1

INTRODUCTION TO MEASUREMENTS
AND UNCERTAINTY THEORY

1. Introduction.
Making a measurement means to ascertain the amount of a physical magnitude in
terms of a standard unit. Therefore, it is essential to express the result of a measurement
with a number and the appropiate units. For example, if somebody says that the speed of
an airplane is 800 we are receiving an incomplete information. What do 800 mean?
Meters per second? Kilometers per hour? Miles per hour?

Only when indicating the measured quantity along with its units, i.e. 800 km/h, useful
information is being given. There are different systems of units, directly intertwined by
arithmetic operations. Nowadays, the most used system of units is the Système
International (SI), and this is the one that we will primarily use in the laboratory. The SI
consists of seven base units: kilogram, meter, second, ampere, Kelvin, mol and candela.

Unfortunately, it is not possible to accomplish a measure with no uncertainty. These

uncertainties can be due to multiple causes. Although some uncertainties are caused by
human errors, some others are inherent to the process of measure and impossible to
avoid. The order of magnitude of the total uncertainty in a measure represents its
accuracy. When measuring a physical magnitude, we should know how reliable the
measurement is. That reliability represents the measure precision, which can be known
after estimating the uncertainty. In this guide we will learn how to estimate the
uncertainty committed when accomplishing a measure in the laboratory.

2. Types of uncertainties.
There are three categories of uncertainties: Precision, systematic uncertainties and
random uncertainties.

• Precision: Every measuring equipment have at least one scale. The smallest scale
graduation determines how precisely one can infer the value of a parameter, i.e.
the accuracy we can reach.

Universidad Carlos III de Madrid ©Physics Department 2011 Physics Laboratory

 2

For example, a conventional ruler is divided into centimeters and millimeters.

Therefore, any length that we measure using that ruler will have an uncertainty of
about 1 mm. This uncertainty can only be decreased if we use a device with higher
accuracy.
Then, the Precision correlates to the equipment accuracy, and it will be referred as
in this guide.

• Systematic errors (or determinate errors): The most common types are
instrumental error, operator error, and method error. Instrumental errors are
usually due to a bad operation or miscalibration of the measuring instrument. They
are often unidirectional, so they slant the result of the measurement consistently
causing the value to be too large or too small always in the same amount. In
principle, systematic errors can be eliminated if the nature of the bias is identified.

An easy example of a systematic error is a losing time clock. The clock will always
be slow from the right time, but the right time can be calculated if we know the
clock delay. Another solution is to repair the clock!

One way to avoid these errors is to check the correct operation and calibration of
the measuring devices.

• Random errors (or indeterminate errors): They are the result of unavoidable
sources of error shifting randomly the measured value to higher or lower values
over the real one. The random errors, unlike the systematic errors, can be reduced,
but never eliminated.

For example, let us imagine that we want to time one minute using a digital clock
which have a precision of a hundredth of a second. It can be seen that we cannot
manage to reach a display of exactly 60 s. Our reaction time always produces an
uncertainty of a few hundredths of a second. In this case the random error is only
due to the operator skills, and not to the operation or calibration of the clock.

The random error will be designated as . in this guide. In next sections we will
see the mathematical treatment of these uncertainties.

The total error of a measurement is usually a combination of the three types of errors
described above. We don’t know the true value of the total error (if we did it wouldn’t be
an error!), but it should be estimated as we are able to. The result of any physical
measurement is expressed with a numerical value, , and the associated degree of
uncertainty, expressed as ∆ .

Universidad Carlos III de Madrid ©Physics Department 2011 Physics Laboratory

 3

3. Reporting the result of an experimental measurement.

Once we have measured a certain magnitude, , and we know that its uncertainty is
∆ , we should express the result as:

∆ [units] (1)

the measured value and the uncertainty being expressed in the same units.

EXAMPLE

With a graduated ruler a person's height has been measured. The result is 1.76 m,
and the uncertainty is 2 cm. The right way to express this measurement is:

1.76 0.02

4. Direct measurements.
A direct measurement is obtained using a measuring device. In this section we will
study the uncertainties associated to this kind of measurements assuming that they are
free from systematic errors.

