CHE 357: Experimental Data Analysis (EDA)

3 Credit Hours

Lecturer: Eunice Sefakor Dogbe

(Department of Chemical Engineering)

LESSON 2: ERROR ANALYSIS

LEARNING OBJECTIVES
At the end of the lecture the student is expected to understand the
following:
▪ Accuracy, precision, random errors, systematic errors
▪ Difference between uncertainty and error
▪ Propagation of errors and uncertainty in computations

1
 Recommended Textbooks
▪ Chapter 3: Holman, J. P. (2012) Experimental Methods for Engineers
(8th Edition). McGraw Hill Inc. New York

▪ Chapters 1 – 4: Taylor, John. An Introduction to Error Analysis, 2nd.

ed. University Science Books. Sausalito, 1997

Introduction
▪ All measurements of physical quantities, however careful and scientific, are
subject to some uncertainty or “error”.
▪ However, experimental methods can be refined to reduce the errors and get
the measurements more closely to the true value (the ideal).
▪ Error analysis is the study and evaluation of uncertainties in physical
measurements. Error analysis
➢allows the scientist to estimate how large his uncertainties are, and
➢helps him reduce the uncertainties when necessary

2
 Accuracy and Precision
▪ Experimental error is the difference between a measurement and the true value or
between two measured values. Experimental error, itself, is measured by its
accuracy and precision
➢Accuracy measures how close a measured value is to the true value or accepted
value. Since a true or accepted value for a physical quantity may be unknown, it
is sometimes not possible to determine the accuracy of a measurement.
➢Precision measures how closely two or more measurements agree with each
other. Precision is sometimes referred to as repeatability or reproducibility.

Random
errors

Systematic
Errors (Bias
Errors):

Figure 2.1 Illustration of Accuracy and Precision

6

3
 Systematic and Random Errors
Systematic Errors (Bias Errors): Random Errors:
▪ They are errors that affect accuracy of measurement, ▪ Random errors are errors that affect the
which cannot be improved by repeating those precision of a measurement, which can be
measurements improved by repeating those measurements.
▪ In the absence of other types of errors, repeated ▪ In the absence of other types of errors, repeated
measurements yield results that differ from the true or measurements yield results that fluctuate above
accepted value by the same amount - one-sided errors and below the true or accepted value - two-
▪ They cannot easily be analysed by statistical analysis sided errors
▪ They can be difficult to detect, but once detected can ▪ They are easily analysed by statistical analysis
be reduced only by refining the measurement method ▪ They can be easily detected, and can be
or technique reduced by repeating the measurement or by
▪ They result from measurement method that is refining the measurement method or technique
inherently wrong such as ▪ Random errors occur due to random,
➢ faulty calibration of measuring instruments unpredictable variations in the measurement
➢ poorly maintained instruments process such as
➢ Faulty reading of instruments by the user, e.g. zero ➢Paralax error
error ➢Incorrectly using the measuring instrument
7

Difference between Error and Uncertainty

▪ Error results from systematic errors and describes the lack of accuracy. It may be reported as
𝑒𝑟𝑟𝑜𝑟
➢𝐹𝑟𝑎𝑐𝑡𝑖𝑜𝑛𝑎𝑙 𝑟𝑒𝑙𝑎𝑡𝑖𝑣𝑒 𝑒𝑟𝑟𝑜𝑟 = 𝑡𝑟𝑢𝑒 𝑣𝑎𝑙𝑢𝑒
or
𝑒𝑟𝑟𝑜𝑟
➢𝑃𝑒𝑟𝑐𝑒𝑛𝑡𝑎𝑔𝑒 𝑒𝑟𝑟𝑜𝑟 = × 100%
𝑡𝑟𝑢𝑒 𝑣𝑎𝑙𝑢𝑒

***Error may be defined as the difference between the reported value and the true value:
𝐸𝑟𝑟𝑜𝑟 = 𝑅𝑒𝑝𝑜𝑟𝑡𝑒𝑑 𝑣𝑎𝑙𝑢𝑒 − 𝑇𝑟𝑢𝑒 𝑣𝑎𝑙𝑢𝑒

