Handbook51version4xUncertentyInEnvLab.pdf

See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/237364262
Handbook for Calculation of Measurement Uncertainty in Environmental
Laboratories
Article · January 2012
CITATIONS
206
READS
8,688
4 authors, including:
Some of the authors of this publication are also working on these related projects:
P2-Measurement uncertainty in medical laboratory View project
UnipHied View project
Bertil Magnusson
Trollboken AB
63 PUBLICATIONS   2,005 CITATIONS
SEE PROFILE
Teemu Näykki
Finnish Environment Institute
24 PUBLICATIONS   295 CITATIONS
SEE PROFILE
Mikael Krysell
Swedish Meteorological and Hydrological Institute
27 PUBLICATIONS   689 CITATIONS
SEE PROFILE
All content following this page was uploaded by Bertil Magnusson on 14 May 2019.
The user has requested enhancement of the downloaded file.

Handbook
for
Calculation of
Measurement Uncertainty
in
Environmental Laboratories
Edition 4 November 2017
Authors
Bertil Magnusson, Trollboken AB, Sweden
Teemu Näykki, SYKE, Finland
Håvard Hovind, Norway
Mikael Krysell, Swedish Agency for Marine
and Water Management
Eskil Sahlin, RISE, Sweden
Drawings by Petter Wang, Norway
Valuable comments on the
contents have been provided by:
Rolf Flykt, Sweden
Irma Mäkinen, Finland
Ulla O. Lund, Denmark
Steve Ellison, UK
Ivo Leito, Estonia
Downloading address:
www.nordtest.info
Recommended citation
This document should be cited* as “B. Magnusson, T. Näykki, H. Hovind, M. Krysell, E. Sahlin,
Handbook for calculation of measurement uncertainty in environmental laboratories, Nordtest Report
TR 537 (ed. 4) 2017. Available from www.nordtest.info.
*Subject to journal requirements

FOREWORD
The approach of using quality control data and validation data to estimate
measurement uncertainty for routine analyses is now well established. This Nordtest
handbook TR537, describing the approach is available in several languages at
www.nordtest.info. First issued in 2003, this 4th
edition is based on experience
gained by many laboratories using this approach to estimate measurement
uncertainty. Major updates in the 2017 version are as follows:
 Uncertainty over the measurement range – a separate section on estimating
measurement uncertainty over the measurement range, in either absolute
units or relative units, has been added, this being one of the major
difficulties of applying this approach.
 Estimating standard deviation from duplicates – a pooled standard deviation
is used instead of a factor applied to the mean range.
 Use of control chart limits for the within-lab reproducibility component
u(Rw) is pointed out more clearly
 Harmonisation with ISO 11352 Water quality — Estimation of measurement
uncertainty based on validation and quality control data. The terminology
has been harmonised with the ISO standard which applies this approach to
measurement uncertainty.
We can also recommend the following resources (see section 2.5) to help users
applying this approach to estimate uncertainty:
 Software MUkit – freely available software following this approach to
estimate uncertainty. An example of a MUkit report is presented in
Appendix 9.
 Online course – a link is given to a course from University of Tartu
presenting in detail this approach to estimate uncertainty.

Contents
1 DEFINITIONS, ABBREVIATIONS AND SYMBOLS..............................................1
2 INTRODUCTION........................................................................................................3
2.1 SCOPE AND FIELD OF APPLICATION...........................................................................3
2.2 COMMENT TO DATA USERS.......................................................................................4
2.3 LADDER OF ERRORS..................................................................................................5
2.4 ABOUT MEASUREMENT UNCERTAINTY....................................................................6
2.5 EXTERNAL RESOURCES ............................................................................................7
3 CALCULATION OF EXPANDED UNCERTAINTY, U - OVERVIEW ..................8
3.1 CUSTOMER NEEDS ....................................................................................................9
3.2 FLOW SCHEME FOR UNCERTAINTY CALCULATIONS ..................................................9
3.3 SUMMARY TABLE FOR UNCERTAINTY CALCULATIONS............................................11
4 UNCERTAINTY OVER THE MEASUREMENT RANGE .....................................12
4.1 RELATIONSHIP BETWEEN MEASUREMENT UNCERTAINTY AND CONCENTRATION....12
4.2 USING REPLICATE RESULTS TO DIVIDE THE MEASUREMENT RANGE........................13
4.3 CALCULATION OF UNCERTAINTY OVER THE MEASUREMENT RANGE ......................14
5 WITHIN-LABORATORY REPRODUCIBILITY - u(Rw) .......................................15
5.1 CUSTOMER DEMANDS.............................................................................................15
5.2 CONTROL SAMPLE COVERING THE WHOLE ANALYTICAL PROCESS..........................16
5.3 CONTROL SAMPLES AND ROUTINE SAMPLE REPLICATES.........................................16
6 METHOD AND LABORATORY BIAS – u(bias)....................................................19
6.1 CERTIFIED REFERENCE MATERIAL.........................................................................20
6.2 PROFICIENCY TESTS ...............................................................................................21
6.3 RECOVERY .............................................................................................................23
7 REPRODUCIBILITY BETWEEN LABORATORIES, sR........................................24
7.1 DATA GIVEN IN A STANDARD METHOD ...................................................................24
7.2 DATA FROM PROFICIENCY TESTING ........................................................................25
8 EXAMPLES...............................................................................................................26
8.1 AMMONIUM IN WATER ...........................................................................................26
8.2 BOD IN WASTEWATER ...........................................................................................26
8.3 PCB IN SEDIMENT ..................................................................................................30
9 REPORTING UNCERTAINTY ................................................................................33
10 REFERENCES...........................................................................................................35
11 APPENDICES............................................................................................................37
11.1 APPENDIX 1: EMPTY FLOW SCHEME FOR CALCULATIONS ..................................37
11.2 APPENDIX 2: EMPTY SUMMARY TABLE..............................................................38
11.3 APPENDIX 3: ERROR MODEL USED IN THIS HANDBOOK......................................39
11.4 APPENDIX 4: UNCERTAINTY OF BIAS FOR NH4-N IN SECTION 3.2......................40
11.5 APPENDIX 5: RAW DATA FOR NH4-N - DUPLICATES ..........................................41
11.6 APPENDIX 6: RAW DATA FOR DISSOLVED OXYGEN IN SECTION 5.3 ...................42
11.7 APPENDIX 7: RAW DATA FOR BOD IN SECTION 8.2 ...........................................43
11.8 APPENDIX 8: TEMPLATE FOR EVALUATION OF UNCERTAINTY ...........................44
11.9 APPENDIX 9: UNCERTAINTY CALCULATIONS USING MUKIT..............................49

of 51
1 Definitions, abbreviations and symbols
The main sources here for edition 4 are edition 3 of this handbook, VIM [10] and
the ISO 11352 Water quality — Estimation of measurement uncertainty based on
validation and quality control data [5].
bias Estimate of systematic measurement error [10].
Difference between mean measured value from a large series of test
results and an accepted reference value (a certified or assigned value).
The measure of trueness [10] is normally expressed in term of bias.
CRM Certified Reference Material
PT Evaluation of participant performance against pre-established criteria
by means of interlaboratory comparisons, also called external quality
assessment
Symbols
RMSbias The root mean square of the individual bias values [5] (bRMS in ref [5])
n
biasi
 2
)
(
s An estimate of the population standard deviation  from a limited
number (n) of observations (xi)
sr Standard deviation under repeatability conditions
repeatability conditions
observation conditions where independent test/measurement results
are obtained with the same method on identical test/measurement
items in the same test or measuring facility by the same operator using
the same equipment within short intervals of time [5]
sRw Standard deviation under within-laboratory reproducibility conditions
within-laboratory reproducibility
intermediate measurement precision where variations within one
laboratory alone are included [5]
Comment: sRw, intermediate measure between sr and sR. An alternative
name is intermediate precision [5]. The sRw can be estimated from a
control sample over a certain period of time, preferably at least one
year.
sR Standard deviation under reproducibility conditions
reproducibility conditions
observation conditions where independent test/measurement results
are obtained with the same method on identical test/measurement
items in different test or measurement facilities with different
operators using different equipment [5]
x Mean value
u(bias) The uncertainty component associated with (possible) method and
laboratory bias [5] (ub in ref [5])

of 51
u(Cref) Standard uncertainty component for the certified or assigned value - a
mean value of the individual uncertainties, u(Crefi) [5]
( Cref
u in ref [5]).
u(Rw) Standard uncertainty component for the within-laboratory
reproducibility [5]
u(x) Standard uncertainty
uc Combined standard uncertainty
U Expanded uncertainty, normally close to 95 % confidence interval
Repeatability, sr
Within-lab reproducibility, sRw
Combined standard uncertainty, uc
Reproducibility between laboratories, sR

of 51
2 Introduction
2.1 Scope and field of application
This handbook is written for environmental testing laboratories in the Nordic
countries. Its purpose is to give support to those implementing the concept of
measurement uncertainty following the principles in GUM [1] for routine
measurements. However, the approach is very general and should be applicable to
most analytical laboratories. The aim is to provide a practical, understandable and
common approach of measurement uncertainty calculation.
This approach is mainly based on already existing quality control and validation
data, according to the European accreditation guideline [2], the Eurolab Technical
Report [3] and the ISO 21748 [4]. The approach is also presented in detail in ISO
11352 [5] and consistent with the requirements of ISO/IEC 17025 [6]. Nordtest has
supported this project financially in order to promote and enhance harmonisation
between Nordic laboratories.
Practical examples, taken directly from the everyday world of environmental
laboratories, are presented and explained. The approach given in this handbook is
presented in detail for determination of acrylamide in snacks with LC-MS in an
online course "Estimation of measurement uncertainty in chemical analysis" from
the University of Tartu1
.
The handbook covers all steps in the analytical chain from the arrival of the test
sample in the laboratory to the reporting of the analytical result. It is important to
notice that vital parts of the total measurement uncertainty are not included, e.g.
sampling, sample transportation and possible gross errors during data
storage/retrieval. Regarding sampling uncertainty there is the Nordtest handbook,
Uncertainty from Sampling [20].
While the recommendations presented do form a valid approach to the evaluation
of measurement uncertainty for many purposes, other suitable approaches may also
be adopted. Especially the Eurolab Technical Report [3] and the Eurachem/CITAC
Guide [7] are useful where several different approaches are presented with detailed
examples and the concept of measurement uncertainty is fully described.
Basic knowledge of terminology [10] and the use of quality control and statistics
[9] are required to follow the calculations presented here. In order to make it
possible for the reader to follow the calculations, raw data is given in appendices.
1
The presentations are given under heading 10. The single-lab validation approach,
www.sisu.ut.ee/measurement/uncertainty, accessed 2017.