4.1. Uncertainty in a single measurement of a magnitude.

When we perform a single measurement of a magnitude , the uncertainty is just the
instrument precision, ∆ . In order to determine we can define two cases
depending on the type of instrument being analogical or digital.

• Analogical instrument: The precision is given as ½ the smallest scale graduation

1
2

For example, the precision of a ruler graduated in millimeters is 0.5 .

• Digital instrument: The precision is given as the smallest value that the device can
measure.

For example, the precision of a digital scale which can measure grams is 1 .

Universidad Carlos III de Madrid ©Physics Department 2011 Physics Laboratory

 4

4.2. Uncertainty in a finite set of measurements of a magnitude.

In general, a single measurement of a magnitude is not reliable enough, as many
factors can drive it wrong, like a misreading of the scale, lack of attention when writing
down the result, etc.
In order to avoid that, one can make repeated measurements of the same magnitude .
The result is a finite set of measurements ( , , … , ), each of them affected by the
precision . But, which result is more accurate? Should we choose only one? In this case,
we will get the best result by giving equal weight to all measurements, i.e., simply
expressing the result as the average or mean, , over the set:
∑
(2)

The standard deviation of the measured values is represented by and is given by

the formula:

∑
(3)

The standard deviation is sometimes referred to as the mean square deviation and it
measures how widely spread the measured values are on either side of the mean.

The final uncertainty, ∆ , in our set of measurements will be the maximum value of
either the precision and the standard deviation:
∆ , (4)
Then, the most precise result will be with an uncertainty ∆ calculated from
equation (4).

EXAMPLE:

15 length measurements from a bar were taken in the laboratory. The results, using a
ruler graduated in millimeters, are listed below:

L ( mm ) L ( mm ) L ( mm )
15.0 14,0 13.5
15.5 15.5 15.5
13.5 15.0 14.0
14.0 14.0 15.5
13.0 14.0 14.0

Universidad Carlos III de Madrid ©Physics Department 2011 Physics Laboratory

 5

Let us try to get the best approximation to the true value of L and estimate its
uncertainty.
As the ruler is an analogical device, the uncertainty in one single measurement is:
∆ 0.5
The best approximation to the true value of L is the mean of the set of
measurement:
∑
14.4
In order to calculate the standard deviation on L , it can be useful to use a table like
the one below:

i Li ± 0.5 mm Li − L ( Li − L) 2
1 15.0 0.6 0.36
2 15.5 1.1 1.21
3 13.5 ‐0.9 0.81
4 14.0 ‐0.4 0.16
5 13.0 ‐1.4 1.96
6 14.0 ‐ 0.4 0.16
7 15.5 1.1 1.21
8 15.0 0.6 0.36
9 14.0 ‐0.4 0.16
10 14.0 ‐0.4 0.16
11 13.5 ‐0.9 0.81
12 15.5 1.1 1.21
13 14.0 ‐0.4 0.16
14 15.5 1.1 1.21
15 14.0 ‐0.4 0.16
∑ 216 10.1

The standard deviation is calculated by means of equation (3).

∑
0.821

Then, the uncertainty of the measurement is:

∆ , 0.5, 0.821 0.821

Universidad Carlos III de Madrid ©Physics Department 2011 Physics Laboratory

 6

5. Significant figures: Rounding off.

When we estimate the uncertainty of a measurement, the result may present a large
number of digits. There is no sense in overstating the precision of the measurement
determination, as we just obtain an estimated value of the uncertainty and the first
nonzero digit determines its magnitude.

In the last example, the uncertainty of the measurement was ∆ 0.821 . The
most significant figure indicates an uncertainty in the range of tenths of millimeter. With
an uncertainty like this, it makes no difference to have an error of 1 thousandth a
millimeter or 2 hundredths a millimeter.

The value of the uncertainty ∆L indicates that the measurement L is not reliable
over tenths a millimeter. That way, it is established that the value of a measurement
cannot have more precision than the uncertainty. The process consists in rounding off the
uncertainty value in order to have only one nonzero digit.

The next criteria will be used to round off the value of a measurement and its
uncertainty:

1. Both numbers should be expressed using the same units.

2. Only one non zero figure should be considered for the uncertainty. For example, an
uncertainty of 0.345 s should be written as 0.3 s and an accuracy of 86 kg as 90 Kg. An
exception is made when the most significant figure is 1 and the next figure is minor or
equal to 5; in this case the number “1” remains, and the next figure remains too. For
example, an error of 0.143 Kg rounds to 0.14 Kg.