▪ Uncertainty results from random errors and describes the lack of precision. It may be
expressed as:
𝑢𝑛𝑐𝑒𝑟𝑡𝑎𝑖𝑛𝑡𝑦
➢𝐹𝑟𝑎𝑐𝑡𝑖𝑜𝑛𝑎𝑙 (𝑅𝑒𝑙𝑎𝑡𝑖𝑣𝑒) 𝑢𝑛𝑐𝑒𝑟𝑡𝑎𝑖𝑛𝑡𝑦 = 𝑏𝑒𝑠𝑡 𝑣𝑎𝑙𝑢𝑒
or
𝑢𝑛𝑐𝑒𝑟𝑡𝑎𝑖𝑛𝑡𝑦
➢𝑃𝑒𝑟𝑐𝑒𝑛𝑡𝑎𝑔𝑒 𝑢𝑛𝑐𝑒𝑟𝑡𝑎𝑖𝑛𝑡𝑦 = 𝑏𝑒𝑠𝑡 𝑣𝑎𝑙𝑢𝑒
× 100%
8

4
 Estimating Experimental Uncertainty for a Single
Measurement
▪ The uncertainty of a single measurement is limited by the precision and accuracy of
the measuring instrument, along with any other factors that might affect the ability of
the experimenter to make the measurement.
▪ The least count is the smallest division that is marked on the instrument.
➢Thus, a meter stick will have a least count of 1.0 mm, a digital stop watch might
have a least count of 0.01 sec.
➢Very good measuring tools are calibrated against standards maintained by the
National Standards Authority.
▪ The instrument limit of error (ILE) is the precision to which a measuring device can
be read, and is always equal to or smaller than the least count.
➢The Instrument Limit of Error is generally taken to be the least count or some
fraction (1/2, 1/5, 1/10) of the least count, depending on whether the divisions are
very closely space, closely spaced or largely spaced.
9

Example problem
For each of the following scales (all in centimeters) determine the least count, the ILE, and
read the length of the gray rod.

10

10

5
 Solution

Least count (cm) ILE (cm) Length (cm)

(a) 1 0.2 9.6

(b) 0.5 0.1 8.5

(c) 0.2 0.1 11.90

11

11

Estimated Uncertainty by Repeated Measurements

▪ The uncertainties of some measurements are more difficult to estimate than
locating a point on a scale (e.g., timing the period of a pendulum in seconds). .
▪ The best estimate of the true value of such measurements is determined by taking
repeated measurements.
▪ The best estimate of the true value of N measurements 𝑥1, 𝑥2, ……, 𝑥𝑁 ;is the
average or the mean, expressed as
𝑥1 + 𝑥2 + ⋯ + 𝑥𝑁
𝐴𝑣𝑒𝑟𝑎𝑔𝑒 𝑚𝑒𝑎𝑛 , 𝑥ҧ =
𝑁
▪ The spread or variation of a data set can be characterized by average deviation
and standard deviation
Average deviation tells us on average (with 50% confidence) how much the
individual measurements vary from the mean.
𝑥1 − 𝑥ҧ + 𝑥2 − 𝑥ҧ + ⋯ + 𝑥𝑁 − 𝑥ҧ
𝐴𝑣𝑒𝑟𝑎𝑔𝑒 𝐷𝑒𝑣𝑖𝑎𝑡𝑖𝑜𝑛, 𝑑ҧ = 12
𝑁
12

6
 Estimated Uncertainty by Repeated Measurements
Standard deviation (most commonly used) is always slightly greater than the average
deviation, and it is associated with the normal distribution that is frequently encountered in
statistical analyses.

𝛿𝑥12 + 𝛿𝑥22 + ⋯ + 𝛿𝑥𝑁2 σ 𝛿𝑥𝑖2

𝑆𝑡𝑎𝑛𝑑𝑎𝑟𝑑 𝐷𝑒𝑣𝑖𝑎𝑡𝑖𝑜𝑛, 𝑠 = =
(𝑁 − 1) (𝑁 − 1)

where 𝛿𝑥𝑖 = 𝑥1 − 𝑥,ҧ the deviation or the uncertainty, for 𝑖 = 1,2, … , 𝑁.