of 51
2.2 Comment to data users
Previously, laboratories reported within-laboratory reproducibility sRw, calculated
from data taken from an internal quality control covering the whole analytical
process. The expanded measurement uncertainty, U, also taking into account
method and laboratory bias variation and using a coverage factor of 2, can give
values which may be a factor of 2 to 5 times higher. However, this does not reflect
a change in the performance of the laboratory, just a much better estimation of the
real variation between laboratories.
In Figure 1, the ammonium nitrogen, NH4-N, results from two laboratories are in
good agreement – the difference is about 5 %. You can see this to the right of the
chart where measurement uncertainty is presented, using the coverage factor of 2,
but not on the left, where the sRw from internal quality control is given.
150
160
170
180
190
200
210
220
230
240
250
0 6
Ammonium
µg/l
Lab 1
Lab 2
Measurement
uncertainty from...
Control sample
Rw
Control sample + CRM or
interlaboratory comparisons
Figure 1. Comparing NH4-N results from two laboratories, Lab 1 = 199 µg L-1
and
Lab 2 = 188 µg L-1
. To the left the error bars are calculated from internal control,
± sRw, and to the right the error bars are expanded measurement uncertainty, ± U.

of 51
2.3 Ladder of errors
As described in Nordtest TR 569 [9] the sources of errors affecting the possible
deviation from a reference value for an analytical result can be described by the
ladder presented in Figure 2.
For an individual determination on a test sample in a certain matrix the four
different steps in the ladder are as follows: 1) the method as such, 2) the method as
it is used in the laboratory, 3) the day-to-day variation in the laboratory, 4) the
variation within an analytical run – repeatability.
Figure 2. Ladder of errors in a measurement
Step 1 - The method bias – a systematic error owing to the method used
Step 2 - The laboratory bias – a systematic error (for an individual laboratory)
Step 3 - The day-to-day variation – a random error occurring between replicated
determinations performed different days in a laboratory over a long period of time
Step 4 - The repeatability – a random error occurring between replicate
determinations performed within a short period of time; inhomogeneity is part of
the repeatability
Each of these steps on the ladder adds to the uncertainty. The measurement
uncertainty normally consists of all four steps. This handbook demonstrates how
certified reference materials (or synthetic control samples), proficiency testing or
recovery tests are used for estimating u(bias), step 1 and 2.
The uncertainty component for within-laboratory reproducibility u(Rw) consists of
step 3 and 4. The u(Rw) can be estimated using repeated measurements of a control
sample over a long period of time, provided that the control sample has similar
matrix and concentration as the test samples and goes through the whole analytical
process. Repeatability can be estimated separately using replicates of routine
samples analysed within the same analytical run.

of 51
2.4 About Measurement Uncertainty
What is measurement uncertainty?
 The number after ±
 All measurements are affected by a certain error. The measurement uncertainty
tells us what size the measurement error might be. Therefore, the measurement
uncertainty is an important part of the reported result.
 Definition: Measurement uncertainty is ”non-negative parameter characterizing the
dispersion of the quantity values being attributed to a measurand, based on the
information used” [10]
Who needs measurement uncertainty?
 The data user/customer needs it together with the result to make a correct
decision. The uncertainty of the result is important, e.g. when looking at
allowable (legal) limits.
 The laboratory, to verify its´ own quality of measurement
Why should the laboratory report measurement uncertainty?
 As explained above, the data user/customer need it in order to be able to make
correct decisions
 An estimation of the measurement uncertainty is required in ISO 17025 [6]
How is measurement uncertainty obtained?
 The basis for the evaluation is a measurement and statistical approach, where
the different uncertainty sources are estimated and combined into a single
value
 “Basis for the estimation of measurement uncertainty is the existing knowledge
(no special scientific research should be required from the laboratories).
Existing experimental data should be used (quality control charts, validation,
interlaboratory comparisons testing, CRM etc.)” [2].
 Guidelines are given in GUM [1], further developed in, e.g., EA guidelines [2],
the Eurachem/CITAC guide [7], in a Eurolab technical report [3] and in ISO
21748 [4]
How is the result expressed with measurement uncertainty?
 Measurement uncertainty should normally be expressed as U, the expanded
measurement uncertainty, with a stated confidence level and a coverage factor,
k. In most cases k = 2, providing a level of confidence of approximately 95 %
 Measurement uncertainty is expressed normally with one digit, maximum 2
digits. In many cases the uncertainty is rounded up e.g. 6.40 % is rounded to
7 % but common sense should prevail so 6.05 % is rounded down to 6 % - see
GUM section 7.2.2 [1]

of 51
 It is often useful to state how the measurement uncertainty was obtained
Example, where ± 7 % is the measurement uncertainty:
Concentration of ammonium nitrogen (NH4-N) = 148 ± 10 µg L-1
. The
measurement uncertainty, 10 µg L-1
(95 % confidence level, coverage factor
k = 2) is estimated from internal control and from proficiency testing
How should measurement uncertainty be used?
 It can be used as in Figure 1, to decide whether there is a difference between
results from different laboratories
 It is necessary when comparing results to allowable limits, e.g. specifications
or allowable (legal) limits, and when using data for classification of ecological
or chemical status as required by various EU directives
2.5 External resources
Software
MUkit (Measurement Uncertainty Kit) is a
measurement uncertainty software application,
its calculation are mainly based on this
handbook, Nordtest TR537. It is a user-friendly
tool, where a laboratory can utilize results from
quality control samples and validation data for
uncertainty estimation. MUkit software is
available for download free of charge at Envical
SYKE website. An example of a MUkit report
is given in Appendix 9.
http://www.syke.fi/envical/en
On-line course
University of Tartu gives an on-line course,
Estimation of measurement uncertainty in
chemical analysis. It is an introductory course
on estimation of measurement uncertainty in
chemical analysis. The course gives the main
concepts and mathematical apparatus of
measurement uncertainty estimation as well as
numerous practical examples. The course
contains lectures, practical exercises and
numerous tests for self-testing.
https://sisu.ut.ee/measurement/uncertainty

of 51
3 Calculation of expanded uncertainty, U - overview
A common way of presenting the different contributions to the total measurement
uncertainty is to use a so-called fish-bone (or cause-and-effect) diagram. We
propose a model (Figure 3), where either the within-laboratory reproducibility is
combined with estimates of the method and laboratory bias, (error model in
Appendix 3) or the reproducibility sR is used more or less directly according to ISO
21748 [4]. The alternative way is to construct a detailed fish-bone diagram and
calculate/estimate the individual uncertainty contributions. This modelling
approach may prove very useful when studying or quantifying individual
uncertainty components. However, it has been shown, that in many cases this
approach underestimates the measurement uncertainty [11], partly because it is
hard to include all possible uncertainty contributions in the modelling approach.
By using experimentally determined quality control (QC) and method validation
data, there is an increased possibility that all uncertainty components are included
in the calculations.
Measurement uncertainty model – fish-bone diagram
Covering the analytical process from sample arrival to report
Figure 3. Measurement uncertainty model (fish-bone diagram), where the within-
laboratory reproducibility standard deviation is combined with estimates of the
method and laboratory bias. Alternatively, according to ISO 21748 [4], the
combined standard uncertainty, uc, can be directly estimated from the
reproducibility between laboratories, sR.
Decision maker
Analytical
Report
QC - Within-laboratory
reproducibility, u(Rw)
(section 5)
Method & lab bias, u(bias)
 Reference material
 Proficiency testing
 Recovery
(section 6)
Value
Customer
Reproducibility
between laboratories, sR
(section 7)

of 51
3.1 Customer needs
Customers or data users are in many cases not used to specifying requirements on
data quality, so often the requirements have to be set in dialogue. Guidance on how
to set the needed uncertainty, the target uncertainty, can be found in the Eurachem
Guide Setting and Using Target Uncertainty in Chemical Measurement [21].
In cases where no requirements have been established, a guiding principle could be
that a tentative expanded uncertainty, U, approximately equal to two times the
reproducibility, sR. The sR can often be obtained from proficiency testing or from
the standard method.
3.2 Flow scheme for uncertainty calculations
The flow scheme presented in this section forms the basis for the method outlined
in this handbook. The flow scheme involves 6 defined steps. The following
example with determination of NH4-N in various types of waters (such as ground,
drinking, surface and waste waters) using the automatic photometric method [12]
shows the way forward for calculating the measurement uncertainty using this flow
scheme. Explanations of the steps and their components will follow in the
succeeding chapters. For each step, there may be one or several options for finding
the desired information.
Background for the NH4-N example – automatic photometric method:
Standard uncertainty for within laboratory reproducibility, u(Rw) - For the internal
quality control a synthetic control sample at a level of 200 µg L-1
is used. Target
control limits are used [9]. These limits are wider than if the limits were based on
the actual sRw obtained in the laboratory. NOTE - The u(Rw) is based on the control
limits not on the actual sRw.
Standard uncertainty for method and laboratory bias, u(bias) - The laboratory has
participated in 6 proficiency testing schemes recently in the concentration range 70
to 270 µg L-1
. All results have been somewhat higher than the assigned value. On
average, the bias has been +2.2 %. This bias is considered small by the laboratory
and is not corrected for in their analytical results, but treated as an uncertainty
component. The raw data is given in Appendix 4.
Absolute or relative uncertainty – In the scope of the method EN ISO 11732 [12] is
stated: the method is suitable for ammonium nitrogen in various types of waters in
mass concentration ranging from 0.1 to 10 mg L-1
in undiluted samples. Relative
uncertainty is here the choice since the starting point of the range is well above the
estimated LOQ of 3 µg L-1
– see further section 4 where the issue of absolute or
relative uncertainty is treated.
For this NH4-N method, the main sources of uncertainty are contamination and
variation in sample handling. These uncertainty sources will be included in the
calculations below since we are using experimental data. An output from the
calculations below using the MUkit software [13] is given in Appendix 9.

