Professional Documents
Culture Documents
929
INTERNATI~NAL ISO
STAN~AR' 5725-2
First edlticn
1994-12-15
...~,
~'1..t.,;¡&"::'j i
>
;,r",~-'\l:;;t~"~""~ .A
J:w .. j jo ~~ ¿~~¿ .0.0.::..
Reference number
lS0 5'725-2:1994IE)
ISO 5725-2:1994(E)
Contents
Page
Scope ..... ,............ " ..... ,....... ,...... ,........... ,,, .. ,..,.... ,............ ,.............. ,
J Definitíons ............................................ 2
" Estimates of the parameters in the basic model ...... " ...... " ..... 2
6.1 Panel ................ ,.............. ,..... ,...... ,.... " ...................... ,...... ,.... , 5
6.5 Operators 6
7.1 The report to, and the decisions te be taken by, the panel 20
() ISO 1994
Al! nghts reserved. Unless otherwise- specified. no par! 01 thls pubhcatlOn rnlly be reproduced
O(uti{¡zed 10 any fO(nl Q( b', any means, eleCtIOI1IC o' mechanu:;al. Includll19 phOtOCOpyll1g "na
Pnnted ,n SWltzerfand
ji
Annexes
data) ..................................................................................... 32
e Bibliography 42
íii
-
,
Foreword
ISO Uhe Internatíonal Organization for Standardizationt is a worldwide
federation of national standards bodies (ISO member bodiesL The wor!<
of preparíng Intemational Standards is normally carríed out through ISO
technícal committees. Each member body interested in a subject for
which a technical committee has been establíshed has the nght te be
represented on that committee. International o rganizatlons, governmental
and non-governmental. ín liaison wíth ISO, also take part in the work. ISO
collaborates c!osely with me International Electrotechnical Commíssíon
(lEC) on all matters of electrotechnícal standardízatíon.
Draft International Stanpards a<lopted by the technical committees are
crrculated to the member bodies for votíng. Publícation as sn International
Standard requires approval by a1 leaSl 75 % of the member bodies casting
a vote.
ISO 5725 consists of the following parts, under the general title Accuracy
(trueness and precísion) of meBsurement methods and resu{ts:
- Pan 2.' Basje method for the determínatíon of rep¡;¡atabilíty Bnd re
producibility of a st~ndard measurement method
Parts 1 to 6 of ISO 5725 together cancel and replace ISO 5725: 1986,
which has been extended to covar trueness (in addítíon to pmcision) and
intermedia,€! precisíon condítions (in addítion to repeatabilíty and repro
ducibility condítions).
Annex A forms an íntegra I part of this part of ISO 5725. Annexes B and
C are tor information only.
IV
Introduction
0.1 ISO 5725 uses two terms u trueness" and "precision" to describe
N
the accuracy of a measurement method. "Trueness refers ta the close
ness of agreement between the arithmetic mean of a large number Qf test
results 8nd tlle trua or accepted reference value. "Precision" refers to the
c10seness of agreement between test results.
0.3 This part of ISO 5725 i5 concerned solely with estimating by mean::;
of the repeatabílity standard deviation and reproducibility standard devi
atíon. Although other types 01 experiment (such as the split-1evel exper
ímenO are usad in certain clrcumstances for the estimatíon of precision.
they are not dealt with ín this part of ISO 5725 but rather are the subject
01 ISO 5725-5 Nor does thís part 01 ISO 5725 considef any other meas
ures of precisíon intermedia te between the two principal measures; those
are 1he subject of ISO 5725-3.
...., .
Tue Jan 16 16:58:57 1996
Part 2:
Basic method for the determination of repeatability and
reproducibility of a standard measurement method
el when the use of a balanced uníform-Ievel layout B is the laboratory component of bias under re
is acceptable. peatability conditions:
ISO 5725-1 :1994, Accuracv (trueness and precision) s~¡ is the estímate of the within-Iaboratory vari
of measurement methods and resufts - Pan 1: anca;
'-- General principies and defrnitions.
8,2 is the arithmetic mean of s~ and is the estí
mate of the repeatability variance; this arith
metic mean is taken over all those
3 Dufinitions rabaratoríes taking part in the aCCUfacy ex
periment which remain after outliers haya
For the purposes of this part of ISO 5725, the defi been excluded:
nitions given in ISO 3534-1 and in ISO 5725-1 apply.
.. (1)
4 Estimates of the parameters in the
basic model
5 Requirements for a precision
4.1 The procedures gilien In this pal1 of ¡SO 5725 experiment
are basad on the statistical modal given in dause 5
of ISO 5725-1 :1994 and elaborated upon in subclausa
1.2 of ISO 5725-1:1994. In panicular, these pro 5.1 Layout of the experiment
cedures are based on equations (2) to (S} of clause 5
of ISO 5725-' :1994.
5,1.1 In tha layout usad in ¡he basic method, sam
The model is pies 1rom q batch es of materials, representing q díf
y=m+B+e ferent levels of the test. are sent to p laboratories
which each obtain exactly n replícate test resuits un
where, for the particular materíal tested. dar repeatability conditions at each of the q levels.
Thls type of experiment ís callad a balanced IJniform
In is the general mean (expectatíonl; level 8xperiment.
5.1.2 The performance of these measuremants shall tween the day the samples are receíved and the
be organizad and instructíons issued as follaws. day the measurements are performed.
al Any preliminary checkin9 of equipment shall be hl AII samples shall be clearly labelled with the name
as speclfied in the standard method, of the experiment ami a sample identification.
b) Each group of It measuremenls belonging to one 5.1.3 In 5. í.2 and elsewhere in this part of
level shall be carriad out under repeatability con ISO 5725, reference is meda to the operator. For
ditions, ¡,e. wíthin a short interval of time and by some measuremants, there may in fact be a team of
the sama operator, and without any íntermediate operatars, each of whom performs so me specific part
recalibration 01 the apparatus unless this is en in of the procedure, In such a case, the team shall be
tegral part oi performíng a measurement. regarded as "the operato(" and any crlange In the
team shall be regarded as providing a differen1 .. op
el It is essential that a group of n tests under re aratar" .
peatability conditions be performed independently
as it they were n tests on dífierent materials. As 5.1.4 In commercial practica, the test results may
a rule, how8'1er, the operator will know that be rcunded rather crudely, but in a precision exper·
he/she is testing identical matenal, but the point ¡ment test results shall be reported to at least one
should be stressed in the instructions that the more digit than specified in the standard method. If
whole purpose 01 the experiment is to determine the method doas not specify the number 01 digits, the
what differences in rasults can occur in actual rounding shall not be coarser than half the repeatabíl
testing. If lt 16 feared that. despite this warning, ity standard deviatlon estímate. When precision may
previous results ma,! lnfluence subsequent test depend on the level m, di-lferent degrees of roul'lding
results and 1hus the repeatability varianee, ít may be needed for dlfferent levels.
should be considered whether to use n separata
samples at each 01 the q levels, coded in such a 5.2 Recruitment of the laboratories
way that the oparator will not know which are the
replícates tor a given level. However, such a pro 5.2.1 The general principies regarding recruítment
cedure could cause problems in ensuring that re of the laboratoríes to participate in an interlaborator'l
peatability conditions will apply between experiment are given in 6.3 of ISO 5725-1: 1994, In
raplleates, This would on(y be possible if the enlistlng the cooperation of the requisite number oi
measuremems were of sur,;h a natUfe tnat all the laboratories, their responsibllitles shall be clearly
qn measurements could be performed wíthin a
stated. An example 01 a suitable enlistment question
short intsNal of time, naire is given in figure 1 .
d} It 15 not essential ,that all the q groups 01 11
5.2.2 Far ths purposes of this part Qf ISO 5725, a
measurements each be pe(formed strictly within
"Iaboratory" is considered to be a combin.3tion of the
a short interval; different groups of measurernents
operator, the equipment and the test slte, One test
may be carríed out on difierent days.
sita {al' laboratory in tha conventional sense) may thus
produce several "¡aboratorles" if it can provide several
e) Measurements of all q levels shall be peñormed
operators each with independent sets 01 equipment
. by one 8nd the sama operator and, in addition, the
and situations in which to perfoml the work.
n measurements at a gíven level shall be per
formed using the same equipment throughout.
