The Expression of

Uncertainty in
Measurements
Traceability and
Uncertainty

Accurate measurement enables us to:

Maintain quality control during
production processes
Comply with and enforce laws and
regulations
Undertake research and development
Calibrate instruments and achieve
traceability to a national measurement
standard
Develop, maintain and compare national
and international measurement standards
Successful measurement
depends on the following:

Accurate instruments
Traceability to national standards
An understanding of uncertainty
Application of good measurement
practice
There are many factors that
can cause inaccuracy:

Environmental effects
Inferior measuring equipment
Poor measuring techniques
 What is the difference between
ACCURACY and PRECISION?
The difference between accuracy and
precision is illustrated below by 4 different
archers… each with varying degree of ability.
The bull's-eye in the target represents the true
value of a measurement.
 Why does uncertainty matter?
Calculating and expressing uncertainty is important to anybody wishing
to make good quality measurements.
It is also crucial where uncertainty can influence a pass or failure in a
particular test, and must therefore be reported on a calibration
certificate.
Inaccurate and imprecise (unrepeatable)
Stone age man missed
the bull's-eye and the 3
attempts were not near
each other
Precise but inaccurate
Robin Hood's Merry
Man missed the bull's-
eye but the 3 attempts
were near each
other.

Accurate but imprecise

American Indian's 3
attempts are near the
bull's-eye, but were not
near each other.
 Accurate and precise
Olympic archer hits the
bull's-eye 3 times!

Accuracy is a qualitative term relating the mean of the

measurements to the true value, while precision is representative of
the spread of these measurements. Even when we are precise and
accurate, there will still be some uncertainty in our measurements.
When the uncertainty of a measurement is evaluated and stated,
then the fitness of purpose for a particular application can be
properly understood
Expressing uncertainty
A measurement result is only complete when accompanied by a
statement of its uncertainty. A statement of uncertainty is required in
order to decide if the result is adequate for its intended purpose and
consistent with other similar results.
It does not matter how accurate a measuring instrument is considered to
be, the measurements made will always be subject to a certain amount of
uncertainty.
In order to express the uncertainty of a measurement, we need to
evaluate as accurately as possible the errors associated with that
particular measurement.
For example - we might say that a particular stick is 200 centimeters
long, plus or minus 1 centimeter, at a 95% confidence level. This is
written: 200 cm ±1 cm at a level of
confidence of 95%
This means we are 95% sure that the length of the
stick is between 199 centimeters and 201
centimeters.
 What is traceability?
Traceability is a method of ensuring that a measurement (even with
its uncertainties) is an accurate representation of what it is trying
to measure.
What is traceability to national standards?
The simple and basic concept behind calibration is that measuring equipment should be tested
against a standard of higher accuracy.
 It should be possible to demonstrate an unbroken
chain of comparisons that ends at a national
standards body such as NIS. This demonstrable
linkage to national standards with known accuracy
is called 'traceability'.
National standards laboratories such as NIS also
routinely undertake international comparisons in
order to establish worldwide consensus on the
accepted value of fundamental measurement units.
 Representatives of seventeen nations signed the Convention of the Metre (Convention
du Mètre) on 20th May 1875 in Paris. This diplomatic treaty provided the foundations
for the establishment of the Système International d'Unités (International System of
Units, international abbreviation SI) in 1960. Since then, national standards
laboratories have cooperated in the development of measurement standards that are
traceable to the SI.
Any organisation can achieve traceability to national standards through the correct use of
an appropriate traceable standard from NIS.
1. Measurement

1.1 What is a measurement?

A measurement tells us about a property of something. It might
tell us how heavy an object is, or how hot, or how long it is.
A measurement gives a number to that property.

Measurements are always made using an instrument of some

kind. Rulers, stopwatches, weighing scales, and
thermometers are all measuring instruments.

The result of a measurement is normally in two parts: a

number and a unit of measurement, e.g.' How long is it?
... 2 metres.’
1.2 What is not a measurement?