of 51
Step Action Ammonium NH4-N by EN ISO 11732
1 Specify measurand,
range, and target U.
Decide rel/abs
calculations.
Concentration of NH4-N > 100 µg L-1
.
Target uncertainty ± 15 %.
Relative uncertainty is calculated.
2 Quantify u(Rw )comp.
A control sample
B possible steps not
covered by the
control sample
A: Control sample 200 µg L-1
. Control limits
(2s) are set to ± 6.68 µg L-1
or 3.34 % relative.
B: The control sample includes all analytical
steps.
3 Quantify u(bias)
components
Proficiency testing results show a % bias2
of
+2.5; +2.7; +1.9; +1.4; +1.8 and +2.9. The root
mean square (RMS) of the bias is 2.26 %. The
uncertainty of the assigned values, u(Cref), is
1.52 %. (see Appendix 4 for explanations)
4 Convert components to
standard uncertainty
u(x)
Conversion to standard uncertainty [1, 7, 16].
u(Rw) = 3.34/2 = 1.67 %
2
2
)
(
)
( Cref
u
RMS
bias
u bias 

%
73
.
2
52
.
1
26
.
2 2
2



5 Calculate combined
standard uncertainty, uc
Standard uncertainties can be combined by
taking the square root of the sum of the squares
 
%
20
.
3
73
.
2
67
.
1
)
(
)
(
2
2
2
2





 bias
u
R
u
u w
c
6 Calculate expanded
uncertainty,
c
u
U 
 2
The reason for calculating the expanded
uncertainty is to reach a high enough confidence
(app. 95 %) in that the “true value” lies within
the interval given by the measurement result 
the uncertainty. 7
40
.
6
20
.
3
2 



U %.
The measurement uncertainty for NH4-N in water will thus be reported as ± 7 %
relative at concentration levels of ammonium > 100 µg L-1
.
2
In case of proficiency testing this is not really a bias since the difference is often based on
a single analytical result.

of 51
3.3 Summary table for uncertainty calculations
The results of the calculations done in the flow scheme will then be summarised in
a summary table.
Concentration of ammonium nitrogen in water by EN ISO 11732
Measurement uncertainty U (95 % confidence interval) is estimated to ± 7 % for
concentration of NH4-N > 100 µg L-1
. The target uncertainty is ± 15 %. The
calculations are based on control chart limits and proficiency testing data.
Value Relative
u(x)
Comments
Within-laboratory reproducibility, u(Rw)
Control sample
X = 200 µg L-1
u(Rw) Control limits is
set to ± 3.34 %
1.67 %
Other components --
Method and laboratory, u(bias)
Reference material u(bias) --
Proficiency test u(bias) RMSbias = 2.26 %
u(Cref) = 1.52 %
2.73 %
2
2
)
(
)
(
Cref
u
RMS
bias
u
bias 

Recovery test u(bias) --
Reproducibility between laboratories, sR
Proficiency tests
70 to 270 µg L-1
.
sR
-- 8.8 % Data – see section
7.2
Interlaboratory
comparison in
standard method
sR 4-10 % Concentration ranges
from 250 to 800 µg L-1
.
Combined standard uncertainty, uc is calculated from the control limits and bias
estimation from proficiency tests. The sR from proficiency testing in a standard
method can also be used (see section 7.2).
Measurand Combined Standard Uncertainty uc Expanded Uncertainty U
Concentration
NH4-N
20
.
3
73
.
2
67
.
1 2
2

 % 2 ·3.20 = 6.4  7 %

of 51
4 Uncertainty over the measurement range
A measurement uncertainty can be given absolute (with the same unit as the unit of
the measured value) or relative (in %) as in the example below.
Measured value Measurement uncertainty, U (95 %)
Absolute Relative
20 µg L-1
2 µg L-1
10 %
4.1 Relationship between measurement uncertainty and
concentration
The relationship between absolute measurement uncertainty and concentration for
many instrumental analytical techniques is shown in Figure 4 below [7, 13, 14].
0 20 40 60 80 100
Absolute
measurement
uncertainty
Concentration
Different types of
contributions to
measurement uncertainty
Total measurement
uncertainty
Figure 4. Relationship between absolute measurement uncertainty and
concentration for many instrumental analytical techniques.
The total measurement uncertainty (solid line) consists of contributions that are
typically proportional to the concentration or independent of the concentration
(dashed lines). The relationship between absolute measurement uncertainty and
concentration can be approximated with a somewhat simpler relationship shown in
Figure 5(a). The corresponding relationship between relative measurement
uncertainty and concentration is shown in Figure 5(b).
From Figure 5 it can be seen that it is appropriate to divide the measurement range
at the dashed line. In the low range it is appropriate to use an absolute measurement
uncertainty, while in the high range it is appropriate to use a relative measurement
uncertainty. For methods applied only in the high measurement range a relative
uncertainty is most appropriate. For some methods, e.g. titrations and physical
methods, it may be appropriate to use an absolute measurement uncertainty in the
whole range, but the choice depends on whether the major errors are absolute or
relative. For pH we strongly recommend an absolute uncertainty.

of 51
Concentration
Absolute
measurement
uncertainty
(a)
(b)
(c)
Measurement range
of the method
Relative measurement
uncertainty
Absolute measurement
uncertainty
Concentration
Relative
measurement
uncertainty
Figure 5. Relationship between (a) absolute measurement uncertainty and
concentration, and (b) relative measurement uncertainty and concentration.
Division of the measurement range (c) at the dashed line into a low range where
the absolute measurement uncertainty is constant and a high range where the
relative measurement uncertainty is approximately constant.
4.2 Using replicate results to divide the measurement range
Results from determination of replicates of samples in the whole measurement
range can be used to divide the measurement range into a low concentration range
where absolute uncertainty is constant and into a high range where relative
uncertainty is constant. This is demonstrated below for 73 samples that have been
analysed for ammonium nitrogen NH4-N as two replicates x1 and x2. For each
sample the calculations shown in the table below were performed.
Table 1 Calculation of relative standard deviation from duplicates. The full data
set is given in Appendix 5.
Sample x1
(µg L-1
)
x2
(µg L-1
)
Mean, x
(µg L-1
)
Relative s
(%)
1 7.46 7.25 7.35 2.019
2 9.01 9.17 9.09 1.245
3 3.60 3.10 3.35 10.554
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
73 31.90 32.36 32.13 1.012
Then the graph in Figure 6 is constructed.

of 51
Figure 6. Plot of relative s vs. x . The dashed line marks a border between two
concentration ranges were it is appropriate to use an absolute and a relative
uncertainty, respectively. For clarity, the concentration scale above 100 µg L-1
is
logarithmic.
Figure 6 indicates that the division of the measurement range is at approximately
20 µg L-1
, as indicated by a dashed line. Above this concentration the relative
standard deviation for the two replicates is independent of the concentration
suggesting that it is appropriate to use a relative uncertainty. Furthermore, at
concentrations below approximately 20 µg L-1
it is appropriate to use an absolute
uncertainty. After the uncertainty in the ranges has been calculated the division
between the two ranges may be adjusted, see the example in section 4.3.
Sometimes the interpretation is not as straightforward as in Figure 6, however,
often useful information about the method is still obtained from such a plot.
4.3 Calculation of uncertainty over the measurement range
After deciding if absolute or relative measurement uncertainty should be used in a
certain measurement range, it is important that 1) all uncertainties used in the
calculations in that range are either absolute or relative, respectively, 2) only data
from that particular measurement range should be used, 3) ideally the data used
should cover the major part of the measurement range and 4) the division of the
measurement range may be adjusted to fit the results.
For NH4-N we obtained an expanded uncertainty for the high concentration range
of 7 % (see section 3.3) and at the low range an expanded uncertainty of 2 µg L-1
was obtained. The level where 2 is about 7 % (2 /0.07) is 28.6 ≈ 30 µg L-1
and the
division of the measurement range is thus adjusted from 20 to 30 µg L-1
. The
measurement uncertainty over the measurement range for NH4-N is given below.
Range Measurement Uncertainty, U
3-30 µg L-1
2 µg L-1
30-1000 µg L-1
7 %
NOTE 1: The absolute uncertainty is equal to the relative uncertainty at 30 µg L-1
where the
measuring range is divided.
NOTE 2: Expanded measurement uncertainty for concentration of NH4-N in water was achieved
by an expert laboratory, which has the method including contamination in full control. Typically
routine laboratories can achieve measurement uncertainties 10-20 % at high concentration level.

