Talanta 64 (2004) 167173

Evaluation of measurement uncertainty in volumetric operations: the

tolerance-based approach and the actual performance-based approach
Rouvim Kadis

Laboratory of State Standards in Analytical Measurements, D.I. Mendeleyev Institute for Metrology, 19 Moskovsky Pr., 198005 St. Petersburg, Russia
Received 8 July 2003; received in revised form 24 December 2003; accepted 6 February 2004
Available online 27 March 2004
Abstract
Although relatively small compared to other sources of uncertainty in an analytical procedure, uncertainty in volumetric operations needs
to be properly evaluated. In this paper, the problem of volumetric uncertainty is addressed with the critical examination of the procedure for
its evaluation recommended in the EURACHEM Guide Quantifying Uncertainty in Analytical Measurement. Some characteristic features of
volumetric apparatus as a measuring device are considered in relation to accuracy specication usually expressed in the form of capacity
tolerances. On the basis of the underlying metrology, written standards to which volumetric ware is manufactured, and evidences available
fromexperimental studies, it is shown that the tolerance already includes a randomerror inherent in the use of volumetric apparatus. Therefore,
no additional allowance, except for temperature effects, needs be made if the uncertainty is derived from the tolerance. A detailed analysis
of relevant uncertainty sources is presented, with two different procedures for evaluating the uncertainty identied; one of them relies on the
prescribed tolerance while the other is based on the experimental estimation of the actual performance in the users hand. The uncertainty
budget for each of these two approaches is evaluated, analysed and illustrated with a numerical example.
2004 Elsevier B.V. All rights reserved.
Keywords: Volumetric operation; Capacity tolerance; Measurement uncertainty; Uncertainty budget
1. Introduction
Simple volumetric operations such as preparation of a so-
lution in a volumetric ask, transferring a required volume
of liquid with a pipette, or delivering a known volume with
a burette in titration techniques are the basis of analytical
work. Being a part of most analytical procedures, these op-
erations are commonly performed using volumetric glass-
ware. This might be called a basic measuring instrument in
an analytical laboratory, after the balance, and with just this
all chemistry students begin their laboratory exercises in a
course of analytical chemistry. And as with every measur-
ing instrument and operation, the question of the accuracy,
which they can ensure, has always been important.
It is a general requirement now to accompany the result
of a measurement with a statement of its uncertainty, which
is needed to judge the results properly. Therefore, much at-
tention is being given to the methods of evaluation of mea-
surement uncertainty. Following the principles laid down in

Tel.: +7-812-323-9644; fax: +7-812-327-9776.