▪ The uncertainty associated with the average value of N measurements is the standard
deviation of the mean, often called the standard error (SE).
𝑠
𝑆𝑡𝑎𝑛𝑑𝑎𝑟𝑑 𝑑𝑒𝑣𝑖𝑎𝑡𝑖𝑜𝑛 𝑜𝑓 𝑡ℎ𝑒 𝑚𝑒𝑎𝑛, 𝑜𝑟 𝑠𝑡𝑎𝑛𝑑𝑎𝑟𝑑 𝑒𝑟𝑟𝑜𝑟 𝑆𝐸 , 𝑠𝑥ҧ =
𝑛

The standard error is smaller than the standard deviation by a factor of 1Τ 𝑛 . This reflects
the fact that we expect the uncertainty of the average value to get smaller when we use a
larger number of measurements, N.
Note: The symbols, s and 𝑠𝑥ҧ are used interchangeably with 𝜎 𝑎𝑛𝑑𝜎𝑥ҧ , respectively 13

13

Estimated Uncertainty by Repeated Measurements

▪ Figure 2.2, which is a plot of data with a
mean of 0.5 shows the meaning of the
standard deviation (SD)
▪ As seen in Figure 2.2, the larger the
standard deviation, the more widely
spread the data is about the mean.
▪ For measurements which have only
random errors, the standard deviation
means that 68% of the measured values
are within s from the mean, 95% are
Figure 2.2: Measured values of x
within 2s from mean, and 99% are
within 3s from the mean.
14

14

7
 Estimated Uncertainty by Repeated Measurements
▪ For a reported value determined by taking the average of a set of independent
readings, the fractional uncertainty (revisited) is given by the ratio of the
uncertainty divided by the average value.
𝑢𝑛𝑐𝑒𝑟𝑡𝑎𝑖𝑛𝑡𝑦
𝐹𝑟𝑎𝑐𝑡𝑖𝑜𝑛𝑎𝑙 𝑢𝑛𝑐𝑒𝑟𝑡𝑎𝑖𝑛𝑡𝑦 =
𝑎𝑣𝑒𝑟𝑎𝑔𝑒 𝑣𝑎𝑙𝑢𝑒
𝑢𝑛𝑐𝑒𝑟𝑡𝑎𝑖𝑛𝑡𝑦
𝑃𝑒𝑟𝑐𝑒𝑛𝑡𝑎𝑔𝑒 𝑢𝑛𝑐𝑒𝑟𝑡𝑎𝑖𝑛𝑡𝑦 = × 100%
𝑎𝑣𝑒𝑟𝑎𝑔𝑒 𝑣𝑎𝑙𝑢𝑒
▪ The fractional uncertainty is also important because it is used in propagating
uncertainty in calculations using the result of a measurement.

15

15

Example problem
The width of a piece of paper is measured at a five points on the sheet using a meter
stick, being careful to keep the meter stick parallel to the edge of the paper (to avoid
a systematic error which would cause the measured value to be consistently higher
than the correct value). The values obtained are; 31.33, 31.15, 31.26, 31.02, and
31.20. Note that the last digit is only a rough estimate, since it is difficult to read a
meter stick to the nearest tenth of a millimeter (0.01 cm). Find the
a. average value
b. average deviation
c. standard deviation
d. standard deviation of the mean
e. percentage uncertainty (standard deviation) 16

16

8
 Propagation of uncertainties
▪ Most physical quantities usually cannot be measured in a single direct measurement.
▪ Instead, they can be calculated from measured values of one or more quantities.
▪ For example, the area of a rectangle, A, can be calculated from measured values of
length, l and height, h. That is 𝐴 = 𝑙ℎ.
▪ Three simple rules are developed for propagation of independent and random
uncertainties including that for
➢Sums and differences
➢Products and quotients
➢Arbitrary functions of one variable

17

17

Propagation of uncertainties: Sums and Differences

Suppose that 𝑥, . . . , 𝑧 are measured with 𝛿𝑥, . . . , 𝛿𝑧 and the measured values used to compute
𝑞 = 𝑥 + ⋯+𝑢 − 𝑤 +⋯+ 𝑧 ,

If the uncertainties in 𝑥, . . . , 𝑧 are known to be independent and random, then the uncertainty in 𝑞
is the quadratic sum (the square-root of the sum of squares) of the original uncertainties;

𝛿𝑞 = 𝛿𝑥 2 + ⋯ + 𝛿𝑢 2 + 𝛿𝑤 2 + ⋯ + 𝛿𝑧 2

Example: x = (2.0 ± 0.2) cm, y = (3.0 ± 0.6) cm, w = (4.5 ± 0.02) cm. Find q = x + y - w and its
uncertainty.

