Contents

1 Introduction 1
1.1. Overview of Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2. Definition of terms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.3. Sampling Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.4. Probability Sampling methods . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4.1. Simple Random Sampling . . . . . . . . . . . . . . . . . . . . . . . 5
1.4.2. Systematic Random Sampling . . . . . . . . . . . . . . . . . . . . . 6
1.4.3. Stratified Sampling . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.4.4. Cluster Sampling . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.5. Non-probability sampling methods . . . . . . . . . . . . . . . . . . . . . . 8
1.5.1. Convinience or Availability . . . . . . . . . . . . . . . . . . . . . . 8
1.5.2. Quota / Proportionate . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.5.3. Expert or Judgemental . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.5.4. Chain referral / Snowballing / Networking . . . . . . . . . . . . . 9
1.6. Errors in sampling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.7. Data Collection Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.7.1. Observation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.7.2. Interview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.7.3. Experimentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

2 Data and Data Presentation 1

2.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
2.2. Data Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
2.2.1. Qualitative random variables . . . . . . . . . . . . . . . . . . . . . 1
2.2.2. Quantitative random variables . . . . . . . . . . . . . . . . . . . . 2
2.3. Data sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.4. Data presentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.4.1. Pie Charts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.4.2. Bar Chart . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.4.3. Histograms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.4.4. Stem and leaf diagram . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.4.5. Frequency Polygons . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.5. Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

3 Measures of Central Tendency 13

3.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.2. Measures of Central Tendency . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.3. Arithmetic Mean . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.3.1. Mean for ungrouped data . . . . . . . . . . . . . . . . . . . . . . . 14

1
2 CONTENTS

3.3.2. Mean for grouped data . . . . . . . . . . . . . . . . . . . . . . . . . 14

3.4. The Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.4.1. Mode for ungrouped data . . . . . . . . . . . . . . . . . . . . . . . . 15
3.4.2. Mode for grouped data . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.5. The Median . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.5.1. Median for ungrouped data . . . . . . . . . . . . . . . . . . . . . . 17
3.5.2. Median for grouped data . . . . . . . . . . . . . . . . . . . . . . . . 17
3.6. Quartiles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.6.1. Quartiles for ungrouped data . . . . . . . . . . . . . . . . . . . . . 19
3.6.2. Quartiles for grouped data . . . . . . . . . . . . . . . . . . . . . . . 19
3.6.3. The second quartile, Q2 (Median) . . . . . . . . . . . . . . . . . . . 20
3.6.4. The upper quartile, Q3 . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.6.5. Percentiles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.7. Skewness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.8. Kurtosis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.9. Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

4 Measures of Dispersion 23
4.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
4.2. Range . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
4.3. Variance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4.4. Standard deviation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
4.5. Coefficient of variation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
4.6. Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

5 Basic Probability 31
5.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
5.2. Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
5.3. Approches to probability theory . . . . . . . . . . . . . . . . . . . . . . . . 31
5.4. Properties of probability . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
5.5. Basic probability concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
5.6. Types of events . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
5.7. Laws of probability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
5.8. Types of probabilities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
5.9. Contigency Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
5.10.Tree diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
5.11.Counting rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
5.11.1. Multiplication Rule . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
5.11.2. Permutations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
5.11.3. Combinations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
5.12.Exercise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

6 Probability Distributions 45
6.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
6.2. Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
6.3. Random variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
6.4. Discrete probability distribution . . . . . . . . . . . . . . . . . . . . . . . 46
6.5. Properties of discrete probability mass function . . . . . . . . . . . . . . 47
6.6. Probability terminology and notation . . . . . . . . . . . . . . . . . . . . . 48
CONTENTS 3

6.7. Discrete probability distributions . . . . . . . . . . . . . . . . . . . . . . . 49

6.7.1. Bernoulli distribution . . . . . . . . . . . . . . . . . . . . . . . . . . 49
6.7.2. Binomial distribution . . . . . . . . . . . . . . . . . . . . . . . . . . 50
6.7.3. Poisson distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
6.8. Continuous probability distributions . . . . . . . . . . . . . . . . . . . . . 53
6.8.1. The Normal distribution . . . . . . . . . . . . . . . . . . . . . . . . 53
6.8.2. The standard normal distribution . . . . . . . . . . . . . . . . . . 54
6.8.3. The Uniform distribution . . . . . . . . . . . . . . . . . . . . . . . 56
6.8.4. The Exponential distribution . . . . . . . . . . . . . . . . . . . . . 57

7 Interval Estimation 59
7.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
7.2. Confidence Intervals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
7.3. Confidence Interval for the Population Mean . . . . . . . . . . . . . . . . 60
7.4. One-Sided Confidence Intervals for the Population Mean . . . . . . . . . 63
7.5. Confidence Interval for the Population Proportion . . . . . . . . . . . . . 67
7.6. Confidence Interval for the Population Variance . . . . . . . . . . . . . . 68
7.7. Confidence Interval for the Population Standard Deviation . . . . . . . . 69
7.8. Confidence Interval for the Difference of Two Populations Means . . . . 70
7.8.1. Case 1: Known Population Variance . . . . . . . . . . . . . . . . . 70
7.8.2. Case 2: Unknown (but assumed Equal) Population Variances . . 70

8 Hypothesis Testing 73
8.1. Important Definitions, and Critical Clarifications . . . . . . . . . . . . . 73
8.2. General Procedure on Hypotheses Testing . . . . . . . . . . . . . . . . . . 75
8.3. Hypothesis Testing Concerning the Population Mean . . . . . . . . . . . 75
8.3.1. Case 1: Known Population Variance . . . . . . . . . . . . . . . . . 75
8.3.2. Guidelines to the Expected Solution . . . . . . . . . . . . . . . . . 76
8.3.3. Case 2: Unknown Population Variance . . . . . . . . . . . . . . . . 76
8.4. Hypothesis Testing concerning the Population Proportion . . . . . . . . . 78
8.5. Comparison of Two Populations . . . . . . . . . . . . . . . . . . . . . . . . 79
8.5.1. Hypothesis Testing concerning the Difference of Two Population
Means . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
8.6. Independent Samples and Dependent/ Paired Samples . . . . . . . . . . 80
8.6.1. Advantages of Paired Comparisons . . . . . . . . . . . . . . . . . . 81
8.6.2. Disadvantages of Paired Comparisons . . . . . . . . . . . . . . . . 82
8.7. Test Procedure concerning the Difference of two Population Proportions 82
8.8. Tests for Independence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
8.9. Ending Remark(s) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85

9 Regression Analysis 87
9.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
9.2. Uses of Regression Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 87
9.3. Abuses of Regression Analysis . . . . . . . . . . . . . . . . . . . . . . . . . 88
9.4. The Simple Linear Regression Model . . . . . . . . . . . . . . . . . . . . . 89
9.4.1. The Scatter Plot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
9.4.2. The Correlation Coefficient . . . . . . . . . . . . . . . . . . . . . . 89
9.4.3. Regression Equation . . . . . . . . . . . . . . . . . . . . . . . . . . 90
9.4.4. Coefficient of Determination, r2 . . . . . . . . . . . . . . . . . . . . 91
4 CONTENTS

10 Index numbers 95
10.1.Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
10.2.Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
10.3.What is an Index Number? . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
10.3.1. Characteristics of an Index Numbers . . . . . . . . . . . . . . . . . 95
10.3.2. Uses of Index Numbers . . . . . . . . . . . . . . . . . . . . . . . . . 96
10.4.Types of Index Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
10.5.Methods of constructing index numbers . . . . . . . . . . . . . . . . . . . 98
10.5.1. Aggregate Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
10.5.2. Merits and demerits of this method . . . . . . . . . . . . . . . . . . 99
10.5.3. Weighted Aggregates Index . . . . . . . . . . . . . . . . . . . . . . 100
10.5.4. Laspeyres Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
10.5.5. Merits and demerits of Laspeyres method? . . . . . . . . . . . . . 101
10.5.6. Paasches Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
10.5.7. Merits and Demerits of Paasches Index . . . . . . . . . . . . . . . 102
10.6.Fisher Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
Chapter 1

Introduction

1.1. Overview of Statistics

Statistics is when individual data values are collected, summarized,analysed and pre-
sented and used for decision making. It is an important tool in transforming raw data
into meaning and usable information. Also statistics can be regarded as a decision
support tool. A table below shows a transformation process of data to information.

Input Process Output

Data Statistical Analysis Information
Raw observation Transformation process Useful, Usable and Meaningful

An understanding of statistics allows managers to: i) Perform simple statistical anal-

ysis. ii) Intelligently prepare and interpret reports expressed in numerical terms. iii)
Communicate effectively with statistical analysts. iv) Good decision making.

1.2. Definition of terms

The following terms shall be used in this module more often.

Statistics
Definition 1
Statistics refers to the methodology [collection techniques] for collection, presentation
and analysis of data and the use of such data [Neter J. et al (1988)].

Definition 2
In common usage, it refers to numerical data. This means an y collection of data or
information constitutes what is referred to as Statistics. Some examples under this
definition are:
2 Introduction

1. Vital statistics - These are numerical data on births, marriages, divorces, com-
municable diseases, harvests, accidents etc.

2. Business and economic statistics - These are numerical data on employment,

production, prices, sales, dismissals etc.

3. Social statistics - These are numeric data on housing, crime, education etc.

Definition 3 - Statistics is making sense of data.

In Statistics (as in real life), we usually deal with large volumes of data making it diffi-
cult to study each observation (each data point), in order to draw conclusions about the
source of the data. We seek a statistical method or methods that can summarise the
data so that we can draw conclusions about these data, without scrutinising each ob-
servations (which is rather difficult). Such methods fall under area of statistics called
descriptive statistics.

A Statistician is an individual who collects dat, analyses it using statistical tech-

niques, interprets the results and makes conclusions and recommendations on the
basis of the data analysis.

Parameter(s) - These are numeric measure(s) derived from a population e.g. popula-
tion mean (µ), population variances (σ 2 ), and population standard deviation (σ).
Data
Data is what is more readily available from a variety of sources and of varying quality
and quantity. Precisely data is individual observations on an issue and in itself con-
veys no useful information.

Information
To make sound decision, one needs good and quality information. Information must
be timely, accurate, relevant, adequate and readily available. Information is defined to
processed data. Table above summarizes relationship between data and information.
GIGO - Garbage In Garbage Out.

Random variable
A variable is any characteristic being measured or observed. Since a variable can take
on different values at each measurement it is termed a random variable. For example,
Sales, Company turnover, Weight, Height, yield, Number of babies born, e.t.c
Introduction 3

Population
A population is a collection of elements about which we wish to make an inference.
The population must be clearly defined before the sample is taken.

Target population
The population whose properties are estimated via a sample or usually the ’total’ pop-
ulation.

Sample
A sample is a collection of sampling units drawn from a population. Data are obtained
from the sample and are used to describe characteristics of the population. A sample
can also be defined as a subset / part of or a fraction of a population.

statistic(s)
These are numeric measure(s) derived from a sample e.g. sample mean (x̄), sample
variances (s2 ), and sample standard deviation (s).

Sampling Frame
A sampling frame is a list of sampling units. A set of information used to identify a
sample population for statistical treatment. It includes a numerical identifier for each
individual, plus other identifying information about characteristics of the individuals,
to aid in analysis and allow for division into further frames for more in-depth analysis.

Sampling.

A process used in statistical analysis in which a predetermined number of observa-

tions will be taken from a larger population. The methodology used to sample from a
larger population depend on the type of analysis being performed, but includes simple
random sampling, systematic sampling and observational sampling. These will be dis-
cussed later.

Sampling Units
Sampling units are non-overlapping collections of elements from the population that
cover the entire population. It is a member of both the sampling frame and of the sam-
ple. The sampling units partition the population of interest for example households or
individual persons for census.
4 Introduction

1.3. Sampling Techniques

We do explore the sampling techniques in order to be able to decide which one is the
most appropriate for each given situation. Sampling techniques are methods of how
data can be collected from the given population.

Types of Sampling

Probability Sampling
Has a distinguishing characteristic that each unit in the population has a known,
nonzero probability of being included in the sample thus, it is clear that every subject
or unit has an equal chance of being selected from the population. These probabilities
are usually equal. It eliminates the danger of being biased in the selection process due
to one’s own opinions or desires.

Non-probability Sampling
Is a process where probabilities cannot be assigned to the units objectively, and hence
it becomes difficult to determine the reliability of the sample results in terms of proba-
bility. A sample is selected according to one’s convenience, or generality in nature. It is
a good technique for pilot or feasibility studies. Examples include purposive sampling,
convenience sampling, and quota sampling. In non-probability sampling, the units
that make up the sample are collected with no specific probability structure in mind
e.g. units making up the sample through volunteering.

Remark:
We shall focus on probability sampling because if an appropriate technique is chosen,
then it assures sample representativeness and hence the errors for the sampling can
be estimated.

Reasons to use Sampling

Sampling is done mostly for reasons of Cost, Time, Accessibility, Utility and Speed.
Expansion on the reasons is left for the lecture. Some points to clearly define when
sampling. Sampling method to be employed.

Sample size
Reliability degree of the conclusions that we can obtain i.e. an estimation of the error
that we are going to have. An inappropriate selection of the elements of the sample
Introduction 5

can cause further errors once we want to estimate the corresponding population pa-
rameters.

1.4. Probability Sampling methods

The four methods of probability sampling are simple random, systematic, stratified
and cluster sampling methods.

1.4.1. Simple Random Sampling

Requires that each element of the population have an equal chance of being selected.
A simple random sample is selected by assigning a number to each element in the
population list and then using a random number table to draw out the elements of the
sample. The element with the number drawn out makes it into the sample. The pop-
ulation is ”mixed up” before a previously specified number, n, of elements is selected
at random. Each member of the population is selected one at a time, independent of
one another. However, it is noted that all elements of the study population are either
physically present or listed.
Also, regardless of the process used for this method, the process can be laborious espe-
cially when the list of the population is long or it is completed manually without the
aid of a computer. A simple random sample can be got using calculator (random key),
computer (using excel function =rand() ), or random number tables.

In this method, every set of n elements in the population has an equal chance of being
selected as the sample.

Advantages

• It eliminates bias due to the personal judgment or discretion of the researcher

• More representative of the population

• Estimates are more accurate

Demerits

• Requires an up to date sampling frame

• Numbering of the population elements may be time consuming eg for large pop-
ulations
6 Introduction

Illustration
An example of simple random sampling may include writing each member of the pop-
ulation on a piece of paper and putting in a hat. Selecting the sample from the hat
is random and each member of the population has an equal chance of being selected.
However, this approach is not feasible for large populations, but can be completed eas-
ily if the population is very small.

1.4.2. Systematic Random Sampling

Selection of sampling units is done in sequences separated on lists by the interval

selection. In this method, every nth element from the list is selected as the sample,
starting with a sample element n randomly selected from the first k elements. For
example, if the population has 1000 elements and a sample size of 100 is needed, then
k would be 1000
100 = 10. Now, if the number 7 is randomly selected from the first ten
elements on the list, then the sample would continue down the list selecting the 7th
element from subsequent groups of ten elements. Care must be taken when using sys-
tematic sampling to ensure that the original population list has not been ordered in a
way that introduces any non-random factors into the sampling.

Illustration
An example of systematic sampling would be if an official from the Academic Registry
of a hypothetical university is to register students for a tour of regional universities.
The official may select at random the 15th student out of the first 20 students in a list
of all students in the university. This official would then keep adding twenty and se-
lecting the 35th student, 55th student, 75th student and so on to register for the tour of
regional universities until the end of the list is reached.

Remark:
In cases where the population is large and the population list is available, systematic
sampling is usually preferred over simple random sampling since it is more convenient
to the experimenter.

1.4.3. Stratified Sampling

It is used when representatives from each homogeneous subgroup within the popu-
lation need to be represented in the sample. The first step in stratified sampling is
to divide the population into subgroups (strata) based on mutually exclusive criteria.
Random or systematic samples are then taken from each subgroup. The sampling
fraction for each subgroup may be taken in the same proportion as the subgroup has
in the population.
Introduction 7

Illustration
As an example, if an owner of a local supermarket conducting a customer satisfaction
survey may wish to select random customers from each customer type in proportion
to the number of customers of that type in the population. Suppose 40 sample units
are to be selected, and 10% of the customers are managers, 60% are users, 25% are
operators and 5% are students from CUT, then 4 managers, 24 users, 10 operators and
2 students from CUT would be randomly selected.

Remark:
Stratified sampling can also sample an equal number of items from each subgroup.

1.4.4. Cluster Sampling

In cluster sampling, the population that is being sampled is divided into naturally oc-
curring groups called clusters. A cluster is as heterogeneous as possible to matching
the population clusters which says that a cluster is representative of the population.
A random sample is then taken from within one or more selected.

Illustration
An organization with 300 small branches providing a service country wide has an em-
ployee at the HQ who is interested in auditing for compliance to some coding standard.
The employee might use cluster sampling to randomly select 40 branches as represen-
tatives for the audit and then randomly sample coding systems for auditing from just
the 40.

Remark:
Cluster sampling can tell us a lot about that particular cluster, but unless the clusters
are selected randomly and a lot of clusters are sampled, generalizations cannot always
be made about the entire population.

Difference between a Cluster & a Stratum

A cluster is a heterogeneous subgroups but a stratum is a homogeneous subgroup. A
summary of probability sampling methods is discussed below.

Simple random
8 Introduction

Each member of the study population has an equal probability of being selected.

Systematic
Each member of the study population is either assembled or listed, a random start is
designated, then members of the population are selected at equal intervals

Stratified
Each member of the study population is assigned to a homogeneous subgroup or stra-
tum, and then a random sample is selected from each stratum.

Cluster
Each member of the study population is assigned to a heterogeneous subgroup or clus-
ter, then clusters are selected at random and all members of a selected cluster are
included in the sample.

1.5. Non-probability sampling methods

There are four methods of non-probability sampling discussed in this module that are
Convienience, Quota, expert and chain referral.

1.5.1. Convinience or Availability

This is sampling which is based on the proximity of the population elements to the de-
cision maker. Being at the right place at the right time. Elements nearby are selected,
and those not in close physical or communication range are not considered.

1.5.2. Quota / Proportionate

This is sampling in which certain distinct or known characteristics in the population

should appear in relatively similar proportions. Eg A population (N ) of 100 comprised
of 60 being females and 40 being males. If a sample of n=20 is to be selected then that
ratio of 6:4 has to be reflected.

1.5.3. Expert or Judgemental

This is sampling in which the decision maker has direct or indirect control over which
elements are to be included in the sample. Appropriate when the decision maker feels
Introduction 9

that some population members have better or more information than others. Or some
members are more representative than others.

1.5.4. Chain referral / Snowballing / Networking

The researcher starts with a person who displays qualities of interest then refers to
the next and so on.

1.6. Errors in sampling

During sampling, errors can be committed by the statistician. These are either sam-
pling or non-sampling errors. Errors can be corrected by sampling without bias.
Some common sources of bias are i) incorrect sampling operation, non-interviews.
Some errors that arises in sampling are discussed below.

Selection error
Selection error occurs when some elements of the population have a higher probability
of being selected than others. Consider a scenario where a manager of a local super-
market wishes to measure how satisfied his customers. He proceeds to interview some
of them from 08:00 to 12:00. Clearly, the customers who do their shopping in the after-
noon are left out and will not be represented making the sample unrepresentative of
all the customers. Such kind of errors can be avoided by choosing the sample so that
all the customers have the same probability of being selected. This is a sampling error.

