See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/225777149

Homogeneity testing of reference materials

Article  in  Accreditation and Quality Assurance · February 1998

DOI: 10.1007/s007690050186

CITATIONS READS

49 1,549

3 authors:

J. Pauwels Andrée Lamberty

European Commission European Commission
130 PUBLICATIONS   3,195 CITATIONS    97 PUBLICATIONS   2,273 CITATIONS   

SEE PROFILE SEE PROFILE

Heinz Schimmel
European Commission
161 PUBLICATIONS   3,868 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Development of Certified Reference Materials for autoimmune antibodies View project

EU scrapie test evaluation 2005 View project

All content following this page was uploaded by Andrée Lamberty on 20 June 2015.

The user has requested enhancement of the downloaded file.

Accred Qual Assur (1998) 3 : 51–55
Q Springer-Verlag 1998 GENERAL PAPER

Jean Pauwels Homogeneity testing of reference

Andrée Lamberty
Heinz Schimmel materials

Received: 18 September 1997

Accepted: 21 November 1997
J. Pauwels (Y) 7 A. Lamberty
H. Schimmel
European Commission, Joint Research
Centre, Institute for Reference Materials
and Measurements (IRMM),
B-2440 Geel, Belgium
Tel.: c32-14-571722
Fax: c32-14-590406
Abstract Homogeneity testing is
of the highest importance for the
certification of reference materials,
as it should demonstrate the validi-
ty of the certified values and their
uncertainties in the analysis of in-
dividual units or portions thereof.
However, the conclusions drawn
from the results of these studies
may often be questioned. It is pro-
posed to improve this situation by
quantifying the effect of homo-
geneity studies in terms of their
impact on the uncertainty of certi-
fied values. Here it should be
noted that the between-units varia-
bility directly affects the certified
uncertainty, whereas the within-
unit inhomogeneity only defines
the minimum representative test
portion.
Key words Reference material 7
Homogeneity testing 7 Within-unit
heterogeneity 7 Between-units
heterogeneity

the statistical tolerance interval of the homogeneity

Introduction
The risk of inhomogeneity is, with few exceptions, in-
herent in any material. Therefore, before certification,
reference materials are normally submitted to homo-
geneity testing [1–4]. The primary aim of such homo-
geneity testing is to verify, on a reasonable number of
units, that all units of the certified reference material
(CRM) may be considered as being “identical” (be-
tween-unit homogeneity testing) when a batch certifica-
tion is envisaged. Moreover, if the material is not pre-
pared in portions for single use, it is, furthermore, gen-
eral practice to verify that the same applies to sub-sam-
ples from a single unit (within-unit homogeneity test-
ing).
However, for both between- and within-unit homo-
geneity testing, it is observed that tests are usually only
carried out with the intention of verifying the hypothe-
sis of no significant between-samples variation. If this
hypothesis is rejected, the material variability is quan-
tified and incorporated in the final uncertainty of the
CRM in case this variability has a significant influence
on the final overall uncertainty of the certified value.
This is done either by basing the CRM uncertainty on
study, as was done e.g. at IRMM (Institute for Refer-
ence Materials and Measurements) for EC-NRM (Eu-
ropean Community-Nuclear Reference Material) 501
[5], which may be considered as a quite conservative
approach, or by including the standard uncertainty due
to material variability in the uncertainty calculated ac-
cording to the Guide to the Expression of Uncertainty
in Measurement [6]. If, on the other hand, the hypothe-
sis is not rejected, it is usually concluded that “no sig-
nificant heterogeneity could be detected” without any
further consequences for the certified uncertainty inter-
val. These arguments are, however, contestable, as in
reality the only significant question is “is the (detected
or not!) material variability significant compared to the
uncertainty interval which has been certified?” (inde-
pendently of whether the above hypotheses are re-
jected or not). One may even state that it is preferable
that a degree of heterogeneity which is negligible com-
pared to the uncertainty interval which has been certi-
fied can be significantly demonstrated (Fig. 1) than to
be unable to detect significant heterogeneity (Fig. 2). In
the latter case a doubt remains about the effective de-
gree of heterogeneity.
52

Fig. 1 Example of a homo-

geneity study in which the de-
tected heterogeneity is signifi-
cant (Fexp 1 F0.05) but negligi-
ble compared to the uncer-
tainty interval of the certifica-
tion measurements (unc. cert.)

