1

DESCRIPTIVE STATISTICS

GENERAL OBJECTIVE OF THE COURSE

The overall objective of the course is to teach students the main techniques of univariate
and bivariate descriptive statistics while giving them some basic notions of operations
research in linear programming. At the end of the course, students should be able to apply
these techniques appropriately in a given context. In other words, they should be able to:
 Solve a linear programming problem by the graphical method
 Understand the basic elements of descriptive statistics
 Pull up statistical tables and graphs for one and two characters
 Perform a statistical analysis using descriptive statistics.
 To study the relationship between two statistical variables
 Calculating indices in descriptive statistics

Specifically, it is about :
 Summarize, synthesize the information contained in the statistical series;
 To highlight its properties;
 Suggest hypotheses about the population from which the sample was drawn.
 2
Tools used :
 Tables (frequency table, contingency table, ...)
 Indicators (average, correlation, ...).
The type of tools used depends on
 The nature of the series (single or multidimensional)
 The nature of the variables (discrete quantitative, continuous or qualitative)
Some of these concepts can be developed and deepened in specific courses.

COURSE OUTLINE

This course spans 4 chapters:

General introduction

Chapter 1: Linear Programming

Chapter 2: General information on descriptive statistics
Chapter 3: Univariate Descriptive Statistics
Chapter 4: Bivariate Descriptive Statistics

GENERAL INTRODUCTION

Statistics is the set of mathematical methods and tools aimed at collecting, describing and
analyzing data afin order to obtain information allowing to make decisions despite the
presence of uncertainty (error, noise). In other words, statistics is the set of mathematical
methods allowing to describe and analyze in a quantitative way repeated phenomena by
numerous elements of the same nature likely to be demonstrated and classified.
Statistics has nowadays invaded all scientific fields. Its teaching within the
framework of a training in education sciences is thus quite naturally part of it.
Sooner or later, in your learning projects or professional activities, you will have to
collect data, process them, describe them and, if possible, generalize them.
Statistics plays an essential role in many disciplines:
 In economics: growth rate, number of patents filed, real estate prices, etc.
 In finance: profitability of an investment,...
 In marketing: market research, ...
 In human resources management: absenteeism, ...
 In medicine: marketing of new drugs, ...
 In social sciences, political sciences, etc.
 3
The application of statistical methods makes it possible to obtain statistics which are in fact
numerical results (collection of figures, set of numbers) presented in the form of tables or
graphs.

CHAPTER 1: LINEAR PROGRAMMING

The objective of this chapter is to familiarize the student with purely graphical tools for
solving linear programming problems.

INTRODUCTION
Linear programming is one of the most important optimization techniques used in
operations research. This is due to the ease of modelling, the efficiency of the algorithms
developed and the existence of numerous software packages on the market. The
generalization of micro-computing has made linear programming available to everyone.
The objective of linear programming is to determine the optimal allocation of scarce
resources between competing activities or products. Economic situations often require
optimizing a function under several constraints in the form of inequalities.

1. FORMULATION OF A MAXIMIZATION PROBLEM

1.1 PROBLEM STATEMENT
A company produces two products A and B, using a machine m and two raw materials p
and q. Each day, 8 hours of m, 10 kg of p and 36 kg of q are available. It is assumed that:
- The production of one unit of A requires 2 kg of p and 9 kg of q, and uses the machine
m for 1 hour;
- The production of one unit of B requires 2 kg of p and 4 kg of q, and uses the machine
m for 2 hours;
 4
- The profits made are 50 FCFA per unit of A and 60 FCFA per unit of B.
The company's objective is to maximize the profit it can make per day from these two
products by making the best use of its resources.
The following table summarizes the data related to this production problem:
A B Available