Non-Response Error
It is possible that some of the elements of the population do not want or cannot answer
certain questions. It may also happen when we have a questionnaire including per-
sonal questions, that some of the members of the population do not answer honestly or
would rather avoid answering. These errors are generally very complicated to avoid,
but in case that we want to check honesty in answers, we can include some questions
called filter questions to detect if the answers are honest. This is a non-sampling error.

Interviewer influence
The interviewer may fail to be impartial i.e. s/he can promote some answers more than
others.

Remark:
A sample that is not representative of the population is called a biased sample.
10 Introduction

Questions relating to selecting out of naturally arise. These are: When concluding
about the population, how many of the population elements is represented by each one
of the sample elements? What proportion of the population are we selecting? The re-
sponses lie in the following factors.

1.7. Data Collection Methods

The three data collection methods are: Observation,Interview and Experimenta-
tion. Depending on the type of research and data to be collected, different methods
can be used to collect that data set.

1.7.1. Observation

This method has the direct and desk research methods. Direct observation involves
collecting data by observing the item in action. Examples for this method are: pedes-
trian flow, vehicle traffic, purchase behavior of a commodity in a shop, quality control
inspection e.t.c. An advantage of this method is that the respondent behaves in a
natural way since he is not aware that he is being observed. A disadvantage is that
it is a passive form of data collection. Also there is no opportunity to investigate the
behavior further. Desk research involves consulting and extracting secondary data
from source documents and collect data from them.

1.7.2. Interview

This method collects primary data through direct questioning. A questionnaire is the
instrument used to structure the data collection process. Three approaches in data
collection using interviews are: personal, postal and telephone interviews.

Personal Interviews
A questionnaire is completed through face-to-face contact with the respondent. Ad-
vantages for this method are: High response rate, it allows probing for reasons, data
collection is immediate, data accuracy is assured, useful for technical data is required,
non-verbal responses can be observed and noted, more questions can be asked, re-
sponses are spontaneous, and use of aided-recall questions is possible. Disadvantage
of this method are that it is time consuming, it requires trained interviewers, fewer
interviews are conducted because of cost and time constraints, biased data can be col-
lected if interviewer is inexperienced.
Introduction 11

Postal Surveys
When target population is large and or geographically dispersed then use of postal
questionnaires is considered most suitable. Advantages of this method is that larger
sample of respondents can be reached, more costs effective, interviewer bias is elimi-
nated, respondents have more time to consider their responses, anonymity of respon-
dents is assured resulting in more honest responses, respondents are more willing to
answer personal questions. The disadvantages for this method are: low response
rate, respondents cannot get clarity on some questions, mailed questionnaires must be
short and simple to complete, limited possibilities of probing or further investigations,
data collection takes long time, no control of who answers the questionnaire, and no
possibilities of validating responses.

Telephone Interviews
The interview is conducted telephonically with the respondent. Advantages of this
method are: it allows quicker contact with geographically dispersed respondents, call-
backs can be made if respondent is not initially available, low cost, interviewer probing
is possible, clarity on questions can be provided by the interviewer, and a larger sample
of respondents can be reached in short space of time. Disadvantages are that respon-
dent anonymity is lost, non-verbal responses cannot be observed, trained interviewers
are required hence more costly, possible interviewer bias, respondent may terminate
interview prematurely, and sampling errors are compounded if more respondents do
not have telephones.

1.7.3. Experimentation
This is when primary data is generated through manipulation of variables under con-
trolled conditions. The method is mostly used in scientific and engineering research.
Data on the primary variable under study is monitored and recorded whilst the re-
searcher controls effects of a number of influencing factors. Examples include: De-
mand elasticity for a product, advertising effectiveness. Advantages of this method
are: quality data is collected and results are generally more objective and valid. The
disadvantages are that the method is costly and time consuming, and may be impos-
sible to control for certain factors which affects the results.
12 Introduction
Chapter 2

Data and Data Presentation

2.1. Introduction

A Statistician collects data (in an appropriate manner) analyses it using statistical

techniques, interprets the results and makes conclusions and recommendations on the
basis of data analysis. The word data keeps turning in our discussion. Data is the
”blood of statistics”.

The world of statistics resolves around data, there is no statistics without data. What
is data? How is it collected? Why do we collect it? These are the questions to be
answered in this chapter.

2.2. Data Types

An understanding of nature of data is necessary for 2 reasons. It enables a user: assess

data quality and to select the appropriate statistical method to use to analyse the data.

Quality of data is influenced by three factors that are: type, source and method
used to collect data. The type of data gathered determines the type of analysis which
can be performed on the data. Certain statistical methods are valid for certain data
types only. An incorrect application of a statistical method to a particular data type
can render the findings invalid.

Data type is determined by the nature of the random variables which the data repre-
sents. Random variables are essentially of two kinds that are Qualitative and Quanti-
tative.

2.2.1. Qualitative random variables

These are variables which yield categorical (non-numeric) responses. The data gener-
ated by qualitative random variables are classified into one of a number of categories.
2 Introduction

The numbers representing the categories are arbitrary i.e. codes: Coded values cannot
be manipulated arithmetically, as it does not make sense.

Examples of qualitative random variables

Random variables Response Categories Data Codes

Supervisor 1
Managerial Level Section Head 2
Departmental Head 3
General Manager 4
Do you like soft drink? Yes 2
No 1
Gender Female 0
Male 1

2.2.2. Quantitative random variables

Quantitative random variables are variables that yield numeric responses. The data
generated for quantitative random variables can be meaningfully manipulated using
conventional arithmetic operations.

Examples of quantitative random variables

Random Variables Response Range Data

Age of employee 17 - 65 years e.g. 39 years
Distance to work 0 - 20 km e.g. 5.3 km
Class size 1, 2, 3 ... e.g. 15 pupils

Each random variable category is associated with a different type of data. There
are two classifications of data types.

Data type 1 - Data measurement scales

Data measurement scales include Nominal, Ordinal, Interval and Ratio-scaled data.

Nominal-scaled data
Objects or events are distinguished on the basis of a name. Nominal-scaled data is
associated mainly with qualitative random variables. Where data of qualitative ran-
dom variables is assigned to one of a number of categories of equal importance, then
such data is referred to as nominal-scaled data. There is no implied ordering between
the groups of the random variable.
Data and Data Presentation 3

Examples of nominal-scaled data

Table below shows examples of nominal scaled data.

Qualitative Random Variables Response Categories Data Code

Gender Male / Female 1/2
Car type owned Mazda/Golf/Toyota/Honda 1/2/3/4
City leaved in Harare/Byo/Mutare/Gweru 1/2/3/4
Marital Status Married/Single/Divorced/Widow 1/2/3/4
Engineering Profession Civil/Electrical/Mechanical 1/2/3

Each observation of the random variables is assigned to only one of the categories
provided. Arithmetic calculations cannot be meaningfully performed on the coded val-
ues assigned to each category. They are only numeric codes which are arbitrarily as-
signed and can be counted. Nominal-scaled data is the weakest form of data, since
only a limited range of statistical analysis can be formed on such data.

Ordinal-scaled data
Objects or events are distinguished on the basis of the relative amounts of some char-
acteristics they posses. The magnitude between measurements is not reflected in the
rank. Such data is associated mainly with qualitative random variables. Like nominal-
scaled data, ordinal-scaled data is also assigned to only one of a number of coded cat-
egories, but there is now a ranking implied between the categories in terms of being
better, bigger, longer, older, taller, or stronger, etc. While there is an implied differ-
ence between the categories, this difference cannot be measured exactly. That is, the
distance between categories cannot be quantified nor assumed to be equal. Ordinal-
scaled data is generated from ranked responses in market research studies.

Examples of Ordinal-scaled data

Qualitative Random Variables Response Categories Data Codes

T-Shirt size Small / Medium / Large 1/2/3
Company turnover Small / Medium / Large 1/2/3
Management levels Lower / Middle / Senior 1/2/3
Work experience Little / Moderate / Extensive 1/2/3
Magazine type Rank the top three magazine 1/2/3
you often read
Sizes of bulbs Smallest / Small / Large / Largest 1/2/3/4

There is a wider range of valid statistical methods (i.e. the area of non-parametric
statistics) available for the analysis of ordinal-scaled data than there is for nominal-
scaled data. Ordinal-scaled data is also generated from a ”counting process”.
4 Introduction

Interval-scaled data
Interval-scaled data is associated with quantitative random variables. Differences
can be measured between values of a quantitative random variable. Thus interval-
scaled data possesses both order and distance properties. Interval-scaled data, how-
ever, does not possess an absolute origin. Therefore the ratio of values cannot be mean-
ingfully compared for interval-scaled data. The absolute difference makes sense when
interval-scaled data has been collected.

Examples of Interval-scaled data

Suppose four places A, B, C and D have temperatures 20o C, 25o C, 35o C and 40o C re-
spectively. Using interval scale we see that the difference between A and B is equal to
that of C and D. However ratios are not used. A value of 0o C does not mean absence of
temperature, also it is not correct to say temperature of D is twice as much as that of
A.

Interval-scaled data is most often generated in marketing studies through rating

responses on a continuum scale. A wide range of statistical techniques can be applied
to interval scaled data as it posses numeric (measurement) properties.

Ratio-scaled data
This data is associated mainly with quantitative random variables. If the full range of
arithmetic operations can be meaningfully performed on the observations of a random
variable, the data associated with that random variable is termed ratio-scaled. It is a
numeric data with a zero origin. The zero origin indicates the absence of the attribute
being measured.

Example 1 of Ratio-scaled data

Quantitative Random Variable Example of data values

Age 42 years
Income $2,500
Distance 35 km
Time 32 minutes
Mass 240g
Price $7.82

Such data are the strongest form of statistical data which can be gathered and
lends itself to the widest range of statistical methods. Ratio-scaled data can be ma-
nipulated meaningfully through normal arithmetic operations. Ratio-scaled data is
Data and Data Presentation 5

gathered through a measurement process. It should be noted that if ratio-scaled data

is grouped into categories, the data type becomes ordinal-scaled. This then reduces the
scope for statistical analysis on the random variable.

Example 2 of Ratio-scaled data

Note: By capturing Age data in categories instead of actual age, the data becomes
ordinal-scaled. However, the random variable remains quantitative in nature. See
table below.

Random Variable Response Category Data code used

Age 0 - 16 1
17 - 24 2
25 - 36 3
37 - 45 4
46 - 55 5

When data capturing instruments are set up, care must be exercised to ensure that
the most useful form of data is captured. However, this is not always possible for
reasons of convenience, cost and sensitivity of information. This applies particularly
to random variables such as age, personal income, company turnover and consumer
behavior questions of a personal nature. The functional area of marketing generates
mostly categorical (i.e. nominal/ordinal) data arising from consumer studies, while the
areas of finance/accounting and production generate mainly quantitative (ratio) data.
Human resources management generates a mix of qualitative and quantitative data
for analysis.

Data type 2

A second classification of data type is either discrete and continuous data.

Discrete data
A random variable whose observations can take on only specific values, usually only
integer values, is referred to as a discrete random variable. In such instances, certain
values are valid, while others are invalid.

Examples of random variables generating discrete data

(i) Number of cars in a parking lot at a given time, (ii) Daily number of hotel rooms
booked for January 1992, (iii) Number of students in a class, (iv) Number of employees
in an organization, (v) Number of paintings in an art collection, (vi) Number of cars
sold in a month by a dealer, (vii) Number of life assurance policies issued in 1990 in
Zimbabwe.
6 Introduction

Continuous data
A random variable whose observations take on any value in an interval is said to gen-
erate continuous data. This means that any value between a lower and an upper limit
is valid.

Examples of random variables generating continuous data

(i) Time taken to travel to work daily, (ii) Age of a bottle of red wine, (iii) Mass of a
caravan, (iv) Tensile strength of material, (v) Speed of an aircraft, (vi) Length of a
ladder.

2.3. Data sources

Data for statistical analysis are available from any different sources. There are two
classification types of data sources that are: Internal/external and Primary/secondary
sources.

Internal data sources

This refers to the availability of data from within an organisation; internal data are
generated during the course of normal business activities. Examples include: i) Finan-
cial data sales vouchers, credit notes, accounts receivable, accounts payable, asset reg-
ister. ii) Production data - production cost records, stock sheets. iii) Human Resource
data - time sheets, wages and salaries schedule, employee personal employment files.
iv) Marketing data sales data, advertising expenditure.

External data sources

Data available from outside an organization is referred to as external data sources.
Such sources may be private institutions, trade/employer/employee associations, profit
motivated organizations or government bodies. The cost of the external data is de-
pendent on the source. Generally, the cost is greater from private bodies than it is
from government or public sources. Examples include: i) Private source include - Com-
mercial and Industrial Association of Business, Research Bureaux. ii) Public domain
sources include Newspapers, journals, trade magazines, reference material in libraries,
The Central Statistical Services (ZimStats) is the Governments data capturing and
dissemination instrument and others such as Universities, Reference Libraries, Banks’
economic reports.

Primary data sources

Data and Data Presentation 7

Data which is captured at the point where it is generated is called Primary data. Such
data is captured for the first time and with specific purpose in mind. Examples of data
sources are: Largely the same as for internal data source, but also includes survey
data (personnel surveys, salary surveys, market research surveys).

Advantages of primary data

Primary data are directly relevant to the problem at hand and generally offer greater
control over data accuracy.

Disadvantages of primary data

Primary data can be time consuming to collect and are generally more expensive to
collect (e.g. Market Research)

Secondary data sources

Data collected and processed by others for a purpose other than the problem at hand
are called secondary data. Such data are already in existence either within or outside
an organisation, i.e. one can get both internal secondary and external secondary data.
The problem at hand determines whether the data are primary or secondary. Exam-
ples of internal secondary data are: Aged market research figures, previous financial
statements of your company and past sales reports. Examples of external secondary
data Reports produced by external data sources.

Advantages of secondary data

Some of the advantages of use of secondary data are: The data are already in exis-
tence, Access time is relatively short, The data are generally less expensive to acquire.

Disadvantages of secondary data

Some disadvantages of secondary data are:

• Data may not be problem specific.

• Data may be outdated and hence inappropriate.

• It may be difficult to assess data accuracy.

• Data may not be subject to further manipulation.

• Combining various sources could lead to errors of collation and introduce bias.
8 Introduction

2.4. Data presentation

Data can be presented in tables or graphs. Graphical techniques are pictorial or graph-
ical representations of data such that the main features of the data are captured. The
various graphical techniques which we will cover in this unit. Pie charts, bar charts,
histograms, box and whisker plots and stem and leaf displays. Some other techniques
which are important are dotplots, Lorenz curve and Z curves are not discussed in this
module.

2.4.1. Pie Charts

A pie chart as the name suggests, is a circle divided into segments like a pie cut into
pieces from the centre outwards. Each segment represents one or more values taken
by a variable. Such charts are used to display qualitative data. Let us now look at an
example, and see how we can construct and interpret a pie chart.

Example 1.1
The ages of 10 students doing BSCAC program at Chinhoyi University of Technology
are: 26, 28, 28, 16, 22, 35, 42, 19, 55, 28. Grouping the ages into classes of 25 and
below, 26-35, 36-45, and above 45, leads to a frequency distribution table below.

Age group Number of Students

Below 25 3
26 - 35 5
36 - 45 1
Above 45 1

We now express these age groups as proportions or percentages and then indicate
the angle in degrees as in table below.

Age group Number of Students Proportions Percentages Angle

3 3 3
Below 25 3 0+3+5+1+1 = 10 10× 100 = 30% 108o
5 5
26 - 35 5 10 10× 100 = 50% 180o
1 1
36 - 45 1 10 10× 100 = 10% 36o
1 1
Above 45 1 10 10× 100 = 10% 36o

There are only 4 groups. What we wish to do is to represent these percentages as

angles in degrees i.e. instead of everything in Table 1.2 (column 2) above adding up
to 10 (or 100 in the case of percentages) we want them to add up to 360o (the total
number of degrees in a circle) as shown in column 5. The calculation of the angle of
Data and Data Presentation 9

the ith category can be done directly from the observations by using.

X
Pn i = 360o
i=1 Xi

i.e. each observation multiplied by 3600 divided by the sum of the observations

2.4.2. Bar Chart

A bar chart, as the name suggests, is a visual presentation of data by means of bars
or blocks put side by side. Each bar represents a count of the different categories of
the data. Although both pie chart and bar graphs (as they are sometimes called) are
used to illustrate qualitative data or discrete qualitative data, bar charts use the ac-
tual counts or frequencies of occurrences of each category of data. We need not use the
actual data; we can use the percentage to come up with the Bar graph. Let us use the
data in example 1.1 to illustrate the bar chart.

Example 1.2
We will now construct the bar chart using the data in example 1.1. We come up with
suitable scales for the height and width of the graph, which are such that the graph
is clear and representative in class example. The bars represent each age group count
in terms of height. You can choose to make the bars thin or wide, it’s up to you all
you need to be certain of is that the bars represent each age group in terms of height.
The bars should be of the sane width. Often, we represent each category by different
colours or shades. This is especially useful when we are comparing several groups. For
instance, we could be comparing the age groups of different intakes that would mean
several graphs all put side by side. In this way we can compare the intakes aged X
over different years.

2.4.3. Histograms

A histogram is a gragh drawn from a frequency distribution. It is used to represent

continuous quantitative data. It usually consists of adjacent, touching rectangles or
bars. The area of each rectangle is drawn in proportion to the frequency corresponding
to that frequency class. When the class intervals are equal, the area of each rectangle
is a constant multiple of height and so the histogram can be drawn as for a bar chart,
except that the rectangles are touching. If the class intervals are not equal, the fre-
quencies are adjusted accordingly to come up with frequency densities for the larger
class intervals.
10 Introduction

Exercise
Consider results of a test written by 45 students and marked out of 70. Data is pre-
sented in categories in table below. Use the data above to draw a histogram for the

Marks Frequencies
10 - 19 7
20 - 29 20
30 - 39 9
40 - 49 3
50 - 59 5
60 - 69 1

mark distribution.

2.4.4. Stem and leaf diagram

A stem and leaf diagram is basically a histogram where the rectangles are built up to
the correct height by individual numbers. Each data value is split up into its stem,
the first digit [or first two digits, etc., depending on the data], and its leaves. Thus, the
number 23 has stem 2 and leaf 3. The number 7 has stem 0 and leaf 7. Perhaps an
example will illustrate this diagram.

Example 1.3
A scientist interested in finding out the age groups of people interested in cultural
movies went to a movie theatre and collected the following information. Ages of people
watching movie is shown below.

7 15 22 38 12 18 14 26 20 15 22 34 12 18 24
19 14 29 21 32 12 17 24 13 25 20 15 31 11 16
23 39 19 14 28 20 9 16 22 39 13 25 19 14 31

The stem 0, 1, 2 and 3 are listed on the left side of a vertical line and the leaves
on the right side opposite the appropriate stem. The stem and leaf diagram of these
data are represented below. A stem and leaf display should always have a key that
indicates how data is displayed. Key: 0|7 = 7, 3|8 = 38.

Table 2.1: Stem Plot of Ages, Key: 1|1 = 11

Stem Leaf
0 79
1 122233444455566788999
2 000122234455689
3 1124899

Also take note that 1st , 2nd , 3rd , e.t.c. number on the leaf side should be in the same
columns for the histogram feature to reveal.
Data and Data Presentation 11

2.4.5. Frequency Polygons

Frequency polygons are one alternative to histograms. The only difference here is that
a frequency polygon is a line plot of the frequencies against the corresponding class
mid-points. The points are joined by straight lines.

