex C are the special cases where the degrees of freedom are
equal to the sample size minus 1 which is the degrees of freedom of the standard deviation based on a
single sample of size n
kn(;pk;)11−=αα(;nn−−;;pt1 )(=−nu ;)n 1 ,
C 1−α p
n
i.e. the special case, where the degrees of freedom of the estimate of the variance is n − 1.
It follows that the one-sided statistical tolerance limits computed for all four batches are as follows.
First batch:   xx=−kn ;;;ναps11− ×= 84,,02−×3471 2,,3232=1294
()
LP11 1
Second batch: xx=−kn();;;ναps11− ×= 41,,02−×3471 2,,3232=864
LP22 2
Third batch:  xx=−kn;;;ναps11− ×= 07,,02−×3471 2,,3232=466
()
LP33
Fourth batch: xx=−kn;;;ναps11− ×= 01,,02−×3471 2,,3232=406
()
LP44
If the upper statistical tolerance limits had been required, the same quantities would be combined
except that the constant times the standard error would be added to the estimated mean.
5.6 Example 4: Two-sided statistical tolerance intervals for separate populations with
unknown common variance
Case 1 — Computation for all batches (m = 4)
Table D.5 in Annex D gives for n = 10, m = 4, f = m(n − 1) = 4(10 − 1) = 36, p = 0,95 and 1 − α = 0,95 and the
value of the two-sided statistical tolerance factor for unknown common variability σ as
kn(;mp;;12−=α),5964.
D
It follows that the two-sided statistical tolerance limits computed simultaneously for all batches are as
follows.
8 © ISO 2014 – All rights reserved

First batch:
xx=−kn;;mp;,11−α ×=s 8402−×,,5964 23232=12,36
()
LD11 P
xx=+kn;;mp;,11−α ×=s 8402+×,,5964 23232=24,44
()
UD11 P
Second batch:
xx=−kn;;mp;,11−α ×=s 4102−×,,5964 23232=80, 6
()
LD22 P
xx=+kn;;mp;,11−α ×=s 4102+×,,5964 23232=20,14
()
UD22 P
Third batch:
xx=−kn();;mp;,11−α ×=s 0702−×,,5964 23232=46, 6
LD33 P
xx=+kn;;mp;,11−α ×=s 0702+×,,5964 23232=16,74
()
UD33 P
Fourth batch:
xx=−41kn;;mp;,−α ×=s 1010−×2,,5964 23232=40, 6
()
LD4 P
xx=+kn;;mp;,11−α ×=s 0102+×,,5964 23232=16,14
()
UD44 P
NOTE The lower limits have been rounded down and the upper limits have been rounded up (in the second
decimal place) to maintain the integrity of the confidence statements.
Case 2 — Individual computation for each batch (m = 1)
It is possible to compute these tolerance limits separately for each batch. For n = 10, m = 1, f = m(n − 1) =
1(10 − 1) = 9, p = 0,95 and 1 − α = 0,95, the value of the two-sided statistical tolerance factor for unknown
common variability σ equals
k (;1010;,95;,0953),= 3935
D
and can be found in Annex D (Table D.4).
Sample standard deviations of four batches:
2 2
ss== 2,,9333=17127 ;   ss== 7,,6556 =27669
11 22
2 2
ss== 4,,2333=20575 ;   ss== 6,,7667 =26013
33 44
Hence the two-sided statistical tolerance limits are as follows:
First batch:
xx=−kn(;ms;,0950;,95)(×=xk−×10;;10,;95 09,)5 s
LD11 11 D 1
=18,40−−×3,,3935 17127 =12,58
xx=+kn(;mp;;11−×α)(sx=+ks01;;09,;50,)95 ×
UD11 11 D 1
=+18,,40 339935×=1,,7127 24 22

Second batch:
xx=−kn(;mp;;11−×αα)(sx=−kp01;; ;)1−×s
LD22 22 D 2
=−14,,10 33935×22,,7669 =470
xx=+kn(;mp;;11−×αα)(sx=+kp01;; ;)1−×s
UD22 22 D 2
=+14,,10 33935×22,,7669 =2350
Third batch:
xx=−kn(;mp;;11−×αα)(sx=−kn;;mp;)−×s
LD33 33 D 3
=−10,,70 3394×2,00575 =37, 1
xx=+kn(;mp;;11−×αα)(sx=+kp01;; ;)1−×s
UD33 33 D 3
=+10,,70 33935×22,,0575 =17 69

Fourth batch:
xx=−kn(;mp;;11−×αα)(sx=−kp01;; ;)1−×s
LD44 44 D 4
=−10,,10 33935×22,,6013 =127
xx=+kn(;mp;;11−×αα)(sx=+kp01;; ;)1−×s
UD44 44 D 4
=+10,,10 33935×22,,6013 =18 93

When comparing the result of both cases it can be declared that the statistical tolerance intervals for
batches 2, 3 and 4 are substantially smaller in Case 1 than in Case 2. But the statistical tolerance interval
for the first batch is only a little larger in Case 2. The explanation is that the constant k is smaller in
D
Case 1 than in Case 2 because the degrees of freedom are larger in Case 1. Batch 1 has the smallest
estimated standard deviation and this compensates for the increase in the constant k .
D
We can conclude that the statistical tolerance intervals computed simultaneously for several populations
can yield intervals shorter than the statistical tolerance intervals computed for each random sample
separately, provided that the underlying normal populations have the same variance. This nice property
follows from the fact that on the average, the estimate of the variance computed from several random
samples is ’better’ than the estimate computed from one random sample, because the latter is based on
a smaller number of observations.
5.7 Example 5: Any distribution of unknown type
Assume we have a sample, x , x ,…, x , of independent random observations on a population (continuous,
1 2 n
discrete, or mixed) and let its order statistics be x ≤ x ≤ … ≤ x .
(1) (2) (n)
It is possible to determine the sample size necessary to achieve at least 100(1 −α) % confidence that at
least 100p % of the population are lying between the vth smallest observation (i.e., order statistic x )
(v)
and the wth largest observation (i.e., order statistic x ).
(n–w+1)
1)  Determine the sample size n necessary to achieve at least 95 % confidence that at least 99 % of the
population’s measured values lie between the minimum and maximum observations, i.e. between the
first (v = 1) and the nth (w = 1) sample order statistics.
Based on the above description v + w = 2, p = 0,99, and 1 − α = 0,95. The minimum sample size determined
from Table E.1 is 473 (the actual confidence level is 95,020 %). A few examples are given below.
2)  Determine the sample size n necessary to achieve at least 95 % confidence that at least 95 % of the
population’s measured values are greater than or equal to the minimum sample order statistic (v = 1
and w = 0).
10 © ISO 2014 – All rights reserved