E-mail address: [email protected] (R. Kadis).
the ISO Guide to the Expression of Uncertainty in Measure-
ment [1], the problemamounts to identifying, evaluating, and
budgeting all practically signicant uncertainty sources in-
volved in a measurement and thus calculating the combined
uncertainty. As applied to analytical measurementby this
term quantitative chemical analysis is meantthe detailed
methodology was developed in a specic guidance docu-
ment, Quantifying Uncertainty in Analytical Measurement,
issued by EURACHEM in 1995 and in revised form (jointly
with CITAC) in 2000 [2]. This guide provides a number of
examples, relating to different analytical problems; they il-
lustrate the uncertainty estimation process in detail.
Although relatively small compared to other sources of
uncertainty in an analytical procedure, volumetric uncer-
tainty needs to be properly evaluated. A practical way of
doing this is to use the capacity tolerances for volumetric
apparatus, taken from the specication. Additionally, an un-
certainty contribution due to the temperature dependence
of volume needs to be accounted for, derived from possi-
ble temperature variations. Note that both of the uncertainty
components can be evaluated based on available informa-
tion, with pen and paper only.
0039-9140/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.talanta.2004.02.005
168 R. Kadis / Talanta 64 (2004) 167173
Meanwhile, the procedure for the estimation of volu-
metric uncertainty, presented in the available examples
([2] (Appendix A), [3,4]), is more complicated than that
outlined above. This suggests accounting not only for the
manufacturers tolerance but also (and obligatory) for the
random variation that should be estimated in a volumet-
ric repeatability experiment. A series of ll-and-weigh or
ll-delivery-weigh operations with water is performed in
such an exercise to get the experimental standard deviation.
Thus, it appears that experimentation is necessary to inform
the uncertainty estimate.
Two questions arise immediately. The rst concerns the
randomvariation inherent in the use of volumetric apparatus.
Has it not really been included in the stated tolerance so
that no additional allowance would be required? The second
question is as follows. If nevertheless the actual performance
has been studied, why should it be used in calculating the
uncertainty along with the tolerance, not instead of it, so
resulting in redundancy in uncertainty estimation?
In subsequent sections, different aspects of specifying per-
formance of volumetric laboratory apparatus are highlighted
in the context of the problem of uncertainty estimation. Us-
ing a cause-and-effect analysis, a detailed uncertainty bud-
get is evaluated, with two alternative procedures identied
for the determination of volumetric uncertainty. One of them
relies on the prescribed tolerance while the other is based on
the experimental estimation of the performance in the users
hand. Which of these two ways to follow depends on actual
circumstances, chiey, the level of accuracy required in the
volume measurement.
In our treatment of the problem, we adhere to the prin-
ciple, known in quality issues as tness for purpose, that
measurement should be made with that level of uncertainty
that is required for its intended application. Adequate ef-
fort is essential in the uncertainty estimation process just as
much. From this point of view, the procedure for evaluat-
ing volumetric uncertainty, recommended in the guide [2],
seems to be excessive. The aim of this note is to make this
clear and avoid overcounting.
2. Volumetric apparatus and its specied accuracy
The measurement of volume in analytical procedures
falls, in the language of metrology, within a category of
so-called direct measurement in which the measurement
process is limited to the use of a single, direct-reading
measuring instrument only. Such an instrument, being
commercially available, shall meet certain metrological re-
quirements, normally in the form of maximum permissible
error, specied by a written standard to which the instru-
ment is made. If the compliance with the specication is
originally guaranteed by the manufacturer or demonstrated
by an independent authority, it is accepted that the errors
produced while the equipment is in use do not exceed its
specied limit of permissible error. This is a basic tenet of
the quality assurance system established for measuring in-
struments in various measurement elds, including volume
measurement. According to this very principle, the limits
of volumetric error laid down in standard specications for
volumetric apparatus have long been used for estimation
of errors occurring in volumetric operations in chemical
analysis [5]. And no extra error contribution, apart from the
capacity tolerance, was required.
Some characteristic features of volumetric apparatus as
measuring device are worth pointing out here. Only in a
broad sense, a volumetric ask, pipette, burette or measuring
cylinder can be called a measuring instrument, though they
are often referred to as such. Actually, all these are mate-
rial measures of volume or, more specically, material mea-
sures of capacity. (The capacity of a vessel is equal to the
volume of liquid contained by or delivered from the vessel
under prescribed conditions.) Material measure, or simply
measure, is a device intended to reproduce or supply, in a
permanent manner during its use, one or more known values
of a given quantity [6]. Among other material measures an
analyst deals with, a balance weight as a measure of mass
and a reference material as a measure of a specic quantity
measured in analytical measurement should be mentioned.
Suppose that a measure of capacity, for instance, a 100 ml
volumetric ask, is used. The quantity 100 ml marked on
the ask expresses the nominal capacity, that is, the nominal
value of the measure. It is this value that is usually recorded
as a result of the volume measurement. The true volume
contained in the ask will naturally not be exactly equal to
the nominal capacity; the difference between the nominal
value and the true value is commonly referred to as the
volumetric error. Since the true value remains unknown in
a volumetric operation, so is the volumetric error that is
considered. In practice, it is important, however, only to
ensure that the error does not exceed an established limit and
the question emerges of how this limit is specied. We will
return to this point in Section 3, restricting the consideration
to some general remarks here.
The total volumetric error is conveniently divided into two
components: (1) intrinsic error apparent when the measure
is used under the reference temperature, usually 20