5.3 Preparation of the materials
fl If in the courss of the measurements sn oparatar
should becoma una'lailable, aoother operatar may 5.3.1 A dlscussion of the poínts that need to be
the mea sureme nts, províded that the considered when selectíng materíals ter use in a pre
changa does not oecur within a group of n cision experiment 1s given in 6.4 of ISO 5725-'1 :1994,
measurements at one level but only occurs be
tween two of the q groups_ Any such change shall 5.3.2 When deciding on the quanbties of material to
be reported with 1he resulta. be provided, allowance shatl be maae tor accidental
spillage or el-rors in obtaining some test results vvhích
g) A time limit shall be gíven within which all may necessitate using extra material. The amount 01
measurements shall be completed. This may be material prepared shall be sufficíent to cover the ex
necessary to limit the time allowed to elapse be periment and allow an adequate stock in reserve.
1. OUf laboratory is wílling to partlcipate in rhe precision experiment for this stan
dard measurement method.
Having studied the method and having mede a fair appraisal of our capabilities and
facilities, we feel that we will be adequately preparad tor cooperative testing of this
method.
3. Comments
(Signed) ...................................................... .
(Companv or
laboratory} ................................................. .
5.3.3 It should be considered whether it is desirable 5.3.5 For the samples a1 each leveL n separate con
fo( so me laboratoríes to obtsin same prelíminary test tainers shall be usad tor each laboratory if there is any
results tor familiarizatíon with the measurement danger of the materíals deteriorating once the con
method befare obtaining the officlal test result and, if tainer has besn opened (e.9. by oxidation. by losing
so. whether addítional material (not precisíon exper vOlatile components, or with hygroscoplc material). In
íment samples) should be provided ior thlS purpose. the case oí unstable materiais, speciol instructíons on
storage and treatment shall be spec1fied. Precautions
may be needed to ensure that samples remaín iden
tical up to the time the measurements are mada. If
5.3.4 When a matenal has to be homogenized, this
the material to be measured consísts of a mixture ot
shall be done in the manner most appropriate for that
powders oí different relativa density or of different
material. When the material to be tested is not
grain sjze. sorne cara is needed because segregatian
homogeneous, it ¡s important to prepare the samples
may result from shakíng, tor examp!e durillg transporto
in tlle manner specifíed in the method, preferably
When reaction with the atmosphere may be ex
startíng with cne batch 01 commercial material for
pacted, the specimens may be sealed into ampoules.
each level. In the case 01 unstable material s, specíal
either evacuated or filiad with an inert gas. For pero
ínstructions on storaga and treatment shall be speci
ishable materíals such as food or blood samples, ii
fiad.
may be necessary to sand them in a deep-frozen state al to contributa hísfher spedalized knowledge in de
to the participating labarataríes with detaíled in signing the experiment;
structions for ihe procedure for thawing.
b) to analyse the data;
el to write a repDrt tor submission to the panel fol
6 Personnel ¡nvolved in a precision lowinQ the instructions comained in 7 I
experiment
NOTE 3 The methods of operation withín differenl lab 6.3 Executive functions
orataries are not expected to be identlcal. Therefore the
contents oí thís clause are anly ¡ntended as a gUlde to be
modified as appropriate to cater for a particular situatioo. 6.3.1 The actual organization of the experiment
should be entrusted to a single laboratory. A member
of the 5t8ft of that laboratory should take full raspen
6.1 Panel sibilí1y; he/she is called the executíve officer and is
appointed by the paneL
6.1.1 The parle: snould consist of experts familiar
wíth the measurement method and its applicatíon.
6.3.2 The tasks of the executive officer are:
6,1.2 The tasks of the panel ara: al to enlist the cooperation of the requisite number
of labaratorías and to ensure that supervisors are
al to plan anó coordinate the experiment; appointed;
b) to decide on the number of laboratoríes, lellels b) to organiza and supslVisa the preparation of the
and measurements to be made. and the number materials and samples and the dispatch of the
of signíficant figures to be required; samples; lor each le1l81, en adequate quantity ot
e) to appoint someone for the statístical funclions material should be set aSIda as a reserve sto.;;k;
(see 6.2);
e) te draft instructions covering aíl the poínts in 5.1.2
d) to appoint someone tar the executlve functions a) to h). and circulate them to the supervlsors
(see 6.3': early enough in advenca for them to raise any
comrnents or querias and to ensure that operators'
e) to conslder the instructions to be issued to the
selected are those who would normal1y carry out
laboratory supervisors in addition 10 the standard
such measurements in routine operat¡ons:
measurement method;
dl to design suítable forms for the operator to use
fl to decide whether some operators may be al
as a working record and for the supervisor to re
!owed to carry out ti few urlofficial rneasurements
port the test results to the requisite number of
in arder to regain experíence 01 the method after
significant figures (such forms may include the
a long imerval (such measurements shall never
name of the oparatar, the dates on wh¡ch sam
be carried out on the affieíal collaboratíve sam
pies wara received and measured, the equipment
pies);
usad and any other relevant ínformatíon);
gl to discuss the raport of the statistical analysis on
e} to deal with aoy queries from laboratones regard
completion of the analysís of the test results;
ing the performance of the measurements:
h) to establísh final valuas for the repeatabilíty stan
dard deviation and the reproducibility standard f) to see that an overall time schedule is maintained:
daviation; g) to callect the data forms and present them to the
i) lo dec,de if flJr1her actions are requlred to improve statistical experto
the standard for the measurement method or with
regard io laboratories whose test results nave
been rejected as outliers. 6.4 Supervisors
6.4.2 The tasks of the supervisor are 6.5.Z Because the object of the experiment is to
determine the precision obtainable by the general
al ensure tha1 tha aparatara setected are those who popula11on of operators workíng from the standard
would normally carry out such measurements in measurement method, in general the operators
routine operatíons; should not be given amplificatíons to the standard for
b) to hand out the samples to the operatorts) in the measuremen1 method. However, ít should be
keeping with the instructions of the executív€ of pOlnted out to the operators tnat the purpose of the
fieer (and to provida material tor famíliarization exercise is to discover the extent to which results can
experiments, if necessary); vary in practica, so that there will be less temptation
e) to supervise the execution of the measurements far them to disca rd or rework results that they feal are
(the supervisor shall not take par! in performing inconsistent.
the measurements);
dí 10 ensure that lhe operators carry out the requirad
6.5.3 Although normally the operators should re
ceive no supplementary amplifícations to the standard
number af measurements;
measurement method. they should bé encouraged to
el lo ensure adherence to the sel timetable for per commant on the standard and, in particular, to stete
forming the measurements; whether the instructions contaíned in ¡t are sufficiently
f) to collec1 the test results recorded to the agreed unambiguous and elear.
number of decimal places. including any anom·
alies and dittículties eX.O€Iríenced. and comments 6.5.4 The tasks of the operators are:
made by the operators.