Based on the above description, v + w = 1, p = 0,95, and 1 − α = 0,95. The minimum sample size determined
from Table E.1 is 59 (the actual confidence level is 95,151 %).
3)  Determine the sample size n necessary to achieve at least 95 % confidence that at least 99 % of the
population’s units are acceptable with at most one permissible nonconforming unit in the sample.
Based on the description in Annex G, v + w = 2 (v + w −1 = 1 because 1 is the maximum permissible number
of nonconforming items in the sample), p = 0,99, and 1 − α = 0,95. The minimum sample size determined
from Table E.1 is 473 (the actual confidence level is 95,020 %). Note that this result is identical to that of
the first example in this section.
4)  Suppose that the distribution of X is expected to have long tails (i.e., produces occasional extreme
positive and negative values) and extra measures are considered necessary to ensure the resulting
statistical tolerance interval is of a useful length. The experimenter decides to exclude lower and upper
order statistics such that the statistical tolerance interval is constructed between the fifth smallest (v
= 5) and fifth largest (w = 5) order statistics. Determine the sample size n necessary to achieve at least
90 % confidence that at least 99 % of the population’s measured values lie within this interval.
Based on the description in Annex G, v + w = 10, p = 0,99, and 1 − α = 0,90. The minimum sample size
determined from Table E.1 is 1418 (the actual confidence level is 90,000 %) and the associated order
statistics are x and x .
(5) (1414)
Annex A
(informative)
Exact k-factors for statistical tolerance intervals for the normal
distribution
Annex A gives the exact k-factors for calculating tolerance intervals based on a single normal sample. In
this annex, a sample of size n from the N(μ, σ) distribution is considered. Let x and s denote the sample
mean and the sample standard deviation, respectively. Initially, we assume that x and s are estimated
2 2 2
from the same sample, and in that case the x -distribution of (n − 1)s /σ has n − 1 degrees of freedom.
But we might have an independent estimate of the standard deviation with degrees of freedom f, where
typically f is greater than n − 1. For example, this would be the case if the estimate of the standard
deviation were based on several independent samples with a common standard deviation. The exact
formulas are easily modified to deal with this situation.
Type of interval Mean Standard deviation Symbol
One-sided Known Unknown k (n;p;1 − α)
Two-sided Known Unknown k (n;p;1 − α)
One-sided Unknown Known k (n;p;1 − α)
Two-sided Unknown Known k (n;p;1 − α)
One-sided Unknown Unknown k (n;p;1 − α)
C
A.1 One-sided statistical tolerance interval with known mean and unknown
standard deviation
The interval [,−∞ μσ+u ] contains a proportion p of the population, and if
p
μμ+>ks +u σ ,
p
then the interval [,−∞ μ+ks] will contain a proportion of the population that is larger than p. We want
to determine k such that this happens with the probability 1 − α, i.e.
u
 
s
p
Pk()μμ+>su+=σ P > =−1 α (A.1)
 
p
σ k
 
2 2 2
The distribution of s /σ is χ /(n − 1) with n − 1 degrees of freedom, so it follows from the last equality
in Formula (A.1) that
u
χ ()n−1
p
α
=
k n−1
so
n−1
ku= (A.2)
p
χ ()n−1
α
12 © ISO 2014 – All rights reserved

Here χ ()n−1 is the α fractile of the χ distribution with n − 1 degrees of freedom, so this is a value that
α
2 2
is exceeded with probability 1 − α by the random variable s (n − 1)/σ .
The variable k in Formula (A.2) is k (n;p;1 − α).
A.2 Two-sided statistical tolerance interval with known mean and unknown
standard deviation
The interval [,μσ++uuμσ] contains a proportion p of the population, and if
1−+pp1
2 2
μμ+>ks +u σ,
1+p
then the interval [,μμ−+ks ks] will contain a proportion of the population that is larger than p. We want
to determine k such that this happens with the probability 1 − α, i.e.
   
s 1
   
Pkμμ+>su+ σ =>P u =−1 α (A.3)
1++pp1
   
σ k
 2   2 
2 2 2
The distribution of s /σ is χ /(n − 1) with n − 1 degrees of freedom, so it follows from the last equality
in Formula (A.3) that
1 χ ()n−1
α
u =
1+p
k n−1
so
n−1
ku= . (A.4)
1+p
χ ()n−1
2 α
Here χ ()n−1 is the α fractile of the χ distribution with n − 1 degrees of freedom, so this is a value that
α
2 2
is exceeded with probability 1 − α by the random variable s (n − 1)/σ .
The variable k in Formula (A.4) is k (n;p;1 − α).
A.3 One-sided statistical tolerance interval with unknown mean and known
standard deviation
Find k such that xk+ σ satisfies that at least a proportion p of the population is below xk+ σ . Note that
μσ+u is the population tolerance limit in the sense that exactly a proportion p of the population is
p
below that limit. So if
xk+≥σμ+u σ ,
p
then the proportion of the population that is smaller than xk+ σ is at least p. Thus the probability that
a proportion of the population is at least p is 1 − α, if
Px()+≥kuσμ+=σα1− . (A.5)
...


NORME ISO
INTERNATIONALE 16269-6
Deuxième édition
2014-01-15
Interprétation statistique des
données —
Partie 6:
Détermination des intervalles
statistiques de dispersion
Statistical interpretation of data —
Part 6: Determination of statistical tolerance intervals
Numéro de référence
©
ISO 2014
DOCUMENT PROTÉGÉ PAR COPYRIGHT
© ISO 2014
Droits de reproduction réservés. Sauf indication contraire, aucune partie de cette publication ne peut être reproduite ni utilisée
sous quelque forme que ce soit et par aucun procédé, électronique ou mécanique, y compris la photocopie, l’affichage sur
l’internet ou sur un Intranet, sans autorisation écrite préalable. Les demandes d’autorisation peuvent être adressées à l’ISO à
l’adresse ci-après ou au comité membre de l’ISO dans le pays du demandeur.
ISO copyright office
Case postale 56 • CH-1211 Geneva 20
Tel. + 41 22 749 01 11
Fax + 41 22 749 09 47
E-mail copyright@iso.org
Web www.iso.org
Publié en Suisse
ii © ISO 2014 – Tous droits réservés