m 1h 2h 8h
p 2 kg 2 kg 10 kg
q 9 kg 4 kg 36 kg
Unit profit 50 60
FCFA FCFA

1.2 Building a linear model

What information does the company manager need to know to consider his problem
solved?
It is enough to know the quantity of product A and the quantity of product B to be
produced daily, isn't it? Let's act as if these quantities were known to us and denote them
with :
x1 = the quantity of product A to be
produced
x2 = the quantity of product B to be
produced
The variables x1 and x2 are called decision variables.
How much profit will the company make from the sale of these two products?
It is a question of adding up the benefits to be gained from each of the two products:
- For product A, it withdraws 50 CFA francs per unit and manufactures x1 units; this
production therefore yields a profit of (50x1) CFA francs;
- Similarly, the x2 quantity of product B allows him to make a profit of (60x2) FCFA.
The total profit to be made from the two products is therefore :
(50x1 + 60x2) FCFA
We will denote this total profit by z and leave the monetary unit implicit:
z = 50x1 + 60x2
We obviously try to make z as large as possible by giving x1 and x2 appropriate values.
The quantity z is a function which, for each production plan (a quantity of A, a
quantity of B), associates the number of dirhams that the firm would earn as profit if it
adopted this plan. This function z, which reflects the objective of our problem, is called the
objective function or economic function. And, since we want to make z as large as
possible, we write :
Maximize z where z = 50x1 + 60x2
which is generally agreed to be abbreviated as follows:
 5
Max z = 50x1 + 60x2
If it were only a question of maximizing z, it would be enough to let x1 or x2 increase so
that z would take on a value as large as it wishes. But expecting such profits is more like a
dream than the situation of our company. There are, of course, natural impediments, called
constraints, that hinder the dream of infinite profit. Let's consider each of these constraints
in turn.

Stress on the machine m

The operating time of the machine m to produce products A and B cannot exceed the
available 8 hours:
Time of use of m ≤ 8.
The time used is the sum of the hours spent on each type of product. For product A, the
time required to produce quantity x1 is calculated as follows
1 hour/(unit of A) × x1 (unit of A) = x1 hours
for product B, the procedure is similar:
2 hour/(unit of B) × x2 (unit of B) = 2x2 hours
The constraint relative to the machine m is therefore written :
x1 + 2x2 ≤ 8 (m)
The sign " ≤"is used, not "=", as it is not mandatory that all available hours be used for the
production of products A and B, although it is not prohibited that this be the case.

Raw material constraints

Based on the machine constraint, these constraints are written quite naturally:
2x1 + 2x2 ≤ 10 (p)
9x1 + 4x2 ≤ 36 (q)
Positivity constraints
They ensure that the solution does not contain negative (unacceptable) values.
x1, x2 ≥ 0,
The model can be summarized as follows:

1.3 Variables of deviation

In order to reduce the constraints to equalities (which are easier to handle than
inequalities), we introduce deviation variables. These variables will always be, like the
decision variables x1 and x2, positive or zero.
After adding the deviation variables x3, x4 and x5 for the constraints (m), (p) and
(q), we obtain the formulation :
 6

The variance variable of a constraint represents the unused available quantity. It is the
difference between the availability and the requirement.
1.4. Generalization: Production problem
Let there be m machines Mi (i = 1,... ,m) which mass produce n types of products Pj (j =
1,...,n). Machine Mi has a maximum capacity of bi time units. The production of one unit of
product Pj requires the use of machine Mi for aij time units.
If cj is the gain from producing one unit of product Pj, solving the problem

will provide the optimal values of the quantities xj to be produced of each of the products;
the economic function represents the total gain; the first member of each constraint (Mi)
represents the total time of use of the machine Mi. The values of the deviation variables
associated with each of these constraints correspond to the unused available time of each
machine.
1.2 FORMULATION OF A MINIMIZATION PROBLEM
1.2.1 Problem statement
A farmer wants his herd to consume the lowest daily ration of three nutrients A, B and C.
The daily requirements are 16 for A, 12 for B and 18 for C. The farmer buys two types of
feed P and Q :
- A unit of P includes 2 units of A, 1 unit of B and 1 unit of C; and it costs 20 FCFA
- A unit of Q consists of 1 unit of A, 1 unit of B and 3 units of C; and it costs 40 FCFA.
The farmer looks for the least expensive combination of P and Q amounts that will meet
the minimum nutrient consumption requirement.
The following table summarizes the data related to this problem:
P Q Minimum
 7
requirements
A 2 1 16
B 1 1 12
C 1 3 18
Unit cost 20 40
FCFA FCFA
1.2.2 Building a linear model
Let us call x1 and x2 the quantities of feed P and Q that need to be purchased. The farmer's
objective is obviously to minimize the total cost of the food that must be purchased.
Mathematically this is written as :
Minimize z = 20x1 + 40x2
which is generally agreed to be abbreviated as follows:
Min z = 20x1 + 40x2
Each of the 3 nutrients to be considered gives rise to a constraint, which aims to require
that the feed, as a whole, satisfies the daily needs of the herd. We obtain:

The above constraints use the sign " ≥"because the minimum consumption requirements
must be met, but may be exceeded.
Finally, we must not forget that we cannot buy negative quantities of P or Q :
x1, x2 ≥ 0
The model can be summarized as follows:

1.2.3. Generalization: Mixing problem

The aim is to find a diet that is as inexpensive as possible, while ensuring an adequate
supply of nutrients.
Let n foods have a price cj (j = 1,...,n) per unit; m nutrients; aij the amount of the ith nutrient
contained in a unit of the jth food; bi (i = 1,...,m) the respective requirements of the m
nutrients.
If the variables xj (j = 1,...,n) represent the quantities of the various foods in the diet, we get
the problem
 8

This problem is nothing more than a mixing problem. It arises in many situations, for
example, in the composition of an animal diet.

