Quantitative

Methods
Professor David Targett

QM-A3-engb 1/2017 (1022)

This course text is part of the learning content for this Edinburgh Business School course.
In addition to this printed course text, you should also have access to the course website in this subject,
which will provide you with more learning content, the Profiler software and past examination questions
and answers.
The content of this course text is updated from time to time, and all changes are reflected in the version
of the text that appears on the accompanying website at http://coursewebsites.ebsglobal.net/.
Most updates are minor, and examination questions will avoid any new or significantly altered material for
two years following publication of the relevant material on the website.
You can check the version of the course text via the version release number to be found on the front
page of the text, and compare this to the version number of the latest PDF version of the text on the
website.
If you are studying this course as part of a tutored programme, you should contact your Centre for
further information on any changes.
Full terms and conditions that apply to students on any of the Edinburgh Business School courses are
available on the website www.ebsglobal.net, and should have been notified to you either by Edinburgh
Business School or by the centre or regional partner through whom you purchased your course. If this is
not the case, please contact Edinburgh Business School at the address below:

Edinburgh Business School

Heriot-Watt University
Edinburgh
EH14 4AS
United Kingdom

Tel + 44 (0) 131 451 3090

Fax + 44 (0) 131 451 3002
Email [email protected]
Website www.ebsglobal.net

The courses are updated on a regular basis to take account of errors, omissions and recent
developments. If you'd like to suggest a change to this course, please contact
us: [email protected].
 Quantitative Methods
The Quantitative Methods programme is written by David Targett, Professor of Information Systems at
the School of Management, University of Bath and formerly Senior Lecturer in Decision Sciences at the
London Business School. Professor Targett has many years’ experience teaching executives to add
numeracy to their list of management skills and become balanced decision makers. His style is based on
demystifying complex techniques and demonstrating clearly their practical relevance as well as their
shortcomings. His books, including Coping with Numbers and The Economist Pocket Guide to Business
Numeracy, have stressed communication rather than technical rigour and have sold throughout the world.
He has written over fifty case studies which confirm the increasing integration of Quantitative Methods
with other management topics. The cases cover a variety of industries, illustrating the changing nature of
Quantitative Methods and the growing impact it is having on decision makers in the Information Technol-
ogy age. They also demonstrate Professor Targett’s wide practical experience in international
organisations in both public and private sectors.
One of his many articles, a study on the provision of management information, won the Pergamon Prize
in 1986.
He was part of the team that designed London Business School’s highly successful part-time MBA
Programme of which he was the Director from 1985 to 1988. During this time he extended the interna-
tional focus of the teaching by leading pioneering study groups to Hong Kong, Singapore and the United
States of America. He has taught on all major programmes at the London Business School and has
developed and run management education courses involving scores of major companies including:
British Rail
Citicorp
Marks and Spencer
Shell
First Published in Great Britain in 1990.
© David Targett 1990, 2000, 2001
The rights of Professor David Targett to be identified as Author of this Work have been asserted in
accordance with the Copyright, Designs and Patents Act 1988.
All rights reserved; students may print and download these materials for their own private study only and
not for commercial use. Except for this permitted use, no materials may be used, copied, shared, lent,
hired or resold in any way, without the prior consent of Edinburgh Business School.
Contents

PART 1 INTRODUCTION AND BACKGROUND

Module 1 Introducing Statistics: Some Simple Uses and Misuses 1/1

1.1 Introduction 1/1

1.2 Probability 1/3
1.3 Discrete Statistical Distributions 1/5
1.4 Continuous Statistical Distributions 1/8
1.5 Standard Distributions 1/11
1.6 Wrong Use of Statistics 1/16
1.7 How to Spot Statistical Errors 1/20
Learning Summary 1/22
Review Questions 1/24
Case Study 1.1: Airline Ticketing 1/26
Case Study 1.2: JP Carruthers Co. 1/26
Case Study 1.3: Newspaper Letters 1/30

Module 2 Basic Mathematics: School Mathematics Applied to Management 2/1

2.1 Introduction 2/1

2.2 Graphical Representation 2/2
2.3 Manipulation of Equations 2/8
2.4 Linear Functions 2/11
2.5 Simultaneous Equations 2/14
2.6 Exponential Functions 2/18
Review Questions 2/25
Case Study 2.1: Algebraic Formulation 2/28
Case Study 2.2: CNX Armaments Co. 2/29
Case Study 2.3: Bonzo Corporation 2/29
Case Study 2.4: Woof Dog Food 2/29

PART 2 HANDLING NUMBERS

Module 3 Data Communication 3/1

3.1 Introduction 3/1

3.2 Rules for Data Presentation 3/3
3.3 The Special Case of Accounting Data 3/12
3.4 Communicating Data through Graphs 3/16
Learning Summary 3/21

Quantitative Methods Edinburgh Business School v

Contents

Review Questions 3/22

Case Study 3.1: Local Government Performance Measures 3/24
Case Study 3.2: Multinational Company’s Income Statement 3/25
Case Study 3.3: Country GDPs 3/25
Case Study 3.4: Energy Efficiency 3/26

Module 4 Data Analysis 4/1

4.1 Introduction 4/1

4.2 Management Problems in Data Analysis 4/2
4.3 Guidelines for Data Analysis 4/6
Learning Summary 4/15
Review Questions 4/16
Case Study 4.1: Motoring Correspondent 4/17
Case Study 4.2: Geographical Accounts 4/18
Case Study 4.3: Wages Project 4/19

Module 5 Summary Measures 5/1

5.1 Introduction 5/1

5.2 Usefulness of the Measures 5/2
5.3 Measures of Location 5/5
5.4 Measures of Scatter 5/14
5.5 Other Summary Measures 5/20
5.6 Dealing with Outliers 5/21
5.7 Indices 5/22
Learning Summary 5/29
Review Questions 5/30
Case Study 5.1: Light Bulb Testing 5/33
Case Study 5.2: Smith’s Expense Account 5/34
Case Study 5.3: Monthly Employment Statistics 5/34
Case Study 5.4: Commuting Distances 5/34
Case Study 5.5: Petroleum Products 5/35

Module 6 Sampling Methods 6/1

6.1 Introduction 6/1

6.2 Applications of Sampling 6/3
6.3 The Ideas behind Sampling 6/3
6.4 Random Sampling Methods 6/4
6.5 Judgement Sampling 6/10
6.6 The Accuracy of Samples 6/12

vi Edinburgh Business School Quantitative Methods

 Contents

6.7 Typical Difficulties in Sampling 6/13

6.8 What Sample Size? 6/15
Learning Summary 6/16
Review Questions 6/18
Case Study 6.1: Business School Alumni 6/20
Case Study 6.2: Clearing Bank 6/20

PART 3 STATISTICAL METHODS

Module 7 Distributions 7/1

7.1 Introduction 7/1

7.2 Observed Distributions 7/2
7.3 Probability Concepts 7/8
7.4 Standard Distributions 7/14
7.5 Binomial Distribution 7/15
7.6 The Normal Distribution 7/19
Learning Summary 7/27
Review Questions 7/29
Case Study 7.1: Examination Grades 7/31
Case Study 7.2: Car Components 7/31
Case Study 7.3: Credit Card Accounts 7/32
Case Study 7.4: Breakfast Cereals 7/32

Module 8 Statistical Inference 8/1

8.1 Introduction 8/1

8.2 Applications of Statistical Inference 8/2
8.3 Confidence Levels 8/2
8.4 Sampling Distribution of the Mean 8/3
8.5 Estimation 8/6
8.6 Basic Significance Tests 8/9
8.7 More Significance Tests 8/18
8.8 Reservations about the Use of Significance Tests 8/24
Learning Summary 8/26
Review Questions 8/28
Case Study 8.1: Food Store 8/30
Case Study 8.2: Management Association 8/31
Case Study 8.3: Textile Company 8/31
Case Study 8.4: Titan Insurance Company 8/31

Quantitative Methods Edinburgh Business School vii

Contents

Module 9 More Distributions 9/1

9.1 Introduction 9/1

9.2 The Poisson Distribution 9/2
9.3 Degrees of Freedom 9/7
9.4 t-Distribution 9/8
9.5 Chi-Squared Distribution 9/14
9.6 F-Distribution 9/19
9.7 Other Distributions 9/22
Learning Summary 9/23
Review Questions 9/25
Case Study 9.1: Aircraft Accidents 9/28
Case Study 9.2: Police Vehicles 9/28

Module 10 Analysis of Variance 10/1

10.1 Introduction 10/1

10.2 Applications 10/2
10.3 One-Way Analysis of Variance 10/5
10.4 Two-Way Analysis of Variance 10/10
10.5 Extensions of Analysis of Variance 10/13
Learning Summary 10/14
Review Questions 10/15
Case Study 10.1: Washing Powder 10/16
Case Study 10.2: Hypermarkets 10/17

PART 4 STATISTICAL RELATIONSHIPS

Module 11 Regression and Correlation 11/1

11.1 Introduction 11/1

11.2 Applications 11/3
11.3 Mathematical Preliminaries 11/4
11.4 Simple Linear Regression 11/7
11.5 Correlation 11/9
11.6 Checking the Residuals 11/13
11.7 Regression on a Computer 11/15
11.8 Some Reservations about Regression and Correlation 11/19
Learning Summary 11/22
Review Questions 11/23
Case Study 11.1: Railway Booking Offices 11/25
Case Study 11.2: Department Store Chain 11/26

viii Edinburgh Business School Quantitative Methods

 Contents

Module 12 Advanced Regression Analysis 12/1

12.1 Introduction 12/1

12.2 Multiple Regression Analysis 12/2
12.3 Non-linear Regression Analysis 12/6
12.4 Statistical Basis of Regression and Correlation 12/12
12.5 Regression Analysis Summary 12/22
Learning Summary 12/23
Review Questions 12/26
Case Study 12.1: CD Marketing 12/28
Case Study 12.2: Scrap Metal Processing I 12/29
Case Study 12.3: Scrap Metal Processing II 12/30

PART 5 BUSINESS FORECASTING

Module 13 The Context of Forecasting 13/1

13.1 Introduction 13/1

13.2 A Review of Forecasting Techniques 13/2
13.3 Applications 13/3
13.4 Qualitative Forecasting Techniques 13/5
Learning Summary 13/16
Review Questions 13/18
Case Study 13.1: Automobile Design 13/19

Module 14 Time Series Techniques 14/1

14.1 Introduction 14/1

14.2 Where Time Series Methods Are Successful 14/2
14.3 Stationary Series 14/2
14.4 Series with a Trend 14/6
14.5 Series with Trend and Seasonality 14/8
14.6 Series with Trend, Seasonality and Cycles 14/8
14.7 Review of Time Series Techniques 14/15
Learning Summary 14/17
Review Questions 14/18
Case Study 14.1: Interior Furnishings 14/20
Case Study 14.2: Garden Machinery Manufacture 14/21
Case Study 14.3: McClune and Sons 14/21

Quantitative Methods Edinburgh Business School ix

Contents

Module 15 Managing Forecasts 15/1

15.1 Introduction 15/1

15.2 The Manager’s Role in Forecasting 15/2
15.3 Guidelines for an Organisation’s Forecasting System 15/4
15.4 Forecasting Errors 15/13
Learning Summary 15/15
Review Questions 15/17
Case Study 15.1: Interior Furnishings 15/19
Case Study 15.2: Theatre Company 15/19
Case Study 15.3: Brewery 15/19

Appendix 1 Statistical Tables A1/1

Appendix 2 Examination Formulae Sheet A2/1
Short-Cut Formula 2/1
Binomial Distribution 2/1
Estimation 2/1
Poisson Distribution 2/2
Normal Distribution 2/2
t-Distribution 2/2
Chi-Squared Distribution 2/2
F-Distribution 2/2
One-Way Analysis of Variance 2/3
Two-Way Analysis of Variance 2/3
Regression 2/3
Correlation Coefficient 2/3
Runs Test 2/3
Exponential Smoothing 2/3
Holt’s Method 2/4
Mean Square Error 2/4

Appendix 3 Practice Final Examinations A3/1

Practice Final Examination 1 3/2
Practice Final Examination 2 3/12

Appendix 4 Answers to Review Questions A4/1

Module 1 4/1
Module 2 4/6
Module 3 4/12
Module 4 4/18
Module 5 4/23

x Edinburgh Business School Quantitative Methods

 Contents

Module 6 4/31
Module 7 4/37
Module 8 4/45
Module 9 4/53
Module 10 4/59
Module 11 4/66
Module 12 4/72
Module 13 4/79
Module 14 4/85
Module 15 4/96

Index I/1

Quantitative Methods Edinburgh Business School xi

PART 1

Introduction and Background

Module 1 Introducing Statistics: Some Simple Uses and
Misuses
Module 2 Basic Mathematics: School Mathematics
Applied to Management

Quantitative Methods Edinburgh Business School

Module 1

Introducing Statistics: Some Simple

Uses and Misuses
Contents
1.1 Introduction.............................................................................................1/1
1.2 Probability ...............................................................................................1/3
1.3 Discrete Statistical Distributions ..........................................................1/5
1.4 Continuous Statistical Distributions .....................................................1/8
1.5 Standard Distributions ........................................................................ 1/11
1.6 Wrong Use of Statistics ...................................................................... 1/16
1.7 How to Spot Statistical Errors ........................................................... 1/20
Learning Summary ......................................................................................... 1/22
Review Questions ........................................................................................... 1/24
Case Study 1.1: Airline Ticketing ................................................................. 1/26
Case Study 1.2: JP Carruthers Co. ............................................................... 1/26
Case Study 1.3: Newspaper Letters ............................................................. 1/30

Prerequisite reading: None

Learning Objectives
This module gives an overview of statistics, introducing basic ideas and concepts at
a general level, before dealing with them in greater detail in later modules. The
purpose is to provide a gentle way into the subject for those without a statistical
background, in response to the cynical view that it is not possible for anyone to read
a statistical text unless they have read it before. For those with a statistical back-
ground, the module will provide a broad framework for studying the subject.

1.1 Introduction
The word statistics can refer to a collection of numbers or it can refer to the
science of studying collections of numbers. Under either definition the subject has
received far more than its share of abuse (‘lies, damned lies…’). A large part of the
reason for this may well be the failure of people to understand that statistics is like a
language. Just as verbal languages can be misused (for example, by politicians and
journalists?) so the numerical language of statistics can be misused (by politicians
and journalists?). To blame statistics for this is as sensible as blaming the English
language when election promises are not kept.

Quantitative Methods Edinburgh Business School 1/1

Module 1 / Introducing Statistics: Some Simple Uses and Misuses

One does not have to be skilled in statistics to misuse them deliberately (‘figures
can lie and liars can figure’), but misuses often remain undetected because fewer
people seem to have the knowledge and confidence to handle numbers than have
similar abilities with words. Fewer people are numerate than are literate. What is
needed to see through the misuse of statistics, however, is common sense with the
addition of only a small amount of technical knowledge.
The difficulties are compounded by the unrealistic attitudes of those who do
have statistical knowledge. For instance, when a company’s annual accounts report
that the physical stock level is £34 236 417 (or even £34 236 000), it conveys an aura
of truth because the figure is so precise. Accompanying the accountants who
estimated the figure, one may have thought that the method by which the data were
collected did not warrant such precision. For market research to say that 9 out of 10
dogs prefer Bonzo dog food is also misleading, but in a far more overt fashion. The
statement is utterly meaningless, as is seen by asking the questions: ‘Prefer it to
what?’, ‘Prefer it under what circumstances?’, ‘9 out of which 10 dogs?’
Such examples and many, many others of greater or lesser subtlety have generat-
ed a poor reputation for statistics which is frequently used as an excuse for
remaining in ignorance of it. Unfortunately, it is impossible to avoid statistics in
business. Decisions are based on information; information is often in numerical
form. To make good decisions it is necessary to organise and understand numbers.
This is what statistics is about and this is why it is important to have some
knowledge of the subject.
Statistics can be split into two parts. The first part can be called descriptive
statistics. Broadly, this element handles the problem of sorting a large amount of
collected data in ways which enable its main features to be seen immediately. It is
concerned with turning numbers into real and useful information. Included here are
simple ideas such as organising and arranging data so that their patterns can be seen,
summarising data so that they can be handled more easily and communicating data
to others. Also included is the now very important area of handling computerised
business statistics as provided by management information systems and decision
support systems.
The second part can be referred to broadly as inferential statistics. This element
tackles the problem of how the small amount of data that has been collected (called
the sample) may be analysed to infer general conclusions about the total amount of
similar data that exist uncollected in the world (called the population). For instance,
opinion polls use inferential statistics to make statements about the opinions of the
whole electorate of a country, given the results of perhaps just a few hundred
interviews.
Both types of statistics are open to misuse. However, with a little knowledge and
a great deal of common sense, the errors can be spotted and the correct procedures
seen. In this module the basic concepts of statistics will be introduced. Later, some
abuses of statistics and how to counter them will be discussed.
The first basic concept to look at is that of probability, which is fundamental to
statistical work. Statistics deals with approximations and ‘best guesses’ because of
the inaccuracy and incompleteness of most of the data used. It is rare to make

1/2 Edinburgh Business School Quantitative Methods

 Module 1 / Introducing Statistics: Some Simple Uses and Misuses

statements and draw conclusions with certainty. Probability is a way of quantifying

the strength of belief in the information derived and the conclusions drawn.

1.2 Probability
All future events are uncertain to some degree. That the present government will
still be in power in the UK in a year’s time (given that it is not an election year) is
likely, but far from certain; that a communist government will be in power in a
year’s time is highly unlikely, but not impossible. Probability theory enables the
difference in the uncertainty of events to be made more precise by measuring their
likelihood on a scale.

Impossible Evens Certain

0 0.5 1

Lifting oneself by A new-born baby There will be at least

one's own bootlaces being male one car accident in
London during the next year

Figure 1.1 Probability scale

The scale is shown in Figure 1.1. At one extreme, impossible events (e.g. that you
could swim the Atlantic) have probability zero. At the other extreme, completely
certain events (e.g. that you will one day die) have probability one. In between are
placed all the neither certain nor impossible events according to their likelihood. For
instance, the probability of obtaining a head on one spin of an unbiased coin is 0.5;
the probability of one particular ticket winning a raffle in which there are 100 tickets
is 0.01.
As a shorthand notation ‘the probability of an event A is 0.6’ is written in this
way:
(A) = 0.6

1.2.1 Measurement of Probability

There are three methods of calculating a probability. The methods are not alterna-
tives since for certain events only one particular method of measurement may be
possible. However, they do provide different conceptual ways of viewing probabil-
ity. This should become clear as the methods are described.
(a) A priori approach. In this method the probability of an event is calculated by a
process of logic. No experiment or judgement is required. Probabilities involving
coins, dice and playing cards can fall into this category. For example, the proba-
bility of a coin landing ‘heads’ can be calculated by noting that the coin has two
sides, both of which are equally likely to fall upwards (pedants, please note: as-
sume it will not come to rest on its rim). Since the coin must fall with one side
upwards, the two events must share equally the total probability of 1.0. There-
fore:

Quantitative Methods Edinburgh Business School 1/3

Module 1 / Introducing Statistics: Some Simple Uses and Misuses

(Heads) = 0.5
(Tails) = 0.5
(b) ‘Relative frequency’ approach. When the event has been or can be repeated a
large number of times, its probability can be measured from the formula:
.
(Event) =
.
For example, to estimate the probability of rain on a given day in September in
London, look at the last 10 years’ records to find that it rained on 57 days. Then:
.
(Rain) =
. ( × )

=
= 0.19
(c) Subjective approach. A certain group of statisticians (Bayesians) would argue
that the degree of belief that an individual has about a particular event may be
expressed as a probability. Bayesian statisticians argue that in certain circum-
stances a person’s subjective assessment of a probability can and should be used.
The traditional view, held by classical statisticians, is that only objective probabil-
ity assessments are permissible. Specific areas and techniques that use subjective
probabilities will be described later. At this stage it is important to know that
probabilities can be assessed subjectively but that there is discussion amongst
statisticians as to the validity of doing so. As an example of the subjective ap-
proach, let the event be the achievement of political unity in Europe by the year
2020 AD. There is no way that either of the first two approaches could be em-
ployed to calculate this probability. However, an individual can express his own
feelings on the likelihood of this event by comparing it with an event of known
probability: for example, is it more or less likely than obtaining a head on the
spin of a coin? After a long process of comparison and checking, the result
might be:
(Political unity in Europe by 2020 AD) = 0.10
The process of accurately assessing a subjective probability is a field of study in
its own right and should not be regarded as pure guesswork.
The three methods of determining probabilities have been presented here as an
introduction and the approach has not been rigorous. Once probabilities have been
calculated by whatever method, they are treated in exactly the same way.
Examples
1. What is the probability of throwing a six with one throw of a die?
With the a priori approach there are six possible outcomes: 1, 2, 3, 4, 5 or 6 show-
ing. All outcomes are equally likely. Therefore:
(throwing a 6) =
2. What is the probability of a second English Channel tunnel for road vehicles being
completed by 2025 AD?
The subjective approach is the only one possible, since logical thought alone cannot
lead to an answer and there are no past observations. My assessment is a small one,
around 0.02.
3. How would you calculate the probability of obtaining a head on one spin of a biased
coin?

1/4 Edinburgh Business School Quantitative Methods

 Module 1 / Introducing Statistics: Some Simple Uses and Misuses

The a priori approach may be possible if one had information on the aerodynamical
behaviour of the coin. A more realistic method would be to conduct several trial
spins and count the number of times a head appeared:
.
(obtaining a head) =
.
4. What is the probability of drawing an ace in one cut of a pack of playing cards?
Use the a priori method. There are 52 possible outcomes (one for each card in the
deck) and the probability of picking any one card, say the ace of diamonds, must
therefore be 1/52. There are four aces in the deck, hence:
(drawing an ace) = =

1.3 Discrete Statistical Distributions

Probability makes it possible to study another essential element of statistical work: the
statistical distribution. It can be thought of either as one of the first steps in descrip-
tive statistics or, alternatively, as a cornerstone of inferential statistics. It will first be
developed as a descriptive technique. Suppose there is a collection of data, which
initially might appear as in Figure 1.2.

53
66
41 71 40
110
83 106

72
20
99 92

75

Figure 1.2 USA sales data

The numbers are all measurements of a variable. A variable is just what the word
implies. It is some entity which can be measured and for which the measurement
varies when several observations are made. The variable might be the number of
serious crimes in each French département or the heights of all 20-year-old males in
Sweden. Figure 1.2 shows the annual sales (in thousands) of a brand of tinned
sweetcorn in different sales regions of the USA. The numbers are referred to as
observations or data points.
It is little more than a mess. A mess can take on different forms, of course. The
first sight of a particular set of data may be a pile of dusty production dockets or it
may be a file of handwritten invoices, but it is always likely to be some sort of mess.
A first attempt to sort it out might be to arrange the numbers in order as in Ta-
ble 1.1.

Table 1.1 Columns of numbers

. 52 59 66
. 54 60 66
41 55 60 .

Quantitative Methods Edinburgh Business School 1/5

Module 1 / Introducing Statistics: Some Simple Uses and Misuses

43 56 60 .
45 57 61 .
46 57 62 .
48 58 62
49 58 63
49 58 65
50 59 65

Table 1.1 is an ordered array. The numbers look neater now but it is still not
possible to get a feel for the data (the average, for example) as they stand. The next
step is to classify the data and then arrange the classes in order. Classifying means
grouping the numbers in bands (e.g. 50–54) to make them easier to handle. Each
class has a frequency, which is the number of data points that fall within that class.
This is called a frequency table and is shown in Table 1.2. This shows that seven
data points were greater than or equal to 40 but less than 50, 12 were greater than or
equal to 50 but less than 60 and so on. There were 100 data points in all.

Table 1.2 A frequency table

Class Frequency
40 ≤ x < 50 7
50 ≤ x < 60 12
60 ≤ x < 70 22
70 ≤ x < 80 27
80 ≤ x < 90 19
90 ≤ x < 100 10
100 ≤ x < 110 3
Total frequency 100
Note: ≤ means ‘less than or equal to’; < means ‘less than’.

It is now much easier to get an overall conception of what the data mean. For
example, most of the numbers are between 60 and 90 with extremes of 40 and 110.
Of course, it is likely that at some time there may be a need to perform detailed
calculations with the numbers to provide specific information, but at present the
objective is merely to get a feel for the data in the shortest possible time. Another
arrangement with greater visual impact, called a frequency histogram, will help
meet this objective.

1/6 Edinburgh Business School Quantitative Methods

 Module 1 / Introducing Statistics: Some Simple Uses and Misuses

30
27

22

Frequency
20 19

12
10
10 7
3

10 20 30 40 50 60 70 80 90 100 110 120 130

Figure 1.3 A frequency histogram

The transition from Table 1.2 to Figure 1.3 is simple and obvious, yet with the
frequency histogram one can see immediately what the data are like. The numbers
are spread symmetrically over a range from 40 to just over 100 with the majority
falling around the centre of the range.
As a descriptive device the frequency histogram works well and it is not necessary
to refine it further. If, on the other hand, there are analytical objectives, the histo-
gram of Figure 1.3 would be developed into a statistical distribution. To be strictly
accurate, all configurations dealt with are statistical distributions, but it is the most
manageable and generally accepted version that is sought.
To carry out this development, notice first the connection between frequencies
and probabilities via the ‘relative frequency’ approach to probability calculations.
The probability that any randomly selected measurement lies within a particular
class interval can be calculated as follows:
(number lies within class ) =

e.g.
(40 ≤ < 50) = = 0.07

The frequency histogram can then be turned into a probability histogram by

writing the units of the vertical axis as probabilities (as calculated above) instead of
frequencies. The shape of the histogram would remain unaltered. Once the histo-
gram is in the probability form it is usually referred to as a distribution, in this case
a discrete distribution. A variable is discrete if it is limited in the values it can take.
For example, when the data are restricted to classes (as above) the variable is
discrete. Also, when a variable is restricted to whole numbers only (an integer
variable), it is discrete.
The probability histogram makes it easier to work out the probabilities associated
with amalgams of classes. For instance, if the probabilities of two of the classes are:
(50 ≤ < 60) = 0.12
(60 ≤ < 70) = 0.22
then:

Quantitative Methods Edinburgh Business School 1/7

Module 1 / Introducing Statistics: Some Simple Uses and Misuses

(50 ≤ < 70) = 0.12 + 0.22

= 0.34
This is true whether working in probabilities or the frequencies from which they
were derived.
Examples
From the data in Figure 1.3, what are the following probabilities?
1. (80 ≤ < 100)
2. ( < 70)
3. (60 ≤ < 100)
Answers
1.
(80 ≤ < 100) = (80 ≤ < 90) + (90 ≤ < 100)
= +
= 0.19 + 0.10
= 0.29
2.
( < 70) = ( ≤ 50) + (50 ≤ < 60) + (60 ≤ < 70)
= 0.07 + 0.12 + 0.22
= 0.41
3.
(60 ≤ < 100) = 0.22 + 0.27 + 0.19 + 0.10
= 0.78

1.4 Continuous Statistical Distributions

To summarise progress so far – there is a probability histogram of a variable, from
which can be determined the probability that any one measurement of the variable
will fall within one of the classes of the histogram. Such a distribution is a discrete
distribution. It is a distribution because the variable is distributed across a range of
values; it is discrete because the values the variables take are in steps rather than
smoothly following one another.
A continuous variable is not limited in the values it can take. It can be whole
numbers, and all values in between; it does not group data in classes, but distin-
guishes between numbers such as 41.73241 and 41.73242. The distribution formed
by a continuous variable is a continuous distribution. It can be thought of as an
extension of a discrete distribution. The extension process is as follows. (The
process is to illustrate the link between discrete and continuous distributions: it is
not a procedure that would ever need to be carried out in practice.)
A discrete distribution like Figure 1.3 is reproduced in Figure 1.4(a). The column
widths are progressively reduced. In (b) the column widths have been halved; for
example, the class 50 ≤ x < 60 is divided into two classes, 50 ≤ x < 55 and
55 ≤ x < 60. In (c) the classes have been further subdivided. As the process contin-

1/8 Edinburgh Business School Quantitative Methods

 Module 1 / Introducing Statistics: Some Simple Uses and Misuses

ues, the distribution becomes smoother, until, ultimately, the continuous distribu-
tion (d) will be achieved.

Variable classes (a) Variable classes (b)

50 60
Continuous variable (d) Variable classes (c)

Figure 1.4 Discrete to continuous

There is now a difficulty concerning the measurement of probabilities. In the
discrete distribution Figure 1.4(a), the probabilities associated with different values
of the variable were equal to the column height. If column heights continue to be
equal to probabilities, the process (a) → (b) → (c) → (d) would result in flatter and
flatter distributions. Figure 1.4(d) would be completely flat since the probability
associated with the now distinct values such as 41.73241 and 41.73242 must be
infinitesimally small. The problem is overcome by measuring probabilities in a
continuous distribution by areas. For example, P(50 ≤ x < 60) is the area under the
part of the curve between 50 and 60, and shaded in Figure 1.4(d).
The argument for using areas is this. In Figure 1.4(a) the column widths are all
the same; therefore probability could be measured just as well by area as by height.
Figure 1.5 gives an example of what happens in the move from (a) to (b), when the
classes are halved. It is supposed that the original data are such that the probabilities
for the new classes can be calculated from them.

Quantitative Methods Edinburgh Business School 1/9

Module 1 / Introducing Statistics: Some Simple Uses and Misuses

50 < x < 55
50 < x < 60
55 < x < 60

P (50 < x < 55) = 0.05

P (50 < x < 60) = 0.12
P (55 < x < 60) = 0.07

Area
0.12 Area Area
0.05 0.07

50 60 50 55 60

Figure 1.5 Reducing class sizes

Using areas to measure probabilities, the column heights of the new classes are
approximately the same as those of the original. The lower probabilities for the new
classes are reflected in the halving of the column widths, rather than changes in the
heights. As the subdivision process continues, there is no tendency for the distribu-
tion to become flatter. In this way a continuous distribution can have a definite
shape which can be interpreted in the same way as the shape of a discrete distribu-
tion, but its probabilities are measured from areas. Just as the column heights of a
discrete distribution sum to 1 (because each observation certainly has some value), so
the total area of a continuous distribution is 1.
The differences between discrete and continuous distributions are summarised in
Table 1.3.

Table 1.3 Differences between discrete and continuous distributions

Discrete Continuous
Variable limited to certain values Variable not limited
Shape is usually stepped Shape is usually smooth
Probabilities are equal to column heights Probabilities are equal to areas under
the curve
Sum of column heights = 1 Total area = 1

1/10 Edinburgh Business School Quantitative Methods

 Module 1 / Introducing Statistics: Some Simple Uses and Misuses

Example

0.01 0.49 0.27 0.21 0.02

60 100 110 135

Figure 1.6 The area under each part of the curve is shown. The total
area is equal to 1.0
Using the continuous distribution in Figure 1.6, what are the probabilities that a
particular value of the variable falls within the following ranges?
1. ≤ 60
2. ≤ 100
3. 60 ≤ ≤ 110
4. ≥ 135
5. ≥ 110
Answers
1. ( ≤ 60) = 0.01
2. ( ≤ 100) = 0.01 + 0.49 = 0.5
3. (60 ≤ ≤ 110) = 0.49 + 0.27 = 0.76
4. ( ≥ 135) = 0.02
5. ( ≥ 110) = 0.21 + 0.02 = 0.23
In practice, the problems with the use of continuous distributions are, first, that
one can never collect sufficient data, sufficiently accurately measured, to
establish a continuous distribution. Second, were this possible, the accurate
measurement of areas under the curve would be difficult. Their greatest
practical use is where continuous distributions appear as standard distributions, a
topic discussed in the next section.

1.5 Standard Distributions

The distribution of US sales data shown in Figure 1.2, Table 1.1, Table 1.2 and
Figure 1.3 is an observed distribution. The data were collected, a histogram
formed and that was the distribution. A standard distribution has a theoretical,
rather than observational, base. It is a distribution that has been defined mathemati-
cally from a theoretical situation. The characteristics of the situation are expressed
mathematically and the resulting situation constructed theoretically. When an actual
situation resembling the theoretical one arises, the associated standard distribution is
applied.

Quantitative Methods Edinburgh Business School 1/11

Module 1 / Introducing Statistics: Some Simple Uses and Misuses

For example, one standard distribution, the normal, is derived from the follow-
ing theoretical situation. A variable is generated by a process which should give the
variable a constant value, but does not do so because it is subject to many small
disturbances. As a result, the variable is distributed around the central value (see
Figure 1.7). This situation (central value, many small disturbances) can be expressed
mathematically and the resulting distribution can be anticipated mathematically (i.e.
a formula describing the shape of the distribution can be found).

497 498 499 500 501 502 503

Average = 500 g

Figure 1.7 Normal distribution of weights of loaves of bread

If an actual situation appears to be like the theoretical, the normal distribution is
applied. Analysis, similar to the probability calculations with the USA sales data, can
then be carried out. Areas under parts of the curve can be found from the mathe-
matical formula or, more easily, from the normal curve tables. The normal
distribution would apply, for instance, to the lengths of machine-cut rods. The rods
should all be of the same length, but are not because of the variation introduced by
vibration, machine inaccuracies, the operator and other factors. A typical analysis
might be to calculate the percentage of production likely to be outside engineering
tolerances for the rods.
The normal distribution can be applied to many situations with similar character-
istics. Other standard distributions relate to situations with different characteristics.
Applying a standard distribution is an approximation. The actual situation is unlikely
to match exactly the theoretical one on which the mathematics were based. Howev-
er, this disadvantage is more than offset by the saving in data collection that the use
of a standard distribution brings about. Observed distributions often entail a great
deal of data collection. Not only must sufficient data be collected for the distribu-
tion to take shape, but also data must be collected individually for each and every
situation.
In summary, using an observed distribution implies that data have been collected
and histograms formed; using a standard distribution implies that the situation in
which data are being generated resembles closely a theoretical situation for which a
distribution has been constructed mathematically.

1/12 Edinburgh Business School Quantitative Methods

 Module 1 / Introducing Statistics: Some Simple Uses and Misuses

1.5.1 The Normal Distribution

The normal distribution, one of the most common, is now investigated in more
detail. Figure 1.7 gives a rough idea of what it looks like in the case of weights of
bread loaves. The principal features are that it is symmetrical and bell-shaped; it has
just one hump (i.e. it is unimodal); the hump is at the average of the variable.
However, not all normal distributions are completely the same. Otherwise, they
could not possibly represent both the weights of bread loaves (with an average value
of 500 g and a spread of less than 10 g) and heights of male adults (with an average
of 1.75 m and a spread of around 0.40 m). All normal curves share a number of
common properties such as those mentioned above but they differ in that the
populations they describe have different characteristics. Two factors, called param-
eters, capture these characteristics and are sufficient to distinguish one normal
curve from another (and conversely specify exactly a normal curve). A parameter is
defined as a measure describing some aspect of a population.
The first parameter is the average or mean of the distribution. Although the
term ‘average’ has not been formally defined yet, it is no more than the expression
in everyday use (e.g. the average of 2 and 4 is 3). Two normal distributions differing
only by this parameter have precisely the same shape, but are located at different
points along a horizontal scale.
The second parameter is the standard deviation. Its precise definition will be
given later. It measures the dispersion, or spread, of the variable. In other words,
some variables are clustered tightly about the average (such as the bread loaves).
These distributions have a low standard deviation and their shape is narrow and
high. Variables that are spread a long way from the average have a high standard
deviation and their distribution is low and flat. Figure 1.8 shows examples of
distributions with high and low standard deviations: salaries in a hospital have a
large spread ranging from those for cleaners to those for consultants; salaries for
teaching staff at a school have a much smaller spread.
A further characteristic of a normal distribution is related to the standard devia-
tion (see Figure 1.9). The data refer to the weights of bread loaves with average
weight 500 g and standard deviation 2 g.
The property of the normal distribution illustrated in Figure 1.9 is derived from
the underlying mathematics, which are beyond the scope of this introduction. In any
case, it is more important to be able to use the normal distribution than to prove its
properties mathematically. The property applies whether the distribution is flat and
wide or high and narrow, provided only that it is normal. Given such a property, it is
possible to calculate the probabilities of events. The example below demonstrates
how a standard distribution (in this case the normal) can be used in statistical
analysis.

Quantitative Methods Edinburgh Business School 1/13

Module 1 / Introducing Statistics: Some Simple Uses and Misuses

(a)

£6000 £33 000 £60 000

(b)

£10 000 £19 000 £28 000

Figure 1.8 Salaries: (a) hospital – high standard deviation; (b) school –
low standard deviation

1/14 Edinburgh Business School Quantitative Methods

 Module 1 / Introducing Statistics: Some Simple Uses and Misuses

68%

1s 1s

498 500 502

Weight (g)
68% of the distribution lies within ±1s of average
68% of bread loaves weigh between 498 g and 502 g

95%

2s 2s

496 500 504

Weight (g)
95% of the distribution lies within ±2s of average
95% of bread loaves weigh between 496 g and 504 g

99%

3s 3s

494 500 506

Weight (g)
99% of the distribution lies within ±3s of average
99% of bread loaves weigh between 494 g and 506 g

Figure 1.9 Characteristics of the standard deviation (s) in a normal

distribution

Quantitative Methods Edinburgh Business School 1/15

Module 1 / Introducing Statistics: Some Simple Uses and Misuses

Example
A machine is set to produce steel components of a given length. A sample of 1000
components is taken and their lengths measured. From the measurements the average
and standard deviation of all components produced are estimated to be 2.96 cm and
0.025 cm respectively. Within what limits would 95 per cent of all components pro-
duced by the machine be expected to lie?
Take the following steps:
1. Assume that the lengths of all components produced follow a normal distribution.
This is reasonable since this situation is typical of the circumstances in which normal
distributions arise.
2. The parameters of the distribution are the average mean = 2.96 cm and the standard
deviation = 0.025 cm. The distribution of the lengths of the components will there-
fore be as in Figure 1.10.

95%

2.91 2.96 3.01

Figure 1.10 Distribution of lengths of steel components

There is a difference between the distribution of all components produced by the
machine (the distribution of the population) and the distribution of the lengths of
components in the sample. It is the former distribution which is of interest and
which is shown in Figure 1.10. The sample has been used to estimate the parame-
ters.
3. From the properties of the normal distribution stated above, 95 per cent of the
distribution of the population (and therefore 95 per cent of all components pro-
duced) will be within two standard deviations of the average. Limits are 2.96 − (2 ×
0.025) and 2.96 + (2 × 0.025), which give 2.91 cm and 3.01 cm.
According to this estimate, 95 per cent of all production will lie between 2.91 cm
and 3.01 cm.

1.6 Wrong Use of Statistics

Statistics are misused whenever statistical evidence is presented in such a way that it
tends to lead to a false conclusion. The Advertising Standards Authority tries to
protect the public from misleading advertising, but the manager has no similar
protection against misleading management data. The presentation may mislead
accidentally or deliberately. In the latter case, the misuse of statistics can be a
creative art. Even so, it is possible to notice a few general types of misuse.

1/16 Edinburgh Business School Quantitative Methods

 Module 1 / Introducing Statistics: Some Simple Uses and Misuses

1.6.1 Definitions
Statistical expressions and the variables themselves may not have precise definitions.
The user may assume the producer of the data is working with a different definition
than is the case. By assuming a wrong definition, the user will draw a wrong
conclusion. The statistical expression ‘average’ is capable of many interpretations. A
firm of accountants advertises in its recruiting brochure that the average salary of
qualified accountants in the firm is £44 200. A prospective employee may conclude
that financially the firm is attractive to work for. A closer look shows that the
accountants in the firm and their salaries are as follows:

3 partners £86 000

8 senior accountants £40 000
9 junior accountants £34 000

The average salary could be:

( × ) ( × ) ( × )
The mean = = £44 200
The ‘middle’ value = £40 000
The most frequent value = £34 000

All the figures could legitimately be said to be the average salary. The firm has
doubtless chosen the one that best suited its purposes. Even if it were certain that
the correct statistical definition was being used, it would still be necessary to ask just
how the variable (salary) is defined. Is share of profits included in the partners’
salaries? Are bonuses included in the accountants’ salaries? Are allowances (a car,
for example) included in the accountants’ salaries? If these items are removed, the
situation might be:

3 partners £50 000

8 senior accountants £37 000
9 junior accountants £32 400

The mean salary is now £36 880. Remuneration at this firm is suddenly not quite
so attractive.