Sommaire Page
Avant-propos .iv
Introduction .v
1 Domaine d’application . 1
2 Références normatives . 1
3 Termes, définitions et symboles . 1
3.1 Termes et définitions . 1
3.2 Symboles . 2
4 Méthodes . 3
4.1 Population normale avec une moyenne et une variance connues . 3
4.2 Population normale avec une moyenne inconnue et une variance connue . 4
4.3 Population normale avec une moyenne inconnue et une variance inconnue . 4
4.4 Populations normales avec des moyennes inconnues et des variances identiques
et inconnues . 4
4.5 Distribution continue quelconque de type inconnu . 4
5 Exemples . 5
5.1 Données pour les Exemples 1 et 2 . 5
5.2 Exemple 1: Intervalle statistique de dispersion unilatéral avec variance et
moyenne inconnues . 5
5.3 Exemple 2: Intervalle statistique de dispersion bilatéral avec moyenne inconnue et
variance inconnue . 6
5.4 Données pour les Exemples 3 et 4 . 6
5.5 Exemple 3: Intervalles statistiques de dispersion unilatéraux pour des populations
séparées avec variance commune inconnue . 8
5.6 Exemple 4: Intervalles statistiques de dispersion bilatéraux pour des populations
séparées ayant une variance commune inconnue . 8
5.7 Exemple 5: Distribution quelconque de type inconnu .10
Annexe A (informative) Facteurs k exacts pour les intervalles statistiques de dispersion pour une
distribution normale .12
Annexe B (informative) Formulaires pour les intervalles statistiques de dispersion .17
Annexe C (normative) Facteurs de la limite statistique de dispersion unilatérale, k (n; p; 1−α),
C
pour un écart-type de la population, σ, inconnu .20
Annexe D (normative) Facteurs de la limite statistique de dispersion bilatérale, k (n; m; p; 1−α),
D
pour un écart-type commun de la population, σ, inconnu (m échantillons) .26
Annexe E (normative) Intervalles statistiques de dispersion non paramétriques (Loi de
distribution indéterminée) .41
Annexe F (informative) Algorithmes de détermination des facteurs pour des intervalles
statistiques de dispersion bilatéraux paramétriques .43
Annexe G (informative) Construction d’un intervalle statistique de dispersion non paramétrique
pour un type de distribution quelconque .45
Bibliographie .47
Avant-propos
L’ISO (Organisation internationale de normalisation) est une fédération mondiale d’organismes
nationaux de normalisation (comités membres de l’ISO). L’élaboration des Normes internationales est
en général confiée aux comités techniques de l’ISO. Chaque comité membre intéressé par une étude
a le droit de faire partie du comité technique créé à cet effet. Les organisations internationales,
gouvernementales et non gouvernementales, en liaison avec l’ISO participent également aux travaux.
L’ISO collabore étroitement avec la Commission électrotechnique internationale (CEI) en ce qui concerne
la normalisation électrotechnique.
Les procédures utilisées pour élaborer le présent document et celles destinées à sa mise à jour sont
décrites dans les Directives ISO/CEI, Partie 1. Il convient, en particulier de prendre note des différents
critères d’approbation requis pour les différents types de documents ISO. Le présent document a été
rédigé conformément aux règles de rédaction données dans les Directives ISO/CEI, Partie 2, www.iso.
org/directives.
L’attention est appelée sur le fait que certains des éléments du présent document peuvent faire l’objet de
droits de propriété intellectuelle ou de droits analogues. L’ISO ne saurait être tenue pour responsable
de ne pas avoir identifié de tels droits de propriété et averti de leur existence. Les détails concernant les
références aux droits de propriété intellectuelle ou autres droits analogues identifiés lors de l’élaboration
du document sont indiqués dans l’Introduction et/ou sur la liste ISO des déclarations de brevets reçues,
www.iso.org/patents.
Les éventuelles appellations commerciales utilisées dans le présent document sont données pour
information à l’intention des utilisateurs et ne constituent pas une approbation ou une recommandation.
Le comité chargé de l’élaboration du présent document est l’ISO/TC 69, Application des méthodes
statistiques.
Cette deuxième édition de l’ISO 16269 annule et remplace la première édition (ISO 16269-6:2005), qui a
fait l’objet d’une révision technique.
L’ISO 16269 comprend les parties suivantes, présentées sous le titre général Interprétation statistique
des données:
— Partie 4: Détection et traitement des valeurs aberrantes
— Partie 6: Détermination des intervalles statistiques de dispersion
— Partie 7: Médiane — Estimation et intervalles de confiance
— Partie 8: Détermination des intervalles de prédiction
iv © ISO 2014 – Tous droits réservés