of 51
5 Within-laboratory reproducibility - u(Rw)
In this section two ways of estimating the within-laboratory reproducibility
component, u(Rw), for the measurement uncertainty calculation are explained:
 Control samples covering the whole analytical process – step 3 and 4 of
the ladder of errors. Normally one sample at low concentration level and
one at a high concentration level.
Here u(Rw) = Rw
s
NOTE – when wider control limits are used (e.g. target control limits) the
u(Rw) is based on the starget used to set the limit in the control chart NOT
on the actual sRw obtained for the control sample.
Here u(Rw) = target
s
 Control samples and routine sample replicates. From control samples not
covering the whole analytical process, step 3 of the ladder of errors and
from duplicate analyses of test samples with varying concentration levels –
step 4 of the ladder of errors.
Here u(Rw) =
2
2
Rw r
s
s 
It is of utmost importance that the estimation covers all steps in the analytical chain
and all types of matrices – worst-case scenario. The control sample should be run
in exactly the same way as the test samples e.g. if the mean of duplicate samples is
used for test samples, then the mean of duplicate control samples should be used
for the calculations.
It is likewise important to cover long-term variations of some uncertainty
components that are systematic in the short-term over time within the laboratory,
e.g. caused by different stock solutions, new batches of critical reagents,
recalibrations of equipment, etc. In order to have a representative basis for the
uncertainty estimation and to reflect any such variation the number of results
should ideally be more than 60 and cover a time period of at least one year [9].
5.1 Customer demands
Some laboratories choose to use the customer demand when setting the limits in
their control charts – target control limits. The actual performance of the method is
not interesting, as long as it meets the customer demands on expanded uncertainty.
If, for example, the customer asks for data with an (expanded) measurement
uncertainty of ± 10 %, then, from our experience, a good starting point is to set the
warning control limits (± 2s) to half that value i.e. ± 5 % [9]. The u(Rw) used in the
calculations will then be 2.5 %3
, providing that the actual sRw is lower. This is just
a proposal and the measurement uncertainty calculations will show if these control
limits are appropriate.
3
Treating the control limits according to GUM [1] as type B estimate with 95 % confidence
limit.

of 51
5.2 Control sample covering the whole analytical process
When a stable control sample is available which covers the whole analytical
process and has a matrix similar to the samples, the within-laboratory
reproducibility at that concentration level can simply be estimated from the
measurements of the control samples. If the measurements performed cover a wide
range of concentration levels, several control samples at different concentration
levels should be used. Example: For NH4-N two control sample concentration
levels (20 and 250 g L-1
) were used. The results for the manual measurement
method are presented in the table below. In this case the u(Rw) is equal to the sRw.
Absolute Relative Comments
Control sample 1
X = 20.01 µg L-1-
1
sRw 0.5 g L-1
2.5 % Measurements in
2002, n = 75
Control sample 2
X = 250.3 µg L-1
sRw 3.8 g L-1
1.5 % Measurements in
2002, n = 50
Other components --
5.3 Control samples and routine sample replicates
A synthetic control solution used for quality control would normally not cover the
whole analytical process and the matrix type is, in most cases, not similar to the
routine samples. Example: To estimate the repeatability in different matrices,
duplicate analysis of ammonium in test samples water is performed, and the sr is
estimated. This gives the repeatability for test samples having a normal matrix
variation in salt and particles at different concentration levels.
Example – different matrices
The data set consists of 73 duplicate analyses in the range of 2 g L-1
– 16000
g L-1
. Most of the results were below 200 g L-1
. The data, given in Appendix 5,
is divided into a lower range, < 30 g L-1
and a higher range > 30 g L-1
.
Absolute u(x) Relative u(x) Comments
Repeatability
Duplicate analyses
2 - 30 g L-1
30 - 16 000 g L-1
sr
sr
0.44 g L-1
3.8 %
n = 47
n = 26
As the estimate from duplicate analyses gives the repeatability component (sr)
only, it should be combined with the synthetic control sample results from Section
5.2 to give a better estimate of u (Rw). It can be noticed that the sample matrix has
some effect on the variation of the results.

of 51
Value u(x) Comments
Within-laboratory reproducibility , u(Rw)
Low level
(2-30 µg L-1
)
u(Rw) 0.5 µg L-1
from control
sample and 0.44 µg L-1
from duplicates
0.7 µg L-
1
Absolute u(Rw) =
2
2
44
.
0
5
.
0 
High level
(> 30 µg L-1
)
u(Rw) 1.5 % from control
sample and 3.8 % from
duplicates
4.1 % Relative u(Rw) =
2
2
8
.
3
5
.
1 
Example – unstable control samples
In this example, duplicate samples have been measured with an oxygen probe on
50 occasions over 2 years. The raw data is given in Appendix 6. The concentration
variation is limited, so an R-chart approach is chosen. Since it is important to look
for systematic differences between the first and the second results the difference
between the first and the second measurement is calculated and plotted in an
R-chart, see Figure 7. The standard deviation for the results can be estimated from
the pooled standard deviation of the duplicate samples (see Appendix 6), and in
this case becomes 0.025 mg L-1
. The control limits 2s are at 07
.
0
83
.
2
025
.
0 

mg L-1
[9]. Mean value is 7.48 and relative u(x) for repeatability is 0.34 %.
Figure 7. R-chart - Determination of dissolved oxygen in sea water - the difference
between the first and the second measurement

of 51
However, this only gives the repeatability, sr for sampling and measurement, but
there will also be a “long-term” uncertainty component from the variation in the
calibration (here the Winkler titration is used for calibration of the oxygen probe).
For this particular measurement, the uncertainty component from the long-term
variation in calibration is hard to measure, as no stable reference material is
available for dissolved oxygen. One method would be to calculate the standard
deviation of differences obtained on different days between the value obtained with
the probe and the value obtained with the Winkler method. Here we choose to
estimate that component by a qualified guess, but laboratories are encouraged to
also try the experimental approach.
The total within-laboratory reproducibility for dissolved oxygen thus becomes:
Value Relative u(x) Comments
Within-laboratory reproducibility, u(RW)
Duplicate analyses of
natural samples,
difference used in R-
chart
sr s = 0.0252 mg L-1
X = 7.50 mg L-1
0.34 % Measurements
in 2000-2002,
n= 51
Estimated variation
from differences in
calibration over time
s = 0.5 % 0.5 % Estimate, based
on experience
Combined relative standard uncertainty uRw
Repeatability + within-
lab reproducibility in
calibration
60
.
0
%
5
.
0
34
.
0 2
2

 %

of 51
6 Method and Laboratory bias – u(bias)
In this chapter three ways of estimating the uncertainty as a result of the bias
component will be outlined, namely; 1) the use CRM, 2) participation in PT and
3) performing recovery tests.
The ISO Guide to the expression of uncertainty in measurement, GUM [1],
assumes that “the result of a measurement has been corrected for all recognised
significant systematic effects” (GUM 3.2.4). This implies that when developing a
measurement method all known sources of bias within the method’s scope should
be investigated and if possible, eliminated. However, in many cases a developed
method may still have a bias and the bias can vary depending on changes in matrix
and concentration. Correcting for any observed bias on one reference only cannot
be generally recommended [15]. The issue of bias correction is also treated in the
leaflet Treatment of an observed bias from Eurachem, www.eurachem.org.
An observed bias can be treated as an uncertainty component as stated in VIM
[10]4
. Bias can in many cases be both positive and negative. Even if the measured
bias is positive in certain matrices and negative in others, all bias values in a
selected concentration range should be used to estimate the uncertainty component,
RMSbias. Even if the bias is not significant or zero one should treat it as an
uncertainty component due to a possible bias – the bias may be small or absent, but
has to be taken into account.
Two bias components have to be estimated:
1) the root mean square (RMS) of the individual bias values [15]
2) the mean of the standard uncertainty of the assigned/certified values,
u(Cref) or u(Crecovery)5
The uncertainty due to bias, u(bias) can then be estimated by
2
2
)
(
)
( Cref
u
RMS
bias
u bias 
 where
CRM
i
bias
n
bias
RMS 

2
)
(
where nCRM is the number CRMs used (or PT or recovery tests).
If only one CRM is used also the sbias (standard deviation of the measured values of
the CRM) has to be included and u(bias) can then be estimated [16, 17] by
  2
2
2
)
(
)
( Cref
u
n
s
bias
bias
u bias









4
Measurement Uncertainty NOTE 1…Sometimes estimated systematic effects are not corrected
for but, instead, associated measurement uncertainty components are incorporated
5
A mean value is used for simplicity – a pooled value would be more correct.

of 51
6.1 Certified Reference Material
Regular measurement of one or several CRM can be used to estimate the bias.
Each reference material should be measured in at least 5 different analytical series
(e.g. on 5 different days) before the values are used.
One CRM - In this example with one CRM the certified value is 11.5 ± 0.5, with a
95 % confidence interval. The analytical results are on average 11.9 with a
standard deviation, sbias, of 2.2 %, n=12.
Uncertainty component from the uncertainty of the certified value
Step Example
Convert the confidence
interval to u(Cref)
The confidence interval is ± 0.5. Divide this by 2 to
convert it to standard uncertainty:
0.5/2 = 0.25
Convert to relative
uncertainty u(Cref)
0.25/11.5 · 100 = 2.16 %
3 Quantify u(bias)
components
%
3.48
11.5)/11.5
-
(11.9
100
bias 


sbias = 2.2 % (n = 12)
u(Cref) = 2.16 %
4 Convert components
to standard
uncertainty u(x)
  








 2
2
2
)
(
)
( Cref
u
n
s
bias
bias
u bias
  %
1
.
4
16
.
2
12
2
.
2
48
.
3 2
2
2









Several CRM - If several CRMs are used, we will get different values for bias.
The uncertainty due to any bias, )
(bias
u will be calculated in the following way.
3 Quantify u(bias)
components
bias CRM1 is 3.48 %, s=2.2 % (n=12), u(Cref1)=2.16 %
bias CRM2 is -0.9 % s=2.0 % (n=7), u(Cref2)=1.8 %
bias CRM3 is 2.5 %, s= 2.8 % (n=10), u(Cref3)=1.8 %
RMSbias = 2.53 %
mean u(Cref)=1,92 %
to standard
uncertainty u(x)
2
2
)
(
)
( Cref
u
RMS
bias
u bias 

%
2
.
3
92
.
1
53
.
2 2
2



of 51
6.2 Proficiency tests
In this case the results from proficiency tests (PT) are used in the same way as
reference materials to estimate )
(bias
u . In order to have a reasonably clear picture
of the bias from proficiency testing results, a laboratory should participate at least 6
times within a reasonable time interval.
The way forward is very similar to that for reference materials. However, for
reference materials a mean value over time is used and for each PT a single
laboratory result is used. Therefore the estimated RMSbias from proficiency test will
usually be higher. Also the certified value of a CRM normally has a lower
uncertainty than an assigned value in a PT. In some cases the calculated uncertainty
u(Cref) from a proficiency testing becomes too high and is not valid for estimating
the u(bias).
Uncertainty component from the uncertainty of the assigned value
Step Example
Find the between laboratory
standard deviations, sR, and
the number of labs, nLab, for
each PT.
In the first PT the sR is 8.7 % and nLab is 23.
NOTE: If the PT provider reports so called ‘robust
standard deviation’, this value must be multiplied by
a factor of 1.25 to correspond to sR described in this
handbook [18].
Calculate u(Crefi) for each
PT %
8
.
1
23
%
7
.
8
)
(
,



i
Lab
Ri
i
n
s
Cref
u
The other five u(Crefi) values are e.g.
2.9 %, 1.7 %, 4.1 %, 3.0 % and 2.1 %
Calculate u(Cref) as the
average of the individual
u(Crefi) values.
Number of PT: N = 6.