1.6.2 Graphics
Statistical pictures are intended to communicate data very rapidly. This speed means
that first impressions are important. If the first impression is wrong then it is
unlikely to be corrected.
There are many ways of representing data pictorially, but the most frequently
used is probably the graph. If the scale of a graph is concealed or not shown at all,
the wrong conclusion can be drawn. Figure 1.11 shows the sales figures for a

Quantitative Methods Edinburgh Business School 1/17

Module 1 / Introducing Statistics: Some Simple Uses and Misuses

company over the last three years. The company would appear to have been
successful.

Sales

2013 2014 2015

Figure 1.11 Sales record (no scale)

However, no scales are shown. In fact, the sales record has been:

2013 £11 250 000

2014 £11 400 000
2015 £11 650 000

A more informative graph showing the scale is given in Figure 1.12. Sales have
hardly increased at all. Allowing for inflation, they have probably decreased in real
terms.

12

10
Sales (£ million)

2013 2014 2015

Figure 1.12 Sales record (with scale)

1.6.3 Sample Bias

Most statistical data are collected as a sample (i.e. they are just a small part of the
total data available (the population)). Conclusions drawn from the sample are
generalised to the population. The generalisation can be valid only to the extent that
the sample is representative. If the sample is not representative then the wrong
conclusions will be drawn. Sample bias can occur in three ways.

1/18 Edinburgh Business School Quantitative Methods

 Module 1 / Introducing Statistics: Some Simple Uses and Misuses

First, it arises in the collection of the data. The left-wing politician who states that
80 per cent of the letters he receives are against a policy of the right-wing govern-
ment and concludes that a majority of all the electorate oppose the government on
this issue is drawing a conclusion from a biased sample.
Second, sample bias arises through the questions that elicit the data. Questions
such as ‘Do you go to church regularly?’ will provide unreliable information. There
may be a tendency for people to exaggerate their attendance since, generally, it is
regarded as a worthy thing to do. The word ‘regularly’ also causes problems. Twice a
year, at Christmas and Easter, is regular. So is twice every Sunday. It would be
difficult to draw any meaningful conclusions from the question as posed. The
question should be more explicit in defining regularity.
Third, the sample information may be biased by the interviewer. For example,
supermarket interviews about buying habits may be conducted by a young male
interviewer who questions 50 shoppers. It would not be surprising if the resultant
sample comprised a large proportion of young attractive females.
The techniques of sampling which can overcome most of these problems will be
described later in the course.

1.6.4 Omissions
The statistics that are not given can be just as important as those that are. A
television advertiser boasts that nine out of ten dogs prefer Bonzo dog food. The
viewer may conclude that 90 per cent of all dogs prefer Bonzo to any other dog
food. The conclusion might be different if it were known that:
(a) The sample size was exactly ten.
(b) The dogs had a choice of Bonzo or the cheapest dog food on the market.
(c) The sample quoted was the twelfth sample used and the first in which as many
as nine dogs preferred Bonzo.

1.6.5 Logical Errors

Statistics allows conclusions about numbers to be drawn. Usually, however, it is the
entities that lie behind the numbers that are of interest. Two of the most common
ways for logical errors to be made are as follows.
First, the numbers may not be the same as the entities. For example, employee
dissatisfaction is sometimes measured through staff turnover. It is the first that is
being studied, but the numbers measure the second. The two may not always
correspond. Financial analysts study the profit figures of companies in order to
judge the profitability of the company. Profit figures are, however, just accounting
measures and are no more than (hopefully, good) approximations to the ‘true
profitability’ of the company, which is difficult both to define and to measure.
Second, conclusions about the numbers do not necessarily imply causal effects in
the entities. For instance, there is a well-established relationship between the average
salary of clergymen and the price of rum. The two variables move together and this
can be verified statistically. However, this does not mean that clergymen support the
price of rum or vice versa. The explanation is that the variables are related via a

Quantitative Methods Edinburgh Business School 1/19

Module 1 / Introducing Statistics: Some Simple Uses and Misuses

third factor, inflation. The variables have increased together as the cost of living has
increased, but they are unlikely to be causally related. This consideration is im-
portant when decisions are based on statistical association. To take the example
further, holding down clergymen’s salaries in order to hold down the price of rum
would work if the relationship were causal, but not if it were mere association.

1.6.6 Technical Errors

Mistakes occur where there is an insufficient understanding of even basic technicali-
ties. An oft-quoted and simplistic case is that of a trade union leader stating his
concern for the lower paid by saying that he would not rest until all his members
earned more than the average salary for the union. (It may be that he was in fact
making a very subtle statement.)
Another simple mistake is in the use of percentages. It would be wrong to sup-
pose that, for example, a 20 per cent increase in productivity this year makes up for
a 20 per cent decrease last year. If the index of productivity two years ago was 100,
then a 20 per cent decrease makes it 80. The 20 per cent increase then makes it 96
(i.e. it has not been returned to its former level).

1.7 How to Spot Statistical Errors

Many types of statistical error can only be dealt with in the context of a particular
quantitative technique, but there are several general questions which can help to
uncover statistical errors and trickery. These questions should be posed whenever
statistical evidence is used.

1.7.1 Who Is Providing the Evidence?

The law provides a good analogy. In a legal case the standing of a witness is an
important consideration in evaluating evidence. One does not expect the defence
counsel to volunteer information damaging to his/her client. In statistics also it is
important to know who is providing the evidence. If the provider stands to gain
from your acceptance of their conclusion, greater care is needed.
It is inconceivable that the makers of Bonzo dog food should ever declare ‘We
have long believed that Bonzo is the finest dog food available. However, recent tests
with a random sample of 2000 dogs indicate that the product made by the Woof
Corporation…’. On the other hand, a report on dog food by an independent
consumer unit carries a greater likelihood of being reliable evidence.

1.7.2 Where Did the Data Come from?

In 2014 ‘on average British people took 2.38 baths per week, compared with 1.15
twenty years ago’ reports a survey of people’s washing habits carried out by a
government department. On the surface this appears to be straightforward evidence,
but how reliable is it?
Where did the data come from? One can assume not from personal observation.
Most probably people were asked. Since not to bath frequently would be a shameful

1/20 Edinburgh Business School Quantitative Methods

 Module 1 / Introducing Statistics: Some Simple Uses and Misuses

admission, answers may well be biased. The figure of 2.38 is likely to be higher than
the true figure. Even so, a comparison with 20 years ago can still be made, but only
provided the bias is the same now as then. It may not be. Where did the 20-year-old
data come from? Most likely from a differently structured survey of different sample
size, with different questions and in a different social environment. The comparison
with 20 years ago, therefore, is also open to suspicion.
One is also misled in this case by the accuracy of the data. The figure of 2.38
suggests a high level of accuracy, completely unwarranted by the method of data
collection. When numbers are presented to many decimal places, one should
question the relevance of the claimed degree of accuracy.

1.7.3 Does It Pass the Common-Sense Test?

Experts in any subject sometimes can become so involved with their work that they
see only the technicalities and not the main issues. Outsiders, inhibited by their lack
of technical expertise, may suppress common-sense questions to the detriment of a
project or piece of research. Anything that does not appear to make sense should be
questioned.
An academic researcher investigated the relationship between total lifetime earn-
ings and age at death, and found that the two variables were closely related. He
concluded that poverty induces early death.
One may question the fact that he is basing a causal conclusion on a statistical
association. Perhaps more importantly, an outsider may think that being alive longer
gives more time to amass earnings and therefore it is at least as valid to conclude
that the causality works in the opposite direction (i.e. an early death causes a person
to have low total lifetime earnings). The researcher was so involved in his work and
also probably had such a strong prior belief that poverty causes early death that he
did not apply the common-sense test.

1.7.4 Has One of the Six Common Errors Been Committed?

Six of the more common types of statistical errors were described in the last section.
Could one of them have been committed? Check through the six categories to see if
one of them could apply:
(a) Is there ambiguity of definition? A statistical term (especially the average)
capable of more than one interpretation may have been used.
(b) Are the pictorial representations misleading? Take a second look to see if
other conclusions could be drawn. Especially check that scales have been includ-
ed.
(c) Is there sample bias? When two samples are compared, is like being compared
with like?
(d) What is missing? Is there any additional information which should have been
included and which could change the conclusion?
(e) Is there a logical error? The numbers may not fully represent the entities they
are intended to measure; a strong associative relationship may not be causal.

Quantitative Methods Edinburgh Business School 1/21

Module 1 / Introducing Statistics: Some Simple Uses and Misuses

(f) Is there a technical error? Have statistical definitions/techniques/methods

been properly applied? Answering this question will usually require a deeper
theoretical knowledge of the subject.

Learning Summary
The purpose of this introduction has been twofold. The first aim has been to
present some statistical concepts as a basis for more detailed study of the subject.
All the concepts will be further explored. The second aim has been to encourage a
healthy scepticism and atmosphere of constructive criticism, which are necessary
when weighing statistical evidence.
The healthy scepticism can be brought to bear on applications of the concepts
introduced so far as much as elsewhere in statistics. Probability and distributions can
both be subject to misuse.
Logical errors are often made with probability. For example, suppose a ques-
tionnaire about marketing methods is sent to a selection of companies. From the
200 replies, it emerges that 48 of the respondents are not in the area of marketing. It
also emerges that 30 are at junior levels within their companies. What is the proba-
bility that any particular questionnaire was filled in by someone neither in marketing
nor at a senior level? It is tempting to suppose that:
Probability = = 39%

This is almost certainly wrong because of double counting. Some of the 48 non-
marketers are also likely to be at a junior level. If 10 respondents were non-
marketers and at a junior level, then:
Probability = = 34%

Only in the rare case where none of those at a junior level were outside the mar-
keting area would the first calculation have been correct.

2000
No. of civil servants

1500

1000

500

0–8 8–16 16–24 24–40 40–60 60+

Salary (£000s)

Figure 1.13 Civil servants’ salaries

1/22 Edinburgh Business School Quantitative Methods

 Module 1 / Introducing Statistics: Some Simple Uses and Misuses

Graphical errors can frequently be seen with distributions. Figure 1.13 shows an
observed distribution relating to the salaries of civil servants in a government
department. The figures give a wrong impression of the spread of salaries because
the class intervals are not all equal. One could be led to suppose that salaries are
higher than they are. The lower bands are of width £8000 (0–8, 8–16, 16–24). The
higher ones are of a much larger size. The distribution should be drawn with all the
intervals of equal size, as in Figure 1.14.
Statistical concepts are open to misuse and wrong interpretation just as verbal
reports are. The same vigilance should be exercised in the former as in the latter.

2000
No. of civil servants

1500

1000

500

0–8 8–16 16–24 24–32 32–40 40–48 48–56 56–64 64+

Salary (£000s)

Figure 1.14 Civil servants’ salaries (amended)

Quantitative Methods Edinburgh Business School 1/23

Module 1 / Introducing Statistics: Some Simple Uses and Misuses

Review Questions
1.1 One of the reasons probability is important in statistics is that, if data being dealt with
are in the form of a sample, any conclusions drawn cannot be 100 per cent certain. True
or false?

1.2 A randomly selected card drawn from a pack of cards was an ace. It was not returned
to the pack. What is the probability that a second card drawn will also be an ace?
A. 1/4
B. 1/13
C. 3/52
D. 1/17
E. 1/3

1.3 Which of the following statements are true?

A. The probability of an event is a number between 0 and 1.
B. Since nothing is ever certain, no event can have a probability equal to 1.
C. Classical statisticians take the view that subjective probability has no validity.
D. Bayesian statisticians take the view that only subjective probability has validity.

1.4 A coin is known to be unbiased (i.e. it is just as likely to come down ‘heads’ as ‘tails’). It
has just been tossed eight times and each time the result has been ‘heads’. On the ninth
throw, what is the probability that the result will be ‘tails’?
A. Less than 1/2
B. 1/2
C. More than 1/2
D. 1

1/24 Edinburgh Business School Quantitative Methods

 Module 1 / Introducing Statistics: Some Simple Uses and Misuses

Questions 1.5–1.7 are based on the following information:

A train station’s daily ticket sales (in £000) over the last quarter (= 13 weeks = 78
days) have been collected in histogram form as shown in Figure 1.15.

25
22

17

8
6

Less than 30–39.9 40–49.9 50–59.9 60 or

30 more

Figure 1.15 Train ticket sales

1.5 On how many days were sales not less than £50 000?
A. 17
B. 55
C. 23
D. 48

1.6 What is the probability that on any day sales are £60 000 or more?
A. 1/13
B. 23/78
C. 72/78
D. 0

1.7 What is the sales level that was exceeded on 90 per cent of all days?
A. £20 000
B. £30 000
C. £40 000
D. £50 000
E. £60 000

1.8 Which of the following statements about a normal distribution is true?

A. A normal distribution is another name for a standard distribution.
B. The normal distribution is an example of a standard distribution.
C. The normal distribution is a discrete distribution.
D. The normal distribution may or may not be symmetrical depending upon its
parameters.

Quantitative Methods Edinburgh Business School 1/25

Module 1 / Introducing Statistics: Some Simple Uses and Misuses

1.9 A normal distribution has mean 60 and standard deviation 10. What percentage of
readings will be in the range 60–70?
A. 68%
B. 50%
C. 95%
D. 34%
E. 84%

1.10 A police checkpoint recorded the speeds of motorists over a one-week period. The
speeds had a normal distribution with a mean 82 km/h and standard deviation 11 km/h.
What speed was exceeded by 97.5 per cent of motorists?
A. 49
B. 60
C. 71
D. 104

Case Study 1.1: Airline Ticketing

As a first step towards planning new facilities at one of its city centre ticket offices, an
airline has collected data on the length of time customers spend at a ticket desk (the
service time). One hundred customers were investigated and the time in minutes each
one was at an enquiry desk was measured. The data are shown below.

0.9 3.5 0.8 1.0 1.3 2.3 1.0 2.4 0.7 1.0
2.3 0.2 1.6 1.7 5.2 1.1 3.9 5.4 8.2 1.5
1.1 2.8 1.6 3.9 3.8 6.1 0.3 1.1 2.4 2.6
4.0 4.3 2.7 0.2 0.3 3.1 2.7 4.1 1.4 1.1
3.4 0.9 2.2 4.2 21.7 3.1 1.0 3.3 3.3 5.5
0.9 4.5 3.5 1.2 0.7 4.6 4.8 2.6 0.5 3.6
6.3 1.6 5.0 2.1 5.8 7.4 1.7 3.8 4.1 6.9
3.5 2.1 0.8 7.8 1.9 3.2 1.3 1.4 3.7 0.6
1.0 7.5 1.2 2.0 2.0 11.0 2.9 6.5 2.0 8.6
1.5 1.2 2.9 2.9 2.0 4.6 6.6 0.7 5.8 2.0

1 Classify the data in intervals one minute wide. Form a frequency histogram. What
service time is likely to be exceeded by only 10 per cent of customers?

Case Study 1.2: JP Carruthers Co.

The JP Carruthers Co. is a medium-sized manufacturing firm. Its sales figures are about
£220 million and its employment level has been around 1100 for the last 10 years. Most
of its sales are in the car industry. JPC’s profit last year was £14 480 000. It has always
enjoyed a reputation for reliability and have generally been regarded as being well
managed.
With few exceptions, JPC’s direct labour force, numbering about 600, is represented
by the TWU, the Transport Workers’ Union. It is the practice in this industry to

1/26 Edinburgh Business School Quantitative Methods

 Module 1 / Introducing Statistics: Some Simple Uses and Misuses

negotiate employee benefits on a company-wide basis, but to negotiate wages for each
class of work in a plant separately. For years, however, this antiquated practice has been
little more than a ritual. Supposedly, the system gives workers the opportunity to
express their views, but the fact is that the wages settlement in the first group invariably
sets the pattern for all other groups within a particular company. The Door Trim Line
at JPC was the key group in last year’s negotiations. Being first in line, the settlement in
Door Trim would set the pattern for JPC that year.
Annie Smith is forewoman for the Door Trim Line. There are many variations of
door trim and Annie’s biggest job is to see that they get produced in the right mix. The
work involved in making the trim is about the same regardless of the particular variety.
That is to say, it is a straight piecework operation and the standard price is 72p per unit
regardless of variety. The work itself, while mainly of an assembly nature, is quite
intricate and requires a degree of skill.
Last year’s negotiations started with the usual complaint from the union about piece
prices in general. There was then, however, an unexpected move. Here is the union’s
demand for the Door Trim Line according to the minutes of the meeting:

We’ll come straight to the point. A price of 72p a unit is diabolical… A fair
price is 80p.
The women average about 71 units/day. Therefore, the 8p more that we want
amounts to an average of £5.68 more per woman per day…
This is the smallest increase we’ve demanded recently and we will not accept
less than 80p.

(It was the long-standing practice in the plant to calculate output on an average daily
basis. Although each person’s output is in fact tallied daily, the bonus is paid on daily
output averaged over the week. The idea is that this gives a person a better chance to
recoup if she happens to have one or two bad days.)
The union’s strategy in this meeting was a surprise. In the past the first demand was
purposely out of line and neither side took it too seriously. This time their demand was
in the same area as the kind of offer that JPC’s management was contemplating.
At their first meeting following the session with the union, JPC’s management heard
the following points made by the accountant:
a. The union’s figure of 71 units per day per person is correct. I checked it against the
latest Production Report. It works out like this:
Average weekly output for the year to date is 7100 units; thus, average daily output
is 7100/5 =1420 units/day.
The number of women directly employed on the line is 20, so that average daily
output is 1420/20 = 71 units/day/woman.
b. The union’s request amounts to an 11.1 per cent increase: (80 − 72)/72 × 100 =
11.1.
c. Direct labour at current rates is estimated at £26 million. Assuming an 11.1 per cent
increase across the board, which, of course, is what we have to anticipate, total
annual direct labour would increase by about £2.9 million: £26 000 000 × 11.1% =
£2 886 000.
Prior to the negotiations management had thought that 7 per cent would be a rea-
sonable offer, being approximately the rate at which productivity and inflation had been
increasing in recent years. Privately they had set 10 per cent as the upper limit to their

Quantitative Methods Edinburgh Business School 1/27

Module 1 / Introducing Statistics: Some Simple Uses and Misuses

final offer. At this level they felt some scheme should be introduced as an incentive to
better productivity, although they had not thought through the details of any such
scheme.
As a result of the union’s strategy, however, JPC’s negotiating team decided not to
hesitate any longer. Working late, they put together their ‘best’ package using the 10
per cent criterion. The main points of the plan were as follows:
a. Maintain the 72p per unit standard price but provide a bonus of 50p for each unit
above a daily average of 61units/person.
b. Since the average output per day per person is 71, this implies that on average 10
bonus units per person per day would be paid.
c. The projected weekly cost then is £5612:
(71 × 0.72) + (10 × 0.50) = 56.12
56.12 × 5 × 20 = £5612
d. The current weekly cost then is £5112:
71 × 0.72 × 5 × 20 = 5112
e. This amounts to an average increase of £500 per week, slightly under the 10 per
cent upper limit:
500/5112 × 100 = 9.78%
f. The plan offers the additional advantage that the average worker gets 10 bonus units
immediately, making the plan seem attractive.
g. Since the output does not vary much from week to week, and since the greatest
improvement should come from those who are currently below average, the largest
portion of any increase should come from units at the lower cost of 72p each. Those
currently above average probably cannot improve very much. To the extent that this
occurs, of course, there is a tendency to reduce the average cost below the 79p per
unit that would result if no change at all occurs:
5612/(71 × 5 × 20) = 79.0p
At this point management had to decide whether they should play all their cards at
once or whether they should stick to the original plan of a 7 per cent offer. Two further
issues had to be considered:
a. How good were the rates?
b. Could a productivity increase as suggested by the 9.8 per cent offer plan really be
anticipated?
Annie Smith, the forewoman, was called into the meeting, and she gave the following
information:
a. A few workers could improve their own average a little, but the rates were too tight
for any significant movement in the daily outputs.
b. This didn’t mean that everyone worked at the same level, but that individually they
were all close to their own maximum capabilities.
c. A number did average fewer than 61 units per day. Of the few who could show a
sustained improvement, most would be in this fewer-than-61 category.
This settled it. JPC decided to go into the meeting with their ‘best’ offer of 9.8 per
cent. Next day the offer was made. The union asked for time to consider it and the next
meeting was set for the following afternoon.
In the morning of the following day Annie Smith reported that her Production Per-
formance Report (see Table 1.4) was missing. She did not know who had taken it but
was pretty sure it was the union steward.

1/28 Edinburgh Business School Quantitative Methods

 Module 1 / Introducing Statistics: Some Simple Uses and Misuses

Table 1.4 Production performance report

Fiscal Week: 10 Cost Centre: 172 Foreperson: Smith
Employee Pay No. Av. Daily Output this Av. Daily Output Y-T-D
Week
11 98 98
13 88 89
17 72 76
23 44 43
24 52 50
26 79 78
30 77 79
32 52 52
34 96 96
35 86 87
40 67 69
42 64 66
43 95 98
45 86 88
47 50 53
48 42 41
52 43 44
54 45 46
55 94 97
59 68 70
Avg. 71
AV. DAILY THIS WEEK – 1398
AV. DAILY YEAR-TO-DATE – 1420

The next meeting with the union lasted only a few minutes. A union official stated his
understanding of the offer and, after being assured that he had stated the details
correctly, he announced that the union approved the plan and intended to recommend
its acceptance to its membership. He also added that he expected this to serve as the
basis for settlement in the other units as usual and that the whole wage negotiations
could probably be completed in record time.
And that was that. Or was it? Some doubts remained in the minds of JPC’s negotiat-
ing team. Why had the union been so quick to agree? Why had the Production
Performance Report been stolen? While they were still puzzling over these questions,
Annie Smith phoned to say that the Production Performance Report had been returned.

1 In the hope of satisfying their curiosity, the negotiating team asked Annie to bring the
Report down to the office. Had any mistakes been made?

Quantitative Methods Edinburgh Business School 1/29

Module 1 / Introducing Statistics: Some Simple Uses and Misuses

Was JPC’s offer really 9.8 per cent? If not, what was the true offer?

Case Study 1.3: Newspaper Letters

The two attached letters appeared recently in a newspaper. In the first letter, Dr X
concludes that dentists should not give anaesthetics. In the second, Mr Y concludes that
dentists are the safest anaesthetists there are.
Danger in the Dental Chair
Sir– As a medically qualified anaesthetist responsible for a large number of
dental anaesthetics I read (17 June) with great distress and despair of the death
under an anaesthetic of Miss A.
It is a source of great concern to me that dentists are permitted to give anaes-
thetics. Any fool can give an intravenous injection, but considerable skill and
experience is needed to handle an emergency occurring in anaesthetics.
For anyone, however qualified, however competent, to give an anaesthetic with
no help whatsoever is an act of criminal folly; the BDA, BMA and all the medical
defence societies would agree with this.
I call upon everyone to boycott anaesthetics given by a dentist under any
circumstances.
Yours faithfully,
Dr X, Colchester, Essex.
A Dental Safety Record That Can’t Be Matched
Sir– Dr X’s feelings (Letters, 25 June) about the tragic death of Miss A will be
shared by many, and they do him credit; but they have also led him astray.
Miss A was not anaesthetised; she was heavily sedated with a combination and
dosage of drugs which produced a severe respiratory depression which the
practitioner was unable to reverse.
In calling for a ban upon the giving of general anaesthetics by dentists, Dr X is
on very unsafe ground. The possession of a medical degree does not of itself
confer immunity from stupidity or negligence; many other people would still be
alive if it did.
If Dr X consults the records produced by the Office of Population Censuses and
Surveys, he will find that, overall, more deaths associated with dental anaesthe-
sia occur when the anaesthetist is medically qualified than when he is a dentist.
Excluding the hospital service (where all anaesthetists are medically qualified but
where nearly 50 per cent of deaths occur), medically qualified anaesthetists give
36 per cent of the dental anaesthetics; they have 45 per cent of the associated
deaths. Not only a balance in favour of the dentist anaesthetist, but one which
shows that mischance can occur to anyone, however skilled.
Not even Dr X, I think, would claim that all the deaths which occurred with
medically qualified anaesthetists were due to misadventure, and all those which
occurred with dentists were negligence.
However, these figures should be put in their proper perspective. In general
dental practice and in the Community Dental Service, about 1.5 million anaes-
thetics are given each year. Over the last 15 years, deaths have averaged 4 a

1/30 Edinburgh Business School Quantitative Methods

 Module 1 / Introducing Statistics: Some Simple Uses and Misuses

year. It is a safety record which cannot be matched by any other form of general
anaesthesia.
Yours faithfully,
Mr Y, (President-Elect) Society for the Advancement of Anaesthesia in Dentis-
try.
1 Comment upon the evidence and reasoning (as given in the letters) that lead to these
two conclusions.

Quantitative Methods Edinburgh Business School 1/31

Module 2

Basic Mathematics: School

Mathematics Applied to Management
Contents
2.1 Introduction.............................................................................................2/1
2.2 Graphical Representation ......................................................................2/2
2.3 Manipulation of Equations .....................................................................2/8
2.4 Linear Functions .................................................................................. 2/11
2.5 Simultaneous Equations ...................................................................... 2/14
2.6 Exponential Functions ......................................................................... 2/18
Review Questions ........................................................................................... 2/25
Case Study 2.1: Algebraic Formulation ....................................................... 2/28
Case Study 2.2: CNX Armaments Co. ........................................................ 2/29
Case Study 2.3: Bonzo Corporation ............................................................. 2/29
Case Study 2.4: Woof Dog Food................................................................... 2/29

Prerequisite reading: None

Learning Objectives
This module describes some basic mathematics and associated notation. Some
management applications are described but the main purpose of the module is to lay
the mathematical foundations for later modules. It will be preferable to encounter
the shock of the mathematics at this stage rather than later when it might detract
from the management concepts under consideration. For the mathematically literate
the module will serve as a review; for those in a rush it could be omitted altogether.

2.1 Introduction
The quantitative courses people take at school, although usually entitled ‘mathemat-
ics’, probably cover several quantitative subjects, including algebra, geometry and
trigonometry. Some of these areas are useful as a background to numerical methods
in management. These include graphs, functions, simultaneous equations and
exponents. Usually, the mathematics is a precise means of expressing concepts and
techniques. A technique may not be complex, but the presence of mathematics,
especially notation, can cause difficulties and arouse fears.
Most of the mathematics met in a management course will be reviewed here.
Although their usefulness in management will be indicated, this is not the prime
purpose. Relevance is not an issue at this stage; the main objective is to deal with

Quantitative Methods Edinburgh Business School 2/1

Module 2 / Basic Mathematics: School Mathematics Applied to Management

basic mathematical ideas now so that they do not interfere with comprehension of
more directly applicable techniques at a later stage.

2.2 Graphical Representation

Graphs are the major pictorial method of representing numbers. Changes in, for
instance, sales figures or financial measures over time are immediately apparent; the
relationship between variables, say supply and demand, are quickly evident. Graphs
are used in most areas of management.
The essence of graphical representation is the location of a point on a piece of
paper by specifying its coordinates. Like many good ideas, this is very simple. As an
example, consider a town map. An entry in the index might read:

Hampton Lane: p. 52, F2

1 2 3

F2

Figure 2.1 Town map

Turning to page 52, we find that the map is divided into rectangles (see Fig-
ure 2.1). To find Hampton Lane, look along row F, and down column 2. Where row
and column meet, we have rectangle F2, somewhere within which those with acute
eyesight will be able to find the lane in question.

2/2 Edinburgh Business School Quantitative Methods

 Module 2 / Basic Mathematics: School Mathematics Applied to Management

3 units
2 (3,2)

y-axis
1 2 units

1 2 3 4
x-axis

Figure 2.2 Coordinates

More generally, any point on a graph is located by two coordinates, which are the
horizontal and vertical distances of the point from a fixed point called the origin.
Figure 2.2 illustrates this for the point (3,2). This point is three units from the origin
in a horizontal direction (the horizontal scale is usually called the x-axis); it is two
units from the origin in a vertical direction (the vertical scale is usually called the y-
axis). The first coordinate is referred to as the x-value and the second coordinate as
the y-value. By convention, this order is always the same.
Logically developing the above, we find that the following facts emerge:
(a) The origin has coordinates (0,0).
(b) Anywhere along the y-axis, the x-value is 0. Similarly, along the x-axis, the y-
value is 0.
(c) The axes can be extended on the other side of the origin. In this case, either or
both coordinates take on negative values. When the scales are extended in all
four directions, the paper is divided into four quadrants (see Figure 2.3).

Quantitative Methods Edinburgh Business School 2/3

Module 2 / Basic Mathematics: School Mathematics Applied to Management

2
x-values negative x-values positive
y-values positive y-values positive
1

–3 –2 –1 0 1 2 3

–1
x-values negative x-values positive
y-values negative y-values negative
–2

Figure 2.3 Quadrants

(d) There is no restriction to whole units; therefore any point on the paper has a
representation.
Figure 2.4 gives some examples of point representation.

y
5
(–7,4)
4

3
(3,2)
2
(612 ,1)
1
(–4,0) (4,0)

–8 –7 –6 –5 –4 –3 –2 –1 0 1 2 3 4 5 6 7 8 x
–1

–2

–3 (0,–3)
(–4,–3)
–4

–5

Figure 2.4 Representing points

Graphical representation is not limited to the location of points. Relationships
can be represented on a graph. Consider the simple example of the relationship
between direct profit and volume sold. For a given price and cost of a product, the
direct profit will vary according to the numbers sold:
Direct profit = (Price − Cost) × Volume

2/4 Edinburgh Business School Quantitative Methods

 Module 2 / Basic Mathematics: School Mathematics Applied to Management

This can be written more concisely with letters in place of the words. Suppose x
is the variable number of products sold, y is the variable direct profit, p the set price
and q the set cost per unit. Then:
=( − ) ﴾2.1﴿
The equation is given a label (Equation 2.1) so that it can be referred to later.
Note that multiplication can be shown in several different ways. For example,
Price (p) times Volume (x) can be written as:
·
( )( )

The multiplication sign (×) used in arithmetic tends not to be used with letters
because of possible confusion with the use of x as a variable.
The use of symbols to represent numbers is the dictionary definition of algebra.
It is intended not to confuse but to simplify. The symbols (as opposed to verbal
labels, e.g. ‘y’ instead of ‘Direct profit’) shorten the description of a complex
relationship; the symbols (as opposed to numbers, e.g. y instead of 2.1, 3.7, etc.)
allow the general properties of the variables to be investigated instead of particular
properties when the symbols take particular numerical values.
The relationship (Equation 2.1) above is an equation. Since price and cost are
fixed in this example, p and q are constants. Depending upon the quantities sold, x
and y may take on any of a range of values; therefore, they are variables. Once x is
known, y is automatically determined, so y is a function of x. Whenever the value
of a variable can be calculated given the values of other variables, it is said to be a
function of the other variables.
If the values of the constants are known, say p = 5, q = 3, then the equation
becomes:
=2 ﴾2.2﴿
A graph can now be made of this function. The graph is the set of all points
satisfying Equation 2.2, i.e. all the points for which Equation 2.2 is true. By looking
at some of the points, the shape of the graph can be seen:
when x = 0, y = 2 times 0 = 0
when x = 1, y = 2 times 1 = 2
when x = 2, y = 2 times 2 = 4, etc.
Therefore, points (0,0), (1,2), (2,4), etc. all lie on this function. Joining together a
sample of such points shows the shape of this graph. This has been done in
Figure 2.5, which shows the graph of the function y = 2x.

Quantitative Methods Edinburgh Business School 2/5

Module 2 / Basic Mathematics: School Mathematics Applied to Management

y = 2x
2

–2 –1 1 2 3 x

–1

Figure 2.5 Linear function

This function is called a linear function since it is a straight line (or, more math-
ematically, it is linear because y and x are not raised to powers; there are no squared,
cubed, logarithmic, etc. terms). More complicated functions can also be graphed.
Figure 2.6 and Figure 2.7 give examples of two such functions.

(–3,7) 7
6
5
4
3
(–2,2) 2
1

–3 –2 –1 1 2 3 x
–1
2
y= x –2 –2
x = –3 –2 –1 0 1 2 3
y = 7 2 –1–2–1 2 7

Figure 2.6 Squared function

2/6 Edinburgh Business School Quantitative Methods

 Module 2 / Basic Mathematics: School Mathematics Applied to Management

y
3 2
y = x + 3x – 2
x = –3 –2 –1 0 1 2 3 4
y = –2 2 0 –2 2 18 52

x
–3 –2 –1 1 2 3

–1

–2

Figure 2.7 Cubed function

Functions are put into graphical form by plotting a selection of points and then
drawing the curve that goes through them. In Figure 2.6, x = −3 might be the
starting point. The corresponding y value is 7 (x = −3 is put into y = x2 − 2 to give
the y value). The point (−3,7) therefore lies on the graph of the function. Trying
x = −2 gives y = 2. The point (−2,2) therefore lies on the graph.
More points are plotted until there are sufficient points to indicate the shape of
the graph. In this case, taking all whole number values for x between −3 and +3
gives sufficient points for the shape to become obvious.

x = −3 −2 −1 0 1 2 3
y= 7 2 −1 −2 −1 2 7

The equation y = x2 − 2 (Figure 2.6) might be used to represent the relationship

between some firm’s profit, y, and output, x (but only for x greater than 0). Profit is
at first negative; a breakeven point is reached at x = 1.4 (approximately); thereafter
profit increases rapidly as economies of scale come into play. Verifying that the
equation (y = x2 − 2) could, in certain circumstances, represent the relationship
between profit and output requires techniques not covered so far.
In Figure 2.7 the same seven x values, from −3 to +3, are sufficient to show the
shape of y = x3 + 3x2 − 2. The choice of these seven points is arbitrary. Any
selection of points that are able to demonstrate the shape can be chosen. It is
important to make sure that a sufficient range of points is included. Had the point
x = −3, y = −2 been excluded, then one could have been led to believe that the
graph was a U shape instead of a ‘Z on its side’ shape.

Quantitative Methods Edinburgh Business School 2/7

Module 2 / Basic Mathematics: School Mathematics Applied to Management

In Figure 2.5 only two points need be plotted since a straight line is defined
completely by any two points lying on it. The number of points which require to be
plotted varies with the complexity of the function.
When we are working with functions, they are usually restricted to their algebraic
form. It is neater and more economical to use them in this form. They are generally
put in graphical form only for illustrative purposes. The behaviour of complex
equations is often difficult to imagine from the mathematical form itself.

2.3 Manipulation of Equations

A simple example of an equation has already been introduced. With more compli-
cated examples it is sometimes necessary to rearrange them. The aim in doing this
might be to simplify an expression (i.e. to shorten it by collecting like terms togeth-
er). This would make it easier to handle in an analysis. Or the aim might be to solve
the equation for a particular variable, say x, when the equation is rearranged in the
form x = ⋯, where the dots represent constants or other variables. For instance, in
economics, the extent to which the sales volumes of a product vary as its price
changes is known as its elasticity. This can be expressed approximately by an
equation:
Elasticity (E) = /
( )/ ( )/

where
Q1 is the quantity sold at price P1
Q2 is the quantity sold at price P2
Suppose the product currently sells at the price P1 and the quantity sold per
month is Q1. A new price is mooted. What is likely to be the quantity sold at this
price? If the elasticity is known (or can be estimated), then the equation can be
rearranged and solved for Q2, i.e. put in the form:
Q2 = function of E,Q1,P1,P2
The likely quantity sold (Q2) at the new price (P2) can then be calculated. But first
the equation would have to be rearranged.
The four rules by which equations can be rearranged are:
(a) Addition. If the same quantity is added to both sides of an equation, the
resulting equation is equivalent to the original equation.
Examples

1 Solve x − 1 = 2 x−1 = 2
Add 1 to both sides of the equation x−1+1 = 2+1
x = 3

2 Solve for x: x − 4 = y + 1 x−4 = y+1

Add 4 to both sides of the equation x−4+4 = y+1+4
x = y+5

2/8 Edinburgh Business School Quantitative Methods

 Module 2 / Basic Mathematics: School Mathematics Applied to Management

(b) Subtraction. If the same quantity is subtracted from both sides of an equation,
the resulting equation is equivalent to the original.
Examples

1 Solve x + 4 = 14 x + 4 = 14
Subtract 4 from both sides of the equation x + 4 − 4 = 14 − 4
x = 10

2 Solve for y: y + x − 5 = 2 y+x−5 = 2

Add 5 to and subtract x from y+x−5+5−x = 2+5−x
both sides of the equation y = 7−x
NB: the addition and subtraction rules combine to give the familiar rule that any part of an
equation can be transferred to the other side of the equals sign as long as its sign is changed
(+ becomes −, − becomes +).
(c) Division. If each side of an equation is divided by the same number (but not
zero), the resulting equation is equivalent to the original.
Examples

1 Solve 8x = 72 8x = 72
Divide both sides by 8 =

x = 9

2 Solve for x: 2y − 4x + 5 = 6x − 3y −
5
Add 5 and 3y to both sides 5y − 4x + 10 = 6x
Add 4x to both sides 5y + 10 = 10x
Divide by 10 +1 = x

This illustrates that the solved variable can appear on either side of the equation.
(d) Multiplication. If both sides of an equation are multiplied by the same number,
except zero, the resulting equation is equivalent to the original.
Examples

1 Solve = 6 = 6
Multiply both sides by 3: x = 18

Quantitative Methods Edinburgh Business School 2/9

Module 2 / Basic Mathematics: School Mathematics Applied to Management

2 Solve =1 = 1
Multiply both sides by (4 − y) 2y + 3 = 4 − y
Add y to both sides 3y + 3 = 4
Subtract 3 from both sides 3y = 1
Divide both sides by 3 y =

2.3.1 Use of Brackets in Algebraic Expressions

Brackets indicate that the expression within the brackets should be treated as if it
were a single symbol. For example, 2(y − 4) means that both the y and the 4 should
be multiplied by 2:
2(y – 4) = 2 times y – 2 times 4
= 2y − 8
When multiplying two brackets, everything in one bracket is multiplied by every-
thing in the other.

For example: (x + 3)(y + 4) = x times y

+ x times 4
+ 3 times y
+ 3 times 4
= xy + 4x + 3y + 12

In carrying out multiplications, remember that a positive multiplied by a negative

is negative, and a negative multiplied by a negative is positive.

For example: (x − 3)(y − 4) = x times y

− x times − 4
− 3 times y
− 3 times − 4
= xy − 4x − 3y + 12

Example

Simplify = =

Multiply both sides by (4 − 3y) and (y − 1) (y + 3)(y − 1) = 4(4 − 3y)

Multiply out the brackets y2 − y + 3y − = 16 − 12y
3
Add 12y, subtract 16 from both sides y + 14y − 19 = 0
2

2/10 Edinburgh Business School Quantitative Methods

 Module 2 / Basic Mathematics: School Mathematics Applied to Management

2.4 Linear Functions

Linear functions are of great importance in management studies. Not only do they
describe many relationships in their own right, but also, because of their simplicity
and ease of use, more complex relationships can sometimes be approximated by the
linear equation. For instance, it might be possible for the total cost of processing
particular benefits payments to be expressed as a linear function of the number of
applications. The equation could then be used, along with others of a similar nature,
in compiling a budget for the department concerned. As another example, one of
the variables might be time. A linear function can then be used to represent growth
or decay. For instance, if sales volume were expressed as a linear function of time,
the equation would represent the increase or decrease in sales over time.
A linear function of x is one in which only constants and multiples of x appear.
There are no terms such as x2, x5, 1, etc. If y is a linear function of x, then the
equation must have the form:
= + ﴾2.3﴿
where m and c are constants.
Equation 2.3 is the general form of a linear equation with two variables; m and c
are merely labels – any other letters would be just as satisfactory. Thus m is the
coefficient of x, the name given to the constant by which a variable is multiplied. In
the benefit payments example, c is the fixed overhead, m the variable cost per
application and x the number of applications; mx + c is then the total cost (= y).
An alternative definition is that the graphical representation of a linear function
must be a straight line, for at all points a change of 1 unit in x gives rise to a change
of m units in y. The following are all straight lines:
y = 2x + 1
y = 3x
y = 4 − 2x

2.4.1 Interpretation of m and c

Figure 2.8 is the graph of y = 2x + 1. It was obtained, as previously, by plotting two
of the points. In this example m = 2 and c = 1.
The value of c is the intercept of the line (i.e. the point on the y-axis where the
line crosses it). This can be seen either from the graph or by putting x = 0 in the
equation.
The value of m is the slope of the line. Alternatively, the slope is referred to as
the gradient. In either case, what is meant is the usual idea of gradient – the ratio
between the distance moved vertically and the distance moved horizontally.