Introduction
Un intervalle statistique de dispersion est un intervalle estimé, d’après un échantillon, pour lequel
il est possible d’affirmer au niveau de confiance 1 − α, par exemple 0,95, qu’il contient au moins une
proportion spécifiée p d’individus de la population. Les limites d’un intervalle statistique de dispersion
sont appelées limites statistiques de dispersion. Le niveau de confiance 1 − α est la probabilité selon
laquelle un intervalle statistique de dispersion construit de la manière spécifiée contiendra au moins
une proportion p de la population. Inversement, la probabilité que cet intervalle contiendra moins que la
proportion p de la population est α. La présente partie de l’ISO 16269 décrit les intervalles statistiques
de dispersion unilatéraux et les intervalles statistiques de dispersion bilatéraux; un intervalle unilatéral
est construit avec une limite inférieure ou une limite supérieure tandis qu’un intervalle bilatéral est
construit avec une limite supérieure et une limite inférieure.
Un intervalle statistique de dispersion dépend d’un niveau de confiance 1 − α et d’une proportion donnée
p de la population. Le niveau de confiance d’un intervalle statistique de dispersion est bien défini à partir
d’un intervalle de confiance pour un paramètre. L’expression de la confiance d’un intervalle de confiance
est la proportion des cas où l’intervalle de confiance contient la valeur vraie du paramètre, une proportion
de 1 − α des cas, dans une longue série d’échantillons aléatoires répétés dans des conditions identiques.
De la même manière, l’expression de la confiance d’un intervalle statistique de dispersion indique qu’au
moins une proportion p de la population est contenue dans l’intervalle dans une proportion de 1 − α
des cas d’une longue série d’échantillons aléatoires répétés dans des conditions identiques. Ainsi, si
nous considérons la proportion établie p de la population comme un paramètre, l’idée sous-jacente aux
intervalles statistiques de dispersion est similaire à l’idée sous-jacente aux intervalles de confiance.
Les intervalles statistiques de dispersion sont fonction des observations de l’échantillon, c’est-à-dire
des statistiques, et leurs valeurs seront généralement différentes pour des échantillons différents. Il est
nécessaire que les observations soient indépendantes pour que les méthodes indiquées dans la présente
partie de l’ISO 16269 soient valables.
Deux types d’intervalles statistiques de dispersion sont définis dans la présente partie de l’ISO 16269:
paramétrique et non paramétrique. L’approche paramétrique se fonde sur l’hypothèse selon laquelle
la caractéristique étudiée dans la population est distribuée selon une loi normale; ainsi, si l’hypothèse
de normalité est avérée, le niveau de confiance avec lequel l’intervalle statistique de dispersion calculé
contient au moins une proportion p de la population ne peut être que de 1 − α. Pour les caractéristiques
distribuées selon une loi normale, l’intervalle statistique de dispersion est déterminé à l’aide de l’un des
formulaires A, B et C donnés dans l’Annexe B.
La présente partie de l’ISO 16269 ne traite pas des méthodes paramétriques s’appliquant à des
distributions autres qu’une loi normale. Si des écarts par rapport à la normalité sont suspectés dans
la population, des intervalles statistiques de dispersion non paramétriques peuvent être construits. La
procédure de détermination d’un intervalle statistique de dispersion pour une distribution continue
quelconque est indiquée dans le formulaire D de l’Annexe B.
Les limites statistiques de dispersion abordées dans la présente partie de l’ISO 16269 peuvent être
utilisées pour comparer la capabilité naturelle d’un processus avec une ou deux limites de spécification
données, soit une limite supérieure, U, soit une limite inférieure, L, ou encore les deux, dans la gestion
statistique d’un processus.
Au-dessus de la limite de spécification supérieure, U, il y a la proportion de non-conformes supérieure,
p (ISO 3534-2:2006, 2.5.4), et en dessous de la limite de spécification inférieure, L, il y a la proportion
U
de non conformes inférieure, p (ISO 3534-2:2006, 2.5.5). La somme p + p = p est appelée proportion
L U L t
de non conformes totale (ISO 3534-2:2006, 2.5.6). Entre les limites de spécification U et L, il y a la
proportion de conformes 1 − p .
t
Les idées sous-jacentes aux intervalles statistiques de dispersion sont plus répandues qu’on ne le pense
généralement, par exemple dans l’échantillonnage pour acceptation par mesures et dans la gestion
statistique d’un processus, comme on le verra dans les deux paragraphes qui suivent.
Dans l’échantillonnage pour acceptation par mesures, les limites U et/ou L sont connues, p , p ou p est
U L t
spécifié en tant que limite de qualité acceptable (LQA), α est implicite et le lot est accepté s’il y a au moins
un niveau de confiance implicite de 100(1 − α) % que la LQA n’est pas dépassée.
Dans la gestion statistique d’un processus, les limites U et L sont fixées à l’avance et les proportions
p , p et p sont soit calculées, lorsque la distribution est supposée connue, soit estimées. Il s’agit d’un
U L t
exemple d’application de contrôle qualité, mais il existe de nombreuses autres applications d’intervalles
[13]
statistiques de dispersion présentées dans de nombreux ouvrages tels que Hahn et Meeker.
Par contre, pour les intervalles statistiques de dispersion dont il est question dans la présente partie de
l’ISO 16269, le niveau de confiance pour l’estimateur de l’intervalle et la proportion de la distribution
située dans les limites de l’intervalle (correspondant à la proportion de conformes mentionnée ci-
dessus) sont fixés à l’avance, et les limites sont estimées. Ces limites peuvent être comparées à U et
à L. Ainsi, l’adéquation des limites de spécification données U et L peut être comparée aux propriétés
réelles du processus. Les intervalles statistiques de dispersion unilatéraux sont utilisés lorsque seule la
limite de spécification supérieure, U, ou la limite de spécification inférieure, L, est appropriée, tandis que
les intervalles bilatéraux sont utilisés lorsque les limites de spécification supérieure et inférieure sont
prises en compte simultanément.
La terminologie relative à ces limites et intervalles différents est confuse car les «limites de
spécification» étaient également autrefois appelées «limites de tolérance» (voir la norme de terminologie
ISO 3534-2:1993, 1.4.3, où ces deux termes, mais aussi le terme «valeurs limites», étaient utilisés comme
synonymes pour désigner ce concept). Dans la dernière révision de l’ISO 3534-2:2006, 3.1.3, seul le terme
«limites de spécification» a été conservé pour ce concept.
La première édition de la présente norme contenait des tableaux détaillés du facteur k pour les intervalles
de dispersion unilatéraux et bilatéraux lorsque la moyenne est inconnue, mais l’écart-type connu. Dans
cette deuxième édition de la norme, ces tableaux sont supprimés. A leur place, les facteurs k exacts sont
donnés à l’Annexe A lorsque l’un des paramètres de la loi normale est inconnu et l’autre paramètre est
connu.
La première édition de la présente norme considérait les intervalles statistiques de dispersion en se
fondant sur un seul échantillon de taille n. La présente édition considère les intervalles statistiques de
dispersion pour m populations avec le même écart-type en se fondant sur des échantillons de chacune
des m populations, chaque échantillon ayant la même taille n.
vi © ISO 2014 – Tous droits réservés

NORME INTERNATIONALE ISO 16269-6:2014(F)
Interprétation statistique des données —
Partie 6:
Détermination des intervalles statistiques de dispersion
1 Domaine d’application
La présente partie de l’ISO 16269 décrit des méthodes permettant d’établir les intervalles statistiques de
dispersion qui comprennent au moins une proportion spécifiée de la population à un niveau de confiance
spécifié. Les intervalles statistiques de dispersion unilatéraux et bilatéraux sont tous deux présentés
l’intervalle statistique de dispersion unilatéral étant caractérisé par une limite supérieure ou par une
limite inférieure, tandis que l’intervalle statistique de dispersion bilatéral possède à la fois une limite
supérieure et une limite inférieure. Deux méthodes sont exposées: une méthode paramétrique, lorsque
la caractéristique étudiée est distribuée selon une loi normale, et une méthode non paramétrique,
lorsqu’on ne sait rien de la distribution si ce n’est qu’elle est continue. Il existe également une procédure
permettant d’établir les intervalles statistiques de dispersion bilatéraux pour plus d’un échantillon
normal avec une variance identique mais inconnue.
2 Références normatives
Les documents suivants, en totalité ou en partie, sont référencés de manière normative dans le présent
document et sont indispensables pour son application. Pour les références datées, seule l’édition citée
s’applique. Pour les références non datées, la dernière édition du document de référence s’applique (y
compris les éventuels amendements).
ISO 3534-1:2006, Statistique — Vocabulaire et symboles — Partie 1: Termes statistiques généraux et termes
utilisés en calcul des probabilités
ISO 3534-2:2006, Statistique — Vocabulaire et symboles — Partie 2: Statistique appliquée
3 Termes, définitions et symboles
Pour les besoins du présent document, les termes et définitions donnés dans l’ISO 3534-1, l’ISO 3534-2
ainsi que les suivants s’appliquent.
3.1 Termes et définitions
3.1.1
intervalle statistique de dispersion
intervalle déterminé à partir d’un échantillon aléatoire de manière qu’un niveau de confiance spécifié
puisse être défini et que l’intervalle couvre au moins une proportion spécifiée de la population
échantillonnée
[SOURCE: ISO 3534-1:2006, 1.26]
Note 1 à l’article: Dans ce contexte, le niveau de confiance est la proportion à long terme des intervalles construits
de cette manière qui incluront au moins la proportion spécifiée de la population échantillonnée.
3.1.2
limite statistique de dispersion
statistique représentant une borne d’un intervalle statistique de dispersion
[SOURCE: ISO 3534-1:2006, 1.27]
Note 1 à l’article: Les intervalles statistiques de dispersion peuvent être
— soit unilatéraux (avec l’une des limites fixée à la limite naturelle de la variable aléatoire), auquel cas ils ont
une limite statistique de dispersion inférieure ou supérieure,
— soit bilatéraux, auquel cas ils ont une limite inférieure et une limite supérieure.
3.1.3
proportion de recouvrement
proportion des individus d’une population se trouvant à l’intérieur d’un intervalle statistique de
dispersion
Note 1 à l’article: Ce concept ne doit pas être confondu avec le concept de « facteur d’élargissement » utilisé dans le
[5]
Guide pour l’expression de l’incertitude de mesure (GUM).
3.1.4
population normale
population distribuée selon une loi normale
3.2 Symboles
Pour les besoins de la présente partie de l’ISO 16269, les symboles suivants s’appliquent.
k (n; p; 1 − α) facteur utilisé pour déterminer les limites des intervalles unilatéraux, c’est-à-dire x
1 L
ou x lorsque μ est connu et σ est inconnu
U
k (n; p; 1 − α) facteur utilisé pour déterminer les limites des intervalles bilatéraux, c’est-à-dire x
2 L
et x lorsque μ est connu et σ est inconnu
U
k (n; p; 1 − α) facteur utilisé pour déterminer les limites des intervalles unilatéraux, c’est-à-dire x
3 L
ou x lorsque μ est inconnu et σ est connu
U
k (n; p; 1 − α) facteur utilisé pour déterminer les limites des intervalles bilatéraux, c’est-à-dire x
4 L
et x lorsque μ est inconnu et σ est inconnu
U
k (n; p; 1 − α) facteur utilisé pour déterminer x ou x lorsque les valeurs de μ et σ sont inconnues
C L U
pour un intervalle statistique de dispersion unilatéral. Le suffixe C est choisi parce
que ce facteur k est tabulé dans l’Annexe C.
k (n; m; p; 1 − α) facteur utilisé pour déterminer x et x (i = 1,2,.,m; m ≥ 2) lorsque les valeurs des
D Li Ui
moyennes μ et la valeur de σ commun sont inconnues pour les m intervalles sta-
i
tistiques de dispersion bilatéraux. Le suffixe D est choisi parce que ce facteur k est
tabulé dans l’Annexe D.
n nombre d’observations dans l’échantillon (taille d’échantillon)
p proportion minimale de la population affirmée comme se trouvant à l’intérieur de
l’intervalle statistique de dispersion
u fractile d’ordre p de la loi normale centrée réduite
p
ème
x j valeur observée ( j = 1,2,.,n)
j
ème ème
x j valeur observée ( j = 1,2,.,n) of i valeur observée (i = 1,2,.,m)
ij
2 © ISO 2014 – Tous droits réservés