Solution: 𝑞 = 2.0 + 3.0 − 4.5 = 0.5𝑐𝑚

𝛿𝑞 = 0.2 2 + 0.6 2 + 0.02 2 = 0.6𝑐𝑚

∴ 𝑞 = 0.5 ± 0.6 𝑐𝑚 18

18

9
Propagation of uncertainties: Products and Quotients
Suppose that 𝑥, . . . , 𝑧 are measured with 𝛿𝑥, . . . , 𝛿𝑧 and Example: w = (4.52 ± 0.02) cm,
the measured values used to compute x = (2.0 ± 0.2) cm. Find z = wx
𝑥 × ⋯× 𝑢 and its uncertainty.
𝑞= ,
𝑤 ×⋯×𝑧 Solution
If the uncertainties in 𝑥, . . . , 𝑧 are known to be z = w x = (4.52) (2.0) = 9.04 cm2
independent and random, then the fractional uncertainty 𝛿𝑧 = 0.905 𝑐𝑚2
in 𝑞 is the sum in quadrature (quadratic sum) of the and
original fractional uncertainties; 𝑧 = 9.0 ± 0.9 𝑐𝑚2

2 2 2 2
𝛿𝑞 𝛿𝑥 𝛿𝑢 𝛿𝑤 𝛿𝑧
= + ⋯+ + + ⋯+
𝑞 𝑥 𝑢 𝑤 𝑧

The same rule holds for multiplication, division, or

19
combinations
19

Propagation of uncertainties: Any

Function of One Variable
If 𝑥 is measured with uncertainty 𝛿𝑥 and is used to calculate the function 𝑞(𝑥), then
the uncertainty 𝛿𝑞 is
𝑑𝑞
𝛿𝑞 = 𝛿𝑥
𝑑𝑥
This rule usually allows us to find 𝛿𝑞 quickly and easily, especially, when 𝑞(𝑥) is not
complicated, e.g., 𝑞 = 𝑥 𝑛 , 𝑞 = 𝐵𝑥 .
Example: Suppose we have measured an angle 𝜃 as 𝜃 = 20 ± 3° and we wish to
find cos𝜃. Find the best estimate of c𝑜𝑠𝜃 and its uncertainty
Solution: cos𝜃 = cos20° = 0.94
𝑑𝑐𝑜𝑠𝜃
𝛿 cos𝜃 = 𝛿𝜃 = sin𝜃 𝛿𝜃 (𝑖𝑛 𝑟𝑎𝑑)
𝑑𝜃
𝛿𝜃 = 3° = 0.05 𝑟𝑎𝑑
cos𝜃 = 0.94 ± 0.02
20
Note: The derivative of cos𝜃 is −sin𝜃 only if 𝜃 is expressed in radians.
20

10
 Propagation of uncertainties: Any Function of
One Variable – More Example problems
A. Measured quantity multiplied by an exact number
Example 1: The thickness T of 200 sheets of paper is 𝑇 = 1.3 ± 0.1 inches, what
is the thickness of a single sheet?
1
Solution: Thicknes of one sheet 𝑡 = 200 × 𝑇 = 0.0065 ± 0.0005 inches
Example 2: The radius, r of a circle is r = (3.0 ± 0.02) cm. Find the circumference
and its uncertainty.
Solution: 𝐶 = 2𝜋𝑟; 𝐶 = (18.8 ± 0.13) 𝑐𝑚
B. Uncertainty in a Power
Example 1: If you measure 𝑥 as 100 ± 6, what should you report for 𝑥, with its
uncertainty?
1
Solution: 𝑥 = 100 = 10 𝛿 𝑥 = 1Τ2 6 = 0.3
𝑥