al to perform 1he measurements accordlng to the
6.4.3 The supervisor of each laboratory should write standard measurement method;
a full report which should contairr the following in1or b) to report any anomalies or difficulties experí
mation: anced; it ¡s better to report a mistake than to ad
just the test resu/ts because one or two missir¡¡:;
al the test results. enterad legibly by their origínator test results will not spoil 1he experiment end
on the forms províded, not transcribed or typed n1any índicate a deficiency in the standard;
(computer or testing machine printout may ba ac
ceptable as an alternativa); e) te comment 01'1 the adequacy of the instructions
in the standard; operators should report any oc
al the original observad values or readings (if any)
casions when they are unable to follow their in
from which the test results were derivad, enterad
structions as this may also indicate a deficiency in
iegibiy by the operator on the forms provided, not
the standard.
transcriaed or typed;
el comments by the operators 01'1 the standard tor
the measurement method; 7 Statistical analysis of a precision
dI informatíon about irregulanties or disturbancss experiment
that may hava occurred during rhe measure
ments, íncludíng any change of opera1or that may 7.1 Preliminary i:onsiderations
have occuned( together with a statement as to
which measurements were performed by which
operator, end the reasans tor any missing results; 7.1.1 The analysis 01 the data, whlch should be
considered as a statistica! problem to be salvad by a
el the dete(s) on which the samples were received; statistical experto ínvol1l8s three successive stages:
ti the date(s) on whieh each sample was measured;
ai critical examination of the data in older lo ¡dentiíy
g) information about the equipment used. if relevant; and treat outlíers or othef írregularities and to test
the sUltabilíty of the model;
h) sny othar relevant information.
b} computation of preliminary I/alues of precision and
6.5 Operators means tor each leve I separately;
e) establishment of final values of precísion and
6.5.1 In each laboratory the measuremenls sha/l be means, íncluding the establishment af a relation
carried out by one oparatar selec1ed as being repre· ship between precision and the leve I m when the
sentative of those likely to perform the measure· analysis indícates that 5IJch a relatíonsh¡p may
mants in normal operations. existo
Cop~ri!Jht h~ the INTERNAT IOHAL ORGAH IZAr ION FOR STANDARD ¡ZAT IOH (ISO)
fue Jan 16 16:58:57 1996
• 48.51903 0594.520 054 •
QISO ISO 5125-2:1994(E)
7.1.2 The analysis first compules. for each level from the comparable entríes in the same tabla that
separately. estimates of they are considered irreconcilable wilh the other dala.
Experience has taught that outliers cannot always be
- the repeatabilíty variance s} avoíded and they have to be taken into consíderation
- the between-Iaboratory varianee .se in a similar way to the treatment of missing data.
7.2 Tabulation 01 the results and notation lhis part of ISO 5725 does not provida a statistical
test by which suspected laboratories may be judged.
used The primary decísion should be the responsibi¡ity of
the stallstical expert, but all rejected !aboratories shall
7.2.1 Cells
be reported to the panel for further action.
Each combínalion of a laboratory and a level is called
a cel! of the precision experimem. In ths ideal case, 1.2.6 Erroneou-S data
the results of 6n experiment with p laboratorles and
q levels consist of a tabla wíth pq cells. each contaín Obviously erroneous data should be ínvestigated and
ing n reolícate test results that can all be usad for corrected or discarded.
computing the repeatability standard deviation and the
reproducibillty standard deviatlon. T;¡is ideal situatiofl 7.2.7 Balanced linifol'm·revel tast results
15 no1. however, always attained in practice. Depar
tures occur owing to redundant data, missíng datá and The ideal case is p labaratories caHed
oUlliers. (í ~ 1. 2, ... p), each testlng q levels called j
(; = 1. 2. .... q) with n replicates at each level (6ach
7.22 Redundant data ij eombinationl. glving a total of pqn test results. Be
cause of missing O.2.3} or deviating (7.2.4) test re
Sornetirnes a laboratory may carr¡ out and rapart sufts, or outlying labaratories (7.2.5l or erroneous daTa
more than the 11 test results officially specified. In that 17.2.6), this ideal situation is no1 always attained. Un
case. the supervisor sha" rsport why this was done der these conditions too notations given irl 7.2.8 to
and whích are the correet test results. If the answer 7.2.10 and the procedures of 7.4 aHow for dHfering
la that they are all equally valid. lhen a random se numbers of test results. Specimens oi recommended
lection should be made from those available test re· forms for the statístical analysis are gíven in figure 2.
sults to choose the planned number of test results for For convenience. they wlll be raferrad to slmply as
analysls. e
forms A. 8 and (of fígure2).
,
l'
. f:/
./ /
-
·.
p ) I
Form e Recammend8d form far 1he eallatian of 1he measures of spread withlo eells
I.anl
LabOfilt1ory I I
1 2 .. .. j .. i , , q 1 q
1
2
.,
¡ $i)
, ,
7.2.9 Cell means (fomo B of figure 2) The cel! means should be recorded 10 one more sig
nificant figure than the test result in form A.
For the general case, use the intracell standard devi oratories or values that appear to be inconsistent with
atloo all other laboratoríes or values may cr.ange the esti
mates, 8nd decisions haya to be made with respect
'l. to these values. Two approaches are íntrcduc~d:
s·:
'/
¿(y,¡.t-Yi~ . " (3) fl, "r:" ,:. ("
r
b) numerical outlier tests. _\\I\(,I¡..'. le G
oro equ!valently
The standard deviatlon should be expressed to one in whieh. tor y,¡ sea 7.2.9. and for ~ see 7A.4.
more significant figure than the results in form A.
Plot tha lit! valuas for each cell in arder of laboratory,
For values of n,} less than 2, a dash should be inserted in groups for each le'lel land separately grouped for
in form C. the several levels examinad by each laboratory) (se9
figure B.7).
~ 7.3.1.3 Examination of the h and k plots may indicate 7.3.1.7 In additíon to these h and k graphs.
1hat specific Jabaratoríes exhíbít patterns of results histograms of cel! means and cel! ranges can reveal
that are markedly different trom the others in the the presence of. for example. two distinct popu
study, This ís índicated by consístently high or low lations. Such a case would require sp9cial treatment
within-cell variatíon and/or extreme cell means across as the general underlyíng principie behind the meth·
many levels. If this occurs. the specific labaratory ods described here assumes a single unimodal popu
should be contacted to try to ascertain the cause of lation.
the discrepant behaviour On the basis of the findings.
the statístícal expert could:
7.3.2 Numerical outller tachnique
al retaín the laboratory's data for the moment;
7.3.2.1 The following practice ís recomrnended for
b) ask the laboratory to rado the measurement (if
dealing with outliers.
feasiblel;
al The tests recommended in 7.3.3 and 7.3.4 are
e) remove the laboratory's data from the study. applied to iden¡ify stragglers or outliers:
7.3.1.4 Various patterns can appear in the h plots. - if the test statístic is less than or aquel to its
Al! laboratoríes can have both positive and negative It 5 % critical vslue, the item tested is accepted
va'ues at dífferent levels of the experimento Individual as correct;
laboratoríes may tend to give either al! positiva or alJ
negative /¡ valuas, and the number of laborataríes - if the test statistic is greater than íts 5 % crit,
giving negativa values is approximately equa! ro those ical value and less than or equal 10 ¡ts 1 %
glVlng posltive valuas. Neither of these patterns IS critlcal value, the ítem testad is callad a
unusual or requires investigation. although tila second straggler and is índicated by a single asterisk;
of these patterns may suggest that a common source
of laboratory bias exists. On the other hand, if all 1he - if the test statistic is greater than its 1 % crit~
h valuas for one laboratory are af one sign and the h ical value, the ítem is called a statistical outlier
valuas for the other laboratoríes are all of the other and ls indicated by a double asterisk.
sigo, then the reason should be 50ught. Ukewise. if
the h valuas·for a laboratory are extreme and appear b) It is next investigated whether the stragglers
to depend on the experimental level in sorne sys and/or statistícal outlíers can be explained by
tematic way. then the rsason should be sought. Unes soma technícal error. for example
are drawn on the h plots corresponding to the indica
tors given in 8.3 (tables 6 and 7), These indicator lines - a slip in performing the measurement.
serve as guides when examíni'.1g patterns in the data.