2.5. Exercises
1. Classify the following data sources as either Primary or secondary and Internal
or external

(a) The economic statistics quoted in The Financial Gazette.

(b) The sum assured values on life assurance polices within your company.
(c) The financial reports of all companies on the Zimbabwean stock exchange
for the purpose of analyzing earnings per share.
(d) Employment statistics published by ZimStats.
(e) Market research findings on driving habits conducted by the ZRP Traffic
section.

2. Define primary and secondary data. Include in your answers the advantages and
disadvantages of both data types. Give two examples of secondary data.

3. What is the difference between primary and secondary data?

4. Areas of continents of the World

Continent Area in millions of km2

Africa 30.3
Asia 26.9
Europe 4.9
North America 24.3
Oceania 8.5
South America 17.9
Russia 29.5

(a) Draw a bar chart of the above information.

(b) Construct a pie chart to represent the total area.

5. The distance (km) travelled by a courier service motorcycle on 30 trips were

recorded by the driver.

24 19 21 27 20 17 17 32 22 26 18 13 23 30 10
13 18 22 34 16 18 23 15 19 28 25 25 20 17 15
12 Introduction

(a) Define the random variable, the data type and the measurement scale.
(b) From the data, prepare:
i. an absolute frequency distribution
ii. a relative frequency distribution and
iii. the (relative) less than ogive.
(c) Construct the following graphs:
i. a histogram of the relative frequency distribution,
ii. stem and leaf diagram of the original data
(d) From the graphs, read off what percentage of trips were:
i. between 25 and 30 km long
ii. under 25km
iii. 22km or more?
Chapter 3

Measures of Central Tendency

3.1. Introduction
From the previous unit, graphical displays were discussed. These are useful means
of communicating broad overviews of the behaviour of a random variable. However,
there is a need for numeral measures, called statistics, which will convey more pre-
cise information about the behaviour pattern of a random variable. The behaviour or
pattern of any random variable can be described by measures of:
• Central tendency and

• Dispersion of observations about a central value.

3.2. Measures of Central Tendency

These are statistical measures which quantify where the majority of observations are
concentrated. They are also called measures of location. A central tendency statistic
represents a typical value or middle data point of a set of observations and are useful
for comparing data sets. These measures may be based on the source that is whether
they are from a population or from a sample. If from a population, we talk of a param-
eter and if from a sample we refer to statistic. The three main measures of central
tendency are:
• Arithmetic mean or average

• Mode and

• Median
Each measure will be computed for ungrouped data and grouped data.

3.3. Arithmetic Mean

Given a set of n sample data values denoted by xi for i = 1, 2, 3, ...n, the arithmetic
mean is defined by x̄. For a population, we refer to a population mean, µ
14 Measures of Central Tendency

3.3.1. Mean for ungrouped data

The arithmetic mean for ungrouped data from a sample is defined as in:

sum of all observations

x̄ =
total number of observations
Xn
xi
i=1
x̄ = (3.1)
n

The population mean is defined as:

n
X
xi
i=1
µ =
N

Where n is the number of observations in the sample, xi is the value of the ith observa-
Pn
tion of random variable x and x̄ is the symbol for a sample arithmetic mean. i=1 xi
is the shorthand notation for the sum of n individual observation i.e.
Xn
xi = x1 + x2 + x3 + ... + xn
i=1

3.3.2. Mean for grouped data

Grouped data is represented by a frequency distribution. All that is known is the

frequency with which observations appear in each of the m classes. Thus the sum
of all the observation cannot be determined exactly. Consequently, it is not possible
to compute an exact arithmetic mean for the data set. The computed mean is an
approximation of the actual arithmetic mean.
m
X
fi xi
x̄ = i=1
P (3.2)
fi

The population mean for grouped data is:

m
X
fi xi
i=1
µ = (3.3)
N

Where m is the number of classes in the frequency distribution. n is the number of

observations in the sample xi is the value of the ith observation of random variable x,
fi is frequency of the ith class, x̄ is the symbol for a sample arithmetic mean. For you
practice attempt examples in the Tutorial Work Sheet.
Measures of Central Tendency 15

Properties of the Mean

• The arithmetic mean uses all values of the data set in its computation.

• The sum of deviation of each observation from the mean value is equal zero.
Pn
i.e. i=1 (xi − x̄) = 0. This makes the mean an unbiased statistical measure of
central location.

Drawbacks of the mean

• It is affected or distorted by extreme values (Outliers) in the data.

• It is not valid to compute the mean for nominal- or ordinal-scaled data. It is

only meaningful to compute the arithmetic mean for ratio-scaled data (discrete
or continuous)

There are other means that can be calculated for different distribution of values
which are Harmonic mean, Geometric mean and Weighted arithmetic mean.
We are not discuss them at this time.

3.4. The Mode

The mode of a given set of data is the observation with the highest frequency. In
other words it is the most occurring value in a data set. It is that observation with the
highest frequence. A distribution can have one mode (unimodal), two models (bimodal)
or many modes(multimodal). Calculations for a mode from a population and sample is
the same.

3.4.1. Mode for ungrouped data

In ungrouped data set, mode is obtained by observing the data carefully then finding
the most frequently occurring observation. However, if the number of observations is
too large, the mode can be found by arranging the data in ascending order and by in-
spection, identify that value that occurs frequently.

Example 1
If B=Blue, G=Green, R=Red and Y=Yellow. Consider a sample: YGBRBBRGYB, picked
from a mixed bag. What is the modal colour?

Solution
The modal colour is Blue, because it appears most, with a frequency of 4.
16 Measures of Central Tendency

3.4.2. Mode for grouped data

In finding mode for grouped data, we first identify the modal class i.e. the class interval
with the highest frequency. The mode lies in this class and then calculate the modal
value using the formula

c(f1 − f0 )
M ode = lmo + (3.4)
2f1 − f0 − f2

where lmo is the lower limit of modal class, f1 is the frequency of the modal class, f0
is the frequency of the class preceeding the modal class, f2 is the frequency of interval
succeeding the modal class and c is the width of the modal class.

Example 2

Find the modal test mark for the following data.

Test mark, x 5 - 10 10 - 15 15 - 20 20 - 25 25 - 30
Frequency 3 5 7 2 4

Solution
We seek to invoke the formula

c(f1 − f0 )
M ode = lmo +
2f1 − f0 − f2

where 15-20 is the modal class with the highest frequency of 7, lmo = 15, f1 = 7,
f0 = 5, f2 = 2, and c = 5. Substituting these in the equation above yields

5(7 − 5)
M ode = 15 +
2(7) − 5 − 2
M ode = 16.42

3.5. The Median

The median is the value of a random variable which divides an ordered (ascending or
descending order) data set into two equal parts. It is also called the second quartile
Q2 or 50th percentile. Half of the observation will fall below this median value and
the other half above it. If the number of observations, n, is odd, then the median is
observation ( n+1 th
2 ) . If the number of observations is even, then median is the mean
of ( n2 )th and ( n2 + 1)th observation. First, we consider a modified scenario of ungrouped
data.
Measures of Central Tendency 17

3.5.1. Median for ungrouped data

Given the following income data presented in a frequency distribution table, find the
median.

Income ($) 3800 4100 4400 4900 5200 5500 6000

Number of workers 12 13 25 17 15 12 6

Solution
The number of observations is 100, which is even thus the median is mean of ( n2 )th
and ( n2 + 1)th observations i.e. the mean of the 50th and 51st observations. To find
these observations we first find the cumulative frequencies of the data set. The 50th
observation is 4400 and 51st observation is 4900. Thus

4400 + 4900
M edian =
2
M edian = 4650

Interpretation
This means 50% of the workers get incomes that are less than $4650 and another 50%
get an income that is more than $4650.

3.5.2. Median for grouped data

Given the following grouped data in a frequency table, find the median.

Income ($) 3601- 3801- 4101- 4401- 4901- 5201- 5501-

3800 4100 4400 4900 5200 5500 6000
Number of workers 12 13 25 17 15 12 6
Cumulative frequency 12 25 50 67 82 94 100

We use a standard formular to calculate the median of the above grouped data which
is

c( n2 − F (<))
M edian = Ome + (3.5)
fme

where me is the median class, Ome is the lower limit of median class, n is the sample
size i.e. total number of observations, F (<) is the cumulative frequency of class prior
to median class and c is the width of the median class.

To use this formula, we calculate the cumulative frequencies and then identify the me-
dian class which is the class containing the ( n+1 th
2 ) observation.
18 Measures of Central Tendency

Example 4:
Calculate the median of following grouped data.

Mark 0-10 10-20 20-30 30-40 40-50

Frequency 2 12 22 8 6

Solution:
First and foremost, order the data set, in this case its already ordered. Then calculate
the cumulative frequencies we get.

Mark 0-10 10-20 20-30 30-40 40-50

Frequency 2 12 22 8 6
Cumulative Frequency 2 14 36 44 50

We wish to use the formula

c( n2 − F (<))
M edian = Ome +
fme

Where the median class is 20 - 30, c = 10, n = 50, F (<) = 14, f = 22, fme = 22 and
Ome = 20. Substituting we have

10[ 50
2 − 14]
Me = 20 +
22
Me = 25

Interpretation
This implies that 50% of the students got less than 25 marks and the other 50% got
more than 25 marks.

The advantage of the median is that it is unaffected by outliers and is a useful measure
of central tendency when the distribution of a random variable is severely skewed. A
disadvantage of the median, however, is that it is inappropriate for categorical data.
It is best suited as a central location measure for interval-scaled data such as rating
scales.

3.6. Quartiles

Quartiles are those observations that divide an ordered data set into quarters (four
equal parts). Lower Quartile, Q1 is the first quartile or 25th percentile. It is that
observation which separates the lower 25 percent of the observations from the top 75
percent of ordered observations. Middle Quartile, Q2 is the second quartile or 50th
percentile or the median. It divides an ordered data set into two equal halves. Upper
Measures of Central Tendency 19

Quartile, Q3 is the third quartile or 75th percentile. It is that observation which di-
vides observations into the lower 75 percent from the top 25 percent.

To compute quartiles, a similar procedure is used as for calculating median. The only
difference lies in (i) the identification of the quartile position, and (ii) the choice of the
appropriate quartile interval. Each quartile position is determined as follows:

For Q1 use n4 , for Q2 use n2 , and for Q3 use 3n

4 to calculate the position of the respective
quartile. The appropriate quartile interval is that interval into which the quartile
position falls. Like the median calculations, this is identified using the less than ogive.
A forumular for Q1 is:

c( n4 − F (<))
Q1 = Oq1 +
fq1

where Oq1 is the lower limit of the class interval with lower quartile value, F (<) is the
cumulative frequency of the class interval before the lower quartile interval and fq1 is
the frequency of the lower quartile interval.

3.6.1. Quartiles for ungrouped data

Ungrouped data situation is easy to calculate the quartiles. Simply identify the quar-
tile positions and identify the value of the variable that lies at that position.

Exercise
Using income data below, find Q1 and Q2 .

Income ($) 3800 4100 4400 4900 5200 5500 6000

Number of workers 12 13 25 17 15 12 6

Solution
n = 100, hence Q1 position is at n4 = 100 th
4 = 25 position. Arranging number of workers
cummulatively i.e. coming up with a cummulative distribution table 25th value lies at
income $4100. Hence Q1 is $4100. Show that Q3 is $5200.

3.6.2. Quartiles for grouped data

In calculating quartiles for grouped data, use of the formular is required since the
position of the quartile will be a interval of numbers. The formular allows us to find
20 Measures of Central Tendency

the exact value. Find the first, second and third quartile values from the distribution
below.

Mark 0-9 10-19 20-29 30-39 40-49

Frequency 2 12 22 8 6

The lower quartile, Q1

Q1 position n4 = 50 th th
4 = 12.5 position. Q1 interval = [10 - 19] because the 12.5 observa-
tion falls within these limits. The formula for Q1 is:

c( n4 − F (<))
Q1 = Oq1 +
fq1

Where Q1 is the lower quartile, Oq1 is the lower limit of Q1 Interval (class), n is the
sample size (total number of observations), F (<) is the cumulative frequency of the in-
terval before the Q1 interval, fq1 is the frequency of the Q1 interval and c is the width
of the Q1 interval.

Thus:
c[ n4 − F (<)] 10[ 50
4 − 2]
Q1 = Oq1 + = 10 + = 18.75
fq1 12
Interpretation:
25 % of the students got below 18.75 marks.

3.6.3. The second quartile, Q2 (Median)

Q2 position, use n2 = 502 = 25
th position. Q interval = [20 - 29] because the 25th
2
observation falls within these limits. The formula for Q2 is:

c[ n2 − F (<)]
Q2 = Oq2 +
fq2

Which gives Q2 = 25 marks.

3.6.4. The upper quartile, Q3

Q3 position 3n
4 =
3×50
4 = 37.5th position. Q3 interval = [30 - 39] because the 37.5th
observation falls within these limits. The formula for Q3 is:

c[ 3n
4 − F (<)]
Q3 = Oq3 +
fq3

where Q3 is the upper quartile, Oq3 is the lower limit of Q3 class interval, n is the
sample size (i.e. total number of observations), F (<) is the cumulative frequency of
the interval before the Q3 interval, fq3 is the frequency of the Q3 interval and c is the
width of the Q3 interval.
Measures of Central Tendency 21

Thus:
c[ 3n
4 − F (<)] 10[ 3×50
4 − 36]
Q3 = Oq3 + = 30 + = 31.875
fq3 8

Interpretation:
75% of the students got below 31.875 marks. Alternatively, 25% of the students got
above 31.875 marks.

3.6.5. Percentiles

In general, any percentile value can be found by adjusting the median formula to:
(i)Find the required percentiles position and from this and (ii) Establish the percentile
interval.

Example
90th percentile position = 0.9 × n, 35th percentile position = 0.35 × n, 25th percentile
position(Q1 ) = 0.25 × n

Uses of percentiles:
Percentiles are used to identify various non-central values. For example, if it is desired
to work with a truncated dataset which excludes extreme values at either end of the
ordered dataset.

3.7. Skewness

Skewness is departure from symmetry. Departure from symmetry is observed by com-

paring the mean, median and mode.

1. If mean = median = mode the frequency distribution is Symmetrical.

2. If mean < median < mode the frequency distribution is negatively skewed i.e.
skewed to the left.

3. If mean > median > mode the frequency distribution is positively skewed i.e.
skewed to the right.

Remark:

1. If a distribution is distorted by extreme values (i.e. skewed) then the median or

the mode is more representative than the mean.
22 Measures of Central Tendency

2. If the frequency distribution is skewed, the median may be the best measure of
central location as it is not pulled by extreme values, nor is it as highly influenced
by the frequency of occurrence.

3.8. Kurtosis
Kurtosis is the measure of the degree of peakedness of a distribution. Frequency dis-
tributions can be described as: leptokurtic, mesokurtic and platykurtic.

• Leptokurtic- highly peaked distribution i.e. a heavy concentration of observa-

tions of around the central location.

• Mesokurtic moderately peaked distribution.

• Platykurtic flat distribution i.e. the observations are widely spread about the
central location.

3.9. Exercises
1. The number of days in a year that employees in a certain company were away
from work due to illness is given in the following table:

Sick days Number of employees

5-6 67
7-8 91
9-10 67
11-12 5

Find the modal class and the modal sick days and interpret.

2. A company employs 12 persons in managerial positions. Their seniority (in years

of service) and sex are listed below:

Sex F M F M F M M F F F F M
Seniority (yrs) 8 15 6 2 9 21 9 3 4 7 2 10

(a) Find the seniority mean, median and mode for the above data.
(b) Which of the mean, median and mode is the least useful measure of location
for the seniority data? Give a reason for your answer.
(c) Find the mode for the sex data. Does this indicate anything about the em-
ployment practice of the company when compared to the medians for the
seniority data for males and females?
Chapter 4

Measures of Dispersion

4.1. Introduction
Spread or Dispersion refers to the extent by which the observations of a random vari-
able are scattered about the central value. Measures of dispersion provide useful infor-
mation with which the reliability of the central value may be judged. Widely dispersed
observations indicate low reliability and less representativeness of the central value.
Conversely, a high concentration of observation about the central value increases con-
fidence in the reliability and representativeness of the central value.

4.2. Range
The range if the difference between the highest and the lowest observed values in a
dataset.

For ungrouped dataset,

Range = Xmax − Xmin

For grouped dataset,

Range = U pper limit of last interval − Lower limit of f irst interval

The range is a crude estimate of spread. It is calculated, but is distorted by extreme

values (outliers). An outlier would either be xmax or xmin . It is therefore a volatile and
unstable measure of dispersion. It also provides no information on the clustering of
observations within the dataset about a central value as it uses only two observations
in its computation.

Example 6:
Given the following data in a frequency distribution table, find the range.
Solution:
24 Measures of Dispersion

Income /$ 3800 4100 4400 4900 5200 5500 6000

Number of Workers 12 13 25 17 15 12 6

Range = Xmax − Xmin = 6000 − 3800 = 2200

For grouped distribution with class intervals, xmin is the lower limit of the lower class
interval and xmax is the upper limit of the highest class interval.

Interquartile range, IQR

Because the range can be distorted by extreme values, a modified range which ex-
cludes these outliers is often calculated. This modified range is the difference between
the upper and lower quartiles i.e.

Interquartile Range = Q3 − Q1

This modified range removes some of the instability inherent in the range if out-
liers are present, but it excludes 50 percent of all observations from further analysis.
This measure of dispersion, like the range, also provides no information on the clus-
tering of observations within the dataset as it uses only two observations.

Quartile deviation
A measure of variation based on this modified range is called quartile deviation (QD)
or the semi-interquartile range. It is found by dividing the interquartile range in half
i.e.
Q3 − Q1
Quartile deviation =
2
Remember when calculating this measure you order your dataset first to calculated Q3
and Q1 . The quartile deviation is an appropriate measure of spread for the median. It
identifies the range below and above the median within which 50 percent of observa-
tions are likely to fall. It is a useful measure of spread if the sample of observations
contains excessive outliers as it ignores the top 25 percent and bottom 25 percent of
the ranked observations.

4.3. Variance
The most useful and reliable measures of dispersion are those that take every observa-
tion into account and are based on an average deviation from a central value. Variance
is such a measure of dispersion. Population variance is denoted by σ 2 whereas sample
variance is denoted by s2 .
Measures of Dispersion 25

Variance for ungrouped data

Sample variance for ungrouped data is given by:
Pn
2 − x̄)2
i=1 (xi
s =
n−1

Population variance is given by:

PN
2 i=1 (xi − µ)2
σ =
N

The main difference being on the denominator of the two. Population variance divides
by N whereas sample variance divide by n − 1.

Consider the ages, in years, of 7 second hand cars: 13, 7, 10, 15, 12, 18, 9. Find the
variance of the ages of cars.

Solution
Step 1: Find the sample mean, x̄ = 84
7 = 12 years.
Step 2: Find the squared deviation of each observation from the sample mean. See
table below.

Car age, xi Mean, x̄ Deviation (xi − x̄) Deviations squared (xi − x̄)2
13 12 +1 1
7 12 -5 25
10 12 -2 4
15 12 +3 9
12 12 0 0
18 12 +6 36
9 12 -3 9
(xi − x̄)2 = 84
P P
(xi − x̄) = 0

Step 3: Find the average squared deviation that is the variance using the formular:

84
S2 = = 14 years2
7−1
Note:
Divison by n would appear logical, but the variance statistic would then be a biased
measure of dispersion. It can be shown to be unbiased if division is by (n−1). For large
samples i.e. n is greater than 30, however this distinction becomes less important.