Quantitative Methods Edinburgh Business School 2/11

Module 2 / Basic Mathematics: School Mathematics Applied to Management

7 y = 2x + 1

6
A
5 (x1,y1)

4
y distance = y1 – y2
3 y1 – y2
B Slope =
(x2,y2) x1 – x2
2 x distance = x1 – x2

1
( 12,0) (0,1)

–1 0 1 2 3 4 x
–1

–2

Figure 2.8 Slope of a line

Figure 2.8 shows two points on the line, A and B, whose coordinates are (x1,y1)
and (x2,y2) respectively. In algebra, subscripts are used to denote unspecified but
fixed values of a variable (e.g. x1 and x2 are the values of x at the two unspecified
points A and B). The slope between A and B is given by:
= =
Since y1 = mx1 + c and y2 = mx2 + c (A and B are on the line):
( )
Slope AB = = =

The same reasoning applies to any two points along the line, confirming the
obvious fact that the slope (and therefore m) is constant along a straight line.
If A and B are two particular points (2,5) and (1,3) (i.e. x1 = 2, y1 = 5, x2 = 1, y2 =
3) then:
Slope AB = =2
For example, if the sales volume of a company were expressed as a linear func-
tion of time, y would be sales and x would be time (x = 1 for the first time period,
x = 2 for the second time period and so on). Then m would be the constant change
in sales volume from one time period to the next. If m = 3, then sales volume would
be increasing by 3 each time period.
A few additional facts are worthy of note.
(a) It is possible for m to be negative. If this is the case, the line leans in a backward
direction, since as x increases, y decreases. This is illustrated in Figure 2.9.
(b) It is possible for m to take the value 0. The equation of the line is then
y = constant, and the line is parallel to the x axis.

2/12 Edinburgh Business School Quantitative Methods

 Module 2 / Basic Mathematics: School Mathematics Applied to Management

(c) Similarly, the line x = constant is parallel to the y axis and the slope can be
regarded as being infinite. The last two lines are examples of constant functions
and are also shown in Figure 2.9.

3
y = –x + 3
x = –2
2

y=1
1

–3 –2 –1 0 1 2 3 4 x

Figure 2.9 Examples of linear functions

The question of how the equation of a particular straight line can be found is
now tackled. The equation is known as soon as the values for m and c are found.
This is possible if any of the following sets of information is available:
(a) The values of m and c.
(b) The value of m and the coordinates of any point on the line.
(c) The value of c and the coordinates of any point on the line.
(d) The coordinates of any two points on the line.
Intuitively, this makes sense, since given any one of the set (a)–(d) one could
draw a graph. In the case of information set (a), the equation is found trivially (e.g. if
m = 2 and c = 4, then the straight line is y = 2x + 4). In cases (b) and (c), one has an
equation with either m or c unknown. The coordinates of the known point are
substituted into the equation, which is solved for the unknown constant (see below
for examples). In case (d), the slope is found as follows:
Slope = where the points are ( , ), ( , )
Then the calculation is carried out as for case (b).
Examples
1. What is the equation of the line with slope 2 which passes through the point (3,4)?
Any straight line has the form y = mx + c. Since slope = m = 2, the equation must
be:
y = 2x + c
Since it passes through (3,4):
4=6+c
c = −2
The line is y = 2x − 2.

Quantitative Methods Edinburgh Business School 2/13

Module 2 / Basic Mathematics: School Mathematics Applied to Management

2. What is the equation of the line with intercept −3 and which passes through the
point (1,1)?
Since the intercept is −3, the line is:
y = mx − 3
Since it passes through (1,1):
1=m–3
m=4
The line is y = 4x – 3.
3. What is the equation of the line passing through the points (3,1) and (1,5)?
The slope between these two points is:
= = = −2
Therefore, the line must be y = −2x + c.
Since it passes through (3,1) (NB (1,5) could just as easily be used):
1 = −6 + c
c=7
The line is y = −2x + 7.

2.5 Simultaneous Equations

Relationships between variables can be described by functions. In particular, a linear
equation can represent the linear relationship between two variables. There are,
however, situations that are described by several equations.
For example, in microeconomics the price and production level of a product can
be written as two equations. First, a relationship between price and quantity will
show the amount consumers will demand at a given price level. Second, a further
relationship between price and quantity will show the quantity a supplier will be
willing to supply at a given price level. Economic theory says that there is an
equilibrium point at which a single price and quantity will satisfy both equations (the
demand curve and the supply curve).
This is the type of problem to be investigated. How can values for variables be
found that satisfy simultaneously more than one equation?
Suppose there are two equations in the two variables x and y as below:
3 + 2 = 18 ﴾2.4﴿
+ 4 = 16 ﴾2.5﴿
What are the values of x and y that satisfy both these equations? Do such values
exist at all? Or are there several of them? For example, can a single price and a single
quantity be found that is correct for both the supply and demand equations?

2/14 Edinburgh Business School Quantitative Methods

 Module 2 / Basic Mathematics: School Mathematics Applied to Management

2.5.1 Types of Solution

If the equations are plotted on a graph, it is easier to understand the solution. The
lines are best plotted by determining the points at which they cross the two axes.

For line (2.4): when x = 0, y = 9

when y = 0, x = 6
For line (2.5): when x = 0, y = 4
when y = 0, x = 16

The values of x and y that satisfy both equations are found from the point of
intersection of the lines (see Figure 2.10). Since this point is on both lines, the x and
y values here must satisfy both equations. From the graph these values can be read:
x = 4, y = 3. That these values do fit both equations can be checked by substituting
x = 4, y = 3 into the equations of the lines.

Line (2.4)

3 Line (2.5)
x
4 6 16

Figure 2.10 Solving simultaneous equations

In this example there is one and only one answer to the problem. There is said to
be a unique solution to the problem.
On the other hand, suppose the two equations are as below:
2x + 3y = 12
2x + 3y = 24
In this case the two equations are inconsistent. The left-hand side of both equa-
tions is the same: 2x + 3y. It is not possible for this to equal simultaneously 12 and
24. They can be plotted as in Figure 2.11. The two lines are parallel and have
therefore no point of intersection and no common solution.

Quantitative Methods Edinburgh Business School 2/15

Module 2 / Basic Mathematics: School Mathematics Applied to Management

8
2x + 3y = 24

4
2x + 3y = 12

x
6 12

Figure 2.11 Inconsistent equations

There is one other possibility that can arise. Suppose the equations are of the
form:
x + 3y = 15
4x + 12y = 60
If these lines are plotted, it is found that they are exactly the same line. The sec-
ond line is four times the first. Dividing through by four gives the same equation.
Any point lying anywhere along the two coincident lines satisfies both equations.
There is an infinite number of solutions and the equations are said to be depend-
ent.
To recap, solving two linear equations in two variables means finding a point
whose coordinates satisfy both equations. One of three outcomes is possible:
(a) There is one common point (i.e. a unique solution).
(b) There is an infinite number of solutions (i.e. the equations are dependent).
(c) There is no solution (i.e. the equations are inconsistent).

2.5.2 Algebraic Solution

In the preceding examples, the solutions, if any, to the equations were found
graphically. The solutions can also be found by mathematical means. The method is
given in Table 2.1.

2/16 Edinburgh Business School Quantitative Methods

 Module 2 / Basic Mathematics: School Mathematics Applied to Management

Table 2.1 Algebraic solution to simultaneous equations

Method Example
3x + 2y = 18
x + 4y = 16
1 Multiply one or both the equations by appropriate 1 Multiply the second equation by 3.
numbers so that the coefficients of x (or y) are the Leave the first equation. The coeffi-
same in both equations. cient of x is now 3 in both equations.
3x + 2y = 18
3x + 12y = 48
2 Add or subtract the equations so that x (or y) 2 Subtract the first from the second.
disappears, leaving one equation in y (or x).
3x + 12y = 48
3x + 2y = 18
3 The remaining equation gives the value of y (or x). 3 10y = 30
y = 3
4 Substitute this answer in either of the original 4 Put y = 3 in.
equations to provide the other half of the solution.
3x + 2y = 18
3x + 6 = 18
3x = 12
x = 4
The answer is x = 4, y = 3

Example
Solve the two simultaneous equations:

5 + 2 = 17 ﴾2.6﴿
2 −3 =3 ﴾2.7﴿
Multiply Equation 2.6 by 3 and Equation 2.7 by 2 so that the coefficients of y are the
same in both equations. (Equation 2.6 could just as well have been multiplied by 2 and
Equation 2.7 by 5 and then x eliminated.)
15x + 6y = 51
4x − 6y = 6
Add the two equations to eliminate y:
19x = 57
x=3
Substitute x = 3 in Equation 2.7 to find the y value:
6 − 3y = 3
3y = 3
y=1

Quantitative Methods Edinburgh Business School 2/17

Module 2 / Basic Mathematics: School Mathematics Applied to Management

The solution is x = 3, y = 1.

2.6 Exponential Functions

Exponential functions are important because of the way they express growth or
decay (e.g. the increase or decrease in a variable over a period of time). Usually they
represent a ‘natural’ pattern of growth or decay.
Think of this in terms of the sales of a new product. Suppose y is the level of
sales and x is time. Under a linear growth function each month shows the same
constant increase in the number (of bottles, packets, boxes, etc.) sold. Under an
exponential growth function each month shows a different sales increase, which is a
constant proportion of the sales level at the start of the month. A linear equation
represents a constant increase each month; an exponential equation represents a
constant percentage increase each month. There are many occasions when exponen-
tial growth is the more realistic representation. Note that exponential functions can
model decreases as well as increases.
An example of exponential growth is that of the growth of money in a bank
deposit account. Suppose £1000 is put into the account, for which the interest rate
is 10 per cent.
At the end of the first year, the balance will have grown to £1000 plus 10 per
cent (i.e. £1100).
At the end of the second year, the balance will have grown to £1100 plus 10 per
cent (i.e. £1210).
At the end of the third year, the balance will have grown to £1210 plus 10 per
cent (i.e. £1331).
And so on …
It is the fact that the interest is given on the total balance at year end (£1100,
£1210, etc.) and not just on the original deposit (£1000) that brings about exponen-
tial growth. The former results in exponential growth, the latter in linear growth.
Figure 2.12 is a graph showing how the balance increases. It shows a shallow but
curved rise. Had the interest rate been greater then the curve would have been
steeper.

2/18 Edinburgh Business School Quantitative Methods

 Module 2 / Basic Mathematics: School Mathematics Applied to Management

2000

Balance (£)
1500

1331

1331
1210
1100
1000
0 1 2 3 4 Time
(years)

Figure 2.12 Bank deposit account

An alternative, more mathematical way of representing the growth of the balance
is as follows:
At the end of year 1: £1000 × 1.10 = £1100
At the end of year 2: £1100 × 1.10 = £1000 × (1.10)2 = £1210
At the end of year 3: £1210 × 1.10 = £1000 × (1.10)3 = £1331
At the end of year n: £1000 × (1.10)n
Exponential or growth functions therefore hinge on expressions of the type
(1.10)5, 23, 104, etc. In business forecasting, equations representing exponential
growth are used. To understand how they work it is also necessary to consider the
algebraic form of such expressions, for example, ax. The properties of these
expressions are explained in the next section.

2.6.1 Exponents
Consider an expression of the form ax. The base is a and the exponent is x. If x is a
whole number, then the expression has an obvious meaning (e.g. a2 = a × a,
a3 = a × a × a, 34 = 81, etc.) It also has meaning for values of x that are not whole
numbers. To see what this meaning is, it is necessary to look at the rules for working
with exponents.
(a) Multiplication. The rule is:
× =
For example:
× =
× =
It can be seen that this makes good sense if one substitutes whole numbers for a,
x and y. For instance:
2 ×2 = 4×8
= 32
= 2

Quantitative Methods Edinburgh Business School 2/19

Module 2 / Basic Mathematics: School Mathematics Applied to Management

Note that the exponents can only be added if the bases are the same. For exam-
ple, a3 × b2 cannot be simplified.
(b) Division. The rule is similar to that for multiplication:
/ =
For example:
/ = = =
Again, the reasonableness of the rule is confirmed by resorting to a specific nu-
merical example.
(c) Raising to a power. The rule is:
( ) =
For example:
( ) =
Several points of detail follow from the rules:
(a) = 1 since 1 = / =
(b) = 1/ since 1/ = / = =
1 1 1 1 1 1
3 +
(c) = √ ;
2 = √ since × =
3 2 2 2 2 = 1=
This last point demonstrates that fractional or decimal exponents do have mean-
ing.
Examples
1. ( × )/
=( )/
= /
=
=
2. Evaluate: 274/3
= (27 )
= ( √27)
=3
= 81
3. Evaluate: 4−3/2
1
= /
4
1
=
(√4)
1
=
2
1
=
8
4. Evaluate: (22)3
=2
= 64

2/20 Edinburgh Business School Quantitative Methods

 Module 2 / Basic Mathematics: School Mathematics Applied to Management

2.6.2 Logarithms
In pursuing the objective of understanding exponential functions, it is also helpful
to look at logarithms. At school, logarithms are used for multiplying and dividing
large numbers, but this is not the purpose here. A logarithm is simply an exponent.
For example, if y = ax then x is said to be the logarithm of y to the base a. This is
written as logay = x.
Examples
1. 1000 = 103 and therefore the logarithm of 1000 to the base 10 is 3 (i.e.
3 = log101000). Logarithms to the base 10 are known as common logarithms.
2. 8 = 23 and therefore the logarithm of 8 to the base 2 is 3 (i.e. 3 = log28). Logarithms
to the base 2 are binary logarithms.
3. e is a constant frequently found in mathematics (just as π is). e has the value 2.718
approximately. Logarithms to the base e are called natural logarithms and are writ-
ten ln (i.e. x = lny means x = logey).
e has other properties which make it of interest in mathematics.
The rules for manipulation of logarithms follow from the rules for exponents:
(a) Addition: log + log = log
(b) Subtraction: log − log = log | / |
(c) Multiplication by a constant: log = log

2.6.3 Exponential Functions

The general form of the exponential function is:
=
In the bank account example of Figure 2.12:
y is the balance at the end of c years
k is the initial deposit
is (1 + interest rate)
So, at the end of three years, the balance is given by the equation:
Balance = 1000 × (1.10)
More generally, putting x = 0 into the equation reveals that k is the intercept.
The exponent constant c is more interesting. If it is positive, then the relationship is
one of growth; that is, as x increases, y also increases. If it is negative, then the
relationship is one of decay; that is, as x increases, y decreases. There is no easy
interpretation of c as a rate of growth, unlike the linear case where m is the constant
increase in each time period. Figure 2.13 gives examples of exponential functions of
growth and decay.
Recall the property of the exponential function that makes it applicable in certain
circumstances. If the relationship is one of growth, then, if x increases by a constant
amount, y increases by a constant percentage of itself (e.g. in Figure 2.13 three
points are (1,4), (2,8), (3,16)); in each case x increases by 1 and y increases by 100
per cent. Similarly, for decay functions, if x increases by a constant amount, then y
decreases by a constant percentage of itself. Contrast this with a linear function

Quantitative Methods Edinburgh Business School 2/21

Module 2 / Basic Mathematics: School Mathematics Applied to Management

where y increases (or decreases) by the same constant amount (= m, in y = mx + c)

each time x increases by 1.

(a) y y=2×2
x

15

10

k=2

0 1 2 3 x

(b) y
–x
y=5×2

15

10

–1 0 1 2 x

Figure 2.13 (a) Growth; (b) Decay

Example
The occurrence of a particular illness has been at a constant level of 100 new cases/year
for some time. A new treatment has been developed and it is thought that the number
of new cases will now decrease at a constant percentage of 25 per annum. What is the
exponential function that describes this pattern?
A graph of the function is given in Figure 2.14. The general form of the exponential
function is:

2/22 Edinburgh Business School Quantitative Methods

 Module 2 / Basic Mathematics: School Mathematics Applied to Management

New cases
Initial level

0 1 2 3 4 Time

Figure 2.14 Decreasing incidence of illness

The base to work with can be chosen. The base 10 is selected here, but any other base
would do as well.
= 10
Initially (at the beginning of the period, when x = 0), the level of new cases is 100.
Therefore:
100 = 10 =
So = 100 · (10 )
After the first year, the new cases have dropped by 25 per cent to 75.
75 = 100(10 )
10 = 0.75
At this point, tables or a calculator have to be consulted. Logarithmic tables show the
logarithm of a number, given the number. In this case, the number is 0.75 and we need
to know its logarithm c. The tables reveal that:
= −0.125
The exponential function relating y and x is therefore:
= 100 · 10 .
Figure 2.14 is a plot of this function. In this situation, estimates were being made for the
future based on judgements of a subjective nature. The judgements were (1) an
exponential function was the correct expression; (2) the rate of decrease was 25 per
cent. The equation gives anticipated numbers of new cases at future points in time.
In other circumstances, the problem might have been the other way round. Historical
records of the numbers of cases might have been available and the task would then have
been to find the function that best described what had happened. After the data were
plotted on a graph there would be two problems. First, it would have to be judged
whether the shape of the pattern of points was most like a straight line or an exponen-
tial curve or one of the many other types of functions that could be used. This
judgement would be based on knowledge of the general shapes of these functions. The
second problem would be that, even if a straight line, say, were judged to be the most
likely shape, not all the points would fall precisely on a straight line (see Figure 2.15).
The second problem is, then, to estimate which straight line best fits the data points. A

Quantitative Methods Edinburgh Business School 2/23

Module 2 / Basic Mathematics: School Mathematics Applied to Management

statistical technique, called regression analysis, is the usual way of dealing with this
second problem. (Regression analysis is a topic covered in Module 11 and Module 12.)

Figure 2.15 Example of regression line

2.6.4 Relationship between Linear and Exponential Functions

The exponential function is:
=
Take logarithms to the base a on either side:
log = log ( )
Apply two of the logarithmic rules (see Section 2.6.2) to the right-hand side:
log = log + log ( ) (Addition)
= log + log (Multiplication by a constant)
Since logaa = 1
log = constant + ﴾2.8﴿
Therefore, the relationship between logay and x is a linear one. (Compare Equa-
tion 2.8 with the equation of a straight line.) If y and x are related by an exponential
function, then logay and x are related by a linear function. In other words, by means
of a transformation (taking logarithms of both sides) an exponential function can
be treated as a linear one. (Transformations will play an important role later in
regression analysis.)

2/24 Edinburgh Business School Quantitative Methods

 Module 2 / Basic Mathematics: School Mathematics Applied to Management

Review Questions
2.1 Which point on the graph shown below is (−1,2)?

A
2

D
1

–2 –1 0 1 2 3 4 x

C B

A. Point A
B. Point B
C. Point C
D. Point D

2.2 What is the equation of the line shown on the graph below?

0 1 2 3 x

A. = +1
B. =1−
C. =− −1
D. = −1

Quantitative Methods Edinburgh Business School 2/25

Module 2 / Basic Mathematics: School Mathematics Applied to Management

2.3 Which of the ‘curves’ shown below is most likely to have the equation y = x2 − 6x + 4?

6
C
A

4
B
2

0 2 4 6 x

A. A
B. B
C. C

2.4 Solve the following for y: 6x + 4 = 2y − 4

Which answer is correct?
A. =3
B. =
C. =3 +4
D. =3 +8

2.5 Solve for y: =

Which answer is correct?
A. = 10 − 5
B. = 10
C. + 2 − 10 = 0
D. =2

2.6 What is the equation of the line with intercept 3 and that goes through point (3,9)?
A. =3 +2
B. =6 +3
C. =4 +3
D. =2 +3

2.7 What is the equation of the line that goes through the points (−1,6) and (3,−2)?
A. =2 +8
B. =4−2
C. =− +5
D. =2 −8

2/26 Edinburgh Business School Quantitative Methods

 Module 2 / Basic Mathematics: School Mathematics Applied to Management

2.8 Solve the two simultaneous equations

4 + =5
2 − =7
Which answer is correct?
A. = −1, = 9
B. = 1, = 1
C. = −1, = −9
D. = 2, = −3

2.9 Solve the two simultaneous equations

2 +7 =3
3 − 2 = 17
Which answer is correct?
A. = , =
B. = 5, = −1
C. = −2, = 1
D. = −1, = 5

2.10 What is (16)−3/2?

A. −64
B.
C.
D. −

2.11 What is log28?

A. 0.9031
B.
C. 3
D. 256

Quantitative Methods Edinburgh Business School 2/27

Module 2 / Basic Mathematics: School Mathematics Applied to Management

2.12 What is the equation of the curve shown in the graph below?

30

20

10

1 2 x

A. = + 10
B. = 10 · 10 .
C. = 10 · 10 .

D. = 100 · 10 .

Case Study 2.1: Algebraic Formulation

1 This ‘case’ consists of a series of formulations (i.e. exercises in turning words into
algebra).
a. A newspaper boy wants to calculate his daily earnings E. Each paper costs him e
pence while he sells a paper for b pence.
Write down the equation relating his earnings to the number of papers he sells per
day, called x.
What are the constants of the equation?
b. In an accounts department, a clerk makes out a bill for a customer by first writing
the customer’s name and address and then, underneath, the details of each of the
customer’s orders. A time and motion study shows that it takes the clerk two
minutes to write the name and address and one minute to write out the details of
each of the customer’s orders.
Find an equation that gives the time it takes for the clerk to make out a bill in terms
of the number of orders that are to be included in the bill.
c. A car salesman has a basic salary of £10 000 a year. In addition, he earns a commis-
sion of £400 for every new car he sells, and £200 for every second-hand car.
What is his total annual salary, as a function of the numbers of new and used cars he
sells during the year?
d. A firm charges a flat rate of £50 per week for the hire of a car. In addition, there is a
surcharge of 9p per mile up to 1000 miles. The surcharge does not apply over 1000
miles.
For hires lasting no more than one week, express this in graphical and algebraic
form.

2/28 Edinburgh Business School Quantitative Methods

 Module 2 / Basic Mathematics: School Mathematics Applied to Management

Case Study 2.2: CNX Armaments Co.

The CNX armaments company has developed two new weapons systems. The costs
and prices associated with each system are as below (in £000):

Fixed cost Variable cost Price

System 1 100 4 5
System 2 1200 4 8

1 What is the breakeven point for each system (i.e. how many of each system need to be
sold so that revenue equals cost)?

Case Study 2.3: Bonzo Corporation

The pet food made by the Bonzo Corporation contains 30 per cent meat and 70 per
cent wheatmeal, whereas the product of the Woof Corporation contains 40 per cent
meat and 60 per cent wheatmeal.
The manager of a large dogs’ home has been told that each dog in his care requires
exactly 6 oz of meat and 10 oz of wheatmeal each day (no more, no less). Express this
in two equations concerning the amounts of the two pet foods each dog should eat.

1 How much of each food should each dog eat?

Case Study 2.4: Woof Dog Food

1 Sales of Woof dog food have increased exponentially (i.e. according to an equation of
the form y = kecx) since six years ago, when the annual sales were 10 000 cases. Last
year the sales were 40 000 cases. What is your estimate of the sales this year? (NB: in
the absence of tables or a calculator, it is necessary to know: log = 1.386 and e1.662=
5.27.)

Quantitative Methods Edinburgh Business School 2/29

PART 2

Handling Numbers
Module 3 Data Communication
Module 4 Data Analysis
Module 5 Summary Measures
Module 6 Sampling Methods

Quantitative Methods Edinburgh Business School

Module 3

Data Communication
Contents
3.1 Introduction.............................................................................................3/1
3.2 Rules for Data Presentation ..................................................................3/3
3.3 The Special Case of Accounting Data ............................................... 3/12
3.4 Communicating Data through Graphs.............................................. 3/16
Learning Summary ......................................................................................... 3/21
Review Questions ........................................................................................... 3/22
Case Study 3.1: Local Government Performance Measures ..................... 3/24
Case Study 3.2: Multinational Company’s Income Statement.................. 3/25
Case Study 3.3: Country GDPs ..................................................................... 3/25
Case Study 3.4: Energy Efficiency ................................................................. 3/26

Prerequisite reading: None

Learning Objectives
By the end of the module the reader should know how to improve data presenta-
tion. This is important both in communicating data to others and in analysing data.
The emphasis is on the visual aspects of data presentation. Special reference is made
to accounting data and graphs.

3.1 Introduction
Data communication means the transmission of information through the medium
of numbers. Its reputation is mixed. Sometimes it is thought to be done dishonestly
(‘There are lies, damned lies and statistics’); at other times it is thought to be done
confusingly so that the numbers appear incomprehensible and any real information
is obscured. Thus far, numbers and words are similar. Words can also mislead and
confuse. The difference seems to be that numbers are treated with less tolerance and
are quickly abandoned as a lost cause. More effort is made with words. One hears,
for instance, of campaigns for the plain and efficient use of words by bureaucrats,
lawmakers, etc., but not for the plain and efficient use of numbers by statisticians,
computer scientists and accountants. Furthermore, while experts spend much time
devising advanced numerical techniques, little effort is put into methods for better
data communication.
This module attempts to redress the balance by looking closely at the question of
data communication. In management, numbers are usually produced in the form of

Quantitative Methods Edinburgh Business School 3/1

Module 3 / Data Communication

tables or graphs. The role of both these modes of presentation will be discussed.
The case of accounting data will be given separate treatment.
It may seem facile to say that data should be presented in a form that is suitable
for the receiver rather than convenient for the producer. Yet computerised man-
agement data often relate more to the capabilities of the computer than to the needs
of the managers – many times accounting information seems to presuppose that all
the receivers are accountants; statistics frequently can only be understood by the
highly numerate. A producer of data should have the users at the forefront of his or
her mind, and should also not assume that the receiver has a similar technical
background to him- or herself.
In the context that the requirements of the users of data are paramount, the aim
now is to show how data might be presented better. How is ‘better’ to be defined?
A manager meets data in just a few general situations:
(a) Business reports. The data are usually the supporting evidence for conclusions
or suggestions made verbally in the text.
(b) Management information systems. Large amounts of data are available on
screen or delivered to the manager at regular intervals, usually in the form of
computer printouts.
(c) Accounting data. Primarily, for a manager, these will indicate the major
financial features of an organisation. The financial analyst will have more detailed
requirements.
(d) Self-generated data. The manager may wish to analyse his own data: sales
figures, delivery performance, invoice payments, etc.
In all these situations speed is essential. A manager is unlikely to have the time to
carry out a detailed analysis of every set of data that crosses his or her desk. The
data should be communicated in such a way that its features are immediately
obvious. Moreover, the features should be the main ones rather than the points of
detail. These requirements suggest that the criterion that distinguishes well-
presented data should be: ‘The main patterns and exceptions in the data should be
immediately evident.’
The achievement of this objective is made easier since, in all the situations above,
the manager will normally be able to anticipate the patterns in the data. In the first
case above, the pattern will have been described in the text; in the other three, it is
unlikely that the manager will be dealing with raw data in a totally new set of
circumstances and he or she will therefore have some idea of what to expect. The
methods of improving data presentation are put forward with this criterion in mind.
In looking at this subject it must be stressed that it is not just communicating
data to others that is important. Communicating data to oneself is a step in coming
to understand them. The role of data communication in analysis is perhaps the most
valuable function of the ideas proposed here. Important results in the sciences,
medicine and economics are not usually discovered through the application of a
sophisticated computerised technique. They are more likely to be discovered
because someone has noticed an interesting regularity or irregularity in a small
amount of data. Such features are more likely to be noticed when the data are well
presented. Sophistication may be introduced later when one is trying to verify the

3/2 Edinburgh Business School Quantitative Methods

 Module 3 / Data Communication

result rigorously, but this should not be confused with the original analysis. In short,
simple, often visual, methods of understanding numbers are highly important. The
role of data communication as part of the analysis of data will be explored in the
next module.

3.2 Rules for Data Presentation

There are seven rules for presentation that should improve the communication of
any information contained in the data. Not all can be applied to any one set of data.
Judgement has to be used in deciding which may be suitable. The rules are based on
psychological research into the way the human mind handles information. If the
mind naturally deals with information in a particular format then this should be
taken into account when presenting information.

3.2.1 Rule 1: Round the Numbers to Two Effective Figures

It has been shown experimentally that, when trying to understand numbers, most
people round them mentally. This brings them to a size that the mind can manipu-
late more easily. In practice, the rounding is to two figures. For instance, someone
wanting to make the calculation 34.8/18.3 would probably round the division to
35/18 to obtain the answer of just less than 2. Of course, there are many situations
– an engineer’s structural estimates, accounting audits, statistical calculations, for
example – in which rounding would be wholly inappropriate. Rounding is being
suggested here for the communication of management information. In this context,
rounding will not usually affect decisions being taken, but it will improve the
effectiveness of the information. If simple numerical patterns exist in the data, they
will be more readily seen as a result of rounding since the numbers can now be
manipulated and assimilated mentally.
Rounding to two effective figures is variable rounding. Not all numbers will be
rounded in the same way. Here are some examples of rounding to two effective
figures:

Original Rounded
(2 effective figures)
1382 1400
721 720
79.311 79
17.1 17
4.2 4.2
2.32 2.3

Quantitative Methods Edinburgh Business School 3/3

Module 3 / Data Communication

These numbers have been rounded to the first two figures. Contrast this to fixed
rounding, such as rounding always to the first decimal place. For example, if the
above numbers were rounded to the first decimal place, the result would be:

Original Rounded
(1st decimal place)
1382 1382.0
721 721.0
79.311 79.3
17.1 17.1
4.2 4.2
2.32 2.3

Rounding to the same number of decimal places may appear to be more con-
sistent, but it does not make the numbers easier to manipulate and communicate.
Rounding to two effective figures puts numbers in the form in which mental
arithmetic is naturally done. The numbers are therefore assimilated more quickly.
The situation is slightly different when a series of similar numbers, all of which
have, say, the first two figures in common, are being compared. The rounding
would then be to the first two figures that are effective in making the comparison
(i.e. the first two that differ from number to number). This is the meaning of
‘effective’ in ‘two effective figures’. For example, the column of numbers below
would be rounded as shown:

Original Rounded
1142 1140
1327 1330
1489 1490
1231 1230
1588 1590

The numbers have been rounded to the second and third figures because all the
numbers have the first figure, 1, in common. Rounding to the first two figures
would be over-rounding, making comparisons too approximate.
Many managers may be concerned that rounding leads to inaccuracy. It is true, of
course, that rounding does lose accuracy. The important questions are: ‘Would the
presence of extra digits affect the decision being taken?’ and ‘Just how accurate are
the data anyway – is the accuracy being lost spurious accuracy?’
Often, one finds that eight-figure data are being insisted upon in a situation
where the decision being taken rests only on the first two figures, and where the
method of data collection was such that only the first two figures can be relied upon
as being accurate. Monitoring financial budgets is a case in point. During a financial
year actual costs are continuously compared with planned. There is room for

3/4 Edinburgh Business School Quantitative Methods

 Module 3 / Data Communication

approximation in the comparison. If the budget were £11 500, it is enough to know
that the actual costs are £10 700 (rounded to two effective figures). No different
decision would be taken had the actual costs been specified as £10 715. The
conclusion would be the same: actual costs are about 7 per cent below budget. Even
if greater accuracy were required it may not be possible to give it. At such an early
stage actual costs will almost certainly rest in some part on estimates and be subject
to error. To quote actual costs to the nearest £1 is misleading, suggesting a level of
accuracy that has not been achieved. In this situation high levels of accuracy are
therefore neither necessary nor obtainable, yet the people involved may insist on
issuing actual cost data to the nearest £1. Where there is argument about the level of
accuracy required, A. S. C. Ehrenberg (1975) suggests in his book that the numbers
should be rounded but a note at the bottom of the table should be provided to
indicate a source from which data specified to a greater precision can be obtained.
Then wait for the rush.

3.2.2 Rule 2: Reorder the Numbers

The pattern in a set of numbers can be understood better when they are in size
order. A table of data will appear more orderly and relationships between the
numbers will often be pinpointed. For instance, in a financial table, listing the
divisions of a company in size order (by capital employed, perhaps) shows at a
glance whether turnover, profit, etc. are in the same order. Divisions that are out of
line in this respect will be highlighted.
The quantity used to do the ordering may be:
(a) one of the columns or rows of the table;
(b) the averages of the columns or rows;
(c) something external to the table, such as ordering geographical regions by
population.
For example, Table 3.1 shows financial data for the geographical divisions of a
company. In Table 3.2 the divisions have been put in size order (and the data
rounded).

Table 3.1 Divisional financial data (£000)

Division Capital employed Turnover Profit
North 1342.42 533.51 78.28
South 1873.45 728.64 96.30
East 1145.71 432.11 89.33
West 1482.56 561.94 82.18

Quantitative Methods Edinburgh Business School 3/5

Module 3 / Data Communication

Table 3.2 Data reordered and rounded (£000)

Division Capital employed Turnover Profit
South 1870 730 96
West 1480 560 82
North 1340 530 78
East 1150 430 89

It is much easier to see the main features of the data from Table 3.2. East now
stands out as a clear exception, its profit being out of line with the other divisions.
The rounding also facilitates the calculation of accounting ratios. The profit margin
is about 14 per cent for all divisions except East, where it is over 20 per cent. While
it is possible to see these features in Table 3.1, they are not so immediately apparent
as in the amended table. When managers have many such tables crossing their
desks, it is essential that attention should be attracted quickly to important infor-
mation that may require action.
Frequently, tables are ordered alphabetically. This is helpful in a long reference
table that is unfamiliar to the user, but not so helpful when management infor-
mation is involved. Indeed, it may be a hindrance. In management information it is
the overall pattern, not the individual entries, that is of interest. Alphabetical order is
more likely to obscure than highlight the pattern. In addition, managers are usually
not totally unfamiliar with the data they receive. For instance, anyone looking at
some product’s sales figures by state in the USA would probably be aware that, for a
table in population order, California would be close to the top and Alaska close to
the bottom. In other words, the loss from not using alphabetical order is small
whereas the gain in data communication is large.

3.2.3 Rule 3: Interchange Rows and Columns

Making a comparison between numbers or seeing a pattern in them is easier if the
numbers lie underneath rather than alongside one another. The reason is that it is
easier to see the difference between, say, two- and three-figure numbers and to
calculate the size of the difference by subtraction if the numbers are in a column
rather than a row. This probably originates from the way children in school are
taught to manipulate numbers. People learn to do sums in this form:

57
−23
34

In a table, then, the more important comparison should be presented down col-
umns, not along rows. Taking the ‘Capital employed’ data from Table 3.2, it could
be presented horizontally, as in Table 3.3, or vertically, as in Table 3.4. In Table 3.3
the differences between adjacent rows are 390, 140, 190 respectively. When the data
are in a column, such calculations are made much more quickly.

3/6 Edinburgh Business School Quantitative Methods

 Module 3 / Data Communication

Table 3.3 Capital employed (by row) (£000)

South West North East
Capital employed 1870 1480 1340 1150

Table 3.4 Capital employed (by column) (£000)

Division Capital employed
South 1870
West 1480
North 1340
East 1150

In many tables, however, comparisons across rows and down columns are equal-
ly important and no interchange is possible.

3.2.4 Rule 4: Use Summary Measures

Summary measures, nearly always averages, for rows and columns provide a focus
for the eye as it moves down or along the numbers. It is a basis for comparison,
making it easier to tell at a glance the degree of variability and to discern whether a
particular element falls above or below the general level of the rest of the row or
column. Table 3.5 shows the number of deliveries per day at a warehouse over a
one-month period (20 days). The average number of deliveries per day provides a
focus, indicating more readily that the numbers are scattered closely and symmetri-
cally about their average.

Table 3.5 Warehouse deliveries

Deliveries Average
8 7 5 11 10 7 8 8 6 7 10 12 5 6 7 9 9 8 8 9 8

The summary measure is usually an average since it is important that the sum-
mary should be of the same size order as the rest of the numbers. A column total,
for example, is not a good summary, being of a different order of magnitude from
the rest of the column and therefore not a suitable basis for comparison. The
summary measure can also be the basis for ordering the rows and columns (see
Section 3.2.2).

3.2.5 Rule 5: Minimise the Use of Space and Lines

There should be as little white space and as few gridlines as possible in a table of
numbers. Lots of white space may look attractive, but the resultant gaps between
the numbers reduce the ability to see patterns, simply because there is a longer
interval as the eye travels from one number to the next. The numbers should be
close together, but not, of course, so close that they become confused.
Likewise, gridlines (horizontal or vertical) in a table interfere with the eye’s
movement from one number to the next. Gridlines should be used to separate one

Quantitative Methods Edinburgh Business School 3/7

Module 3 / Data Communication

type of number from another (e.g. to separate a summary row from the rest of the
numbers). Table 3.6 shows the data of Table 3.2 but with white space and gridlines
introduced. Table 3.7 is a repeat of Table 3.6 but with an acceptable use of space
and gridlines.

Table 3.6 Data of Table 3.2 with white space and gridlines
Division Capital Turnover Profit
employed
South 1870 730 96
West 1480 560 82
North 1340 530 78
East 1150 430 89

Table 3.7 White space and gridlines removed

Division Capital employed Turnover Profit
South 1870 730 96
West 1480 560 82
North 1340 530 78
East 1150 430 89

The patterns and exceptions in these data are much more clearly evident once the
white space and gridlines have been removed. The purpose of many tables is to
compare numbers. White space and gridlines have the opposite effect. They
separate numbers and make the comparison more difficult.
3.2.6 Rule 6: Labelling Should Be Clear but Unobtrusive
Care should be taken when labelling data; otherwise the labels may confuse and
detract from the numbers. This seems an obvious point, yet in practice two labelling
faults are regularly seen in tables. First, the constructor of a table may use abbreviat-
ed or obscure labels, having been working on the project for some time and falsely
assuming that the reader has the same familiarity with the numbers and their
definitions. Second, gaps may be introduced in a column of numbers merely to
accommodate extra-long labels. Labels should be clear and not interfere with the
understanding of the numbers.
Table 3.8 is an extract from a table of historical data relating to United Kingdom
utilities prior to their privatisation in the 1980s and 1990s. The extract shows ‘Gross
income as % of net assets’ for a selection of these organisations. First, the length of
the label relating to Electricity results in gaps in the column of numbers; second,
this same label includes abbreviations, the meaning of which may not be apparent to
the uninitiated. In Table 3.9 the labels have been shortened and unclear terms
eliminated. If necessary, a footnote or appendix could provide an exact definition of
the organisations concerned.

3/8 Edinburgh Business School Quantitative Methods

 Module 3 / Data Communication

In Table 3.9 the numbers and labels are clearer. Watertight and lengthy defini-
tions of the organisations do not belong within the table. The purpose of this rule is
to assert the primary importance of tables in communicating numbers. As far as
possible, the labels should give unambiguous definitions of the numbers but should
not obscure the information contained in the numbers.