Therefore, 𝑥 = 10 ± 0.3 21

21

Propagation of uncertainties: Step-by-step

Any calculation can be broken down into a sequence of steps, each involving just one of the
following types of operations (1) sums and differences; (2) products and quotients; and (3)
computation of a function of one variable such as n𝑥, 𝑥 𝑛 , sin𝑥, 𝑒 𝑥 , 𝑜𝑟 ln 𝑥 .
For example, if we calculate q = x(y − z sin w) from measured quantities 𝑥, 𝑦, 𝑧, 𝑎𝑛𝑑 𝑤 with
their uncertainties 𝛿𝑥, 𝛿𝑦, 𝛿𝑧, and 𝛿𝑤, we could calculate the uncertainty in 𝑞 as follows;
a. Find the uncertainty in sin 𝑤
b. Knowing the answer in (a) above, find the uncertainty in the product z sin w
c. Then, find the uncertainty in the difference y − z sin w
d. Finally, find the uncertainty in the complete product x(y − z sin w)
Class Exercise: Given the following; 𝑥 = 200 ± 2, 𝑦 = 50 ± 2, 𝑧 = 40 ± 2, where the three
uncertainties are independent and random,. Use the step-by-step propagation to find the
quantity 𝑞 = 𝑥Τ(𝑦 − 𝑧)
Note: The step-by-step approach cannot apply to a function such as 𝑞 = y − 𝑥 sin y because
the function is a difference of two terms (y and 𝑥 sin y), both of which depend on y. They
22
are
not independent.
22

11
 The law of propagation of uncertainty
The law of propagation of uncertainty is the single general formula from which all of the
rules for propagation of uncertainties discussed so far can be derived and with which any
problem in uncertainty propagation can be solved.
Suppose that 𝑥, . . . , 𝑧 are measured with 𝛿𝑥, . . . , 𝛿𝑧 and the measured values used to
compute the function 𝑞(𝑥, . . . , 𝑧). If the uncertainties in 𝑥, . . . , 𝑧 are independent and
random, then the uncertainty in 𝑞 is

2 2
𝜕𝑞 𝜕𝑞
𝛿𝑞 = 𝛿𝑥 + ⋯+ 𝛿𝑧
𝜕𝑥 𝜕𝑧

Note: Although this formula is rather cumbersome to use, it is useful theoretically.

23

23

Example Problems (The law of propagation of uncertainty)

1. (Solved in class) Measuring g with a simple pendulum, L – Length, T- oscillation
period 𝑇 = 2𝜋 (𝐿Τ𝑔)1Τ2 ⟹ 𝑔 = 4𝜋 2 𝐿Τ𝑇 2
Given 𝐿 = 92.95 ± 0.1 𝑐𝑚, 𝑇 = 1.936 ± 0.004 𝑠
Find g and its uncertainty (using the step-by-step approach and the law of propagation of
uncertainty).
(Answer: g = 979 ± 4 𝑐𝑚)

2. (Solved during tutorials) A temperature measuring device works according to resistance

in a copper wire which is correlated as 𝑅 = 𝑅𝑜 1 + 𝛼 𝑇 − 20
where
Ro is the resistance at 20℃ = 6Ω ± 0.3%
α is the temperature coefficient of the resistance = 0.004℃−1 ± 1%
T is the temperature = 30℃ ± 1℃
Compute the resistance of the circuit and its uncertainty.
24
(Answer: 𝑅 = 6.24 ± 0.03 Ω or 𝑅 = 6.24 Ω ± 0.49% )
24

12
 READING ASSIGNMENT

25

Significant Figures
The significant figures of a number are the digits The number of significant figures in the
from the first nonzero digit on the left to either reported value of a measured or calculated
quantity provides an indication of the precision
(a) the last digit (zero or nonzero) on the right if with which the quantity is known, that is, the
there is a decimal point, or more significant figures, the more precise is the
(b) the last nonzero digit of the number if there is value.
no decimal point. The number of significant figures suggests a
For example, rough estimate of the relative uncertainty as
2300 or 2.3 x 103 has two significant figures. summarised in the table below.
2300. or 2.300 x 103 has four significant figures. Number of Corresponding relative
2300.0 or 2.3000 x 103 has five significant figures. significant uncertainty is
figures between Or roughly
23,040 or 2.304 x 104 has four significant figures.
0.035 or 3.5 x 10−2 has two significant figures. 1 10 % and 100 % 50 %