Fig. 2 Example of a homo-

geneity study in which no het-
erogeneity can be detected
(Fexp~F0.05) but possible het-
erogeneity may be significant
compared to the uncertainty
interval of the certification
measurements (unc. cert.)

Moreover, it is stated by Schiller in [4] that samples
used for homogeneity testing should be at the minimum
mass recommended for use in certificates. This state-
ment is, at least, discussable. In reality, the problems
are quite different for the cases of between-units and
within-units homogeneity testing.
There is less disagreement about the fact that the
principal requirement for any method to be used for
homogeneity testing of reference materials must be its
high repeatability, as this property determines the abili-
ty of the method to detect small differences between
analyzed samples. Therefore, it is even accepted that a
method that would not be acceptable for use in certifi-
cation measurements (because of a lack of traceability)
but that is able to achieve a very low coefficient of var-
iation is fully appropriate for homogeneity testing.
Moreover, the chosen method should be applied under
optimal conditions of repeatability, i.e. that, as far as
possible, all samples should be measured on the same
day, with the same instrument, by the same operator
and against the same calibrants, and that, if the number
of samples is too large to be measured in one day,
measures must be taken in order to minimize the con-
sequences of any possible drift in the measurements
[3].
Between-units homogeneity testing
Unless homogeneity testing is included in the certifica-
tion exercise itself in the form of a “nested design”, be-
tween-units homogeneity testing is generally carried
out prior to certification and after the candidate refer-
ence material has been packaged in its final form.