C; and
(2) inuence error caused by the departure from that tem-
perature. As the contribution of the inuence error is mainly
determined by the properties of the liquid to be measured,
the intrinsic error alone is specied as a tolerance limit. This
is a standard way of expressing the accuracy of volumet-
ric apparatus (and measuring instruments); it is also used
for classifying them according to their accuracy. So, the ca-
pacity tolerances for volumetric glassware have been estab-
lished corresponding to two accuracy classes: class A and
class B, with the tolerance of class B volumetric glassware
being approximately twice those of class A.
The use of volumetric apparatus involves a series of
operations such as lling the vessel, setting or reading the
meniscus against a reference line or scale, and draining if
the device is intended for delivery. In other words, the value
R. Kadis / Talanta 64 (2004) 167173 169
of capacity is reproduced by the measure by following
some operating procedure that an analyst has to implement.
This leads to the situation where the accuracy inherent in
this material measure, as opposed, say, to the accuracy of
a weight, cannot be evaluated in isolation, by ignoring the
contribution of a procedural error that is inseparable.
If however, all signicant error sources in a measurement
process are known and kept under control by following an
operating procedure, the limits to the procedural error and
hence the total measurement error may be inferred from rel-
evant consideration without a great risk of being incorrect.
It is therefore conventional for handeye-operated volumet-
ric apparatus to incorporate the random error contribution,
typical for its proper use, into the limit of volumetric error.
Normal variations in manipulating and reading volumetric
glassware are supposed to be included in the specied tol-
erance.
By contrast, a different situation arises with the mechan-
ical action, piston-operated volumetric apparatus, for in-
stance, piston pipettes that dispense their specied volume.
Besides the operators technique, many other factors such
as the instruments state of repair, environmental conditions,
physical properties of the liquid being delivered, affect the
performance of those pipettes; and as with any mechanical
device it deteriorates over time. Further, the factory-set ad-
justment of a piston pipette can be altered by the user, di-
rectly creating a bias in the instrument. Therefore, for such
delivering devices the tolerance limits on both random and
systematic errors are separately set in the specication. Since
the performance of a mechanically operated volumetric ap-
paratus needs be determined and controlled experimentally
on a regular basis, evaluation of the uncertainty in its use is
based on the performance tests rather on the specications.
Being a separate issue, this problem is beyond the scope of
the paper.
3. Capacity tolerance and random error: requirements
for meeting standards, and experimental evidence
Let us turn to written standards that have been interna-
tionally established for different aspects of design, spec-
ications and application of laboratory glass volumetric
apparatus [79]. The standards set out principles of speci-
cations applicable to any article of volumetric glassware.
The question of how the limit of volumetric error is spec-
ied is of our concern now.
The limit of error is xed by taking into account of several
factors [7,9]. First, this is a design valuea volume that
should occupy a readily visible length of tube (e.g. the ask
neck) of the maximum diameter allowed. From operational
considerations, the basic linear equivalent of the class A
tolerance has been set equal to 0.4 mm. And an additional
linear allowance for potential parallax error in reading the
meniscus is made, which is also based on the diameter of the
tube. Accordingly, a formula has been derived ([7], Annex
Table 1
One-mark pipette precision and capacity tolerances
Pipette
nominal
capacity (ml)
Repeatability
standard
deviation
a
(ml)
Class A
tolerance
b
(ml)
99.7% condence
interval as a
fraction of the
tolerance
2 0.0018 0.01 0.5
5 0.0029 0.015 0.6
0.0040 0.8
10 0.0041 0.02 0.6
0.0045 0.7
25 0.0065 0.03 0.6
0.0068 0.7
50 0.0092 0.05 0.6
0.0113 0.7
a
The numbers in the rst (and single) line in each row are taken from
[10] (Table XXIV); the numbers in the second line are taken from [11]
as a pooled estimate for three different ways of draining a pipette.
b
Taken from [14] with the symbol omitted.
B), made up of the two components, which relates the limit
of volumetric error to the maximum internal diameter of the
tube at the meniscus.
Concurrently with this, another requirement comes into
play for glassware intended for delivery, where an error may
be signicant due to a variation in the technique of draining