0.03500 or 3.500 x 10−2 has four significant 2 1 % and 10 % 5%

figures. 3 0.1 % and 1 % 0.5 %
(Note: The number of significant figures is easily 26
shown and seen if scientific notation is used.)
26

13
 Significant Figures
▪ Generally, if you report the value of a ▪ When two or more quantities are combined by
measured quantity with three significant multiplication and/or division, the number of
figures, for instance, you indicate that the significant figures in the result should equal the
value of the third of these figures may be off lowest number of significant figures of any of the
by as much as a half-unit. Thus, if you report a multiplicands or divisors.
mass as 8.3 g (two significant figures), you ▪ If the initial result of a calculation violates this rule,
indicate that the mass lies somewhere between you must round off the result to reduce the number
8.25 and 8.35 g, whereas if you give the value of significant figures to its maximum allowed
as 8.300 g (four significant figures) you value, although if several calculations are to be
indicate that the mass lies between 8.2995 and performed in sequence it is advisable to keep extra
8.3005 g. significant figures of intermediate quantities and to
▪ Note, however, that this rule applies only to round off only the final result.
measured quantities or numbers calculated
from measured quantities. If a quantity is
known precisely-like a pure integer (2) or a
counted rather than measured quantity (16
oranges), its value implicitly contains an
infinite number of significant figures (5 cows (The raised quantities in parentheses denote the
really means 5.0000 ... cows). number of significant figures in the given numbers.)
27

27

Significant Figures
The rule for addition and subtraction
concerns the position of the last significant
figure in the sum, that is, the location of this
figure relative to the decimal point. The rule
is;
When two or more numbers are added or
subtracted, the positions of the last
significant figures of each number relative
to the decimal point should be compared.
Of these positions, the one farthest to the
left is the position of the last permissible
significant figure of the sum or difference.

Finally, a rule of thumb for rounding off numbers in which the digit to be dropped is a 5 is
always to make the last digit of the rounded-off number even:
1.35 ===> 1.4
1.25 ===> 1.2 28

28

14
Reporting of measured or calculated quantities
We should note that 𝛿𝑥 is an estimate of an uncertainty An important qualification to Rules 1 and 2
and should not be stated with too much precision. In above
high-precision work, uncertainties are sometimes stated
To reduce inaccuracies caused by rounding, any
with two significant figures. However, the following are
the basic rules;
numbers to be used in subsequent calculations
should normally retain at least one significant
1. Rule for stating uncertainties figure more than is finally justified.
Experimental uncertainties should be almost always be Notes:
rounded to one significant figure (however, if the leading
figure in the uncertainty is a 1, we use two significant 1. The uncertainty has the same dimension as
figures). For example, if we measure a quantity 𝑔 and its the measured quantity, therefore, it is clearer
−2
uncertainty is calculated as 𝛿𝑔 = 0.02385 𝑚𝑠 , this to write the units after both the answer and
answer should be rounded to 𝛿𝑔 = 0.02 𝑚𝑠 −2 the uncertainty.
2. Rule for stating Answers 2. If a measured quantity is so large or so small
that it calls for scientific notation, then it is
The last significant figure in any stated answer simpler and clearer to put the answer and
(measured or computed quantity) should usually be of uncertainty in the same form. For example,
the same order of magnitude (in same decimal position) measured charge = (1.61 ± 0.05) ×
as the uncertainty. For example, measured speed = 10−19 Coulombs, instead of 1.61 × 10−19 ±
6051.78 ± 30 𝑚Τ𝑠 should be corrected to 5 × 10−21 Coulombs.
measured speed = 6050 ± 30 𝑚Τ𝑠 29

29

Class Exercises:

30

30

15
 Thank you

31

16