When appropriate, the assessment of the homogeneity
of the candidate CRM may be performed on randomly
chosen units, but in general it is preferred to carry out
between-units homogeneity testing on samples which
are taken from production either in a systematic [2] or
in a stratified random way [3], e.g. 1 from every 50 or
100 units produced in a continuous process. To investi-
gate the existence of trends due to the preparation or
filling process, the homogeneity measurements are then
carried out in such a way that drifts in the applied test-
ing method can be detected and corrected for. This as-
pect is not further discussed in this paper. In cases
where physical phenomena which might lead to possi-
ble inhomogeneity are well understood, sample selec-
tion in a systematic manner is used to increase the pow-
er of the test. If the material is prepared in one single
batch, the number of units to be tested, e.g. as recom-
mended by BCR (Community Bureau of Reference) is
10 to 30, depending on the estimated likelihood of in-
homogeneity. If, on the contrary, there is in principle a
reason to define sub-batches, e.g. when the material is
composed of several individual preparations sampled
into units without prior mixing or when parallel lines
are used to fill containers, at least 6 units from each
sub-batch should be examined.
As the intention of any between-units homogeneity
testing should be to demonstrate that the difference be-
tween the units is negligible as compared to the certi-
fied uncertainty interval, the best estimate of the rela-
tive value of each unit must be determined. This can be
achieved by using a method with the highest repeatabil-
ity (smeth), by using an optimal sample size, and by in-
creasing the number of replicates (n) carried out on
each unit. The experimentally measured uncertainty
(uexp) is always an overestimation of the real between-
units variation (ubetw), as it combines the uncertainties
of both the analytical measurement (umeas) and the be-
tween-units variation itself:
u 2exppu 2meascu 2betw
where it should be noted that the analytical measure-
ment uncertainty umeas corresponds to:
umeaspsmeth / ;n
n being the number of replicates carried out on each
unit.
However, if the method uncertainty smeth can be esti-
mated separately, a better estimate of the between-
units variation can be obtained after correction of uexp
for umeas. In non-destructive analysis this can be ob-
tained by carrying out repeated measurements on a sin-
gle sample. In destructive analysis, however, the esti-
mation of smeth is more difficult: if a homogeneous sam-
ple (e.g. an aqueous solution) is available, it is possible
to carry out repeated measurements, but if the estima-
tion of smeth is limited to the detection step only, smeth is
53
underestimated, thus leading to a conservative estima-
tion of ubetw. In most cases, the sample is solid and the
analysis comprises different steps such as digestion, ex-
traction, derivatization, separation etc, and it is very
difficult to distinguish between these contributions and
the material inhomogeneity. An estimate of this inho-
mogeneity contribution can, however, be made by
carrying out repeated measurements on samples of var-
ious sizes, provided smeth is kept constant, e.g. by vary-
ing the volumes of chemicals used proportionally. The
final judgement on how to correct for umeas must, how-
ever, remain with the certifying officer (as defined in in
the draft revised ISO Guide 31 [7]) taking responsibili-
ty for the certification of the reference material.
However, if ubetw is calculated as
u 2betwpu 2expPu 2means
the estimate of ubetw carries an uncertainty equal to the
combined uncertainties of the estimates of uexp and
umeas, i.e.
u(ubetw) 2p(uexp / ;2n ) 2c(umeas / ;2n ) 2
where npdegrees of freedom.
Consequently, ubetw can never be assigned a value
smaller than the value of the combined uncertainties of
uexp and umeas, i.e.
ubetw6;(uexp / ;2n ) 2c(umeas / ;2n ) 2
The possible outcome of between-units homogeneity
testing is that
– either the between-units variability ubetw is negligi-
ble as compared to the certified uncertainty interval
(Fig. 1), in which case inhomogeneity can be ignored;
the uncertainty determined for the mean value of the
batch can then be used to describe the uncertainty of
the CRM, or
– the between-units variability ubetw is a significant
factor in the total uncertainty (Fig. 3), in which case the
batch should either be rejected or reworked, or each
unit should be certified individually.
Most frequently, however, the between-units varia-
bility ubetw is of the same order as the uncertainty of the
mean value (Fig. 4), and as in general it is preferred to
assign a single certified value to all units of the entire
CRM batch, one of the terms making up the uncertain-
ty of the CRM should be this uncertainty associated
with the (possible) between-units inhomogeneity of the
material.
This can be done by basing the CRM uncertainty on
the so-called “statistical tolerance interval” of the ho-
mogeneity study:
x̄Bkb7s
2
where
x̄pan estimate of the mean m
span estimate of the measure of the dispersion s
54

Fig. 3 Example of a homo-

geneity study in which not
only is a significant hetero-
geneity detected (Fexp 1 F0.05),
but it is also significant com-
pared to the uncertainty inter-
val of the certification measur-
ements (unc. cert.)

Fig. 4 Example of a homo-

geneity study in which the de-
tected heterogeneity is signifi-
cant (Fexp 1 F0.05) and compar-
able to the uncertainty inter-
val of the certification measur-
ements (unc. cert.)

kbpthe
2 factor for two-sided tolerance limits for normal
distributions for a number of samples (n), a proportion
(p) of the total population to be covered (e.g. 0.95) and
a probability level (1-a) to be specified (e.g. 0.95)
Generally, the 95%/95% tolerance interval, which cor-
responds to the interval containing the value of 95% of
all units with a probability of 95%, is used [8].
This approach is, however, considered as being quite
conservative, and, since the publication of the Guide to
the Expression of Uncertainty in Measurement, a more
appropriate approach would certainly be to include the
between-units variability ubetw in the combined uncer-
tainty (uc) calculated according to the GUM (Guide to
the Expression of Uncertainty in Measurement) [6]:
Upk7ucpk7;S u 2i cu 2betw
with: ui being the various uncertainty contributions of
other origin.
This issue is further discussed in a paper on “The
Determination of the Uncertainty of Reference Materi-
als Certified by Laboratory Intercomparison” [9].
Within-unit homogeneity testing
Very often, CRMs are not prepared in portions for sin-
gle use, and (small) test portions of a single unit are
used for analysis. In such cases, it is equally necessary
to check whether sub-samples from the same unit are
equally “identical” in terms of their certified properties
and, moreover, down to which test portion this identity
remains valid. Here also, the same general principles
apply as for between-units homogeneity testing, be it
that either different test portion sizes may be examined
(see e.g. [3]) or that microanalytical techniques may be
used to derive experimentally the minimum test por-
 55