N
Cref
u
Cref
u
N
i
i
1
)
(
)
( 2.7 %

of 51
The bias has been +2 %, +7 %, -2 %, +3 %, +6 % and +5 %, in the 6 PT where the
laboratory has participated.
3 Quantify u(bias)
components
RMSbias = 4.6 %,
u(Cref)= 2.6 %
to standard
uncertainty u (x)
%
3
.
5
6
.
2
6
.
4
)
(
)
(
2
2
2
2





 Cref
u
RMS
bias
u bias
If the PT provider estimates the uncertainty, U, of the assigned value e.g. according
to the procedure described in ISO 13528 [18], then U/2 should be used as u(Crefi)
for each PT instead of calculating uncertainty via sR and nLab.
NOTE: The drawback with PT is that the laboratory result is based on one
measurement which results in an increased uncertainty compared to a mean value.
If it is possible to measure several times under a longer time interval on PT
samples we recommend using the mean values of these measurements.

of 51
6.3 Recovery
Recovery tests, for example the recovery of a standard addition to a sample, can be
used to estimate one component6
of the bias [15]. The guidance given in this
section is applicable for test methods that do not include a recovery correction in
the procedure.
In an experiment the recoveries for an added spike were 95 %, 98 %, 97 %, 96 %,
99 % and 96 % for 6 different sample matrices. The average is 96.8 %. The spike
of 0.5 mL was added with a micropipette.
Uncertainty component from recovery, u(Crecovery)
Step Example
Main components are the
uncertainty of the
concentration, u(conc) of
standard and volume added
u(vol)
u(conc) - Certificate ± 1.2 % (95 % conf. limit) gives =
0.6 %
u(vol) - This value can normally be found in the
manufacturer’s specifications, or better still use the
limits specified in your laboratory. Max bias 1 %
(rectangular interval) and repeatability max 0.5 %
76
.
0
5
.
0
3
1
)
( 2
2











vol
u %
Calculate u(Crecovery)
0
.
1
76
.
0
6
.
0
)
(
)
( 2
2
2
2



 vol
u
conc
u %
3 Quantify u(bias)
components
RMSbias = 3.44 %
u(Crecovery)=1.0 %
to standard
uncertainty u(x) %
6
.
3
0
.
1
44
.
3
)
(
)
(
2
2
2
2





 ery
cov
Cre
u
RMS
bias
u bias
6
Some bias components are not included in a recovery e.g. bias due to low selectivity,
contamination.

of 51
7 Reproducibility between laboratories, sR
If the demand on uncertainty is low, it may be possible to directly use the sR as an
approximation of uc [3, 4]. The expanded uncertainty .
2 R
s
U 
 This may be an
overestimate depending on the quality of the laboratory. It may also be an
underestimate due to sample inhomogeneity or matrix variations.
7.1 Data given in a standard method
In order to use a figure taken directly from the standard method, the laboratory
must prove that they are able to perform in accordance with the standard method
[4], i.e. demonstrating control of bias and verification of the repeatability, sr.
Reproducibility data in the standard method can either be given as a standard
deviation sR or as reproducibility limit R and then sR = R/2.8.
The example below is taken from ISO 15586 Water Quality — Determination of
trace elements by atomic absorption spectrometry with graphite furnace. The
matrix is wastewater. Combined uncertainty in wastewater, uc, is taken from the sR
from interlaboratory comparisons performed according to ISO 5725 [19] quoted in
the ISO method7
.
Table 2. ISO 15586 - Results from the proficiency testing – Cd in water with
graphite furnace AAS. The wastewater was digested by the participants.
Cd nLab Outliers Assigned value Mean Recovery sr sR
µg L-1
µg L-1
% % %
Synthetic Lower 33 1 0.3 0.303 101 3.5 17.0
Synthetic Higher 34 2 2.7 2.81 104 1.9 10.7
Fresh
water
Lower 31 2 0.572 2.9 14.9
Fresh
water
Higher 31 3 3.07 2.1 10.4
Waste
water
27 2 1.00 3.1 27.5
Measurand Matrix Combined Uncertainty uc Expanded Uncertainty U
Cd Waste
water
uc = 27.5 % 2·uc = 55 %  60 %
7
In ISO 15586 the relative s is given as coefficient of variation, CV.

of 51
7.2 Data from proficiency testing
Proficiency tests (PT) are valuable tools in uncertainty evaluation. The
reproducibility between laboratories is normally given directly in reports from the
exercises as sR.
Data may well be used by a laboratory having performed satisfactorily in the PT as
provided that this PT covers all relevant uncertainty components and steps - see
17025 section 5.4.6.3 [6].
Table 3. Summary results (mean or pooled values) from 10 PT that Lab A has
participated in. The reproducibility standard deviation is given in absolute units
for pH, sR and in relative units sR % for the other parameters.
Parameter Assigned
value
Lab A %
deviation
sR
(abs)
sR
%
No. of
labs
Excluded
pH 7.64 -0.037 0.101 - 90 5
Conductivity, mS/m 12.5 -2.8 - 3.2 86 6
Alkalinity, mmol L-1
0.673 +2.3 - 3.9 60 3
Turbidity, FNU 1.4 -9.1 - 14.2 44 3
NH4-N, µg L-1
146 +2.2 - 8.8 34 5
NO3-N, µg L-1
432 -1.6 - 3.7 39 6
In Table 3 we find that for conductivity, the mean assigned value for the results
from 10 PT is 12.5 mS/m. The reproducibility standard deviation is 3.2 %, which is
a pooled standard deviation between the laboratories in the different PT and this
value can be taken as an estimate of combined standard uncertainty i.e.
uc (conductivity) = sR = 3.2 % and U = 2·3.2 = 6.4 ≈ 7 % at 12.5 mS/m.
If we take the ammonium results, we have a mean value of 146 µg L-1
, and we find
that the reproducibility, sR, is 8.8 %. Thus U = 2·8.8 = 17.6 ≈ 18 % at this
concentration level.
Comment: In Section 3 the expanded uncertainty, U, for ammonium is 7 % using
an automated method in one highly qualified laboratory.
mS/m

of 51
8 Examples
In this chapter, practical examples are presented on how measurement uncertainty
can be calculated using the approach in this handbook.
8.1 Ammonium in water
Ammonium in water has already been treated in section 3.2, 4 and section 7.2 . The
results in the upper measurement range are summarised in Table 4.
Table 4. Measurement uncertainty of ammonium in water – comparison of different
calculations
Uncertainty calculations
based on
Relative expanded
uncertainty, U
Comment
Control sample + PT ± 7 % Uncertainty for one good
laboratory- range > 30 µg L-1
.
PT ± 18 % Uncertainty in general among
laboratories – level 150 µg L-1
8.2 BOD in wastewater
Biological Oxygen Demand, BOD, is a standard parameter in the monitoring of
wastewater. This example shows how data from internal quality control can be
used together with CRM results or data from PT to calculate the measurement
uncertainty. The results for expanded uncertainty are summarised in Table 5
Table 5. Measurement uncertainty of BOD in water - comparison of different
calculations
Uncertainty calculations
based on
Relative expanded
uncertainty, U
Comment
Control sample + CRM ± 11 %
Control sample + PT ± 10 % n = 3, unreliable
estimate
PT ± 16 % Uncertainty in general
among laboratories
For BOD at high concentrations, using the dilution analytical method, the major
error sources are the actual oxygen measurement and variation in the quality of the
seeding solution. These errors will be included in the calculations.
Raw data from the internal quality control, using a CRM, is shown in Appendix 7.

of 51
The laboratory has only participated in three proficiency testing exercises the last 2
years (Table 6). At least six would be needed, so here we estimate the bias in two
different ways – with a CRM and with PT.
Table 6. BOD - results from PT
Exercise Assigned
value
Laboratory
result
Bias sR Number
of labs
u(Crefi)
mg L-1
mg L-1
% % %
1 154 161 + 4.5 7.2 23 1.50
2 219 210 - 4.1 6.6 25 1.32
3 176 180 +2.3 9.8 19 2.25
X +0.9 7.878
u(Cref) 1.69
RMSbias 3.76 - - -
8
If sR or the number of participants vary substantially from exercise to exercise, then a
pooled standard deviation will be more correct to use. In this case, where the variation in sR
is limited, we simply calculate the mean sR (the corresponding pooled standard deviation
becomes 7.82, an insignificant difference).