x valeur maximale des valeurs observées: x = max {x , x , …, x }
max max 1 2 n
x valeur minimale des valeurs observées: x = min {x , x , …, x }
min min 1 2 n
x limite inférieure de l’intervalle statistique de dispersion
L
x limite supérieure de l’intervalle statistique de dispersion
U
n
moyenne de l’échantillon, x = x
j
x ∑
n
j = 1
n
ème
moyenne du i échantillon ()im= 12,,., , x = x
x
ii∑ j
i
n
j=1
n n
 
 
n xx−
∑∑j j
n
 
s écart-type de l’échantillon,
1 j ==11j
 
s = xx− =
()
∑ j
n−11nn−
()
j = 1
n
ème
écart-type du i échantillon im= 12,,., , s = ()xx−
()
s ii∑ ji
i
()n−1
j=1
écart-type de l’échantillon établi sur un ensemble de données;
s
P
m n m
1 1
2 2
s = ()xx−= s
P ∑∑∑ ij i i
mn()−1 m
i==j= i 1
1 − α niveau de confiance de l’affirmation selon laquelle la proportion de la population se
trouvant à l’intérieur de l’intervalle de dispersion est supérieure ou égale au niveau
spécifié p
μ moyenne de la population
ème
μ moyenne de la i population (i = 1,2,.,m)
i
σ écart-type de la population
4 Méthodes
4.1 Population normale avec une moyenne et une variance connues
Lorsque les valeurs de la moyenne, μ, et de la variance, σ , d’une population distribuée selon une loi
normale sont connues, la distribution de la caractéristique étudiée est complètement déterminée. Il y a
exactement une proportion p de la population:
a) à droite de x = μ −μ × σ (intervalle unilatéral inférieur);
L p
b) à gauche de x = μ + μ × σ (intervalle unilatéral supérieur);
U p
c) entre x = μ −μ × σ et x = μ + μ × σ (intervalle bilatéral).
L (1+p)/2 U (1+p)/2
Dans les équations ci-dessus, μ est le fractile d’ordre p de la loi normale centrée réduite.
p
NOTE Dans la mesure où l’on sait que ces déclarations sont justes, elles sont faites au niveau de confiance
de 100 %.
4.2 Population normale avec une moyenne inconnue et une variance connue
Lorsque l’un ou les deux paramètres de la loi normale sont inconnus, mais estimés à partir d’un
échantillon aléatoire, des intervalles ayant des propriétés similaires à ceux du paragraphe 4.1 peuvent
encore être construits. Supposons, par exemple, que la moyenne soit inconnue, mais que la variance soit
connue. Il est alors possible de déterminer une constante k telle que l’intervalle, compris entre
xx=−kxσσ et =+xk
LU
contienne au moins une proportion p de la population au niveau de confiance spécifié de 1 − α. Noter
qu’il existe deux distinctions importantes par rapport à la situation de 4.1 où les paramètres étaient
supposés connus. Premièrement, lorsqu’un ou plusieurs paramètres sont estimés, l’intervalle contient
au moins une proportion p de la population, mais pas exactement une proportion p de la population.
Deuxièment, lorsque des paramètres sont estimés, la déclaration est vraie uniquement au niveau de
confiance préalablement spécifié de 1 − α. Le facteur k dans l’expression des limites ci-dessus dépend
des paramètres inconnus de la loi normale, de la proportion p, du niveau de confiance 1−α et du nombre
d’observations dans l’échantillon aléatoire lorsque l’un des paramètres de la loi normale est inconnu et
que l’autre paramètre est connu, l’Annexe A donne les facteurs k exacts.
4.3 Population normale avec une moyenne inconnue et une variance inconnue
Les formulaires A et B, donnés dans l’Annexe B, sont applicables lorsque la moyenne et la variance
de la population normale sont inconnues. Le formulaire A s’applique aux cas unilatéraux tandis que
le formulaire B s’applique aux cas bilatéraux. Le formulaire A est utilisé avec les tableaux de facteurs
k de l’Annexe C, ou en variante en utilisant la formule exacte relative au facteur k donnée en A.5. Le
formulaire B est utilisé avec les facteurs k donnés dans la première colonne des tableaux de l’Annexe D.
Les détails relatifs au calcul des facteurs k de l’Annexe D sont donnés à l’Annexe F.
4.4 Populations normales avec des moyennes inconnues et des variances identiques et
inconnues
Le formulaire C, donné dans l’Annexe B, est applicable lorsque les moyennes et les variances des
populations normales sont inconnues. En outre, les variances sont supposées être identiques pour toutes
les populations étudiées, auquel cas on parle de variance commune.
4.5 Distribution continue quelconque de type inconnu
Si la caractéristique étudiée est une variable d’une population de forme inconnue, alors un intervalle
statistique de dispersion peut être déterminé à partir des statistiques d’ordre x d’un échantillon de
(i)
n observations aléatoires indépendantes. La méthode indiquée dans le formulaire D, utilisée conjointement
avec les Tableaux E.1 et E.2, donne les étapes permettant de déterminer la taille d’échantillon requis sur
la base des statistiques d’ordre à utiliser, du niveau de confiance souhaité et du contenu souhaité.
NOTE 1 Les intervalles statistiques de dispersion dans lesquels le choix des bornes (sur la base des statistiques
d’ordre) ne dépend pas de la population échantillonnée sont appelés intervalles statistiques de dispersion non
paramétriques.
NOTE 2 La présente Norme internationale ne préconise pas de méthode pour les distributions d’un type connu
autre qu’une loi normale. Toutefois, si la distribution est continue, la méthode non paramétrique peut être utilisée.
Une sélection de références à la littérature scientifique pouvant aider à déterminer les intervalles de dispersion
pour d’autres distributions est également fournie à la fin de ce document.
4 © ISO 2014 – Tous droits réservés