the vessel. For this apparatus, the limit of volumetric error
is prescribed to be not less than four times the experimental
standard deviation obtained under repeatability conditions.
Hence, it follows that a possible run-to-run variation that
can reasonably be expected to be in use has been taken into
account in the tolerance value. Expressed as 95% or 99.7%
condence interval, this error contribution may evidently
range up to a half or, respectively, three-fourths of the limit
of volumetric error.
An example may be taken of One-marked (transfer)
pipettes, which were examined over the years. Table 1
summarises the repeatability standard deviations and the
calculated 99.7% condence intervals compared with the
corresponding class A tolerances for pipettes of capacity
from 2 to 50 ml. The original data have been taken from two
extensive studies [10,11] on random errors made in deliver-
ing from pipettes of the ordinary type. (Another study [12],
in which different contributions to the overall precision
were also investigated, resulted in values of the standard
deviation that were 510 times lower than those found in
the rst two studies, owing to special means of improving
precision. In one more paper [13], precision estimates were
presented, which were associated with the use of 10 and
20 ml pipettes by a class of students, with the repeatability
standard deviation for a 10 ml pipette near to that indicated
in Table 1.) One can see from the data of Table 1 that the
contribution of random error falls between 0.5 and 0.8 of
the tolerance value, which agrees well with the proportion
(3/4) expected from the standard specication.
170 R. Kadis / Talanta 64 (2004) 167173
We emphasise that this feature is characteristic of volu-
metric glassware normally used in a laboratory. Still smaller
random variation should be expected when a higher level
of accuracy is aimed at, particularly, in calibration cervices.
This is quite apparent from the data on long-term precision
in calibration of laboratory glassware, which was gathered
at the US National Bureau of Standards [15,16]. With that
data, presented as a difference three-standard deviation lim-
its, it can be easily shown that the 99.7%condence intervals
do not exceed approximately 0.2 of the class A tolerances
for all articles processed.
The above ndings are evidences in favour of the fact that
the capacity tolerance is the limit to allowable error in the
use of volumetric ware, not specically in its calibration as
the EURACHEM Guide designates in worked examples and
in the summary of Common Sources and Values of Uncer-
tainty ([2], Appendix G). It stands to reason that the pro-
cedures for proper use of the volumetric glassware must be
followed by qualied and motivated personnel. They have
all been written down in the ISO standards cited [7,8] as
well as applicable national standards, not to mention many
textbooks on quantitative analysis (e.g. [17,18]) which in-
clude chapters describing the subject in detail.
4. Identifying uncertainty sources
Using a cause-and-effect diagram, also known as the
shbone or Ishikawa diagram is an effective means of
uncertainty analysis, which helps to identify, explore and
display relevant uncertainty sources. This is a useful teach-
ing tool as well, since it shows the relationships between
the effect and causes responsible for it. Such a diagram for
a volumetric operation is drawn in Fig. 1.
In the diagram, four main branches are depicted for which
contributory cause factors are added where necessary.
Branch 1 (Procedure): It is representative of the procedu-
ral contribution to the total uncertainty, incorporates sev-
eral factors such as cleanliness of apparatus, setting or
Procedure Calibration
balance
performance
air buoyancy
Temperature effects
Volume
mass of water
procedure
cleanness
of vessel
setting/reading
the meniscus
drainage
effects
delivery
time
delivery
technique
temperature
density of water
variation of density
of the liquid
change in capacity
of the vessel
Physical properties
of the liquid
Fig. 1. Cause-and-effect diagram for a volumetric operation.
reading the meniscus, and drainage effects for the appa-
ratus used to deliver liquids. The last component, as in-
uenced by the time of delivery and delivery technique,
is due to the tendency of retaining the liquid on the walls
of the tube during delivery. (For instance, an error due
to after-drainage in using a burette is negligible if it has
a delivery time long enough to allow the liquid to drain
and rejoin the main column. The burette shall meet this
requirement, and minimum times for tubes of different
graduated lengths have been specied in the standards.)
Branch 2 (Temperature effects): Two different effects are
taken into consideration when the temperature of mea-
surement differs from the reference temperature of 20