Table 3.8 Gross income as % of net assets

1960 1965.......
National Coal Board 8.3 ...................
Gas Council and Boards 7.8 ...................
Electricity (BEA, CEA, EC and 10.7 ...................
Boards)
South of Scotland Electricity 9.5 ...................
Board
North of Scotland Hydro-Electric 5.4 ...................
Board
… …

Table 3.9 Gross income as % of net assets (relabelled)

1960 1965 …
National Coal Board 8.3 … …
Gas Council/Boards 7.8 … …
Electricity (ex Scotland) 10.7 … …
S. of Scotland Electricity 9.5 … …
N. of Scotland Hydro-Electric 5.4 … …
… … … …

3.2.7 Rule 7: Use a Verbal Summary

A verbal summary can help achieve the objective of communicating information
quickly by directing attention to the main features of the data. The summary should
be short and deal with the main pattern, not the points of detail or minor eccentrici-
ties in the data. It should not, of course, mislead.
In a management report, the verbal summary will probably be there already since
the data are likely to be present to provide evidence of a conclusion drawn and
described in the text. In other circumstances where a manager meets data, a verbal
summary should be produced, if possible, along with the data. A verbal summary
for Table 3.7 might be: ‘For all divisions turnover is just less than 40 per cent of
capital employed; profit is approximately 14 per cent of turnover except in East,
where the figure is over 20 per cent.’

Quantitative Methods Edinburgh Business School 3/9

Module 3 / Data Communication

Example: Facts and Figures

Table 3.10 GDP of nine EC countries plus Japan and the USA (€ thou-
sand million)
1965 1975 1987 1990 1995 1997
United 99.3 172.5 598.8 766.4 846.3 1133.3
Kingdom
Belgium 16.6 46.2 123.5 154.4 209.0 213.7
Denmark 10.1 26.9 88.8 99.6 129.4 140.2
France 96.8 253.3 770.2 940.8 1169.1 1224.4
Germany* 114.3 319.9 960.9 1182.2 1846.4 1853.9
Ireland 2.7 5.9 27.2 35.9 49.4 65.1
Italy 53.4 130.2 657.4 861.2 832.0 1011.1
Luxem- 0.7 1.7 6.0 8.1 13.2 13.9
bourg
Nether- 18.7 61.2 188.9 222.3 301.9 315.6
lands
Japan 83.0 372.8 2099.4 2341.5 3917.9 3712.1
USA 690.0 1149.7 3922.3 4361.5 5374.3 6848.2
* Germany includes the former GDR in 1995 and 1997.

Table 3.10 is taken from a UK company’s management report. It is a table of

historical economic data intended as background information for managers.
Knowledge of economic changes during earlier years is an important guide to
understanding current economic performance. The table shows the gross domestic
product of the nine countries of the European Community for various years
between 1965 and 1997, with the USA and Japan included for comparison. Since it
is intended as background information only, an economist doing a detailed study of
any or all of the countries would refer to reference sources. Typically, a reader may
wish to ask questions such as:
(a) How does the economic growth of Germany compare with that of the UK?
(b) How do the economic sizes of the major EC countries compare with one
another?
(c) Is Japan catching up with the USA? Just how small are the smaller EC countries?
The table is presented neatly, but typical questions such as these take a surpris-
ingly long time to answer.
A few changes to the layout along the lines of the seven rules make the infor-
mation much easier to grasp. After applying the seven rules, the amended data are
shown in Table 3.11.

3/10 Edinburgh Business School Quantitative Methods

 Module 3 / Data Communication

Table 3.11 The GDP of 11 countries (€ thousand million) (amended)

1997 1995 1990 1987 1975 1965
USA 6800 5400 4400 3900 1100 690
Japan 3700 3900 2300 2100 370 83
Germany* 1900 1800 1200 960 320 110
France 1200 1200 940 770 250 97
United Kingdom 1100 850 770 600 170 100
Italy 1000 830 860 660 130 53
Netherlands 320 300 220 190 61 19
Belgium 210 210 150 120 46 17
Denmark 140 130 100 89 27 10
Ireland 65 49 36 27 6 3
Luxembourg 14 13 8 6 2 1
* Germany includes the former GDR in 1995 and 1997.

The rules were applied as follows:

(a) Rule 1. Rounding to two effective figures is straightforward except for the two
smallest countries, Ireland and Luxembourg. For these two countries, the data
have been over-rounded. Two effective figures should mean that the decimal
place is retained. To do so, however, would upset the balance of the table and
attract excessive attention to these countries which would be the only ones to
have decimal places. On the other hand, loss of the decimal place involves the
loss of real accuracy. A trade-off between communication and accuracy has to be
made. On balance, communication is placed first and some numbers for Ireland
and Luxembourg rounded to one figure. Comparisons involving them can still
be made. In any case, students of the economic progress of Luxembourg would
doubtless not be using this table as a source of data.
(b) Rule 2. The rows are ordered by size. The previous order was UK first (it is a
UK company’s report) followed by the other EC countries alphabetically, fol-
lowed by the two non-EC countries. Alphabetical order does not help in
answering the questions, nor does the separation of the USA and Japan (does
anyone doubt that they are not EC members?). Size order makes it easier to
compare countries of similar size as well as immediately indicating changes in
ranking. The ordering in Table 3.11 is based on the 1997 GDP. It could equally
have been based on the 1965 GDP or the populations. There is often no one
correct method of ordering. It is a matter of taste.
(c) Rule 3. Interchanging rows and columns is not necessary. One is just as likely to
wish to compare countries (down the columns) as one country across the years
(along the rows).
(d) Rule 4. The use of summary measures is not particularly helpful. Row and
column averages would have no intuitive meaning. For example, the average
GDP for a country over a selection of years is not a useful summary.
(e) Rule 5. The vertical lines in the original table hinder comparisons across the
years and have been omitted.

Quantitative Methods Edinburgh Business School 3/11

Module 3 / Data Communication

(f) Rule 6. The labelling is already clear. No changes have been made.
(g) Rule 7. It would be very difficult to make a simple verbal summary of these
data. Moreover, in the context, the publishers would probably not wish to be
appearing to lead the reader’s thinking by suggesting what the patterns were.
The typical questions that might be asked of these data can now be applied to
Table 3.11. It is possible to see quickly that Germany’s GDP increased by 1900/110
= just over 17 times; Italy’s by 1000/53 = just over 19 times; Japan’s by just under
45 times; the UK’s by 11; Japan has overtaken Germany, France and the UK;
Ireland is over four times the size of Luxembourg economically. The information is
more readily apparent from Table 3.11 than from Table 3.10.

3.3 The Special Case of Accounting Data

The communication of accounting data requires special treatment. The main factors
that make accounting data different are:
(a) they have a logical sequence (e.g. a revenue and expenditure statement builds up
to a final profit);
(b) there is a tradition of how accounts should be presented (e.g. negative amounts
are shown in brackets);
(c) there are legal requirements as to what should be presented.
These factors do not mean that the presentation of accounts cannot be im-
proved, but it does mean that extra care must be taken and that proposed changes
may be met with resistance. The difficulty seems to be that what constitutes a well-
presented set of accounts to an accountant may differ from what constitutes a well-
presented set of accounts to a layperson. The accountant’s criteria are probably
based on what is good practice when drawing up accounts and on the technical
approval of other accountants. For example, most final accounts use brackets to
indicate a negative. This may make sense when preparing accounts and when
dealing with other accountants, but most laypeople are more used to a minus sign to
indicate a negative, and most financial data are for laypeople. Published accounts are
for shareholders; internal company financial information is for managers (not all of
whom are former accountants). It is more important that accounting data should be
in a form suitable for accounting laypeople than suitable for accountants.
The accounts in Table 3.12 are a case in point. They are a disguised version of
the accounts of a multinational shipping company, and an accounting association
awarded them a prize for their excellence. While they doubtless represent fine
accounting practice, it is almost impossible to tell what, financially, befell the
company between the two years. Communication of the data can be improved by
applying some of the presentation ‘rules’. (Only rules 1, 5, 6 and 7 are appropriate
for accounts; rules 2, 3 and 4 cannot be applied to tables that have a logical se-
quence.)
Some interesting facts emerge when the data are re-presented. Table 3.13 shows
the same accounts re-presented (in just one language). The new table reveals that it
was an unusual year for the company. Incomes and expenditures changed by large
and extremely variable amounts.

3/12 Edinburgh Business School Quantitative Methods

 Module 3 / Data Communication

Table 3.12 Shipping company accounts

2015 2016
£ £000

Gross Freight Income 98 898 684 73 884

Voyage Expenses
Own Vessels 31 559 336 (24 493)
Vessels on Timecharter 42 142 838 (73 702 174) (28 378)

Operating Expenses:
Crew Wages and Social Security 7 685 965 (9 010)
Other Crew Expenses 541 014 (633)
Insurance Premiums 1 161 943 (1 367)
Provisions and Stores 1 693 916 (2 268)
Repairs and Maintenance 1 685 711 (3 297)
Other Operating Expenses 60 835 (27)
(12 829 384)

Result from Vessels 12 367 126 4 411

Results from Parts in Limited Partner- 1 793314 (163)
ships
Result from Operation Business 167 343 0
Building
Management Expenses
Salaries, Fees, Employees’ 1 426 607 (1 208)
Benefits, etc.
Other Management Expenses 502 815 (635)
(1 929 422)

Depreciation Fixed Assets (7 106 305) (5 365)

Transferred from Classification 112 401 652
Survey Fund
Set aside for Classification Survey (612 401) (27)
Fund
(500 000)

4 792 056 (2 335)

Capital Income and Expenses
Misc. Interest Income 318 601 364
Misc. Interest Expenses (8 450 307) (6 482)

(8 131 706)

Quantitative Methods Edinburgh Business School 3/13

Module 3 / Data Communication

2015 2016
£ £000
Net Profit/(Loss) Currency Exch. (190 836) (680)
Dividends 35 732 47
(8 286 810)

Other Income and Expenses

Net Profit from Sales Fixed 5 553 047 5 593
Assets
Balance Classification Survey and 1 414 132 858
Self Insurance Fund upon Sale

6 967 179

Net Profit/(Loss) from Sale of Shares (38 647) 0

Misc. Income/(Expenses) relating to 629 630 1 408
previous years
Adjustm. Cost Shares 226 197 (206)
Loss from Receivables (219 189) 0
Reversal write up Fixed Assets (2 026 067) 0
previous years
5 539 103

Result before Taxes and Allocations 2 044 349 (1 433)

Allocations (1 929 400)
114 949 (1 433)

Reserves (11 495) 1 071

Dispositions (103 454) 362
(114 949) 1 433

(a) Freight income up 33%

(b) Voyage expenses up 40%
(c) Operating expenses down 20%
(d) Operating profit up 180%
(e) Partnership shares Small −ve to large +ve
(f) Management expenses equal
(g) Capital income/expense up 20%
(h) Other income/expense down 25%
(i) Profit before tax −ve to +ve

3/14 Edinburgh Business School Quantitative Methods

 Module 3 / Data Communication

Table 3.13 Shipping company accounts (amended)

2015 2016
Statement of revenue and expenses (£ million)
Gross freight income 98.9 73.9
Voyage expenses −73.7 −52.9
Operating expenses −12.8 −16.6
Profit from vessels 12.4 4.4
Shares in partnership 1.8 −0.16
Operation business building 0.2 0
Management expenses −1.9 −1.8
Depreciation −7.1 −5.4
Classification survey (net) −0.5 0.6
Capital income and expenses −8.3 −6.8
Other income and expenses 5.5 7.7
Profit before tax 2.0 −1.4
Allocations −1.9 0
To reserves 0.1 −1.4

These features of the company’s finances are remarkable. Even when one knows
what they are, it is very difficult to see them in the original table (Table 3.12). Yet it
is this volatility that is of major interest to shareholders and managers.
The question of rounding creates special difficulties with accounting data. The
reason is that rounding and exact adding up are not always consistent. It has to be
decided which is the more important – the better communication of the data or the
need to allow readers to check the arithmetic. The balance of argument must weigh
in favour of the rounding. Checking totals is a trivial matter in published accounts
(although not, of course, in the process of auditing). If a mistake were found in the
published accounts of such a large company, the fault would almost certainly lie
with a printer’s error. But ‘total checking’ is an obsessive pastime and few compa-
nies would risk the barrage of correspondence that would undoubtedly ensue even
though a note to the accounts explained that rounding was the cause. Because of
this factor the two effective figures rule may have to be broken so that adding and
subtracting are exact. This has been done in Table 3.13. The only remaining
question is to wonder why totals are exactly right in company accounts which have
in any case usually been rounded to some extent (the nearest £ million for many
companies). The answer is that figures have been ‘fudged’ to make it so. The same
considerations apply, but to a lesser degree, with internal company financial infor-
mation.
Communicating financial data is an especially challenging area. The guiding prin-
ciple is that the main features should be evident to the users of the data. It should
not be necessary to be an expert in the field nor to have to carry out a complex
analysis in order to appreciate the prime events in a company’s financial year. Some
organisations are recognising these problems by publishing two sets of (entirely

Quantitative Methods Edinburgh Business School 3/15

Module 3 / Data Communication

consistent) final accounts. One is source material, covering legal requirements and
suitable for financial experts; the other is a communicating document, fulfilling the
purpose of accounts (i.e. providing essential financial information). Other organisa-
tions may, of course, have reasons for wanting to obscure the main features of their
financial year.

3.4 Communicating Data through Graphs

The belief that a picture speaks a thousand words (or numbers) is widely held. For
most people a picture is more interesting and attractive than numbers. A graph is a
form of picture that can, in certain circumstances, be very helpful in communicating
data. They are not, however, always helpful. It is important to distinguish between
the helpful and unhelpful uses of a graph.

Deposit rates: annual percentage average

14

12

10
Rate %

0
86 87 88 89 90 91 92 93 94 95 96
Year

Figure 3.1 USA interest rates

Source: International Monetary Fund

3/16 Edinburgh Business School Quantitative Methods

 Module 3 / Data Communication

Deposit rates: annual percentage average

14

12

10

Italy
6
USA

4 Canada
France
Netherlands
Denmark
2 Belgium
Germany
Japan
0
86 87 88 89 90 91 92 93 94 95 96
Year

Figure 3.2 International interest rates

Source: International Monetary Fund
Figure 3.1 is a graph of interest rates in the USA in the years 1986–96. It works
well. The pattern can be seen quickly and easily. This is no longer so if the graph is
used for comparing interest rates in several countries. Figure 3.2 is Figure 3.1 with
several more countries added. The graph ceases to function well. There are too
many lines and they tend to cross over. As a result, no overall message emerges.
Furthermore, the graph cannot be used for reference purposes without difficulty
and inaccuracy. What, for instance, was the interest rate in Belgium in 1993?
This example illustrates the principles underlying the use of graphs for data
communication:
(a) Graphs are helpful when:
 attracting attention and adding variety in a series of tables;
 communicating very simple patterns.
(b) Graphs are not helpful when:
 communicating even slightly complex patterns;
 being used for reference purposes.
These principles are straightforward and it may appear at first sight that little
could go wrong in the use of graphs in organisations. However, things do go very
wrong with surprising frequency. Figure 3.3 is a disguised version of part of the
management information system of a large company. It was one of a series of about
40 similar graphs.
It is not difficult to see that it has been produced automatically from a computer
with little thought to communication (the vertical scale is an indication that it comes
hot from the computer). Nor is it difficult to say what is wrong.

Quantitative Methods Edinburgh Business School 3/17

Module 3 / Data Communication

Key: Denmark
Spain
Italy
Netherlands
Belgium
1500.0

1239.0

978.0

717.0

456.0

195.0
JAN JAN JAN JAN JAN
2011 2012 2013 2014 2015

Figure 3.3 Annual coal imports as monthly averages

3/18 Edinburgh Business School Quantitative Methods

 Module 3 / Data Communication

No pattern of any sort is remotely evident. To extract an individual number (say,

the imports for Denmark in March 2013) requires first being able to find the line for
the country and, second, to estimate the imports by a calculation along the lines of
‘about 2/5 of the way between 456 and 717’. In short, the graph is good neither for
showing patterns nor for reference purposes. Since the graph is unusable, at the very
least, expense could be saved by not issuing it.
To be fair, these data are difficult to communicate by any means because there
are so many. Distinct improvements can nevertheless be made, using the rules of
presentation. Table 3.14 gives the same data as are on the graph in Figure 3.3 but in
tabular form. The table can certainly be used for reference purposes. In addition, it
is possible, although not easy, to see the seasonal pattern in the data.

Table 3.14 Annual coal imports as monthly averages

Italy Belg. Den. Neth. Spain Italy Belg. Den. Neth. Spain
2011 Jan. 910 610 230 280 260 2013 Jan. 620 310 360 420 290
Feb. 1200 850 360 360 240 Feb. 1000 520 250 320 440
Mar. 1300 720 340 420 440 Mar. 1100 630 460 430 260
Apr. 1200 590 340 350 430 Apr. 990 660 430 540 410
May 1100 520 310 410 360 May 1200 480 590 290 420
June 1100 570 330 440 320 June 790 580 560 390 330
July 1300 380 380 240 290 July 1300 470 520 370 370
Aug. 1100 390 330 320 250 Aug. 640 540 570 400 330
Sept. 800 360 370 300 210 Sept. 1000 540 510 270 330
Oct. 920 360 380 390 500 Oct. 1000 490 320 500 490
Nov. 1100 390 400 370 240 Nov. 900 570 470 440 380
Dec. 970 500 370 340 290 Dec. 1100 680 530 420 370

2012 Jan. 570 450 230 420 380 2014 Jan. 930 470 470 300 360
Feb. 800 550 300 280 310 Feb. 780 510 530 390 420
Mar. 1100 770 330 400 430 Mar. 900 590 440 260 440
Apr. 910 690 540 270 380 Apr. 780 530 440 490 440
May 970 690 290 390 420 May 1100 510 420 400 440
June 910 660 350 520 240 June 1000 650 440 510 350
July 900 690 370 430 300 July 1100 550 390 350 400
Aug. 900 580 330 240 300 Aug. 870 580 570 350 360
Sept. 750 520 480 430 340 Sept. 1100 610 460 380 360
Oct. 1400 640 410 380 430 Oct. 1100 750 730 360 530
Nov. 1100 590 360 560 430 Nov. 950 660 750 530 410
Dec. 1000 450 340 570 430 Dec. 1200 600 650 500 400

2015 Jan. 570 790 490 500 390

Feb. 1100 520 360 470 450
Mar. 880 870 460 520 450

Average of all years 980 560 420 390 370

Quantitative Methods Edinburgh Business School 3/19

Module 3 / Data Communication

If the general pattern over the years or a comparison between countries is re-
quired, Table 3.15 is suitable. This shows the average monthly imports of coal for
each year. It can now be seen that four of the countries have increased their imports
by a factor of 30–45 per cent. Italy is the exception, having decreased coal imports
by 20 per cent. The level of imports in the countries can be compared.

Table 3.15 Monthly coal imports (average per month in thousand

tonnes)
Italy Belg. Den. Neth. Spain
2011 1100 520 340 350 320
2012 940 610 340 400 370
2013 970 490 460 400 370
2014 980 590 510 400 410
2015 870 750 450 480 430
Average 980 560 420 390 370

In these terms, Italy is the largest, followed by Belgium, followed by the other
three at approximately the same level.
Table 3.15 can be transferred to a graph, as shown in Figure 3.4. General patterns
are evident. Italy has decreased its imports, the others have increased theirs; the level
of imports is in the order Italy, Belgium … The difference between the table and the
graph becomes clear when magnitudes have to be estimated. The percentage change
(−20 per cent for Italy, etc.) is readily calculated from the table, but not from the
graph. In general, graphs show the sign of changes but a table is needed to make an
estimate of the size of the changes. The purpose of the data and personal prefer-
ence would dictate which of the two were used.

1200

1000
Italy
800 Belgium

600 Netherlands
Denmark
400 Spain

200

2011 2012 2013 2014 2015

Figure 3.4 Monthly coal imports as annual averages

Graphs are the most important but not the only pictorial method of communi-
cating numbers. A wide range of possibilities exists, all of which have their
advantages and disadvantages. The underlying principles are the same as for graphs.

3/20 Edinburgh Business School Quantitative Methods

 Module 3 / Data Communication

Pictures are useful for attracting attention and for showing very general patterns.
They are not useful for showing complex patterns or for extracting actual numbers.

Learning Summary
The communication of data is an area that has been neglected, presumably because
it is technically simple and there is a tendency in quantitative areas (and perhaps
elsewhere) to believe that only the complex can be useful. Yet in modern organisa-
tions there can be few things more in need of improvement than data
communication.
Although the area is technically simple, it does involve immense difficulties. What
exactly is the readership for a set of data? What is the purpose of the data? How can
the common insistence on data specified to a level of accuracy that is not needed by
the decision maker and is not merited by the collection methods be overcome? How
much accounting convention should be retained in communicating financial
information to the layperson? What should be done about the aspects of data
presentation that are a matter of taste? The guiding principle among the problems is
that the data should be communicated according to the needs of the receiver rather
than the producer. Furthermore, they should be communicated so that the main
features can be seen quickly. The seven rules of data presentation described in this
module seek to accomplish this.
Rule 1: round to two effective digits.
Rule 2: reorder the numbers.
Rule 3: interchange rows and columns.
Rule 4: use summary measures.
Rule 5: minimise use of space and lines.
Rule 6: clarify labelling.
Rule 7: use a verbal summary.
Producers of data are accustomed to presenting them in their own style. As al-
ways there will be resistance to changing an attitude and presenting data in a
different way. The idea of rounding especially is usually not accepted instantly.
Surprisingly, however, while objections are raised against rounding, graphs tend to
be universally acclaimed, even when not appropriate. Yet the graphing of data is the
grossest form of rounding. There is evidently a need for clear and consistent
thinking in regard to data communication.
This issue has been of increasing importance because of the growth in usage of
all types and sizes of computers and the development of large-scale management
information systems. The benefits of this technological revolution should be
enormous but the potential has yet to be realised. The quantities of data that
circulate in many organisations are vast. It is supposed that the data provide
information which in turn leads to better decision making. Sadly, this is frequently
not the case. The data circulate, not providing enlightenment, but causing at best
indifference and at worst tidal waves of confusion. Poor data communication is a
prime cause of this. It could be improved. Otherwise, one must question the
wisdom of the large expenditures many organisations make in providing untouched

Quantitative Methods Edinburgh Business School 3/21

Module 3 / Data Communication

and bewildering management data. One thing is clear: if information can be assimi-
lated quickly, it will be used; if not, it will be ignored.

Review Questions
3.1 In communicating management data, which of the following principles should be adhered
to?
A. The requirements of the user of the data are paramount.
B. Patterns in the data should be immediately evident.
C. The data should be specified to two decimal places.
D. The data should be analysed before being presented.

3.2 The specification of the data (the number of decimal places) indicates the accuracy. True
or false?

3.3 The accuracy required of data should be judged in the context of the decisions that are
to be based upon the data. True or false?

3.4 3732.578 when rounded to two effective figures becomes:

A. 3732.58
B. 3700
C. 3730
D. 3732

3/22 Edinburgh Business School Quantitative Methods

 Module 3 / Data Communication

3.5 The column of numbers:

1732
1256.3
988.42
38.1

when rounded to two effective figures becomes:

A.
1730
1260
988
38

B.
1730
1260
990
38

C.
1700
1300
990
38

3.6 Which are correct reasons? It is easier to compare numbers in a column than in a row
because:
A. The difference between two- and three-figure numbers is quickly seen.
B. Subtractions of one number from another are made more quickly.
C. The numbers are likely to be closer together and thus easier to analyse quickly.

3.7 When the rows (each referring to the division of a large company) of a table of numbers
are ordered by size, the basis for the ordering should be:
A. The numbers in the left-hand column.
B. The averages of the rows.
C. The capital employed in the division.
D. The level of manpower employed in the division.

Quantitative Methods Edinburgh Business School 3/23

Module 3 / Data Communication

3.8 What criticisms can be made of the following table of numbers?

Sales region Sales (£000)

Eastern 1230
Northern 1960
South-West 1340
South of Scotland 1030
Wales 1220

A. Numbers not rounded to two effective figures.

B. Regions not in size order.
C. Gap in column of numbers because of labelling.
D. There should be a vertical line between Sales region and Sales.

3.9 Only some of the presentation rules can be applied to financial accounts. This is
because:
A. Rounding cannot be done because the reader may want to check that the
auditing has been correct.
B. Rounding cannot be done because it is illegal.
C. An income statement cannot be ordered by size since it has to build up to a
final profit.
D. Published annual accounts are for accountants; therefore their presentation is
dictated by accounting convention.

3.10 Graphs should be used in which of the following circumstances?

A. Where changes over time are involved.
B. To make data visually more attractive.
C. To distinguish fine differences in just a few variables.
D. To communicate very simple patterns.

Case Study 3.1: Local Government Performance Measures

Table 3.16 shows some performance measurement data relating to local area govern-
ments. The data are percentages that show the extent to which each of the six
authorities has attained predefined objectives in each of five months. For example,
Southam had achieved only 32.5 per cent of its objectives in December 2015 but had
increased this to 70.0 per cent by June 2016.

Table 3.16 Local government performance measures

December February April May June
Eastham 25.0 77.9 74.7 79.0 94.0
Northington 14.7 66.7 85.9 88.0 88.0
Plumby 56.0 67.4 75.0 55.0 71.0
Southam 32.5 49.2 64.6 66.0 70.0
Tyneham 40.0 100.0 84.3 100.0 100.0
Westerley 43.7 86.0 89.5 72.0 87.0

3/24 Edinburgh Business School Quantitative Methods

 Module 3 / Data Communication

1 This table is one of many that elected representatives have to consider at their monthly
meetings. The representatives need, therefore, to be able to appreciate and understand
the main features very quickly. In these circumstances, how could the data be presented
better? Redraft the table to illustrate the changes.

Case Study 3.2: Multinational Company’s Income Statement

Table 3.17 is taken from the annual accounts of a large multinational company. A second
income statement later in the accounts is more detailed and presumably satisfies legal
requirements and the needs of financial specialists. Table 3.17 is therefore intended to
communicate the main features of the company’s financial year to non-specialists who
may be shareholders, trade unionists, etc.

Table 3.17 Income statement summary of combined figures

£ million 2015 2016
Results for the year ended 31 December
Sales to third parties 9147 9842

Operating profit 541 601

Concern share of associated companies’ profit 59 64
Non-recurring and financial items 50 56

Profit before taxation 550 609

Taxation 272 315

Profit after taxation 278 294

Outside interests and preference dividends 20 21

Profit attributable to ordinary capital 258 273

Ordinary dividends 95 106

Profit of the year retained 163 167

1 Compared to many accounting statements Table 3.17 is already well presented, but
what further improvements might be made?

Case Study 3.3: Country GDPs

1 Consider again Table 3.10 and Table 3.11 earlier in the module. Table 3.10 was the
original and Table 3.11 an amended version. Could the data have been presented better
by means of a graph?

Quantitative Methods Edinburgh Business School 3/25

Module 3 / Data Communication

Case Study 3.4: Energy Efficiency

An organisation of energy consultants advises companies on how they might save money
by making more efficient use of oil. Table 3.18 is part of a recommendation prepared for
the senior managers of a client company and shows the financial return from a capital
investment in oil conservation. The investment is evaluated by estimating the internal
rate of return (IRR) on the project; the IRR being a measure of the profitability of a
capital project; in general terms, the higher the IRR, the more profitable the project
(although it is not necessary to understand the meaning of IRR in order to answer this
case). Some of the inputs to the calculation (e.g. the life of the project and the level of
initial investment) are uncertain. Therefore, some sensitivity analysis has been done. In
other words, the IRR has been calculated several times. Each time one of the assump-
tions is varied, all others remain fixed at their original values.

Table 3.18 Sensitivity analysis for oil conservation investment

Base IRR 24.93% Variation in New Difference

parameter IRR in IRR

Rebuilding investment + £0.5m (£4.342m) 22.71% – 2.22%

cost (£3.842m) – £0.5m (£3.342m) 30.49% + 5.56%

Outstanding life (13 years) – 1 yr (12 years) 23.10% – 1.83%

– 2 yrs (11 years) 22.62% – 2.31%
– 3 yrs (10 years) 22.06% – 2.87%
– 4 yrs (9 years) 21.27% – 3.66%

Fuel consumption + 5T/day (48T/day) 27.73% + 2.80%

differential (43 tonnes/day) – 5T/day (38T/day) 21.96% – 2.97%

Fuel price + 1%/yr (13%/year) 26.42% + 1.49%

escalation rate (12%/year) – 1%/yr (11%/year) 23.51% – 1.42%

1 Table 3.18 gives the results of this sensitivity analysis. It shows the extent to which the
assumptions have been varied and the new IRR for each variation. The ‘base rate’ is the
IRR for the original calculation. How could it be better presented? (Note that it is not
necessary to understand the situation fully in order to propose improvements to the
data communication.)

References
Ehrenberg, A. S. C. (1975). Data Reduction. New York: John Wiley and Sons.

3/26 Edinburgh Business School Quantitative Methods

Module 4

Data Analysis
Contents
4.1 Introduction.............................................................................................4/1
4.2 Management Problems in Data Analysis .............................................4/2
4.3 Guidelines for Data Analysis ..................................................................4/6
Learning Summary ......................................................................................... 4/15
Review Questions ........................................................................................... 4/16
Case Study 4.1: Motoring Correspondent ................................................... 4/17
Case Study 4.2: Geographical Accounts ...................................................... 4/18
Case Study 4.3: Wages Project ..................................................................... 4/19

Prerequisite reading: Module 3

Learning Objectives
By the end of this module the reader should know how to analyse data systematical-
ly. The methodology suggested is simple, relying very much on visual interpretation,
but it is suitable for most data analysis problems in management. It carries implica-
tions for the ways information is produced and used.

4.1 Introduction
What constitutes successful data analysis? There is apparently some uncertainty on
this point. If a group of managers are given a table of numbers and asked to analyse
it, most probably they will ‘number pick’. Individual numbers from somewhere in
the middle of the table which look interesting or which support a long-held view
will be selected for discussion. If the data are profit figures, remarks will be made
such as: ‘I see Western region made £220 000 last year. I always said that the new
cost control system would work.’ A quotation from Andrew Lang, a Scottish poet,
could be applied to quite a few managers: ‘He uses statistics as a drunken man uses
lamp posts – for support rather than illumination.’
Real data analysis is concerned with seeking illumination, not support, from a set
of numbers. Analysis is defined as ‘finding the essence’. A successful data analysis
must therefore involve deriving the fundamental patterns and eliciting the real
information contained in the entire table. This must happen before sensible remarks
can be made about individual numbers. To know whether the cost system in the
above example really did work requires the £220 000 to be put in the context of
profit and cost patterns in all regions.

Quantitative Methods Edinburgh Business School 4/1

Module 4 / Data Analysis

The purpose of this module is to give some guidelines showing how illumination
might be derived from numbers. The guidelines give five steps to follow in order to
find what real information, if any, a set of numbers contains. They are intended to
provide a framework to help a manager understand the numbers he or she encoun-
ters.
One might have thought that understanding numbers is what the whole subject
of statistics is about, and so it is. But statistics was not developed for use in man-
agement. It was developed in other fields such as the natural sciences. When it is
transferred to management, there is a gap between what is needed and what
statistics can offer. Certainly, many managers, having attended courses or read
books on statistics, feel that something is missing and that the root of their problem
has not been tackled. This and other difficulties involved in the analysis of manage-
ment data will be pursued in the following section, before some examples of the
types of data managers face are examined. Next, the guidelines, which are intended
to help fill the statistics gap, will be described and illustrated. Finally, the implica-
tions of this gap for the producers of statistics will be discussed.

4.2 Management Problems in Data Analysis

Managers face a unique set of problems in their general task of understanding
numbers. This may partly explain the poor or non-existent data analysis seen (or not
seen) in a lot of companies. Rich sources of information, of great potential value in
decision making, are often ignored. The first step in making an improvement is
accepting that these problems exist. The set of problems includes the following
main features:
(a) The statistical gap. The subject of statistics does not provide all the techniques
and methods that a manager would like to have at his or her disposal. For in-
stance, much of statistics is concerned with the rigorous testing of hypotheses.
The manager faced with last month’s sales figures or some accounting data has
first to find a hypothesis (i.e. has first to find what might be the real information
in the data). Statistics gives no clear help with this task. Having found some real
information by whatever means, the manager is less concerned with rigorous
statistical testing than with taking action. He or she will no doubt test the validity
of the information, but by other informal, perhaps qualitative, means. Contrast
this with the natural sciences where the emphasis is on thoroughly testing under
different conditions hypotheses derived by non-statistical methods (such as be-
ing hit on the head by an apple). Statistics plays a different role in the natural
sciences than in management. What the manager primarily needs, and what sta-
tistics does not fully provide, are techniques for first detecting patterns and
regularities in data, without having to wait for an external impetus.
(b) A lack of confidence. This manifests itself in different forms: from the free
admission of a fear of numbers to the aggressive statement that management is
solely a matter of instinct and that methodical evaluations of information are
unnecessary. Whatever the manifestation, the effect is the same: little or no data
analysis is done. In fact the majority of numbers problems require only a mini-
mal knowledge of technical matters but a lot of common sense. Data analysis is

4/2 Edinburgh Business School Quantitative Methods

 Module 4 / Data Analysis

rather like reading. When looking at a business report, a manager will usually
read it carefully, work out exactly what the author is trying to say and then decide
whether it is correct. The process is similar with a table of numbers. The data
have to be sifted, thought about and weighed. To do this, good presentation (as
stressed in Module 3 in the rules for data presentation) may be more important
than sophisticated techniques. Most managers could do excellent data analyses
provided they had the confidence to treat numbers more like words. It is only
because most people are less familiar with numbers than words that the analysis
process needs to be made more explicit (via guidelines such as those in Section
4.3 below) in the case of numbers.
(c) Over-complication by the experts. The attitude of numbers experts (and other
sorts of experts as well) can confuse managers. The experts use jargon, which is
fine when talking to their peers but not when talking to a layperson; they try
sophisticated methods of analysis before simple ones; they communicate results
in a complicated form, paying little regard to the users of the data. For example,
vast and indigestible tables of numbers, all to five decimal places, are often the
output of a management information system. The result can be that the experts
distance themselves from management problems. In some companies specialist
numbers departments have adopted something akin to a research and develop-
ment role, undertaking solely long-term projects. Managers come to believe that
they have not the skills to help themselves while at the same time believing that
no realistic help is available from experts.

4.2.1 Examples of the Problems Faced

The following three examples are all situations in which experts have over-
complicated the presentation or analysis of data. In each case a simpler treatment of
the data reveals important information.

Accounting Data
In Module 3 Table 3.12 showed the income statement of a multinational shipping
company. It is difficult to analyse (i.e. it is difficult to say what the significant
features of the company’s business were). Some important happenings are obscured
in Table 3.12, but they were revealed when the table was re-presented in Table 3.13.

Quantitative Methods Edinburgh Business School 4/3

Module 4 / Data Analysis

Table 4.1 Budgeting data from an MIS

PORT - OCEAN PORT TERMINAL COSTS - SHIPSIDE OPERATIONS SCHEDULE 2
PERIOD - (CURRENCY US DOLLARS)

MONTH CUMULATIVE
TERMINAL COSTS
ESTIMATE STANDARD VARIANCE VAR % ESTIMATE STANDARD VARIANCE VAR % BUDGET
LO-LO
STEVEDORING
STRAIGHT TIME - FULL 131 223 143 611 1 288 8.6 1 237 132 1 361 266 124 134 9.1 1 564 896
STRAIGHT TIME - M.T. 13 387 14 651 1 264 8.6 256 991 281 399 24 408 8.7
(UN)LASHING 78 (78) 78 (78)
SHIFTING 801 (801) 11 594 (11 594)
OVERTIME, SHIFT TIME OF
WAITING & DEAD TIME 7 102 (7 102) 190 620 (190 620)

RO-RO
STEVEDORING
TRAILERS
STRAIGHT FULL 20 354 26 136 5 782 22.1 167 159 215 161 48 002 22.3 330 074
STRAIGHT M.T. 178 228 50 21.9 14 846 18 993 4 147 21.8
RO-RO COST PLUS
VOLVO CARGO
ROLLING VEHICLES 14 326 19 515 5 189 26.6 98 210 157 163 58 951 37.5
BLOCKSTONED 29 27 (2) (7.4) 613 674 61 9.1
(UN) LASHING RO-RO 355 (355) 355 (355)
SHIFTING 977 (977) 3 790 (3 790)
OVERTIME, SHIFT TIME OF
WAITING & DEAD TIME 1 417 (1 417) (28 713) (28 713)
HEAVY LIFTS (OFF STANDARD) 2 009 (2 009)

CARS
STEVEDORING
STRAIGHT TIME 6 127 6 403 276 4.3 38 530 35 328 (3 202) (9.1) 168 000
(UN) LASHING 2 (2)
SHIFTING 795 (795) 1 288 (1 288)
OVERTIME, SHIFT TIME OF
WAITING & DEAD TIME 7 573 (7 573)
OTHER SHIPSIDE OF COSTS 3 422 (3 422) 24 473 (24 473)

TOTAL TERMINAL COSTS 200 571 210 571 16 000 4.5 2 083 976 2 069 984 (13 992) (.7) 2 062 970

4/4 Edinburgh Business School Quantitative Methods

 Module 4 / Data Analysis

Management Information System Output

Table 4.1 is an example of perfectly accurate and meaningful data which are
presented in a form convenient for the producer of the data but not for the user.
A manager would take a considerable time to understand the table. Since this is
part of a Management Information System (MIS) which contains many such tables,
it is essential that a manager should be able to assimilate the information contained
in the table very rapidly. A little thought given to simplifying the table in accordance
with the needs of the user would bring an enormous improvement (see Table 4.5).

Market Research
Table 4.2 indicates what can happen when experts over-complicate an analysis. The
original data came from interviews of 700 television viewers who were asked which
British television programmes they really like to watch. The table is the result of the
analysis of this relatively straightforward data. It is impossible to see what the real
information is, even if one knows what correlation means. However, a later and
simpler analysis of the original data revealed a result of wide-ranging importance in
the field of television research. (See Ehrenberg, 1975, for further comment on this
example.)

Table 4.2 Result of the analysis of some market research data

Adults who ‘Really Like to Watch’: Correlations to 4 decimal places
(Programmes ordered alphabetically within channel)
PrB ThW Tod WoS GrS LnU MoD Pan RgS 24H
ITV PrB 1.0000 0.1064 0.0653 0.5054 0.4741 0.0915 0.4732 0.1681 0.3091 0.1242
ITV ThW 0.1064 1.0000 0.2701 0.1474 0.1321 0.1885 0.0815 0.3520 0.0637 0.3946
ITV Tod 0.0653 0.2701 1.0000 0.0926 0.0704 0.1546 0.0392 0.2004 0.0512 0.2437
ITV WoS 0.5054 0.1474 0.0926 1.0000 0.6217 0.0785 0.5806 0.1867 0.2963 0.1403
BBC GrS 0.4741 0.1321 0.0704 0.6217 1.0000 0.0849 0.5932 0.1813 0.3412 0.1420
BBC LnU 0.0915 0.1885 0.1546 0.0785 0.0849 1.0000 0.0487 0.1973 0.0969 0.2661
BBC MoD 0.4732 0.0815 0.0392 0.5806 0.5932 0.0487 1.0000 0.1314 0.3267 0.1221
BBC Pan 0.1681 0.3520 0.2004 0.1867 0.1813 0.1973 0.1314 1.0000 0.1469 0.5237
BBC RgS 0.3091 0.0637 0.0512 0.2963 0.3412 0.0969 0.3267 0.1469 1.0000 0.1212
BBC 24H 0.1242 0.3946 0.2437 0.1403 0.1420 0.2661 0.1221 0.5237 0.1212 1.0000

In all three examples any message in the data is obscured. They were produced by
accountants, computer scientists and statisticians respectively. What managers
would have the confidence to fly in the face of experts and produce their own
analysis? Even if they had the confidence, how could they attempt an analysis? The
guidelines described below indicate, at a general level, how data might be analysed.
They provide a starting point for data analysis.

Quantitative Methods Edinburgh Business School 4/5

Module 4 / Data Analysis

4.3 Guidelines for Data Analysis

The guidelines comprise five stages. A manager who follows them should be able to
understand better the data he or she is faced with. They are, however, guidelines and
not a rigorous methodology which guarantees complete success every time.