2 GRAPHICAL METHOD

In this section we present a technique for solving a linear program with two variables x1
and x2. In this case we can use a graphical representation of the linear program. The
graphical representation will be useful to gain an intuitive understanding of the basic
principles of linear programming. The canonical formulation of a linear program with
two variables can be written in one of two forms:

2.1 SOME GEOMETRY REMINDERS

2.1.1. The construction of the feasible region
Each of the equations aix1 + bix2 = ci defines a line that divides the plane into two half-planes
P1 and P2 of equation :
 9

Each constraint determines one of the two half-planes (P1) or (P2) that can be found by
checking whether a particular point (the origin (0,0) for example) is contained in it or not.
The intersection of all the half-planes corresponding to the constraints constitutes the set of
feasible points: these are the solutions common to all the constraints. This set corresponds
to a region D of the plane and is often called a feasible region. This region is sometimes
empty or unbounded.
y y

D
D

o x o x
Bounded regionUnbounded region
2.1.2 The search for an optimal solution
We have just constructed the feasible region of a linear program in two variables. This
set contains an infinite number of feasible solutions. It remains to find, among these
feasible solutions, the one(s) which give(s) the best value to z.
Fixing z at an arbitrarily chosen value p, we obtain the line :
(Dp) a0x1 + b0x2 = p
This line is called the z-function iso value line. It represents the points in the plane that
give z the value p.
We are interested in the family of lines (Dp) (p parameter). They are all parallel lines with

slope : , which can be written :

 10

- If b0 > 0, maximizing z is equivalent to maximizing p and thus yp. So the maximum of

z is obtained for the line having at least one common point with the feasible region
and having the highest possible intercept.
y direction z croissants y direction z croissants

D
B
D D

A
C
o x o x
Case b0 > 0 and a0 > 0 Case b0 > 0 and a0 < 0
- If b0 < 0, maximizing z is equivalent to maximizing p and thus minimizing yp.
Therefore, the maximum of z is obtained for the line having at least one common
point with the feasible region and having the lowest possible intercept.
y direction z croissants y direction z croissants

D
B
D D

A
C
o x o x
Case b0 < 0 and a0 > 0 Case b0 < 0 and a0 < 0
All the points on the same line ensure the same value for z. When one passes from one
line to another, the values of z vary: they increase if one moves in the direction of the
normal vector to the iso value lines: hence the meaning of the arrow
indicating the "direction of increasing z".
2.2 MAXIMIZATION PROBLEM
2.2.1 Problem statement
A chassis manufacturing company is planning to produce two new models using the
capacities of its three workshops. These are an aluminium frame and a wooden frame
 11
respectively. The first product requires the first workshop to manufacture the aluminium
frame and the third workshop where the glass is mounted on the frame. The second
product requires the second workshop to manufacture the wooden frame and the third
workshop where the glass is mounted on the frame. The unit profits, the manufacturing
times of each product in each workshop and the weekly capacities of these workshops
are given in the following table:
Product 1 Product 2 Available capacity
(hours/product) (hours/product) (hours/week)
Worksho 1 0 4
p1
Worksho 0 2 12
p2
Worksho 3 2 18
p3
Profit 300 FCFA 500 FCFA
The question that arises is, "How many frames of each type should be produced per week
in order to achieve maximum profit?"
2.2.2 The linear model
What are the quantities of chassis that the company should produce per week, if it wants
to maximize its profit? The decision variables will be:
x1 = number of type 1 chassis to be produced per week
x2 = number of type 2 chassis to be produced per week
The problem of planning the production of chassis is translated into the following linear
model:

To each couple of values of the decision variables x1 and x2 we associate a point (x1,x2)
of the R2 plan: the point (x1,x2) is interpreted as the proposal of a production plan
indicating the number of frames to be manufactured per week. Generally, we speak
indifferently of the point (x1,x2) or the production plan (x1,x2) or the solution (x1,x2).
2.2.3. Construction of the feasible region
The constraints x1 ≥ 0 and x2 ≥ 0 mean that all points (x1,x2) representing acceptable
solutions must be in the first quadrant, i.e. to the right of the Ox2 axis and above the Ox1
axis.
Note that all the constraints of the model, including the non-negativity constraints, can be
written in the following form:
 12
ax1 + bx2 ≤ c
To clarify the procedure that leads to the graphical representation of the feasible region,
let us first consider the constraint on Workshop 1. The line associated with this
constraint is the line with equation :
x1 = 4 (1)
The graphical representation of equation (1) is obtained by drawing the vertical line which
passes through the point (4,0). The constraint (A1) is satisfied for the points of the plane
on the same side of the line (1) as the origin of the reference frame.

Since all the constraints in the linear model are of the same form, it is sufficient to
consider them in turn, proceeding, for each constraint, as follows:
- Choose 2 points of the associated line. In order to facilitate the calculations, we can
choose the two points having each one a null coordinate. Then draw this line ;
- Determine on which side of the associated line the constraint is satisfied. To quickly
find the right side, simply look to see if the point (0,0) is on the right side;
- Draw an arrow pointing to that side.
Let's illustrate this procedure again using the constraint (A2). The line associated with
this constraint is the line of equation :
x2 = 6 (2)
- The points (0,6) and (8,6) belong to the line (2) associated with (A2) ;
- The point (0,0) verifies the latter and the points that are feasible according to the
only constraint (A2) are under the line (2) ;
- The arrow pointing to the right side is placed.
The graphical result of considering constraints (A1) and (A2) is shown in the following
figure:
 13

The set of points that have not been eliminated after all constraints have been taken into
consideration constitutes the feasible region. In economics, this set is also called the
production set.
2.2.4 Finding an optimal solution
The feasible region we have just constructed does not determine a solution, but a whole
domain containing an infinite number of points. We will have to find an optimal
production plan, i.e. a point verifying all the constraints and allowing z to take its
largest value on the feasible region.
For this, we will consider successive values of the economic function:
z = p.
This corresponds graphically to parallel lines
300x1 + 500x2 = p. (4)
 14
The points of one of these lines are therefore the locus of all points giving the same value
of profit (hence the name of iso-profits lines).
Drawing a first iso-profit line gives an illustration of the slope of z. Drawing a second
iso-profit line allows us to determine the direction in which the value of z increases.
Thus, by gradually increasing the value of p in equation (4), we obtain parallel lines,
each further from the origin than the previous ones.

Since the economic function z is sought to be maximized here, an optimal solution is

determined by searching for the highest iso-profit line that has at least one point in the
feasible region.

Highest iso-profit line Graphical solution alignment

Here, there is only one optimal solution which is the point

This point lies at the intersection of lines (2) and (3). The available capacities of
workshops 1 and 2 are therefore fully used. On the other hand, for this optimal solution,
the available capacities of workshop 1 will be more than sufficient. At the optimal point,

z∗ = (300 × 2) + (500 × 6) = 3600

the economic function takes the value 3600 :
 15
2.3 MINIMIZATION PROBLEM
2.3.1 Problem statement
A company needs three raw materials A, B and C to produce a product. It needs at least
14 kg of A, 12 kg of B and 18 kg of C. It can only buy mixtures X and Y . Product X
contains 2 kg of A, 1 kg of B and 1 kg of C. Product Y contains 1 kg of A, 1 kg of B and 3
kg of C. Product X costs 20 FCFA and Y costs 40 FCFA. Moreover, he cannot buy more
than 14 kg of X and 16 kg of Y
The question that arises is, "How much of X and Y should be purchased to meet the
company's needs at the lowest cost?"
2.3.2 Formulation and graphical resolution
Let us call x1 and x2 the quantities of X and Y to be purchased. The following linear
program is easily obtained:

Constraints (A), (B) and (C) are written as " ≥" because the minimum raw material
requirements must be met, but these can be exceeded.
This program differs from the previous one in 2 aspects: the economic function is to
minimize; some constraints are of sign " ≥"and the origin O = (0,0) of the plane is not a
feasible solution.
The graphical resolution technique remains the same, but since it is a search for a
minimum, the iso value line closest to the origin and which remains in contact with the
feasible region, provides the minimum.
First, we construct the feasible region. The hatched area in the graph below is the
set of points satisfying all constraints.
 16