There are some processes that might seem to

be measurements, but are not. For example,
comparing two pieces of string to see which
is longer is not really a measurement. Counting
is not normally viewed as a measurement.
Often, a test is not a measurement: tests
normally lead to a ‘yes/no’ answer or a
‘pass/fail’ result. (However, measurements
may be part of the process leading up to a test
result.)
2 Uncertainty of
measurement
2.1 What is uncertainty of measurement?
The uncertainty of a measurement tells us something about its
quality.
Uncertainty of measurement is the doubt that exists
about the result of any measurement. You might think
that well-made rulers, clocks and thermometers should
be trustworthy, and give the right answers. But for
every measurement - even the most careful - there is
always a margin of doubt. In everyday speech, this
might be expressed as ‘give or take’ ... e.g. a stick
might be two metres long ‘give or take a centimetre’.
2.2 Expressing uncertainty of measurement

Since there is always a margin of doubt about any measurement,

we need to ask ‘How big is the margin?’ and ‘How bad is the
doubt?’ Thus, two numbers are really needed in order to quantify
an uncertainty. One is the width of the margin, or interval. The
other is a confidence level, and states how sure we are that the
‘true value’ is within that margin.
For example:
We might say that the length of a certain stick measures 20
centimetres plus or minus 1 centimetre, at the 95 percent
confidence level. This result could be written: 20 cm ±1 cm, at a
level of confidence of 95%.
The statement says that we are 95 percent sure that the stick is
between 19 centimetres and 21 centimetres long. There are other
ways to state confidence levels. More will be said about this later
on, in Section 7.
2.3 Error versus uncertainty
It is important not to confuse the terms ‘error’ and
‘uncertainty’.
Error is the difference between the measured value
and the ‘true value’ of the thing being measured.
Uncertainty is a quantification of the doubt about the
measurement result.
Whenever possible we try to correct for any known
errors: for example, by applying corrections from
calibration certificates. But any error whose value we
do not know is a source of uncertainty.
2.4 Why is uncertainty of measurement important?

You may be interested in uncertainty of measurement simply

because you wish to make good quality measurements and
to understand the results. However, there are other more
particular reasons for thinking about measurement uncertainty
You may be making the measurements as part of a:

• calibration - where the uncertainty of measurement must be

reported on the certificate.
• test - where the uncertainty of measurement is needed to
determine a pass or fail or to meet a
• tolerance - where you need to know the uncertainty before
you can decide whether the tolerance is met
... or you may need to read and understand a calibration
certificate or a written specification for a test or
measurement.
 3 Basic statistics on sets of numbers

3.1 ‘Measure thrice, cut once’ ... operator error

There is a saying among craftsmen, ‘Measure thrice, cut once’.
This means that you can reduce the risk of making a mistake
in the work by checking the measurement a second or third
time before you proceed.

In fact it is wise to make any measurement at

least three times. Making only one measurement
means that a mistake could go completely
unnoticed. If you make two measurements and
they
do not agree, you still may not know which is ‘wrong’. But if you
make three measurements, and two agree with each other while
the third is very different, then you could be suspicious about the
third.
3.2 Basic statistical calculations

The two most important

statistical calculations are to find the average or
arithmetic mean, and
standard deviation
for a set of numbers.

3.3 Getting the best estimate - taking the average

of a number of readings
 3.4 How many readings should you average?
The ideal would be to find the mean from an infinite set of values.
The more results you use, the closer you get to that ideal estimate
of the mean.

What is a good number?

As a rule of thumb usually between 4 and 10 readings is sufficient.
Example 1. Taking the average or arithmetic mean of a number of
values

Suppose you have a set of 10 readings. To find the average, add them
together and divide by the number of values (10 in this case).
The readings are: 16, 19, 18, 16, 17, 19, 20, 15, 17 and 13
The sum of these is: 170
The average of the 10 readings is: 170 = 17
10
 3.5 Spread ... standard deviation
When repeated measurements give different results, we want
to know how widely spread the readings are. The spread of
values tells us something about the uncertainty of a
measurement.
By knowing how large this spread is, we can begin to judge the
quality of the measurement or the set of measurements.
The usual way to quantify spread is standard deviation.
The standard deviation of a set of numbers tells us about how
different the individual readings typically are from the average of
the set.
As a ‘rule of thumb’, roughly two thirds of all readings will fall
between plus and minus (±) one standard deviation of the
average. Roughly 95% of all readings will fall within two
standard deviations. This ‘rule’ applies widely although it is
by no means universal.
The ‘true’ value for the standard deviation can only be found
from a very large (infinite) set of readings. From a moderate
number of values, only an estimate of the standard deviation can
be found. The symbol s is usually used for the estimated
standard deviation.