of 51
Example A: BOD with internal quality control + one CRM
Step Action BOD in wastewater with EN 1899-1
(method with dilution, seeding and ATU)
Decide rel/abs
calculations.
Concentration of BOD < 2 µg L-1
in
wastewater.
The target uncertainty is ± 20 %.
A control sample
covered by the control
sample
A: The control sample, a CRM gives an s of
2.6 % at an O2 level of 206 mg L-1
. This s is
when setting the control chart limits.
B: The measurement of the control sample
includes all analytical steps.
3 Quantify u(bias)
components.
The CRM is certified to 206  5 mg L-1
O2. The
average result is 214.8 mg L-1
. Thus, there is a
bias of +8.8 mg L-1
or 4.3 %.
The sbias is 2.6 % (n=19)
The u(Cref) is 5 mg L-1
/ 2 = 2.5 mg L-1
and relative %
2
.
1
100
206
/
5
.
2 

standard
uncertainty u(x)
u (Rw) = 2.6 %
%
5
.
4
2
.
1
19
6
.
2
3
.
4
)
(
)
(
2
2
2
2
2
2












 Cref
u
n
s
bias
bias
u bias
uc = 2
2
5
.
4
6
.
2  = 5.2 %
uncertainty,
c
u
U 
 2
%
11
4
.
10
2
.
5
2 



U

of 51
Example B: BOD with internal quality control + PT
Step Action BOD in wastewater with EN 1899-1
(method with dilution, seeding and ATU)
Decide rel/abs
calculations.
Concentration of BOD < 2 µg L-1
in
wastewater. The target uncertainty is ± 20 %.
A control sample
covered by the
control sample
A: The control sample, a CRM, gives an s of
2.6 % at an O2 level of 206 mg L-1
. The s of
2.6 % is also used when setting the limits in the
control chart.
includes all analytical steps after sampling
3 Quantify u(bias)
components.
Data from Table 6
RMSbias = 3.76 %
69
.
1
)
( 
Cref
u %
to standard
uncertainty u(x)
u(Rw) = 2.6 %
%
12
.
4
69
.
1
76
.
3
)
(
)
(
2
2
2
2




 Cref
u
RMS
bias
u bias
standard
uncertainty, uc
uc = 87
.
4
12
.
4
6
.
2 2
2

 %
uncertainty,
c
u
U 
 2
%
10
7
.
9
87
.
4
2 



U

of 51
8.3 PCB in sediment
In these examples, the u(Rw) is estimated from a quality control sample and the
u(bias) is estimated from two different sources: in the first example using a CRM
and in the second example by participation in proficiency tests. The sample-work
up is a major error source, and it is thus crucial that this step is included in the
experimental data. The number of proficiency tests is too few to get a reliable
estimate.
Example C: PCB with internal quality control + one CRM
Step Action PCB in sediment by GC-MS
Decide rel/abs
calculations.
Sum of concentration of 7 PCB analytes.
The range is dependent on matrix and analyte.
Demand on expanded uncertainty is ± 20 %.
A control sample
covered by the
control sample
A: The control sample, which is a CRM, gives
a sRw = 8 % at a level of 150 µg kg-1
dry matter.
The sRw of 8 % is also used when setting the
control chart limits.
includes all steps except for drying the sample
to determine the dry weight. The uncertainty
contribution from that step is considered small.
3 Quantify u(bias)
components.
The CRM is certified to 152  14 µg kg-1
. The
average result of the control chart is 144. Thus,
there is a bias = 5.3 %. The sbias = 8 % (n=22)
u(Cref) 14/2, which is 4.6 % relative.
u(x)
u(Rw) = 8 %
22
.
7
6
.
4
22
8
3
.
5
)
(
)
(
2
2
2
2
2
2












 Cref
u
n
s
bias
bias
u bias
standard uncertainty uc
uc = 2
2
22
.
7
8  = 10.8 %
uncertainty,
c
u
U 
 2
%
22
6
.
21
8
.
10
2 



U

of 51
Example D: PCB with internal quality control + proficiency testing
Step Action PCB in sediment by GC-MS
Decide rel/abs
calculations.
Sum of concentration of7 PCB analytes.
The range is dependent on matrix and analyte.
Demand on expanded uncertainty is ± 20 %.
A control sample
sample
A: The control sample, which is a stable in-
house material, gives sRw = 8 % at a level of
150 µg kg-1
dry matter. The sRw of 8 % is also
used as s when setting the control chart limits.
includes all steps except for drying the sample
to determine the dry weight. The uncertainty
contribution from that step is considered small
and is not accounted for.
3 Quantify u(bias)
components.
Participation in 3 PT with concentration levels
similar to the internal quality control. The bias
in the 3 exercises has been –2 %, -12 % and –
5 %. RMSbias = 7.6 %
The u(Crefi) values in the three exercises have
been 2.7 %, 2.5 % and 3.5 %.
On average u(Cref) = 2.9 %
u(x)
The u(Rw) is 8 %
%
1
.
8
9
.
2
6
.
7
)
(
)
(
2
2
2
2




 Cref
u
RMS
bias
u bias
uc = 2
2
1
.
8
8  = 11.4 %
uncertainty,
c
u
U 
 2
%
23
8
.
22
4
.
11
2 



U

of 51
Summary table for PCB measurement uncertainty
calculations
PCB in sediment by extraction and GC-MS
Measurement uncertainty U (95 % confidence interval) is estimated to ± 23 %
(relative) for 7 PCB in sediments at a level of 150 µg kg-1
dry weight. The target
uncertainty is ± 20 %. The calculations are based on internal quality control using a
stable sample, CRM and the participation in a limited number of PT.
Value u(x) Comments
Within-laboratory reproducibility , u(Rw)
Control sample
X = 160 µg kg-1
dry weight
u(Rw) 12.8 µg kg-1
dry
weight
8 %
Other components too small to be considered
Method and laboratory, u(bias)
CRM u(bias) Bias: 5.3 %
sbias = 8 ; n = 22
u(Cref) = 4,7 %
u(bias) = 7.29 u(bias)=
2
2
2
)
(Cref
u
n
s
bias bias


PT
n = 3
u(bias) RMSbias = 7.6
u(Cref) = 2.9 %
u(bias) = 8.1 u(bias)=
2
2
)
(Cref
u
RMSbias 
Combined uncertainty, uc, is calculated from internal quality control and the
maximum u(bias) from PT.
Measurand Combined Uncertainty uc Expanded Uncertainty U
PCB uc = 2
2
1
.
8
8  = 11.4 %
23
8
.
22
4
.
11
2
2 




 c
u
U
Conclusion: In this case the calculation of the u(bias) gives similar results
regardless of whether CRM or proficiency testing results are used. Sometimes
proficiency tests will give considerably higher values. If the CRM is similar to test
samples and at an appropriate concentration level it might in such cases be more
correct to use the CRM results.

of 51
9 Reporting uncertainty
This is an example of how an analytical report could look, when measurement
uncertainty is reported together with the data. The company and accreditation body
logotypes are omitted, and the report does not contain all information normally
required for an accredited laboratory. It is recommended to use either relative or
absolute values for the benefit of the customer. Here is reported absolute
uncertainty.
Analytical Report
Sample identification: P1 – P4
Samples received: 14 December 2002
Analysis period: 14 –16 December 2002
Results
NH4-N (µg L-1
):
Sample Result U Method
P1 103  7 23B
P2 122  9 23B
P3 12  2 23B
P4 14  2 23B
TOC (mg L-1
)
Sample Result U Method
P1 40  4 12-3
P2 35  4 12-3
P3 10  1 12-3
P4 9  1 12-3
Signed: Dr Analyst

of 51
The laboratory should also prepare a note explaining how the measurement
uncertainty has been calculated for the different parameters. Normally, such an
explanatory note should be communicated to regular customers and other
customers who ask for information. An example is given below:
Note on measurement uncertainty from Dr Analyst’s laboratory
Measurement uncertainty:
U = expanded measurement uncertainty, estimated from control sample
results, proficiency testing and the analyses of CRMs, using a coverage factor
of 2 to reach approximately 95 % confidence level.
NH4-N: U is estimated to 7 % above 30 µg L-1
and 2 µg L-1
below 30 µg L-1
.
TOC: U is estimated to 10 % over the whole concentration range.
References:
 B. Magnusson, T. Näykki, H. Hovind, M. Krysell, E. Sahlin,
Handbook for calculation of measurement uncertainty in
environmental laboratories, Nordtest Report TR 537 (ed. 4) 2017.
Available from www.nordtest.info
 ISO 11352:2012 Water quality — Estimation of measurement
uncertainty based on validation and quality control data.

of 51
10 References
1 Guide to the Expression of Uncertainty in Measurement. ISO, Geneva (1993).
(ISBN 92-67-10188-9) Reissued as ISO Guide 98-3 (2008), also available from
www.bipm.org as JCGM 100:2008.
2 EA-4/16: EA guideline on the expression of uncertainty in quantitative testing,
(2003) www.european-accreditation.org
3 EUROLAB Technical Report 1/2007, Measurement uncertainty revisited:
Alternative approaches to uncertainty evaluation. EUROLAB (2007). Available
from www.eurolab.org.
4 ISO 21748:2017, Guide to the use of repeatability, reproducibility and trueness
estimates in measurement uncertainty estimation. ISO, Geneva.
5 ISO 11352:2012, Water quality -- Estimation of measurement uncertainty based
on validation and quality control data, ISO, Geneva.
6 EN ISO/IEC 17025:2005, General Requirements for the Competence of
Calibration and Testing Laboratories.
7 S. L. R. Ellison and A. Williams (Eds). Eurachem/CITAC guide: Quantifying
Uncertainty in Analytical Measurement, (3rd
edition, 2012) ISBN 978-0-948926-
30-3. Available from www.eurachem.org.
8 V J Barwick and E Prichard (Eds), Eurachem Guide: Terminology in Analytical
Measurement – Introduction to VIM 3 (2011). ISBN 978-0-948926-29-7.
Available from www.eurachem.org.
9 H. Hovind, B. Magnusson, M. Krysell, U. Lund, and I. Mäkinen, Internal Quality
Control - Handbook for Chemical Laboratories, Nordtest technical report 569
Edition 4 (2011). www.nordtest.info.
10 ISO/IEC Guide 99:2007, International vocabulary of metrology - Basic and
general concepts and associated terms (VIM). ISO, Geneva, (2007). Available from
http://www.bipm.org as JGCM 200:2008.
11 M. Thompson and S.L.R. Ellison, Dark uncertainty, Accred Qual Assur (2011)
16: 483.
12 ISO 11732:2005, Water quality -- Determination of ammonium nitrogen by
flow analysis (CFA and FIA) and spectrometric detection.
13 T. Näykki, A. Virtanen and I. Leito, Software support for the Nordtest method
of measurement uncertainty evaluation, Accred Qual Assur, (2012) 6: 603–612.
14 B. Magnusson and M. Koch, Use of characteristic functions derived from
proficiency testing data to evaluate measurement uncertainties, Accred Qual Assur
(2013) 17:399–403.
15 B. Magnusson, S. L. R. Ellison, Treatment of uncorrected measurement bias in
uncertainty estimation for chemical measurements, Anal Bioanal Chem, (2008)
390:201-213.