4.3.1 Stage 1: Reduce the Data

Many tables contain too many numbers. Data are often included on the grounds
that ‘someone, somewhere, may need them’. For the sake of these probably
mythical people, all others wanting to understand the data have to distinguish what
is important from an enveloping confusion of other irrelevant data.
It is true that the producer of the data has to cater for the many requirements of
the people who receive them. Even so, producers tend to err on the side of over-
supply and include more than is necessary, even given a wide range of users. The
effect is multiplied if receivers of data, on the occasions when they are asked what
they require, overstate their needs ‘just in case’.
The first stage of an analysis is, then, to cut down on the numbers. This means
omitting those that are redundant or irrelevant. This is very much a matter of
judgement, but it is important to resist the assumption that every single piece of data
must be important because it has been included. The producer is unlikely to have
included them because he or she knows them to be important (how would they
know?). They will probably have no idea whether they are important or not. They
will have been included because they are available and the producer is being
cautious. Moreover, it is easier to include everything than to have to agonise over
what should be excluded. By reducing the data the analyst may well find that he or
she is dealing with just a fraction of the original set.

4.3.2 Stage 2: Re-present the Data

The visual approach is central to the understanding of numbers. A pattern can often
be seen quickly in a set of well-arranged data which otherwise (for example, when
computerised) appears to be no more than a random jumble. Recall the similarity
between handling words and numbers. In the same way that a verbal report is sifted,
meditated upon and mused over, so time also needs to be spent thinking about a
table of numbers. If the numbers are skilfully presented, the thinking process should
be shorter and there will be a greater chance that they will be understood. Good
presentation of numbers is analogous to clarity of style and the absence of jargon in
verbal reports. However, the visual aspect of data analysis and the need to present
data well is usually neglected, perhaps because it does not have a high technical
content, but the approach has been shown to work in practice.

4/6 Edinburgh Business School Quantitative Methods

 Module 4 / Data Analysis

The re-presentation being recommended does not refer just to data that are liter-
ally a random jumble. On the contrary, the assumption is that the data have already
been assembled in a neat table. Neatness is preferable to messiness but the patterns
may still be obscured. When confronted by apparent orderliness one should take
steps to re-present the table in a fashion which makes it easy to see any patterns
contained in it. The ways in which data can be rearranged were explored in detail in
the previous module. Recall that the seven steps were:
(a) Round the numbers to two effective figures.
(b) Put rows and columns in size order.
(c) Interchange rows and columns where necessary.
(d) Use summary measures.
(e) Minimise use of gridlines and white space.
(f) Make the labelling clear and do not allow it to hinder comprehension of the
numbers.
(g) Use a verbal summary.

4.3.3 Stage 3: Build a Model

Building a model is a mathematical-sounding expression for a straightforward task.
It is this stage that can lead the analyst to an unnecessarily technical and mathemati-
cal solution. The aim in building a model is essentially to find a pattern in the
numbers and some way of summarising them. The model may be a verbal one, an
arithmetic one or an algebraic expression, either simple or complex. Frequently the
simpler models prove sufficient. Even so, this stage is the most difficult and skilful,
requiring an element of creativity. The following are examples of very different
models, all of which proved adequate for the particular data and context concerned.
(a) Row averages all equal but with a ±20 per cent variation within each column.
(b) Real profits increased by 5 per cent p.a. between 2005 and 2010 but down by 2
per cent p.a. between 2011 and 2016.
(c) Sales per salesperson approximately equal for each northern region; sales per
salesperson approximately equal for each southern region; but the northern fig-
ure is 25 per cent higher than the southern.
(d) Column 1 (represented by y) is related to column 2 (represented by x) by the
expression: y = 0.75x + 3.
The principal benefit of a model is that just a few numbers or words have to be
dealt with instead of a profusion. If the model is a good one, it can be used as a
summary, for spotting exceptions, for making comparisons and for taking action.
The vast amount of information now available to managers (particularly in view
of recent growth in the use of all sizes of computers) means that when looking for a
model or pattern one should look at the simple ones first. There is usually neither
time nor sufficient specialist help to do otherwise. It has to be a simple model or
nothing. In any case the simple models are usually the more useful since they are
easier to handle and communicate with. There are sometimes objections to simple
models purely because they are simple and, it is supposed, could not represent the
harsh complexities of the world. Yet most of the best established scientific models
have been simple: Newton’s Laws and Boyle’s Law, for instance. They may not

Quantitative Methods Edinburgh Business School 4/7

Module 4 / Data Analysis

encapsulate every nuance of reality, but for some time they have been able to
summarise and predict reality to a generally acceptable level of approximation. In
management the objective is usually no more than this.
Only if the simple approach fails are complex methods necessary, and then ex-
pert knowledge may be required. As a last resort, even if the numbers are random
(random means they have no particular pattern or order), this is a model of a sort
and can be useful. For example, the fact that the day-to-day movements in the
returns from quoted shares are random is an important part of modern financial
theory.

4.3.4 Stage 4: Exceptions

Once a pattern is established, it is feasible to look at the exceptions to it. In manage-
rial situations the exceptions can be more important than the pattern. For the
marketing divisions of a company, the latest monthly sales figures may be seen to
have a pattern in which the sales volume per salesperson is approximately the same
in every sales region except for region X, where the figure is lower. The exception
can now be investigated. Did it occur because it is a large region geographically and
thus more time is taken up travelling? Is it because someone is falling down on the
job? Is it on account of reasons within or beyond the control of the company? One
reason for an extreme exception is that someone made a mistake with the numbers,
(e.g. a typist omitted a zero). This last possibility should always be considered.
When the exceptions have been noticed, they can be corrected or regretfully
disregarded (in the case of items beyond the company’s control) or management
action can be taken. However the exceptions are investigated and dealt with, it is the
recognition of a pattern or model in the numbers that makes it possible to see their
existence.
It is a common error to deal with exceptions before discerning the pattern. This
is the basis of the ‘number picking’ mentioned earlier. Not only is this illogical (since
an exception must by definition have something to be an exception from), but it
also leads to true exceptions being missed while spurious ones are discussed at
length and in depth.
A further possibility may arise. The exceptions may be so numerous and inexpli-
cable that the only conclusion has to be that the model chosen for the data is not
good enough. It does not explain the fluctuations in the numbers sufficiently well to
be regarded as a general model. The exceptions disprove rather than prove the rule.
In this situation the only option is to return to the data and attempt, in the light of
knowledge of the inadequacies of the failed model, to find a new model which
better represents the data. Several iterations of this process may take place before a
satisfactory model is found.

4.3.5 Stage 5: Comparisons

Having detected the pattern and explained or corrected the exceptions to it, one
may compare the results with other relevant information. Rarely is any analysis done
completely in isolation. Nearly always, there are other results with which to make a

4/8 Edinburgh Business School Quantitative Methods

 Module 4 / Data Analysis

comparison. The other results may be from another year, from another company,
from another country or from another analyst. In other words, reference can usually
be made to a wider set of information. In consequence, questions may be prompted:
Why is the sales mix different this year from the previous five? Why do other
companies have less brand switching for their products? Why is productivity higher
in the west of Germany? Making comparisons such as these provides a context in
which to evaluate results and also suggests the consistencies or anomalies which
may in turn lead to appropriate management action.
If the results coincide with others, then this further establishes the model and
may mean that in future fewer data may need to be collected – only enough to see
whether the already established model still holds. This is especially true of manage-
ment information systems where managers receive regular printouts of sets of
numbers and they are looking for changes from what has gone before. It is more
efficient for a manager to carry an established model from one time period to the
next rather than the raw data.
Example: Consumption of Distilled Spirits in the USA
As an example of an analysis of numbers that a manager might have to carry out,
consider Table 4.3 showing the consumption of distilled spirits in different states of the
USA. The objective of the analysis would be to measure the variation in consumption
across the states and to detect any areas where there were distinct differences. How
can the table be analysed and what information can be gleaned from it? The five stages
of the guidelines are followed.
Stage 1: Reduce the data. Many of the data are redundant. Are percentage fig-
ures really necessary when per capita figures are given? It is certainly possible, with
some imaginative effort, to conceive of uses of percentage data, but they are not
central to the purposes of the table. It can be reduced to a fraction of its original
size without any loss of real information.
Stage 2: Re-present. To understand the table more quickly, the numbers can be
rounded to two effective figures. The original table has numbers, in places, to eight
figures. No analyst could possibly make use of this level of specification. What con-
clusion would be affected if an eighth figure were, say, a 7 instead of a 4? In any
event, the data are not accurate to eight figures. If the table were a record docu-
ment (which it is not) then more than two figures may be required, but not eight.
Putting the states in order of decreasing population is more helpful than alphabetical
order. Alphabetical order is useful for finding names in a long list, but it adds nothing
to the analysis process. The new order means that states are just as easy to find.
Most people will know that California has a large population and Alaska a small one,
especially since no one using the table will be totally ignorant of the demographic
attributes of the USA. At the same time, the new order makes it easy to spot states
whose consumption is out of line with their population.
The end result of these changes, together with some of a more cosmetic nature, is
Table 4.4. Contrast this table with the original, Table 4.3.

Quantitative Methods Edinburgh Business School 4/9

Module 4 / Data Analysis

Table 4.3 Consumption of distilled spirits in the USA

Per cent to US
Rank in Consumption in Total
consumption wine gallons Percent consumption Per Per
Licence states increase capita capita
2015 2014 2015 2014 decrease 2015 2014 2015 2014

Alaska 46 47 1 391 172 1 359 422 2.3 0.33 0.32 3.64 3.86
Arizona 29 30 4 401 883 4 144 521 6.2 1.03 0.98 1.94 1.86
Arkansas 38 38 2 534 826 2 366 429 7.1 0.60 0.56 1.20 1.12
California 1 1 52 529 142 52 054 429 0.9 12.33 12.32 2.44 2.46
Colorado 22 22 6 380 783 6 310 566 1.1 1.50 1.49 2.47 2.49

Connecticut 18 18 7 194 684 7 271 320 (−1.1) 1.69 1.72 2.31 2.35
Delaware 45 43 1 491 652 1 531 688 (−2.6) 0.35 0.36 2.56 2.65
Dist. of 27 27 4 591 448 4 828 422 (−4.9) 1.08 1.14 6.54 6.74
Columbia
Florida 4 4 22 709 209 22 239 555 1.7 5.33 5.28 2.70 2.67
Georgia 13 13 10 717 681 9 944 846 7.8 2.52 2.35 2.16 2.02

Hawaii 41 40 2 023 730 1 970 089 2.7 0.48 0.47 2.28 2.28
Illinois 3 3 26 111 587 26 825 876 (−2.7) 6.13 6.35 2.33 2.41
Indiana 19 20 7 110 382 7 005 511 1.5 1.67 1.66 1.34 1.32
Kansas 35 35 2 913 422 2 935 121 (−0.7) 0.68 0.70 1.26 1.29
Kentucky 26 26 4 857 094 5 006 481 (−3.0) 1.14 1.19 1.42 1.47

Louisiana 21 21 7 073 283 6 699 853 5.6 1.66 1.59 1.84 1.77
Maryland 12 12 10 833 966 10 738 731 0.9 2.54 2.54 2.61 2.62
Massachusetts 10 10 13 950 268 14 272 695 (−2.3) 3.28 3.38 2.40 2.45
Minnesota 15 15 8 528 284 8 425 567 1.2 2.00 1.99 2.15 2.15
Missouri 20 17 7 074 614 7 697 871 (−7.9) 1.66 1.82 1.48 1.61

Nebraska 36 36 2 733 497 2 717 859 0.6 0.64 0.64 1.76 1.76
Nevada 30 31 4 360 172 4 095 910 6.5 1.02 0.97 7.15 6.92
New Jersey 8 8 15 901 587 16 154 975 (−1.6) 3.73 3.82 2.17 2.21
New Mexico 42 41 1 980 372 1 954 139 1.3 0.47 0.46 1.70 1.70
New York 2 2 41 070 005 41 740 341 (−1.6) 9.64 9.88 2.27 2.30

North Dakota 47 46 1 388 475 1 384 311 0.3 0.33 0.33 2.16 2.16
Oklahoma 33 29 3 904 574 4 187 527 (−6.8) 0.92 0.99 1.41 1.54
Rhode Island 39 39 2 073 075 2 131 329 (−2.7) 0.49 .50 2.24 2.30
South 23 25 5 934 427 5 301 054 11.9 1.39 1.26 2.08 1.88
Carolina
South Dakota 48 48 1 312 160 1 242 021 5.6 0.31 0.29 1.91 1.82

Tennessee 24 24 5 618 774 5 357 160 4.9 1.32 1.27 1.33 1.28
Texas 5 6 17 990 532 17 167 560 4.8 4.22 4.06 1.44 1.40
Wisconsin 11 11 10 896 455 10 739 261 1.5 2.56 2.54 2.36 2.33

Total licence 319 583 215 317 874 435 0.5 75.04 75.22 2.13 2.13

4/10 Edinburgh Business School Quantitative Methods

 Module 4 / Data Analysis

Table 4.4 Consumption of distilled spirits in the USA (amended)

Licence states Consumption
(in order of population) Wine gallons (millions) Per capita (gallons)
2015 2014 2014
California 53.0 52.0 2.4
New York 41.0 42.0 2.3
Texas 18.0 17.0 1.4
Illinois 26.0 27.0 2.3
Florida 23.0 22.0 2.7

New Jersey 16.0 16.0 2.2

Massachusetts 14.0 14.0 2.4
Indiana 7.1 7.0 1.3
Georgia 11.0 9.9 2.2
Missouri 7.1 7.7 1.5

Wisconsin 11.0 11.0 2.4

Tennessee 5.6 5.4 1.3
Maryland 11.0 11.0 2.6
Minnesota 8.5 8.4 2.2
Louisiana 7.1 6.7 1.8

Kentucky 4.9 5.0 1.4

Connecticut 7.2 7.3 2.3
S. Carolina 5.9 5.3 2.1
Oklahoma 3.9 4.2 1.4
Colorado 6.4 6.3 2.5

Kansas 2.9 2.9 1.3

Arizona 4.4 4.1 1.9
Arkansas 2.5 2.4 1.2
Nebraska 2.7 2.7 1.8
New Mexico 2.0 2.0 1.7

Rhode Island 2.1 2.1 2.2

Hawaii 2.0 2.0 2.3
D. Columbia 4.6 4.8 6.5
S. Dakota 1.3 1.2 1.9
N. Dakota 1.4 1.4 2.2

Nevada 4.4 4.1 7.2

Delaware 1.5 1.5 2.6
Alaska 1.4 1.4 3.6
Average 9.7 9.6 2.3

Quantitative Methods Edinburgh Business School 4/11

Module 4 / Data Analysis

Stage 3: Build a model. The pattern is evident from the transformed table. Con-
sumption varies with the population of the state. Per capita consumption in each
state is about equal to the figure for all licence states with some variation (±30 per
cent) about this level. The pattern a year earlier was the same except that overall
consumption increased slightly (1 per cent) between the two years. Refer back to
Table 4.3 and see if this pattern is evident even when it is known to be there. There
may of course be other patterns but this one is central to the objectives of the anal-
ysis.
Stage 4: Exceptions. The overall pattern of approximately equal per capita con-
sumption in each state allows the exceptions to be seen. From Table 4.4, three
states stand out as having a large deviation from the pattern. The states are District
of Columbia, Nevada and Alaska. These states were exceptions to the pattern in the
earlier year as well. Explanations in the cases of District of Columbia and Nevada are
readily found, probably being to do with the large non-resident populations. People
live, and drink, in these states who are not included in the population figures (diplo-
mats in DC, tourists in Nevada). An explanation for Alaska may be to do with the
lack of leisure opportunities. Whatever the explanations, the analytical method has
done its job. The patterns and exceptions in the data have been found. Explanations
are the responsibility of experts in the marketing of distilled spirits in the USA.
Stage 5: Comparison. A comparison between the two years is provided by the
table. Other comparisons will be relevant to the task of gaining an understanding of
the USA spirits market. The following data would be useful:
(i) earlier years, say, five and ten years before;
(ii) a breakdown of aggregate data into whisky, gin, vodka, etc.;
(iii) other alcoholic beverages: wine, beer, etc.
Once data from these other sources have been collected they would be analysed in
the manner described, but of course the process would be shorter because the
pattern can be anticipated. Care would need to be taken that like was being com-
pared with like. For example, it would have to be checked that an equivalent
definition of consumption was in force ten years earlier.

4.3.6 Implications for the Producers of Statistics

The methodology is intended for use by managers who are not statistical specialists.
However, if the problems that the methodology is intended to counter are real and
if it does indeed prove useful in overcoming them, there are implications for the
producers of data as well as for the users.
The obvious implication for producers of statistics is that more attention should
be paid to the uses to which the data are being put. This does not mean asking
managers what they want since this will just lead to an over-provision of data.
Likewise, the practice of issuing all the data that are held (albeit broken down into
finance, marketing, etc.) results in a surfeit of data which merely confuses the
aspiring analyst. Designers of information systems, writers of reports and the like
should consider exactly what contribution their data are supposed to make and how
they are expected to improve decision making. What is needed is a more systematic
approach whereby data are provided not in isolation but in the context of the
management processes they are intended to serve.

4/12 Edinburgh Business School Quantitative Methods

 Module 4 / Data Analysis

The second implication is more direct. Data should be presented in forms which
enable them to be analysed speedily and accurately. Much of the reduction and re-
presentation stages of the guidelines could, in most instances, be carried out just as
well by the producer of the data as by the user. It would then need to be done only
once rather than many times by the many users of the data. Unfortunately, when
time is spent thinking about the presentation of statistics, it is usually spent in
making the tables look neat or attractive rather than making them amenable to
analysis.
There is much that the producers of data can do by themselves. For example,
refer back to the extract from a management information system shown in Ta-
ble 4.1: if thought were given to the analysis of the data through the application of
the guidelines, a different presentation would result (see Table 4.5).
(a) Some data might be eliminated. For instance, full details on minor categories of
expenditure may not be necessary. This step has not been taken in Table 4.5
since full consultation with the receivers would be necessary.
(b) The table should be re-presented using the rules of data presentation. In
particular, some rounding is helpful. This is an information document, not an
auditing one, and thus rounding is appropriate. In any case, no different conclu-
sions would be drawn if any of the expenditures were changed by one unit. In
addition, improvement is brought about by use of summary measures and a
clearer distinction between such measures and the detail of the table.
(c) A model derived from previous time periods would indicate when changes were
taking place. There is a good case for including a model or summary of previous
time periods with all MIS data. This has not been done for Table 4.5 since previ-
ous data were not available.
(d) Exceptions can be clearly marked. It is, after all, a prime purpose of budget data
to indicate where there have been deviations from plan. This can be an automat-
ic process. For example, all variances greater than 10 per cent could be marked.
This might even obviate the need for variance figures to be shown.
(e) The making of comparisons is probably not the role of the data producer in this
example, involving as it does the judgement of the receivers in knowing what the
relevant comparisons are. The task of the producer has been to facilitate these
comparisons.
Making the suggested changes does of course have a cost attached in terms of
management time. However, the cost is a small fraction of the cost of setting up and
operating the information system. The changes can transform the system and make
it fully operational. If an existing system is being largely ignored by managers, there
may be no alternative.

Quantitative Methods Edinburgh Business School 4/13

Module 4 / Data Analysis

Table 4.5 Budgeting data from an MIS (amended from Table 4.1)
Port: Liverpool OCEAN PORT TERMINAL COSTS – SHIPSIDE OPERATIONS
Period: December (U.S Dollars: Conversion rate 1.60)

MONTH CUMULATIVE
ESTIMATE STANDARD VARIANCE VAR % ESTIMATE STANDARD VARIANCE VAR % BUDGET

LO-LO: Stevedoring (STR-FULL) 131 000 144 000 12 000 9 1 240 000 1 360 000 120 000 9 1 560 000
Stevedoring (STR-MT) 13 400 14 700 1 300 9 257 000 281 000 24 000 9
Unlashing 78 0 −78 * 78 0 −78 *
Shifting 200 0 −200 * 12 000 0 −12 000 *
Overtime, etc. 7 100 0 −7 100 * 191 000 0 −191 000 *

RO-RO: Stevedoring TR (STR-FULL) 20 400 26 100 5 800 22 167 000 215 000 48 000 22 330 000
Stevedoring TR (STR-MT) 180 230 50 22 15 000 19 000 4 100 22
Stevedoring cost plus 0 0 0 0 0 0 0 0
Stevedoring Volvo 0 0 0 0 0 0 0
Stevedoring rolling 14 300 19 500 5 200 27 98 000 157 000 59 000 37
Stevedoring blockstow 29 27 −2 −7 610 670 60 9
Unlashing 350 0 −350 * 350 0 −350 *
Shifting 980 0 −980 * 3 800 0 −3 800 *
Overtime, etc. 1 400 0 −1 400 * 29 000 0 −29 000 *
Heavy lifts 0 0 0 0 2 000 0 −2 000 *

CARS: Stevedoring (STR) 6 100 6 400 280 4 38 000 35 000 −3 000 −9 170 000
Unlashing 0 0 0 0 2 0 −2 *
Shifting 800 0 −800 * 1 300 0 −1 300 *
Overtime, etc. 0 0 0 0 7 600 0 −7 600 *

OTHER: Shipside costs 3 400 0 −3 400 * 24 000 0 −24 000 *

TOTALS: LO-LO 152 000 158 000 5 900 4 1 700 000 1 640 000 −59 000 4
RO-RO 38 000 46 000 8 300 18 315 000 392 000 77 000 20
CARS 6 900 6 4001 −520 −8 47 000 35 000 −12 000 −34
OTHER 3 400 0 −3 400 * 24 000 0 −24 000

GRAND 201 000 211 000 10 000 5 2 080 000 2 070 000 −14 000 −1 2 060 000
TOTAL
Totals may not agree because of rounding.
*Zero standard cost, therefore variance not calculable.

4/14 Edinburgh Business School Quantitative Methods

 Module 4 / Data Analysis

Few managers will not admit that there is currently a problem with the provision
and analysis of data, but they rarely tell this to their IT systems manager or whoever
sends them data. Without feedback, inexpensive yet effective changes are never
made. It must be the responsibility of users to criticise constructively the form and
content of the data they receive. The idea that computer scientists/statisticians
always know best, and if they bother to provide data then they must be useful, is
false. The users must make clear their requirements, and even resort to a little
persistence if alterations are not forthcoming.

Learning Summary
Every manager sees the problem of handling numbers differently because each sees
it mainly in the (probably) narrow context with which he or she is familiar in his or
her own work. One manager sees numbers only in the financial area; another sees
them only in production management. The guidelines suggested here are intended
to be generally applicable to the analysis of business data in many different situa-
tions and with a range of different requirements. The key points are:
(a) In most situations managers without statistical backgrounds can carry out
satisfactory analyses themselves.
(b) Simple methods are preferable to complex ones.
(c) Visual inspection of well-arranged data can play a role in coming to understand
them.
(d) Data analysis is like verbal analysis.
(e) The guidelines merely make explicit what comes naturally when dealing with
words.
The need for better skills to turn data into real information in managerial situa-
tions is not new. What has made the need so urgent in recent times is the
exceedingly rapid development of computers and associated management infor-
mation systems. The ability to provide vast amounts of data very quickly has grown
enormously. It has far outstripped the ability of management to make use of the
data. The result has been that in many organisations managers have been swamped
with so-called information which in fact is no more than mere numbers. The
problem of general data analysis is no longer a small one that can be ignored. When
companies are spending large amounts of money on data provision, the question of
how to turn the data into information and use them in decision making is one that
has to be faced.

Quantitative Methods Edinburgh Business School 4/15

Module 4 / Data Analysis

Review Questions
4.1 Traditional statistical techniques do not help managers in analysing data. True or false?

4.2 The need for new management skills in data analysis arises because so many data come
from computers, which means that they have to be presented in a more complicated
style. True or false?

4.3 Which of the following reasons is correct? The first step in data analysis is to reduce the
data. It is done because:
A. Most data sets contain some inaccuracies.
B. One can only analyse a small amount of data at a time.
C. Most data sets contain some items which are irrelevant.

4.4 Data recorded to eight decimal places can be rounded down since such a degree of
accuracy will not affect the decision being taken. True or false?
A. True
B. False

4.5 Which of the following reasons are correct? A model or pattern is used to summarise a
table because:
A. Exceptions can be seen more easily and accurately.
B. It is easier to make comparisons with other sets of data.
C. The model will be more accurate than the original data.

4.6 Which statement below best describes the following data?

Year Sales (£m)

2013 3.2
2014 4.0
2015 5.0
2016 6.2

A. Growth of about £1 million p.a.

B. Growth of about 25 per cent p.a.
C. 2016 sales were the highest.
D. Average sales = 4.6 million.

4/16 Edinburgh Business School Quantitative Methods

 Module 4 / Data Analysis

4.7 A company has four divisions. The profit and capital employed by each are given in the
table below. Which division is the exception?

Profit Capital employed

Division 1 4.8 80.3
Division 2 7.2 191.4
Division 3 3.6 59.4
Division 4 14.5 242.0

A. Division 1
B. Division 2
C. Division 3
D. Division 4

4.8 A confectionery manufacturer’s production level for a new chocolate bar is believed to
have increased by 5 per cent per month over the last 36 months. However, for 11 of
these months this model does not fit. The exceptions were as follows: for five months
strikes considerably reduced production; the three Decembers had lower figures, as did
the three Augusts, when the production plant is closed for two weeks. You would be
right in concluding that the 5 per cent model is not a good one because 11 exceptions
out of 36 is too many. True or false?

4.9 Towards the completion of an analysis of the consumption of distilled spirits across the
different states of the USA in a particular year, the results are compared with those of
similar studies. Which of the following other analyses would be useful?
A. Consumption of distilled spirits across the départements of France.
B. Consumption of wine across the départements of France.
C. Consumption of wine across the states of the USA.
D. Consumption of whisky across the states of the USA.

4.10 A simple model is used in preference to a sophisticated one in the analysis of data
because sophisticated models obscure patterns. True or false?

Case Study 4.1: Motoring Correspondent

1 This statement was made in a British national newspaper by its motoring correspondent
in February 2015:

Ten years ago, in 2004, the death rate on the roads of this country was running
at 0.1 death for every 1 million miles driven. By 2009 a death occurred every
12 million miles. Last year, according to figures just released, there were 6400
deaths, whilst a total of 92 000 million miles were driven.

a. Analyse these data.

b. Was driving safer in 2014 than in 2004?
c. Make a forecast of the death rate in 2018.
d. What reservations are there regarding the prediction?

Quantitative Methods Edinburgh Business School 4/17

Module 4 / Data Analysis

Case Study 4.2: Geographical Accounts

1 Table 4.6, taken from the annual accounts of a major UK company, shows the
expenditure by geographical divisions, broken down into categories. For instance, the
Wessex region spent £48 545 000 in all, split into £573 000 on finance, £13 224 000 on
raw materials, etc. Analyse the table, showing particularly, for each expenditure
category, the regions that have unusual expenditure levels.

Table 4.6 Expenditure by geographical division (£000 (%))

Total Finance Raw Manpower Transport Fuel
materials
England and Wales 1 109 896 34 406 447 161 249 820 318 430 8 968
(100.0) (3.1) (40.3) (22.5) (28.7) (0.8)
Divisions
North West 149 831 1 105 75 517 28 050 38 679 1 112
(100.0) (0.7) (50.4) (18.7) (25.8) (0.7)

Northumbria 39 426 121 18 406 9 963 9 689 346

(100.0) (0.3) (46.7) (25.3) (24.6) (0.9)

Severn-Trent 187 005 8 401 68 861 39 136 62 301 11 732

(100.0) (4.5) (36.8) (20.9) (33.3) (0.9)

Yorkshire 109 607 1 270 51 462 18 537 33 916 969

(100.0) (1.2) (47.0) (16.9) (30.9) (0.9)

Anglia 137 898 4 184 47 721 33 487 41 157 811

(100.0) (3.0) (34.6) (24.3) (29.8) (0.6)

Thames 227 745 8 055 87 221 58 572 62 573 1 671

(100.0) (3.5) (38.3) (25.7) (27.5) (0.7)

South 76 539 1 216 22 596 20 292 25 709 693

(100.0) (1.6) (29.5) (26.5) (33.6) (0.9)

Wessex 48 545 573 13 224 16 379 15 481 476

(100.0) (1.2) (27.2) (33.7) (31.9) (1.0)

South West 39 230 950 18 083 8 132 9 564 307

(100.0) (2.4) (46.1) (20.7) (24.4) (0.8)

Wales 94 070 8 531 44 070 17 272 19 361 851

(100.0) (9.1) (46.8) (18.4) (20.6) (0.9)
(Note: This table is only half the original. Several expenditure categories have been left out to keep it to a
manageable size. Although totals for each expenditure type are correct, the sum of expenditure for each
region is less than the total.)

4/18 Edinburgh Business School Quantitative Methods

 Module 4 / Data Analysis

Case Study 4.3: Wages Project

1 Table 4.7 was taken from the report on a research project conducted for a UK
government department. The project was intended to find out the extent to which
employers were paying wages that were lower than nationally agreed minimum wages. It
involved a sample survey of 11 trades in four geographical areas. For each of the trades,
the table shows:
a. the percentage of those employers sampled who were found to be underpaying;
b. the percentage of employees included in the survey who were being underpaid;
c. the total amount underpaid in the year of the survey in each trade by the employers
included in the sample.
Analyse the data and suggest what the main conclusions might be. If no general pattern
exists, suggest what extra data are required to further the analysis.

Table 4.7 Wages underpayment

Trades Proportion of Proportion of Amount
employers employees’ underpaid
inspected who wages examined
were found to that were
be underpaying entitled to
arrears
Retail bread 32.99 23.22 2 516.01
Bookselling 50.00 20.07 1 205.48
Drapery and outfitting 31.85 17.20 11 922.26
Retail food 21.74 11.89 16 850.33
Furnishing and allied trades 20.19 7.86 14 085.62
Newsagency and tobacco 32.93 21.63 9 564.66
Hairdressing 26.23 11.30 4 394.04
Licensed non-residential 22.01 10.24 6 218.02
hotels
Licensed restaurants 30.77 8.79 3 939.20
Unlicensed restaurants 47.58 32.78 5 315.83
Other trades 17.86 8.20 157.30
Total 27.08 13.91 76 168.75

References
Ehrenberg, A. S. C. (1975). Data Reduction. New York: John Wiley and Sons.

Quantitative Methods Edinburgh Business School 4/19

Module 5

Summary Measures
Contents
5.1 Introduction.............................................................................................5/1
5.2 Usefulness of the Measures ....................................................................5/2
5.3 Measures of Location..............................................................................5/5
5.4 Measures of Scatter ............................................................................. 5/14
5.5 Other Summary Measures ................................................................. 5/20
5.6 Dealing with Outliers .......................................................................... 5/21
5.7 Indices ................................................................................................... 5/22
Learning Summary ......................................................................................... 5/29
Review Questions ........................................................................................... 5/30
Case Study 5.1: Light Bulb Testing............................................................... 5/33
Case Study 5.2: Smith’s Expense Account .................................................. 5/34
Case Study 5.3: Monthly Employment Statistics ........................................ 5/34
Case Study 5.4: Commuting Distances ........................................................ 5/34
Case Study 5.5: Petroleum Products ........................................................... 5/35

Prerequisite reading: None

Learning Objectives
By the end of the module, the reader should know how large quantities of numbers
can be reduced to a few simple summary measures that are much easier to handle
than the raw data. The most common measures are those of location and scatter.
The special case of summarising time series data with indices is also described.

5.1 Introduction
When trying to understand and remember the important parts of a lengthy verbal
report, it is usual to summarise. This may be done by expressing the essence of the
report in perhaps a few sentences, by underlining key phrases or by listing the main
subsection headings. Each individual has his own method, which may be physical (a
written précis) or mental (some way of registering the main facts in the mind).
Whatever the method, the point is that it is easier to handle information in this way,
by summarising and storing these brief summaries in one’s memory. On the few
occasions that details are required, it is necessary to turn to the report itself.
The situation is no different when it is numerical rather than verbal information
that is being handled. It is still better to form a summary to capture the salient
characteristics. The summary may be a pattern, simple or complex, revealed when

Quantitative Methods Edinburgh Business School 5/1

Module 5 / Summary Measures

analysing the data, or it may be based on one or more of the standard summary
measures described in this module.
Inevitably some accuracy is lost. In the extreme, if the summarising is badly done,
it can be wholly misleading. (How often do report writers claim to have been totally
misunderstood after hearing someone else’s summary of their work?) Take the case
of a recent labour strike in the UK about wages payment. In reporting the current
levels of payment, newspapers/union leaders/employers could not, of course, state
the payments for all 173 000 employees in the industry. They had to summarise.
Five statements as to the ‘average’ weekly wage were made:
The average weekly wage is …

Quote 1: £241 (union leader)

Quote 2: £248 (newspaper)
Quote 3: £271 (newspaper)
Quote 4: £298 (newspaper)
Quote 5: £323 (employers’ organisation)

All these quoted wages were said to be the same thing: the average weekly wage.
Are the employees in the industry grossly underpaid or overpaid? It is not difficult
to choose an amount that reinforces one’s prejudices. The discrepancies are not
because of miscalculations but because of definitions. Quote 1 is the basic weekly
wage without overtime, shift allowances and unsocial hours allowance, and it has
been reduced for tax and other deductions. Since the industry is one that requires
substantial night-time working for all employees, no one actually takes home the
amount quoted. Quote 2 is the same as the first but without the tax deduction.
Quote 3 is the average take-home pay of a sample of 30 employees in a particular
area. Quote 4 is basic pay plus unsocial hours allowance but without any overtime
or tax deductions. Quote 5 is basic pay plus allowances plus maximum overtime
pay, without tax and other deductions.
It is important when using summary measures (and in all of statistics) to apply
common sense and not be intimidated by complex calculations. Just because
something that sounds statistical is quoted (‘the average is £41.83’) does not mean
that its accuracy and validity should be accepted without question. When summary
measures fail, it is usually not because of poor arithmetic or poor statistical
knowledge but because common sense has been lacking.
In this context, the remainder of this module goes on to describe ways of sum-
marising numbers and to discuss their effectiveness and their limitations.

5.2 Usefulness of the Measures

There are several types of summary measure. Each type of measure summarises a
different aspect of the data. For management purposes, it is usually possible to
summarise adequately a set of data using just two or three types of measure. For
example, the managers of the assembly line for a certain make of car receive

5/2 Edinburgh Business School Quantitative Methods

 Module 5 / Summary Measures

monthly a computer report of the two previous months’ production. The report for
June (22 working days) and May (19 working days) is given in Table 5.1.

Table 5.1 Car assembly data

Monthly production report
This month: June Previous month: May
Date Production Date Production
Mon 1 235
Tues 2 240
Wed 3 210
Thurs 4 240
Fri 5 225 Fri 1 238
Weekly total 1150 238
Mon 8 238 Mon 4 0
Tues 9 247 Tues 5 248
Wed 10 242 Wed 6 253
Thurs 11 241 Thurs 7 256
Fri 12 228 Fri 8 242
Weekly total 1196 999
Mon 15 226 Mon 11 245
Tues 16 231 Tues 12 242
Wed 17 233 Wed 13 238
Thurs 18 220 Thurs 14 247
Fri 19 215 Fri 15 239
Weekly total 1125 1211
Mon 22 230 Mon 18 249
Tues 23 234 Tues 19 244
Wed 24 225 Wed 20 241
Thurs 25 225 Thurs 21 247
Fri 26 220 Fri 22 236
Weekly total 1134 1217
Mon 29 234 Mon 25 0
Tues 30 220 Tues 26 252
Wed 27 246
Thurs 28 241
Fri 29 235
Weekly total 454 974
MONTHLY TOTAL 5059 4639

The data as shown are useful for reference purposes (e.g. what was the produc-
tion on 15 May?) or for the background to a detailed analysis (e.g. is production
always lower on a Friday and, if so, by how much?). Both these types of use revolve
around the need for detail. For more general information purposes (e.g. was May a
good month for production? What is the trend of production this year?) the amount

Quantitative Methods Edinburgh Business School 5/3

Module 5 / Summary Measures

of data contained in the table is too large and unwieldy for the manager to be able to
make the necessary comparisons. It would be rather difficult to gauge the trend of
production levels so far this year, from six reports, one for each month, such as that
in Table 5.1. If summary measures were provided, then most questions, apart from
the ones that require detail, could be answered readily. A summary of Table 5.1
might be as shown in Table 5.2.
The summary information provided in Table 5.2 enables a wide variety of man-
agement questions to be answered and, more importantly, answered quickly.
Comparisons are made much more easily if summary data for several months, or
years, are available on one report.
Three types of summary measure are used in Table 5.2. The first, average pro-
duction, measures the location of the numbers and tells at what general level the
data are. The second, the range of production, measures scatter and indicates how
widely spread the data are. The third indicates the shape of the data. In this case,
the answer ‘symmetrical’ says that the data fall equally on either side of the average.

Table 5.2 Summary of May and June production

Summary Production Report
June May
Average production/day* 230 244
Range of daily production* 210–247 235–256
Shape of distribution Symmetrical Symmetrical
* Excluding holidays, when there was no production.

The three measures reflect the important attributes of the data. No important
general features of the data are omitted. If, on the other hand, the measure of scatter
had been omitted, the two months could have appeared similar. In actual fact, their
very different ranges provide an important piece of information that reflects
production planning problems.
For each type of measure (location, scatter, shape) there is a choice of measure to
use (for location, the choice is between arithmetic mean and other measures). The
different types of measures are described below. The measures have many uses
other than as summaries and these will be indicated. They will also be found in
other subject areas. For example, the variance, a measure of scatter, plays a central
role in modern financial theory.

5/4 Edinburgh Business School Quantitative Methods

 Module 5 / Summary Measures

5.3 Measures of Location

Measures of location are also referred to as measures of central tendency. Their
purpose is to show, in general terms, the size of the data in question.
5.3.1 Arithmetic Mean
The most common and useful measure is the well-known arithmetic mean. It is
defined as:
Arithmetic mean =

This can be put in shorthand (or mathematical) notation:

∑
=

where:
refers to the data in the set
is standard notation for the arithmetic mean
∑ is the Greek capital sigma and, mathematically, means ‘sum of’
is standard notation for the numbers of readings in the set.

For example, for the nine numbers 3, 3, 4, 5, 5, 6, 6, 6, 7:

=
=
Arithmetic mean = 5
5.3.2 Median
The median is the middle value of a set of numbers. There is no mathematical
formula for calculating it. It is obtained by listing the numbers in ascending order,
and the median is that number at the halfway point.
For example, using the same nine numbers as above which are already in ascend-
ing order:

3, 3, 4, 5, 5, 6, 6, 6, 7
↑
Middle number

Median = 5
If there is an even number of readings, then there can be no one middle number.
In this case, it is usual to take the arithmetic mean of the middle two numbers as the
median.
For example, if the set of nine numbers above was increased to ten by the pres-
ence of ‘8’, the set would become:

Quantitative Methods Edinburgh Business School 5/5

Module 5 / Summary Measures

3, 3, 4, 5, 5, 6, 6, 6, 7, 8
︸
Middle two numbers

Median = (5+6)/2
Median = 5.5

5.3.3 Mode
The third measure of location is the mode. This is the most frequently occurring
value. Again, there is no mathematical formula for the mode. The frequency with
which each value occurs is noted and the value with the highest frequency is the
mode.
Again, using the same nine numbers as an example: 3, 3, 4, 5, 5, 6, 6, 6, 7

Number Frequency
3 2
4 1
5 2
6 3
7 1
Mode = 6

5.3.4 Calculating Measures of Location

A set of data might simply be recorded as a list of numbers in a spreadsheet.
Calculating measures of location then involves the easy task of highlighting the
numbers and applying one of the standard spreadsheet functions. On other occa-
sions a set of data may be assembled and displayed in the form of a frequency table.
For example, the data used above to illustrate measures of location were in the form
of a frequency table when the mode was calculated. For large sets of data the
frequency table may show the data in groups. For example, Table 5.3 refers to the
number of customer complaints received daily by a railway company over a one-year
(350 days) period. It shows the number of days on which there were fewer than ten
complaints, the number of days when there were between ten and 19 complaints
and so on.