5 Exemples
5.1 Données pour les Exemples 1 et 2
Les formulaires A à B, donnés dans l’Annexe B, sont illustrés par les Exemples 1 et 2 à l’aide des valeurs
[2]
numériques de l’ISO 2854:1976, Article 2, paragraphe 1 des remarques introductives, Tableau X, fil 2:
12 mesures de la charge de rupture du fil en coton. Il convient de noter que le nombre d’observations,
n = 12, indiqué pour ces exemples, est considérablement plus faible que celui recommandé dans
[1]
l’ISO 2602. L’unité de mesure pour exprimer les données numériques et les calculs dans les différents
exemples est le centinewton (voir Tableau 1).
Tableau 1 — Données pour les Exemples 1 et 2
Valeurs en centinewtons
x 228,6 232,7 238,8 317,2 315,8 275,1 222,2 236,7 224,7 251,2 210,4 270,7
Ces mesures proviennent d’un lot de 12 000 bobines, issu d’une même série de fabrication, emballées
dans 120 boîtes contenant chacune 100 bobines. Douze boîtes de ce lot ont été prélevées au hasard et
une bobine a été prise au hasard dans chacune de ces boîtes. Des éprouvettes de 50 cm de long ont été
découpées dans le fil de ces bobines, à environ 5 m de l’extrémité libre. Les essais proprement dits ont
été réalisés sur les parties centrales de ces éprouvettes. Des informations antérieures permettent de
penser raisonnablement que les charges de rupture mesurées dans ces conditions ont une distribution
pratiquement normale. Il est démontré, dans l’ISO 2854:1976, que les données ne contredisent pas
l’hypothèse d’une distribution normale.
En utilisant le test graphique par diagrammes à surfaces pour les valeurs aberrantes décrit dans
l’ISO 16269-4, il est également possible de conclure qu’aucune des valeurs de données ne peut être
déclarée comme une valeur aberrante avec un niveau de signification α = 0,05.
Les données du Tableau 1 donnent les résultats suivants:
Taille d’échantillon: n = 12
n
Moyenne de l’échantillon: x ==x 3 024,1/12= 252,01
∑ j
n
j = 1
n n
 
 
n xx−
∑∑j j
 
j ==11j 166 772,27
 
Écart-type de l’échantillon: s = = = 1 263,426 333= 5,545
nn−1 12×11
()
La présentation formelle des calculs sera donnée dans l’Exemple 1 en utilisant le formulaire A de
l’Annexe B (intervalle unilatéral, variance et moyenne inconnues).
5.2 Exemple 1: Intervalle statistique de dispersion unilatéral avec variance et moyenne
inconnues
Une limite x est requise de manière à ce qu’il soit possible d’affirmer au niveau de confiance
L
1 − α = 0,95 (95 %) qu’au moins 0,95 (95 %) des charges de rupture des éléments du lot, mesurées
dans les mêmes conditions, sont supérieures à x . Les résultats sont présentés de manière détaillée ci-
L
dessous.
Détermination de l’intervalle statistique de dispersion de la proportion p:
a) intervalle unilatéral « à droite » (intervalle statistique de dispersion unilatéral inférieur)
Valeurs déterminées:
b) proportion de la population choisie pour l’intervalle statistique de dispersion: p = 0,95
c) niveau de confiance choisi: 1 − α = 0,95
d) Taille d’échantillon: n = 12
Valeur du facteur de dispersion, provenant du Tableau C.2: kn(;p;)12−=α ,736 4
C
Calculs:
n
x ==x
j

252,01
n
j = 1
n n
 
 
n xx−
∑∑j j
 
j ==11j
 
s =
= 35,545
nn−1
()
kn(;ps;)19−×α = 7,265 3
C
Résultats: intervalle unilatéral «à droite» (intervalle statistique de dispersion unilatéral inférieur)
L’intervalle de dispersion qui contiendra au moins une proportion p de la population au niveau de
confiance 1 − α a une limite inférieure:
xx=−kn(;ps;)1−×α = 154,7
L C
5.3 Exemple 2: Intervalle statistique de dispersion bilatéral avec moyenne inconnue et
variance inconnue
Supposons qu’il soit requis de calculer les limites x et x de manière qu’il soit possible d’affirmer au
L U
niveau de confiance 1 − α = 0,95 que dans une proportion du lot au moins égale à p = 0,90 (90 %), la
charge de rupture est comprise entre x et x .
L U
Le Tableau D.4 donne pour un nombre d’échantillon, m = 1 ayant pour taille n = 12, le facteur de la limite
statistique de dispersion bilatérale, k :
D
kn(;11;;p −=α),2 6703
D
d’où
xx=−kn(;11;;ps−×α ),=−252 01 2,,6703×=35 545 157,0
LD
xx=+kn(;11;;ps−×α ),=+252 01 2,,6703×=35 545 347,0
UD
5.4 Données pour les Exemples 3 et 4
Supposons que le pourcentage de solides dans chacun de quatre lots de levure de bière humide, provenant
chacun d’un fournisseur différent, doive être déterminé. Les pourcentages des quatre lots sont distribués
6 © ISO 2014 – Tous droits réservés