of 51
16 V. J. Barwick and S. L: R. Ellison, Measurement uncertainty: Approaches to the
evaluation of uncertainties associated with recovery, Analyst, (1999) 124, 981-990.
17 E. Hund, D.L. Massart and J. Smeyers-Verbeke, Operational definitions of
uncertainty. TrAC, (2001) 20 (8).
18 ISO 13528:2015 Statistical methods for use in interlaboratory comparison by
proficiency testing.
19 ISO 5725-1-6:1994, Accuracy (trueness and precision) of measurement
methods.
20 C. Grøn, J. Bjerre Hansen, B. Magnusson, A. Nordbotten, M. Krysell, K.
Jebjerg Andersen, U. Lund Nordtest 604, Uncertainty from sampling - A
NORDTEST handbook for sampling planners on sampling quality assurance and
uncertainty estimation Nordtest technical report 604 (2007), www.nordtest.info.
21 R. Bettencourt da Silva and A. Williams (Eds), Eurachem/CITAC Guide:
Setting and Using Target Uncertainty in Chemical Measurement, (1st
ed. 2015).
Available from www.eurachem.org.

of 51
11 Appendices
11.1 Appendix 1: Empty flow scheme for calculations
Before starting: Always identify the main error sources, to make sure that they are
included in the calculations.
Step Action Parameter in matrix by method:
Decide rel/abs
calculations.
(measurand) in range xx.
The target uncertainty is ± _ %.
Relative/absolute uncertainty is calculated.
A control sample
sample
A:
B:
3 Quantify u(bias)
components
standard uncertainty u(x)
uncertainty,
c
u
U 
 2

of 51
11.2 Appendix 2: Empty summary table
(measurand) in (matrix) by (method)
Measurement uncertainty U (95 % confidence interval) is estimated to ± _ (unit)
for (measurand) in (matrix) at a level of _ (unit). The target uncertainty is
± _ (unit). The calculations are based on (control samples/control limits/CRM/
/PT/other).
Value u(x) Comments
Control sample
X = (conc) (unit)
sRw
Other components
Method and laboratory bias
Reference material bias
Proficiency testing bias
Recovery test bias
Reproducibility between laboratories
Proficiency testing sR
Standard method sR
Combined uncertainty, uc, is calculated from __ and bias from __.
Measurand Combined Uncertainty uc Expanded Uncertainty U
c
u

2 =

of 51
11.3 Appendix 3: Error model used in this handbook
This model is a simplification of the model presented in the ISO guide [4]:
e
B
m
y 


 )
(
y measurement result of a sample
m expected or “true” value for y
 method bias
B laboratory bias
e random error at within-laboratory reproducibility conditions
Uncertainty estimation in section 3 -6
2
2
2
)
(
)
( bias
u
Rw
u
uc 

2
)
(Rw
u The estimated variance of e under within-laboratory reproducibility
conditions – intermediate precision. In the ISO guide the repeatability,
sr is used as an estimate of e.
2
)
(bias
u The estimated variance of method bias and laboratory bias.
Uncertainty estimation in section 7
The combined uncertainty uc can also be estimated from reproducibility data.
2
2
2
2
R
r
L
c s
s
s
u 

 - equation A6 ref. [4]
Where sR
2
is the estimated variance under reproducibility conditions and where sL
2
is either the estimated variance of B if one method is used by all laboratories or an
estimated variance of B and  if several different methods have been used in the
collaborative study (interlaboratory comparison) and sr
2
is the estimated variance
of e.
Comment
For samples that are more inhomogeneous and have big variations in matrix
opposed to control samples, the estimation of the measurement uncertainty of the
method can become too low. However we recommend the use of repeatability limit
for duplicate analyses r
s
r 
 8
.
2 in order to control sample inhomogeneity.

of 51
11.4 Appendix 4: Uncertainty of bias for NH4-N in section 3.2
Results for a laboratory from proficiency tests of NH4-N in water.
PT Assigned
value xref
Laboratory
result xi
Bias9
(difference)
sR Number
of labs
)
( i
Cref
u
µg L-1
µg L-1
% % %
1999 1 81 83 2.5 10 31 1.80
2 73 75 2.7 7 36 1.17
2000 1 264 269 1.9 8 32 1.41
2 210 213 1.4 10 35 1.69
2001 1 110 112 1.8 7 36 1.17
2 140 144 2.9 11 34 1.89
X + 2.20 Mean u(Cref) 1.52
RMSbias 2.26
Number of proficiency tests: N = 6
RMS of the bias 26
.
2
6
9
.
2
...
7
.
2
5
.
2 2
2
2
2






N
biasi
% (rel)
%
52
.
1
6
89
.
1
17
.
1
69
.
1
41
.
1
17
.
1
80
.
1
)
(
)
( 1










N
Cref
u
Cref
u
N
i
i
A t-test shows that the bias (+2.20 %) is not significant taking into account the
standard uncertainty of the assigned value of 1.52 %.
If the assigned value is a median or a robust mean it is recommended to multiply
the standard deviation by factor of 1.25 to correspond to sR described in this
handbook [18].
If PT provider reports expanded uncertainty, U, of the assigned value according to
ISO 13528 [18] or similar , then U/2 should be used as u(Crefi) for individual PT
instead of calculating uncertainty via sR and nLab.
9
See note in section 6.2.

of 51
11.5 Appendix 5: Raw data for NH4-N - duplicates
Raw data for section 4.2 and 5.3 - Estimation of the sr from duplicates, absolute s at
low levels and relative s at ‘high’ levels.
Concentration < 30 µg L
-1
x1 x2 s
µg L-1
µg L-1
µg L-1
7.46 7.25 0.15
9.01 9.17 0.11
3.60 3.10 0.35
6.48 6.48 0.00
14.49 14.12 0.26
10.84 10.89 0.04
4.61 5.00 0.28
2.60 2.42 0.13
2.80 2.62 0.13
5.84 6.19 0.25
2.12 2.50 0.27
2.30 2.11 0.13
2.52 2.89 0.26
3.71 3.71 0.00
7.43 7.43 0.00
8.83 8.51 0.23
9.12 8.79 0.23
8.24 7.90 0.24
2.62 2.78 0.11
3.33 3.33 0.00
2.69 2.69 0.00
12.09 12.09 0.00
4.24 4.24 0.00
10.49 10.64 0.11
3.68 3.52 0.11
9.37 9.37 0.00
2.22 2.06 0.11
6.10 6.10 0.00
2.96 2.86 0.07
14.02 13.70 0.23
4.24 3.62 0.44
5.10 4.61 0.35
2.78 2.62 0.11
8.52 6.81 1.21
12.82 14.05 0.87
3.17 2.40 0.54
11.28 11.43 0.11
14.31 13.82 0.35
4.01 4.48 0.33
3.27 3.58 0.22
9.98 10.29 0.22
12.56 13.66 0.78
16.2 16.6 0.28
28.8 28.7 0.07
15.8 18.5 1.91
17.7 16.7 0.71
3.35 2.88 0.33
Pooled s 0.44
Concentration > 30 µg L
-1
x1 x2 s Relative s
µg L-1
µg L-1
µg L-1
%
37.6 36.9 0.49 1.3
4490 4413 54.45 1.2
136 125 7.78 6.0
62.6 62.3 0.21 0.3
159 159 0.00 0.0
16540 16080 325.27 2.0
31.3 30.1 0.85 2.8
58.5 60.1 1.13 1.9
741 796 38.89 5.1
130 127 2.12 1.7
29.4 29.2 0.14 0.5
1372 1388 11.31 0.8
36.6 44.7 5.73 14.1
22.6 23.4 0.57 2.5
34.8 33.2 1.13 3.3
92.9 94.0 0.78 0.8
40.6 42.2 1.13 2.7
80.4 86.4 4.24 5.1
78.2 73.8 3.11 4.1
48.9 50.9 1.41 2.8
36.6 35.3 0.92 2.6
51.9 52.2 0.21 0.4
198 207 6.36 3.1
70.3 69.2 0.78 1.1
29.9 30.6 0.49 1.6
31.9 32.4 0.35 1.1
Pooled s % 3.8