In order to get these results we have rounded off the values according to the
following rules:

a) If the first dropped digit is greater than or equal to 5, the last kept figure should
be increased in 1 unit. For example, if we round off 0.861342 s, we should add
one unit to the “8” and express this time period as 0.9 s, since the first dropped
number is 6 > 5.

b) If the first dropped digit is minor than 5, the last kept figure remains the same.
That way, 234.38 m rounded off in the most significant figure will be 200 m.

c) If the first dropped figure equals 5, we may find two situations:

Universidad Carlos III de Madrid ©Physics Department 2011 Physics Laboratory

 7

• In the next dropped figures, there are nonzero values. In this case, last
preserved figure grows in 1 unit. For example, 35.234 s is round to 40 s.

• All the drop figures are zero except a 5 number. In this case, the last kept
figure remains the same. For example, 35.000 N rounds off 30N.

These round off rules are applicable both to the total error value and to the
measurement value.

3. The measurement value should have the same precision than the uncertainty.
Rounding off figures implies to turn into zero all the figures whose order of magnitude is
minor than the uncertainty one. However, differing from the uncertainty, the result can
have more than one nonzero digit. To round off the measurement value it is necessary to
round off the uncertainty value before. For example, if the uncertainty is 0.7 Kg and the
value of the measure is 25.784535 Kg, the final result is expressed as (25.8 ± 0.7) kg. If the
uncertainty was 7 kg, the result would be (26 ± 7) Kg

Let us see the following example as a resume of the previous explanation.

EXAMPLE:

Round off and express these measurements and uncertainties according to the
previous criteria.

Rounded
Uncertainty (m) Measure (m) Result (m)
uncertainty (m)
0.018 0.02 0.987 0.99±0.02
0.068 0.07 25.8251 25.83±0.07
0.072 0.07 25.825 25.82±0.07
0.66 0.7 0.88 0.9±0.7
0.52 0.5 12 12.0±0.5
0.942 0.9 1.867 1.9±0.9
0.987 1 26.97 27±1
11.897 12 356.257 356±12
26 30 364 360±30
340 300 588.6 600±300
370.86 400 25.82 0±400

Universidad Carlos III de Madrid ©Physics Department 2011 Physics Laboratory

 8

6. Absolute and relative errors: Analysis of the uncertainty.

So far we have only considered the absolute error ∆x . In order to compare the error
with a measurement itself, we can use the relative error, calculated by the following
formula:
∆
(5)

Or:
∆
% 100 (6)

If we express it in %:

EXAMPLE:

Suppose that we have measured the distance from the Earth to the Sun ( RTS ) and
from Marte to the Sun ( RMS ), and that the obtained results are:

1.5 0.4 10

22.8 0.4 10

In both results, the absolute error is the same: 0.4·108 km. However, the uncertainty
is much higher in the first case ( RTS ) than in the second one ( RMS ), as the relative error
indicates:

0.4 10
100 27%
1.5 10
0.4 10
100 2%
22.8 10

7. Indirect measurements.
We may find in the laboratory physical magnitudes that cannot be directly measured
but can be calculated from two or more directly measured quantities. In this case we are
making indirect measurements. For example, an indirect measurement is the surface of a
rectangle calculated from the measurement of its sides’ lengths at the laboratory.

Universidad Carlos III de Madrid ©Physics Department 2011 Physics Laboratory

 9

The uncertainty in the indirect measure derives from the uncertainties of the direct
measurements used to calculate it. The method to calculate this uncertainty is known as
propagation of uncertainty.

7.1. Propagation of uncertainty.

Suppose that we want to calculate the value of a magnitude y , dependent on a
series of magnitudes x1 , x2 ,..., xn , which can be known by a direct measurement in the
laboratory:

y = f ( x1 , x2 ,..., xn ) (7)

First, we will calculate the mean and the uncertainty values x i ± ∆xi . The best
estimation of y is obtained by substituting in equation (3) the obtained values for xi .

y = f ( x1, x 2,..., x n) (8)

In order to estimate the uncertainty of ∆xi we can use the differential analysis,
described below.