Table 5.3 Railway complaints

No. of complaints No. of days
0 to 9 24
10 to 19 33
20 to 29 68
30 to 39 54

5/6 Edinburgh Business School Quantitative Methods

 Module 5 / Summary Measures

No. of complaints No. of days

40 to 49 53
50 to 59 42
60 to 69 31
70 to 79 19
80 to 89 10
90+ 16
Total 350

The calculation of measures of location for grouped or classified data such as

those in Table 5.3 is based on the principle of representing each class of data by its
mid-point. So, for example, all 53 observations in the 40–49 class are treated as if
they were equal to the mid-point, 44.5. The calculation of the arithmetic mean then
proceeds as in Table 5.4.

Table 5.4 Railway complaints

No. of complaints Mid-point No. of days Frequency × Mid-point
0 to 9 4.5 24 108.0
10 to 19 14.5 33 478.5
20 to 29 24.5 68 1666.0
30 to 39 34.5 54 1863.0
40 to 49 44.5 53 2358.5
50 to 59 54.5 42 2289.0
60 to 69 64.5 31 1999.5
70 to 79 74.5 19 1415.5
80 to 89 84.5 10 845.0
90+ 94.5 16 1512.0
Total 350 14 535.0
Arithmetic mean = ∑(Frequency × Mid-point) for all classes/350
= 14 535/350
= 41.53

Treating all data in a class as if each observation were equal to the mid-point is of
course an approximation, but it is done to simplify the calculations. However, on
some occasions the data may only be available in this form anyway. For example, in
measuring the lengths of machined car components as part of a quality check, the
observations would probably be recorded in groups such as ‘100.5 to 101.0’ rather
than as individual measurements such as ‘100.634’. The most serious approximation
in Table 5.4 is in taking the mid-point of the 90+ class as 94.5, since this class could
include days when complaints had been much higher, say 150, because of some
special circumstances such as severe disruption on account of a derailment. For

Quantitative Methods Edinburgh Business School 5/7

Module 5 / Summary Measures

open-ended groups such as this it may be necessary to examine the outliers to test
the validity of the mid-point approximation.
Calculating the mode and median from a frequency table is more straightforward.
The median class is the one in which the middle observation lies. In this case the
175th and 176th observations lie in the 30–39 class (i.e. the median is 34.5). The
mode is the mid-point of the class with the highest frequency. In this case the class
is 20–29 and the mode is therefore 24.5.

5.3.5 Choosing the Measure of Location

Figure 5.1, Figure 5.2 and Figure 5.3 show three different sets of data generated
from three different situations. Each has a different shape of histogram. Check your
understanding of their definitions by calculating mean, median and mode for each
set and decide which measure of location is the most representative for each set.

Context: Marks out of 15 scored by each of 20 participants in a driving competition.

A symmetrical distribution.
Readings: 5, 5, 6, 6, 7, 7, 7, 7, 8, 8, 8, 8, 8, 9, 9, 9, 10, 10, 11, 12
Shape:

4
No. of readings

1 2 3 4 5 6 7 8 9 10 11 12
Marks

Figure 5.1 Driving competition results

Table 5.5 lists the results. Note that the arithmetic mean of all three distributions
is the same: 8.0. However, the data are very different in the three cases, indicating a
need for care in summarising.

Table 5.5 Symmetrical, U shape and Reverse J distributions

Readings Mean Median Mode
Fig. 5.1: Symmetrical 8.0 8 8
Fig. 5.2: U shape 8.0 2 0 and 19
Fig. 5.3: Reverse J 8.0 1 0

For the symmetrical distribution (Figure 5.1) all three measures are equal. This is
always approximately the case for symmetrical data. Whichever measure is chosen, a

5/8 Edinburgh Business School Quantitative Methods

 Module 5 / Summary Measures

similar answer results. Consequently, it is best to use the most well-known measure
(i.e. the arithmetic mean) to summarise location for this set of data.
Calculations for Figure 5.1:

Mean =
=
= 8

Median = middle value of set

= average of 10th and 11th values
= 5, 5, 6, 6, 7, 7, 7, 7, 8, 8, 8, 8, 8, 9, 9, 9, 10, 10, 11, 12
=8
Mode = most frequently occurring value
=8

Context: Number of episodes in a 19-part serial seen by a sample of 20 viewers.

A U-shaped distribution.
Readings: 0, 0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 4, 17, 18, 18, 19, 19, 19, 19, 19
Shape:
5

4
No. of readings

0 1 2 3 4 5 17 18 19
Episodes seen

Figure 5.2 Television viewing data

The U-shaped data (Figure 5.2) are typical for certain television series and also
magazine readership where many people see hardly any programmes/issues, many
people see all or most, and just a very few fall between the two extremes. Here the
arithmetic mean is not very helpful, since no one actually sees between five and 16
episodes. As a summary of the data, it would mislead. The median is not a particu-
larly good summariser either. Data in which virtually all readings are at one of two
extremes cannot easily be reduced to a single measure that is the middle value of the
set. It is also very sensitive to small changes. A serial that was only slightly more
popular could have resulted in a median of around 18. The best measure is the
mode. In this case, there are two, 0 and 19. Even when technically there is only one
mode for the data, it is usual, when the histogram has more than one definite peak,

Quantitative Methods Edinburgh Business School 5/9

Module 5 / Summary Measures

to quote more than one mode, each mode corresponding to one peak. For example,
in Figure 5.2, had the frequencies for 0 and 19 episodes been 5 and 4 respectively,
technically there would have been one mode at 0, but because the histogram still
would have two peaks, the data should be reported as having two modes.
Calculations for Figure 5.2:

Mean =
=
= 8
Median = middle value of set
= average of 10th and 11th values
= 0, 0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 4, 17, 18, 18, 19, 19, 19, 19, 19
=2
Mode = most frequently occurring values
= 0 and 19

Context: Weeks away from work through sickness in a one-year period for a sample of 20 employees in
a particular company.
Readings: 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 3, 5, 18, 28, 44, 52
Shape:
7

5
No. of readings

0
0 1 2 3 5 18 28 44 52
Weeks of sickness

Figure 5.3 Sickness records

The reverse J-shaped distribution (Figure 5.3) arises with distributions that are
truncated at one end (no values less than 0) but which have a few outliers at the
other end of the scale. It is typical of sickness records, where most workers are
absent very little or not at all, but where one or two employees with a major illness
may be absent for most of the year. The best measure is the median. It tells how
many weeks off the ‘middle’ employee had. There are major defects for the other
two measures. The arithmetic mean is distorted by the outliers. A very different

5/10 Edinburgh Business School Quantitative Methods

 Module 5 / Summary Measures

value would be obtained if the outlier of 52 weeks had not been present. Then the
mean would have been reduced from 8.0 to 5.7.
In all situations, the arithmetic mean can be misleading if there are just one or
two extremes in the data. The mode is not misleading, just unhelpful. Most sickness
records have a mode of 0, therefore to quote ‘mode 0’ is not providing any more
information than merely saying that the data concern sickness records.
Calculations for Figure 5.3:
Mean =
=
= 8
Median = middle value of set
= average of 10th and 11th values
= 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 3, 5, 18, 28, 44, 52
=1
Mode = most frequently occurring value
=0
Mean, medium and mode are the major measures of location and are obviously
useful as summaries of data. Equally obviously, they do not capture all aspects of a
set of numbers. Other types of summary measure are necessary. However, before
we leave measures of location, their uses, other than as summarisers, will be
described.

5.3.6 Other Uses of Measures of Location

As a Focus for the Eye

The importance of the visual approach in data analysis has already been stressed.
One aspect of this is the use of summary measures of location as a focus to guide
the eye and give a more speedy impression of the meaning of the data. Consider the
following two sets of data. It is difficult to see the shape or pattern in them.

Set 1: 8, 7, 5, 11, 10, 7, 8, 8, 6, 7, 10, 12, 5, 6, 7, 9, 9, 8, 8, 9

Set 2: 2, 1, 0, 18, 5, 0, 52, 2, 1, 0, 1, 44, 3, 0, 0, 28, 0, 0, 2, 1

Adding the mean of each set of data, as in the next two sets, allows the shape of
the distribution to become apparent more quickly to the eye. As you can see along
the rows:

Set 1: 8, 7, 5, 11, 10, 7, 8, 8, 6, 7, 10, 12, 5, 6, 7, 9, 9, 8, 8, 9 Mean = 8.0

Set 2: 2, 1, 0, 18, 5, 0, 52, 2, 1, 0, 1, 44, 3, 0, 0, 28, 0, 0, 2, 1 Mean = 8.0

Quantitative Methods Edinburgh Business School 5/11

Module 5 / Summary Measures

In the first case, the focus enables one to see that the numbers are scattered
closely and about equally either side of the mean; in the second case, one sees that
most numbers are below the mean with just a few considerably above.
In fact, the two sets are the symmetrical data and the reverse J-shaped data intro-
duced earlier in Figure 5.1 and Figure 5.3. In the latter case the arithmetic mean was
judged not to be the most useful measure to act as a summary; nevertheless it has a
value when used as a focus for the eye. One meets this usage with row and column
averages in tables of numbers.

For Comparisons
Measures of location can be used to compare two (or more) sets of data regardless
of whether the measure is the best summary measure for that set.

Set 1: 5, 7, 8, 9, 9, 10 Mean = 8
Set 2: 5, 5, 5, 6, 6, 7, 8, 10 Mean = 6.5

The two sets of data above contain a different number of readings. The arithme-
tic mean may or may not be the correct summary measure for either set.
Nevertheless, a useful comparison between them can be effected through the mean.
Similarly, the sickness records of a group of people (reverse J shape) over several
years can be compared using the arithmetic mean, even though one would not use
this measure purely to summarise the data.

5.3.7 The Pre-eminence of the Arithmetic Mean

For each type of summary measure, several actual measures are available for use. In
the case of measures of location, the possibilities are arithmetic mean, median and
mode. Which measure should be used in which situation? What are the strengths
and weaknesses of each measure?
The choice between arithmetic mean, median and mode is often an easy one. The
reason is that the arithmetic mean is pre-eminent. It is easy to calculate and use and
it is widely understood and recognised. The arithmetic mean, therefore, is always
used… unless there are good reasons not to. Three examples of good reasons are
given below.

Distortion of the Mean by Outliers

The arithmetic mean is sensitive to large outliers. This was the case with the reverse
J-shaped distribution in Figure 5.3. Whenever large outliers are present, use of the
median should be considered.
Another example of this effect would be in calculating the average salary in a
small engineering firm where one finds:

5/12 Edinburgh Business School Quantitative Methods

 Module 5 / Summary Measures

Salary
(inc. bonuses)
1 Founder/managing director £60 000
4 Skilled workers £14 000
5 Unskilled workers £12 000
Arithmetic mean salary = £17 600

The average salary as calculated is totally unrepresentative of salaries earned in

the firm, purely because of the distorting presence of the high earnings of the
managing director. The median salary would be £13 000 and thus more representa-
tive.

Distortion of the Mean by Clusters

The arithmetic mean can be unrepresentative if the data splits into two or more
distinctly different clusters. The mean may well fall in the middle and be some
distance from any of the data.
Such is the case with the U-shaped distribution in Figure 5.2, which referred to
television viewing. People saw almost none or almost all the episodes. The mean
was calculated as coming between the two clusters at the extremes. In such cases the
mode may be more representative.

Error by Taking Averages of Averages

When an average of an average is taken, the arithmetic mean can be incorrect.
Consider the following case of two streams of 15-year-old pupils at a high school
and their average marks in a national examination.

Stream A: 50 pupils, average mark 74%

Stream B: 30 pupils, average mark 50%

What is the average mark for both streams together?

It is tempting to say:
Overall average =
= 62%
However, this figure is the average of two averages. To get the true average, one
has to go back to the first principles:
( )
Overall average =
( × ) ( × )
=
=
=
= 65%

Quantitative Methods Edinburgh Business School 5/13

Module 5 / Summary Measures

The lesson is: when averaging averages where groups of different size are in-
volved, go back to the basic definition of the average.
Except where one of the difficulties described above applies, the arithmetic mean
is the first choice of measure of location.

5.4 Measures of Scatter

Measures of scatter do exactly as their name implies. They are a measure of the
extent to which the readings are grouped closely together or scattered over a wide
interval. They are also called measures of dispersion.

5.4.1 Range
The best-known and certainly the simplest measure of scatter is the range, which is
the total interval covered by the numbers.
Range = Largest reading − Smallest reading
For example, for the nine numbers 3, 3, 4, 5, 5, 6, 6, 6, 7:
Range = 7 − 3
=4

5.4.2 Interquartile Range

The range is defined entirely on the two extreme values, and this might be mislead-
ing if it were used as a general measure of scatter. To overcome this problem, the
interquartile range is the range of the numbers after having eliminated the highest
and lowest 25 per cent. This measure is no longer sensitive to extremes.
Interquartile range = Range of middle 50% of readings
For example, consider the following numbers: 3, 3, 4, 5, 5, 6, 6, 6, 7.
The top 25 per cent of readings (= approximately the top two numbers) and the
bottom 25 per cent (= approximately the bottom two numbers) are eliminated:

3, 3, 4, 5, 5, 6, 6, 6, 7
︸ ︸
remove remove

Interquartile range = 6 − 4
=2

5.4.3 Mean Absolute Deviation

A measure that involves all the readings is the mean absolute deviation (MAD). It
is the average distance of the readings from their arithmetic mean:
( )
MAD =

5/14 Edinburgh Business School Quantitative Methods

 Module 5 / Summary Measures

Or, more mathematically:

∑| |
MAD =

where:
is the arithmetic mean
is the number of readings in the set

(The notation | | (pronounced ‘the absolute value of’) means the size of the
number disregarding its sign).
For example, calculate the MAD of: 3, 3, 4, 5, 5, 6, 6, 6, 7.
From the previous work: = 5

3 3 4 5 5 6 6 6 7
− −2 −2 −1 0 0 1 1 1 2
| − | 2 2 1 0 0 1 1 1 2
∑| − | = 2 +2 +1 +0 +0 +1 +1 +1 +2
= 10

MAD =
= 1.1
The concept of absolute value used in the MAD is to overcome the fact that
( − ) is sometimes positive, sometimes negative and sometimes zero. The
absolute value gets rid of the sign. Why is this necessary? Try the calculation without
taking absolute values and see what happens.

5.4.4 Variance
An alternative way of eliminating the sign of deviations from the mean is to square
them, since the square of any number is never negative. The variance is the average
squared distance of readings from the arithmetic mean:
Variance =

Or, more mathematically:

∑( )
Variance =
=
= 2
The n − 1 in the denominator seems surprising. One would expect it to be n so
that the variance is the ‘average squared deviation from the mean’. If the variance is
being calculated from the population (the set of all possible values of the variable),
then the denominator should indeed be n.
However, in virtually every practical situation, the calculation is being made from
a sample (a subset of the population). In general, samples are less diverse than the

Quantitative Methods Edinburgh Business School 5/15

Module 5 / Summary Measures

population. We can see this intuitively because the one-in-a-million extreme outlier
that is present in the population will not usually be present in the sample. Extreme
outliers have a large impact on the variance since it is based on squared deviations.
Consequently, when the variance is calculated from a sample, it tends to underesti-
mate the true population variance.
Dividing by n − 1 instead of n increases the size of the calculated figure, and this
increase offsets the underestimate by just the right amount. ‘Just the right amount’
has the following meaning. Calculating the variance with n − 1 as the denominator
will give, on average, the best estimate of the population variance. That is, if we
were to repeat the calculation for many, many samples (in fact, an infinite number
of samples) and take the average, the result would be equal to the true population
variance. If we used n as the denominator this would not be the case. This can be
verified mathematically but goes beyond what a manager needs to know – consult a
specialist statistical text if you are interested.
Section 9.3 on ‘Degrees of Freedom’ in Module 9 gives an alternative and more
technical explanation.
Unless you are sure you are in the rare situation of dealing with the whole popu-
lation of a variable, you should use the n − 1 version of the formula. Calculators and
popular spreadsheet packages that have a function for calculating the variance
automatically nearly always use n − 1, although there may be exceptions.
Taking the usual example set of numbers, calculate the variance of 3, 3, 4, 5, 5, 6,
6, 6, 7 (Mean = 5).

3 3 4 5 5 6 6 6 7
− −2 −2 −1 0 0 1 1 1 2
( − ) 4 4 1 0 0 1 1 1 4
∑( − ) = 4 +4 +1 +0 +0 +1 +1 +1 +4
= 16

∑( )
Variance =
=
= 2
The variance has many applications, particularly in financial theory. However, as
a pure description of scatter, it suffers from the disadvantage that it involves
squaring. The variance of the number of weeks of sickness of 20 employees is
measured in square weeks. However, it is customary to quote the variance in
ordinary units (e.g. in the above example the variance is said to be two weeks).

5/16 Edinburgh Business School Quantitative Methods

 Module 5 / Summary Measures

5.4.5 Standard Deviation

The standard deviation attempts to overcome this problem by taking the square
root of the variance. It also has important applications of its own, particularly in
connection with the normal distribution.
Standard deviation = √Variance
Or, in mathematical form:
∑( )
Standard deviation =

In the example given in Section 5.4.4 above, variance = 2; therefore:

Standard deviation = √2
= 1.4

5.4.6 Calculating Measures of Dispersion

Just as for measures of location, calculating measures of dispersion from a spread-
sheet (or other statistical package) involves the easy task of highlighting the numbers
and applying one of the standard spreadsheet functions. In the absence of a
spreadsheet, the formula for the variance can be awkward to use because of the
need to calculate ( − ) for each observation. The task is made easier by using a
short-cut formula that is exactly equivalent to the original – some simple algebra
transforms one into the other. The short-cut formula is:
Variance = ∑( )− · / −1
The calculations for an earlier example would be as in Table 5.6.

Table 5.6 Short-cut formula

3 9
3 9
4 16
5 25
5 25
6 36
6 36
6 36
7 49
Total 45 241

Mean = =5

Variance = ∑( )− · / − 1
= [241−9×25]/8
= 2

Quantitative Methods Edinburgh Business School 5/17

Module 5 / Summary Measures

5.4.7 Comparing the Measures of Scatter

In the case of measures of location, the choice between measures is relatively easy.
The arithmetic mean is pre-eminent and it is chosen unless there is a good reason to
do otherwise. For measures of scatter, the relative merits of the measures are more
evenly balanced and the choice is more difficult. Table 5.7 shows the major
strengths and weaknesses of each measure.

Table 5.7 Comparison of measures of scatter

Measure Advantages Disadvantages
Range Easily understood Distorted by outliers
Familiar Descriptive measure only

Interquartile range Easily understood Not well known

Descriptive measure only

Mean absolute deviation Intuitively sensible Unfamiliar

Difficult to handle mathematically

Variance Easy to handle mathematically Wrong units

Used in other theories No intuitive appeal

Standard deviation Easy to handle mathematically Too involved for descriptive purposes
Used in other theories

All the measures have their particular uses. No single one stands out. When a
measure of scatter is required purely for descriptive purposes, the best measure is
probably the mean absolute deviation, although it is not as well known as it deserves
to be. When a measure of scatter is needed as part of some wider statistical or
mathematical theory, then the variance and standard deviation are frequently
encountered.
Further Example
A company’s 12 salespeople in a particular region last month drove the following
number of kilometres:

Salesperson Kilometres
(hundreds)
1 34
2 47
3 30
4 32
5 38
6 39
7 36
8 43

5/18 Edinburgh Business School Quantitative Methods

 Module 5 / Summary Measures

Salesperson Kilometres
(hundreds)
9 31
10 40
11 42
12 32

Calculate:
(a) range
(b) interquartile range
(c) MAD
(d) variance
(e) standard deviation.
Which measure is the most representative of the scatter in these data?
(a) Range = Highest – Lowest = 47 – 30 = 17
(b) Putting the numbers in ascending order:

30, 31, 32, 32, 34, 36, 38, 39, 40, 42, 43, 47
Lowest quartile (25%) Highest quartile (25%)
Interquartile range = 40 – 32 = 8
(c) To calculate MAD, it is first necessary to find the arithmetic mean:
Mean =
=
= 37
Next calculate the deviations:

34 47 30 32 38 39 36 43 31 40 42 32
− −3 10 −7 −5 1 2 −1 6 −6 3 5 −5
| − | 3 10 7 5 1 2 1 6 6 3 5 5
∑| − | = 54

∑| |
MAD =
= 4.5
(d) The variance

34 47 30 32 38 39 36 43 31 40 42 32
− −3 10 −7 −5 1 2 −1 6 −6 3 5 −5
( − ) 9 100 49 25 1 4 1 36 36 9 25 25
∑( − ) = 320

Quantitative Methods Edinburgh Business School 5/19

Module 5 / Summary Measures

∑( )
Variance =
=
= 29.1
(e) Standard deviation
= √Variance
= √29.1
= 5.4
The best descriptive measure of scatter in this situation is the mean absolute
deviation. The average difference between a salesperson’s travel and the average travel
is 4.5 (450 kilometres). This is a sensible measure that involves all data points. The range
is of great interest, but not as a measure of scatter. Its interest lies in indicating the
discrepancy between the most and least travelled. It says nothing about the ten in-
between salespeople. The interquartile range is probably the second choice. The
variance and standard deviation are probably too complex conceptually to be descrip-
tive measures in this situation, where further statistical analysis is not likely.

5.4.8 Coefficient of Variation

When there are differences in the means of two groups, a measure of scatter must
be ‘standardised’ before comparison of relative variation can be made. The coeffi-
cient of variation does this in the case of the standard deviation.
Coefficient of variation =

This is useful when sets of data with very different characteristics are being com-
pared. For example, suppose one is looking at the number of passengers per day
passing through two airports. Over a one-year period the average number of
passengers per day, the standard deviations and the coefficients of variation are
calculated.

Mean Standard deviation Coefficient of variation

Airport 1 4 200 1 050 0.25
Airport 2 15 600 2 250 0.14

Consideration of the standard deviations alone would suggest that there was more
scatter at Airport 2. In relation to the number of passengers using the two airports, the
scatter is smaller at Airport 2 as revealed by the coefficient of variation being 0.14 as
against 0.25 at Airport 1.

5.5 Other Summary Measures

Measures of location and scatter are the most useful and frequently seen summaries.
There are two other measures that are occasionally encountered.

5/20 Edinburgh Business School Quantitative Methods

 Module 5 / Summary Measures

5.5.1 Skew
Skew measures the extent to which a distribution is non-symmetrical. Figure 5.4(a)
is a distribution that is left-skewed; Figure 5.4(b) is a symmetrical distribution with
zero skew; Figure 5.4(c) is a distribution that is right-skewed.

(a) (b) (c)

Figure 5.4 Skew: (a) left skew; (b) zero skew; (c) right skew
The concept of skew is normally used purely descriptively and is assessed visually
(i.e. one looks at the distribution and assesses whether it is symmetrical or right- or
left-skewed). Skew can be measured quantitatively but the formula is complex and
the accuracy it gives (over and above a verbal description) is rarely necessary in
practice. However, the measurement of skew gives rise to the alternative labels
positively skewed (right-skewed) and negatively skewed (left-skewed).

5.5.2 Kurtosis
Kurtosis measures the extent to which a distribution is ‘pinched in’ or ‘filled out’.
Figure 5.5 shows three distributions displaying increasing levels of kurtosis. As with
skew, a qualitative approach is sufficient for most purposes (i.e. when one looks at
the distribution, one can describe it as having a low, medium or high level of
kurtosis). Kurtosis can also be measured quantitatively, but, again, the formula is
complex.

(a) (b) (c)

Figure 5.5 Kurtosis: (a) high; (b) medium; (c) low

5.6 Dealing with Outliers

The data from which summary measures are being calculated may well include one
or more outliers, which may have a disproportionate effect on the result. This is

Quantitative Methods Edinburgh Business School 5/21

Module 5 / Summary Measures

particularly true of the variance and standard deviation, which use squared values.
How does one deal with the outliers? Are they to be included or excluded? Three
basic situations arise.
(a) Twyman’s Law. This only half-serious law states that any piece of data that
looks interesting or unusual is wrong. The first consideration when confronted
by an outlier is whether the number is incorrect, perhaps because of an error in
collection or a typing mistake. There is an outlier in these data, which are the
week’s overtime payments to a group of seven workers (in £s):
13.36, 17.20, 16.78, 15.98, 1432, 19.12, 15.37
Twyman’s Law suggests that the outlier showing a payment of 1432 occurs be-
cause of a dropped decimal point rather than a fraudulent claim. The error
should be corrected and the number retained in the set.
(b) Part of the pattern. An outlier may be a definite and regular part of the pattern
and should be neither changed nor excluded. Such was the case with the sickness
record data of Figure 5.3. The outliers were part of the pattern and similar ef-
fects were likely to be seen in other time periods and with other groups of
employees.
(c) Isolated events. Outliers occur that are not errors but that are unlikely to be
repeated (i.e. they are not part of the pattern). Usually they are excluded from
calculations of summary measures, but their exclusion is noted. For example, the
following data, recorded by trip wire, show the number of vehicles travelling
down a major London road during a ten-day period:
5271, 5960, 6322, 6011, 7132, 5907, 6829, 741, 7098, 6733
The outlier is the 741. Further checking shows that this day was the occasion of
a major royal event and that the road in question was closed to all except state
coaches, police vehicles, etc. This is an isolated event, perhaps not to be repeated
for several years. For traffic control purposes, the number should be excluded
from calculations since it is misleading, but a note should be made of the exclu-
sion. Hence, one would report:
Mean vehicles/day =
=
= 6363 (excluding day of Royal Event)
The procedure for outliers is first to look for mistakes and correct them; and
second, to decide whether the outlier is part of the pattern and should be includ-
ed in calculations or an isolated event that should be excluded.

5.7 Indices
An index is a particular type of measure used for summarising the movement of a
variable over time. When a series of numbers is converted into indices, it makes the
numbers easier to understand and to compare with other series.
The best-known type of index is probably the cost of living index. The cost of
living comprises the cost of many different goods: foods, fuel, transport, etc.
Instead of using the miscellaneous and confusing prices of all these purchases, we

5/22 Edinburgh Business School Quantitative Methods

 Module 5 / Summary Measures

use an index number, which summarises them for us. If the index in 2014 is 182.1
compared with 165.3 in 2013, we can calculate that the cost of living has risen by:
. .
× 100 = 10.2%
.

This is rather easier than having to cope with the range of individual price rises
involved.
Every index has a base year when the index was 100 (i.e. the starting point for
the series). If the base year for the cost of living index was 2008, then the cost of
living has risen 82.1 per cent between 2008 and 2014. This could be said in a
different way: the 2014 cost of living is 182.1 per cent of its 2008 value.
The index very quickly gives a feeling for what has happened to the cost of living.
Comparisons are also easier. For example, if the Wages and Salaries Index has 2008
as its base year and stood at 193.4 in 2014, then, over the six years, wages out-
stripped the cost of living: 93.4 per cent as against 82.1 per cent.
The cost of living index is based on a complicated calculation. However, there
are some more basic indices.

5.7.1 Simple Index

At the most primitive level, an index is just the result of the conversion of one series
of numbers into another, based on 100. Suppose the original numbers refer to the
average price of a new house in some region in each of 10 years and, with 2010
arbitrarily chosen as the base year, the original series and the index are as follows:

2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Price (£000s) 6.1 8.2 8.6 10.1 11.8 12.4 16.9 19.0 19.7 19.4
Index 49 66 69 81 95 100 136 153 159 156

The index for 2010, being the base year, is 100. The other data in the series are
scaled up accordingly. For instance, the index for 2007 is:
8.6 × = 69
.
where 8.6 is the original datum and 12.4 is the original datum for the base year. And
for 2013:
19.7 × = 159
.
The choice of the base year is important. It should be such that individual index
numbers during the time span being studied are never too far away from 100. As a
rule of thumb, the index numbers are not usually allowed to differ from 100 by
more than the factor of 3 (i.e. the numbers are in the range 30 to 300). If the base
year for the numbers in the series above had been chosen as 2005, then the index
series would have been from 100 to 318.
In long series, there might be more than one base year. For example, a series
covering more than 30 years from 1982 to 2014 might have 1982 as a base year with

Quantitative Methods Edinburgh Business School 5/23

Module 5 / Summary Measures

the series then rising to 291 in 2001, which could then be taken as a second base
year:

1982 2001 2014

Index 100 → 291
100 → 213

Care obviously needs to be taken in interpreting this series. The increase from
1982 to 2014 is not 113 per cent. If the original series had been allowed to continue,
the 2014 index would have been 213 × 2.91 = 620. The increase is thus 520 per
cent.

5.7.2 Simple Aggregate Index

The usefulness of an index is emphasised when it is used to summarise several
factors. A monthly index for meat prices could not be based on monthly prices for,
say, beef alone. It would have to take into account prices of other types of meat. An
aggregate index does just this.
For example, the table below shows the price per kilogram at market for beef,
pork and lamb. A simple aggregate combines the different prices by adding them up
and forming an index from the total. For instance, the total price for January is
148 + 76 + 156 = 380. For February it is 150 + 80 + 167 = 397.

Price (p per kilogram)

Month Cattle Pigs Lambs
Jan. 148 76 156
Feb. 150 80 167
Mar. 155 75 180
Apr. 163 79 194
May 171 81 186
June 179 76 178
July 184 82 171
Aug. 176 76 168
Sept. 142 79 163
Oct. 146 84 160
Nov. 149 87 159
Dec. 154 94 162

If January is taken as the base, then the indices for the months are:

Month Total Index

Jan. 380 100
Feb. 397 104

5/24 Edinburgh Business School Quantitative Methods

 Module 5 / Summary Measures

Mar. 410 108

Apr. 436 115
May 438 115
June 433 115
July 437 114
Aug. 420 111
Sept. 384 101
Oct. 390 103
Nov. 395 104
Dec. 410 108

One possible disadvantage of this index is that livestock with a low price will
have much less influence than livestock with a high price. For instance, in February
a 20 per cent change in the price of cattle would change the index much more than
a 20 per cent change in the pig price:

February 150 + 80 + 167 = 397 Index = 104.5

With 20% cattle price change 180 + 80 + 167 = 427 Index = 112.4
With 20% pig price change 150 + 96 + 167 = 413 Index = 108.7

However, this may be a desirable feature. If the price level of each factor in the
index reflects its importance, then the higher-priced elements should have more
effect on the index. On the other hand, this feature may not be desirable. One way
to counter this is to construct a price relative index. This means that the individual
prices are first converted into an index and then these individual indices are aver-
aged to give the overall index.

5.7.3 Weighted Aggregate Index

Whether to use prices or price relatives in the index described above is really a
question of weighting. How much influence should the individual parts have,
relative to one another, on the overall index? This issue is of greatest importance in
a cost of living index. A simple aggregate index would not be a suitable method of
combining the prices of milk, bread, fruit, tobacco, electricity and so on. A
weighted aggregate index allows different weights to be given to the different
prices. What the weights should be is still a matter of judgement, but in the case of
price indices the quantities purchased are often used.
If we return to the livestock index, suppose the quantities purchased at the live-
stock markets from which the price data were obtained are as below (quantities in
thousand tonnes):

Quantitative Methods Edinburgh Business School 5/25

Module 5 / Summary Measures

Cattle Pigs Lambs

Price Quantity Price Quantity Price Quantity
Jan. 148 1.6 76 0.25 156 0.12
Feb. 150 1.5 80 0.23 167 0.11
Mar. 155 1.4 75 0.25 180 0.10

Instead of adding them up, we first weight the prices by the quantities, and the
final index is formed from the resulting monthly total. The quantities used for the
weighting should be the same for each month, since this is a price index. Otherwise
the index would measure price and volume changes. If the quantities used for
weighting are the base month (January) quantities, then the index is known as a
Laspeyres Index and is calculated as follows:

Month Weighted total Index

Jan. (148 × 1.6) + (76 × 0.25) + (156 × 0.12) = 274.5 100
Feb. (150 × 1.6) + (80 × 0.25) + (167 × 0.12) = 280.0 280/274.5 × 100 = 102
Mar. (155 × 1.6) + (75 × 0.25) + (180 × 0.12) = 288.3 288.3/274.5 × 100 = 105

A Laspeyres Index, like other indices, can be used for quantities as well as prices.
For a quantity index the role of price and quantity in the above example (of a price
index) would be reversed, with prices providing the weightings to measure changes
in quantities. The weights (prices) remain at the constant level of some base period
while the quantities change from time period to time period. For example, the UK
Index of Manufacturing Production shows how the level of production in the
country is changing as time goes by. The quantities refer to different types of
product – consumer goods, industrial equipment, etc. – and the prices are those of
the products in a base year. The use of prices as weights for quantities gives the
most expensive products a heavier weighting.
A major criticism of the Laspeyres Index is that the weights in the base year may
soon become out of date and no longer representative. An alternative is the
Paasche Index, which takes the weights from the most recent time period – the
weightings therefore change from each time period to the next. In the livestock
example a Paasche Index would weight the prices in each month with the quantities
relating to December, the most recent month. A Paasche Index always uses the
most up-to-date weightings, but it has the serious practical disadvantage that, if it is
to be purely a price index, every time new data arrive (and the weightings change)
the entire past series must also be revised.
A fixed weight index may also be used. Its weights are from neither the base
period nor the most recent period. They are from some intermediate period or from
the average of several periods. It is a matter of judgement to decide which weights
to use.
The cost of living index has already been introduced. It indicates how the cost of
a typical consumer’s lifestyle changes as time goes by, and it has many practical uses.
For example, it is usually the starting point in wage negotiations, since it shows how
big a wage increase is needed if an employee’s current standard of living is to be

5/26 Edinburgh Business School Quantitative Methods

 Module 5 / Summary Measures

maintained. A wage increase lower than the cost of living index would imply a
decrease in real wages.
What type of index should be used for the cost of living index? Table 5.8 and
Table 5.9 show the simplified example from Economics Module 12 (Tables 12.1 and
12.2).

Table 5.8 Consumer price index in base period t

Good/Service Quantity Price ($) Total expenditure ($)
Loaf of bread 10 loaves 1.00/loaf 10.00
Glass of wine 5 glasses 1.20/glass 6.00
Haircut 2 7.00 14.00
30.00

Table 5.9 Consumer price index in base period t + 1

Good/Service Quantity Price ($) Total expenditure ($)
Loaf of bread 10 loaves 1.10/loaf 11.00
Glass of wine 5 glasses 1.20/glass 6.00
Haircut 2 8.00 16.00
33.00

Price index for time t + 1 = 33.00/30.00 × 100

= 110
Increase in price index = (110 – 100)/100
= 10%
This is a Laspeyres Index, but it is not enough for a cost of living index, which
must reflect, in addition to prices, the fact that people’s purchasing behaviour
changes in response to changes in prices, wage levels, technology, fashion and so
on. For example, the arrival into the market of good-quality wines from a wider
range of countries than in the past provides extra choice, possibly reduces prices
and therefore may lead to more wine being consumed. A change in fashion from
longer to shorter hair may result in greater spending on haircuts, possibly at the
expense of bread purchases. A cost of living index should reflect changes such as
these, since it is intended to indicate ‘typical’ living costs.
A Paasche Index should be used since it incorporates changes in both prices and
weights. In practice it is based on a lengthy and detailed list of expenditure catego-
ries and their relative weights. The list might look like Table 5.10.

Quantitative Methods Edinburgh Business School 5/27

Module 5 / Summary Measures

Table 5.10 List of expenditures and their weights

Expenditure £ Weight
All items 100.00
Food and beverages
Food at home 12.87
Food not at home 6.34
Alcoholic beverages 1.77 20.98
Housing
Rent 3.66
Home ownership 23.54
Fuel 5.21
Furnishings 6.43 38.84
Clothing
Male 1.07
Female 1.87
Infant 0.34
Footwear 0.85 4.13
Transportation
Cars
Petrol
Public
…
… 18.72
Medical 2.11
Entertainment 3.22
Other 12.00

The weights are multiplied by the prices to give the index, as in the livestock
example. The weights are changed regularly as a result of government surveys of
expenditure patterns. It is important to note that, for the cost of living index,
previous values of the index are not changed as the weights change (i.e. the index
remains as it was when first calculated). Consequently, in comparing the cost of
living now with that of 20 years ago, a cost of living index reflects changes in
purchasing behaviour as well as the inevitably increasing prices.
A price index such as the cost of living index can be used to deflate economic
data. For example, the measure of a country’s economic activity is its GNP (gross
national product), the total value of the goods and services an economy produces in
a year. The GNP is measured from statistical returns made to the government and is
calculated in current prices (i.e. it is measured in terms of the prices for the goods
and services that apply in that year). Consequently, the GNP can rise from one year
to another because prices have risen through inflation, even though actual economic
activity has decreased. It would be helpful to neutralise the effect of prices in order

5/28 Edinburgh Business School Quantitative Methods

 Module 5 / Summary Measures

to have a more realistic measure of economic activity. This is done by deflating the
series so that the GNP is in real, not current, terms. The deflation is carried out by
using a price index. Table 5.11 shows the GNP of a fictional country for 2008–14,
together with a price index for those years. Current GNP is converted to real GNP
by dividing by the price index.

Table 5.11 Deflating GNP

Year GNP(current) Price index GNP(real)
2008 320 100 320
2009 335 104 322
2010 347 107 324
2011 358 110 325
2012 376 113 333
2013 389 116 335
2014 403 120 336

For example, for 2012:

GNP(real) = GNP(current)/Price index × 100
= (376/113) × 100
= 333

The changes in GNP(real) over the years 2008–14 show that economic activity
did increase in each of those years but not by as much as GNP(current) suggested.
It is important that an appropriate price index is used. It would not have been
appropriate to use the cost of living index for GNP since that index deals only with
consumer expenditure. A price index that incorporates the prices used in the GNP –
that is, investment and government goods as well as consumer goods – should be
used.

Learning Summary
In the process of analysing data, at some stage the analyst tries to form a model of
the data, as suggested previously. ‘Pattern’ or ‘summary’ are close synonyms for
‘model’. The model may be simple (all rows are approximately equal) or complex
(the data are related via a multiple regression model). Often specifying the model
requires intuition and imagination. At the very least, summary measures can provide
a model based on specifying for the data set:
(a) number of readings;
(b) a measure of location;
(c) a measure of scatter;
(d) the shape of the distribution.
In the absence of other inspiration, these four attributes provide a useful model
of a set of numbers. If the data consist of two or more distinct sets (as, for example,
a table), then this basic model can be applied to each. This will give a means of

Quantitative Methods Edinburgh Business School 5/29

Module 5 / Summary Measures

comparison between the rows or columns of the table or between one time period
and another.
The first attribute (number of readings) is easily supplied. Measures of location
and scatter have already been discussed. The shape of the distribution can be found
by drawing a histogram and literally describing its shape (as with the symmetrical, U
and reverse-J distributions seen earlier). A short verbal statement about the shape is
often an important factor in summarising or forming a model of a set of data.
Verbal statements have a more general role in summarising data. They should be
short, no more than one sentence, and used only when they can add to the sum-
mary. They are used in two ways: first, they are used when the quantitative measures
are inadequate; second, they are used to point out important features in the data.
For example, a table of a company’s profits over several years might indicate that
profits had doubled. Or a table of the last two months’ car production figures might
have a note stating that 1500 cars were lost because of a strike.
It is important, in using verbal summaries, to distinguish between helpful state-
ments pointing out major features and unhelpful statements dealing with trivial
exceptions and details. A verbal summary should always contribute to the objective
of adding to the ease and speed with which the data can be handled.