of 51
11.6 Appendix 6: Raw data for dissolved oxygen in Section 5.3
Data plotted in Figure 7 - oxygen in sea water. Range equals the difference
between Result 1 and 2. A pooled s is calculated from the s calculated for each
duplicate.
x1 x2 Range s
mg L-1
mg L-1
mg L-1
mg L-1
8.90 8.91 -0.01 0.007
8.99 9.01 -0.02 0.014
8.90 8.90 0.00 0.000
9.11 9.12 -0.01 0.007
8.68 8.64 0.04 0.028
8.60 8.51 0.09 0.064
8.61 8.61 0.00 0.000
8.02 8.00 0.02 0.014
7.05 7.08 -0.03 0.021
6.98 7.01 -0.03 0.021
7.13 7.10 0.03 0.021
6.79 6.78 0.01 0.007
6.55 6.55 0.00 0.000
6.53 6.53 0.00 0.000
4.68 4.68 0.00 0.000
5.28 5.33 -0.05 0.035
7.42 7.40 0.02 0.014
7.62 7.63 -0.01 0.007
5.88 5.88 0.00 0.000
6.03 6.06 -0.03 0.021
6.33 6.33 0.00 0.000
5.90 5.90 0.00 0.000
6.24 6.27 -0.03 0.021
6.02 6.02 0.00 0.000
9.13 9.11 0.02 0.014
9.10 9.14 -0.04 0.028
8.50 8.44 0.06 0.042
8.73 8.71 0.02 0.014
8.09 8.09 0.00 0.000
7.56 7.58 -0.02 0.014
6.30 6.32 -0.02 0.014
6.43 6.44 -0.01 0.007
7.25 7.34 -0.09 0.064
7.25 7.31 -0.06 0.042
8.00 8.03 -0.03 0.021
8.38 8.29 0.09 0.064
9.23 9.29 -0.06 0.042
9.09 9.08 0.01 0.007
9.37 9.36 0.01 0.007
9.38 9.37 0.01 0.007
9.32 9.25 0.07 0.049
8.47 8.49 -0.02 0.014
8.27 8.28 -0.01 0.007
8.37 8.31 0.06 0.042
8.09 8.15 -0.06 0.042
8.05 8.03 0.02 0.014
7.38 7.4 -0.02 0.014
7.49 7.49 0.00 0.000
4.52 4.49 0.03 0.021
4.45 4.44 0.01 0.007
4.29 4.27 0.02 0.014
Pooled s 0.0252

of 51
11.7 Appendix 7: Raw data for BOD in section 8.2
Results in mg L-1
O2 consumption. The certified value and expanded uncertainty of
the CRM is 206 ± 5 mg L-1
. As the average of two results is always reported for
test samples, the sRw is also calculated from the average of each sample pair in the
internal quality control.
Date x1 x2 Average
mg L-1
mg L-1
mg L-1
2000-12-09 219 215 217
2001-03-01 206 221 214
2001-03-01 221 210 216
2001-04-01 215 207 211
2001-09-05 199 218 209
2001-09-19 224 212 218
2001-10-16 216 213 215
2001-01-07 196 215 206
2001-11-28 210 207 209
2001-12-11 228 223 226
2001-12-13 207 229 218
2002-01-15 207 208 208
2002-01-22 224 214 219
2002-01-30 201 214 208
2002-02-11 219 223 221
2002-03-06 217 218 218
2002-09-18 206 228 217
2002-10-01 215 226 221
Average 214.8
s 5.6
s % 2.6

of 51
11.8 Appendix 8: Template for evaluation of uncertainty
Measurement Uncertainty
from
Quality Control and Validation Data
Name of analytical procedure:
Analyte:
Measurement ranges Calculations -
Relative in % or
absolute in concentration units
Measurement range 1
Measurement range 2
Measurement range 3
Short description of the analytical procedure
Corresponding standard procedure/method
Target uncertainty

of 51
Within laboratory reproducibility u(Rw)
Control sample: Low Middle High
Composition of control sample
Mean value
Standard deviation, s
Number of determinations, n
Number of months
Assigned value
Estimate of sRw from the warning limits of the control chart
Warning limits ±


2
limits
warning
sRw
Conc. (abs)
% (relative)
List of differences in the procedure/method or property of control samples
compared with test samples and if possible also an indication of size. From
size an estimate of standard uncertainty, u can be made.
Difference Size u
1
2
3
4
Differences could e.g. be sample amount or matrix, instability, temperature,
inhomogeneity, impurities that influence the measurement result.
Inhomogeneity of test samples can be assessed by running duplicates. If
there are important differences increased, within-lab reproducibility can be
calculated below.
Estimation of an increased sRw
Control sample


 2
2
)
(
)
( s
difference
Rw s
s
Rw
u

of 51
Bias – method and laboratory bias from CRM
Bias is a systematic error or mean difference from an assigned value.
Use one page for each matrix or concentration level where appropriate.
(Here you have a choice to do the calculations in concentration (absolute) or
relative in %. Please fill in unit used for each column)
Concentration Range:
One CRM. Uncertainty in the assigned value u(Cref) = U(Cref)/2.
CRM Own lab results Cert.
value
U(Cref) u(Cref) n bias =
Lab –
CRM
Relative bias =
(Lab-CRM)/CRM · 100
Mean Sbias
If there is only one CRM there is only one bias value but several
measurements and the following equation is applied:
  2
2
2
)
(
)
( Cref
u
n
s
bias
bias
u CRM











Where n = number of measurement on the CRM and sbias is the obtained
standard deviation from measurements on the CRM.
Several CRM – uncertainty in assigned value is u(Cref) = U(Cref)/2
CRM Own lab results Cert.
value
U(Cref) u(Cref) bias =
Lab –
CRM
Relative bias =
(Lab-CRM)/CRM · 100
Mean sCRM
RMSbias
Number of CRM samples, N =
Root Mean Square is
N
bias
RMS
i
bias


2
)
(
=
Mean value of )
(Cref
u =
Estimate from several CRM: 2
2
)
(
)
( Cref
u
RMS
bias
u bias 
 =

of 51
Bias – method and laboratory bias from PT
Bias is a systematic error or mean difference from an assigned value.
Use one page for each matrix or concentration level where appropriate.
(Here you have a choice to do the calculations in concentration (absolute) or
relative in %. Please fill in unit used for each column)
Concentration range:
Proficiency Testing (PT)
Data from last 10 PT – minimum six!
Year Sample
¤
Own
lab
value
PT value bias =
Lab –PT
Relative bias =
(Lab-PT)/PT · 100
sR
10
nLab u(Crefi)
=SR/NLab
RMSbias
Number of PT, NPT =
Root Mean Square,
PT
i
bias
N
bias
RMS


2
)
(
=
Uncertainty in assigned PT value
PT
N
i
i
N
Cref
u
Cref
u


 1
)
(
)
( where
i
Lab
Ri
i
n
s
Cref
u
,
)
( 
Calculation of u(bias)
See section 6 in Nordtest handbook.
From PT: 2
2
)
(
)
( Cref
u
RMS
bias
u bias 
 =
If PT provider reports expanded uncertainty of the assigned value (U)
according to procedure described in ISO 13528, then U/2 should be used as
u(Crefi) for individual PT instead of calculating uncertainty via sR and nLab.
10
If the assigned value is a median or if a robust mean PT value is used, the ‘robust
standard deviation’ reported by PT provider must be multiplied by factor of 1.25 [18].

of 51
Evaluation of expanded measurement uncertainty
 2
2
)
(
)
(
2
2 bias
u
Rw
u
u
U c 




where uc= combined standard uncertainty
Low range – Measurement uncertainty:
Bias from u(Rw) u(bias) uc U = 2 · uc
CRM
PT
Recovery
Middle range – Measurement uncertainty:
CRM
PT
Recovery
High range – Measurement uncertainty:
CRM
PT
Recovery
List over the main contributions to measurement uncertainty and if possible
also an indication of size of uncertainty source in concentration (e.g. mg L-1
)
or in % (relative).
Source Size
1
2
3
4
5

of 51
11.9 Appendix 9: Uncertainty calculations using MUkit.
Step Action
Determination of
ammonium in water
2017-10-22
1
Specify
Measurand
Analyte measured: Ammonium
Concentration range: 30 – 1000 µg L-1
Matrix: Water
Analysis method: EN/ISO 11732
2
Quantify
within-
laboratory
reproducibility
,
Control sample
that covers all
the steps in
the analytical
process
Control samples:
Matrix: Water
Period of measurements: 2001-01-01 -
2002-01-01
Number of control samples: 135
Average concentration: 200 µg L-1
Standard deviation, : 1,67 %
= 1,67 %

of 51
Step Action Determination of ammonium in water 2012-11-22
3
Quantify
method
and
laboratory
bias,
Method and laboratory bias from
Interlaboratory comparisons:
Interlaboratory comparison count, : 6
I 1 2 3 4 5 6
Assigned
concentration,
81
µg L-1
73 µg
L-1
264
µg L-1
210
µg L-1
110
µg L-1
140
µg L-1
Measured
concentration,
83
µg L-1
75
µg L-1
269
µg L-1
213
µg L-1
112
µg L-1
144
µg L-1
2,47
%
2,74
%
1,89
%
1,43
%
1,82
%
2,86
%
Between
laboratories
standard
deviation,
10,0
%
7,00
%
8,00
%
10,0
%
7,00
%
11,0
%
Consensus
value robust
mean or
median,
No No No No No No
Fixed standard
deviation,
10,0
%
7,00
%
8,0
%
10,0
%
7,0
%
11,0
%
Number of
participating
laboratories,
31 36 32 35 36 34
1,80
%
1,17
%
1,41
%
1,69
%
1,17
%
1,89
%
Analyte
measured
ammonium
Matrix water water water water water water
Date
1999-
03-01
1999-
09-01
2000-
03-03
2000-
10-04
2001-
04-04
2001-
10-11
Arranger NIVA NIVA NIVA NIVA NIVA NIVA

of 51
Additional
information
= 1,52 %
= 2,26 %
= 2,73 %
4
Convert
components
to standard
uncertainty
= 1,67 %
= 2,73 %
5
Calculate
combined
standard
uncertainty = 3,20 %
6
Calculate
expanded
uncertainty = 6,4 %
View publication stats
View publication stats

Handbook51version4xUncertentyInEnvLab.pdf

More Related Content

Similar to Handbook51version4xUncertentyInEnvLab.pdf (20)

Recently uploaded (20)

Handbook51version4xUncertentyInEnvLab.pdf