Variance can be also calculated using a formular below. It gives similar results as
the above formular:
26 Measures of Dispersion

x2i − nx̄2
P
S2 =
n−1
x2 = 1092,
P P
x = 84, n=7 and x̄ = 12, substituting the values in the above
formular:
[1092 − 7(122 )] 84
S2 = = = 14 years2
7−1 6
Variance for grouped data
Grouped data is data presented in a frequency distribution table. Sample variance for
such grouped data is calculated using the formular:
Pn
2 − x̄)2
i=1 f (xi
S =
n−1

or,
f x2i − nx̄2
P
2
S =
n−1
Population variance is given by:

f x2i − N µ2
P
2
σ =
N

Example 7:
Consider data for student marks obtained from Test 1. Calculate the variance of the
student marks.

Marks 0-10 10-20 20-30 30-40 40-50

Frequency 2 12 22 8 6

Solution
The midpoint in an inteval is calculated as:

Lower limit + U pper limit

M idpoint =
2

Marks, x Frequency, fi Midpoint, xi fi xi x2 f x2

0+10
0-10 2 2 =5 10 25 50
10-20 12 15 180 225 2700
20-30 22 25 550 625 13750
30-40 8 35 280 1225 9800
40-50 6 45 270 2025 12150
Total 50 1290 38450

P
fx 1290
f x2 = 38450. Using the above formular, the
P
Mean, x̄ = n = 50 = 25.8 and
variance,
Measures of Dispersion 27

f x2i − nx̄2
P
S2 =
n−1
38450 − 50(25.8)2 5168
S2 = = = 105.47 marks2
50 − 1 49
The variance is a measure of average squared deviation about the arithmetic mean.
It is expressed in squared units. Consequently, the meaning in a practical sense is
obscure. To provide meaning, the measure should be expressed in the original units of
the random variable.

4.4. Standard deviation

A standard deviation is a measure which expresses the average deviation about the
mean in the original units of the random variable. The standard deviation is the
square root of the variance. Mathematically:

A sample standard deviation is:

p √
Sx = Sample variance = s2
sP
f x2i − nx̄2
Sx =
n−1
A population standard deviation is
rP
2 f x2i − N µ2
σ =
N

The standard deviation is a relatively stable measure of dispersion across different

samples of the same random variable. It is therefore a rather powerful statistic. It
describes how the observations are spread about the mean.

4.5. Coefficient of variation

From a sample, the coefficient of variation is defined as follows

S
CV = × 100%
x̄

whereas a population coefficient of variation is

σ
CV = × 100%
µ

.
This ratio describes how large the measure of dispersion is relative to the mean of
28 Measures of Dispersion

the observation. A coefficient of variation value close to zero indicates low variability
and a tight clustering of observations about the mean. Conversely, a large coefficient
of variation value indicates that observations are more spread out about their mean
value.

From our example above,

S 10.27
CV = × 100% = × 100% = 39.8%.
x̄ 25.8

4.6. Exercises
1. Find the mean and the standard deviation for the following data which records
the duration of 20 telephone hotline calls on the 0772 line for advice on car re-
pairs.

Duration Number of calls

0-≤1 7
1-≤2 0
2-≤3 3
3-≤4 1
4-≤5 9

At a cost of $2.60 per minute, what was the average cost of a call, and what
was the total cost paid by the 20 telephone callers. Calculated the coefficient of
variation and interpret it.

2. Employee bonuses earned by workers at a furniture factory in a recent month

(US$) were:

47 31 42 33 58 51 25 28
62 29 65 46 51 30 43 72
73 37 29 39 53 61 52 35

From the table above, find the

(a) Mean and standard deviation of bonuses.

(b) Interquartile range and quartile deviation.
(c) Coefficient of variation and comment.

3. Give three reasons why the standard deviation is regarded as a better measure
of dispersion than the range.

4. Discuss briefly which measure of dispersion would you use if the:

(a) mean is used as the measure of central location and why?

Measures of Dispersion 29

(b) median is used as a measure of central location and why?

5. Discuss the limitations of the range as a measure of dispersion.

6. Define the following terms as they are used in statistics.

(a) Outliers
(b) Skewness
(c) Kurtosis
30 Measures of Dispersion
Chapter 5

Basic Probability

5.1. Introduction
This unit will introduce to you simple concepts and terminologies in probability. These
include events, types of probabilities and rules of probabilities. Probability theory
is fundamental to the area of statistical inference. Inferential statistics deals with
generalising the behaviour of random variables from sample findings to the broader
population. Probability theory is used to quantify the uncertainties involved in making
these generalisations.

5.2. Definition
An Event is a collection of possible outcomes from an experiment or a trial. For ex-
ample Heads or Tails are events which can be obtained from tossing a fair coin.

An Experiment is a process which generates events or outcomes. For instance, toss-

ing a fair die three times constitutes an experiment.

Probability is the chance or likelihood of a particular outcome out of a number of

possible outcomes occurring for a given event. Thus probability is a number between
0 and 1, which quantifies likelihood of an event occurring or not occurring. Therefore
probability range is 0 ≤ P (A) ≤ 1, where A is an event of a specific type.

Most decisions are made in the face of uncertainty. Probability is therefore, concerned
with uncertainty.

5.3. Approches to probability theory

There are two broad approaches to probability, namely Subjective and Objective.
32 Basic Probability

Subjective probability
It is probability which is based on a personal judgement that a given event will occur.
There is no theoretical or empirical basis for producing subjective probabilities. In
other words this probability of an event based on an educated guess, expert opinion or
just plain intuition. Subjective probabilities cannot be statistically verified and there
are not extensively used, hence will not be considered further.

Examples

1. When commuters board a commuter omnibus, they assume that they will arrive
safely at their destinations, so P(arriving safely) = 1.

2. If you invest, you assume that you will get a good return, so P (good return) =
0.9.

Objective probabilities
These are probabilities that can be verified, through repeated experimentation or em-
pirical observations. Mathematically it is defined as a ration of two numbers:

r
P (A) =
n

Where A is event of a specific type, r is number of outcomes of event A, n is total num-

ber of possible outcomes (also called the sample space) and P(A) is probability of event
A occurring.

Objective probabilities are derived either:

• a priori - that is when possible outcomes are known in advance such as tossing
a coin,selecting cards from a deck of cards. Classical probability.

N umber of outcomes f avouring event A

P (A) =
T otal number of possible outcomes

For example the probability of a Head if a fair coin is tossed once is, P (Head) =
1
2 = 0.5

• Empirically that is when the values of r and n are not known in advance and
have to be observed through data collection or from a relative frequency table you
can deduce probability of the different outcomes.

N umber of times event A has occured

P (A) =
N umber of times event A could have occured
Basic Probability 33

For instance, If out of a random sample of 90 customers 50 said they prefer Bak-
ers Inn bread, then relative frequency that a randomly selected customer will
50
prefer Bakers Inn bread is 90 = 0.56

• Theoretically - that is through use of theoretical distribution functions (math-

ematical formula that can be used to compute probabilities for certain event
types).

These probabilities are used extensively in statistical analysis.

5.4. Properties of probability

1. A probability value lies only between 0 and 1 that is 0 ≤ P (A) ≤ 1.

2. If an event A cannot occur (i.e. an impossible event), then P (A) = 0

3. If an event A is certain to occur, then P (A) = 1

4. The sum of probabilities of all possible outcomes of a random experiment equals

n
X
one, that is P (Ei ) = 1.
i=1

5. Complementary probabilities: If P (A) is the probability of event A occurring,

then the probability of event A not occurring is defined as: P (Ac ) = 1P (A). Note:
P (Ac ) is also sometimes written as P (Ā)orP (A0 ).

Example
Consider random process of drawing cards from a card deck. These probabilities are
called a priori probabilities.

26 1
1. Let A = event of selecting a red card. Then P (Red card) = 52 = 2 (26 possible red
cards out of 52 cards).

13 1
2. Let B = event of selecting a spade. Then P (Spade) = 52 = 4 (13 possible spades
out of a total of 52).

4 1
3. Let C = event of selecting an ace. Then P (Ace) = 52 = 13 (4 possible ace out of a
total of 52 cards).

1 12
4. Let D = event of selecting ’not an ace’. Then P (not an ace) = 1P (ace) = 1− 13 = 13 .
34 Basic Probability

5.5. Basic probability concepts

1. Intersection of Two events
The intersection of two events A and B is the set of outcomes that belongs to both
A and B simultaneously. It is written as A ∩ B i.e. A and B and the keyword is
and.

2. Union of Two events

The union of Events A and B is the set of outcomes that belongs to either A or B
or both and the key word is or. It is written as A ∪ B.

3. Complement of an event
The complement of an event A is the collection of all possible outcomes that are
not contained in event A. That is P (Ac ) = 1P (A). Note P (Ac ) is also sometimes
written as P (Ā) or P (A0 ). In other words P (A) + P (A0 ) = 1.

5.6. Types of events

1. Mutually Exclusive or disjoint events
These are events which cannot occur at the same time. The occurrence of one
event automatically prevents the occurrence of the other event. For mutually ex-
clusive events the intersection of events is empty i.e. there are no common events.

Examples

(a) Passing and failing the same examination are mutually exclusive. In other
words its not possible to pass and fail at the same time one examination.
(b) In tossing a fair die once, getting a 3 and a 5 are mutually exclusive. You get
one outcome at time and not both.

2. Non-Mutually Exclusive Events

These are events which can occur simultaneously. The occurrence of one event
does not prevent the occurrence of the other events. The intersection set of the
events has got elements.

Examples

(a) In tossing a fair die once, getting an odd number or a number greater than
2 are non mutually exclusive events i.e. it is possible for the number to be
odd and at the same time being greater than 2.
(b) An individual can have more than one bank account i.e. if you open a bank
account it does not prevent you from opening another account with another
bank.
Basic Probability 35

3. Collectively exhaustive events

Events are said to be collectively exhaustive when the union of all possible events
is equal to the sample space. This means that, in a single trial of a random ex-
periment, at least one of these events is certain to occur.

Example
Consider a random experiment of selecting companies from the Zimbabwe Stock
Exchange (ZSE). Let event A = small company, event B = medium company and
event C = large company. Then (A ∪ B ∪ C) = sample space (small, medium, large
companies) = all ZSE companies.

4. Statistically Independent events

Two events are said to statistically independent if the occurrence of event A has
no effect on the outcome of event B occurring and vice versa.

Example
Let A = event that an employee is over 30 years of age, B = event that the em-
ployee is female. If it can be assumed or empirically verified that a randomly
selected employee is over 30 years of age from a large organisation is equally
likely to be either male or female employee, then the two events A and B are
statistically independent.

5. Statistically Dependent events

Events are dependent if the occurrence of one of the event A affects the occur-
rence of the second event B. These will be discussed under conditional probability.

Remark
The terms Statistically independent events and mutually exclusive events should
not be confused. They are two very different concepts. When two events are
mutually exclusive, they are NOT Statistically independent. They are dependent
in the sense that if one event happens, then the other event cannot happen. In
probability terms, the probability of the intersection of two mutually exclusive
events is zero, while the probability of two independent events is equal to the
product of the probabilities of the separate events.

5.7. Laws of probability

There are generally two laws in probability theory, namely, Addition Laws and Mul-
tiplication Laws. Addition laws pertain to mutually and non-mutually exclusive
events only. The key word is OR. What is the probability that event A OR B will
occur:
36 Basic Probability

• For Mutually Exclusive events:

P (AorB) = P (A) + P (B)
P (A ∪ B) = P (A) + P (B)
Note: Union (∪) means OR

Example
What is the probability of getting a 5 or 6 if a fair die is tossed once?

Solution The sample space has six possible outcomes 1, 2, 3, 4, 5, 6. Therefore

P (5or6) = P (5) + P (6) = 16 + 61 = 13

• For Non-mutually Exclusive events

P (A or B) = P (A) + P (B) − P (A and B)
P (A ∪ B) = P (A) + P (B)P (A ∩ B)

The intersection sign, ∩ means the joint probability of events A and B. P (A and B)
is subtracted to avoid double counting.

Example
What is the probability of getting an even number or a number less than four if
a fair die is tossed once?

Answer
Let event A = getting an even number and the elements are 2, 4, 6 and event B
= getting a number less than four and the elements are 1, 2, 3. Then P (A) =
3 3 1
6 and P (B) = 6 . Thus P (A and B) = 6 . There is only one element which is
common in A and B and the number is 2. Therefore
P (A or B) = P (A) + P (B) − P (A ∩ B) = 36 + 36 − 16 = 56

Exercise 1
Sixty per cent of the population of a town read either magazine A or magazine B and
10% read both. If 50% read magazine A, what is the probability that one person, se-
lected at random, read magazine B?

Multiplication Laws
Multiplication Laws pertain to dependent and independent events. The key word is
AND

a) For Independent events: P (A and B) = P (A) × P (B)

Basic Probability 37

Example
What is the probability of getting a tail when two fair coins are tossed at the
same time?

Answer
P (T 1 and T 2) = P (T 1) × P (T 2)
T1 = the probability of getting a tail from coin 1. T2 = the probability of getting a
tail from coin 2. The two outcomes do not affect each other. Therefore
P (T 1 and T 2) = 12 × 12 = 41

b) Dependent events will be discussed on the section on conditional probability.

5.8. Types of probabilities

Objective probabilities can be classified into 3 categories, namely:

• Marginal Probability

• Joint Probability

• Conditional probabilities

Marginal Probability
It is the probability of only a single event A occurring regardless of certain conditions
prevailing. It is written as P (A). A frequency distribution describes the occurrence of
only one characteristic of interest at a time and is used to estimate marginal probabil-
ities.

Joint Probability
It is the chance that two or more events will occur simultaneously. It is the occurrence
of events at the same time. If the joint probability on any two events is zero, then the
events are mutually exclusive.

Conditional Probability
It is the probability that a given event occur under the conditions that another event
has already occurred. The symbol P (A|B) is the probability that the event A will occur
given that event B has already occurred.

P (A ∩ B)
P (A|B) = (5.1)
P (B)
38 Basic Probability

Sex
Payment Method Total
Male Female
Credit Card 10 15 25
Cash 8 6 14
Total 18 21 39

provided P (B) > 0, and similarly

P (B ∩ A)
P (B|A) = (5.2)
P (A)

provided P (A) > 0

Note: P (A ∩ B) = P (B|A).P (A) = P (A|B).P (B). P(A and B) is the joint probability of
events A and B. P (B) is the probability of event B, which is a marginal probability.
Example: Joint Marginal and Conditional Probabilities
Consider the table of fees payment methods by sex
What is the probability of getting a person who is

a) (i) Female and uses a credit card?

(ii) Male and uses cash?

b) (i) Credit card user?

(ii) Female?

c) (i) Female given that she uses cash?

(ii) Credit card user given that he is a male?

Answer

(a) Question a) is joint probability of the events. The sample space has 39 people
altogether.
15
(i) P(female and credit card) = 39 = 0.3846.

Note: The two events should not be confused with independent events. In this
case find the value in the intersection set of female column and credit card row
which is 15.

8
(ii) P(male and cash) = 39 = 0.2051.
It is the chance of two events occurring at the same time.

(b) Question b) is marginal probability, the conditions prevailing is sex or payment

method.
Basic Probability 39

25
(i) P(credit card user) = 39 = 0.6.

Note: The prevailing condition which has been ignored is sex.

(ii) P(female) = 21
39 = 0.5385.

The condition which has been ignored is payment method. For joint probabilities,
consider values inside the table as a ration of the grand total 39. For marginal
probabilities consider row and column totals as ratios of grand total.

c) Question c) is conditional probability.

6
P (F emale and Cash) 6
i) P(female — cash user) = P (Cash) = 39
14 = 14 = 0.4286
39

10
P (Credit card and M ale) 5
ii) P(credit card — male)= P (M ale) = 39
18 = 9 = 0.5556
39

Exercise 2
A golfer has 12 golf shirts in his closet. Suppose 9 of these shirts are white and the
others are blue. He gets dressed in the dark, so he just grabs a shirt and puts it on.
He plays golf two days in a row and does not do laundry. What is the likelihood both
shirts selected are white?

5.9. Contigency Tables

A contingency table is a cross-tabulation that simultaneously summarise two variables

of interest and their relationship. The level of measurement can be nominal. It is a
table that is used to classify sample observations according to two or more identifiable
characteristics.

Example
A survey of 150 students classified each as to gender and the number of movies at-
tended last month. Each respondent is classified according to two criteria, that is, the
number of movies attended and gender.

Gender
Movies Attended Total
Male female
0 20 40 60
1 40 30 70
2 or more 10 10 20
Total 70 80 150
40 Basic Probability

5.10. Tree diagram

A tree diagram is a graph that is helpful in organising calculations that involves sev-
eral stages. Each branch in the tree is one stage of the problem. The branches of the
tree diagram are weighted by probabilities.

Exercise 3
Using the previous example of 150 students? Using a tree diagram what is the proba-
bility of selecting a male student given that he has seen one movie?

5.11. Counting rules

Probability computations involve counting the number of successful outcomes (r) and
the total number of possible outcomes (n) and expressing them as a ratio. Often the
values of rand n are not feasible to count because of the large number of possible
outcomes involved. Counting rules assist in finding values of r and n. There are three
basic counting rules:

• Multiplication rule

• Permutations

• Combinations.

5.11.1. Multiplication Rule

The multiplication rule is applied in two ways:

a) The total number of ways in which n objects can be arranged (ordered) is given
by:
n! = nf actorial = n(n − 1)(n − 2)(n − 3)...............3.2.1
Note that 0! = 1.

Example
The number of different ways in which 7 horses can complete a race is given by:
7! = 7.6.5.4.3.2.1 = 5040 different arrangements.

b) If a particular random process has

• n1 possible outcomes on the first trial

• n2 possible outcomes on the second trial e.t.c.
• nj possible outcomes on the first trial.
Basic Probability 41

Then the total number of outcomes for the j trials is: n1 × n2 × n3 × n4 × ............ × nj
Example
A restaurant menu has a choice of 4 starters, 10 main courses and 6 desserts. What is
the total number of meals that can be ordered in this restaurant.

Solution
The total numbers of possible meals that can be ordered are: 4 × 10 × 6 = 240 meals.

5.11.2. Permutations
A permutation is number of distinct ways in which a group of objects can be arranged.
Each possible arrangement (ordering) is called a permutation. The number of ways of
arranging r objects selected from n objects where ordering is important, is given by the
formula:

n!
Prn = (5.3)
(n − r)!

Where n! = n f actorial = n(n − 1)(n − 2)(n − 3)....3.2.1. and r = numbers of objects

selected at a time and n = total number of objects.

Example
10 horses compete in a race.

(i) How many distinct arrangements are there of the first 3 horses past the post?

(ii) What is the probability of predicting the order of the first 3 horses past the post?

Answer

(i) Since the order of 3 horses is important, it is appropriate to use the permutation
formula.
10!
That is: Prn = P310 = (10−3)! = 720
There are 720 distinct ways of selecting 3 horses out of 10 horses.

(ii) The probability of selecting the first 3 horses past the post is:

1 1
P (f irst 3 horses) = Selecting 3 out of 10 horses = 730 chance of winning.

