Step 4. Calculating the Combined Uncertainty
8. Step 4. Calculating the Combined Uncertainty
8.1. Standard uncertainties
appropriate to assume a triangular distribution, with a standard deviation of a/ √ 6 (see Appendix
8.1.1. Before combination, all uncertainty
E).
contributions must be expressed as standard uncertainties, that is, as standard deviations. This
EXAMPLE
may involve conversion from some other measure
A 10 ml Grade A volumetric flask is certified to
of dispersion. The following rules give some
within ±0.2 ml, but routine in-house checks
guidance for converting an uncertainty
show that extreme values are rare. The standard
component to a standard deviation.
uncertainty is 0.2/ √ 6 ≈
0.08 ml.
8.1.6. Where an estimate is to be made on the evaluated experimentally from the dispersion of
8.1.2. Where the uncertainty component was
basis of judgement, it may be possible to estimate repeated measurements, it can readily be
the component directly as a standard deviation. If expressed as a standard deviation. For the
this is not possible then an estimate should be contribution to uncertainty in single
made of the maximum deviation which could measurements, the standard uncertainty is simply
reasonably occur in practice (excluding simple the observed standard deviation; for results
mistakes). If a smaller value is considered subjected to averaging, the standard deviation
substantially more likely, this estimate should be of the mean [B.24] is used.
treated as descriptive of a triangular distribution.
8.1.3. Where an uncertainty estimate is derived If there are no grounds for believing that a small from previous results and data, it may already be
error is more likely than a large error, the expressed as a standard deviation. However
estimate should be treated as characterising a where a confidence interval is given with a level
rectangular distribution.
of confidence, (in the form ±a at p % ) then divide
8.1.7. Conversion factors for the most commonly the value a by the appropriate percentage point of
used distribution functions are given in Appendix the Normal distribution for the level of
E.1.
confidence given to calculate the standard deviation.
8.2. Combined standard uncertainty
EXAMPLE
8.2.1. Following the estimation of individual or
A specification states that a balance reading is
groups of components of uncertainty and
within ±0.2 mg with 95 % confidence. From
expressing them as standard uncertainties, the
standard tables of percentage points on the
next stage is to calculate the combined standard
normal distribution, a 95% confidence interval
uncertainty using one of the procedures described
is calculated using a value of 1.96 σ . Using this
below.
figure gives a standard uncertainty of (0.2/1.96) ≈ 0.1.
8.2.2. The general relationship between the combined standard uncertainty u c (y) of a value y
8.1.4. If limits of ±a are given without a and the uncertainty of the independent parameters confidence level and there is reason to expect that
x 1 ,x 2 , ...x n on which it depends is extreme values are likely, it is normally appropriate to assume a rectangular distribution, * u
c (y(x 1 ,x 2 ,.. .)) =
with a standard deviation of a/ √ 3 (see Appendix
E). where y(x 1 ,x 2 ,.. ) is a function of several
EXAMPLE
parameters x 1 ,x 2 ..., c i is a sensitivity coefficient evaluated as c i = ∂ y/ ∂ x i , the partial differential of y
A 10 ml Grade A volumetric flask is certified to
with respect to x
i and u(y,x i ) denotes the uncertainty in y arising from the uncertainty in x .
within ±0.2 ml. The standard uncertainty is
0.2/ √ 3 ≈ 0.12 ml.
Each variable's contribution u(y,x i ) is just the
8.1.5. If limits of ±a are given without a
confidence level, but there is reason to expect that
* The ISO Guide uses the shorter form u i (y) instead of
extreme values are unlikely, it is normally
u (y,x i )
Page 25
Quantifying Uncertainty Step 4. Calculating the Combined Uncertainty
square of the associated uncertainty expressed as that this method, or another appropriate computer-
a standard deviation multiplied by the square of based method, be used for all but the simplest the relevant sensitivity coefficient. These
cases.
sensitivity coefficients describe how the value of
8.2.6. In some cases, the expressions for
y varies with changes in the parameters x 1 ,x 2 etc.
combining uncertainties reduce to much simpler
N OTE : Sensitivity coefficients may also be evaluated
forms. Two simple rules for combining standard
directly by experiment; this is particularly
uncertainties are given here.
valuable where no reliable mathematical description of the relationship exists.
Rule 1
8.2.3. Where variables are not independent, the For models involving only a sum or difference of relationship is more complex:
quantities, e.g. y=(p+q+r+...), the combined standard uncertainty u c (y) is given by u y x
i , j ... )) =
u ( y ( p , q .) = u ( p ) 2 + u ( q ) 2 + .....
Rule 2
where u(x i ,x k ) is the covariance between x i and x k and c i and c k are the sensitivity coefficients as
For models involving only a product or quotient, described and evaluated in 8.2.2. The covariance
e.g. y=(p × q × r × ...) or y= p / (q × r × ...) , the is related to the correlation coefficient r ik by
combined standard uncertainty u c (y) is given by u (x i ,x k ) = u(x i ) ⋅ u (x k ) ⋅ r ik
where -1 ≤ r ik ≤ 1.
p
q
8.2.4. These general procedures apply whether where (u(p)/p) etc. are the uncertainties in the the uncertainties are related to single parameters,
parameters, expressed as relative standard grouped parameters or to the method as a whole.
deviations.