where xi is the result of the ith measurement and x bar is the

arithmetic mean of the n results considered.
Example 2. Calculating the estimated standard deviation of a
set of values
It is rarely convenient to calculate standard deviations by hand,
with pen and paper alone. But it can be done as follows:
Suppose you have a set of n readings (let’s use the same set of
10 as above).
Start by finding the average:
For the set of readings we used before, 16, 19, 18, 16, 17, 19, 20,
15, 17 and13, the average is 17.
Next, find the difference between each reading and the average,
i.e. -1 +2 +1 -1 0 +2 +3 -2 0 -4,
and square each of these,
i.e. 1 4 1 1 0 4 9 4 0 16.
Next, find the total and divide by n-1 (in this case n is 10, so n-1 is 9),
i.e. 1 + 4 + 1 + 1 + 0 + 4 + 9 + 4 + 0 + 16 40
= = 4 .44
9 9
The estimated standard deviation, s, is found by taking the square
root of the total, i.e. s = 4 .44 = 2 .1
3.7 How many readings do you need to find an estimated
standard deviation?

the more readings you use, the better the estimate will
be. In this case it is the estimate of
uncertainty that improves with the number of readings
(not the estimate of the mean or ‘end
result’). In ordinary situations 10 readings is enough.
For a more thorough estimate, the results
should be adjusted to take into account the number of
readings.
4 Where do errors and uncertainties come from?

• The measuring instrument

• The item being measured
• The measurement process
• ‘Imported’ uncertainties
• Operator skill
• Sampling issues
• The environment
Visual alignment is an operator skill. A
movement of the observer can make an
object appear
to move. ‘Parallax errors’ of this kind can
occur when reading a scale with a pointer.
5 The general kinds of uncertainty in any
measurement

5.1 Random or systematic

The effects that give rise to uncertainty in measurement can be
either:
• random - where repeating the measurement gives a randomly
different result. If so, the more measurements you make, and
then average, the better estimate you generally can expect to
get. or
• systematic - where the same influence affects the result for
each of the repeated measurements (but you may not be able
to tell). In this case, you learn nothing extra just by repeating
measurements. Other methods are needed to estimate
uncertainties due to systematic effects, e.g. different
measurements, or calculations.
5.2 Distribution - the ‘shape’ of the errors

The spread of a set of values can take different forms, or

probability distributions.

5.2.1 Normal distribution

5.2.2 Uniform or rectangular distribution
 5.2.3 Other distributions
More rarely, distributions can have other shapes, for
example, triangular, M-shaped (bimodal or two-peaked)

5.3 What is not a measurement uncertainty?

*Mistakes made by operators

*Tolerances
*Specifications
*Accuracy
*Errors
*Statistical analysis
6 How to calculate uncertainty of measurement
*Identify the sources of uncertainty in the measurement.
*Estimate the size of the uncertainty from each source.
*The individual uncertainties are combined to give an
overall figure.

6.1 The two ways to estimate uncertainties

Type A evaluations - uncertainty estimates using
statistics (usually from repeated readings)
Type B evaluations - uncertainty estimates from any
other information. This could be information from past
experience of the measurements, from calibration
certificates, manufacturer’s specifications, from
calculations, from published information, and from
common sense.
6.2 Eight main steps to evaluating uncertainty

1. Decide what you need to find out from your measurements.

Decide what actual measurements and calculations are
needed to produce the final result.

2. Carry out the measurements needed.

3. Estimate the uncertainty of each input quantity that feeds

into the final result. Express all uncertainties in similar
terms.

4. Decide whether the errors of the input quantities are

independent of each other. If you think not, then some
extra calculations or information are needed.
5. Calculate the result of your measurement
(including any known corrections for things such as
calibration).

6. Find the combined standard uncertainty from all

the individual aspects.

7. Express the uncertainty in terms of a coverage

factor together with a size of the uncertainty
interval, and state a level of confidence.

8. Write down the measurement result and the

uncertainty, and state how you got both of these.
 7 Other things you should know before making
an uncertainty calculation
Uncertainty contributions must be expressed in similar terms
before they are combined. Thus, all the uncertainties must be
given in the same units, and at the same level of confidence.