5.11.3. Combinations
A combination is the number of different ways of arranging a subset of objects selected
from a group of objects where the ordering is not important. Each possible arrange-
ment is called a combination. The number of ways of arranging r objects selected from
42 Basic Probability

n objects, not considering order, is given by the formula:

n!
Crn = (5.4)
(n − r)!r!

Where n! = n f actorial = n(n − 1)(n − 2)(n − 3)...3.2.1, r! = r(r − 1)(r − 2)(r − 3)...3.2.1,
r = number of objects selected and n = total number of objects.

Example
10 horses complete in a race.

(i) How many arrangements are there of the first 3 horses past the post, not consid-
ering the order in which the first three pass the post?

(ii) What is the probability of predicting the first 3 horses past the post, in any order?

Answer

(i) The order of the first 3 horses is not important, hence apply the combination for-
mula.
n! 10!
Crn = (n−r)!r! = (10−3)!7! = 120
There are 120 different ways of selecting the first 3 horses out of 10 horses, with-
out regard to order.

(ii) The probability of selecting the first 3 horses past the post, disregarding order is
given by
1
P (f irst 3 horses) = Selecting1 3 horses = 120 chance of winning.

5.12. Exercise
1 . Find the values of:

(a) P47
(b) C28

2 . There are 5 levels of shelving in a supermarket. If 3 brands of soup must each

be placed on a separate shelf, how many different ways can a packer arrange the
soup brands?

3 . In an examination a student is asked to answer three questions from an ex-

amination paper containing eight questions. How many deferent selections are
possible?

4 . A company has 12 products in its product range. It wishes to advertise in the

local newspaper, but due to space constraints, it is only allowed to display 7 of
Basic Probability 43

its products at a time. How many different ways can this company compose a
display in the local newspaper?
44 Basic Probability
Chapter 6

Probability Distributions

6.1. Introduction

This unit will study probability distributions. A probability distribution gives the en-
tire range of values that can occur based on an experiment. A probability distribution
is similar to a relative frequency distribution. However in steady of describing the
past, it describes a likely future event. For instance a drug manufacturer may claim
a treatment will cause weight loss for 80% of the population. A consumer protection
agency may test the treatment on a sample of six people. If the manufacturers claim
is true, it is almost impossible to have an outcome where no one in the sample loses
weight and its most likely that 5 out of the 6 do lose weight.

6.2. Definition

A probability distribution or probability mass function is a listing of all the possible

outcomes of an experiment and the probability of each of these outcomes.

6.3. Random variables

A random variable is a function whose value is a real number determined by each ele-
ment in the sample space. In other words its a quantity resulting from an experiment
that, by chance, can assume different values. There are two types of random variables,
discrete random variables and continuous random variables.

Discrete Random Variable is a variable that can assume a countable number of

values. Examples include:

• Number of defective light bulbs obtained when three light bulbs are selected at
random from a consignment could be 0, 1, 2, or 3.

• The number of employees absent from the day shift on Monday.

46 Probability Distributions

• The daily number of accidents that occur in the city of Harare.

Continuous Random Variable is a variable that can assume values corresponding

to any of the points contained in one or more intervals. Examples include:

• The waiting time for customers to receive their order at a manufacturing com-
pany.

• Tire pressure measure in kilo Pascal (KPa) of an automobile.

• The height of each student is this class.

The choice of a particular probability distribution function depends primarily on the

nature of the random variable (i.e. discrete or continuous) under study. Thus we have
Discrete Probability distributions and Continuous Probability distributions.

6.4. Discrete probability distribution

The probability distribution or probability mass function of a discrete random variable

is a graph, table or formula that specifies the probability associated with each possible
value the random variable can assume.

Example
Find the probability mass function of the numbers when a pair of dice is thrown.

Answer
Let X be a random variable whose values of x are the possible totals of the outcomes
of the two dies. Then x can be an integer from 2 to 12. Two dice can fall in 6 × 6 ways
1 2
each with a probability 36 . For example, P (X = 3) = 36 since a total of 3 can occur in
two ways, that is (1,2) or (2, 1). The probability distribution (mass) function is:

x 2 3 4 5 6 7 8 9 10 11 12
1 2 3 4 5 6 5 4 3 2 1
P(X= x) 36 36 36 36 36 36 36 36 36 36 36

Exercise 1

1 Three coins are tossed. Let X be the number of heads obtained. Find the function
of X.

2 Suppose you are interested in the number of Tails showing face up of coin. What
is the probability distribution for the number of Tails?
Probability Distributions 47

6.5. Properties of discrete probability mass function

Let X be a discrete random variable that can assume values x1 , x2 , x3 , . . . , xn . Then

P (X = x) is the probability mass function if:

i. 0 ≤ P (X = x) ≤ 1 for all values of x.

n
X
ii. P (X = x1 ) + P (X = x2 ) + . . . + P (X = xn ) = =1
i=1

This means that probability of any value of x is never negative and also the sum
of the probabilities of the discrete random variable should be equal to 1.

iii. The mean, or the expected value of a discrete random variable X is µ = and given
by:
X
E(X) = xi P (X = xi )
all x

iv. Variance of a discrete random variable X is

V ar(X) = σ 2 = E(X − µ)2

X
σ2 = (xi − µ)2 P (X = xi ) (6.1)
all x

Note that: σ 2 = E(X 2 ) − E(X)2 = E(X 2 ) − µ2

Example
Consider the following probability distribution for a discrete random variable. Verify

x 0 1 2 5 10
P (X = xi 0.05 0.25 0.30 0.20 0.20

the probability properties and find the standard deviation of the distribution.

Solution

i. All P (X = xi ) are between 0 and 1, for i = 0, 1, 2, 5, 10.

ii. Sum of probabilities should be equal to 1.

X
P (X = xi ) = P (X = x1 ) + P (X = x2 ) + . . . + P (X = x5 )
= 0.05 + 0.25 + 0.30 + 0.20 + 0.20
= 1
48 Probability Distributions

iii. The mean is given by:

X
µ = xi P (X = xi )
= 0 × 0.05 + 1 × 0.25 + 2 × 0.30 + 5 × 0.20 + 10 × 0.20
= 3.85

iv. The variance is given by

n
X
V ar(X) = E(X − µ)2 = (xi − µ)2 P (X = xi )
i=1
= (0 − 3.85) × 0.05 + (1 − 3.85)2 × 0.25 + . . . + (10 − 3.85)2 × 0.20.
2

= 11.6275

p √
v. Standard deviation, σ = V ar(X) = 11.6275 = 3.410

6.6. Probability terminology and notation

a) At most 3, means not more than 3. Here 3 is an arbitrary number, it therefore
means 3 is the maximum discrete value which can be assumed by a random
variable. Let X be the random variable, taking x = 0, 1, 2, 3, . . . , n, where n is the
sample size. Notation: P (X ≤ 3) = P (X = 0) + P (X = 1) + P (X = 2) + P (X = 3).

b) At least 3, means not less than 3. The minimum that can be assumed is 3 since
3 is not less than itself. Notation: P (X ≥ 3) = P (X = 3) + P (X = 4) + P (X =
5) + . . . + P (X = n)

c) Less than 3 this effectively means values below 3, and 3 is actually excluded.
Notation: P (X < 3) = P (X = 0) + P (X = 1) + P (X = 2)

d) More than 3, means values above 3, in discrete terms it is from 4 upwards. No-
tation: P (X > 3) = P (X = 4) + P (X = 5) + P (X = 6) + . . . + P (X = n) or using
the complimentary rule it is given as 1 − P (X ≤ 3)

e) Exactly 3, it means equals. Notation: P (X = 3).

f) Between 3 and 6 means the discrete values between 3 and 6, which are 4 and 5.
However it should be noted that the limits can be exclusive or inclusive. Notation
for exclusive: P (3 < X < 6) = P (X = 4) + P (X = 5). Notation for inclusive:
P (3 ≤ X ≤ 3) = P (X = 3) + P (X = 4) + P (X = 5) + P (X = 6)

Exercise 2
Consider the following probability distribution that characterises a marketing ana-
lysts belief concerning the probabilities associated with the number, x of sales that a
company might expect per month for a new super computer:
Probability Distributions 49

x 0 1 2 3 4 5 6 7 8
P(X=x) 0.02 0.08 0.15 0.19 0.24 0.17 0.10 0.04 0.01

i. What is the probability that the company will sell as in (a), (b), (c), (d), (e) and (f)
above.

ii. Find the mean, variance and standard deviation of X.

6.7. Discrete probability distributions

For the purpose of this course we will focus on three special and commonly used dis-
crete probability distributions that are Bernoulli, Binomial and Poisson distributions.

6.7.1. Bernoulli distribution

In probability theory and statistics, the Bernoulli distribution, named after Swiss sci-
entist Jacob Bernoulli, is the probability distribution of a random variable which takes
value 1 with success probability and value 0 with failure probability. A random vari-
able X which has two possible outcomes, say 0 and 1, is called a Bernoulli random
variable. The probability distribution of X is:

P (X = 1) = p

P (X = 0) = 1 − p

i.e. P (X = 0) = 1 − P (X = 1) = 1 − p

This distribution best describes all situations where a ”trial” is made resulting in ei-
ther ”success” or ”failure,” such as when tossing a coin, or when modeling the success
or failure of a surgical procedure. The Bernoulli distribution is defined as:

f (X = x) = px (1 − p)1−x , for x = 0, 1 (6.2)

Where, p is the probability that a particular event (e.g. success) will occur.

Example: tossing a fair coin, you get a head or a tail each with probability of 0.5.
Thus, if a head is labelled 1 and a tail 0, and a tail the random variable X representing
the outcome takes values 0 or 1. If the probability that X = 1 is p, then we have that:

1
P (X = 1) = .
2
1 1
P (X = 0) = 1 − = ,
2 2
since events X = 1 and X = 0 are mutually exclusive.
50 Probability Distributions

6.7.2. Binomial distribution

Data often arise in the form of counts or proportions which are realizations of a discrete
random variable. A common situation is to record how many times an event occurs
in n repetitions of an experiment, i.e. for each repetition the event either occurs (a
”success”) or it does not (a ”failure”) occur. More specifically, consider the following
experimental process:

1. There are n trials.

2. Each trial results in a success or a failure.

3. The probability of a success, p is constant from trial to trial.

4. The trials are independent.

An experiment satisfying these four conditions is called a binomial experiment. The

outcome of this type of experiment is the number of successes, i.e., a count. The dis-
crete variable X representing the number of successes is called a binomial random
variable. The possible counts, X = 0, 1, 2, . . . n, and their associated probabilities de-
fine the binomial distribution, denoted by Bin(n, p).

Suppose we repeat a Bernoulli p experiment n times and count the number X of suc-
cesses, the distribution of X is called the Binomial, Bin(n, p) random variable. The
quantities n and p are called parameters and they specify the distribution.

If X = X1 + X2 + . . . + Xn , where Xi are independent and identically distributed

Bernoulli random variables, then X is called a Binomial random variable. Thus the
probability mass function is:

P (X = x) = Cxn px (1 − p)n−x (6.3)

n!
Where k = 0, 1, 2, . . . , n, 0 < p < 1 and Cxn = (n−x)!x!

Mean and variance of the Binomial Distribution

1. Mean, µx = E(X) = np

2. Variance, σx2 = V ar(X) = np(1 − p) = npq

Example:
The notation Bin(n, p) means a Binomial distribution with parameters n and p. Find:
Probability Distributions 51

i) Probability of getting 4 heads in 6 tosses of a fair coin

ii) Mean and

iii) Variance.

Solution:

i) Let X is the number of heads (success) when a fair coin is tossed. Thus:
4
P (X = 4) = C46 12 (1 − 21 )64 = 15
64

1
ii) Mean= np = 6 × 2 =3

1
iii) Variance= npq = 6 × 2 × (1 − 12 ) = 1.5

Exercises:

1. A manufacturer of nails claims that only 3% of its nails are defective. A random
sample of 24 nails is selected, what is the probability that 5 of the nails are
defective?

2. A certain rare blood type can be found in only 0.05% of people. If the population
of a randomly selected group is 3000, what is the probability that at least two
persons in the group have this rare blood type?

6.7.3. Poisson distribution

The Poisson distribution, named after French mathematician Simon Denis Poisson, is
a discrete probability distribution that expresses the probability of a given number of
events occurring in a fixed interval of time and/or space if these events occur with a
known average rate and independently of the time since the last event. The Poisson
distribution can also be used for the number of events in other specified intervals such
as distance, area or volume.

Examples binomial experiments

i) The number of cars arriving at a parking garage in one-hour time interval.

ii) The number of defective screws per consignment.

iii) Number of typing errors per page.

iv) Number of particles of a given chemical in a litre of water.

The Poisson question: What is the probability of r occurrences of a given outcome be-
ing observed in a predetermined time, space or volume interval?
52 Probability Distributions

A Poison random variable is a discrete random variable that can take integer values
from 0 up to infinite (∞). The parameter for this distribution is λ, i.e. P o(λ). The
Poisson probability mass function is given by:

λx e−λ
P (X = x) = (6.4)
x!

Where x = 0, 1, 2, . . . ∞ and 0 < λ < ∞.

Example:
The number of students arriving at a takeaway every 15 minutes is a Poisson random
variable with parameter λ = 0.2. Find the probability that zero, at most one and more
than two students arrive at the takeaway.

Solution:
Using the formular that:
λx e−λ
P (X = x) =
x!

i) Probability that no students arrive:

0.20 e−0.2
P (X = 0) =
0!
= 0.8187

ii) Probability that at most one student arrive

P (X ≤ 1) = P (X = 0) + P (X = 1)
0.20 e−0.2 0.21 e−0.2
= +
0! 1!
= 0.9824

iii) Probability that at least two students arrive)

P (X ≥ 2) = P (X = 2) + P (X = 3) + . . .
= 1 − (P (X = 0) + P (X = 1))
= 1 − (0.8187 + 0.1637)
= 0.0176

Properties of Poisson Random variable

Probability Distributions 53

1. Mean = E(X) = λ

2. Variance = V ar(X) = λ

Exercises

1. A textile producer has established that a spinning machine stops randomly due
to thread breakages at an average rate of 5 stoppages per hour. What is the
probability that in a given hour:

i) 3 stoppages will occur on this spinning machine?

ii) at most 2 stoppages will occur on this spinning machine?
iii) more than 4 stoppages will occur on this spinning machine?
iv) between 2 and 6 stoppages will occur on this spinning machine?
v) What is the probability that no more than 1 stoppage will occur in a given
two-hour interval?

2. The arrival of patients at a rural clinic is 2 per hour. In any given hour what is
the probability that:

a) no patient will arrive?

b) exactly six patients will arrive?
c) not less than 2 patients will arrive.
d) Determine the variance.

Remark: As a general rule, always check that the time, space or volume interval over
which occurrences of the random variable are observed is the same as the time, space
or volume interval corresponding to the average rate of occurrences, λ. When they
differ, adjust the rate of occurrences to coincide with the observed interval.

6.8. Continuous probability distributions

We will discuss three continuous random variable distributions that are the Normal,
Uniform and Exponential distributions only.

6.8.1. The Normal distribution

One of the most useful and frequently encountered continuous random variable distri-
butions is called the Normal distribution. Its graph is called the normal curve, which
is bell shaped. The curve describes the distribution of so many sets of data that occur
in nature, industry and research.

Characteristics of the Normal Distribution

54 Probability Distributions

i. It is bell-shaped.

ii. It is symmetrical about a central value µ.

iii. The tails of the distribution never touch the axis (i.e. asymptotic).

iv. A normally distributed random variable if fully described by two parameters

namely µ (the population mean) and σ 2 (the population variance).

v. The area under the curve is equal to 1.

A random variable, X is said to be normally distributed with a probability density

function given by:

1 −1 x−µ 2
f (x) = √ e2( σ ) , −∞ < x < ∞ (6.5)
2πσ

where µ = mean of the random variable X and σ 2 = variance of the random variable X.
The random variable X is represented as: X ≈ N (µ, σ 2 ). µ and σ 2 are said to be the
parameters of X.

It is difficult to use the probability density function of the normal distribution to cal-
culate the probabilities for X. Hence the process of standardisation is used so that the
probability values are taken directly from the standard nornal distribution table. This
table indicates the probabilities corresponding to different values of Z starting at -3.
The process of standardisation involves calculating the value of Z by use of a formular:

X −µ
Z= (6.6)
σ

6.8.2. The standard normal distribution

The standard normal distribution is a special kind of a normal distribution with mean
zero and variance 1. Z is called a standard normal distribution random variable and
written as: Z ≈ N (0, 1). The cumulative distribution function of Z is denoted by Φ(Z),
and is denoted by P (Z < z) = Φ(Z). The values of this are found in the Standard
normal distribution tables.

Use Standard normal distributiion tables to find the probabilities below.

a. P (Z ≥ −2)

b. P (Z > 0.79)

c. P (−1.11 < Z < −0.7)

d. P (−1.3 < Z < 2.1)

Probability Distributions 55

e. P (Z ≤ −3)

f. P (0.04 < Z < 0.46)

Solution

a.

P (Z ≥ −2) = 1 − P (Z ≤ −2)
= 1 − Φ(−2)
= 1 − 0.0228
= 0.9772

b.

P (Z > 0.79) = 1 − P (Z < 0.79)

= 1 − Φ(0.79)
= 1 − 0.7852
= 0.2148

c.

P (−1.11 < Z < −0.7) = Φ(−0.79) − Φ(−1.11)

= 0.2420 − 0.1335
= 0.1085

d.

P (−1.3 < Z < 2.1) = Φ(2.1) − Φ(−1.3)

= 09821 − 0.0968
= 0.8853

e.

P (Z ≤ −3) = Φ(−3)
= 0.0013

f.

P (0.04 < Z < 1.46) = Φ(1.46) − Φ(0.04)

= 0.9278 − 0.5160
56 Probability Distributions

= 0.4118

Finding probabilities using Z-Distribution

The normal random variable, X with a mean µ and variance σ (N (µ, σ)) has a standard
normal distribution
x−µ
z=
σ

6.8.3. The Uniform distribution

This distribution is also known as the rectangular distribution. A continuous uniform

variable has a probability density over an interval. Its probability distribution density
function is:
(
1
f (X = x) = b−a a < x < b (6.7)
0 elsewhere

Properties of the Uniform distribution

i. Mean is given by

1
E(X) = (b + a) (6.8)
2

ii. Variance is given by

(b − a)2
V ar(X) = (6.9)
12

NB
The probability that X falls in some interval say (c,d) can be easily calculated by inte-
grating the density function
1
f (x) =
b−a
to obtain
c−d
P (X = (c, d)) =
b−a
Example:
The marks of students from a certain examination are uniformly distributed in the
interval 50 to 75. The density function for the marks is given by:
(
1
75−50 75 < x < 50
f (X = x) =
0 elsewhere

Find the mean and variance of this distribution.

Solution:
Probability Distributions 57

1. The Mean is given by E(X) = 21 (b + a) = 12 (75 + 50) = 62.5

(b−a)2 (75−50)2
2. The Variance is given by 12 = 12 = 52.083

Interpretation:
The average mark for the examination was 62.5 with a variance of 52.083.

Exercise:
For the continuous uniform distribution defined of the interval [a, b], where b > a,
show that
i) Mean = 12 (b + a) and
2
ii) Variance = (b−a)
12

6.8.4. The Exponential distribution

An exponential distribution variable is a continuous random variable that can take on

any positive value. X is said to be an exponential random variable with a parameter λ
and a probability density function
(
λe−λx x > 0
f (x) = (6.10)
0 otherwise

The exponential distribution function ogten arises in practice as the distribution of

waiting time i.e. the amount of time until a specified event occurs. Examples include
time until a customer arrives, or time until a machine fails.