The polyhedron ABCDEF, which constitutes the feasible region, is shown in the
following figure, along with the isocost lines z = 0, z = 100 and z = 440. Since the
objective is to minimize z, the optimum is achieved by finding the lowest possible
isocost line that touches the feasible region D.

isocosts z = 0, z = 200 and z = 440. the vertex B corresponds to the optimal solution
The graph shows vertex B=(9,3) as the point where the optimal solution is obtained. The
cost at point B is lower than the cost at all other vertices in the feasible region. This
result is confirmed by evaluating the economic function at each of the vertices adjacent
to B :
in A = (14,1) z = (20 × 14) + (40 × 1) = 320
in B = (9,3) z = (20 × 9) + (40 × 3) = 300
in C = (2,10) z = (20 × 2) + (40 × 10) = 440
Suppose that the cost of product X increases by 20 FCFA. The economic function
becomes :
Min z = 40x1 + 40x2 Let's find
the minimum value of z graphically. We obtain
 17

Since the lowest possible isocost line is tangent to constraint (B), the minimum value
would be expected at vertices B(9,3) and C(2,10). Moreover, any point on the segment
[B, C] will minimize the economic function under the constraints. This multiplicity of
optimal solutions comes from the fact that the isocost lines are parallel to the line
associated with the constraint (B).

CHAPTER 2: THE BASICS OF DESCRIPTIVE STATISTICS

Objectives:
The general objective of this chapter is to provide the basic tools of descriptive statistics.
Specifically, the student should be able to, by the end of this chapter:
 Distinguish between the different branches of statistics ;
 18
 Differentiate between qualitative and quantitative variables;
 Differentiate between discrete and continuous statistical variables;
 Master the basic vocabulary of descriptive statistics.
1 DEFINITION
Statistics is the study of data collection, analysis, processing, interpretation and
presentation to make the data understandable to all. It is at the same time a science, a
method and a set of techniques.
Or, it is also the set of methods that allow to collect, order, reduce and condense data.
It uses parameters (or synthesizers), graphs, and data analysis methods (programmable
calculators, computers...).
There are two branches of statistics: descriptive (definitive) statistics and inferential
(inductive) statistics. It should be noted that these branches do not oppose each other, but
complement each other.
Descriptive statistics: Descriptive statistics is a prerequisite to inferential statistics, in
which statistical methods are applied to infer properties of a population from a sample. Its
purpose is to describe and summarize in a formalized way the information collected in an
exhaustive way.
Inferential statistics (inductive) :
 Make an estimate
Ex: Estimate the average weight of individuals in the population from the average
weight of individuals in the sample.
 Testing a hypothesis
Ex: Test from the sample data whether the average weight in the population is greater than
70 kg.
 In general, draw conclusions about a population using results obtained from a
sample.
Note: Inferential statistics only makes sense when you work on a sample drawn
(randomly) from a reference population.

Figure 1: Classification of Statistics

 19
2 BASIC CONCEPT
2.1 POPULATION
2.1.1 Definition :
The population refers to the set of statistical units concerned by the quantitative
study. In other words, it is the set of units or individuals on which a statistical analysis is
carried out. Alternatively, the population refers to the set of objects or individuals
(statistical units) that one wishes to study. Statistical units, also called individuals, are the
abstract entities that represent people, animals or objects.
2.1.2 Examples :
The set of marriages celebrated in a town hall or during a period.
Students from Cameroonian universities.
Employees of any company.
The inhabitants of any neighbourhood.
2.2 SAMPLE OF A POPULATION.
2.2.1 Definition :
In statistics, a sample is a set of individuals representative of a population. That is to say, a
handful of individuals taken from a given population to conduct a statistical study.
2.2.2 Examples :
Cars registered in OU.
The university establishments of Douala.
The inhabitants of the city of Dschang.
Employees of the Civil Service.

Figure 2: Statistical units, population, sample

Population
Sample 2

Sample 1

Units
 20
2.3 THE E-MODALITIES OF A CHARACTERISTIC OR A VARIABLE.
2.3.1 Definition :
A characteristic can have two or more values or present two or more different situations
called the modalities of the characteristic. In other words, the modalities are different
possible values or situations that a characteristic (or a statistical variable) can take or
present.
The modalities of a character must have the following characteristics: they must be
incompatible, exhaustive and unambiguous. Example: for the variable gender, the terms
are male / female.
Note: The terms of a variable are meaningful only if the variable has an operational
definition, that is, a clear and universally accepted meaning.