- an error in computation,
7.3.1.5 If one laboratory stands out on the k plot as
- a simple clerical error in transcribing a test re
having many large values, then the reason should be
sult, or
sought: this indlcates that it has a poorer repeatability
than !he other laborataríes. A laboratory could give - analysis of the wrong sample.
rise lo consistently small k values because of such
factors as excessive rounding of its data or an insen Where the error was one of the computation or
sitíve measurement scale. Lines are drawn on the k transcríption type, the suspeet result should be
plots correspondíng to the indícators given in 8.3 (ta raplaced by the correet value: where the error
bies 6 and 7}. These indicator Hnes serve as guides was fmm analysíng a wrong sample. the result
when examining patterns in the data. should be placad in íts correct cell. After sLlch
correctíon has been made. the examínatlon for
7.3.1.6 When I€ n h or k plot grouped by laboratory slragglers or outliers should be repeated. If the
sllggests that one laboralory has several h or k values explanation of the technical error is sueh that it
near the critica I value line, the correspondíng plot proves impossible to replace the suspeet test re
grouped by level should be stlldíed, Often a valua that sult, then it should be discarded as a "genuine"
appears larga in a plot grouped by laboratory will turn outlier 1hat does not be!ong to the experiment
out to be reasonably consis1ent with olnar labora· proper.
torles tor the same level. If ít ís revealed as strongly
dif1erent from values tor the other laboratories, then e) When any straggler::¡ and/or statistical outliers re
the reason should be sought. main that hsve not been explained or reiected as
10
belongíng to an outlying laboratory, the stragglers Cochran'::; test has to be applied to form C of figure 2
are retained as correet i1ems and the statistical at each level separately.
outliers are discarded unless the statistician for
good reason decides to rataín them, 7.3.3.3 Cochran's critaríon applies Slríctly only when
all the standard deviatlons are derived from the same
d) When the data for a cell have been rsiectad for number (n) of test results obtaíned under repeatability
fQrm 8 of figure 2 under 1he aboye procedure, condl1íons, In actual cases, this number may vary
then the corresponding data shall be rejected for owing to missing or discarded data, This part of
forme of figure 2, and vice versa. ISO 5725 8ssumes, however, that in a properly or
ganizad 9xperiment su eh vanations in the number of
1.3.2,2 The tests given in 7.3,3 and 7,3.4 are of two test results par cell will be limitad and can be ignorad,
types. Cochran's test is a test of the within-Iaboratory and therefore Cachran's critanon ís applied using for
variabilities and should be applied first, then any n the number of test results occurring in the majority
neeessary action should be taken, with repoatad tes1s of cells.
il necessar'l. The other test (Grubbs') is primarily a
test of between-Iaboratory variebility, and can al50 be
used (if ti> 2) where Cochran's test has raised sus· \7.3,3.4 Cochran's criteríon tests only the híghest
píclons as to whether the high within-Iaboratol)' varí : value in a set 01 standard daviatíons and io.; therefore
ation ís attríbutable to only one of the test results in 1 a ane-sided outlier test. Varlance heterogeneity may
~'--also, of course, manifest itself in some of the stan
the cell,
dard devia1ions being comparatively too low. How
aver, sma!1 values of standard deviation may be very
7.3.3 Cochran's test strongly influeneed by tha degree of rounding Qf the
original data and are for that reason not very relíable.
7.3.3.1 Thls pan of ISO 5725 assumes that between In addition, it seems unreasonable to reject t"e data
laboratorjes only 5mall differences exis1 in the within from a laboratory beca use it has accomplished a
laboratory variances, Experience, however, shows higher precision in its test results than the other I¿¡b
that thís is not always the case" so that a test has oratories. Hence Cochran's criterían is consídered ad
been included here to test the validity of this as equete,
sumption. Several tests could be used for this pur
pose, but Cochrao's test has been chosen.
7.3.3.5 A critical examination of fonn e of figure 2
may sometimes reveal that the standard deviations
7.3.3.2 Given a set 01 p standard devíations Si' all for a particular laboratory are at an or al mast levels
computed from the same number (ll) of replicate re lower than those for other laborataríes. This may in·
sults, Cochran's test s1atis1ic, e, ís dicate that the laboratory works with a lower repeat
abílity standard deviation than the other labarataríes,
e ... (8) which ir, turn may be caused either by bertar tech
nique ano equipment or by a modified or incorrect
application of the standard maasuremenl method. If
thís occurs it should be reported to the panel, which
where Smax is Ihe highest standard devíation in 1he set. should then decide whether the poínt is worthy of a
more detailed inves1igation, (An example of this is
al If the test s1atistic ís less than or equa I to its 5 % taboratory 2 in the experime n1 detailed In B, 1.)
critical valuB, the ítem tested is accepted as cor
ree!.
7,3.3,6 If the highest s1andard deviat'ron ís classed
bl If ¡he tes1 statísllc is greater than its 5 % critica! as an autlier, then the value should be omitted and
value and (ess then or equel te its 1 % critical Cochran's test repeated on the remaining va!ues. This
value, the ítem tested is called a straggler and is process can be repestad but í1 may fead 10 excessiv!';l
indicated by .si single asterlsk, rejeetíons wheo, as is sometimes the case, the
underlyíng assumption of normality is not sufficíently
el If the test statistic ís greater than its 1 % critical welJ approximated lO. The repeated applicauon of
valua, the ítem is callad a statistical outlíer and is Cochran's test is here proposad only as a he(pful tooi
índicated by a double asterisk. in view of the lack of a statístical test designed for
testing severa! outliers together, Cochran's test is not
Critical values for Cochran's test are gíven in 8.1 designed for this purpose and great cautlOn shourd be
(table 4). exercised in drawíng conclusions. When two or three
11
and (18)
s= ... (11) Critical values for Grubbs' test are given in 8.2
(lable 5).
12
Grubbs' test tor two outlying obseNótions de· 7.4.4 Calculation of the general mean m
scribed in 7.3.42. It the Grubbs' test doss not
show a cel! mean to be an outlier, then apply the For level j, the general mean is
double-Grubbs' test described in 7.3.4.2.
¿)n" l)s~
i=' _ _ __ __ (20)
7.4.2 Basie data f'
¿(III} 1)
The basic data needed for the computations are pre i= 1
sented in the three tables given in figure 2:
where
The number of non-ernpty cells may be different tor These celculatlons are ¡Uustrated in the examples in
different levels. hence the index j in Pi' 8.1 and 8.3 in annex B.
13
7.4.5.3 For 1he particular case where all ni} == 11 = 2, 111: Ig Sr ¡; + d Ig m lar s, = Cm d); d .¡;.1 (an expo
the simpler formulae may be used, gíving nential relationship)
1: $, = bm (a straight line through the origin) 7.5.6.1 A good estirnate of the parameters of the
regression líne requires a weíghted regression be
11. Sr = a + bm (e straight line with a positiva inter cause the standard error 01 ,f is proportional 10 1he
capt) predícted value of Sj (~).
14
The weightíng factors haya to be proportional to 7.5.6.4 For relatíonshlp 11, the mitial values Sal are the
l/(si. where ~ is the predicted repeatabirlty standard original values of s as obtained by the procedures
deviatíon fQr level j. However si
depends on par gjven in 7.4. These are used to calcutate
ameters that have yet to be calculated
WOj = 11(so/ U=:; 1, 2, .... q)
A mathematícally correet proeedure fer fíndíng esti
mates cerresponding to the weighted least-squares and to calculate a, and b¡ as in 7.5.6.2.
of residuals may be comp¡¡cated. The following pro
cedure, wliich has preved to be satisfactory in prac This leads to
tice, is recommended.