7.1.1. Differential analysis

Assuming that the uncertainty ∆xi of the variables xi is small enough, it can be
demonstrated that the uncertainty ∆ y can be calculated as:

∂f ∂f ∂f n
∂f
∆y = ∆x1 + ∆x2 + ... + ∆xn = ∑ ∆xi (9)
∂x1 ∂x2 ∂xn i =1 ∂xi
xi xi xi xi

Be aware that all terms in equation (9) are absolute values and ∆xi is a positive
number.

∂f
NOTE: is the partial derivative of f with respect to the variable xi ; i. e. it is the
∂xi
derivative of the function f with respect to xi with the other variables held constant (see
the examples). The subindex xi indicates that this value should be evaluated in the result
of the partial derivative.

Universidad Carlos III de Madrid ©Physics Department 2011 Physics Laboratory

 10

EXAMPLE:

A physical magnitude y is determined by the equation: y = a1 x1 − a2 x2 where a1 and

a2 are constants without uncertainty, and the uncertainties of x1 and x2 are ∆x1 and ∆x2
respectively. Calculate an equation for the uncertainty of y .

The partial derivatives are:

and

hence:

∆ | | ∆ | | ∆

EXAMPLE:

A physical magnitude y is determined by the equation: y = x1n ⋅ x2m , where m and n are
constants without uncertainty, and the uncertainties of x1 and x2 are ∆x1 and ∆x2
respectively. Calculate an equation for the uncertainty of y .

Solution:

The partial derivatives are:

and

hence:

| | ∆ 1
∆ 2 1 ∆

EXAMPLE:

Suppose that we have measured the diameter of a sphere with a precision of 1 cm:
150 1 cm. Calculate the area (A) and the volume (V) of the sphere and the
corresponding uncertainties.

The surface and the volume of the sphere can be calculated as:

A = 4π R 2
V = 43 π R 3

Universidad Carlos III de Madrid ©Physics Department 2011 Physics Laboratory

 11

The radius R is R = D
2 = 75cm and ∆R = ∆D
2 = 0.5cm . Therefore, R = 75.0 ± 0.5 cm . For
the area and the volume we will have:

A = 4π R 2 = 70685.8 cm 2
∆A = ∂A
∂R ∆R = 8π R ∆R = 942.5 cm 2

V = 43 π R 3 = 1767145.9 cm 3
∆V = ∂V
∂R ∆R = 4π R 2 ∆R = 35342.9 cm3

The final results will be:

A = 70700 ± 900 = (70.7 ± 0.9)103 cm 2

V = 1770000 ± 40000 = (17.7 ± 0.4)105 cm 3

EXAMPLE:

Two resistors, R1 and R2 , have been measured 5 times each: R1 =9.5, 9.8, 10.2, 9.9,
10.1 Ω (ohms); =15,5, 15,2, 14,8, 15,2, 15,0 Ω . The precision of an individual
measurement is 0.1 Ω . Let us find the most likely value of R1 and R2 and estimate their
corresponding uncertainties. If these resistors are connected in parallel, calculate the
equivalent resistor of this circuit.

We can resume the measurements of R1 and R2 in the following tables:

i R1i ± 0.1Ω R1i − R1 ( R1i − R1 )2 i R2i ± 0.1 Ω R2i − R2 ( R2i − R2 ) 2

1 9.5 ‐ 0.4 0.16 1 15.5 0.36 0.1296
2 9.8 ‐ 0.1 0.01 2 15.2 0.06 0.0036
3 10.2 0.3 0.09 3 14.8 ‐ 0.34 0.1156
4 9.9 0.0 0.00 4 15.2 0.06 0.0036
5 10.1 0.2 0.04 5 15.0 ‐ 0.14 0.0196
∑ 49.5 0.30 ∑ 75.7 0.272

Universidad Carlos III de Madrid ©Physics Department 2011 Physics Laboratory

 12

The mean values of R1 and R2 are:

R1 = ∑5R1i = 9.9 Ω

R2 = ∑5R2 i = 15.14 Ω

The standard deviations are:

ε acc ( R1 ) =
∑( R1i − R1 ) 2
= 0.24 Ω
5

ε acc ( R2 ) =
∑( R 2i − R2 ) 2
= 0.23 Ω
5

The final results are:

∆ R1 = max(ε p ( R1 ), ε acc ( R1 )) = max(0.1, 0.24) = 0.24 Ω

R1 = 9.9 ± 0.2 Ω

∆ R2 = max(ε p ( R2 ), ε acc ( R2 )) = max(0.1, 0.23) = 0.23 Ω

R2 = 15.1 ± 0.2 Ω

For resistors in parallel, the equivalent resistor is:

1 1 1 RR
= R + R ⇒Req = R1+R2
Req 1 2 1 2

And the best value of Req is obtained substituting R1 and R2 for their corresponding
R1 and R2 mean values, previously calculated:

R1 R2
Req = = 5.986 Ω
R1 + R2
The uncertainty of Req is:

∂Req ∂Req
∆ Req = ∆ R1 + ∆ R2
∂R1 ∂R2
2 2
R2 R1
= ∆ R1 + ∆ R2 = 0.104 Ω
( R1 + R2 ) 2
( R1 + R2 ) 2

Universidad Carlos III de Madrid ©Physics Department 2011 Physics Laboratory

 13

Where mean values R1 and R2 have been substituted when evaluating the partial
derivatives. The final result for Req is:

Req = 6.0 ± 0.1 Ω .

8. Physical laws: Analysis of the dependence between

variables.
Many physical laws establish the dependence of one variable in terms of another one.
For example, the relation between the speed and the acceleration of an object affected by
gravity is:

g t (10)

This equation establish a linear relation between and , where is the intercept
and is the slope.

A physical law is valid if the predicted relation between variables is experimentally

verified. In the example, we should measure the speed v(ti ) of an object at different
sampling instants ti and verify if the linear relation in equation (10) is true. If so, we can
determine the values of v0 and g from the experimental data.

The next two sections explain two methods to analyze the linear relations between
two variables and to calculate the parameters of proportionality.

50

40

δy
30
Y axis

20 δx

10 m=δ x / δ y
b
0
0 1 2 3
X axis
Figure 1. Linear fitting: m is the slope of the straight line and b is the intercept.

Universidad Carlos III de Madrid ©Physics Department 2011 Physics Laboratory

 14

8.1. Graphical method.

When a variable y is linearly proportional to another variable x , the plot of a given
set of pairs , in a graph leads to a straight line. The general equation of a straight
line is:

y = mx + b (11)

Where m is the straight line's slope and b is the intercept.

Therefore, if the experimental data collected in the laboratory are supposed to obey a
linear relation, when plotting them a spotted graph will be obtained, and a straight line
can be drawn, as shown in Figure 1.

We can draw the “best” straight line through all the points, and we can calculate the
numerical values of m and b as shown in Figure.1. It is important to see that the units for
m and b are:

[ y]
[ m] = , [b] = [ y ] (12)
[ x]

If a linear relation is not found when plotting the experimental data, it is evident that
they do not obey a linear relation, and is not worth calculating m and b.

40

2
m=δ x / δ y=14.2/1.5=9.5 m/s
30
Speed (m/s)

20 δ y=14.2 m/s

10 δ x=1.5 s

b=0.3 m/s
0
0 1 2 3 4
Time (s)
Figure 2. See the example in the text.

In the proposed example, the result of a set of measurements in the laboratory is a

series of pairs , .

Universidad Carlos III de Madrid ©Physics Department 2011 Physics Laboratory

 15

t (s) v (t ) (m/s)
0.94 13.21
1.58 17.23
1.96 23.99
2.66 26.74
2.91 35.57
3.76 38.43

Figure 2 shows the data plot.

It can be observed that the experimental data spread as a straight line, as predicted
by the physical law described in equation (10). Then, applying the described graphical
method, and values can be calculated as: 9.5 ⁄ and 3 ⁄ .

8.2. Least Squares Fitting

The graphical method is a rough way to fit data to a straight line and obtain its slope
and intercept. It is easy, but not rigorous enough. These magnitudes can be obtained
more accurately by using the least squares method.