selon une loi normale avec des moyennes inconnues μ i = 1,2,3,4. Sur la base de l’expérience acquise
i
avec ces fournisseurs, il peut être supposé que les variances sont identiques. Un essai pour les données
suivantes ne donne aucune raison de supposer le contraire. Les données sont donc supposées avoir
une variance commune σ . Le chercheur souhaite déterminer les intervalles statistiques de dispersion
bilatéraux pour les pourcentages de solides dans chaque lot.
[14]
Les valeurs relatives à des échantillons aléatoires de taille n = 10 prélevés dans les quatre lots sont
indiquées dans le Tableau 2:
Tableau 2 — Données pour les Exemples 3 et 4
Valeurs en pourcentage
j/i 1 23456789 10
1 20 18 16 21 19 17 20 16 19 18
2 19 14 17 13 10 16 14 12 15 11
3 11 12 14 10 8 10 13 9 12 8
4 10 7 11 96 11 8 12 13 14
ème ème
Noter que la j valeur du i échantillon est désignée par: x .
ij
Les résultats suivants sont obtenus:
Taille des échantillons: n = 10
Nombre d’échantillons: m = 4
Moyennes des échantillons de chacun des quatre lots:
x ==184 /,10 18 4 ;   x ==141/,10 14 1 ;    x ==107 /,10 10 7 ;   x ==101/,10 10 1
1 2 3 4
Variances des échantillons de chacun des quatre lots:
2 2
n  n  n  n 
2 2
   
nx − x nx − x
1jj1 2jj2
   
j=1 j=1 264 j=1 j=1 689
2   2  
s = = = 2,9333 ;   s = = = 7,655 6
1 2
nn()−1 10×9 nn()−1 10×9
2 2
n n n n
   
2 2
   
nx − x nx − x
3jj3 4jj4
   
j=1 j=1 381 j=1 j=1 609
2   2  
s = = = 4,2333 ;   s = = = 6,766 7
3 4
nn()−1 10×9 nn()−1 10×9
Écart-type de l’échantillon établi sur l’ensemble des données:
m
s ==s (,2,9333++7 655 64,,23336+=766 72),323 2
P i

m 4
i=1
Degrés de liberté de l’écart-type établi sur l’ensemble des données:
f = m(n − 1) = nm − m = 36
∑∑ ∑∑
∑∑ ∑∑
5.5 Exemple 3: Intervalles statistiques de dispersion unilatéraux pour des populations
séparées avec variance commune inconnue
Supposons que l’on souhaite calculer les intervalles statistiques de dispersion inférieurs pour les
quatre fournisseurs, c’est-à-dire que l’on souhaite calculer les intervalles qui contiennent au moins une
proportion p pour tous les fournisseurs. Le Tableau C ne peut pas fournir la réponse, mais les intervalles
ont la même forme que celle indiquée dans l’Exemple 1, à savoir une constante multipliée par l’écart-type
estimé soustrait de la moyenne estimée:
xx=−kn(;fp;;1−×α ),s
LPii i
ème
où la constante k(n ;f;p;1−α) dépend de la taille du i échantillon et des degrés de liberté de l’écart-
i
type établi sur l’ensemble des données. L’expression relative à la constante est issue de l’Article A.5, voir
l’Équation (A.14):
kn(;fp;;1−=α )(tfnu ;),
i 1−α ip
n
i
où tf(;nu ) désigne le quantile d’ordre 1 − α de la loi de t non centrée avec le paramètre de
1−α ip
décentrement nu et f degrés de liberté. La loi de t non centrée et en particulier ses quantiles sont
ip
disponibles dans des logiciels statistiques. Supposons qu’une proportion p = 0,95 et un niveau de
confiance 1 − α = 0,95 soient souhaités. Dans ce cas, n = 10 et f = m(n − 1) = nm − m = 36, et la constante
i
est donc
kt(;10 36;;0,95 0,95)(==10×1,6449;36),2,3471
0,95
où le quantile d’ordre 0,95 de la loi normale centrée réduite u = 1,6449 est inséré.
0,95
Les valeurs données dans les tableaux de l’Annexe C sont les cas particuliers où les degrés de liberté sont
égaux à la taille de l’échantillon moins 1, c’est-à-dire les degrés de liberté de l’écart-type sur la base d’un
seul échantillon de taille n
kn(;pk;)11−=αα(;nn−−;;pt1 )(=−nu ;)n 1 ,
C 1−α p
n
c’est-à-dire le cas particulier où le nombre de degrés de liberté de l’estimation de la variance est n − 1
Il s’ensuit que les limites statistiques de dispersion unilatérales calculées pour l’ensemble des quatre
lots sont les suivantes:
Premier lot:   xx=−kn ;;;ναps11− ×= 84,,02−×3471 2,,3232= 12 94
()
LP11 1
Deuxième lot:  xx=−kn ;;;ναps11− ×= 41,,02−×3471 2,,3232= 864
()
LP22 2
Troisième lot:  xx=−kn;;;ναps11− ×= 07,,02−×3471 2,,3232= 466
()
LP33
Quatrième lot:  xx=−kn;;;ναps11− ×= 01,,02−×3471 2,,3232= 406
()
LP44
Si les limites statistiques de dispersion supérieures étaient requises, les mêmes grandeurs devraient
être combinées, excepté que le produit de la constante par l’erreur-type devrait être ajouté à la moyenne
estimée.
5.6 Exemple 4: Intervalles statistiques de dispersion bilatéraux pour des populations
séparées ayant une variance commune inconnue
Cas 1 — Calcul pour tous les lots (m = 4)
8 © ISO 2014 – Tous droits réservés