7.5.6.2 With weighting factor W¡ equal to 1/(~l, The computations are (hen repeatad with
where N '" O, 1, 2 ... for successive íterations. tlien Mi,) = l/{s,l
to produce
tlie calculated formulae are:
;2j=lJ:l + b¿~j
For relatíonship 11 (8 -= a +
a= ... (25)
and
... (26)
T4 ... ¿ (19 mj) (lg Sj)
i
15
',,-- 7.5.9 Exampres of fittirtg relationships 1, I! and 111 of 7.5.9.1 An example of fitting relatio0shlp I is given
7.5.2 to the same set of data are now given in 7.5.9.1 in table 1.
B.3 and have been used here only to iIIustrate the 7.5.9.2 An exampre 01 fitting retationship II is given
numerical procedure. They will be further discussed in tabla 2 (mi' si are as in 7.5.9.1).
in B.3.
Table 1 - Relationshlp 1: .r = bm
mi 3,94 8,28 14 18 15.59 2D,41
0,092
1 .
0,127
sI 0.179 0.337 0.393
~(Jjlmj) 0.094 B
= 0,019
b --q 5
31 0,058 + 0,009 O m
El = 0,030 + 0,015 6 m
,
S2J 0,092 0,159 0,251 0,273 0,348
w 118 40 16 13 B
"
SJ 0.032 + 0,015 4 m
NOTE - The values of the weighting factors are not crítical; íwo significant figures suffice.
19 1ÍiJ
19 so,
I + 0.595
- 1,036
+ D,91S
- 0,747
+ 1,152
- 0,896
I + 1,193
0.472
I
!
'1 1.310
-0.406
19 s = 1,506 5 + 0,772 Ig m
or & - 0,031 mO,}]
16
1.6 Statistical analysis as a step-by-step 7.6.6 Investigate whether there ís or may be some
procedure technical explanation for the stragglers and{or stat·
istical outlíers and, if possible, verfty su eh an expla
NOTE 5 Figure 3 índicates in a slepwíse fashlon the pro· nation. Correet or discard. as required, those
cedure given in 7.6. stragglers andjor statistical outliers tha1 have been
satisfactorily explained, and apply corresponding ccr
rections to the forms. If there are no stragglers or
7.6.1 Colreet all available test resul1s in ene form, statistical outlíers lef1 that nave not been explaíned,
10rm A of fjgure2 (sea 7.2). It is recommended that ignore steps 7.6.7 to 7.6.10 and proceee direCtly with
this form be arranged ¡nto p rows, indexed 7.6.11.
i = 1, 2, ... p (representing the p laboratorías that
have contributed data) and q columns, índaxed NOTE 6 A large number 01 stragglers and{or statistical
j = 1, 2, ", q (representing the q levels in ínereasing outliers may indícale a pronounced variance inhomogeneity
arder). or pronounced dífferences between laborataríes and may
lhereby cast doubt on the suitability of the measurement
In a uniform-Ievel experiment the test results within method. This should be reported lo the panel.
a cel! of form A need not be distínguished and may
be put in aoy desired crder. 7.6.7 If the distríbution of the unexpkllned stragglers
e
or statistical outliers in form B or does not suggest
any outlying laboratoríes {see 7.2.5L ignore step 7.6.8
7.6.2 Inspect form A tor any obvlous írregularities, and proceed directly with 7.6.9.
írwestígate ando if necessary, discard any obviously
erroneous data (tor example, data outside the range 7.6.8 If tne evidence against sorne suspected
of the measuring mstrument or data which are ím autlying laboratDríes is cOf"Isidered strong enough to
possible for technical reasons) and rsport to the panel. iustífy the rejectían of some or all the data from those
It is sometimes ímmediately evident that the test re laboratoríes, than discard the requisite dala and report
sults 01 a particular laboratory or in a particular cell líe to the panel.
at a level ínconsistent with the other da1a. Such obvi
ously discordant data shall be discarded immedlatelv, The decision to rejeet some or al1 data from a partic
but the fact shall be reported to the panel fer further ular laboratory is the responsibility of tlle stalistical
consíderation (see 7.7.1 ). expert carrying out the analysis, but shall be reported
to the panel tor further consideration (see 7.7.1).
7.6.3 From form A, corrected according to 7.6.2 7.6.9 lf any stragglers andlor statistical outliers re
when needed, compute form B containing cel! means main that have not been explained or attributed to an
e
and form containing. measures of wíthin·ceil spread. outlying laboratory, discard tha statlstical ou1liers but
retain the stragglers.
When a cell in form A contains only a single test re
sult, one of the options of 7.4.3 should be adoptad. 7.8.10 If in the previous steps any entry in form B
has been rejected. then the corresponding entry in
form C has lo be rejected also. and vice versa.
1.6.4 Prepare the Mandel h and k plots as descríbed
ln 7.3.1 and examine tham for consístency oi t.he 7.6.11 From the entries that have been retained as
ctata~ TheseRlots may indicate the suitabilíty Qf tlle correet in forms B and e, compute, by the procedures
data for further analysis, the pre,sence of any possíbie given in 7.4, for each leve l separately, the mean level
outlying values or outlying laboratories. However, :'la In) and tlle repeatabilit'{ and reproducibilíty standard
definite decisions are taken at 1his stage, but are de· deviations.
layed until completion of 7.6.5 to 7.6.9.
¡Ji.12 If the experímerlt only used a sll1gle level, or
if 11 has been decided that the repeatabiiiw and re
7.6.5 Inspect forms B end e (see flgure 2) level by
producibility standard deviations should be given sep
level fo r possible stragglers and/or statíst]cal outliers
arately for each level (see 7.5.1) and not as functions
[sea 7.3.2.1 al]. Apply the statistical tests given in 7.3
of the level, ignore steps 7.6.13 to 7,6,18 and proceed
te all suspect ítems, marking the stragglers wíth a
dírectly with 7.6.19.
single asterisk and the statistical outliers with a dou
ble asterisk. lf there are no stragglers or statistícal NOTE 7 The following steps 7.6,1 3 lo 7.6.1 7 are applíed
oulliers, 19nore steps 7.6.6 tú 7.6.10 and pracead di 10 ;sr aOO separately, but lor oreviw they are wrilten out
SR
rectly with 7.6.11. only in terms al sr
11
Ora", up forrn Á.
No
Oiscc.rd correspondfng
entry in lorm 8 01" C.