7.1 Standard uncertainty

All contributing uncertainties should be expressed at the same

confidence level, by converting them into standard uncertainties.
A standard uncertainty is a margin whose size can be thought
of as ‘plus or minus one standard deviation’. The standard
uncertainty tells us about the uncertainty of an average (not just
about the spread of values). A standard uncertainty is usually
shown by the symbol u (small u), or u(y) (the standard uncertainty
in y).
7.1.1 Calculating standard uncertainty for a Type A evaluation

When a set of several repeated readings has been taken (for a

Type A estimate of uncertainty), the mean, , and estimated
standard deviation, s, can be calculated for the set. From these,
the estimated standard uncertainty, u, of the mean is calculated
from:

where n was the number of measurements in the set. (The

standard uncertainty of the mean has historically also been
called the standard deviation of the mean, or the standard error
of the mean.)
 7.1.2 Calculating standard uncertainty for a Type B
evaluation
Where the information is more scarce (in some Type B
estimates), you might only be able to estimate the upper
and lower limits of uncertainty. You may then have to
assume the value is equally likely to fall anywhere in
between, i.e. a rectangular or uniform distribution. The
standard uncertainty for a rectangular distribution is found
from:

where a is the semi-range (or half-width) between the upper and

lower limits.
7.1.3 Converting uncertainties from one unit of
measurement to another

Uncertainty contributions must be in the same units before

they are combined. As the saying
goes, you cannot ‘compare apples with pears’.

For example, in making a measurement of length, the

measurement uncertainty will also eventually be stated in
terms of length. One source of uncertainty might be the
variation in room temperature. Although the source of this
uncertainty is temperature, the effect is in terms of length,
and it must be accounted for in units of length. You might
know that the material being measured expands in length
by 0.1 percent for every degree rise in temperature. In
that case, a temperature uncertainty of ±2 °C would give a
length uncertainty of ±0.2 cm in a piece of the material 100
cm long.
7.2 Combining standard uncertainties

Individual standard uncertainties calculated by Type A or Type

B evaluations can be combined validly by ‘summation in
quadrature’ (also known as ‘root sum of the squares’). The
result of this is called the combined standard uncertainty,
shown by uc or uc(y).

Summation in quadrature is simplest where the result of a

measurement is reached by addition or subtraction. The more
complicated cases are also covered below for the
multiplication and division of measurements, as well as for
other functions.
7.2.1 Summation in quadrature for addition and
subtraction
The simplest case is where the result is the sum of a series
of measured values (either added together or subtracted).
For example, you might need to find the total length of a
fence made up of different width fence panels. If the standard
uncertainty (in metres) in the length of each fence panel was
given by a, b, c, etc., then the combined standard uncertainty
(in metres) for the whole fence would be found by squaring
the uncertainties, adding them all together, and then taking
the square root of the total ,i.e.
7.2.2 Summation in quadrature for multiplication or division

For more complicated cases, it can be useful to work in terms of

relative or fractional uncertainties to simplify the calculations. For
example, you might need to find the area, A, of a rectangular
carpet, by multiplying the length, L, by the width, W (i.e. A=L×W).
The relative or fractional uncertainty in the area of the carpet can
be found from the fractional uncertainties in the length and width.
For length L with uncertainty u(L),the relative uncertainty is u(L)/L.
For width W, the relative uncertainty is u(W)/W. Then the relative
uncertainty u(A)/A in the area is given by

This equation can also be used (in exactly the same form) for a case where the
result is a quotient of two values (i.e. one number divided by another, for
example,length divided by width). In other words, this form of the equation
covers all cases where the operations are multiplication or division.
7.2.3 Summation in quadrature for more complicated
functions
Where a value is squared (e.g. Z2) in the calculation of the final
measurement result, then the
relative uncertainty due to the squared component is in the form

Where a square root (e.g. √Z) is part of the calculation of a result,

then the relative uncertainty due to that component is in the form

For example, you might measure electrical resistance R and

voltage V, and then calculate the resulting power P using the
relationship
In this case, the relative uncertainty u(P)/P in the value of power
would be given by
 7.3 Coverage factor k

we may wish to have an overall uncertainty stated at another

level of confidence, e.g. 95 percent. This re-scaling can be done
using a coverage factor, k. Multiplying the combined standard
uncertainty, uc , by a coveragefactor gives a result which is
called the expanded uncertainty, usually shown by the symbol
U,i.e.

This means that expanded uncertainties given on calibration

certificates, if properly expressed, can be ‘decoded’ into
standard uncertainties.
8 How to express the answer

It is important to express the answer so that a reader can use the

information. The main things to mention are:

• The measurement result, together with the uncertainty figure,

e.g. ‘The length of the stick was 20 cm ±1 cm.’