C.
This is the variation of density of liquid with tempera-
ture and the change in the capacity of the vessel itself
with the change of temperature. The two effects act in
opposition, with the former usually of much greater mag-
nitude than the latter.
Branch 3 (Calibration): It includes its own subsidiary
branches: procedure and temperature, the latter with the
two arrows as explained above, and some additional ef-
fects, specically those associated with mass determina-
tion by weighing. Among them are balance performance
and differential air buoyancy between the weighted ob-
ject (water) and the balance weights. For a small volume
delivered, particularly with a micropipette, it may be also
important to consider the evaporation loss (not shown in
the diagram).
Branch 4 (Physical properties of the liquid): It is mainly
concerned of the delivery processes, takes account of a
difference in properties such as the viscosity and surface
tension of the liquid being measured and water, which can
cause a departure of the measured volume from that stated
in calibration. It is usual to assume that for dilute aqueous
solutions ordinary employed in volumetric analysis these
effects are so small that they can be disregarded. (The
truth of this statement was experimentally demonstrated
[19] as early as at the end of the 19th century. However,
exceptions have long been intimated [20]; and further, the
use of non-aqueous solvents may call into question the
R. Kadis / Talanta 64 (2004) 167173 171
validity of water-related calibration, although the error
was not found to be high in a case study [21].)
Not all of the factors identied in the diagram act as
sources of uncertainty as such, that is, a randomvariable with
zero mean value. Some of them, particularly those involved
in calibration, result in a systematic deviation of the vol-
ume measured subsequently, i.e. a measurement bias. Oth-
ers, such as drainage effects, bring about uncertainty and
bias contribution alike. Still others may appear either as an
uncertainty or as a bias, depending on circumstances. This
last case is true, for instance, in regard to temperature effects
where a possible variation about a mean room temperature
(for a totality of measurements) results in an uncertainty,
but a certain deviation from the reference temperature (for
a particular measurement) results in a bias.
5. Quantifying volumetric uncertainty
Let us consider a typical case where volumetric glassware
with some nominal capacity and manufacturers tolerances
is used, and our task is now to evaluate measurement un-
certainty in the volumetric operation. The question is to
what extent the capacity tolerance carries built-in infor-
mation about possible error, from which the uncertainty
can reasonably be derived. From what has been said in the
previous sections it appears that all the inuence factors re-
lating to the upper branches, procedure and calibration, in
the cause-and-effect diagram are covered in the stated toler-
ance. Only the bottom branches, particularly that one which
represents the possible temperature variation, must addition-
ally be allowed for. With this supplementary contribution,
the use of the capacity tolerance converted to the standard
uncertainty following the principles of the ISO Guide [1] is
well justied in estimating volumetric uncertainty.
One may say that this is true when volumetric apparatus
from a reliable manufacturer is dealt with and operated in
the proper way. That is right indeed. If there is a doubt that
an item of volumetric glass has been calibrated properly, it is
generally recommended to test it for accuracy by making a
single calibration. There may be, of course, situations where
it is reasonable to assess the actual single apparatus operator
performance in a precision (repeatability) experiment that
is a calibration experiment in a way. This is necessary, for
example, when a high level of accuracy, not attainable with
the specied tolerances, is required or when manufacturers
specication cannot be condently relied upon (specically,
with mechanical piston-operated apparatus).
Such a calibration experiment is based on the gravimet-
ric method repeatedly applied to determine the actual vol-
ume contained or delivered, with apparatus manipulated in
the same manner as in its use and the number of replicates
not less than seven to ensure a meaningful uncertainty esti-
mate. In this way, the factory calibrated volumetric appara-
tus now becomes in-house calibrated. Then, we can use in
subsequent work the estimated true capacity instead of the
nominal capacity and also the experimental standard devia-
tion of the mean as a more specic estimate of calibration
uncertainty than that derived from the tolerance limit. Minor
contributions to the uncertainty from remaining effects such
as the uncertainty in weighing and the uncertainty of refer-
ence values of the density of water (due to small temper-
ature variation in calibration process) should be combined
with the standard deviation of the mean to form a combined
calibration uncertainty.
An important point is that the uncertainties originating
at the measurement procedure and calibration are both es-
timated here in terms of the experimental standard devia-
tion, so that there is no longer any necessity for utilising
the manufacturers tolerances. Thus, we come to recognise
that in quantifying volumetric uncertainty the specied tol-
erances and the estimated variability shall be used alter-
natively, not conjointly as the guide [2] recommends. We
are dealing with two different procedures that can be re-
ferred to as the tolerance-based approach and the actual
performance-based approach according to whether the vol-
umetric apparatus has been factory calibrated or in-house
calibrated.
An uncertainty budget for these two approaches is drawn
up in Tables 2a and 2b respectively, where different uncer-
tainty components are formulated and quantied with a nu-
merical example taken from actual practice. The items in-
cluded in the budget are just the same as the ones displayed
in the cause-and-effect diagram except for the branch Phys-
ical properties of the liquid, whose contribution is assumed
negligible. With such analysis the distinction in methodol-
ogy becomes clearer for two modes of calibration. In one
case, a tolerance-based uncertainty is associated with the
measurement result equal to the nominal capacity, while in
the other case the estimated true capacity (of a particular
volumetric apparatus) is taken as the result, which is usually
provided with much smaller uncertainty.
6. Uncertainty budget analysis
In general, the accuracy requirements arising from an in-
tended application of a measurement result determine the
method, the instrument, and the conditions under which
the measurement is performed, and as a consequence, the
way the uncertainty is calculated. In Section 5, two differ-
ent procedures for evaluating volumetric uncertainty were
referred to as the tolerance-based approach and the actual
performance-based approach. The uncertainty budget for the
two approaches (Table 2) provides a means of comparing
different contributions to the total uncertainty, which allows
us to judge the efciency and suitability one way or the other.
It is apparent that there is little point in mixing the two
procedures by combining the standard uncertainty derived
from the tolerance with the repeatability standard deviation
from the experiment.
172 R. Kadis / Talanta 64 (2004) 167173
Table 2
Uncertainty budget for a volumetric operation
Uncertainty source Standard uncertainty
Formula
a,b
Example
c
(ml)
(a) The tolerance-based approach
Procedure