However, when an uncertainty contribution is associated with the whole procedure, it is usually
N OTE Subtraction is treated in the same manner as
expressed as an effect on the final result. In such
addition, and division in the same way as
cases, or when the uncertainty on a parameter is
multiplication.
expressed directly in terms of its effect on y, the
8.2.7. For the purposes of combining uncertainty sensitivity coefficient ∂ y / ∂ x i is equal to 1.0.
components, it is most convenient to break the original mathematical model down to expressions
EXAMPLE
which consist solely of operations covered by one
A result of 22 mg l -1 shows a measured standard
of the rules above. For example, the expression
deviation of 4.1 mg l -1 . The standard uncertainty u(y) associated with precision under
these conditions is 4.1 mg l . The implicit
model for the measurement, neglecting other
should be broken down to the two elements (o+p)
factors for clarity, is
and (q+r). The interim uncertainties for each of
y = (Calculated result) + ε
these can then be calculated using rule 1 above;
where ε represents the effect of random
these interim uncertainties can then be combined
variation under the conditions of measurement.
using rule 2 to give the combined standard
∂ y / ∂ε is accordingly 1.0
uncertainty.
8.2.8. The following examples illustrate the use of sensitivity coefficient is equal to one, and for the
8.2.5. Except for the case above, when the
the above rules:
special cases given in Rule 1 and Rule 2 below,
EXAMPLE 1
the general procedure, requiring the generation of partial differentials or the numerical equivalent
y = (p-q+r) The values are p=5 . 02, q=6 . 45 and
must be employed. Appendix E gives details of a
r =9 . 04 with standard uncertainties u(p)=0 . 13,
numerical method, suggested by Kragten [H.12], u (q)=0 . 05 and u(r)= 0 . 22. which makes effective use of spreadsheet
y = 5.02 - 6.45 + 9.04 = 7.61
software to provide a combined standard uncertainty from input standard uncertainties and
a known measurement model. It is recommended
Page 26
Quantifying Uncertainty Step 4. Calculating the Combined Uncertainty
EXAMPLE 2
• Any knowledge of the number of values used
y = (op/qr). The values are o=2.46, p=4.32,
to estimate random effects (see 8.3.3 below).
q =6.38 and r=2.99, with standard uncertainties
8.3.3. For most purposes it is recommended that k
of u(o)=0.02, u(p)=0.13, u(q)=0.11 and u(r)=
is set to 2. However, this value of k may be
insufficient where the combined uncertainty is
y =( 2.46 × 4.32 ) / (6.38 ×
based on statistical observations with relatively
2 2 few degrees of freedom (less than about six). The
choice of k then depends on the effective number
u ( y ) = 0 . 56 ×
of degrees of freedom.
8.3.4. Where the combined standard uncertainty
is dominated by a single contribution with fewer
⇒ u(y) = 0.56 × 0.043 = 0.024
than six degrees of freedom, it is recommended that k be set equal to the two-tailed value of
8.2.9. There are many instances in which the Student’s t for the number of degrees of freedom magnitudes of components of uncertainty vary
associated with that contribution, and for the with the level of analyte. For example,
level of confidence required (normally 95%). uncertainties in recovery may be smaller for high
Table 1 (page 28) gives a short list of values for levels of material, or spectroscopic signals may
vary randomly on a scale approximately proportional to intensity (constant coefficient of
EXAMPLE:
variation). In such cases, it is important to take
A combined standard uncertainty for a weighing
account of the changes in the combined standard
operation is formed from contributions
uncertainty with level of analyte. Approaches
u cal =0.01 mg arising from calibration
include:
uncertainty and s obs =0.08 mg based on the
• standard deviation of five repeated Restricting the specified procedure or
observations. The combined standard
uncertainty estimate to a small range of
2 uncertainty u 2 c is equal to 0 . 01 + 0 . 08 analyte concentrations.
=0.081 mg. This is clearly dominated by the
• Providing an uncertainty estimate in the form
repeatability contribution s obs , which is based on
of a relative standard deviation.
five observations, giving 5-1=4 degrees of freedom. k is accordingly based on Student’s t.
• Explicitly calculating the dependence and
The two-tailed value of t for four degrees of
recalculating the uncertainty for a given
freedom and 95% confidence is, from tables,
result.
2.8; k is accordingly set to 2.8 and the expanded uncertainty U=2.8 × 0.081=0.23 mg.
Appendix E4 gives additional information on these approaches.
8.3.5. The Guide [H.2] gives additional guidance on choosing k where a small number of
8.3. Expanded uncertainty
measurements is used to estimate large random effects, and should be referred to when estimating
8.3.1. The final stage is to multiply the combined degrees of freedom where several contributions standard uncertainty by the chosen coverage
are significant.
factor in order to obtain an expanded uncertainty. The expanded uncertainty is required to provide
8.3.6. Where the distributions concerned are an interval which may be expected to encompass
normal, a coverage factor of 2 (or chosen
a large fraction of the distribution of values which according to paragraphs 8.3.3.-8.3.5. using a level could reasonably be attributed to the measurand.
of confidence of 95%) gives an interval containing approximately 95% of the distribution
8.3.2. In choosing a value for the coverage factor of values. It is not recommended that this interval k , a number of issues should be considered. These
is taken to imply a 95% confidence interval include:
without a knowledge of the distribution • The level of confidence required
concerned.
• Any knowledge of the underlying distributions
Page 27
Quantifying Uncertainty Step 4. Calculating the Combined Uncertainty
Table 1: Student’s t for 95% confidence (2-tailed)
Degrees of freedom
νν t
Page 28
Quantifying Uncertainty Reporting Uncertainty