• The statement of the coverage factor and the level of

confidence. A recommended wording is: ‘The reported uncertainty
is based on a standard uncertainty multiplied by a coverage factor
k = 2, providing a level of confidence of approximately 95%.’and

• How the uncertainty was estimated

9 Example - a basic calculation of uncertainty

Below is a worked example of a simple uncertainty

analysis. It is not realistic in every detail, but
it is meant to be simple and clear enough to illustrate
the method. First the measurement and the
analysis of uncertainty are described. Secondly, the
uncertainty analysis is shown in a table (a
‘spreadsheet model’ or ‘uncertainty budget’).
9.1 The measurement - how long is a piece of string?

Suppose you need to make a careful estimate of the

length of a piece of string. Following the steps listed in
Section 6.2, the process is as follows.
Example 3: Calculating the uncertainty in the length of a
piece of string
Step 1. Decide what you need to find out from your
measurements. Decide what actual
measurements and calculations are needed to produce
the final result. You need to make a measurement of the
length, using a tape measure. Apart from the actual length
reading on the tape measure, you may need to consider:
• Possible errors of the tape measure
- Does it need any correction, or has calibration shown it to
read correctly - and
what is the uncertainty in the calibration?
- Is the tape prone to stretching?
- Could bending have shortened it? How much could it have
changed since it was calibrated?
- What is the resolution, i.e. how small are the divisions on
the tape (e.g. millimetres)?
• Possible errors due to the item being measured

- Does the string lie straight? Is it under- or over-stretched?

-Does the prevailing temperature or humidity (or anything else)
affect its actual length?

- Are the ends of the string well-defined, or are they frayed?

• Possible errors due to the measuring process, and the person

making the measurement
- How well can you line up the beginning of the string with the
beginning of the tape measure?
- Can the tape be laid properly parallel with the string?
- How repeatable is the measurement?
Step 2. Carry out the measurements needed.

You make and record your measurements of length. To be extra

thorough, you repeat the measurement a total of 10 times,
aligning the tape measure freshly each time (probably not very
likely in reality!). Let us suppose you calculate the mean to be
5.017 metres (m), and the estimated standard deviation to be
0.0021 m (i.e. 2.1 millimetres).
For a careful measurement you might also record:
- when you did it
- how you did it, e.g. along the ground or vertically, reversing the
tape measure or not, and other details of how you aligned the
tape with the string.
- which tape measure you used
- environmental conditions (if you think these could affect your
results)
- anything else that could be relevant.
Step 3. Estimate the uncertainty of each input quantity that
feeds into the final result.

Express all uncertainties in similar terms (standard uncertainty, u).

You would look at all the possible sources of uncertainty and estimate the
magnitude of each. Let us say that in this case:
• The tape measure has been calibrated. It needs no correction, but the
calibration uncertainty is 0.1 percent of reading, at a coverage factor k = 2
(for a normal distribution). In this case, 0.1 percent of 5.017 m is close to 5
mm. Dividing by 2 gives the standard uncertainty (for k = 1) to be u = 2.5 mm.
• The divisions on the tape are millimetres. Reading to the nearest division
gives an error of no more than ±0.5 mm. We can take this to be a uniformly
distributed uncertainty (the true readings could lie variously anywhere in the 1
mm interval - i.e.±0.5 mm). To find the standard uncertainty, u, we divide the
half-width (0.5 mm) by√3, giving u = 0.3 mm, approximately.
• The tape lies straight, but let us suppose the string unavoidably has a few
slight bends in it. Therefore the measurement is likely to underestimate the
actual length of the
string. Let us guess that the under estimate is about 0.2 percent, and that the
uncertainty in this is also 0.2 percent at most. That means we should correct
the result by adding 0.2 percent (i.e. 10 mm). The uncertainty is assumed to be
uniformly distributed, in the absence of better information. Dividing the half-
width of the uncertainty (10 mm) by √3 gives the standard uncertainty u = 5.8
mm (to the nearest 0.1 mm).

The above are all Type B estimates. Below is a Type A estimate.

• The standard deviation tells us about how repeatable the placement of the
tape measure is, and how much this contributes to the uncertainty of the mean
value. The estimated standard deviation of the mean of the 10 readings is
found
Step 4. Calculate the result of your measurement (including
any known corrections for things such as calibration). The
result comes from the mean reading, together with the correction
needed for the string lying slightly crookedly,
i.e. 5.017 m + 0.010 m = 5.027 m.