Review Questions
5.1 Which of the following statements about summary measures are true?
A. They give greater accuracy than the original data.
B. It is easier to handle information in summary form.
C. They are never misleading.
D. Measures of location and scatter together capture all the main features of data.
Questions 5.2 to 5.4 refer to the following data:
1, 5, 4, 2, 7, 1, 0, 8, 6, 6, 5, 2, 4, 5, 3, 5

5.2 The arithmetic mean is:

A. 4
B. 5.5
C. 4.25
D. 4.5
E. 5

5.3 The median is:

A. 4
B. 5.5
C. 4.25
D. 4.5
E. 5

5/30 Edinburgh Business School Quantitative Methods

 Module 5 / Summary Measures

5.4 The mode is:

A. 4
B. 5.5
C. 4.25
D. 4.5
E. 5

5.5 Which of the following applies? As a measure of location, the arithmetic mean:
A. is always better than the median and mode.
B. is usually a misleading measure.
C. is preferable to mode and median except when all three are approximately
equal.
D. should be used when the data distribution is U shaped.
E. None of the statements applies.

5.6 An aircraft’s route requires it to fly along the sides of a 200 km square (see figure
below). Because of prevailing conditions, the aircraft flies from A to B at 200 km/h, from
B to C at 300 km/h, from C to D at 400 km/h and from D to A at 600 km/h. What is
the average speed for the entire journey from A to A?

A B

200 km

D C

A. 325 km/h
B. 320 km/h
C. 375 km/h
D. 350 km/h

5.7 Which of the following statements about measures of scatter are true?
A. Measures of scatter must always be used when measures of location are used.
B. A measure of scatter is an alternative to a measure of location as a summary.
C. One would expect a measure of scatter to be low when readings are close
together, and high when they are further apart.
D. A measure of scatter should be used in conjunction with a measure of disper-
sion.

Quantitative Methods Edinburgh Business School 5/31

Module 5 / Summary Measures

Questions 5.8 to 5.12 refer to the following data:

23, 27, 21, 25, 26, 22, 29, 24, 27, 26

5.8 The range is:

A. 7
B. 8
C. 9
D. 10

5.9 The interquartile range is:

A. 2
B. 4
C. 8
D. 5

5.10 The mean absolute deviation is:

A. 3
B. 2
C. 1.8
D. 2.2

5.11 The variance is:

A. 5.6
B. 6.2
C. 5.2
D. 6.5

5.12 The standard deviation is:

A. 6.2
B. 5.6
C. 2.4
D. 2.5

5.13 Which of the following statements is true?

A. No measure of scatter is pre-eminent, and therefore it does not matter which
is used for a particular set of data.
B. Interquartile range is not distorted by extreme readings.
C. Variance and standard deviation measure scatter but in squared units.
D. Mean absolute deviation is preferable to interquartile range because it is easier
to handle mathematically.

5/32 Edinburgh Business School Quantitative Methods

 Module 5 / Summary Measures

5.14 Which of the following statements are true regarding the presence of an outlier in a
data set of which summary measures are to be calculated?
A. An outlier should either be retained in or excluded from the set.
B. An outlier that is part of the pattern of the data should always be used in
calculating the arithmetic mean.
C. An outlier that is not part of the pattern should usually be excluded from any
calculations.

5.15 Two governmental indices have the following values:

2011 2012 2013 2014

Cost of living 100 109.7 118.3 128.6
Wages, salaries 100 111.2 123.1 133.5

Which of the following is correct? Between 2013 and 2014, the growth of the cost of
living compared to wages and salaries was:
A. much greater.
B. slightly greater.
C. equal.
D. slightly less.
E. much less.

Case Study 5.1: Light Bulb Testing

1 Samples of light bulbs were obtained from two suppliers and tested to destruction in
the laboratory. The following results were obtained for the length of life:

Length of life (hours)

700–899 900–1099 1100–1299 1300–1499 Total
Supplier A 12 14 24 10 60
Supplier B 4 34 19 3 60

a. Which supplier’s bulbs have greater average length of life?

b. Which supplier’s bulbs are more uniform in quality?
c. Which supplier would you prefer to use?

Quantitative Methods Edinburgh Business School 5/33

Module 5 / Summary Measures

Case Study 5.2: Smith’s Expense Account

1 Mr Smith’s expense account for six trips made in the past month is given in Table 5.12.

Table 5.12 Mr Smith’s expense account

Trip Days Expenses (£) Expenses per day (£)
1 2 64 32
2 8 128 16
3 0.5 40 80
4 9 108 12
5 4 80 20
6 0.5 25 50
Totals 24 445 210

Mr Smith’s boss felt that these expenses were excessive because, he said, the average
expense per day was £35. Other salespeople away on weekly trips submitted expenses
that averaged at around £20 a day. Where did the £35 come from? How can Smith
argue his case?

Case Study 5.3: Monthly Employment Statistics

1 The statistics for monthly employment in four different departments of Company X are
as follows:

Department
A B C D
Mean monthly employment level 10 560 4891 220 428
Standard deviation 606 302 18 32

Is the monthly employment level more stable in some departments than in others?

Case Study 5.4: Commuting Distances

1 The distribution of the distance between home and place of work for workers in the
London region was found in a survey to be:

Distance (to nearest mile)

0 1 2 3 4–5 6–7 8–9 10–11 12–14 15–19 20–29 30+
% workers 3 7 9 10 16 12 11 8 8 4 6 6

The mean distance from work was found to be 10.5 miles. Calculate the mode, the
median and two measures of scatter. How would you summarise these data succinctly?

5/34 Edinburgh Business School Quantitative Methods

 Module 5 / Summary Measures

Case Study 5.5: Petroleum Products

1 The following table shows the wholesale prices and production totals of three
petroleum products, prices being given in pence per gallon and production figures in
millions of barrels.

Prices Quantities
2012 2013 2014 2012 2013 2014
Car petrol 26.2 27.1 27.4 746 768 811
Kerosene 24.8 28.9 26.5 92 90 101
Paraffin 23.0 24.1 24.8 314 325 348
a. Use 2012 = 100 to construct a simple aggregate index for the years 2012 to 2014
for the prices of the three petroleum products.
b. Use 2012 quantities as weights and 2012 = 100 to construct a weighted aggregate
index for the years 2012 to 2014 for the prices of the three petroleum products.
c. Does it matter which index is used? If so, which one should be used?
d. How else could our index be constructed?

Quantitative Methods Edinburgh Business School 5/35

Module 6

Sampling Methods
Contents
6.1 Introduction.............................................................................................6/1
6.2 Applications of Sampling .......................................................................6/3
6.3 The Ideas behind Sampling ....................................................................6/3
6.4 Random Sampling Methods ...................................................................6/4
6.5 Judgement Sampling ........................................................................... 6/10
6.6 The Accuracy of Samples ................................................................... 6/12
6.7 Typical Difficulties in Sampling .......................................................... 6/13
6.8 What Sample Size? .............................................................................. 6/15
Learning Summary ......................................................................................... 6/16
Review Questions ........................................................................................... 6/18
Case Study 6.1: Business School Alumni ..................................................... 6/20
Case Study 6.2: Clearing Bank ...................................................................... 6/20

Prerequisite reading: None

Learning Objectives
By the end of this module the reader should know the main principles underlying
sampling methods. Most managers have to deal with sampling in some way. It may
be directly in commissioning a sampling survey, or it may be indirectly in making
use of information based on sampling. For both purposes it is necessary to know
something of the techniques and, more importantly, the factors critical to their
success.

6.1 Introduction
Statistical information in management is usually obtained from samples. The
complete set of all conceivable observations of a variable is a population; a subset
of a population is a sample. It is rarely possible to study a population. Nor is it
desirable, since sample information is much less expensive yet proves sufficient to
take decisions, solve problems and answer questions in most situations. For
example, to know what users of soap powder in the UK think about one brand of
soap powder, it is hardly possible to ask each one of the 15 million of them his or
her opinion, nor would the expense of such an exercise be warranted. A sample
would be taken. A few hundred would be interviewed and from their answers an
estimate of what the full 15 million are thinking would be made. The 15 million are
the population, the few hundred are the sample. A population does not have to refer

Quantitative Methods Edinburgh Business School 6/1

Module 6 / Sampling Methods

to people. In sampling agricultural crops for disease, the population might be 50 000
hectares of wheat.
In practice, information is nearly always collected from samples as opposed to
populations, for a wide variety of reasons.
(a) Economic advantages. Collecting information is expensive. Preparing ques-
tionnaires, paying postage, travelling to interviews and analysing the data are just
a few examples of the costs. Taking a sample is cheaper than observing the
whole population.
(b) Timeliness. Collecting information from a whole population can be slow,
especially waiting for the last few questionnaires to be filled in or for appoint-
ments with the last few interviewees to be granted. Sample information can be
obtained more quickly, and sometimes this is vital, for example, in electoral opin-
ion polls when voters’ intentions may swing significantly as voting day
approaches.
(c) Size and accessibility. Some populations are so large that information could
not be collected from the whole population. For example, a marketing study
might be directed at all teenagers in a country and it would be impossible to
approach them all. Even in smaller populations there may be parts that cannot
be reached. For example, surveys of small businesses are complicated by there
being no up-to-date lists because small businesses are coming into and going out
of existence all the time.
(d) Observation and destruction. Recording data can destroy the item being
observed. For example, in a quality test on electrical fuses, the test ruins the fuse.
It would not make much sense to destroy all fuses immediately after production
just to see whether they worked. Sampling is the only possible approach to some
situations.
There are two distinct areas of theory associated with sampling, illustrated in
Figure 6.1. In the example of market research into soap powder above, the first area
of theory would show how to choose the few hundred interviewees; the second
would show how to use the sample information to draw conclusions about the
population.

There is a need for The sample results are

statistical information A sample is used to make estimates
about a population. collected. about the population.

Theory
Theory How to make
How to choose inferences about
a sample. the population
from the sample.

Figure 6.1 Areas of sampling theory

6/2 Edinburgh Business School Quantitative Methods

 Module 6 / Sampling Methods

Only the first area, methods of choosing samples, is the topic here. The methods
will be described, some applications will be illustrated and technical aspects will be
introduced. The second area, making inferences, is the subject of Module 8.

6.2 Applications of Sampling

Examples of the use of sampling methods are common, so much so that the skill
required in choosing the samples may not be appreciated.
(a) Opinion polls. Newspapers regularly feature the results of opinion polls, which
typically are concerned with issues such as: ‘For which party would you vote if
there were a national election tomorrow?’ and ‘Are you satisfied with the gov-
ernment’s handling of the economy?’ The population to which such questions
are addressed is the entire electorate of a country. For the UK the population
would comprise approximately 47 million people. Surveying the population
would be expensive, time-consuming and practically impossible. Opinion poll
companies base their results on questioning a sample, usually of 1000–2000 peo-
ple in size, of the electorate. The poll is therefore much cheaper and can be
obtained sufficiently quickly to be up to date when published. As will be shown
later, the results are only slightly less accurate than a population survey.
(b) Quality control. A large food processing company receives many tonnes of raw
food materials every day. It would be very expensive, if not impossible, to in-
spect all the incoming supplies. Instead, samples are taken. When a consignment
arrives, a sample from it is inspected. On the basis of the results, the whole con-
signment may be rejected. The rules that govern the acceptance or rejection are
based on statistical theory.
(c) Checking invoices. Large organisations issue hundreds of thousands of
invoices in the course of a financial year. When checking their accuracy, either
for reasons of control or in auditing, they cannot all be investigated. A sample is
taken, and from the results estimates of the overall errors in invoicing are made.

6.3 The Ideas behind Sampling

In these three examples, the methods by which the samples are selected have to be
carefully thought out. In order to draw correct conclusions about the population, a
sample should be representative of it (i.e. the sample should be a microcosm of the
population). No method of collecting samples exists that will guarantee to produce
representative samples every time. All methods will occasionally produce an
unrepresentative sample, but all methods have representativeness as their aim.
However, if samples are chosen in such a way that they are consistently not repre-
sentative, there is said to be sampling bias.
The fundamental method of achieving representativeness is simple random
sampling, in which the sample is chosen from the population in such a way that
each element of the population has an equal chance of being chosen. Imagine that
each element is numbered, the numbers are written on labels that are placed in a
(possibly large) hat and a sample is picked out by a blindfolded person. Simple

Quantitative Methods Edinburgh Business School 6/3

Module 6 / Sampling Methods

random sampling is equivalent to this procedure. The selection of the sample is left
to chance and one would suppose that, on average (but not always), the sample
would be reasonably representative.
The major disadvantage of simple random sampling is that it can be expensive. If
the question to be answered is the likely result of a UK general election, then the
sampler must find some means of listing the whole electorate, choose a sample at
random and then visit the voters chosen (perhaps one in Inverness, one in Pen-
zance, two in Norwich, one in Blackpool, etc.). Both selection and questioning
would be costly.
Variations on simple random sampling can be used to overcome this problem.
For example, multi-stage sampling in the opinion poll example above would
permit the sample to be collected in just a few areas of the country, cutting down on
the travelling and interviewing expenses.
The variations on simple random sampling also make it possible to use other
information to make the sample more representative. In the soap powder example a
stratified sample would let the sampler make use of the fact that known percent-
ages of households use automatic machines, semi-automatic machines, manual
methods and launderettes.
Some situations do not allow or do not need any randomisation in sampling.
(How would a hospital patient feel about a ‘random’ sample of blood being taken
from his body?) Judgement sampling refers to all methods that are not essentially
random in character and in which personal judgement plays a large role. They can
be representative in many circumstances.
Figure 6.2 shows diagrammatically the main sampling methods and how they are
linked. In the next sections these methods will be described in more detail.

6.4 Random Sampling Methods

6.4.1 Simple Random Sampling
Simple random sampling is a procedure of sampling in which each member of the
population has an equal chance of being selected. Numbering the elements of the
population and drawing the numbers from a hat is simple random sampling. The
method does not rely on the ownership of a large hat. More usually, it works with
the help of a computer or a random number table.
Table 6.1 shows 28 two-digit random numbers. They were generated by a comput-
er program, which makes each selection in such a way that each number from 00 to
99 has an equal chance of being chosen. Alternatively, they could have been taken
from a random number table (which is generated in the same way and can be found in
most sets of statistical tables).

6/4 Edinburgh Business School Quantitative Methods

 Module 6 / Sampling Methods

All sampling methods

Random sampling Judgement sampling

Simple random Variations on Systematic Convenience Quota

simple random:
- Multi-stage
- Cluster
- Stratified
- Weighting
- Probability
- Variable
- Area

Figure 6.2 Relationship between the main sampling methods

Table 6.1 Two-digit random numbers

31 03 62 98 05 88 69
07 99 26 31 17 74 47
85 53 22 14 60 44 93
58 42 75 72 61 55 02

Suppose a simple random sample of five is to be selected from the 90-strong

workforce of a factory. The five are to be asked to take part in a detailed survey of
attitudes to work.
From payroll records a list of the workforce is made and each person given a
number between 1 and 90, as shown in Table 6.2.

Table 6.2 Numbered list of workforce

1. Appleyard, R. 31. Lester, E. 61. Sharpe, P.
2. Bainbridge, B. 32. Leyland, M. 62. Sutcliffe, H.
3. Binks, J. … …
4. Bowes, W. … …
5. Close, B. … …
6. Cowan, R. … …
… … …
… … 88. Wardle, J.
29. Hutton, L. … 89. Wilson, D.
30. Illingworth, R. 60. Ryan, M. 90. Yardley, N.

Quantitative Methods Edinburgh Business School 6/5

Module 6 / Sampling Methods

To choose the sample, the random numbers are taken one at a time from Ta-
ble 6.1 and each associated with the corresponding number and name from
Table 6.2. The first number is 31, so the corresponding employee is Lester, E. The
second number is 03, so the employee is Binks, J. This is continued until the
necessary sample is collected. The third number is 62, and thus the name is Sutcliffe,
H. The fourth number is 98, which does not correspond to any name and is
ignored. Table 6.3 shows the sample of five.

Table 6.3 Sample of five from workforce

Number Name
31 Lester, E.
03 Binks, J.
62 Sutcliffe, H.
05 Close, B.
88 Wardle, J.

The drawbacks of simple random sampling are that the listing of the population
can prove very expensive, or even impossible, and that the collection of data from
the sample can also be expensive, as with opinion polls. Variations on simple
random sampling have been developed to try to overcome these problems. These
methods still have a sizeable element of random selection in them and the random
part uses a procedure such as that described above.

6.4.2 Variations on Simple Random Sampling

Multi-Stage
In multi-stage sampling, the population is split into groups and each group is split
into subgroups, each subgroup into subsubgroups, etc. A random sample is taken at
each stage of the breakdown. First a simple random sample of the groups is taken;
of the groups chosen a simple random sample of their subgroups is taken and so on.
For example, suppose the sample required is of 2000 members of the workforce of
a large company, which has 250 000 employees situated in offices and factories over
a large geographical area. The objective is to investigate absenteeism. The company
has 15 regional divisions; each division has an average of ten locations at which an
office or factory is situated. Each location has its own computerised payroll. Since
the company operates on a decentralised basis, no company-wide list of employees
exists. Figure 6.3 illustrates how the population is split.

6/6 Edinburgh Business School Quantitative Methods

 Module 6 / Sampling Methods

Company

15 geographical divisions

About ten locations in each division

About 1700 employees at each location

Figure 6.3 Population split

If sampling were to be done by simple random sampling, two difficulties would
be encountered. First, a company-wide list of employees would need to be compiled
– no easy task with payrolls on different computer systems. Second, it is likely that
to collect the necessary information from personnel records would require visits to
virtually all the divisions and locations of the company, which would be time-
consuming and expensive.
Multi-stage sampling overcomes these problems. First, a single random sample is
taken of four regional divisions from the list of 15 using the method of simple
random sampling described above. For each of the four divisions chosen, a list of its
locations (on average ten per division) is made. A simple random sample of two of
the locations is taken. Four divisions and two locations per division mean a total of
eight locations in all. At each of these eight locations a simple random sample of
250 employees is taken from the already available computerised payroll listing. The
final sample is then 250 employees from both of two locations, in each of four
divisions, making a total of 2000 as required. However, there have been no prob-
lems of listing elements of a population and the expense of collecting data has been
reduced by restricting work to eight locations instead of 150.
There are no statistical rules for deciding how big a sample should be chosen at
each stage. Common sense and knowledge of the circumstances are usually suffi-
cient. For instance, choosing just one division out of 15 could lead to an
unrepresentative sample because of a complete bias to just one geographical area;
choosing eight divisions out of 15 means that cost reduction advantages may be lost.
Choosing four is no more than a reasonable compromise.

Cluster Sampling
Cluster sampling is closely linked with multi-stage sampling in that the population
is divided into groups, the groups into subgroups and so on. The groups are
sampled, the subgroups sampled, etc. The difference is that, at the final stage, each
individual of the chosen groups is included in the final sample. Pictorially, the final
sample would look like a series of clusters drawn from the population.

Quantitative Methods Edinburgh Business School 6/7

Module 6 / Sampling Methods

For example, suppose the company described in the previous section on multi-
stage sampling is slightly different. The locations are not offices and factories but
small retail establishments at which an average of six staff are employed. The four
divisions and two locations per division could be sampled as before, but, because
there are only six staff, further samples at each location would not be taken. All six
would be included in the sample, carrying the further advantage of ensuring that all
grades of employee are in the sample. In this case the total sample size is 4 × 2 × 6
= 48. This is two-stage cluster sampling, because two stages of sampling, of
divisions and locations, are involved. If all locations, ignoring divisions, had been
listed and sampled, the process would involve only one sampling stage and would be
called ordinary cluster sampling. If a total sample larger than 48 were required
then more divisions or more than two locations per division would have to be
selected.

Stratified Sampling
In stratified sampling prior knowledge of a population is used to make the sample
more representative. If the population can be divided into subpopulations (or strata)
of known size and distinct characteristics then a simple random sample is taken of
each subpopulation such that there is the same proportion of subpopulation
members in the sample as in the whole population. If, for example, 20 per cent of a
population forms a particular stratum, then 20 per cent of the sample will be of that
stratum. The sample is therefore constrained to be representative at least as regards
the occurrence of the different strata. Note the different role played by the strata
compared to the role of the groups in multi-stage and cluster sampling. In the
former case, all strata are represented in the final sample; in the latter, only a few of
the groups are represented in the final sample.
For an example of stratified sampling, let’s return to the previous situation of
taking a sample of 2000 from the workforce of a large company. Suppose the
workforce comprises 9 per cent management staff, 34 per cent clerical, 21 per cent
skilled manual and 36 per cent unskilled manual. It is desirable that these should all
be represented in the final sample and in these proportions. Stratified sampling
involves first taking a random sample of 180 management staff (9 per cent of 2000),
then 680 clerical, then 420 skilled and finally 720 unskilled. This does not preclude
the use of multi-stage or cluster sampling. In multi-stage sampling the divisions and
locations would be selected first, then samples taken from the subpopulations at
each location (i.e. take 22 management staff – 9 per cent of 250 – at location one, 22
at location two and so on for the eight locations).
The final sample is then structured in the same way as the population as far as
staff grades are concerned. Note that stratification is only worth doing if the strata
are likely to differ in regard to the measurements being made in the sample. Other-
wise the sample is not being made more representative. For instance, absenteeism
results are likely to be different among managers, clerks, skilled workers and
unskilled workers. If the population had been stratified according to, say, colour of
eyes, it is unlikely that absenteeism would differ from stratum to stratum, and
stratified sampling would be inapplicable.

6/8 Edinburgh Business School Quantitative Methods

 Module 6 / Sampling Methods

Weighting
Weighting is a method of recognising the existence of strata in the population after
the sampling has been carried out instead of before, as with stratification. Weighting
means that the measurements made on individual elements of the sample are
weighted so that the net effect is as if the proportions of each stratum in the sample
had been the same as those in the population. In the absenteeism investigation,
suppose that at each location the computerised payroll records did not indicate to
which staff grade those selected for the sample belonged. Only when the personnel
records were examined could this be known. Stratification before sampling is
therefore impossible. Or, at least, it would be extremely expensive to amend the
computer records. The sample of 2000 must be collected first. Suppose the strata
proportions are as in Table 6.4.

Table 6.4 Strata proportions

Stratum Population Sample Weighting
Management 9% 12% 9/12
Clerical 34% 39% 34/39
Skilled 21% 15% 21/15
Unskilled 36% 34% 36/34

The table shows the weighting that should be given to the elements of the sam-
ple. The weighting allocated to each stratum means that the influence each stratum
has on the results is the same as its proportion in the population. If the measure-
ments being made are of days absent for each member of the sample, then these
measurements are multiplied by the appropriate weighting before calculating average
days absent for the whole sample.

Probability Sampling
In simple random sampling each element of the population has an equal chance of
being selected for the sample. There are circumstances when it is desirable for
elements to have differing chances of being selected. Such sampling is called
probability sampling.
For example, when a survey is carried out into the quality and nutritional value of
school meals, a random sample of schools will have to be taken and their menus
inspected. If every school has an equal chance of being chosen, children at large
schools will be under-represented in the sample. The probability of choosing menus
from small schools is greater when a sample of schools is taken than when a sample
of schoolchildren is taken. This may be important if there is a variation in meals
between large and small establishments.
The issue hinges on what is being sampled. If it is menus, then school size is
irrelevant; if it is children subjected to different types of menu, then school size is
relevant. Probability sampling would give schools different chances of being chosen,
proportional to the size of the school (measured in terms of the number of children

Quantitative Methods Edinburgh Business School 6/9

Module 6 / Sampling Methods

attending). In the final sample, children from every size of school would have an
equal chance that their menu was selected.

Variable Sampling
In variable sampling some special subpopulation is over-sampled (i.e. deliberately
over-represented). This is done when the subpopulation is of great importance and
a normal representation in the sample would be too small for accurate information
to be gleaned.
Suppose the project is to investigate the general health levels of children who
have suffered from measles. The population is all children, of specified ages, who
have had measles. A small subpopulation (perhaps 1 per cent) is formed of children
who have suffered permanent brain damage as part of their illness. Even a large
sample of 500 would contain only five such children. Just five children would be
insufficient to assess the very serious and greatly variable effects of brain damage.
Yet this is a most important part of the survey. More of such children would
purposely be included in the sample than was warranted by the subpopulation size.
If calculations of, for instance, IQ measurements were to be made from the sample,
weighting would be used to restore the sample to representativeness.

Area Sampling
Sometimes very little may be known about the population. Simple random sampling
may be difficult and expensive, but, at the same time, lack of knowledge may
prevent a variation on simple random sampling being employed. Area sampling is
an artificial breaking down of the population to make sampling easier. The popula-
tion is split into geographical areas. A sample of the areas is taken and then further
sampling is done in the few areas selected.
The survey of users’ attitudes to a particular soap powder is such a case. A listing
of the whole population of users is impossible, yet not enough is known to be able
to use multi-stage or cluster sampling. The country/district where the sampling is to
be conducted is split into geographical areas. A sample of the areas is selected at
random and then a further sample is taken in the areas chosen, perhaps by listing
households or using the electoral roll. The difficult operation of listing the popula-
tion is reduced to just a few areas.

6.5 Judgement Sampling

Judgement sampling methods involve a significant degree of personal judgement.
They may be used when random methods are impossible (e.g. when taking a blood
sample). In other situations random sampling may be extremely expensive and a
non-random method is a cheaper alternative.
The major judgement sampling methods are as follows.
(a) Systematic sampling. In systematic sampling the sample is taken at regular
intervals from the population. If the population consisted of 50 000 and a sample
of 1000 were required, then every fiftieth item on a list of the population would
be taken. This would constitute a systematic sample. To avoid always taking the

6/10 Edinburgh Business School Quantitative Methods

 Module 6 / Sampling Methods

first name on the list, the starting point would be selected at random from the
first 50.
To return to the workforce absenteeism survey, the computerised payroll could
be sampled systematically. If the payroll had 1750 names and a sample of 250
was required, then every seventh name could be taken. Provided that the list is in
random order, the resulting sample is, in effect, random, but all the trouble of
numbering the list and using random numbers has been avoided. Systematic
sampling would usually provide a random sample if the payroll were in alphabet-
ical order.
There are, however, dangers in systematic sampling. If the payroll were listed in
workgroups, each group having six workers and one foreman, then taking one
name in seven might result in a sample consisting largely of foremen or consist-
ing of no foremen at all.
A systematic sample can therefore result in a sample that is, in effect, a random
sample, but time and effort has been saved. At the other extreme, if care is not
taken, the sample can be hopelessly biased.
(b) Convenience sampling. This means that the sample is selected in the easiest
way possible. This might be because a representative sample will result or it
might be that any other form of sampling is impossible.
In medicine, a sample of blood is taken from the arm. This is convenience sam-
pling. Because of knowledge of blood circulation, it is known that the sample is
as good as random.
As a further example, consider the case of a researcher into the psychological
effects on a family when a member of the family suffers from a major illness.
The researcher will probably conduct the research only on those families that are
in such a position at the time of the research and who are willing to participate.
The sample is obviously not random, but the researcher has been forced into
convenience sampling because there is no other choice. He or she must analyse
and interpret the findings in the knowledge that the sample is likely to be biased.
Mistaken conclusions are often drawn from such research by assuming the sam-
ple is random. For instance, it is likely that families agreeing to be interviewed
have a different attitude to the problem of major illness from families who are
unwilling.
(c) Quota sampling. Quota sampling is used to overcome interviewer bias. It is
frequently used in market research street interviews. If the interviewer is asked to
select people at random, it is difficult not to show bias. A young male interviewer
may, for instance, select a disproportionate number of attractive young females
for questioning.
Quota sampling gives the interviewer a list of types of people to interview. The
list may be of the form:
10 males age 35–50
10 females age 35–50
10 males age 50+
10 females age 50+
Total sample = 40
Discretion is left to the interviewer to choose the people in each category.

Quantitative Methods Edinburgh Business School 6/11

Module 6 / Sampling Methods

Note that quota sampling differs from stratified sampling. A stratum forms a
known proportion of the population; quota proportions are not known. Any
conclusions must be interpreted in the context of an artificially structured, non-
random sample.

6.6 The Accuracy of Samples

If a method of sample collection existed that could guarantee to produce representa-
tive samples every time, then the problem of inference would be a simple one. For
example, suppose the average income of all the 20 000 students at London Universi-
ty is to be estimated. By some means a sample of 200 students that is fully
representative of the population (the 20 000 students) has been collected. If the
average income measured over the sample is £6000, then, since the sample is fully
representative, the average income for the population must also be £6000.
The fact that no sample can definitely be known to be representative means that
there is a need to estimate the errors that may arise. Estimating these errors is a
fundamental part of statistical inference. The statistical theory by which this can be
done will be covered later, but its result can be appreciated now. In the student
income example the sample mean was £6000. The true population mean was
unknown, but statistical theory allows statements to be made such as: ‘There is a 95
per cent probability that the true population average income lies in the range £6000
±£200, i.e. between £5800 and £6200.’ Such statements can only be made when
random sampling methods are used. In other words, random sampling not only
gives the single estimate of £6000 but, in addition, gives a measure of the accuracy
of the estimate (±£200). The statistical theory on which such accuracy calculations
are based rests on the random selection of the sample. Accuracy levels cannot be
estimated from judgement samples since they are not essentially random in nature.
Stratified sampling, besides its other advantages, increases the level of accuracy of
estimates as against that of a simple random sample. This can be demonstrated with
a simplified example. Suppose the population being sampled comprises four
managers. Their average salary is being estimated. The population is stratified by
virtue of two of the managers having accounting qualifications.

Manager Salary (£000s) Stratum

A 32 1 (Accounting qualification)
B 24 1 (Accounting qualification)
C 20 2 (No qualification)
D 16 2 (No qualification)

If samples of size 2 are taken and the average salary of each sample is computed,
then using simple random sampling there will be six possible samples:

6/12 Edinburgh Business School Quantitative Methods

 Module 6 / Sampling Methods

Sample Average salary

A, B 28 (= (32 + 24)/2)
A, C 26
A, D 24 Range of sample averages: 18 → 28
B, C 22
B, D 20
C, D 18

Under stratified sampling, where each sample of two comprises one manager
from each stratum, there are four possible samples:

Sample Average salary

A, C 26
A, D 24 Range of sample averages: 20 → 26
B, C 22
B, D 20

If a sample of size 2 were being used to estimate the population average salary
(23), simple random sampling could give an answer in the range 18 → 28, but
stratified sampling is more accurate and could only give a possible answer in the
range 20 → 26. On a large scale (in terms of population and sample size), this is the
way in which stratified sampling improves the accuracy of estimates.
In summary, simple random sampling allows (in a way as yet unspecified) the
accuracy of a sample estimate to be calculated. Stratified sampling, besides its other
advantages, improves the accuracy of sample estimates.

6.7 Typical Difficulties in Sampling

6.7.1 Sampling Frame
The sampling frame is the complete list from which the sample is selected. This
sounds like the population, but in practice the two are likely to be different. A
sizeable difference between the two may mean that the results of the sampling are
unreliable.
In the usual example of the absenteeism survey, the population is the entire
workforce of the company. However, computer records will not be precisely up to
date as people join and leave the company. In addition, the payroll listings may
differ from location to location as regards such things as the inclusion or exclusion
of part-time staff. Factors such as these result in a sampling frame that is likely to
differ from the population. The difference may be important. Absenteeism among
part-time staff may not be the same as for full-time staff and care would need to be
taken in the interpretation of the results.

Quantitative Methods Edinburgh Business School 6/13

Module 6 / Sampling Methods

6.7.2 Non-response
Envisage a sample of households with which an interviewer has to make contact to
ask questions concerning the consumption of breakfast cereals. If there is no one at
home when a particular house is visited, then there is non-response. If the house is
revisited (perhaps several times are necessary) then the sampling becomes very
expensive; if the house is not revisited then it will be omitted from the sample and
the sample is likely to be biased. The bias occurs because the households where no
one is in may not be typical of the population. For instance, if the interviewer made
his calls during the daytime then households where both marriage partners are at
work during the day would tend to be omitted, possibly creating bias in the results.
The information from the sample could not properly be applied to the whole
population but only parts of it.
Non-response does not occur solely when house visits are required, but whenev-
er the required measurements cannot be made for some elements of the sample. In
the absenteeism example, missing personnel records for some of the staff selected
for the sample would constitute a non-response problem. If efforts are made to find
or reconstruct the missing records it will be expensive; if they are left out of the
sample it will be biased since, for instance, the missing records may belong mainly to
new employees.

6.7.3 Bias
Bias is a systematic (i.e. consistent) tendency for a sampling method to produce
samples that over- or under-represent elements of the population. It can come from
many sources, and sampling frame error and non-response, as described above, are
two of them. Other sources include:
(a) Inaccurate measurement. This might be physical inaccuracy, such as a
thermometer that is calibrated 1 degree too high; it might be conceptual, such as
a salary survey that excludes perks and commission bonuses.
(b) Interviewer bias. This is when the interviewer induces biased answers. For
example, an interviewer may pose a question aggressively: ‘You don’t believe the
new job conditions are advantageous, do you?’, signalling that a particular answer
is required.
(c) Interviewee bias. Here the interviewee injects the bias. The interviewee may be
trying to impress with the extent of his knowledge and falsely claim to know
more than he does. Famously, a survey of young children’s video viewing habits
seemed to indicate that most were familiar with ‘adult’ videos until it was realised
that the children interviewed were anxious to appear more grown-up than they
were and made exaggerated claims. People of all ages tend to give biased answers
in areas relating to age, salary and sexual practice.
(d) Instrument bias. The instrument refers to the means of collecting data, such as
a questionnaire. Poorly constructed instruments can lead to biased results. For
example, questions in the questionnaire may be badly worded. ‘Why is Brand X
superior?’ raises the question of whether Brand X actually is thought to be supe-
rior. Bad questions may not just be a question of competence: they may indicate

6/14 Edinburgh Business School Quantitative Methods

 Module 6 / Sampling Methods

a desire to ‘rig’ the questionnaire to provide the answers that are wanted. It is
always advisable to ask about the source of funding for any ‘independent’ survey.
Bias is most dangerous when it exists but is not recognised. When this is the case
the results may be interpreted as if the sample were truly representative of the
population when it is not, possibly leading to false conclusions.
When bias is recognised, it does not mean that the sample is useless, merely that
care should be taken when making inferences about the population. In the example
concerned with family reactions to major illness, although the sample was likely to
be biased, the results could be quoted with the qualification that a further 20 per
cent of the population was unwilling to be interviewed and the results for them
might be quite different.
In other circumstances a biased sample is useful as a pilot for a more representa-
tive sampling study.

6.8 What Sample Size?

One of the most important sampling questions concerns the size of the sample. In
our example, why choose a sample of 2000? Why not 1000 or 5000? There are two
approaches to this question.
The first is to ask what level of accuracy is needed in the results. For example, an
opinion poll to predict the level of support for a political party may give the answer
‘53% of the electorate intends to vote for the party’. What level of accuracy do we
require to be associated with this figure? If the accuracy of the poll was ‘53% ±
20%’ then the result would not be very useful in predicting the result of an election
or gauging changes in support as time went by. On the other hand, if the accuracy
was ‘53% ± 1%’ then the result would be useful for both purposes. Similarly, in a
survey of family expenditure a result that said ‘The average family spends £120 per
week on food’ would be useful if the accuracy was ‘± £1.50’ but useless if the
accuracy was ‘± £50’.
Once we know what accuracy we require to meet the purposes of the enquiry, we
can turn to statistical theory to calculate the sample size – which is related to
accuracy. The statistical theory is described in later modules but it works roughly as
follows.
With a random sample, the accuracy of the results can be calculated. It is possible
to say, with 95 per cent probability, how close to the calculated sample statistic the
true population value is likely to be. The theory (to be explained later) shows that
the larger the sample size, the closer is the estimate to the true population. In fact,
the accuracy increases (i.e. error decreases) in proportion to the square root of the
sample size. Consequently, increasing the sample from 100 to 400 (a factor of four)
doubles the accuracy (i.e. halves the error: a factor of √4 = 2); increasing it from
400 to 1600 doubles it again, and so on. There are, therefore, decreasing returns
from enlarging the sample. The answer to the question ‘What should the sample size
be?’ is this: decide what accuracy is required for the objectives of the study and then
use statistical theory to calculate the sample size that will provide this accuracy.

Quantitative Methods Edinburgh Business School 6/15

Module 6 / Sampling Methods

The second answer to the sample size question is not theoretical but is more
frequently met in practice. It is to collect the largest sample that the available budget
allows. Then the results can be interpreted in the light of the accuracy given by the
sample size.
It may seem surprising that the population size has no bearing on the sample
size. However, the population size has an effect in a different way. The definition of
simple random sampling (which is also involved in other types of sampling) is that
each member of the population should have an equal chance of being chosen.
Suppose the population is small, with only 50 members. The probability of the first
member being chosen is 1/50, the second is 1/49, the third is 1/48 and so on (i.e.
the probabilities are not equal). For a large population the difference in probabilities
is negligible and the problem is ignored. For a small population there are two
options. First, sampling could be done with replacement. This means that when an
element of the sample has been chosen, it is then returned to the population. The
probabilities of selection will then all be equal. This option means that an element
may be included in the sample twice. The second option is to use a different theory
for calculating accuracy and sample size. This second option is based on sampling
without replacement and is more complicated. A small population, therefore,
while not affecting sample size, does make a difference to the nature of sampling.

Learning Summary
It is most surprising that information collection should be so often done in apparent
ignorance of the concept of sampling. Needing information about invoices, one
large company investigated every single invoice issued and received over a three-
month period: a monumental task. A simple sampling exercise would have reduced
the cost to around 1 per cent of the actual cost with little or no loss of accuracy.
Even after it is decided to use sampling, there is still, obviously, a need for careful
planning. This should include a precise timetable of what and how things are to be
done. The crucial questions are: ‘What are the exact objectives of the study?’ and
‘Can the information be provided from any other source?’ Without this careful
planning it is possible to collect a sample and then find the required measurements
cannot be made. For example, having obtained a sample of 2000 of the workforce,
it may be found that absence records do not exist, or it may be found that another
group in the company carried out a similar survey 18 months before and their
information merely needs updating.
The range of uses of sampling is extremely wide. Whenever information has to
be collected, sampling can prove valuable. The following list gives a guide to the
applications that are frequently encountered:
(a) opinion polls of political and organisational issues;
(b) market research of consumer attitudes and preferences;
(c) medical investigations;
(d) agriculture (crop studies);
(e) accounting;
(f) quality control (inspection of manufactured output);

6/16 Edinburgh Business School Quantitative Methods

 Module 6 / Sampling Methods

(g) information systems.

In all applications, sampling is a trade-off between accuracy and expense. By
sampling at all, one is losing accuracy but saving money. The smaller the sample, the
greater the accuracy loss, but the greater the saving. The trade-off has to be made in
consideration of the accuracy required by the objectives of the study and the budget
available. Even when the full population is investigated, however, the results will not
be entirely accurate. The same problems that occur in sampling – non-response, the
sampling frame and bias – will occur with the full population. The larger the sample
size, the closer the accuracy to the maximum accuracy that is obtainable with the full
population. It may even be the case that measurement errors may overwhelm
sampling errors. For instance, even a slightly ambiguous set of questions in an
opinion poll can distort the results far more than any inaccuracy resulting from
taking a sample rather than considering the full population. The concern of many
managers that sample information must be vastly inferior to population information
is ill-founded. A modest sample can provide results that are of only slightly lower
accuracy than those provided by the whole population and at a fraction of the cost.

Quantitative Methods Edinburgh Business School 6/17

Module 6 / Sampling Methods

Review Questions
6.1 Which reason is correct?
The need to take samples arises because:
A. it is impossible to take measurements of whole populations.
B. sampling gives more accurate results.
C. sampling requires less time and money than measuring the whole population.

6.2 Which reason is correct?

Randomness of selection is incorporated into many sampling methods in order that:
A. the accuracy of inferences can be calculated.
B. the sample will be totally representative of the population.
C. the cost of sampling is reduced.
D. it is not necessary to list the entire population.

6.3 A British company divides the country into nine sales regions. Three of them are to be
selected at random and included in an information exercise. Using the table of random
numbers, what are the regions chosen? (Take the numbers starting at the top left, a row
at a time.)