Figure 3 - Flow diagram o, the principal steps in the statistical analysi:s (continued Q() page 19)
18
19
7.6.13 PIQt sJ agamst n~ and Judge frOn! tnis plot panel. in thls repon me followíng informatíon should
whether s depends on m or not. If s is considered to be given:
depend on m, ignore step 7.6.14 and proceed with
7.6.15. If s is judged to be independent of m, proceed al a full account of the observatíolls received f[Qm
with step 7.6.14. If there should be doubt, it is best the operators and/or supervisors concerning the
to work out both cases and let the panel decide. standard tor the measurement method;
Thare exists no useful statistical test appropriate for b) a fuI! account of the laboratories that have bse/l
thís probJem, but ¡he technical experts familiar with reíected as outlying labarataries in steps 7_6.2 and
the measurement method should have sufficient ex 7.6.8. together wíth the reasons for their re-
perience to take a decision. jection;
7.6.14 Use +LSJ= sr as the final value of the re c) a full account of any stragglers and/or statístical
outliers tha! were discovered, and whether these
peatability standard deviation. Ignore steps 7.6.15 to
7.6 18 and proceed directly wíth 7.6.19. were explained and corrected, or díscarded;
Mj , .~,. and ,'iN and an ac
d) a form of the final rasults
1.6.15 Judge from the plot of 7.6.13 whether the count of the conclusions rsached in steps 7.6.13,
relalionship between s and m can be representad by 7.6.15 or 7.616. illustrated by one of the plots
a straight line and, if so, whether relationshíp I recommended in these steps;
(s hm) or relationship 1I (.1 = a + bm) is appropri<'lte
(sea 7.5.2). Determine the parameter b, or the two e) forms A, B and e (figure 2) usad in the statlstlcal
parameters a and b, by the procedure of 7.5.6. If the analysis. possibly as an annex.
linear reJatíonship is considered satisfactory, ignore
atep 7.6.16 afld proceed directfy wfth 7.6.17. If noto
prooeed with step 7.6.16. 1.7.2 Decisions 10 be taken by the panel
7.6.16 Plot Ig Sj against Ig fñj and judge from this The panel should then discuss this report and take
whe1har the ralationship between 19 s and 19 m can decisions conceming the following questions.
reasonably be representad by a stralght Ime. It thls is
al Are the discordant ri3sults, stragglers or outliers.
considerad satisfactory, fit the relationship 111
if any, due to defects in the descrjption of the
(Ig s = e + d 19 m) using me procedure given in 7.5.8.
standard for the measurement method?
7.6.17 11 a sa1isfactory relation has been estabrished b) What aotíon should be taken w[th respect to re
in step 7.6.15 or 7.6.16, then the final values of Sr (or jected outlying ¡abaratarías?
SR) are the smoothed values obrained from this re
el Do tlle r9sults of the outlying laboratories andlor
lationship tor given. values of m. Ignore step 7.6.18
the commen1s received from the operators and
and proceed with 7.6.19,
supervisors indicate the need to ímprove the
standard for tha meaSurement method? If so,
7.6.18 If no satisfactory relation has been estab
what are the improvements required?
lished in step 7.6_15 or 7.6.16, the statistical expert
should decide whether some other relation between d) Do the results of the precíslon experiment justify
< Ano m GI'!f1 b~ established. N alternatilfel'l \,vhethe r the establishment of values of the P9peatabilit'i
the data are so irregular that the establishment of a standard deviation and reproducíbility standard
functíonal relationshíp ia considered to be ímpossible. deviation? If so, what are those valves, in what
form should they be published, and what is the
7.6.'19 Prepare a report showing the basic data and regian in which the precisíon data apply?
the results and conclusions from the statistical analy
siso and present this to the panel. The graphícal pres
entatíons of 7.3.1 may be useful in presenting thtt 7.7.3 Full report
consístency or variabilíty oE the results.
A report setting out the reasons for the wQrk and how
7.7 The report tD, l,lnd the decisions to be it was organized, including the rapart by the statisti
taken by, the panel cían and setting out agreed conclusions. should be
prepared by the executive officer for approval by the
7.1.1 Report by the statistical expert panel. Sorne graohical presentation of consistencv or
variability is oiten useful. The report should be circu
Havlng completad tha statistlcal analysis, the statisti lated to those responsible for authorizing tlle work
cal expert should wdte a report to be submitted te the and to cther interested parties.
20
8 Statistical tables
8.1 Critical values for Cochran's test (see 7.3,3} are
given in table 4.
¡ 5
6
0,928
0,883
0,841
0,781
0,788
0,722
0,684
0,616
0,696
0,626
0,598
0,532
0.63.'3
0,564
0.544
0,480
0.589
0,520
0,50e
0,445
l 0,838 0,727 0,664 O,56'~ 0,568 0,480 0,508 0,431 0,466 0,397
8 0,794 0.680 0.615 0.516 0,521 0,438 0,463 0,391 0,423 0.360
9 0,754 0,638 0,573 0,478 0,481 0,403 0,425 0,358 0,387 0,329
10 0,71$ 0.602 0.536 0.445 0.447 0,373 0.393 0,331 0,357 0,303
11 0,684 0.570 0.504 0.417 0,418 0,348 0.366 0,308 0,332 0,281
12 Q,S53 0,541 0,475 0,392 0,392 0.326 0.343 0,288 0,310 0,262
13 0,624 0,515 ~.450 0,371 0,369 0,307 0.322 0,271 0,291 0,:243
14 0,599 0,482
• 0,427 0.352 0,349 0,291 0,304 0.255 0 274
1 0,232
15 0,575 , 0,471 0,407 0,335 0.332 0.276 0,288 0,242 0,259 0,220
16 0,553 0,452 0,388 0.319 0,316 0,262 0,274 0,230 0,246 0.208
17 0.532 0,434 0.372 0,305 0,301 0,250 0.261 0.219 0.234 0,198
18 0,514 0,418 0,356 0.293 0,288 0.240 0,249 0.209 0,223 0,1!3S
19 0,496 0.403 0,343 0,291 0,276 0,230 0,238 0,200 0,214 0.181
20 OA80 0,389 0,330 0,270 0,265 0,220 0,229 0.192 0,205 Q.174
21 DAS!: 0,377 0,318 Oj2o~ O.,"'",
,_...r,,", 0,212 0.220 0,185 0,197 0.167
22 0.450 0,365 . 0,307 0,252 0.246 0,204 0,212 0,178 0,189 0.160
23 0,437 0,354 0,297 0,243 0.238 0,197 0,204 0.172 0,182 0,155
24 0,425 0,343 0,287 0.235 0.230 0,191 0.197 0,166 0,176 0.149
25 0,413 0,334 0,278 0,228 0,222 0,185 0.190 0.150 0.170 0.144
26 OA02 0.325 0.270 0.221 0,215 0,179 0.184 0,155 0,164 0,140
27 0,391 0,316 0.262 0,216 0.209 0,173 0,179 0.150 0.159 0,135
28 0,382 0,308 0,255 0,2Q9 0,202 0,163 0.173 0.148 0,154 0.131
29 0,372 ' 0,300 0,248 0,203 0.196 0,164 0,168 0.142 0,150 0,127
30 0,363 0.293 0,241 0,198 0,191 0,159 0,164 0,138 0,145 0.124
31 0,355 0,286 0,235 0,193 0,186 0,155 0,159 0,134 0,141 0.120
32 0,347 0,280 0,229 0,188 0,181 0,151 0,155 0,131 0,138 0,117
33 0,339 0,273 0,224 0,184 0,177 0,147 0,151 0,127 0,134 0,114
34 0,332 0,267 0.218 0,179 0,172 0.144 0,147 0,124 0,131 0,111
35 0.325 0,262 0.213 0.175 0.168 0.140 0.144 0,121 0,127 0,108
0,318 0,208 1 0,172 0,165 0,137 0,140 0,116 0,124 0,106
36 0.256
37 0,312 0.251 0,204 0,168 0,161 0,134 0,137 0.116 0,121 0,103
38 0,306 0,246 0.200 0.154 0,157 0.131 0,134 0.113 0,119 0.101
39 0,300 0,242 0,196 0,161 0,154 0.129 0,131 0,111 0,116 0,099
40 0,294 0,237 0,192 0,158 0,151 0.126 0,128 0,108 0,114 0,097
p = number 01 laboratories at a given level
21
"-.-. 8.2 CritiCal values for Grubbs' test (sea 7.3.4) are For the Grubbs' test tor two outlying observations.
given ir: table 5. Qutliers and stragglers give rise to values which are
smaller than the tabulated 1 % and 5 % critical values
For the GrubbS' test for one outlying observation, respectívely.
outliers and stragglers give rise to values whlch are
larger than the tabulated 1 % and 5 % critical values 8.3 Indicators tor Mandel's h and k statistícs (ses
respectively. 7.3.1) are given in tables 6 and 7.