Let us consider the equation of a straight line:

y = m x+b (13)

The least squares method is a mathematical procedure for finding the best‐fitting
straight line to a given set of points by minimizing the sum of the squares of the distances
from the experimental points to the straight line. If the experimental points are
( x1 , y1 ),L , ( xn , yn ) , the best‐fitting line will be found for values of m (the slope of the
straight line) and b (the intercept) obeying the condition:

∑ (14)

Universidad Carlos III de Madrid ©Physics Department 2011 Physics Laboratory

 16

It can be demonstrated than the values m and b obeying this condition are the
solution of the following system of equations:





∑y i = m ∑ xi + bn
(15)



∑x y i i = m ∑ xi2 + b ∑ xi

So the values of m and b are given by:

∑ y n(8) i

m=
∑ x y ∑x = ∑ y ∑ x − n ∑ x y
i i i i i i i
(16)
∑ x n(9) (∑ x ) − n ∑ x
i i
2 2
i

∑ x ∑x 2
i i

∑ x ∑ y (11)
i i

b=
∑ x ∑x y = ∑ x ∑ x y − ∑ y ∑ x
2
i i i i i i i
2
i
(17)
∑ x n(12) (∑ x ) − n ∑ x
i i
2 2
i

∑ x ∑x 2
i i

m and b are obtained from the experimental data by applying (16) and (17), so they
have uncertainties ( ∆m , ∆b ) associated to them. These uncertainties can be estimated by
applying the following equations:

Universidad Carlos III de Madrid ©Physics Department 2011 Physics Laboratory

 17

nσ 2
∆m = (18)
n ∑ xi2 −  ∑ xi 
2
 

 

σ 2 ∑ xi2
∆b = (19)
n ∑ xi2 −  ∑ xi 
 2

 

These equations have been simplified by supposing that only the yi values have
uncertainties and that these uncertainties are the same for all data, that is: ∆ ( xi ) = 0 ∀xi
, and ∆ ( yi ) = σ ∀yi . In the case that the uncertainties σ yi are not equal for all the yi data,

∑
n
the σ value will be obtained as the average value of the data: σ = σ y = 1
n i =1
σy .
i

Before applying the least squares method, it is very advisable to plot the data in order
to check that the relation between variables is approximately linear. If that is not the case,
and the relation between variables is not linear, the method is it NOT applicable (see
section 8.3 ).

When making a least‐squares fitting, it is important to take into account all the
decimal digits obtained during the intermediate steps of the process to calculate m, b,
∆m and ∆b . Only the final values should be rounded.

EXAMPLE:

Suppose that we measure several tension and intensity pairs of values at an electrical
circuit in order to calculate the resistance of the circuit, and that the following values are
obtained:

V (V) I (A)
21 ± 1 20 ± 1
29 ± 1 30 ± 1
39 ± 1 40 ± 1
55 ± 2 50 ± 1
59 ± 2 60 ± 1

Universidad Carlos III de Madrid ©Physics Department 2011 Physics Laboratory

 18

We know that V = IR . If we compare the equation V = RI + b (being b = 0) with the

equation of a straight line: y = m x + b we deduct that R corresponds to the slope of the
straight line, which can be fitted using the least squares method.

m and b can be obtained by applying equations (16) and (17). In order to do this, we have
to calculate the following data:

y =V x=I xy x2
21 20 420 400
29 30 870 900
39 40 1560 1600
55 50 2750 2500
59 60 3540 3600
∑y i = 203 ∑xi = 200 ∑x y
i i = 9140 ∑x
2
i = 9000

That lead to values of m and b of:

m = R = 1.02 Ω, b = −0.2 V

The uncertainties of these values also have to be calculated. In order to do so, σ has
to be obtained:

1 n
σ= ∑ σ y = 1.4V
n i =1 i

After using equations (19) and (20), the following results are obtained:

∆m = 0.0443 Ω

∆b = 1.878 V

So finally, the calculated resistance is: m = R = (1.02±0.04) Ω

And the intercept is b = (0±2) V. This value for b was the one expected from the
theoretical equation V = IR .

Universidad Carlos III de Madrid ©Physics Department 2011 Physics Laboratory

 19

8.3. Linearization of equations.

Both the graphical method and least squares method can be only applied when the
relation between the two variables is linear. However, there are many physical laws
described by nonlinear relations. For example, the equation relating the position of an
object and the time in a motion with constant acceleration, provided that the initial
velocity is zero, is given by:

1 2
s (t ) = at (20)
2

being a the acceleration of the object.