2.4 CHARACTERS (OR VARIABLES)

2.4.1 Definition :
A characteristic is a particular aspect of the individual (in the statistical sense of the term,
an individual can be a company, a student, etc.) in which the statistician is interested. Each
characteristic has modalities.
2.4.2 Examples :
The power and colour of a registered vehicle OR Characters (or Variables) C=
power and colour.
* Tuition fees and location of Douala's academic institutions Characters (or
Variables) C= Tuition fees and location.
The salary and gender of an employee in a company Characters (or Variables) C=
salary and gender.
The age and marital status of an inhabitant of the city of Dschang Characters (or
Variables) C= age and marital status.
There are two types of characteristics: qualitative and quantitative.
Most statisticians use the term "characteristic" or "variable" interchangeably.
However, some prefer to use the term "variables" to refer only to quantitative
characteristics.

2.4.3 The 2 Character Types (or Variables) :

A character (or variable) can be :
a) Quantitative: A variable is said to be quantitative when the characteristic observed on
the statistical individual can be expressed in numerical form, i.e. measurable.
Examples:
The power of a vehicle numbered LT Quantitative characteristic (or variable) C=
power.
Number of children in a household C = number of children
Tuition fees in Douala universities Quantitative Characteristic (or variable) C=
tuition fees.
The salary of a civil servant Quantitative characteristic (or variable) C= salary.
 21
The age of an inhabitant of the city of Dschang Quantitative characteristic (or
variable) C= age.

b) Qualitative: Qualitative characteristics are present when the characteristic observed on

the individual is not measurable by means of a number. In other words, it can be identified
without being measurable.
Examples:
The colour of a numbered vehicle LT Qualitative characteristic (or variable) C=
colour.
The location of Douala's academic institutions Qualitative characteristic (or variable)
C= location.
The gender of an employee Qualitative characteristic (or variable) C= gender.
The marital status of an inhabitant of the city of Dschang Qualitative characteristic
(or variable) C= marital status.

Case of a Quantitative Characteristic (or variable) :

A quantitative character (or variable) can be :
a) Discreet:
A variable is said to be discrete when it can only take integer values within an interval.
Example: Number of children per family = { M i }= { x i }= { 0; 1 ; 2 ; 3 ; … … . }.
b) Continuous :
A variable is considered to be continuous when it can take all the values of an interval. In
other words, the measure of the character (or variable) is a decimal number taken from a
set of possible values grouped into intervals called classes.
Example:
Size of an individual= { M i }={ x i }=¿.
Weight of an individual= { M i }={ x i }=¿.

Case of a Qualitative Characteristic (or variable) :

The measure of the qualitative characteristic (or variable) can be a list, a nomenclature,
or a code... established before or after the survey.
Example:
Socio-professional category: Executive (C), Employee (E), Worker (O), ...
Marital status of an individual: Married (M), Single (S), Widowed (W), Divorced
(D)….

Diagram3 summarizing different types of variables

22
 23
CHAPTER 3: UNIVARIATE DESCRIPTIVE STATISTS
In this chapter we will follow the following plan

I) GRAPHIC REPRESENTATIONS
I.1) Case of a Qualitative Characteristic (or Variable).
I.2) Case of a Quantitative Characteristic (or Variable).

II) CENTRAL TENDENCY CHARACTERISTICS (POSITIONAL PARAMETERS)

II.1) The m0 Mode and the Modal Class.
II.2) The Median me.
II.3) Quartiles Q1 and Q3.
II.4) Deciles D1 and D9.
II.5) The Arithmetic Mean m = x of a Statistical Series.

(III) CHARACTERISTICS OR PARAMETERS OF DISPERSION

III.1) The Scope e.
III.2) The Interquartile rangeq.
III.3) The Interdecile Gapd.
III.4) The Variance V(x) and the Standard Deviation x of a Statistical Series (xi ; ni).

(IV) SHAPE CHARACTERISTICS

V) APPLICATION EXERCISES

Objectives:
The general objective of this chapter is to characterize univariate statistics.
Specifically, the student must master :
 The characteristics of central tendencies of a statistical series and to be able to
determine them and especially to interpret them in front of a statistical problem;
 Dispersion characteristics and how to calculate and interpret them;
 Form and concentration characteristics: calculate and interpret them in relation to a
statistical study.