3
1,155 1.155 -
4
h496 ¡.<!Si 0.000 O 0,0002
5
1.764 1,715 0,001 8
0,009 O
6
1,973 1,887 0,011 6
0,0349
7
2,139 2,020 0,0308
0,0708
B
2,274 2,126 0,0563
0.110 1
9
2,387 2,215 0.0851 0.1492
10
2,482 2,290 0.115 O 0.1864
11
2,564 2.355 0,144 8
Ot22i 3
12
2,636 2.412 0,1738 0.2537
13
2.699 2,462 0,2016 0,2836
14
2,755 2.507 0.228 O 0.311 2
15
2.806 2.549 0.253 O 0.3367
16
2,852 2.585 0.2767 0.360 :;
17
2,894 2,620 0,299 O
0,382 :2
18
2,932 2,651 0,320 O
0,4025
19
2,968 2,681 0,339 a
0,421 4
20
3,001 2,709 0,358 5
0,439 1
21
3.031 2.733 0.3761 0,455 e
22
3,060 2,758 0,3927 0,471 1
23
3,087 2,781 0,4085 0,4857
24
3.112 2.802 0,4234 0,4994
25
3,135 2,822 0,4376 0,5123
26
3.\57 2.841 0,451 o 0,5245
27
3,178 2,859 0.4638 0.536 o
28
3,199 2,876 0,4759 0,547 o
29
3,218 2,893 0,4875 0,5574
30
3,2313 2,908 0,4985 0,5672
31
3.253 2,924 0.5091 0,5766
32
3.270 2,938 0,5192 0,5856
33
3,286 2,952 0,5288 0.5941
34
3.301 2,965 0,5381 0.6023
35
3,316 2,979 0,5469 0,610 1
36
3,330 2,991 0,5554 0,6175
37
3,343 3,003 0,5636 0,6247
38
3,356 3,014 0,5714 0,6316
39
3.369 3,025 0.5789 0,636 :4
40
3,381 3,036 0,5862 0,6445
Reproduced, wíth the permission of the American Statistícal Assocíation. from re/arenea
(4) in annex C.
22
ISO 5725.2:1994(E)
23
p /¡
"
n
2
3
4
5
5
7
8
9
10
3
1,15 1,65 1,53 1,45 1,40 1.37 1,34 1,32 1,30 1,29
4
1,42 1,76 1,59 1,50 1,44 1,40 1,37 1,35 1,33 1.31
!5 1,57 1,81 1,62 1,53 1,46 1,42 1,39 1,36 1,34 1,32
IS 1,66 1,85 1,64 1,54 1,48 1,43 1,40 1,37 1,35 1,33
7
1,71 1,87 1,66 1.55 1.49 1,44 1,41 1,38 1,36 1,34
8
1.75 1.88 1,67 1.56 1.60 1,45 1,41 1,38 1.36 1,34
9
1,78 1,90 1.6B 1.57 1,50 1,45 1,42 1.39 1,36 1,35
10
1.80 1.90 1,68 1,57 1.50 1,46 1.42 1,39 1.37 1,35
11
1,82 1,91 1,69 1,58 1,51 1,46 1,42 1.39 1,37 1.35
12
1,83 1,92 1,69 1,58 1.51 1,46 1,42 1,40 1,37 1,35
13
1,84 1,92 1,69 1,58 1,51 1.46 1,43 1,40 1.37 1,35
14
1,85 1,92 1,70 1,59 1,52 1,47 1,43 1,40 1,37 1,35
15
1,86 1,93 1,70 1,59 1.52 1,47 1,43 1,40 1.38 1.36
16
1,86 1.93 1,70 1.59 1,52 1,47 1,43 1,40 1,38 1.36
17
1,87 1.93 1,70 1,59 1,52 1.47 1.43 1,40 1.38 1.36
18
1,88 1.93 1,71 1,59 1.52 1.47 1.43 1.40 1,38 1,36
19
1,88 1,93 1,71 1.59 1,52 1.47 1.43 1,40 1,38 1.36
: 20
1,89 1,94 1,71 1,59 1,52 1,47 1,43 1,40 1,38 1,36
21
1,89 1.94 1,71 1,60 1.52 1,47 1,44 1,41 1,38 1,36
22
1,89 1.94 1,71 1,60 1,52 1,47 1,44 1.41 1,38 1,36
23
1.90 1.94 1,71 1,60 1,53 1,47 1,44 1,41 1,38 1,36
24
1.90 1.94 1,71 1,60 1,53 1,48 1,44 1,41 1,38 1,38
25
1,90 1.94 ' 1,7i 1,60 1,53 1.48 1,44 1,4'1 1.. 38
t36
26
1,90 1.94 1,71 1,60 1,53 1.48 1.44 1.41 ¡ 1.38 1,36
27
1.91 1,94 1,71 1,60 1.53 1,48 1,44 1,41 1,38 1,36
2B 1,91 1,94 1.71 1,60 1,53 1,48 1,44 1.41 1,38 1,36
29
1.91 1.94 1,72 1,60 1,53 1,48 1,44 1.41 1,38 1,36
30
1,91 1,94 1,72 1,60 1,53 1,48 1,44 1,41 1,38 1.36
p = number of laboratoríes at a gíven leve!
24
Annex A
{normative)
Component in a test result repre UCL Upper controllimit (either actíon límit or warning
senting the random error occurríng límítl
in every test result
W \'Yeighting factor used in calculating a weighted
f Critical range factor regression
p-quantile of the F-distribution with
VI and iI.
degrees of freedom
w Range of a set of test results
25
26
Annex B
(informative)
27
Copyr igl.t b\l tbe INTERNAr IONAL ORGAN lZAflON FOR STANDARD ¡ZATlON (ISO)
Tue Jan 16 16:58:57 1996
.. ~8519D3 0594541 589 . .
1$0 5725-2: 1994rE) o ISO
6.1.3 Computation of can means lY;j) For lavel 1, largest value 01 s ís in laboratary 8:
The cell means are given, as a percentage by mass ri = 0,001 82: test value = 0,347
[% lmlmlJ. in table 6.2 in the format of form B of fig·
ure2 (see 7.2.9).
For level 2, largas! value af s is ¡n labora1ory 5:
8.1.4 Computation of standard deviations
(sy) r:s 2 0,006 36; test value = 0,287
Cochrao's test with n = 3 tor p = 8 laboratoríes gives Fer level 4, largest value of s ís in laboratory 4:
critical valuas of 0,516 for 5 % and 0,615 tor 1 %.
I:s
2
= 0,004 63; test value = 0,310
laboratory i 1 i 2 ! 3 4
I
slj nv &,¡ ti¡¡ slí lIij ,f
V nü
28
1 i xx
X
X
2 xx
3 xx
4 XXX
5~ x XX
x
6~ xx
x
7~ x
8 x X X
XxX)(
X
2~ xx
3 XI<X
xx x
""
51 h x
)<Xx
'"
,~ x !5
e xx x
m",(
29
xxxx
1~
x
z{ xx
x
3{ xx
x
4{ x x
si x x ) xx
6~ x ~
? xxx
e{ ~
1{ x ~ Ix
z{ x
X
3 xxx
x
xx I
S xxx x x
6 xxx
7 x> x
8~ xx
m.
30
analysis.
S,. = 0,015 24
Sil:; 0,026 32
13 LJl¡ 27
8.1.8 Conclusions
T4
= r.n,2 = 95 The precision of the measurement method should be
quoted, as a percentage by mass, as
T5 '= L(n¡ 1)s¡2 = 0,004411
Level
Single
low I Singl.
hlgh
Double
low I Double
high
Type oftest
Table B.5 - Computed values of mi' Sr] and SRi for sulfur content o, coa.