Example
Suppose that the length of a phone call in minutes is exponentially distributed with
parameter, λ = 0.1. If someone arrives immediately ahead of you at a public telephone
booth, what is the probability that you will wait for at least 20 minutes.

Solution
Let X be the length of a phoe call made in front of you. Then
Z ∞
P (X > 20) = 0.1e−0.1x dx
20
= −e−0.1x |∞
20

= e−2 = 0.2706
58 Probability Distributions
Chapter 7

Interval Estimation

7.1. Introduction
We are now in the knowledge that a population parameter can be estimated from sam-
ple data by calculating the corresponding point estimate. This chapter is motivated by
the desire to understand the goodness of such a point estimate. However, due to sam-
pling variability, it is almost never the case that the population parameter equals the
sample statistic. Further, the point estimate does not provide any information about
its closeness to the true population parameter. Thus, we cannot rely on point estimates
for decision making and policy formulation in day to day living and or in any organisa-
tion, institution or country. We need bounds that represent a range of plausible values
for a population parameter. Such ranges are called interval estimates.

To obtain the interval estimates, the same data from which the point estimate was
obtained is used. Interval estimates may be in the form of a confidence interval whose
purpose is to bound population parameters such as the mean, the proportion, the vari-
ance, and the standard deviation; a tolerance interval which bounds a selected propor-
tion of the population; and a prediction interval which places bounds on one or more
future observations from a population.

In this course, we will focus on confidence intervals.

7.2. Confidence Intervals

It is noted that we cannot be certain that an interval contains the true but unknown
population parameter since only a sample from the full population is used to compute
both the point estimate and, the interval estimate! A confidence interval is constructed
so that there is high confidence that it does contain the true but unknown population
parameter. Generally, a 100(1 − α)% confidence interval equals
60 Interval Estimation

point estimate ± reliability coefficient × s.e.(parameter)

where α is the level of significance between zero and one; 1 − α is a value called
the ”confidence coefficient”; 100(1 − α)% is the confidence level; parameter estimate
is a value for the point estimate such as for the sample mean, x, or for the pop-
ulation proportion, pb; reliability coefficient is a probability point obtained from an
appropriate table as dictated by, for example, zα or t α2 ,n−1 ; and s.e.(parameter) read
standard error of the parameter, measures the closeness of the point estimate to the
true population parameter i.e. it measures the precision of an estimate in getting the
parameter.

7.3. Confidence Interval for the Population Mean

The overall assumption made is that the sample comes from a normal population.

Case 1: Known Population Variance

Suppose that, in addition to the overall assumption, the variance of the pop-
ulation, σ 2 , is known. Then a random variable called the sample mean, X, is
defined such that
σ2
X ∼ N (µ, ),
n
whose standardised result is
√
n(X − µ)
Z= ∼ N (0, 1).
σ
The 100(1 − α)% confidence interval estimate for the population mean may also
take the form `1 ≤ µ ≤ `2 where the end points `1 and `2 are called lower- and
upper-confidence limits respectively and are computed from the sample data.
Different samples will produce different values for the end points.

Also, observe that,

σ
`1 = x − z α2 × √
n
and
σ
`2 = x + z α2 × √
n
Thus, a 100(1 − α)% confidence interval for the population mean is
σ σ
x − z α2 × √ ≤ µ ≤ x + z α2 × √ .
n n

Example
Interval Estimation 61

Consider the following data.

64.3 64.6 64.8 64.2 64.5 64.3 64.6 64.8 64.2 64.3

Assume that it is normally distributed with unit population variance. For these
data, construct a 95% confidence interval for the population mean.

Solution
Using the data, n = 10, x = 64.46, the level of significance, α = 5% = 0.05, and
from the given assumption, σ 2 = 1. Now, the resulting 95% confidence interval
(CI) for the population mean is
σ σ
x − z0.025 × √ ≤ µ ≤ x + z0.025 × √
n n

Substituting we have

1 1
64.46 − 1.96 × √ ≤ µ ≤ 64.46 + 1.96 × √ .
10 10

Simplifying we then have the 95% CI for the population mean as

63.84 ≤ µ ≤ 65.08.

Interpretation
From the above estimation of the confidence interval for the population mean,
it is tempting to conclude that the population mean, µ, is within 63.84 ≤ µ ≤
65.08 with a probability of 0.95. To be blunt, this statement is not true.

Well, the true value of the population mean, µ, is unknown (!), and the con-
fidence interval is a random interval that is a function of the sample mean (!).
In this scenario, to say that µ is within 63.84 ≤ µ ≤ 65.08 with a probability
0.95 is totally off the mark. Now, if you follow the argument, the statement
63.84 ≤ µ ≤ 65.08 is either correct with probability 1 or incorrect with proba-
bility 1. By and large, the correct interpretation of a 100(1 − α)% confidence
interval for a population parameter is that:

if a very large number of random samples are collected and a 100(1 − α)% confi-
dence interval for the population parameter is computed from each sample, then
100(1 − α) percent of these intervals will contain the true value of the population
parameter with confidence 100(1 − α).
62 Interval Estimation

In terms that are loose and respecting the frequency approach to probability,
for our case, this general interpretation says

we don’t know if the statement 63.84 ≤ µ ≤ 65.08 is true for this specific sam-
ple, but that, in repeated sampling, the method used to obtain the confidence
interval for µ yields correct statements 95% of the times with a 0.95 confidence.

To illustrate this interpretation from the other end, let us consider a thousand
samples taken, and their 100(1 − α)% confidence intervals for a specified pop-
ulation parameter constructed. Conceivably, at least fifty of these confidence
intervals will fail to contain the true value of the population parameter. Sup-
pose that these were 95% confidence intervals. Then, only 5% of the confidence
intervals would fail to contain the true value of the population parameter. This
says that 950 of the times would contain the true population parameter of in-
terest with a 0.95 confidence.

We point out that in practice we obtain only one random sample and calculate
only one confidence interval. From the preceding standpoint, this confidence
interval either contains or does not contain the true population parameter. In
the end, one MUST therefore reject the obvious temptation!

Lecture Exercise
For the above example, separately construct a 90% and 99% confidence interval
for the population mean.

Further, the length or width of a confidence interval is given by `2 − `1 . Now,

how long are the resulting confidence intervals? Of the three confidence inter-
vals constructed so far, which one is the most precise?

Task: Starting from the cases considered above, what is the general relation-
ship between confidence levels and their precision?
Interval Estimation 63

Remark:
The precision of a confidence interval is inversely proportional to the confidence
level. It is desirable to obtain a confidence interval that is short enough for
purposes of decision making and that also has adequate confidence. This is
easily the reason why the 95% confidence level is the default confidence level
chosen by researchers and practitioners.

7.4. One-Sided Confidence Intervals for the Population Mean

Using similar assumptions, one-sided confidence limits for the population mean,
µ, are obtained by setting either `1 to -∞ or `2 to ∞ and replacing z α2 by zα .

We therefore have a 100(1 − α)% upper confidence limit for µ given by

σ
µ ≤ x + zα × √ ,
n

and a 100(1 − α)% lower confidence limit for µ given by

σ
x − zα × √ ≤ µ.
n

Lecture Exercise
For the data in the above example, construct the 90%, 95%, 99% lower-, and
upper- confidence limits. What observations can you make?

Case 2: Unknown Population Variance

(a): Large Samples (n > 30)

It was assumed in the foregoing discussion that the population distribution is

normal with an unknown µ and a known standard deviation σ. However, these
assumptions may be dropped when dealing with large-samples.

Let the observations X1 , X2 , ..., Xn be a random sample from a population with

unknown mean, µ and an unknown variance, σ 2 . If n is large, then

σ2
X ∼ N (µ, )
n
64 Interval Estimation

and it follows that √

n(X − µ)
Z= ∼ N (0, 1).
σ
In this case n is large and so it is permissible to replace the unknown σ by s.
This has close to no effect on the distribution of Z.

For large n, the quantity √

n(X − µ)
s
follows a standard normal distribution with mean zero and unit standard de-
viation.

Then the 100(1 − α)% confidence interval for µ is

s s
x − z α2 × √ ≤ µ ≤ x + z α2 × √
n n

which is true regardless of the sample’s underlying distribution.

Example
A study was carried out in Zimbabwe to investigate pollutant contamination
in small fish. A sample of small fish was selected from 53 rivers across the
country and the pollutant concentration in the muscle tissue was measured
(ppm). The pollutant concentration values are shown below. Construct a 95%
confidence interval for the population mean, µ.

1.230 1.330 0.040 0.044 1.200 0.270 0.490 0.190 0.940 0.520 0.830
0.810 0.710 0.500 0.490 1.160 0.050 0.150 0.400 0.190 0.650 0.770
1.080 0.980 0.630 0.560 0.410 0.730 0.430 0.590 0.340 0.340 0.270
0.840 0.500 0.340 0.280 0.340 0.250 0.750 0.870 0.560 0.100 0.170
0.180 0.190 0.040 0.490 0.270 1.100 0.160 0.210 0.860

Solution
Since n > 30, then the 95% confidence interval for µ is

0.3486 0.3486
0.5250 − 1.96 × √ ≤ µ ≤ 0.5250 + 1.96 × √
53 53

which simplifies to
0.431 ≤ µ ≤ 0.619

Lecture Exercise
Interval Estimation 65

Construct the 90% and the 99% CI for µ using the above data. Further, using
the above data construct the 90%, 95%, and the 99% lower- and upper- CI for
the population mean.

(b): Small Samples ( n ≤ 30)

It is now necessary to introduce a new confidence interval construction proce-

dure that addresses the scenario of small samples. In many cases, it is reason-
able to assume that the underlying distribution is normal and that moderate
departure from normality will have little effect on validity of the result.

Remark: In the equally likely event that the assumption is unreasonable, an

alternative (not discussed here) is to use the non-parametric procedures which
are valid regardless of underlying populations.

For our purposes, it will be reasonable to assume that the population of interest
is normal with an unknown mean, µ, and an unknown variance, σ 2 . A small
random sample of size n is drawn. Let X and S2 be the sample mean and
sample variance, respectively. We wish to construct a two-sided confidence
interval on µ . The population variance, σ 2 , is unknown and it is a reasonable
procedure to use S2 to estimate σ 2 . Then the random variable Z is replaced
with T which is given by √
n(X − µ)
T =
s
which is a random variable that follows the student’s t distribution with n − 1
degrees of freedom which are associated with the estimated standard devia-
tion.

Notation
We let tα ,n−1 and t α2 ,n−1 be the value of the random variable T with n − 1 de-
grees of freedom above which we find a probability α or α2 respectively.

The 100(1 − α)% CI for µ is given by

s s
x − t α2 ,n−1 × √ ≤ µ ≤ x + t α2 ,n−1 × √
n n
66 Interval Estimation

where t α2 ,n−1 is the upper 100 α2 percentage point of the t- distribution with n − 1
degrees of freedom.

Example
Consider the following data obtained from a local Transport Logistics company.

19.8 10.1 14.9 7.5 15.4 15.4 15.4 18.5 7.9 12.7 11.9
11.4 11.4 14.1 17.6 16.7 15.8 19.5 8.8 13.6 11.9 11.4

Construct a 95% confidence interval for the population mean, µ.

Solution
Since our sample is small, n = 22, then the 95% confidence interval for the
population mean is given by
s s
x − t α2 ,n−1 × √ ≤ µ ≤ x + t α2 ,n−1 × √
n n

Substituting yields

3.55 3.55
13.71 − 2.080 × √ ≤ µ ≤ 13.71 + 2.080 × √
22 22

and simplifying we have

12.1 ≤ µ ≤ 15.3

as the 95% confidence interval for µ.

Lecture Exercise
For the above data,construct the 90% and the 99% confidence intervals on the
population mean and interpret the two confidence intervals. Further, construct
the 90%, the 95% and the 99% lower - and upper - confidence limits. Give an
interpretation of each and all of them.

Remark: One-sided confidence intervals for the mean of a normal population

are constructed by choosing the appropriate lower- or upper-confidence limit
and then replacing t α2 ,n−1 by tα,n−1 .
Interval Estimation 67

7.5. Confidence Interval for the Population Proportion

Suppose that a random sample of size n, large n, has been taken from a large
population and that x but less than n observations in this sample belong to a
class of interest. Then pb calculated as nx is a point estimator of the proportion of
the population p that belongs to this class. It is noted that n and p are the pa-
rameters of a binomial distribution (refer to earlier discussions). The sampling
distribution of pb is approximately normal with mean p and variance p(1−p) n
if p
is not too close to either 0 or 1 and if n is relatively large. To apply this, it is
required that np and n(1-p) be greater than or equal to 5. We are saying that:
If n is large, then the distribution of

pb − p
Z=q ∼ N (0, 1).
p(1−p)
n

For large samples, which usually is the case when dealing with proportions, a
satisfactory 100(1 − α)% confidence interval on the population proportion p is
r r
pb(1 − pb) pb(1 − pb)
pb − z α2 × ≤ p ≤ pb + z α2 ×
n n
α
where pb is the point estimate of p, and z α2 is the upper 2
probability point of the
standard normal distribution.

Example
In a random sample of 85 stone sculptures, 10 have a surface finish that is
rougher than the expected. Construct a 95% confidence interval for the popu-
lation proportion of stone sculptures with a surface finish that is rougher than
the expected.

Solution
A 95% two-sided confidence interval for p is
r r
0.12(1 − 0.12) 0.12(1 − 0.12)
0.12 − 1.96 × ≤ p ≤ 0.12 + 1.96 ×
85 85
which simplifies to
0.05 ≤ p ≤ 0.19

Remark: The one-sided lower - and upper - confidence intervals are respec-
68 Interval Estimation

tively given as r
pb(1 − pb)
pb − zα × ≤ p
n
and r
pb(1 − pb)
p ≤ pb + zα × .
n
Lecture Exercise
In the above example, construct and interpret the 95% and the 99% lower - and
upper - confidence limits for the population proportion.

7.6. Confidence Interval for the Population Variance

Let X1 , X2 , ..., Xn be a random sample from a normal distribution with mean µ

and variance σ 2 , and let S2 be the sample variance. Then the random variable

(n − 1)S 2
V =
σ2
has a chi-square (χ2 ) distribution with n − 1 degrees of freedom.

Now, if s2 is the sample variance from a random sample of n observations from

a normal distribution with unknown variance, σ 2 , then a 100(1−α)% confidence
interval on σ 2 is
(n − 1)S 2 2 (n − 1)S 2
≤ σ ≤
χ2α ,n−1 χ21− α ,n−1
2 2

where χ2α ,n−1 and χ21− α ,n−1 are the upper and lower 100 α2 percentage points of
2 2
the χ2 distribution with n − 1 degrees of freedom, respectively.

Illustration
An Entrepreneur has got an automatic filling machine that she uses to fill
bottles with liquid detergent. A random sample of 20 bottles results in a sample
variance of fill volume of s2 equal to 0.0153 (f luid ounces)2 . Assume that the fill
volume is normally distributed. Then a 95% upper- CI is

(n − 1)S 2
σ2 ≤
χ21−α,n−1

substituting yields
(20 − 1) × 0.0153
σ2 ≤
χ21−0.05,20−1
Interval Estimation 69

simplifying we have
19 × 0.0153
σ2 ≤
χ20.95,19

where χ20.95,19 is 10.117.

so we get
19 × 0.0153
σ2 ≤
10.117
giving
σ 2 ≤ 0.0287

NB: The statistical tables round off 10.117 to 3 s.f.

7.7. Confidence Interval for the Population Standard De-

viation

The one-sided lower and upper- confidence intervals for σ 2 are

(n − 1)S 2
≤ σ2
χ2α,n−1

and
(n − 1)S 2
σ2 ≤
χ21−α,n−1

Remark: Clearly, the lower- and upper- confidence intervals/ limits for σ are
the square roots of the corresponding limits in the above equations.

We state that σ 2 ≤ 0.0287, is converted into an upper - confidence interval/ limit

for the population standard deviation σ by taking the square root of both sides.
The resulting 95% confidence interval is σ ≤ 0.17.

Lecture Exercise
Using the information from the above illustration, construct a 90% lower- and
upper- confidence limits for the population standard deviation, σ.
70 Interval Estimation

7.8. Confidence Interval for the Difference of Two Popu-

lations Means

The overall assumption remains in place. And, the same with everything else.
We are simply considering two populations and constructing confidence inter-
vals for the difference in two population means, µ1 − µ2 .

7.8.1. Case 1: Known Population Variance

Illustrative Example
An entrepreneur is interested in reducing the drying time of a wall paint. Two
formulations of the paint are tested; formulation 1 is the standard, and formu-
lation 2 has a new drying ingredient that should reduce the drying time. From
experience, it is known that the standard deviation of drying time is 8 min-
utes, and this inherent variability should be unaffected by the addition of the
new ingredient. Ten specimens are painted with formulation 1, and another
10 specimens are painted with formulation 2; the 20 specimens are painted in
random order. The two sample mean drying times are 121 minutes and 112
minutes, respectively. Construct a 99% confidence interval for the difference in
the two population means.

Solution
To be provided in the lecture.

7.8.2. Case 2: Unknown (but assumed Equal) Population Variances

This is called the Homogeneous Variance Assumption.

Illustrative Example
The following data is from two populations, A and B. Ten samples from A had
a mean of 90.0 with a sample standard deviation of s1 = 5.0, while 15 sam-
ples from B had a mean of 87.0 with a sample standard deviation of s2 = 4.0.
Assume that the populations, A and B are normally distributed and that both
normal populations have the same standard deviation. Construct a 95% confi-
dence interval on the difference in the two population means.
Interval Estimation 71

Solution
To be provided in the lecture.
72 Interval Estimation
Chapter 8

Hypothesis Testing

8.1. Important Definitions, and Critical Clarifications

Hypotheses
A hypothesis is a statement about a population. Testing of hypotheses eval-
uates two hypotheses called the null and the alternative denoted H0 and H1
respectively. An H0 is the assertion that a population parameter takes on a
particular value. On the other hand, an H1 expresses the way in which the
value of a population parameter may deviate from that specified under H0 . The
direction of deviation may be specified (one - sided/ tailed tests) or may not be
specified(two - sided/ tailed tests).

We take time to point out that the language and grammar of testing of hy-
potheses does not use the word ”accept” or any of its numerous synonyms. This
is beyond semantics. To say one ”accepts” the null hypothesis is to imply that
they have proved the null hypothesis to be true. This practice is incorrect. The
null hypothesis is the claim that is usually set up with the expectation of re-
jecting it. The null hypothesis is assumed true until proven otherwise. If the
weight of evidence points to the belief that the null hypothesis is unlikely with
high probability, then there exists a statistical basis upon which we may reject
the null hypothesis. The design of hypotheses tests is such that they are with
the null hypothesis until there is enough evidence that suggest support for the
alternative hypothesis. Clearly, the design is never about selecting the more
likely of the two hypotheses. Let’s take this to our legal system. One is consid-
ered not guilty until proven otherwise. It is the job of the prosecutor to build
a case i.e. put evidence before the court of law that the person in question is
guilty. The jury or the judge will give their verdict as guilty or not guilty but
will NEVER give their verdict with an import of being innocent. By and large,
the courts of law are a classical example of constant testing of hypotheses pro-
cedure. So, let it be clear that on the basis of the data from the sample, we
74 Hypothesis Testing

either reject the null hypothesis or fail to reject the null hypothesis.