v

6
0.082

Calibration
Temperature
d
Vt

6
0.034
Combined standard uncertainty 0.089
Expanded uncertainty (k = 2) 0.18
Result of a measurement 100.00 0.18
(b) The actual performance-based approach
Procedure s 0.014
Calibration
e
Procedure
s

n
0.0044
Balance performance
f
1

s
2
b
+2

2
nl
3
0.00013
Density of water/temperature
g
V

d
w
dt

6
0.0017
Temperature
d
Vt

6
0.034
Combined standard uncertainty 0.037
Expanded uncertainty (k = 2) 0.074
Result of a measurement 100.15 0.07
a
In these formulae:
v
is the capacity tolerance specied by manufacturer, V the nominal capacity for a volumetric apparatus factory calibrated
or the estimated true capacity for that in-house calibrated, the coefcient of cubical thermal expansion of the liquid measusred, t the limit of
possible temperature variation about the mean working temperature, s the standard deviation from the repeatability (calibration) experiment with the
number of replicates equal to n,
w
the density of water as calibrating uid, s
b
the repeatability of the balance specied as the standard deviation,
nl
the non-linearity of the balance specied as the maximum allowable deviation from the linear characteristic function, and
t
is the limit of possible
temperature variation in calibration process.
b
The value of

6 in the denominator in the formulae is used to convert the capacity tolerance,
v
, and also the limits of the temperature ranges,
t and
t
, to the respective standard uncertainties based on a symmetric triangular probability distribution of the occurring errors. Yet the value of

3 based on a symmetric rectangular distribution was used to calculate the standard uncertainty associated with the balance non-linearity in the same
manner as in Ref. [22]. For details in choosing an appropriate model probability distribution when uncertainty estimates are made through professional
judgement, so called Type B evaluation, see the ISO Guide [1] (clauses 4.3 and F.2.3.3) and also Ref. [23].
c
As an example, the uncertainties in use of a 100 ml class B volumetric ask were calculated based on the specication [24] (
v
= 0.20 ml) and
the repeatability (calibration) experiment. This experiment involved a series of 10 ll-and-weigh operations with distilled water at 20.4