Random Numbers
5 8 5 0 4
7 2 6 9 3
6 1 4 7 8

Sales regions
1. North West England
2. Eastern England
3. Midlands
4. London
5. South East England
6. South West England
7. Wales
8. Scotland
9. Northern Ireland
The sample will comprise:
A. SE England, Scotland, SE England.
B. SE England, Scotland, London.
C. Scotland, SE England, NW England.
D. SE England, Wales, SW England.
E. NW England, N Ireland, NW England.

6/18 Edinburgh Business School Quantitative Methods

 Module 6 / Sampling Methods

6.4 The advantages of multi-stage sampling over simple random sampling are:
A. fewer observations need to be made.
B. the entire population does not have to be listed.
C. it can save time and effort by restricting the observations to a few areas of the
population only.
D. the accuracy of the results can be calculated more easily.

6.5 Which of the following statements about stratified sampling are true?
A. It is usually more representative than a simple random sample.
B. It cannot also be a cluster sample.
C. It may be more expensive than a simple random sample.

6.6 Which of the following statements about variable sampling fractions is true?
A. Measurements on one part of the sample are taken extra carefully.
B. A section of the population is deliberately over-represented in the sample.
C. The size of the sample is varied according to the type of items so far selected.
D. The results from the sample are weighted so that the sample is more repre-
sentative of the population.

6.7 Where are non-random (judgement) methods of sampling used? When:

A. random sampling is impossible.
B. non-random methods provide a representative sample at lower cost.
C. there is a likelihood of interviewer bias.

6.8 What is the essential difference between stratified and quota sampling?
A. Quota sampling refers only to interviewing, whereas stratification can apply to
any sampling situation.
B. Stratification is associated with random selection within strata, whereas quota
sampling is unconnected with random methods.
C. The strata sizes in the sample correspond to the population strata sizes,
whereas the quota sizes are fixed without necessarily considering the popula-
tion.

Quantitative Methods Edinburgh Business School 6/19

Module 6 / Sampling Methods

6.9 A sample of trees is to be taken (not literally) from a forest and their growth monitored
over several years. In the forest 20 per cent of the trees are on sloping ground and 80
per cent are on level ground. A sample is taken by first dividing the forest into geo-
graphical areas of approximately equal size. There are 180 such areas (36 sloping, 144
level). Three sloping areas and 12 level ones are selected at random. In each of these 15
areas, 20 trees are selected at random for inclusion in the sample.
The resulting sample of 300 (20 × 15) trees can be said to have been selected using
which of the following sample methods?
A. Multi-stage.
B. Cluster.
C. Stratification.
D. Weighting.
E. Probability.
F. Area.
G. Systematisation.
H. Convenience.

6.10 A random sample of 25 children aged 10 years is taken and used to measure the average
height of 10-year-old children with an accuracy of ±12 cm (with 95 per cent probability
of being correct). What would have been the accuracy had the sample size been 400?
A. ±3 cm
B. ±0.75 cm
C. ±48 cm
D. ±4 cm

Case Study 6.1: Business School Alumni

1 A French business school wishes to gain some information on the careers of its
graduates. It decides to take a sample of its graduates and send them a questionnaire.
The sample is selected as follows.
After a random start, every twentieth name on the list of the Alumni Association is
taken, together with every twentieth name on the list of graduates whose addresses are
known but who did not join the Association. Both lists are alphabetical. In all, 1200
usable replies were received. Does this constitute a random sample?

Case Study 6.2: Clearing Bank

1 A UK clearing bank with more than 2200 branches and 7 million customers wants to
take a representative sample of personal customer chequebook accounts. Measure-
ments taken from the sample will include average balances, numbers of transactions,
profitability and usage of services. This information will be used for setting bank charges,
assessing the effect of variations in interest rates and formulating marketing plans.
The bank is split into 14 geographical regions. Each region has between 100 and 500
branches of varying size. Over the whole country the profile of bank sizes is as follows:

6/20 Edinburgh Business School Quantitative Methods

 Module 6 / Sampling Methods

Branch size No. of branches %

Large (average 5000 accounts) 697 32
Medium (average 2500 accounts) 728 34
Small (average 1000 accounts) 740 34
2165 100

It is thought that branch size affects the profitability of customer accounts, because of
differing staff ratios and differing ranges of services being available at branches of
different size. Each branch has between 100 and 15 000 chequebook accounts. All
accounts are computerised regionally. Each region has its own computer, which can
provide a chronological (date of opening account) list of chequebook accounts and from
which all the necessary information on average balances and so on can be retrieved.
a. Why should the bank adopt a sampling approach rather than taking the information
from all account holders?
b. In general terms, what factors would influence the choice of sample size?
c. If a total sample of 2000 is to be collected, what sampling method would you
recommend? Why?
d. In addition, the bank wants to use the information from the sample to compare the
profitability of chequebook accounts for customers of different socioeconomic
groups. How would you do this? What extra information would you require? NB:
An account holder’s socioeconomic group can be specified by reference to his/her
occupation. The bank will classify into five socioeconomic groups.
e. What practical difficulties do you foresee?

Quantitative Methods Edinburgh Business School 6/21

PART 3

Statistical Methods
Module 7 Distributions
Module 8 Statistical Inference
Module 9 More Distributions
Module 10 Analysis of Variance

Quantitative Methods Edinburgh Business School

Module 7

Distributions
Contents
7.1 Introduction.............................................................................................7/1
7.2 Observed Distributions ..........................................................................7/2
7.3 Probability Concepts ..............................................................................7/8
7.4 Standard Distributions ........................................................................ 7/14
7.5 Binomial Distribution .......................................................................... 7/15
7.6 The Normal Distribution .................................................................... 7/19
Learning Summary ......................................................................................... 7/27
Review Questions ........................................................................................... 7/29
Case Study 7.1: Examination Grades ........................................................... 7/31
Case Study 7.2: Car Components................................................................. 7/31
Case Study 7.3: Credit Card Accounts ........................................................ 7/32
Case Study 7.4: Breakfast Cereals ................................................................ 7/32

Prerequisite reading: Module 1, Module 5

Learning Objectives
By the end of the module the reader should be aware of how and why distributions,
especially standard distributions, can be useful. Having numbers in the form of a
distribution helps both in describing and in analysing. Distributions can be formed
from collected data, or they can be derived mathematically from knowledge of the
situation in which the data are generated. The latter are called standard distributions.
Two standard distributions, the binomial and the normal, are the main topics of the
module.
Proof of some of the formulae requires a high level of mathematics. Where pos-
sible, mathematical derivations are given but, since the purpose of this course is to
explore the practical applications of techniques, not their historical and mathemati-
cal development, there are situations where the mathematics is left in its ‘black box’.
Note on technical sections: Section 7.3 ‘Probability Concepts’, Section 7.5.3
‘Deriving the Binomial Distribution’ and Section 7.6.3 ‘Deriving the Normal
Distribution’ are technical and may be omitted on a first reading.

7.1 Introduction
Several examples of distributions have been encountered already. In Module 5, on
summary measures, driving competition scores formed a symmetrical distribution;
viewing of television serial episodes formed a U-shaped distribution; sickness

Quantitative Methods Edinburgh Business School 7/1

Module 7 / Distributions

records formed a reverse J-shaped distribution. These distributions, particularly their

shapes, were used as part of the description of the data. A closer look will now be
taken at distributions and their use in the analysis of data.
There are two general types of distribution, observed and standard. The latter is
also known as theoretical or probability distribution. An observed distribution is
derived from collecting data in some situation; the distribution is peculiar to that
one situation. A standard distribution is derived mathematically and is, theoretically,
applicable to all situations exhibiting the appropriate mathematical characteristics.
Standard distributions have the advantage that time and effort spent on data
collection are saved. The formation and application of both types of distribution will
be described in this module.

7.2 Observed Distributions

An observed distribution starts with the collection of numbers. The numbers are all
measurements of a variable, which is just what the word implies. It is some entity
that can be measured and for which the measurement varies when observations are
made of it. For instance, the variable might be the market share of a well-known
soft drink at 100 of the producer’s European marketing centres, as shown in
Figure 7.1

7/2 Edinburgh Business School Quantitative Methods

 Module 7 / Distributions

6
5

9
11
8

11 12
12 11 10

18 16 14
16 17 15
17 15
16
19
17 13 23 15
18 19
20 21 16 19 17
20
19 18 18
19 20 18
20 18
21 19
13 23 23
22 24 19
13 22 20
23 23
24 20
25 24 26 25
22 24 24
23 25 24 25 20 25
27 24
18 25
30 16 28 25
25 29 25 25
26
16 29 28 26 27
29 27 27
26 27 28 28 31
4
34 32
8
28
36

Figure 7.1 Market share (%) of soft drink product throughout Europe
Figure 7.1 is a mess. At first, most data are. They may have been taken from large
market research reports, dog-eared production dockets or mildewed sales invoices;
they may be the output of a computer system where no effort has been made at data
communication. Some sorting out must be done. A first attempt might be to arrange
the numbers in an ordered array, as in Table 7.1.

Table 7.1 Ordered array

4 11 15 16
5 11 15 …
6 12 15 …
8 12 16 …
8 13 16 …
9 13 16 …
10 13 16 …
11 14 16 …

Quantitative Methods Edinburgh Business School 7/3

Module 7 / Distributions

The numbers look neater now, but it is still difficult to get a feel for the data –
the average, the variability, etc. – as they stand. The next step is to classify the data.
Classifying means grouping the numbers in bands, such as 20–25, to make them
easier to appreciate. Each class has a frequency, which is the number of data points
(market shares) that fall within that class. A frequency table is shown in Table 7.2.

Table 7.2 Frequency table

Class Frequency
0<x≤5 2
5 < x ≤ 10 5
10 < x ≤ 15 12
15 < x ≤ 20 31
20 < x ≤ 25 27
25 < x ≤ 30 19
30 < x ≤ 35 3
35 < x ≤ 40 1
Total 100

30

25
Frequency

20

15

10

0 5 10 15 20 25 30 35 40
Market shares (%)

Figure 7.2 Frequency histogram of soft drink market shares

Table 7.2 shows that 12 data points (market shares at 12 centres) were greater
than 10 but less than or equal to 15, 31 were greater than 15 but less than or equal
to 20 and so on. Data points are sometimes referred to as observations or read-
ings. It is now easier to get an overall impression of what the data mean. Most of
the numbers are between 15 and 25 with extremes of just above 0 and just below
40. Another arrangement, a frequency histogram as in Figure 7.2, has even greater
visual impact.
Figure 7.2 makes it easy to see that the data are spread symmetrically over their
range with the majority falling around the centre of it. As a descriptive device the
frequency histogram is satisfactory without further refinement. If, on the other
hand, the exercise had analytical objectives, then it would be developed into a

7/4 Edinburgh Business School Quantitative Methods

 Module 7 / Distributions

probability histogram. The transformation uses the relative frequency method of

probability measurement. The probability that any randomly selected measurement
lies within a particular class interval can be calculated:
(number lies in class ) =
The frequency histogram becomes a probability histogram by making the units of
the vertical axis probabilities instead of frequencies. Since the total frequency is 100,
the frequency 2 becomes the probability 0.02, 4 becomes 0.04, 12 becomes 0.12, etc.
The shape of the histogram remains the same. Once the histogram is in probability
form it is referred to as a distribution, although, to be strictly accurate, all configura-
tions in Figure 7.1 and Figure 7.2, and in Table 7.1 and Table 7.2, are distributions
of some sort. Analyses, based on the manipulation of probabilities, can now be
carried out. For example, what is the probability that any market share will be in the
range 10–20? From Figure 7.2:
(10 < ≤ 15) = 0.12
(15 < ≤ 20) = 0.31
These probabilities are calculated from frequencies of 12 and 31 respectively for
each class. The frequency for the combined class 10–20 is therefore 43 (= 12 + 31).
Consequently:
( )
(10 < ≤ 20) = = 0.43

Alternatively, this probability can be calculated by adding the probabilities of the

individual classes:
(10 < ≤ 20) = 0.12 + 0.31 = 0.43
As a further example, what is the probability that any market share is 15 or less?
( ≤ 15) = (0 < ≤ 5) + (5 < ≤ 10) + (10 < ≤ 15)
= 0.02 + 0.05 + 0.12
= 0.19
The calculations can be made in either frequency or probability form. Probability
calculations such as these form the basis of analyses with observed distributions.
Example
Table 7.3 shows the number of deliveries of food made to a hypermarket (the number
of trucks arriving at the hypermarket) per day over a one-year period (300 working
days). The frequencies have been turned into percentages and then probabilities. For
instance, on 53 days there were between zero and nine deliveries and therefore the
probability of there being deliveries in this range on any day is 53/300 = 0.18 (approx.).
Figure 7.3 shows a probability histogram of the same data.

Table 7.3 Frequency table of deliveries

No. of No. of % Probability
deliveries days
0–9 53 18 0.18
10–19 82 27 0.27

Quantitative Methods Edinburgh Business School 7/5

Module 7 / Distributions

20–29 75 25 0.25
30–39 42 14 0.14
40–49 29 10 0.10
50+ 19 6 0.06
Total 300 100 1.00

Probability 0.25

0.20

0.15

0.10

0.05

0–9 10–19 20–29 30–39 40–49 50+

Figure 7.3 Probability histogram of deliveries

(a) The facilities at the hypermarket are only sufficient to handle 29 deliveries per day.
Otherwise, overtime working is necessary. What is the probability that on any day
overtime will be required?
From Figure 7.3:
P(deliveries are 50 or more) = 0.06
P(deliveries 40–49) = 0.10
P(deliveries 30–39) = 0.14
Therefore:
P(deliveries exceed 29) = 0.06 + 0.10 + 0.14
= 0.30
There is a 30 per cent probability that overtime will be needed on any day.
(b) The capacity can be increased from the present 29 by taking on more staff and
buying more handling equipment. To what level should the capacity be raised if there
is to be no more than a 10 per cent chance of overtime being needed on any day?
A capacity level X is to be found such that:
P(deliveries exceed X) = 0.10
Since:
P(deliveries 50 or more) = 0.06
P(deliveries 40–49) = 0.10
X must lie between 40 and 49. To find X, assumptions have to be made. Making the
assumption that deliveries in the range of 40–49 are spread evenly over this range,
then:
P(deliveries 40–45) = 0.06
P(deliveries 46–49) = 0.04

7/6 Edinburgh Business School Quantitative Methods

 Module 7 / Distributions

Consequently:
P(deliveries exceed 45) = P(deliveries 46–49) + P(deliveries 50 or more)
= 0.04 + 0.06
= 0.10
Therefore, X = 45.
The capacity of 45 will result in overtime being needed on 10 per cent of days.
An alternative representation of a frequency distribution is as a cumulative frequency
distribution. Instead of showing the frequency for each class, a cumulative frequency
distribution shows the frequency for that class and all smaller classes. For example,
instead of recording the number of days on which the deliveries of food to a hypermar-
ket were in the ranges 0–9, 10–19, 20–29, etc., a cumulative distribution records the
number of days when deliveries were less than 9, less than 19, less than 29, etc.
Table 7.4 turns Table 7.3 into cumulative form.

Table 7.4 Cumulative frequencies of deliveries

No. of Cumulative % Probability
deliveries (f) no. of days P(no. deliveries ≤ f)
≤9 53 18 0.18
≤ 19 135 45 0.45
≤ 29 210 70 0.70
≤ 39 252 84 0.84
≤ 49 281 94 0.94
All 300 100 1.00

Table 7.4 shows cumulative frequencies in ‘less than or equal to’ format. They could just
as easily be in ‘more than or equal to’ format, as in Table 7.5.

Table 7.5 Cumulative frequencies of deliveries

No. of Cumulative % Probability
deliveries (f) no. of days P(no. deliveries ≥ f)
≥0 300 100 1.00
≥ 10 247 82 0.82
≥ 20 165 55 0.55
≥ 30 90 30 0.30
≥ 40 48 16 0.16
≥ 50 19 6 0.06

These cumulative frequency tables can be put into the form of graphs. They are then
known as ogives. Figure 7.4 shows Table 7.4 and Table 7.5 as ogives.

Quantitative Methods Edinburgh Business School 7/7

Module 7 / Distributions

less than
more than

300

250

200
Frequency

150

100

50

0
0 9 19 29 39 49
Deliveries

Figure 7.4 Ogives

Typical questions, such as ‘If the facilities can handle 29 deliveries per day, what is the
probability of this capacity being exceeded?’, can be answered more easily when the data
are in cumulative form or displayed as an ogive.

7.3 Probability Concepts

Before we move from observed distributions to standard distributions, some
preliminary work on probability concepts is needed. This will be of value in coming
to understand the derivation of standard distributions.
The fundamentals of statistical probability were discussed in the first module.
Probability is the likelihood of an event taking place; it takes values between 0 and 1;
it can be measured by three methods: a priori, relative frequency and subjective
assessment. These basic ideas will now be extended to some further properties of
probability. At first this will be done in abstract by thinking of hypothetical events
referred to as A, B, C, … Examples later will make the ideas more concrete.
Events A, B, C, … are mutually exclusive if the occurrence of one automatical-
ly rules out the occurrence of others. For example, suppose the events refer to a
company’s profit in one year’s time: A is a loss or breakeven, B is a profit greater
than 0 but less than £100 000, C is a profit of at least £100 000 but less than
£200 000, D is a profit of at least £200 000 but less than £300 000, and so on. The
events are mutually exclusive since the occurrence of one automatically rules out the
others. If an additional event (AA) were the company being made bankrupt, then
the events AA, A, B, C, … are not mutually exclusive, since, for instance, it is

7/8 Edinburgh Business School Quantitative Methods

 Module 7 / Distributions

possible for the company to make a loss (event A) and be made bankrupt (event
AA).
The addition law for mutually exclusive events is:
(A or B or C or … ) = (A) + (B) + (C) + ⋯
This law has already been used implicitly in the example on hypermarket deliver-
ies. The probabilities of different classes were added together to give the probability
of an amalgamated class. The law was justified by relating the probabilities to the
frequencies from which they had been derived.
Example
Referring to the example on hypermarket deliveries (Table 7.3), what is the probability
of fewer than 40 deliveries on any day?
The events (the classes) in Table 7.3 are mutually exclusive. For example, given that the
number of deliveries on a day was in the range 10–19, it is not possible for that same
number of deliveries to belong to any other class. The addition law, therefore, can be
applied:
P(fewer than 40 deliveries) = P(0–9 deliveries or 10–19 or 20–29 or 30–39
= P(0–9) + P(10–19) + P(20–29) + P(30–39)
= 0.18 + 0.27 + 0.25 + 0.14
= 0.84

Equally, the result could have been obtained by working in the frequencies from which
the probabilities were obtained.
Note that if events are not mutually exclusive, this form of the addition law cannot be
used.
A conditional probability is the probability of an event under the condition that
another event has occurred or will occur. It is written in the form P(A/B), meaning the
probability of A given the occurrence of B. For example, if P(rain) is the probability of
rain later today, P(rain/dark clouds now) is the conditional probability of rain later
given that the sky is cloudy now.
The probabilities of independent events are unaffected by the occurrence or non-
occurrence of the other events. For example, the probability of rain later today is not
affected by dark clouds a month ago and so the events are independent. On the other
hand, the probability of rain later today is affected by the presence of dark clouds now.
These events are not independent. The definition of independence is based on condi-
tional probability. Event A is independent of B if:
(A) = (A/B)
This equation is merely the mathematical way of saying that the probability of A is not
affected by the occurrence of B. The idea of independence leads to the multiplication
law of probability for independent events:
(A and B and C and … ) = (A) × (B) × (C) × …
The derivation of this law will be explained in the following example.

Quantitative Methods Edinburgh Business School 7/9

Module 7 / Distributions

Example
Twenty per cent of microchips produced by a certain process are defective. What is the
probability of picking at random three chips that are defective, defective, OK, in that
order?
The three events are independent since the chips were chosen at random. The multipli-
cation law can therefore be applied.
(1st chip defective and 2nd chip defective and 3rd chip OK)
= (defective) × (defective) × (OK)
= 0.2 × 0.2 × 0.8
= 0.032
This result can be verified intuitively by thinking of choosing three chips 1000 times.
According to the probabilities, the 1000 selections are likely to break down as in
Figure 7.5. Note that in a practical experiment the probability of, for instance, 20 per
cent would not guarantee that exactly 200 of the first chips would be defective and 800
OK, but it is the most likely outcome. Figure 7.5 shows 32 occasions when two
defectives are followed by an OK; 32 in 1000 is a proportion of 0.032 as given by the
multiplication law.

1st chip 2nd chip 3rd chip

128 OK
160 OK
32 defective
32 OK
200 defective 40 defective
8 defective
1000 selections
512 OK
800 OK 640 OK
128 defective
128 OK
160 defective
32 defective

Figure 7.5 Breakdown of probabilities

Closely associated with probability calculations is the idea of a combination. It is
concerned with the sequencing of objects selected in a sample. A combination is
defined as the number of different ways in which r objects can be chosen from a
total of n objects. is the mathematical notation for a combination. An example
will explain the idea further: how many ways are there of choosing a subcommittee
of two people from a full committee of four?
The problem is tackled by first looking into how many orderings or sequences
the four can be put. Suppose the four people are labelled A, B, C, D. There are four
possible candidates for the first place. Once the first place has been decided, there
remain three possible candidates for the second place, leaving two for the third and
finally there is just one possibility for the last place. In total, this is 4 × 3 × 2 × 1 =
24 ways of sequencing the four people. This is shown diagrammatically in Fig-
ure 7.6. The expression 4 × 3 × 2 × 1 is known as 4 factorial (written ‘4!’).

7/10 Edinburgh Business School Quantitative Methods

 Module 7 / Distributions

4 possibilities for 1st place

A B C D

3 possibilities for 2nd place

B C D A C D A B D A B C
2 possibilities for 3rd place

C D B D B C C D A D A C B D A D A B B C A C A B
1 possibility for 4th place

D C D B C B D C D A C A D B D A B A C B C A B A

Figure 7.6 Ways of ordering four people

More generally:
factorial = ! = × −1× − 2 × …× 3 × 2 × 1
From the bottom row of Figure 7.6, there are seen to be 24 ways of sequencing
four people. If the first two are separated from the others to form the subcommit-
tee, as in Table 7.6, it may appear to the cursory view that there are 4! ways of
forming the subcommittee. This is not so, because there is repetition. First, repeti-
tion occurs because the order of the two people left out does not affect the
subcommittee. In Table 7.6, the first two sequences (numbers 1 and 2) are the same
as regards the subcommittee. Both start with A, B. They appear as separate se-
quences only because the two people left out are in the order C, D and D, C
respectively. This situation occurs whatever two people are in the subcommittee
places. Each of the pairs of orderings 1,2; 3,4; 5,6; 7,8, etc. provides just one
subcommittee. The number of different subcommittees can be halved from 24 to 12
as a result of this effect.

Table 7.6 Selecting two from four

A A A A A A B B B B B B C C C C C C D D D D D D
B B C C D D A A C C D D A A B B D D A A B B C C
C D B D B C C D A D A C B D A D A B B C A C A B
D C D B C B D C D A C A D B D A B A C B C A B A
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24

Second, repetition occurs because the order of the two selected for the subcom-
mittee does not affect its constitution. It is the same subcommittee of A and B
whether they were selected first A then B or vice versa. In Table 7.6 each pair of
orderings 1,7; 2,8; 3,14 etc. provides just one subcommittee. Consequently, the
number of subcommittees can be further halved from 12 to 6. Since the process
started with all possible sequences of the four candidates, and since allowance has
now been made for all repetitions, this is the final number of different subcommit-
tees. They are:
AB, AC, AD, BC, BD, CD

Quantitative Methods Edinburgh Business School 7/11

Module 7 / Distributions

Put more concisely, the number of ways of choosing these six was calculated
from:

Using factorial notation this can be written:

!
= = =6
!× ! ×

More generally, the number of ways of selecting r objects from a larger group of
n objects is:
!
=
( !×( )!)

This can be verified intuitively by using numbers instead of n and r in a similar

manner to the subcommittee example. (Proof of the combination formula involves
some difficult mathematics and is not given.)
Example
What are the probabilities of obtaining 0, 1, 2, 3, 4 sixes when four dice are rolled
simultaneously?
The probability of four sixes is obtained from the multiplication law for independent
events. The events (the outcome for each die) are independent since the rolling of one
is not affected by the rolling of any other. The law is therefore applicable.
(4 sixes) = (6 on 1st die) × (6 on 2nd) × (6 on 3rd) × (6 on 4th)
= × × ×
= 0.001 (approx. )
The probability of zero sixes is calculated similarly:
(0 sixes) = (1 … 5 on 1st die) × (1 … 5 on 2nd)
× (1 … 5 on 3rd) × (1 … 5 on 4th)
= × × ×
= 0.482
An attempt to calculate the probability of one six could be made as follows:
(1 six) = (6 on 1st die) × (1 … 5 on 2nd)
× (1 … 5 on 3rd) × (1 … 5 on 4th)
= × × ×
= 0.0965
However, this only allows for the one six being on the first die; the six could have been
on any die. Using the addition law:
(the 6 is on 1st or 2nd or 3rd or 4th die) = (1 six)
= (the 6 on 1st die) + (the 6 on 2nd) + (the 6 on 3rd) + (the 6 on 4th)
= 0.0965 × 4
= 0.386
An attempt to calculate the probability of two sixes could be as follows:

7/12 Edinburgh Business School Quantitative Methods

 Module 7 / Distributions

(2 sixes) = (6 on 1st die) × (6 on 2nd) × (1 … 5 on 3rd) × (1 … 5 on 4th)

= × × ×
= 0.0193
Once again this refers to the sixes occurring in a specific order. The two sixes could
occur on any of the dice. In how many ways could this happen? The answer is , the
number of ways to select two objects (showing sixes) from a total of four. Consequent-
ly:
(2 sixes) = 0.0193 ×
= 0.0193 × 6 (from the subcommittee example)
= 0.116
Similarly, the probability of three sixes:
(3 sixes) = No. of ways the 3 sixes could occur × (3 sixes in a specific order)
= × × × ×
!
= × 0.00386
( !× !)
= 0.015
All the calculations, not just those for two and three sixes, can be generalised using the
combination notation:
( sixes) = × (1/6) × (5/6)
This formula has been shown to work for r = 2 and 3. It also works for one six since
= 4. For r = 0 the formula is:
(0 sixes) = × (5/6)
requires knowledge of 0! Mathematicians have defined this as 1 since this is the only
value that makes consistent sense. Given this definition:
!
= × 0!
!
= 1
Similarly, using 0! = 1 makes the formula work for four sixes. The complete answer to
the question is:

No. sixes Probability

0 0.482
1 0.386
2 0.116
3 0.015
4 0.001

These calculations are not of immediate practical value, but they are the basis for
the derivation of the binomial distribution later in the module.

Quantitative Methods Edinburgh Business School 7/13

Module 7 / Distributions

7.4 Standard Distributions

Observed distributions often entail a great deal of data collection. Not only must
sufficient data be collected for the distribution to take shape and probabilities to be
measurable, but also the data must be collected individually for each and every
situation. Standard distributions can reduce the amount of data collection.
A standard distribution is one that has been defined mathematically from a theo-
retical situation. The characteristics of the situation were expressed mathematically
and the resulting distribution (that is, the probabilities of the variable taking differ-
ent values) calculated. In other words, the probabilities were measured not by the
relative frequency method (as in observed distributions) but by the a priori method.
When an actual situation resembling the theoretical one arises, the associated
standard distribution is applied without the need for a lot of data collection.
For example, what distribution results when an unbiased coin is tossed? In this
case the variable takes on just two values, heads or tails. To obtain the standard
distribution of this situation is easy. Both heads and tails are equally likely (the coin
is unbiased). Therefore the distribution is:
P(heads) = 0.5
P(tails) = 0.5
An observed distribution would be obtained by tossing the coin many times,
counting the number of heads and tails and calculating probabilities from these
frequencies. The standard and observed distributions should be approximately
(although they may not be exactly) the same.
One much-used standard distribution, the normal, was derived from the theoret-
ical situation of a variable being generated by a process that should give the variable
a constant value but does not do so because it is subject to many small disturbances.
At a result the variable is distributed around the constant value. The mathematical
expression of such a situation can be used to calculate the probabilities of the
variable taking different values. The normal distribution would then apply, for
instance, to the lengths of machine-cut steel rods. The rods should all be of the
same length but are not because of variations introduced by vibration, machine
setting, the operator, etc. Knowledge of the properties of the normal distribution
allows analyses concerning, for instance, the percentage of rods whose lengths fall
within given tolerance ranges to be found. The normal distribution also applies to
many other situations with similar characteristics. It will be discussed in detail later
in the module. There are other standard distributions that derive from situations
with other characteristics.
In summary, therefore, using an observed distribution implies that data have
been collected, probabilities calculated and histograms formed; using a standard
distribution implies that the situation in which the data are being generated resem-
bles closely a theoretical situation for which a distribution has been constructed
mathematically.

7/14 Edinburgh Business School Quantitative Methods

 Module 7 / Distributions

7.5 Binomial Distribution

One of the earliest standard distributions to be developed was the binomial
distribution.

7.5.1 Characteristics
The binomial distribution is discrete (the values taken by the variable are distinct),
giving rise to stepped shapes. Figure 7.7 illustrates this and also that the shape can
vary from right-skewed through symmetrical to left-skewed depending upon the
situation in which the data are collected.

Figure 7.7 Binomial distribution

7.5.2 Situations in which the Binomial Occurs

Like all standard distributions, the binomial is constructed mathematically from a
theoretical situation. For the binomial the situation is as follows:

The elements of a statistical population are of two types. Each element must be
of one but only one type. The proportion, p, of the population that is of the
first type is known (and the proportion of the second type is therefore 1 − p).
A random sample of size n is taken. Because the sample is random, the number
of elements of the first type it contains is not certain (it could be 0, 1, 2, 3, …
or n) depending upon chance.

From this theoretical situation the probabilities that the sample contains given
numbers (from 0 up to n) or elements of the first type can be calculated. If a great
many samples were actually collected, a histogram would be gradually built up, the
variable being the number of elements of the first type in the sample. Probabilities
measured from the histogram should match those theoretically calculated (approxi-
mately but not necessarily exactly, as with the coin tossing example). The
probabilities calculated theoretically are binomial probabilities; the distribution
formed from these probabilities is a binomial distribution.
For example, a machine produces microchip circuits for use in children’s toys.
The circuits can be tested and found to be defective or OK. The machine has been
designed to produce no more than 20 per cent defective chips. A sample of 30 chips
is collected. Assuming that overall exactly 20 per cent of chips are defective, the

Quantitative Methods Edinburgh Business School 7/15

Module 7 / Distributions

binomial probabilities give the likelihood that the sample contains 0, 1, 2, … or 30

defective chips. The probabilities can be found in a binomial distribution table (such
as Appendix 1, Table A1.1, the use of which will be explained later). If several
samples of size 30 are taken, a distribution of the number of defectives per sample
will emerge. The frequencies of the distribution should match the theoretically
calculated probabilities. If they do not, this may be purely by chance. On the other
hand, it may be an indication that the assumption of 20 per cent defectives is not
correct. Standard distributions are often used in this way. Theoretical probabilities
are compared with observed frequencies and discrepancies used to indicate where
assumptions (in this example the defective rate of the machine) may be false.
To summarise, the binomial relates to situations in which a sample is taken from
a population that consists of two types of element (hence the name ‘binomial’).
Binomial probabilities are the probabilities of obtaining different numbers of each
type in the sample.
Actual situations in which the binomial is used include:
(a) inspection schemes (as above);
(b) opinion polls (the two-way split of the population results from agreement/
disagreement with statements made by the pollster);
(c) selling (the split is the outcome of a contact being a sale/no sale).

7.5.3 Deriving the Binomial Distribution

Suppose a population is split into type 1 elements (proportion p = 1/6) and type 2
elements (proportion 1 − p = 5/6) and that a sample of size 4 is taken. How can the
probabilities that the sample contains 0, 1, 2, 3 or 4 elements of type 1 be calculated?
These probabilities are binomial probabilities. The histogram formed from them is a
binomial distribution.
Fortunately these probabilities have already been calculated. The earlier example
involving dice was exactly the situation described above. The population was split
into two parts: sixes (proportion 1/6) and non-sixes (proportion 5/6). A sample of
4 (4 dice were rolled) was taken. The formula used to calculate the probabilities in
this situation was:
( sixes) = × (1/6) × (5/6)
This formula for the specific case of sample size 4 and proportion of type 1, p =
1/6, can be extended to the general binomial case of sample size n and proportion p
by substituting p for 1/6, 1 − p for 5/6 and n for 4. The general formula by which
binomial probabilities are calculated is:
( of type 1 in sample) = × × (1 − )
The symbols are:
= sample size
= proportion of type 1 in population
1 − = proportion of type 2 in population
= mathematical shorthand for a ‘combination’ = n!/(r ! × (n − r)!)
! (pronounced ‘n factorial’) = n × (n − 1) × (n − 2) × … × 2 × 1

7/16 Edinburgh Business School Quantitative Methods

 Module 7 / Distributions

The following example shows how this formula can be used. If 40 per cent of the
USA electorate are Republican voters, what is the probability that a randomly
assembled group of three will contain two Republicans?
Using the binomial formula:
(2 Republicans in group of 3)
= ( = 2)
= × × (1 − )
= × × (1 − ) since sample size, , equals 3
= × 0.4 × 0.6 since the proportion of type 1 (Republicans) is 0.4
!
= × 0.16 × 0.6
( !× !)

= 3 × 0.16 × 0.6
= 0.288
There is a 28.8 per cent chance that a group of three will contain two Republican
voters.
Making similar calculations for r = 0, 1 and 3, we obtain the binomial distribution
of Figure 7.8.

43.2%

28.8%
21.6%

6.4%

0 1 2 3

Figure 7.8 Binomial distribution for groups of three, p = 40%

7.5.4 Using Binomial Tables

Most analyses involving the binomial distribution are based not on mathematics but
on binomial tables. The probabilities calculated in the Republican voter example can
be obtained from tables. The answers to this example may be checked using
binomial tables for sample size 3 (for this and other sizes from 1 to 8 see Appendix
1, Table A1.1). Look down the left-hand side of the table until the correct sample
size, n = 3, is found; look across the table to find p = 0.4; read down the column;
the 4 probabilities for r = 0, 1, 2, 3 are the same as in Figure 7.8.

Quantitative Methods Edinburgh Business School 7/17

Module 7 / Distributions

Example
A manufacturer has a contract with a supplier that no more than 5 per cent of the
supply of a particular component will be defective. The component is delivered in lorry
loads of several hundred. From each delivery a random sample of 20 is taken and
inspected. If three or more of the 20 are defective, the load is rejected. What is the
probability of a load being rejected even though the supplier is sending an acceptable
proportion of 5 per cent defective?
The binomial distribution is applicable, since the population is split two ways and the
samples are random. Table A1.1 in Appendix 1 is a binomial table for samples of size 20.
As with Table A1.1, the rows refer to the number of defective components in the
sample; the columns refer to the proportion of defectives in the population of all
components supplied (0.05 here). The following is based on the column headed 0.05 in
Table A1.1 in Appendix 1:

P(0 defectives) = 35.8%

P(1 defective) = 37.7%
P(2 defectives) = 18.9%
P(3 defectives) = 6.0%
P(more than 3) = 1.6%
Total = 100.0%

The manufacturer rejects loads when there are three or more defectives in the sample.
From the above, the probability of a load being rejected even though the supplier is
sending an acceptable proportion of defectives is (6.0% + 1.6% =) 7.6 per cent.

7.5.5 Parameters
The distributions in Figure 7.7 differ because the parameters differ. Parameters fix
the context within which a variable varies. The binomial distribution has two
parameters: the sample size, n, and the population proportion of elements of the
first type, p. Two binomial distributions with different parameters, while still having
the same broad characteristics shown in Figure 7.7, will be different; two binomial
distributions with the same sample size and the same proportion, p, will be identical.
The variable is still free to take different values but the parameters place restrictions
on the extent to which different values can be taken.
The binomial probability formula in Section 7.5.3 demonstrates why this is so.
Once n and p are fixed, the binomial probability is fixed for each r value; if n and p
were changed to different values, the situation would still be binomial (the same
general formula is being used), but the probabilities and the histogram would differ.
Right-skewed shapes occur when p is small (close to 0), left-skewed shapes when p
is large (close to 1), and symmetrical shapes when p is near 0.5 or n is large (how
large depends upon the p value). This can be checked by drawing histograms for
different parameter values.

7/18 Edinburgh Business School Quantitative Methods

 Module 7 / Distributions

7.5.6 Deciding whether Data Fit a Binomial

The theoretical situation on which the binomial distribution is defined involves
several assumptions, particularly the assumption that the sample is selected at
random. In an actual situation, if the assumptions are not met, then the theoretical
probabilities will not reflect reality. In practice it is difficult to know the extent to
which the assumptions are met. It is prudent, therefore, to check whether the
theoretical situation matches the actual. This is done by making some observations
and comparing them with what would be expected theoretically if the binomial
applied.
This check is not part of the analysis itself but is to decide whether the situation
is binomial. For example, the purpose of the sampling scheme in the previous
example involving components supplied under contract was to check that one of
the parameters (p) was within acceptable limits. A preliminary check should be made
that the situation is truly binomial and that binomial probabilities can be applied
legitimately. This initial check could be done by taking samples from lorries contain-
ing a known proportion of defective items. If this proportion were 5 per cent, then
35.8 per cent of the samples should have no defectives, 37.7 per cent one defective,
18.9 per cent two defectives and so on in accordance with the extract taken from
Table A1.1 in Appendix 1 in the example above. If this were not approximately the
case then, since the parameters are known with certainty, the conclusion would have
to be drawn that the situation was not binomial. There is a further example of this
procedure in the case studies at the end of the module.

7.6 The Normal Distribution

The most used and best-known standard distribution is probably the normal.

7.6.1 Characteristics
The normal distribution is bell shaped and symmetrical, as in Figure 7.9. It is also
continuous. The distributions met up to now have been discrete. A distribution is
discrete if the variable takes distinct values such as 1, 2, 3, … but not those in
between. The delivery distribution (Table 7.3) is discrete since the data are in
groups; the binomial distribution is discrete since the variable (r) can take only
whole number values. Continuous means that all values are possible for the variable.
It could take the values 41.576, 41.577, 41.578, … rather than all these values being
grouped together in the class 40–49 or the variable being limited to whole numbers
only.

Quantitative Methods Edinburgh Business School 7/19

Module 7 / Distributions

68%

1s 1s
2s 2s
3s 3s

Figure 7.9 Normal distribution

A further implication of continuity is that probabilities are measured not by the
height of the distribution above the x-axis (as in the discrete case), but by areas
under the curve. For example, P(15 ≤ x ≤ 25) is the area under the part of the
curve between x = 15 and x = 25 as a proportion of the whole area of the distribu-
tion. The reason for this is mathematical, but intuitively it can be seen that since
P(x = 18.629732…), say, must be infinitesimally small, a continuous distribution in
which probabilities are measured by the height of the curve above points (as with
the discrete distributions) would be virtually flat. (Refer back to Module 1 for a
fuller discussion of continuity.)
The normal distribution also has important attributes with respect to the standard
deviation (see Figure 7.9):
68% area (68% of readings) within ±1 standard deviation of mean
95% area (95% of readings) within ±2 standard deviations of mean
99.7% area (99.7% of readings) within ±3 standard deviations of mean
For example, it is known that the IQ of children is normally distributed with
mean 100 and standard deviation 17. These attributes imply that:
68% children have an IQ in range 83–117 (mean ±1 standard deviation)
95% children have an IQ in range 66–134
99.7% children have an IQ in range 49–151
Tables exist that give the area under any part of a normal curve, not just within
ranges specified by whole numbers of standard deviations. Table A1.2 from
Appendix 1, which will be explained later, is an example.

7.6.2 Situations in Which the Normal Occurs

The normal distribution is constructed mathematically from the following theoreti-
cal situation:

Repeated observations or measurements are taken of the same constant quan-

tity. Each time an observation is taken, the quantity is subject to many sources

7/20 Edinburgh Business School Quantitative Methods

 Module 7 / Distributions

of disturbance. Each source gives rise to