Step 5. Find the combined standard uncertainty from all the

individual aspects. The only calculation used in finding the result
was the addition of a correction, so summation in quadrature can
be used in its simplest form

=6.4 mm (to one decimal place)

Step 6. Express the uncertainty in terms of a coverage factor together
with a size of the uncertainty interval, and state a level of confidence.
For a coverage factor k = 2, multiply the combined standard uncertainty by 2, to
give an expanded uncertainty of 12.8 mm (i.e. 0.0128 m). This gives a level of
confidence of about 95 percent.

Step 8. Write down the measurement result and the uncertainty, and state
how you got both of these.
You might record: ‘The length of the string was 5.027 m ±0.013 m. The
reported expanded uncertainty is based on a standard uncertainty multiplied by
a coverage factor k = 2, providing a level of confidence of approximately 95%.
‘The reported length is the mean of 10 repeated measurements of the string
laid horizontally. The result is corrected for the estimated effect of the string not
lying completely straight when measured. The uncertainty was estimated
according to the method in
9.2 Analysis of uncertainty - preadsheet model

To help in the process of calculation, it can be useful to

summarise the uncertainty analysis or‘uncertainty budget’ in a
spreadsheet as in Table 1 below.
 10 Other statements (e.g. compliance with specification)

In Case (a), both the result and the uncertainty fall inside the specified limits. This
is classed as a ‘compliance’.

In Case (d), neither the result nor any part of the uncertainty band falls within the
specified limits. This is classed as a ‘non-compliance’.

Cases (b) and (c) are neither completely inside nor outside the limits. No firm
conclusion about compliance can be made.
11 How to reduce uncertainty in measurement

• Calibrate measuring instruments (or have them calibrated for you) and
use the calibration corrections which are given on the certificate.
• Make corrections to compensate for any (other) errors you know about.
• Make your measurements traceable to national standards
• Choose the best measuring instruments, and use calibration facilities
with the smallest uncertainties.
• Check measurements by repeating them, or by getting someone else to
repeat them from time to time, or use other kinds of checks. Checking by
a different method may be best of all.
• Check calculations, and where numbers are copied from one place to
another, check this too.
• Use an uncertainty budget to identify the worst uncertainties, and
address these.
• Be aware that in a successive chain of calibrations, the uncertainty
increases at every step of the chain.
12 Some other good measurement practices

Overall, use recognised good practices in measurements, for

example:
• Follow the maker’s instructions for using and maintaining
instruments.
• Use experienced staff, and provide training for measurement.
• Check or validate software, to make sure it works correctly.
• Use rounding correctly in your calculations.
• Keep good records of your measurements and calculations.
Write down readings at the time they are made. Keep a note of
any extra information that may be relevant. If past
measurements are ever called into doubt, such records can be
very useful.
13 Use of calculators

When using calculators or computers to work out uncertainties,

you need to know how to avoid mistakes when using them.

13.1 Calculator keys

Is the average (arithmetic mean)

is the estimate of standard deviation you
gives the standard deviation of the sample itself
 13.2 Calculator and software errors

A calculator is useful for complicated arithmetic, but it can also be a source of error.
Calculators can make mistakes! In particular, they can
sometimes give unexpected results when dealing with very
long numbers. For example, some calculators would wrongly
give:
0.000 000 2 × 0.000 000 2 = 0 (exactly),
when the correct answer is 0.000 000 000 000 04. (Of course,
this would be better expressed
as 2×10-7 × 2×10-7 = 4×10-14.) Even computers can suffer
from this form of rounding error.
To identify this problem, spreadsheet software should be
checked by running through a typical calculation ‘by hand’ to
make sure both methods agree. To avoid these problems with
rounding, it is good practice to work with ‘transformed’ numbers
in your calculations (this is sometimes called ‘scaling’ or
‘coding the data’).
Example 4: Find the average and estimated standard
deviation of 1.000 000 03, 1.000 000 06 and 1.000 000 12
Working in whole numbers, you can find the average of 3, 6
and 12 (which is 7), and then deduce that the average of the
original numbers is 1.000 000 07.
Step by step: you subtract the whole number (1) from
1.000 000 03, 1.000 000 06 and 1.000 000 12, giving
13.4 Rounding
Calculators and spreadsheets can give an answer to many decimal places.
There are some recommended practices for rounding the results:
e.g. 20.1 cm ± 0.2 cm.
• Use a meaningful degree of rounding in calculations. The uncertainty in a
measurement result may define how many decimal places you should report.
For example, if the uncertainty in your result is in the first decimal place, then
the measurement result should probably also be stated to one decimal place,
• Make your calculations to at least one more significant figure
than you eventually require. Be aware of how many significant
figures you need to use when multiplying or dividing or carrying
out more complex calculations.