Introduction
Unlike qualitative characteristics, where graphical representation is the only means of
studying their distribution, quantitative characteristics lend themselves to several other
means of studying their distribution. Thus, the analysis of a quantitative variable can be
done on the basis of so-called central tendency characteristics (or positional characteristics)
such as the mean, mode and median. The analysis can also be based on dispersion
characteristics such as the range, variance and interquartile ranges, and even on shape
characteristics such as the skewness coefficient, the skewness coefficient and the kurtosis
coefficient.
 24

I GRAPHIC REPRESENTATIONS.
These are graphs of the numbers ni or frequencies fi of the modalities xi in proportion
to the lengths or areas of the graphs.
I.1 CASE OF A QUALITATIVE CHARACTERISTIC (OR VARIABLE) :
In the case of a statistical distribution with a qualitative character, the modalities must be
arranged in the table according to the logic that facilitates the understanding and
interpretation of the phenomenon studied.
Three types of graphical representation are used:
- The strip chart ;
- The representation by organ pipes or in bar ;
- The pie chart or circular diagram or pie chart.
NB: The curve obtained by joining the vertices of the bars is called a statistical
polygon or polygonal diagram.
1) The Strip Chart :
(a) Method: We carry in :
* on the x-axis: the modalities xi, arbitrarily ;
on the ordinate: line segments whose length is proportional to the number ni or the
frequency fi of each modality xi.
b) Example:
Let xi be the measures of the marital status of individuals in a given population:
Married (M), Single (C), Widowed (V), Divorced (D), Married Life (MV).
Modalities xi M C V D VM
Staffing and 350 150 100 75 200

2) The Organ Pipe or Bar Graph :

(a) Method: We carry in :
* on the x-axis: the modalities xi, arbitrarily ;
on the ordinate: rectangles whose length is proportional to the number of people or
to the frequency fi of each modality xi.

b) Example: In the previous example, let's give the histogram of this series:
 25

or

3
0
2
0
1
0

0 D x
C V V
i
M
3) The Sector or Pie Chart :
(a) Method :
Divide a disk into sectors corresponding to the modalities xi of area proportional to
the number ni or the frequency fi of each modality.
b) Example:
In the previous example, let's give the pie chart of this series:

Situation Matrimoniale

Marié(M)
Célibataire(C )
Veuf(V)
Divorcé(D)
Vie Maritale(VM)

I.2 CASE OF A QUANTITATIVE CHARACTERISTIC (OR VARIABLE) :

Two types of graphical representation are used:
- The bar chart for a discrete quantitative variable which is a representation of the
numbers ni or frequencies fi; the histogram for a continuous quantitative variable.
 26

Exercises :
Let the following series be

1) a- Represent the series A by a table of effectives, and complete it with the frequency
distribution ;
b- Still for the series A, group the data by classes of amplitude 5 points;
c- Draw the bar graph representing the series of notes of the series A ;
2) For the B series, for example, draw the histogram
3) Draw the pie chart, organ pipe chart and strip chart.
 27
II POSITION OR CENTRAL TENDENCY PARAMETERS.
Central tendency characteristics (or positional parameters) are values of the statistical
variable located in the centre of the distribution. There are three types of central tendency
characteristics: mean, median and mode.
II.1 THE M0 MODE AND THE MODAL CLASS.
1) Definitions :
* The mode refers to the value of the variable that corresponds to the maximum frequency
observed on the distribution. In other words, the mode is the distinct value corresponding
to the highest effective noted M0. Depending on the nature of the quantitative variable, the
way to determine it is not the same.
When the variable is discrete, the mode value can be determined by looking at the
highest frequency - absolute or relative. The mode is the value of the variable that has that
frequency. In the case of a bar chart, the mode corresponds to the highest chart. If 02
successive values of the variable have the highest frequency, a modal interval is defined.
When the mode is not unique, it loses all meaning.
When the variable is continuous, a modal class is defined. It is such that the average
frequency per unit of amplitude is the highest when the classes do not have the same
amplitude. By convention, the mode can be the center of the class or determined
graphically.
NB: When the variable is discrete, we can determine the value of the mode by
looking at the highest frequency - absolute or relative. The mode is the value of the
variable which has this frequency.
For a continuous quantitative character, i.e. the data are grouped in classes, we call
modal class, the class of the character xi having the greatest number of numbers and its
center is the modal value. By convention, we can retain for the mode the center of the
modal class.
Note1 : It is generally assumed that the distribution is uniform within each class.
Note 2:
- The mode can be calculated for all types of variables, quantitative and qualitative.
- The mode is not necessarily unique.
- When a continuous variable is partitioned into classes, we can definite a modal class
(class corresponding to the highest effectif).
Note: it is possible that a series does not have a
 28

