Introduction to measurement
uncertainty
Science
for a safer world
1
Overview
• What is measurement uncertainty?
• Why is measurement uncertainty important?
• The ISO approach
• General approach to evaluating uncertainty
• Rules for converting and combining uncertainties
Measurement uncertainty
ISO definition
“A parameter, associated with the result of a
measurement, that characterises the dispersion of the
values that could reasonably be attributed to the
measurand”
The part of the result after the ±
What information does it give?
10
The concentration of the acid is:
20
0.102 ± 0.0016 mol dm-3
30
40
50
The concentration is between
0.1004 and 0.1036 mol dm-3
A RANGE containing the TRUE VALUE
Why do we need to know the uncertainty?
• Indication of quality of results
– are results “fit for purpose”?
• Interpretation of results
– are these results really different?
– what confidence can be placed in decisions based on results?
• Knowledge of the uncertainty should
– improve understanding of test methods
– give confidence to users of results
– increase acceptability of results
… For these reasons, uncertainty estimation is a requirement
of ISO/IEC 17025
Where does uncertainty come from?
Random and systematic effects
• Random effects
– cause results to vary from one measurement to the next
– cannot be correct for
– can usually be reduced by increasing the number of
measurements and reporting an average
• Systematic effects
– remain constant over a series of measurements
– cause results to be consistently higher (or lower) than expected
– can be corrected for (but not reduced by averaging)
• The effect of gross errors (mistakes!) is not included in
an uncertainty estimate
ISO recommendations
• Uncertainties arise from several contributions
• Two ways of evaluating uncertainty components
– statistical (Type A) and otherwise (Type B)
– should be treated in the same way
• Express as standard deviations
• Combine by “.. the usual method for the combination of
variances”
• Multiply by a (stated) factor if required
– coverage factor (k)
Evaluating uncertainty – general
procedure
Step 1
Be clear about what Write down equation used to calculate result.
is being measured
Step 2
Parameters appearing in the equation will
Identify the sources contribute to the uncertainty. What other factors
of uncertainty will influence the result?
Estimate the size of each uncertainty
Step 3 component (the effect it will have on the result).
Quantify uncertainty Convert all estimates to the same form
components
(standard uncertainty, u).
Step 4
Combine using rules for combination of
Combine the
uncertainties variances. u u 2 u 2 u 2 ....
c 1 2 3
Step 5 Multiply the combined uncertainty by a coverage
Expand combined
factor to obtain an expanded uncertainty. U = [Link]
uncertainty
Uncertainty components
• Uncertainty in a result will arise from several contributions
• Uncertainty contributions combined to give a “total” uncertainty
– uncertainty components must be expressed in the same form to allow
combination
– express as standard deviations => standard uncertainty (u)
• BUT…uncertainty information comes in different forms so may need
to convert
– standard deviations - use as is
– confidence intervals - convert
– expanded uncertainties - convert
– stated ranges - convert
u(x) from a standard deviation s
• For uncertainties associated with single results
– use the standard deviation s
• For uncertainties associated with a mean of n
observations
– convert to standard deviation of the mean
s/ n
u(x) from a confidence interval
• Example:
“Concentration = 1000 mg L-1 ± 3 mg L-1
with a level of confidence of not less than 95%”
t s
x
n
• Rule:
Divide the half-range (3 mg L-1) by 1.96
u(x) from an expanded uncertainty
• Example:
“The concentration is 1000 ± 3 mg L-1 where the
reported uncertainty is an expanded uncertainty,
calculated using a coverage factor of k=2, which gives a
level of confidence of approximately 95%”
• Rule:
Divide the expanded uncertainty by the coverage factor,
k
U( x )
u x
k
u(x) from a stated range
A range (±a) is given without an Assume rectangular distribution
estimate of the distribution or divide a by 3 2a
confidence level. No reason to
a
expect values close to mean are u x
more likely than values at the 3
extremes.
A range (±a) is given without an
Assume triangular distribution
estimate of the distribution or
divide a by 6 2a
confidence level. Values close to
mean are more likely than values a
at the extremes. u x
6
Combination of uncertainties (1)
• Addition or subtraction
y = a + b - c +…..
Uncertainties are combined as standard deviations
u( y ) u(a)2 u(b)2 u(c )2 ....
ua2 ub2
ub
ua
Combination of uncertainties (2)
• Product or quotient
ab
y
c
• Uncertainties are expressed as relative standard
deviations
2 2 2
u( y ) u (a ) u ( b ) u (c )
y a b c
Combination of uncertainties (3)
• Example: 0.05 g pure chemical dissolved in water and made up to
50 mL in a volumetric flask
– what is the uncertainty in the concentration of the solution (g mL-1)?
Parameter Value Standard uncertainty u(xi) Relative uncertainty
xi u(xi)/xi
Mass of
0.05 g 0.00020 g 0.004
chemical (m)
Volume (v) 50 mL 0.066 mL 0.00132
u c
2 2
m 0.0002 0.066
c 0.001 g mL-1 0.00421
v c 0.05 50
0.00421 0.001 =
u(c)
4.2 x 10-6 g mL-1
Expanded uncertainty U
uc -> 68%
U = 2uc -> 95%
U = [Link]
Summary
• Measurement uncertainty is a range containing the
true value
– has contributions from all significant factors that cause the
result to vary
• The ISO Guide describes the principles of uncertainty
estimation:
– uncertainty components are expressed as standard deviations
– combined according to mathematical rules
– combined uncertainty may require to be increased by a factor
(expanded uncertainty)