In the words of R. A. Fisher:

In relation to any experiment we may speak of ... the ”null hypothesis,” and it
should be noted that the null hypothesis is never proved or established, but is
possibly disproved, in the course of experimentation. Every experiment may be
said to exist only in order to give the facts a chance of disproving the null hy-
pothesis.

Remark 1: The H0 reflects the position of no change and will always be worded
as an equality.

Remark 2: The language which implies ”acceptance” of the null hypothesis

is both misleading and against the grammar of the testing of hypotheses.

Test Statistic
This is a value calculated from sample data and is used to decide on rejecting
H0 .

Critical Region
This is a range of values which is such that when the test statistic falls into it
then H0 would be rejected.

Critical Value
Is a value that separates the rejection region and the non-rejection region.

Type I error
Occurs when a true null hypothesis is rejected. A null hypothesis is rejected
when in actual fact it is true.

Type II error
It occurs when a false null hypothesis is not rejected. Alternatively, it is when
a null hypothesis is not rejected when in actual fact it is false.

Level of significance of a Test

Is the probability of making a type I error expressed as a percentage. It is de-
noted by α.
Hypothesis Testing 75

Power of a Statistical Test

It is the probability that the testing of hypotheses procedure rejects the null
hypothesis when the null hypotheses is indeed false.

8.2. General Procedure on Hypotheses Testing

The following steps are recommended in applying the testing of hypotheses

procedure.

• From the problem context, identify the parameter of interest.

• Clearly state the hypotheses i.e. H0 and H1 .

• Identify or choose the level of significance, α.

• Determine an appropriate test statistic.

• Obtain the critical value from appropriate tables.

• Compute the test statistic by substituting necessary statistics into an ap-

propriate equation.

• Decide on the basis of a decision criterion that rejects H0 if, upon compar-
ison, the test statistic is more extreme than a critical value.

• Conclude on the basis of the decision’s import, and report in the context of
the problem.

8.3. Hypothesis Testing Concerning the Population Mean

8.3.1. Case 1: Known Population Variance

The overall normality assumption is made. In this case, we further assume

that the population variance, σ 2 , is known. The sample mean X which is a point
estimator of µ is a random variable with population mean µ and population
2
variance σn . The test statistic is
√
n(x − µ0 )
Zcal = ∼ N (0, 1)
σ
Exercise
Consider the following data where the population mean is claimed to be 50. σ
= 2, α = 0.05, n = 25, and x = 51.3 What conclusions should be drawn about the
76 Hypothesis Testing

claim?

8.3.2. Guidelines to the Expected Solution

It is given that the population is normal and the population standard deviation
is known. Z− score is the test statistic. The testing of hypothesis procedure is
two sided. At the 0.05 level of significance and based on the sample evidence,
we conclude that the population mean is different from 50.

Exercise
For the above exercise, instead of using the testing of hypothesis procedure,
construct a 95% confidence interval. Test the same hypothesis using the con-
fidence interval. Is the value specified under H0 contained in the confidence
interval? Or, is zero contained in the confidence interval? What conclusions
should be drawn?

Hint on the Decision Criterion

If the value specified under H0 is contained in the confidence interval, then we
fail to reject H0 . And, if zero is contained in the confidence interval, then we
fail to reject H0 .

8.3.3. Case 2: Unknown Population Variance

(a) Large Samples Scenario (n > 30)

The Z- score. We find the critical value using the level of significance specified
or 0.05 if not specified on the basis of the test being a single-sided or a two
sided.

Example
Let the mean cost of an Introduction to Statistics textbook be µ. In testing the
claim that the population mean is not USD34.50 a sample of 36 current text-
books had selling costs with a sample mean USD32.00 and a sample standard
deviation of USD6.30. Using a 10% level of significance, what conclusion can
be made?
Hypothesis Testing 77

Solution
A two-tailed test, n > 30, α = 0.1 and, thus, the critical value is ±1.96. Detailed
solution in the lecture.

(b) Small Sample Scenario (n ≤ 30)

We consider now the case of hypothesis testing on the mean of a population
with an unknown variance, σ 2 . The test statistic is
√
n(x − µ0 )
tcal =
s
which follows a t - distribution with n − 1 degrees of freedom.

Exercise
The increased availability of light materials with high strength has revolution-
ized the design and manufacture of golf clubs, particularly drivers. Clubs with
hollow heads and very thin faces can result in much longer tee shots, especially
for players of modest skills. This is due partly to the spring-like effect that the
thin face imparts to the ball. Firing a golf ball at the head of the club and mea-
suring the ratio of the outgoing velocity of the ball to the incoming velocity can
quantify this spring-like effect. The ratio of velocities is called the coefficient
of restitution of the club. An experiment was performed in which 15 drivers
produced by a particular club maker were selected at random and their coeffi-
cients of restitution measured. In the experiment the golf balls were fired from
an air cannon so that the incoming velocity and spin rate of the ball could be
precisely controlled. The sample mean and sample standard deviation are x =
0.83725 and s = 0.02456. Determine if there is evidence at the α = 0.05 level to
support the claim that the mean coefficient of restitution exceeds 0.82.

Guidelines to the Expected Solution

The mean is the parameter of interest. The population standard deviation is
unknown and the sample size is small. Therefore, the appropriate test statistic
to be used is the t - statistic and the corresponding critical value is tcrit = 1.76.

Exercise
For the above exercise, instead of using the testing of hypothesis procedure,
construct a 95% confidence interval. Test the same hypothesis using the confi-
78 Hypothesis Testing

dence interval approach.

8.4. Hypothesis Testing concerning the Population Pro-

portion

Illustrative Example
The advertised claim for batteries for cell phones is set at 48 operating hours,
with proper charging procedures. A study of 5000 batteries is carried out and
15 stop operating prior to 48 hours. Do these experimental results support the
claim that less than 0.2 percent of the company’s batteries will fail during the
advertised time period, with proper charging procedures? Use a hypothesis
testing procedure with α = 0.01. Is the conclusion the same at the 10% level of
significance?

Solution
15
We are testing H0 : p = 0.002 against H1 : p < 0.002 with pb = 5000
= 0.003.

NB: By claiming a value of the population proportion to be 0.002 implies that

the population proportion. Hence letting p0 to be the hypothesised population
proportion value yields

pb − p0
Zcal = q = 1.5827223454 < Zcrit = Z0.01 = 2.3263
p0 (1−p0 )
n

We fail to reject H0 and conclude that, at the 1% level of significance, there

is no enough evidence to suggest that less than 0.2 percent of the company’s
batteries will fail during the advertised time period, with proper charging pro-
cedures.

Exercise
Let p be the proportion of new car loans having a 48 months period. In some
year p = 0.74. Suppose it is believed that this has declined and accordingly we
wish to test this belief using a 1% level of significance. What is the conclusion
if 350 of a sample of 500 new car loans have a time period of 48 months?
Hypothesis Testing 79

8.5. Comparison of Two Populations

We now extend the previous one population results to the difference of means
for two populations.

8.5.1. Hypothesis Testing concerning the Difference of Two Popula-

tion Means

Case 1: Known Population Variances

Example Consider the following gasoline mileages of two makes of light trucks.
The trucks 1 and 2 have the population means and populations standard devi-
ations 28 and 6, and 24 and 9 respectively. If 35 of truck 1 and 40 of truck 2
are tested, test the claim that the mean difference is 4.

Solution
Exercise in the lecture.

Remark: In inferential applications the population variances σ12 and σ22 are
generally not known and must be estimated by s21 and s22 . The standard error is
estimated by s
s21 s2
s.e. = + 2
n1 n2
Case 2: Unknown Population Variance, and Small Sample (n1 + n2 ≤ 31)

We assume that the variances of both distributions σ12 and σ22 are unknown but
equal. This common variance is estimated by a quantity called pooled variance
denoted s2p and calculated as

(n1 − 1)s21 + (n2 − 1)s22

s2p =
n1 + n2 − 2

Thus, the test statistic is given by

(x1 − x2 ) − (µ1 − µ2 )
tcal = q
s2p [ n11 + n12 ]
80 Hypothesis Testing

which follows a t- distribution with n1 + n2 − 2 degrees of freedom.

Example
Consider the following data. n1 = 10, x1 = 90, s1 = 5, n2 = 15, x2 = 87 and
s2 = 4. Assume that the populations are normally distributed and that both
populations have the same standard deviation. At the 5% level of significance,
can we conclude that there is a difference in the two population means?

Solution
Left as an exercise for the lecture.

8.6. Independent Samples and Dependent/ Paired Sam-

ples

In testing for the equality of two population means, we may choose to select
two random samples one from each population and compare their means. If
these sample means exhibit a difference, then we reject the null hypothesis
that H0 : µ1 − µ2 = 0. Another approach is to try and match the subjects from
the two populations according to variables which will be expected to have an
influence on the variable under study. The two samples are no longer indepen-
dent and the inferences are now based on the differences of the observations
from the matched pairs.

Case 3: Independent Samples An illustrative example will be vital in exposing

the testing of hypothesis procedure.

Example
Samples of two brands of pork sausage are tested for their fat content. The
results of the percentage of fat are summarised as follows: Brand A (n=50,
x=26.0, s=9.0) and Brand B (n=46, x=29.3, s=8.0). Can we conclude that there
is sufficient evidence to suggest that there is a difference in the fat contented
of the two brands of pork sausage? Use a 5% level of significance.
Hypothesis Testing 81

Solution
Left as an exercise for the lecture.

Case 4: Dependent/ Paired Samples

Given two paired samples X11 , X12 , · · · , X1n1 and X21 , X22 , · · · , X2n2 we form a sin-
gle sample of the differences d1 , d2 , · · · , dn where d1 = X11 − X21 , d2 = X12 − X22 , · · · , dn = X1n1 − X

For the new single sample, we find its mean, d, that estimates the population
mean for the differences, µd and standard deviation, sd . Assuming that the
original populations are normally distributed with equal means i.e. µ1 = µ2
and equal variances, the population mean for the differences µd is zero and a
standard error that is estimated by √sdn .

d
The test statistic in this case is tcal = s.e.

The hypotheses tests concerning µ1 and µ2 are now based on the sample mean
using the single sample and we have a modified null hypothesis, H0 : µd = 0
against an appropriate alternative hypothesis as instructed by the situation.

Example
Five automachines are tested for wind resistance with two types of grills. Their
drag coefficients were determined and recorded as follows.

Automachine 1 2 3 4 5
Grill A 0.47 0.46 0.40 0.44 0.43
Grill B 0.50 0.45 0.47 0.44 0.48

Using a 5% level of significance test for the difference in the drag coefficients
due to type of grill.

Solution
Left as an exercise during the lecture.

8.6.1. Advantages of Paired Comparisons

• By pairing, we remove the additional source of variation (i.e. the differ-

ence between experimental units) and hence reduce the random variation
82 Hypothesis Testing

as measured by s2p in the case without pairing and s2d in the case of pairing.
Therefore, s2d > s2p implies a gain in precision due to pairing.

• The confidence interval based on the paired comparison is much narrower

than that from two sample analysis using unpaired observations. This
also implies a gain in precision due to pairing.

• It may be less expensive since in most cases fewer experimental units are
used when compared to a two sample design.

8.6.2. Disadvantages of Paired Comparisons

• There is a substantial loss in the degrees of freedom in a paired compari-

son than in a two sample t - test.

• A rest period may be required between applying the first and second treat-
ment in order to minimise the carry over effect from the first treatment.
Even, then the carry over effect may not be completely eliminated.

8.7. Test Procedure concerning the Difference of two Pop-

ulation Proportions

Suppose that two independent random samples of sizes n1 and n2 are taken
from two populations, and let x1 and x2 represent the number of observations
that belong to the class of interest in sample 1 and sample 2, respectively. In
testing the hypotheses

H0 : p1 − p2 = 0
H1 : p1 − p2 6= 0,

the test statistic is

(pb1 − pb2 ) − (p1 − p2 )
Zcal = q
p1 (1−p1 )
n1
+ p2 (1−p
n2
2)

which is approximately standard normal.

If H0 : p1 − p2 = 0 is true, then p1 = p2 . Thus p1 = p2 = p such that the

test statistic becomes
(pb1 − pb2 ) − (p1 − p2 )
Zcal = q
p(1 − p)[ n11 + n12 ]
Hypothesis Testing 83

which still is approximately standard normal. The common population propor-

tion, p, is estimated by
x1 + x2
pb = .
n1 + n2

Assuming that H0 : p1 − p2 = 0 is true, the test statistic is therefore given by

(pb1 − pb2 )
Zcal = q
pb(1 − pb)[ n11 + 1
n2
]

Example
Consider the following situation in which comparison is made of two concept
exposition methods. Method A is the standard and method B is the proposed.
A class of 200 CUMT105 students at the Chinhoyi University of Technology is
used. The students were randomly assigned to two groups of equal size. One
group was exposed to method A, and the other group was exposed to method
B. At the end of the semester, 19 of the students exposed to method B showed
improvement, while 27 of those exposed to method A improved. At the 5% level
of significance, is there sufficient reason to believe that method A is effective in
concept exposition?

Solution
First, we state the hypotheses:

H0 : pA − pB = 0
H1 : pA − pB 6= 0

Then, we extract the given data: nA = nB = 100, pc

A = 0.27, p
c B = 0.19, xA = 27,
and xB = 19. Thus, pb = 0.23.

The test statistic and the critical value are Zcal = 1.35 and Zcrit = 1.96 respec-
tively.

After comparing Zcal and Zcrit , the decision is that we fail to reject H0 . From
this decision, we therefore conclude that, at the 5% level of significance, there
is no sufficient evidence to support the assertion that method A is effective in
concept exposition.
84 Hypothesis Testing

Exercise
A study is made of business support of the immigration enforcement practices.
Suppose 73% of a sample of 300 cross border traders and 64% of the light man-
ufacturers said they fully supported the policies being proposed. Is there suffi-
cient evidence to conclude that the proposed policies are equally supported by
the two groups sampled. Use a 1% level of significance.

8.8. Tests for Independence

Tests for independence are performed on categorical data such as when testing
for independence of opinion on a public policy and gender. The data is con-
tained in what is called a contingency table. The hypotheses are tested using a
Chi - square test statistic, χ2cal .

Illustrative example
A company operates four machines three shifts each day. From production
records, the following data on the number of breakdowns are collected:
Machines
Shifts A B C D
1 4 3 2 1
2 3 1 9 4
3 1 1 6 0

Using 5% level of significance, test the hypothesis that breakdowns are inde-
pendent of the shift.

Solution: To be provided in the lecture.

Exercise
Grades in Statistics and Communication Skills taken simultaneously were
recorded as follows for a particular group of students.
Com. Skills Grade
Stats Grade 1 2.1 2.2 Other
1 25 6 17 13
2.1 17 16 15 6
2.2 18 4 18 10
Other 10 8 11 20

Are the grades in Statistics and Communication Skills related? Use α = 0.01
Hypothesis Testing 85

8.9. Ending Remark(s)

It is required that one may demonstrate that hypothesis testing and confidence
intervals are equivalent procedures insofar as decision making or inference
about population parameters is concerned. However, each procedure presents
different insights. What is the major difference between these two cousin pro-
cedures?
86 Hypothesis Testing
Chapter 9

Regression Analysis

9.1. Introduction

It is important to note that the approach used here first exposes the useful con-
cepts of the regression analysis technique, gives an illustrative example on the
application of these concepts, and then wraps up with a practice question.

Many problems that are encountered in everyday life involve exploring the
relationships between two or more variables. Without attempting to formally
define what regression analysis is, regression analysis is a statistical tool that
is very useful for these types of problems. For example, in the clothing indus-
try, the sales obtained from selling particular designer outfits is related to the
amount of time spent advertising the label. Regression analysis can be used
to build a model to predict the sales given the amount of time devoted to ad-
vertising the label. In the sciences, regression analysis models can be used for
process optimization. For instance, finding the temperature levels that max-
imise yield, or for puproses of process control.

After non-superficial, but serious and rigorous studying of this chapter, the stu-
dent is expected to be able to use simple linear regression for building models
to everyday data, apply the method of least squares to estimate the param-
eters in a linear regression model, use the fitted regression model to make a
prediction of a future observation and interpret the scatter plot, the correlation
coefficient, the coefficient of determination, and the regression parameters.

9.2. Uses of Regression Analysis

The uses of regression are but not limited to:

• understanding underlying processes
88 Regression Analysis

• prediction

• forecasting

• optimisation

• control purposes

9.3. Abuses of Regression Analysis

Regression analysis is widely used and frequently misused. Several common

abuses of regression include developing statistically significant relationships
among variables that are completely unrelated in a cause - effect sense. A
strong observed association between variables does not necessarily translate
into a cause - effect relationship between the variables. Therefore, care must
be exercised when choosing variables on which to perform regression analysis.

Regression relationships are valid only for values of the explanatory variable
within the range of the original data. The linear relationship that we have
assumed may be valid over the original range of X, but may unlikely remain
so as we extrapolate i.e. if we use values of X beyond the range in question
to estimate the value of Y. Alternatively put, as we stride from the range of
the values of X for which data were collected, our certainty about the validity
of the assumed model tend to fade away. We caution that linear regression
models are not necessarily valid for extrapolation purposes. Clearly, this is not
saying NO to extrapolation. Note that in many life situations extrapolation of
a regression model may be the only way to approach a given problem. We are
strongly warning that there is need to be alive to the potential abuses of the
treasure. To dilute the preceding a bit, a modest extrapolation may be quite
fine in most situations, however large extrapolations will almost always pro-
duce unacceptable results.

Well, we will concentrate on two random variables: the explanatory variable

or the independent variable or the cause variable among a host of other names
which is denoted X, and the response variable or the dependent variable or the
effect variable among a host of other names denoted by Y. These two variables
vary together. X causes Y to vary or X explains the response in Y. Such sit-
uations are modeled using the simple linear regression analysis technique be-
Regression Analysis 89

cause they have only one explanatory variable or independent variable. Specif-
ically, this will be our focus in this chapter.

9.4. The Simple Linear Regression Model

The simple linear regression model is an equation of a straight line given by

Y = a + bX + 

where Y is the response / dependent variable, a is a regression coefficient / re-

gression parameter called the intercept, b is a regression coefficient / regression
parameter called the slope, X is the explanatory/ independent variable and e is
the random error term.

The random error term follows a normal distribution with a mean zero and
an unknown variance σ 2 . For completeness, we state that the random errors
corresponding to different observations are also assumed to be uncorrelated
random variables. To determine the appropriateness of employing simple lin-
ear regression we use (1) the scatter plot, and or (2) the correlation coefficient
techniques.

9.4.1. The Scatter Plot

The choice of the model is based on inspection of a scatter diagram. We merely

use our eyes to inspect the nature of the relationship exhibited by the points.
The slope of the points will tell us the direction of the relationship and the
distances between the points will tell us the nature of the magnitude of the
relationship. Invariably, we note that there is a deep seated tendency to join
the points on the scatter plot. We therefore take the unprecedented step of
telling you what is not correct. Let us be clear that the points need not be
joined and that there is NO line of whatever form to be drawn on a scatter
diagram.

9.4.2. The Correlation Coefficient

A correlation coefficient measures the strength and direction of the relation-

ship between two variables that vary together. We will compute the Pearson
90 Regression Analysis

product moment correlation coefficient denoted by r and given by

P P P
n xy − x y
r=p P P P P (9.1)
[n x2 − ( x)2 ][n y 2 − ( y)2 ]

Note that −1 ≤ r ≤ 1. Though seemingly intimidating, this formula is quite

user friendly. Employ it and discover more!