Le Tableau D.5 donne, pour n = 10, m = 4, f = m(n − 1) = 4(10 − 1) = 36, p = 0,95 et 1 − α = 0,95, la valeur
du facteur statistique de dispersion bilatéral pour une variabilité commune inconnue σ , comme étant
égale à
kn(;mp;;12−=α ),5964.
D
Il s’ensuit que les limites statistiques de dispersion bilatérales calculées simultanément pour l’ensemble
des lots sont les suivantes:
Premier lot:
xx=−kn;;mp;,11−α ×=s 8402−×,,5964 2 3232= 12,36
()
LD11 P
xx=+kn;;mp;,11−α ×=s 8402+×,,5964 2 3232= 24,44
()
UD11 P
Deuxième lot:
xx=−kn;;mp;,11−α ×=s 4102−×,,5964 2 3232= 80, 6
()
LD22 P
xx=+kn;;mp;,11−α ×=s 4102+×,,5964 2 3232= 20,14
()
UD22 P
Troisième lot:
xx=−kn;;mp;,11−α ×=s 0702−×,,5964 2 3232= 46, 6
()
LD33 P
xx=+kn();;mp;,11−α ×=s 0702+×,,5964 2 3232= 16,74
UD33 P
Quatrième lot:
xx=−41kn;;mp;,−α ×=s 10 10−×2,,5964 2 3232= 40, 6
()
LD4 P
xx=+kn;;mp;,11−α ×=s 0102+×,,5964 2 3232= 16,14
()
UD44 P
NOTE Les limites inférieures ont été arrondies par défaut et les limites supérieures ont été arrondies par
excès (à la deuxième décimale près) pour maintenir l’intégrité des formulations de la confiance.
Cas 2 — Calcul individuel pour chaque lot (m = 1)
Il est possible de calculer ces limites de dispersion séparément pour chaque lot. Pour n = 10, m = 1, f =
m(n − 1) = 1(10 − 1) = 9, p = 0,95 et 1 − α = 0,95, la valeur du facteur statistique de dispersion bilatéral
pour une variabilité commune inconnue σ est égale à
k (;10 10;,95;,0953),= 3935
D
et peut être trouvée à l’Annexe D (Tableau D.4).
Écarts-types d’échantillon pour les quatre lots:
2 2
ss== 2,,9333 = 1 7127 ;   ss== 7,,6556 = 2 7669
11 22
2 2
ss== 4,,2333 = 2 0575 ;   ss== 6,,7667 = 2 6013
33 44
Il s’ensuit que les limites statistiques de dispersion bilatérales sont les suivantes:
Premier lot:
xx=−kn(;ms;,0950;,95)(×= xk−×10;;10,;95 09,)5 s
LD11 11 D 1
= 18,40−−×3,,3935 1 7127 = 12,58
xx=+kn(;mp;;11−×α )(sx=+ks01;;09,;50,)95 ×
UD11 11 D 1
=+18,,40 339935×=1,,7127 24 22
Deuxième lot:
xx=−kn(;mp;;11−×αα)(sx=−kp01;; ;)1−×s
LD22 22 D 2
=−14,,10 3 3935×22,,7669 = 470
xx=+kn(;mp;;11−×αα)(sx=+kp01;; ;)1−×s
UD22 22 D 2
=+14,,10 3 3935×22,,7669 = 23 50
Troisième lot:
xx=−kn(;mp;;11−×αα)(sx=−kn;;mp;)−×s
LD33 33 D 3
=−10,,70 3 394×2,00575 = 37, 1
xx=+kn(;mp;;11−×αα)(sx=+kp01;; ;)1−×s
UD33 33 D 3
=+10,,70 3 3935×22,,0575 = 17 69
Quatrième lot:
xx=−kn(;mp;;11−×αα)(sx=−kp01;; ;)1−×s
LD44 44 D 4
=−10,,10 3 3935×22,,6013 = 127
xx=+kn(;mp;;11−×αα)(sx=+kp01;; ;)1−×s
UD44 44 D 4
=+10,,10 3 3935×22,,6013 = 18 93
Lorsque l’on compare les résultats obtenus dans les deux cas, il est possible de déclarer que les intervalles
statistiques de dispersion pour les lots 2, 3 et 4 sont sensiblement plus petits dans le Cas 1 que dans le
Cas 2. Cependant, l’intervalle statistique de dispersion pour le premier lot n’est que légèrement plus
grand dans le Cas 2. Ceci s’explique par le fait que la constante k est plus petite dans le Cas 1 que dans
D
le Cas 2 parce que les degrés de liberté sont plus grands dans le Cas 1. Le lot 1 a le plus faible écart-type
estimé et cela compense l’augmentation de la constante k .
D
Il est possible de conclure que les intervalles statistiques de dispersion calculés simultanément pour
plusieurs populations peuvent donner des intervalles plus courts que les intervalles statistiques de
dispersion calculés séparément pour chaque échantillon aléatoire, à condition que les populations
normales sous-jacentes aient la même variance. Cette propriété intéressante découle du fait qu’en
moyenne, l’estimation de la variance calculée à partir de plusieurs échantillons aléatoires est «meilleure»
que l’estimation calculée à partir d’un seul échantillon aléatoire parce que cette dernière est fondée sur
un plus petit nombre d’observations.
5.7 Exemple 5: Distribution quelconque de type inconnu
Supposons que nous disposions d’un échantillon, x , x ,…, x , d’observations aléatoires indépendantes
1 2 n
sur une population (continue, discrète ou combinée) et que ses statistiques d’ordre soient x ≤ x ≤ …
(1) (2)
≤ x .
(n)
Il est possible de déterminer la taille d’échantillon nécessaire pour atteindre un niveau de confiance
ème
d’au moins 100(1 − α) % pour qu’au moins 100p % de la population soient compris entre la v plus
ème
petite observation (c’est-à-dire statistique d’ordre x ) et la w plus grande observation (c’est-à-dire
(v)
statistique d’ordre x ).
(n–w+1)
1) Déterminer la taille d’échantillon n nécessaire pour atteindre un niveau de confiance d’au
moins 95 % pour qu’au moins 99 % des valeurs mesurées sur la population soient comprises entre les
ème
observations minimale et maximale, c’est-à-dire entre la première (v = 1) et la n (w = 1) statistiques
d’ordre de l’échantillon.
10 © ISO 2014 – Tous droits réservés

Sur la base de la description ci-dessus, v + w = 2, p = 0,99 et 1 − α = 0,95. La taille minimale d’échantillon
déterminé à partir du Tableau E.1 est de 473 (le niveau de confiance réel est de 95,020 %). Quelques
exemples sont donnés ci-dessous.
2) Déterminer la taille d’échantillon n nécessaire pour atteindre un niveau de confiance d’au moins
95 % pour qu’au moins 95 % des valeurs mesurées sur la population soient supérieures ou égales à la
statistique d’ordre minimale de l’échantillon (v = 1 et w = 0).
Sur la base de la description ci-dessus, v + w = 1, p = 0,95 et 1 − α = 0,95. La taille minimale d’échantillon
déterminé à partir du Tableau E.1 est de 59 (le niveau de confiance réel est de 95,151 %).
3) Déterminer la taille d’échantillon n nécessaire pour atteindre un niveau de confiance d’au moins
95 % pour qu’au moins 99 % des individus de la population soient acceptables avec au plus un individu
non conforme admissible dans l’échantillon.
Sur la base de la description donnée à l’Annexe G, v + w = 2 (v + w − 1 = 1 car 1 est le nombre maximal
admissible d’individus non conformes dans l’échantillon), p = 0,99 et 1 − α = 0,95. La taille minimale
d’échantillon déterminé à partir du Tableau E.1 est de 473 (le niveau de confiance réel est de 95,020 %).
Noter que ce résultat est identique à celui obtenu dans le premier exemple de ce paragraphe.
4) Supposons que la distribution de X ait de longues queues (c’est-à-dire qu’elle produise des valeurs
positives et négatives extrêmes occasionnelles) et que des mesures supplémentaires soient jugées
nécessaires pour faire en sorte que l’intervalle statistique de dispersion obtenu ait une longueur utile.
L’expérimentateur décide d’exclure des statistiques d’ordre inférieures et supérieures de telle sorte que
l’inte
...

Questions, Comments and Discussion

Ask us and Technical Secretary will try to provide an answer. You can facilitate discussion about the standard in here.

Loading comments...