• Rounding of values should be carried out only at the end of the

calculation, to avoid rounding errors. For example, if 2.346 is
rounded up to 2.35 at an early stage in a calculation, it could
later be rounded up to 2.4. But if 2.346 is used throughout a
calculation it would be correctly rounded to 2.3 at the final stage.

• Although results are finally rounded either up or down,

depending on which is the nearest figure, the rule for rounding
uncertainties is different. The final uncertainty is rounded up
to the next largest figure, not down.
 Understanding the terminology

accuracy
closeness of the agreement between measurement result and
true value. (Accuracy is a qualitative term only.)
bias (of a measuring instrument)
systematic error of the indication of a measuring instrument
calibration
comparison of an instrument against a reference or standard, to
find any errors in the values indicated by the instrument. In
some cases, calibration assigns a relationship between the
input and output of an instrument; for example, calibration of a
resistance thermometer could relate its output (in ohms) to an
input temperature (in degrees Celsius, or in kelvins).
confidence level
number (e.g. 95 %) expressing the degree of confidence in a
result
correction (calibration correction)
number added to an instrument reading to correct for an error,
offset, or bias. (Similarly, a reading may be multiplied or divided by
a correction factor to correct the value.)
correlation
interdependence, or relationship, between data or measured
quantities
coverage factor
number which is multiplied by the combined standard uncertainty
to give an expanded uncertainty for a particular level of
confidence
error
offset or deviation (either positive or negative) from the correct
value
estimated standard deviation
estimate of the standard deviation of the ‘population’ based on a
limited sample
expanded uncertainty
standard uncertainty (or combined standard uncertainty)
multiplied by a coverage factor k, to give a particular level of
confidence
Gaussian distribution
(See normal distribution)
interval (confidence interval)
margin within which the ‘true value’ being measured can be
said to lie, with a given level of
confidence
level of confidence
number (e.g. 95 %) expressing the degree of confidence in
the result
mean (arithmetic mean)
average of a set of numbers
measurand
particular quantity subject to measurement
normal distribution
distribution of values in a characteristic pattern of spread
(Gaussian curve) with values more likely to fall near the mean
than away from it
operator error
mistake
precision
a term meaning ‘fineness of discrimination’ but often misused to
mean ‘accuracy’ or ‘uncertainty’. Its use should be avoided if
possible.
random error
error whose effects are observed to vary randomly
range
difference between the highest and the lowest of a set of values
reading
value observed and recorded at the time of measurement
rectangular distribution
distribution of values with equal likelihood of falling anywhere
within a range
repeatability (of an instrument or of measurement results)
closeness of the agreement between repeated measurements of the same property
under the
same conditions
reproducibility (of an instrument or of measurement results)
closeness of the agreement between measurements of the same property carried out
under
changed conditions of measurement (e.g. by a different operator or a different method,
or at a
different time)
resolution
smallest difference that can be meaningfully distinguished (e.g. a change of one (1) in
the last
place of a digital display)
result (of a measurement)
value obtained from a measurement, either before or after
correction or averaging
sensitivity
change in response (of an instrument) divided by the
corresponding change in the stimulus
standard deviation
a measure of the spread of a set of results, describing how values
typically differ from the average of the set. Where it is not possible
to obtain an infinite set of results (in practice it never is) we
instead use the estimated standard deviation.
standard uncertainty
uncertainty of a measurement expressed as a margin equivalent
to plus and minus (±) one standard deviation.
systematic error
bias or offset (either positive or negative) from the correct value
true value
the value that would be obtained by a perfect measurement
Type A evaluation of uncertainty
evaluation of uncertainty by statistical methods
Type B evaluation of uncertainty
evaluation of uncertainty by non-statistical methods
uncertainty budget
summary of the uncertainty calculations
uncertainty of measurement
quantified doubt about the result of a measurement
uniform distribution
distribution of values with equal likelihood of falling anywhere
within a range