Interpretation of r

When interpreting r note should be taken to mention the magnitude / size of

the correlation and the direction of the linear relationship. The absence of the
other component renders the interpretation incomplete. We now give typical
interpretations for some values of r. If r = −1, then we say that there is a
perfect negative linear relationship between X and Y. If r = 1, then there is a
perfect positive linear relationship between X and Y. If r = 0, then there is a no
linear relationship between X and Y. Note that this implies that other forms
of relationships may best model the situation. If r = −0.9, then that there is a
very strong negative linear relationship between X and Y. If r = 0.9, then there
is a very strong positive linear relationship between X and Y. If r = −0.75, then
that there is a strong negative linear relationship between X and Y. If r = 0.75,
then there is a strong positive linear relationship between X and Y. If r = −0.5,
then there is a fair negative linear relationship between X and Y. If r = 5, then
there is a fair positive linear relationship between X and Y. Any value of r that
is less than −0.5 or 0.5 says that there is a weak negative or weak positive
linear relationship between X and Y. Thus, using simple linear regression to
model the situation is not advisable.

Remark: The interpretation of r must clearly state the magnitude/ size and
direction of the relationship between the random variables X and Y.

9.4.3. Regression Equation

Having established that a linear relationship exists between the random vari-
ables X and Y, we proceed to fit the linear regression model/ line/ equation. To
fit a regression model is to estimate the regression coefficients a and b. The
estimates are denoted b
a and bb. The fitted model is written in the form

Yb = b
a + bbX
Regression Analysis 91

Now, we have fitted a model and we wish to determine how good it is and then
use it for prediction of new values for the system in question. To determine how
good our model is we calculate the values of the response variable in reverse
for each and every value of the explanatory variable and then note the differ-
ence. This difference obtained by subtracting the value of the fitted model from
the actually observed is the error in our model for that observation and it is
called the residual. By performing what is called residual analysis we are able
to come up with a statement on the adequacy of our fitted regression model.

After establishing the adequacy of our model we then proceed to predict future
values of the response variable for the system in question. This is technically
called forecasting.

Computation of the Regression Coefficients

The method used to compute the regression coefficients is called the least squares
method.

We first estimate the slope, b as

P P P
bb = n xy − x y
P 2 P 2
n x − ( x)

And then we immediately estimate a by

a = y − bbx
b

Interpretation of the Regression Coefficients

The intercept, a is the value of Y when X = 0. The slope, b indicates the value of
Y when X changes by a unit value.

Naturally, how much of the variability in the response variable has been ex-
plained by fitting the regression model? To answer this question we need to
compute the following coefficient.

9.4.4. Coefficient of Determination, r2

The coefficient of determination establishes the amount of variability in the

response variable that has been explained or accounted for by fitting a regres-
sion model. It is obtained by squaring the correlation coefficient, r and we have
92 Regression Analysis

0 ≤ R2 ≤ 1. Expressing r2 as a percentage

r2 = r2 × 100%

gives the amount of variability in the response variable that has been explained
by fitting a regression model.

Illustrative Example
Consider the following set of observations. Take X to be the exploratory vari-
able and Y to be the response variable.

Y 1 0 1 2 5 1 4 6 2 3 5 4 6 8 4
X 60 63 65 70 70 70 80 90 80 80 85 89 90 90 90

1. Draw a scatter plot for the above data. Comment on the suitability of
using simple linear regression to describe the relationship.

2. Calculate and comment on the Pearson correlation coefficient.

3. Fit the regression model using the method of least squares. Interpret the
regression coefficients.

4. State how much of the variation in Y has been accounted for by fitting the
linear regression model.

5. Using the fitted regression model, what is the value of Y when X = 60?
What is the residual?

6. What is the value of Y when X = 95?

Solution
A scatter diagram of the above data is shown in the figure below.
Exercise
Consider the following quantities for two random variables X and Y. Let X be
the cause variable and Y be the effect variable.

y 2 = 170045, x2 = 29 and
P P P P P
n = 20, x = 24, y = 1843, xy = 2215

1. Is it appropriate to employ simple linear regression analysis on these

data?

2. Fit the regression model using the method of least squares. What is the
meaning of the regression coefficients?
Regression Analysis 93

Figure 9.1: A scatter diagram of X and Y values

3. How much of the variability in Y has been explained by fitting the linear
regression model above.

4. Using the fitted regression model, what is the value of Y when X = 2?

What is the residual?

5. What is the value of Y when X = 25?

6. Comment on the usefulness of the values in parts (d) and (e) given that
P
for twenty observations x = 24. Hint: You are expected to reflect on the
uses and abuses of the regression analysis technique.
94 Regression Analysis
Chapter 10

Index numbers

10.1. Objectives

After reading this chapter, you will be conversant with

1. The Concept of Index Numbers

2. Uses of Index Number

3. Different Types of Index Numbers

4. Aggregates Method of Constructing Index Numbers

10.2. Introduction

Index numbers are today one of the most widely used statistical indicators.
Generally used to indicate the state of the economy, index numbers are aptly
called barometers of economic activity. Index numbers are used in comparing
production, sales or changes exports or imports over a certain period of time.
The role-played by index numbers in Indian trade and industry is impossible
to ignore. It is a very well known fact that the wage contracts of workers in our
country are tied to the cost of living index numbers.

10.3. What is an Index Number?

By definition, an index number is a statistical measure designed to show changes

in a variable or a group of related variables with respect to time, geographic
location or other characteristics such as income, profession, etc.

10.3.1. Characteristics of an Index Numbers

1. These are expressed as a percentage: Index number is calculated as a

ratio of the current value to a base value and expressed as a percentage.
96 Introduction

It must be clearly understood that the index number for the base year is
always 100. An index number is commonly referred to as an index.

2. Index numbers are specialized averages: An index number is an average

with a difference. An index number is used for purposes of comparison
in cases where the series being compared could be expressed in different
units i.e. a manufactured products index (a part of the whole sale price
index) is constructed using items like Dairy Products, Sugar, Edible Oils,
Tea and Coffee, etc. These items naturally are expressed in different units
like sugar in kg, milk in litres, etc. The index number is obtained as a
result of an average of all these items, which are expressed in different
units. On the other hand, average is a single figure representing a group
expressed in the same units.

3. Index numbers measures changes that are not directly measurable: An

index number is used for measuring the magnitude of changes in such
phenomenon, which are not capable of direct measurement. Index num-
bers essentially capture the changes in the group of related variables over
a period of time. For example, if the index of industrial production is 215.1
in 1992-93 (base year 1980-81) it means that the industrial production in
that year was up by 2.15 times compared to 1980-81. But it does not,
however, mean that the net increase in the index reflects an equivalent
increase in industrial production in all sectors of the industry. Some sec-
tors might have increased their production more than 2.15 times while
other sectors may have increased their production only marginally.

10.3.2. Uses of Index Numbers

Index numbers are used for:

1. Establishes trends - Index numbers when analyzed reveal a general trend

of the phenomenon under study. For eg. Index numbers of unemployment
of the country not only reflects the trends in the phenomenon but are
useful in determining factors leading to unemployment.

2. Helps in policy making - It is widely known that the dearness allowances

paid to the employees is linked to the cost of living index, generally the
consumer price index. From time to time it is the cost of living index,
which forms the basis of many a wages agreement between the employees
union and the employer. Thus index numbers guide policy making.

3. Determines purchasing power of the dollar Usually index numbers are

used to determine the purchasing power of the dollar. Suppose the con-
Index numbers 97

sumers price index for urban non-manual employees increased from $100
in 2004 to $202 in 2006, the real purchasing power of the dollar can be
found out as follows:
100
= 0.495
202
It indicates that if dollar was worth $100 in 2004 its purchasing power is
$49.5 in 2006.

4. Deflates time series data - Index numbers play a vital role in adjusting
the original data to reflect reality. For example, nominal income (income
at current prices) can be transformed into real income(reflecting the ac-
tual purchasing power) by using income deflators. Similarly, assume that
industrial production is represented in value terms as a product of vol-
ume of production and price. If the subsequent years industrial produc-
tion were to be higher by 20% in value, the increase may not be as a result
of increase in the volume of production as one would have it but because
of increase in the price. The inflation which has caused the increase in the
series can be eliminated by the usage of an appropriate price index and
thus making the series real.

10.4. Types of Index Numbers

There are three principal types of indices, these are:

• Price index,

• Quantity index

• Value index.

1. Price Index - The most frequently used form of index numbers is the price
index. A price index compares charges in price of edible oils. If an attempt
is being made to compare the prices of edible oils this year to the prices
of edible oils last year, it involves, firstly, a comparison of two price situ-
ations over time and secondly, the heterogeneity of the edible oils given
the various varieties of oils. By constructing a price index number, we are
summarizing the price movements of each type of oil in this group of edi-
ble oils into a single number called the price index. The Whole Price Index
(WPI) and the Consumer Price Index (CPI) are some of the popularly used
price indices.

2. Quantity Index - A quantity index measures the change in quantity from

one period to another. If in the above example, instead of the price of
98 Introduction

edible oils, we are interested in the quantum of production of edible oils

in those years, then we are comparing quantities in two different years or
over a period of time. It is the quantity index that needs to be constructed
here. The popular quantity index used in this country and elsewhere is
the index of industrial production (HP). The index of industrial production
measures the increase or decrease in the level of industrial production in
a given period compared to some base period.

3. Value Index - The value index is a combination index. It combines price

and quantity changes to present a more spatial comparison. The value
index as such measures changes in net monetary worth. Though the value
index enables comparison of value of a commodity in a year to the value of
that commodity in a base year, it has limited use. Usually value index is
used in sales, inventories, foreign trade, etc. Its limited use is owing to the
inability of the value index to distinguish the effects of price and quantity
separately. What are the methods of constructing Index Numbers?

10.5. Methods of constructing index numbers

There are two approaches for constructing an index number namely

1. Aggregate method

2. Average of relatives method.

Note: The index constructed in either of these methods could be either an

unweighted index (and unweighted index is an index where equal weights are
implicitly assigned) or a weighted index (a weighted index is an index where
weighted are assigned to the various items constituting the index.)

10.5.1. Aggregate Method

Under the aggregates method of constructing an index number, we could have

unweighted and weighted aggregates index.

Unweighted Aggregates Index

An unweighted aggregates index is calculated by totalling the current year/given
years elements and then dividing the result by the sum of the same elements
during the base period. To construct a price index, the following mathematical
Index numbers 99

formula may be used.

P
P1
U nweighted Aggregate P rice Index = P × 100% (10.1)
P0
P
Where, P1 = Sum of all elements in the composite for current year and
P
P0 = Sum of all elements in the composite for base year.

10.5.2. Merits and demerits of this method

Merit - This is the simplest method of constructing index numbers.

Demerits - It does not consider the relative importance of the various com-
modities involved. The unweighted index doesnt reflect the reality since the
price changes are not linked to any usage/consumption levels.

Example
Construct an unweighted index for the three commodities taking 2010 as the
base year.
Prices
Commodities
2010 (P0 ) 2012 (P1 )
Oranges (Dozen) 20 28
Milk (Ltr) 5 8
Gas 76 100

The unweighted aggregate price index (UAPI) by:

P
P1
U AP I = P × 100%
P0
136
= × 100%
101
= 134.65% (10.2)

Above we measured changes in general price levels on the basis of changes

in prices of a few items. While the year 1990 was taken as the base year, a
comparison has been made between the prices of 1992 and that of the base
year 1990.

Interpretation
As evident, the price index was 134.65, which means that the prices rose by
34.65 percent from 1990 to 1992. By no means should this price index be
100 Introduction

interpreted as a reflection of the price changes of all goods and services, as

this calculation is a rough estimate. On inclusion of other items/elements and
varying weights in the composite, with 1990 as the base year and 1992 as the
current year, there is every possibility that the calculated price index would
be different from the price index calculated earlier. This factor can be cited
as one of the drawbacks of the simple unweighted index. The unweighted in-
dex doesnt reflect the reality since the price changes are not linked to any us-
age/consumption levels. On the other hand, a weighted index attaches weights
according to their significance and hence is preferred to the unweighted index.

To make this clear, let us calculate the price index with the same data provided
above but by changing the milk consumption from 1 liter to 100 liters. The
following table provides the calculation of the price index.

10.5.3. Weighted Aggregates Index

In a weighted aggregates index, weights are assigned according to their signif-

icance and consequently the weighted index improves the accuracy of the gen-
eral price level estimate based on the calculated index. Generally, the level of
consumption is taken as a measure of its importance in computing a weighted
aggregates index. There are various methods of assigning weights to an index.
The more important ones are:
1. Laspeyres Method

2. Paasche Method

3. Fixed Weight Aggregates Method

4. Fishers Ideal Method and

10.5.4. Laspeyres Method

Laspeyres method uses the quantities consumed during the base period in com-
puting the index number. This method is also the most commonly used method
which incidentally requires quantity measures for only one period. Laspeyres
index can be calculated using the following formula:
P
P1 Q 0
Laspeyres P rice Index(LP I) = P × 100% (10.3)
P0 Q 0

Where, P1 = Prices in the current year, P0 = Prices in the base year, Q0 = Quan-
tities in the base year.
Index numbers 101

In general, Laspeyres price index calculates the changes in the aggregate value
of the base years list of goods when valued at current year prices. In other
words, Laspeyres index measures the difference between the theoretical cost
in a given year and the actual cost in the base year of maintaining a standard
of living as in the base year. Also, Laspeyres quantity index can be calculated
by using the formula:
P
P0 Q1
Laspeyres Quantity Index(LQI) = P × 100% (10.4)
P0 Q0

Where, Q1 = Quantities in the current year and Q0 , P0 are as defined earlier.

Let us understand this with the help of an example. Calculate the Laspeyres
Price and Quantity Indices for the following production data.

Production Prices
Product Q0 Q1 P0 P1 P 0 Q0 P1 Q0 P0 Q1
1985 1990 1985 1990
Rice 46.60 58.00 700 910 32620.00 42406.00 40600.00
Sugar 14.57 17.92 620 950 9033.40 13841.50 11110.40
Salt 69.46 85.10 205 300 14239.30 20838.00 17445.50
Wheat 33.84 40.30 330 470 11167.20 15904.80 13299.00

Solution

Laspeyres price index is:

P
P1 Q0
Laspeyres P rice Index(LP I) = P × 100%
P0 Q0
9299030
LP I = × 100%
6705990
LP I = 138.67%

Laspeyres quantity index is:

P
P0 Q 1
Laspeyres Quantity Index(LQI) = P × 100%
P0 Q 0
82454.90
= × 100%
67059.90
= 122.96%

10.5.5. Merits and demerits of Laspeyres method?

Merits - A laspeyres index is simpler in calculation and can be computed once

the current year prices are known, as the weight are base year quantities in a
102 Introduction

price index. This enables us an easy comparability of one index with another.

Demerits - Laspeyres tends to overestimate the rise in prices or has an upward

bias. Let us see how it overestimates the rise in prices or has an upward bias.
There is usually a decrease in the consumption of those items for which there
has been a considerable price hike and the usage of base year quantities will
result in assigning too much weight to prices that have increased the most
and the net result is that the numerator of the Laspeyres index will be too
large. Similarly, when the prices go down, consumers tend to demand more of
those items that have declined the most and hence the usage of base period
quantities will result in too low weight to prices that have decreased the most
and the net result is that the numerator of the Laspeyres index will again be
too large. This is a major disadvantage of the Laspeyres index. However, the
Laspeyres index remains most popular for reason of its practicability. In most
countries, index numbers are constructed by using Laspeyres formula.

10.5.6. Paasches Method

Paasche index is similar to that of computing a Laspeyres index. The differ-

ence is that the Paasche method uses quantity measures for the current period
rather than for the base period. The Paasche index can be calculated using the
following formula.

P
P1 Q1
P aasche P rice Index(P P I) = P × 100% (10.5)
P0 Q1

Where, P1 = Prices in the current year P0 = Prices in the base year Q1 = Quan-
tities in the current year. The Paasche quantity index is given by:
P
P1 Q1
P aasche Quantity Index(P QI) = P × 100% (10.6)
P1 Q0

10.5.7. Merits and Demerits of Paasches Index

Merit - Paasches index attaches weights according to their significance.

Demerits - Paasche index is not frequently used in practice when the number
of commodities is large. This is because for Paasche index, revised weights or
quantities must be computed for each year examined. Such information is ei-
Index numbers 103

ther unavailable or hard to gather adding to the data collection expense, which
makes the index unpopular. Paasche index tends to underestimate the rise in
prices or has a downward bias.

Let us understand the Paasches method with the help of an example. The
table below represents the calculation of Paasche index. In general Paasche
index reflects the change in the aggregate value of the current years (given
periods) list of goods when valued at base period prices. From the table below
calculate the Paasche price and quantity indices.

1992 1993
Commodity Price Quantity Price Quantity P0 Q0 P0 Q1 P 1 Q0 P1 Q1
A 3 18 4 15 54 45 72 60
B 5 6 5 9 30 45 30 45
C 4 20 6 26 80 104 120 156
D 1 14 3 15 14 15 42 45
178 209 264 306

Solution
Paasche Price Index (PPI) is:
P
P1 Q 1
P aasche P rice Index(P P I) = P × 100%
P0 Q 1
P P I = 146.41%

Paasche Quantity Index (PQI)

P
P1 Q 1
P aasche Quantity Index(P QI) = P × 100%P QI = 115.91%
P1 Q 0

Paasche price index is 146.41%. Laspeyres Price Index, when calculated is

148.31%

The difference between the purchase index and Laspeyres index reflects the
change in consumption patterns of the commodities A, B, C and D used in that
table. As the weighted aggregates price index for the set of prices was 148.31%
using the Laspeyres method and 146.41 using the Paasche method for the same
set, it indicates a trend towards less expensive goods. Generally, Laspeyres
and Paasche methods tend to produce opposite extremes in index values com-
puted from the same data. The use of Paasche index requires the continuous
use of new quantity weights for each period considered. As opposed to the
104 Introduction

Laspeyres index, Paasche index generally tends to underestimate the prices or

has a downward bias. Because people tend to spend less on goods when their
prices are rising, the use of the Paasche which bases on current weighting, pro-
duces an index which does not estimate the raise in prices rightly showing a
downward bias. Since all prices or all quantities do not move in the same order,
the goods which have risen in price more than others at a time when prices in
general are rising will tend to have current quantities and they will thus have
less weight in the Paasche index.

10.6. Fisher Index

Prof. Irving Fisher has proposed a formula for constructing index numbers, as
the geometric mean of the Laspeyres and Paasche indices i.e. Fisher’s quantity
and price index aare calculated as:
p
F isher0 s Quantity Index = (Laspeyres Quantity Index × P aasche Quantity Index)

p
F isher0 s P rice Index = (Laspeyres P rice Index × P aasche P rice Index)

The following advantages can be cited in favor of Fishers Index:

1. Theoretically, geometric mean is considered the best average for the con-
struction of index numbers and Fishers index uses geometric mean.

2. As already noted, Laspeyres index and Paasche index indicate opposing

characteristics and Fishers index reduces their respective biases. In fact,
Fishers ideal index is free from any bias. This has been amply demon-
strated by the time reversal and factor reversal tests.

3. Both the current year and base year prices and quantities are taken into
account by this index. The Index is not widely used owing to the practical
limitations of collecting data. Fishers Ideal Quantity Index can be found
out by the formula.