In some cases such as these it is possible to linearize this equation in order to obtain a
straight line y = mx + b . There are two methods for linearizing an equation:

• Changing variables: We can perform a change of variables in order to obtain the

equation of a straight line. In our example, if we rename the variables as:

t2 → x , s → y

The resulting equation is:

y = 12 ax

By comparing with the equation of a straight line, we deduct that:

m → 12 a , b → 0

So a = 2m .

• Using logarithms: This method involves taking logarithms in the equation and
using their properties to linearize the expression. In our example, if we take
logarithms on the two sides of the equation:

log ( s ) = log ( 12 at 2 )

and apply the properties of the logarithms, we obtain:

log ( s ) = log ( 12 a) + 2 log (t )

Universidad Carlos III de Madrid ©Physics Department 2011 Physics Laboratory

 20

Our new variables are:

x → log (t ) , y → log ( s )

So we have a linear equation:

y = log ( 12 a ) + 2 x

where:

m → 2 , b → log ( 12 a)

So finally: a = 2 eb

Once the equation has been linearized, a new table with the data of the new variables
should be made. These data will be used to calculate m and b , and the plot of x vs. y
should be approach to a straight line.

In our example and in the case of having used logarithms, we would construct the
following table:

x y
log (t1 ) log ( s1 )
log (t2 ) log ( s2 )
MM MM
log(t n ) log ( sn )

9. Graphs
9.1. General requirements.
One of the most useful ways of presenting experimental results is by plotting graphs.
In order to do this, it is convenient to have in mind that:

• Experimental points should be clearly visible.

• Scales should be chosen so that data occupy the maximum space possible in the
graph. The vertical and horizontal axes can be scaled different if it is necessary and the
origin of the graph does not have to be (0, 0) .

Universidad Carlos III de Madrid ©Physics Department 2011 Physics Laboratory

 21

• Mark the axes at periodic intervals (and not for each experimental point). Use a
comfortable scale to make calculations easy.

• When representing experimental points, add the uncertainty of each of them by

drawing vertical and horizontal error bars. In order to do so, draw a cross centered on
each experimental data. The arms of the cross should have a length of ∆x (for the
horizontal error bar) and ∆y (for the vertical error bar).

• Add the units on each axis.

• If the experimental data can be fitted by a straight line, draw the line of best fit.
This straight line is obtained by applying y = m x + b , being m and b the ones obtained
from the least squares fitting.

9.2. Graph papers.

The use of millimeter paper (see figure 3a) makes it easier to draw a plot. When
marking intervals on a millimeter paper, the distance between points is proportional to
the difference between the values.

For example, if we represent 3 values being: x1 = 1 , x2 = 10 , and x3 = 100

corresponding to points P1 , P2 , and P3 , the distance between P1 and P2 will be proportional
to x2 − x1 = 9 and the distance between P2 and P3 will be proportional to x3 − x2 = 90 .

There exists another type of paper, called logarithmic paper (see figure 3b). In this
paper, the distance between points is proportional to the difference between the
logarithms of the values. In a logarithmic graph, the distance between P1 and P2 will be
proportional to Log ( x2 ) − Log ( x1 ) = 1 − 0 = 1 , and the distance between P3 and P2 will be
proportional to Log ( x3 ) − Log ( x2 ) = 2 − 1 = 1 . In this way, the logarithmic paper makes the
transformation to the logarithm scale, without having to calculate the actual logarithms.
This represents an advantage in many cases. Representing log y vs. log x in millimeter
paper is the same as representing y vs. x in logarithmic paper.

Universidad Carlos III de Madrid ©Physics Department 2011 Physics Laboratory

 22

4 100

3
10
Y axis

Y axis
2

1
1

0 0,1
0 1 2 3 4 0,1 1 10 100

X axis X axis
(a) (b)

Figure 3. (a) Graph paper. (b) Logarithmic paper.

10. Bibliography.
• “Statistics”. R. J. Barlow. John Wiley and Sons, 1989

• "Análisis de errores". C. Sánchez del Río. EUDEMA, 1989

Universidad Carlos III de Madrid ©Physics Department 2011 Physics Laboratory