tion down to which the certified uncertainties still re-
main valid [10]. Also here, the experimental within-unit
uncertainty (uexp) is an overestimate of the real within-
unit uncertainty, as it combines the uncertainties of
both the analytical measurement (umeas) and the mate-
rial inhomogeneity (uinh):
u 2exppu 2meascu 2inh
and, if the uncertainty of the analytical measurement
can be estimated separately, a better estimate of the in-
homogeneity within each unit can be obtained after
correction of uexp for umeas. One must, however, be
aware that the uncertainty of the analytical measure-
ment used for the determination of uinh is in general
different from the one used to correct the between-
units uncertainty, as, except in non-destructive analysis,
each portion can only be analyzed once, and therefore
umeaspsmeth.
Here also, if uinh is calculated as
u 2inhpu 2expPu 2meas
the estimate of uinh can never be assigned a value smal-
ler than the value of the combined uncertainties of uexp
and umeas, i.e.
uinh6;(uexp / ;2n ) 2c(umeas / ;2n ) 2
An essential difference between within-unit and be-
tween-units homogeneity testing is, however, that if the
observed within-unit inhomogeneities are significantly
larger than the certified uncertainties for test portions
of a given size, one may recommend the use of larger
test portions on the basis of the fact that the uncertain-
ty due to material inhomogeneity (uinh) is inversely
proportional to the square root of the mass of the ana-
lyzed test portion [11]. It is on the basis of this property
that microanalysis was effectively proposed to deter-
mine experimentally the minimum test portion down to
which CRM certificates remain valid [10].
If it is planned to carry out several analyses on each
individual unit tested for between-units homogeneity, it
may be preferable to estimate both the within- and the
between-unit homogeneity from a one-way analysis of
variance. In this case one must, however, be aware that
one is then obliged to compromise on the chosen test
portion size, as the optimal test portion sizes for within-
and between-unit homogeneity testing are quite contra-
dictory.
In the case that the methods used to carry out the
certification measurements are sufficiently precise and
the number of data sufficiently large, expensive sepa-
rate homogeneity testing may be avoided by demon-
strating homogeneity via a so-called “nested design” of
the certification exercise. If such a procedure is used,
which can however only be recommended if the risk of
inhomogeneity is very low and if all methods used in
the certification exercise have a high repeatability, each
laboratory analyzes several portions from each of sev-
eral units. The relevant uncertainties are then derived
either from a two-way analysis of variance if only be-
tween-units homogeneity is tested or from a three-way
analysis of variance if both within-unit and between-
units homogeneity are tested [3].

References
1. Certification of reference materials –
General and statistical principles
(1989) ISO Guide 35
2. Quality system guidelines for the pro-
duction of reference materials (1996)
ISO Guide 34
3. Guidelines for the production and
certification of BCR reference mate-
rials (1997) Document BCR/01/97
4. Schiller SB (July 1996) Statistical as-
pects of the certification of chemical
batch SRMs. NIST Special Publica-
tion 260-125
5. Pauwels J, Lievens F, Ingelbrecht C
(1989) Certification of a Uranium-238
dioxide Reference Material for neu-
tron dosimetry. EUR-12438 EN
6. Guide to the Expression of Uncer-
tainty in Measurement (1995) ISO,
ISBN 92-67-10188-9
7. Contents of certificates, certification
reports and labels of reference mate-
rials, ISO Guide 31 (under revision in
1997)
8. Steyaert H (1988) Statistical Toler-
ance Intervals. EUR-11639
9. Pauwels J, Lamberty A, Schimmel H,
Accred Qual Assur (submitted for
publication)
10. Pauwels J, Vandecasteele C (1993)
Fresenius J Anal Chem 345 : 121–123
11. Ingamells CO, Switzer P (1973) Tal-
anta 20 : 547–568

View publication stats