C (
t
= 0.2

C),
which gave the mean value of capacity V = 100.152 ml and the repeatability standard deviation s = 0.009 ml. The results of testing other pieces
of the same glassware were available, so it was reasonable to use the pooled estimate of the standard deviation, s = 0.014 ml, based on more than
90 observations. The following values of parameters were put to the calculation: = 2.1 10
4
C
1
, t = 4

C,
w
= 1.0 g ml
1
, s
b
= 0.05 mg,

nl
= 0.15 mg, the latter two taken from the balance (Sartorius MC210P) specication.
d
The coefcient of cubical thermal expansion of liquids is at least one order of magnitude higher than that of glass. Therefore, only the rst effect
of the two relating to the temperature inuence was accounted for in the evaluation of uncertainty caused by ambient temperature variation.
e
The sources of bias in calibration such as temperature inuence and air buoyancy are not included in the budget; these effects have normally been
built in the calculation of the volume, adjusted to 20

C, from the observed mass of water by reference to the appropriate table [8,16], with water
temperature, air temperature, and air pressure as input parameters. So, in-house calibration is held to have no bias.
f
In estimating the uncertainty originating from analytical balance, only the two most signicant contributions, due to repeatability and non-linearity,
were taken into account out of others that are relevant according to Ref. [22].
g
In estimating the uncertainty caused by the temperature inuence on the density of water, we set

dw
dt

= 2.1 10
4
g ml
1
C
1
, a good
approximation in the room temperature range.
As is clear from the example relating to a 100 ml class
B volumetric ask, the estimate of uncertainty can be re-
duced signicantly if it is based on the actual performance
rather than the specication. So the combined standard
uncertainty decreases from 0.089 to 0.037 ml in the ex-
ample. This is evidently due to the small uncertainty that
the random variation actually introduces in the budget, and
still smaller remaining contributions to calibration uncer-
tainty. Furthermore, the signicant calibration bias equal
to 0.15 ml is revealed here, emphasising the importance of
R. Kadis / Talanta 64 (2004) 167173 173
in-house calibration. Nevertheless, it does not follow from
these facts that the analyst must necessarily be engaged in
a performance exercise with his volumetric glassware in
order to estimate the (lower) uncertainty. An adequate effort
is required in evaluating the uncertainty in a measurement.
It can be seen that the combined procedurecalibration
contribution to the budget is reduced so that the uncertainty
caused by the temperature variation becomes the largest in
the total uncertainty (Table 2(b)). The standard uncertainty
due to the temperature variation (0.034 ml) was calculated
with the normal assumption that uctuations in temperature
are possible within 4

C about the mean room tempera-

ture. The inference from this graphic example is as follows:
in-house calibration is unjustied when the temperature con-
trol in use of volumetric apparatus is lacking. It is unreason-
able to make calibration with the highest level of accuracy
that is not required and not ensured in the measurement.
The situation may however reverse where the temperature
is under regulatory control owing to more stringent require-
ments being placed on accuracy of the measurement. So, in
the special case that the actual temperature of volumetric
solution is measured, the corresponding uncertainty contri-
bution (0.034 ml) will be reduced to a negligible uncertainty
of the temperature measurement, with the combined uncer-
tainty essentially decreased to the level determined by the
procedure contribution (0.014 ml). The in-house calibration
would be advantageous in such cases. Note that if the actual
temperature is measured, the effect caused by its difference
from the reference temperature is manifested as bias to be
corrected for. Alternatively, thermostatting at 20