Level} f A
31
These values may be applied within a range order to provida for another type of graphical
0,69 % (m/m) to 3,25 % (mlm). They were deter presentation of data. Mandel's piats are fully
mined from a uniform-Ieve! experíment involving 8 illustrated and discussed in the example given in
laboratories covering that range of values, in which 8.3.
four stragglers were detectad 8nd retained.
8,2.2 Original data
These are presentad in table B.6. in degrees Celsius,
B_2 Example 2: Softening point of pitch in the format of form A of figure 2 (see 7.2.8).
(Severallevels with miss;ng data)
Table 8.6 - Original data: Softening point of
8.2.1 Background pítch (oC)
b) Source 2 i
89.7
89,8 I 98,5
97,2
97.2
97,0
102,6
103.6
32
Table 8.8 AbsoJute differences within cells: Softening point of pitch ("C)
Level}
Laboratory i I 1 2
I
3 4
i
33
Number 01 laboratories, p 15
The values for all tour levels are given in tabla B.l1.
34
Table 8.11 - Computed values of mJ' Sr! and SR] for $oftening point of piteh
!
Level¡ ~
m) (OCI sd s"J
:v:
15 9 ti
110 12 7 4
1~1116fS ) 6 z 11HI leveL 1,
16
n
9
~
14 ~
15 9 4
12 9 '2
111116 10 ) 1 131
level 2
i5 ro;
;¡ 7
116 12 7 51,'
110111 3 4 2 1 1
80 90 100 110
Temperature, oc
35
~~LL~~~~____~~~~~________________~3~______""_____________ leve!2
2 3 S
Temperature. 'C
36
measured having been selected so as to cavar the 'P1e test resul1s for laboratory 1 were always higher,
normal ranga expected to be encountered in gen· and at sorne levers consicterably higher, than those 01
eral commercial applicatíon. These were chosen the other /aboratarías.
to líe at t1'1e approximate leve!s of 4, 8, 12, 16 and
20 ['Yo Cm/m)]. The usual practice would be to re The second test result for laboratory 6 at level 5 is
cord test results to only one decimal place, but for suspect; the vafue recordad would fít much battsr a1
this experiment operators were instructed to work level4.
to two decimal places,
These points are discussed further in B,3,5
37
'-.. 8.3.4 Absoluta dlfferences within cells At level 5, the absolute difference 1,98 gave a test
t
statistlc valve of 1.9S ¡6.166 3 =.: 0,636.
These are gíven In tabla B. 14, as Wjj' as a percentage
by mass, in the format of form e of figure 2 (sea For p 9, the critical values for Cochran's lest are
7.2.10). 0,638 for 5 %. and 0,754 for 1 %.
38
Becí:luse of these test results, it was dedded to reject Wíthout these test results, 1he Cochran's test statístíc
tha pair of test results from laboratary 6 for level 5 at level 4 was then compared with the critical value
because it was uncertaín what meteríal had been for 8 laborataries (O,680 at 5 %) :and this no longer
measured and to rejeet all the test resul1s from lab appeared as a straggler and was retained.
oratory 1 as coming from an outlying laboratory.
.., 3
:i ---------
~ 2
E
.,
------
'" 1
'¡j
.
-o
c:
¿
O
- 1
2 _.___________________________________••____________________________________
3 7 8
Loboro.tory I
Figure B.1 - Titration of creosote oi!: Mandet's between·laboratory consistency statistic, h, grouped by
laboratoríes
39
6~--------------------------------------------------------------,
- 1
:1 5 6 1 9
labora tory f
Figure B.8 THration of creosote oil: Mandel's within-laboratory consistency statistic, k, grouped by
laboratories
B.3.6 Computation of mí' Sr) and ~Rj Someone familiar with the requiremen1s lar a stan
dard measurement method for creosote oil may be
The va!ues 01 ~, S¡j and Sflj computed without the test able to select the most suitable relationshíp.
results of laboratory 1 and the pair of test results from
laboratory 6, level 5, are given in table 8.16, as a per
csntage by mas::;, calculated as in 7.4.4 and 7.4.5. B.3.8 Final values el precision
From figure 8.9 it ís evident that lhe valva for level 3 The reason for the outlying 1esr results of laboratory
is strongly divergant and could not be ImprQved by 1 should be investigated.
any alternatíveprocedures {see 7.5.2).
This 5eems to have been a rather unsatisfaclory pre
For repeatability, a straight line through the origin cision experiment. One of the 9 laboratarías had to
seems adequate. be reiected as an outlier, and another laborator)' had
testad a wrong specimen. The material tor !evel 3
For reproducibility. al! three fines show adequate tit 5eems to have been wrongly selected, having almost
with the data, relationship 111 showing the best fit the same value as leve! 4 instead of Iying midway
40
between levels 2 and 4. Moreover, the material for other material. 1t might be wor1hwhile to repeat this
leve I 3 seems to have been sQmewhat dífferent in experiment, taklng more cara over the selection of the
nature, perhaps beíng more homogeneous than 1he materíals for the different levels.
Table B.16 - Computad values 01 ~jl .':'j and sR} for thermometric titration of
creosote oi!
"
LevelJ P, m¡ sn '\'Ri
".
o~·t
SR,O,04m '''-..
0,6
0.5
0.4
0.3
0,2
0,1
a
o s 10 1S zo m
Figure 8.9 - Plot of sr) and Sil) Hgainst mJ o. the data from table 8.16, showing the functional relationships
fittad in 7.5 from these data
41
Annex e
{informatível
Bibliography
[1] ISO Guide 33:1989, Uses of cereitíed reference [8J ISO 3534-2: 1983, Statistícs Vocabular¡ and
materia/s. symbols - Pare 2: Statistícal quafity control.
[2] ISO Guide 35:1989, Certification of reference [9] ISO 3534-3:1985, Statistics - VocabulalY and
materia!s -- General and ststistical principIes. symbo/s - Pan 3: Design of experiments.
(3) ASTM E691-87, Scandard Practice for Conduct [10J ISO 5725-3:1994, Accuracy (trueness and pre
ing an ¡nterlaborarory Study to Determine che cision) of measuremenf rnethods and resu/ts -
Precísíon of 8 Test Method. American Society Part 3: Intermedia te measures of the precision
for Testing and Materials, Philadelphia. PA, of a standard me8surement method.
USA.
[11] ISO 5725-4:1994, Accuracy (trueness and pre
[4J GRUB6S, F.E. Qnd BECK. G. Extension of sample cisíon) of measurement methods and resulrs -
SiZS5 and percentage points tor sígnificance Part 4: Basle methods for the determination of
tests of outlying obseNations. Technometrics, the trueness of 8 standard measurement
14, 1972, pp. 847-854. method.
[5] "Standard Methods fer Testing Tar and its [12) ISO 5725-6:-1., AcCtlraey (trueness and pre-
Products", 7th Ed. Stsndardisation ot Tar Prod ciSÍon) of measurement methods and results
ucts Tests Commi1tee, 1979. Pare 5: Alternative methods for the determj·
¡¡ation of the precisíon of a standard measure
[6J TOMKINS, S.S. Industrial and Engineering Chem ment method.
istry (Analytical edition), 14, 1942, pp. 141-145.
[13J ISO 5725-6:1994, Accuracy (trueness and pre-
[7] GRUBBS, F.E. Procedures for detecrlng oU11ying císíon) of measurement methods and results
obS8Nations in sample:>. Technometrics, 11, Pare 6: Use in practice of accuracy values.
1969, pp. 1-21.
1) To be published.
42
les 03.120.30
Descrlptors: measurem¡¡nt tests, tesl results, accuracy, reproduclbllir:y. statlsllcal analyslS.