C before
setting the meniscus is recommended in the most accurate
work; this cancels the bias directly.
7. Conclusions
The capacity tolerance specied for volumetric apparatus
can suitably be used in evaluating volumetric uncertainty.
This procedure, here named as the tolerance-based approach,
requires only the tolerance value to be converted into the
standard uncertainty following the established rules. It is
essential that normal variations in manipulating and reading
volumetric glassware are understood to be included in the
stated tolerance, and thus no additional allowance, except
the temperature effects, is needed for obtaining the total
uncertainty.
Alternatively, the actual performance has to be estimated
in a repeatability (calibration) experiment, irrespective of the
prescribed tolerance. Much lower uncertainty is attained this
way because of the small uncertainty that the random vari-
ation actually introduces in the budget, and the still smaller
remaining contributions to calibration uncertainty that are
relevant. As a consequence, the uncertainty caused by the
temperature variation may become the dominating contri-
bution to the total uncertainty where the temperature in use
of volumetric apparatus varies within several degree celcius.
Thus, the actual performance-based approach is advanta-
geous provided the precise temperature control is exercised.
References
[1] BIPM, IEC, IFCC, ISO, IUPAC, IUPAP, OIML, Guide to the Ex-
pression of Uncertainty in Measurement, International Organization
for Standardization, 1993.
[2] EURACHEM/CITAC Guide, Quantifying Uncertainty in Analytical
Measurement, second ed., 2000.
[3] N.T. Crosby, et al., F.E. Prichard (co-ordinating author), in: E.J.
Newman (Ed.), Quality in the Analytical Chemistry Laboratory (An-
alytical Chemistry by Open Learning), Wiley, Chichester, 1995,
pp. 286289.
[4] R. Lawn, F.E. Prichard (co-ordinating author), Practical Laboratory
Skills Training Guides: Measurement of Volume, LGC, Teddington,
2003, pp. 1821.
[5] K. Eckschlager, Fehler bei chemischen Analysen, Geest & Portig,
Leipzig, 1964, pp. S.3436.
[6] BIPM, IEC, IFCC, ISO, IUPAC, IUPAP, OIML, International Vo-
cabulary of Basic and General Terms in Metrology, second ed., In-
ternational Organization for Standardization, 1993.
[7] ISO 384 Laboratory GlasswarePrinciples of Design and Construc-
tion of Volumetric Glassware, International Organization for Stan-
dardization, 1978.
[8] ISO 4787 Laboratory GlasswareVolumetric GlasswareMethods
for Use and Testing of Capacity, International Organization for Stan-
dardization, 1984.
[9] ASTM E 694 Standard Specication for Laboratory Glass Volumetric
Apparatus, ASTM International, 1999.
[10] E.J. Conway, Microdiffusion Analysis and Volumetric Error, fth
ed., Crosby Lockwood, London, 1962, Chapter LV, pp. 356371.
[11] J. Boche nska, A. Rokosz, Chem. Anal. (Warsaw) 8 (1963) 327.
[12] G.A. Dean, J.F. Herringshaw, Analyst 86 (1961) 434.
[13] A.R. Rogers, J. Pharm. Pharmacol. 10 (Suppl.) (1958) 98T.
[14] ISO 648 Laboratory GlasswareOne-Mark Pipettes, International
Organization for Standardization, 1977.
[15] J. Lembeck, The Calibration of Small Volumetric Laboratory Glass-
ware (NBSIR 74461), Appendix 2, National Bureau of Standards,
Washington, DC, 1974.
[16] ASTM E 542 Standard Practice for Calibration of Laboratory Volu-
metric Apparatus, ASTM International, 2001.
[17] I.M. Kolthoff, E.B. Sandell, E.J. Meehan, S. Bruckenstein, Quanti-
tative Chemical Analysis, fourth ed., Macmillan, London, 1969.
[18] A.I. Vogel, Vogels Textbook of Quantitative Chemical Analysis, fth
ed. (revised by G.H. Jeffery et al.), Longman Sci & Techn, Harlow,
1989.
[19] J. Wagner, Z. Physik. Chem. 28 (1899) 193.
[20] W. Schloesser, C. Grimm, Chem.-Ztg. 30 (1906) 1071.
[21] E.B. Buchanan, Talanta 13 (1966) 1599.
[22] A. Reichmuth, Weighing AccuracyEstimating Measurement Bias
and Uncertainty of a Weighing, Mettler Toledo, 2001 (down-
loaded as weighing accuracy 2.1.pdf at http://www.mtlabservice.com/
general 04measurement uncert.htm).
[23] R.L. Kadis, Meas. Tech. 43 (2000) 403 (translated from
Izmeritelnaya Tekhnika, Moscow).
[24] ISO 1042 Laboratory GlasswareOne-Mark Volumetric Flasks, In-
ternational Organization for Standardization, 1998.