2) Examples:
(a) Example 1:
Let xi be the measures of the marital status of individuals in a given population: Married
(M), Single (C), Widowed (W), Divorced (D), Married Life (MV).
Modalities xi M C V D VM
Staffing and 350 150 100 75 200
The mode is m0 = xi = M
(b) Example 2:
Number of calls 0 1 2 3 4 5 6 7 8 Total
xi
No. of minutes or 93 261 416 393 308 174 93 42 20 N=1800
ni 5,2 14,5 23,1 21,8 17,1 9,7 5,2 2,3 1,1 100
Frequency fi=
N
The mode is m0 = x2 = 2.
(c) Example 3:
The distances xi (in thousands of km) travelled by each of the 150 cars of a car rental
company are recorded.
The results of this survey are :
Distances xi 0 ; 5  5 ; 7  7 ; 9  9 ; 15  Total
a+b 2,5 6 8 12
Centers here = of classes
2
Staffing and 15 78 36 21 150
The modal class is m0 = x2 = 5; 7 /2 = 6 and the modal value is ci = 6 .

II.2 THE MEDIAN ME.

1) Definition:

The median of a variable noted Me or x1/2 is the value which divides the distribution of
the values arranged in ascending or descending order into two parts of equal numbers.
Or, It is the observed or possible value of the series ordered in ascending or descending
order, which divides this series into two sub-series, each comprising the same number of
observations

2) Practical determination of the Median (me) for a discrete variable :

Application exercise :
Student A received 8 marks which are: x1 = 5; x2 = 5; x3 = 6; x4 = 9; x5 = 10; x6 = 12; x7
= 13; x8 = 13.
Student B received 9 marks which are: x1 = 2; x2 = 3; x3 = 5; x4 = 6; x5 = 8; x6 = 9; x7 =
9; x8 = 10; x9 = 10.
 29
Student C received 10 marks which are: x1 = 6; x2 = 6; x3 = 10; x4 = 12; x5 = 12; x6 = 13;
x7 = 14; x8 = 15; x9 = 16; x10 = 16.
Student D received 11 marks which are: x1 = 0; x2 = 0; x3 = 1; x4 = 4; x5 = 5; x6 = 8; x7
= 10; x8 = 12; x9 = 13; x10 = 16; x11 = 17.
Let's find the median me of each student:
xN +xN x8 + x 8 x 4 + x 5 9+10
+1 +1
For student A (N= 8 : even) : we have me = 2 2
= 2 2 = 2 = 2 = 9,5
2 2
Let me = 9.5.
For student B (N= 9 : odd) : we have me = x N +1 = x 9+1 = x 5 = 8. So me = 8.
2 2
xN +xN x 10 + x 10 x5 + x 6 12+ 13
+1 +1
For student C (N= 10 : even) : we have me = 2 2
= 2 2 = = =
2 2 2 2
12,5. So me = 12.5.
For student D (N= 11 : odd) : we have me = x N +1 = x 11+ 1 = x 6 = 8. Let me = 8.
2 2

3) Practical determination of the median for continuous variables

In this case, the median can be determined by linear interpolation and graphically. By
linear interpolation, the median :
1
−F (xi)
Me = xi + (xi+1 - xi)* 2
F ( xi+ 1 )−F (xi)
Graphically, the distribution function which is the cumulative curve is a continuous curve.
We consider three points of this curve A, B, C. Point A (xi, F (xi)), point B corresponds to
the median and point C (xi+1; F (xi+1)). Then, we consider the triangles ABC and ADE.
They are 02 similar triangles having a common base. According to the Thales property, we
AB AD
have : =
BC DE
AB represents Me - Xi
AD represents Xj - Xi, which is the amplitude of the median class.
BC is 0.5 - F(xi), and De is F(xj) - F(xi).
AB AD Me −Xi
Let be the relative frequency fi. Thus legality = becomes =
BC DE 0 ,5−F (xi)
xj−xi
.
F ( xj )−F ( xi)
By positing xj -xi = ai and F(xj) - F(xi) = fi
 30
1
0 ,5−F (xi) −F (xi)
Me = xi +ai [ ] so AB AD Me−xi 2
fi = é quivaut à = =Me
BC DE xi+1−xi F ( xi+1 )−F (xi)
n
Note: the values 0.5 or values appear between two rows in the table, the median class is
2
the lower of the two.
Determining the median: graphical method

NB: the median is obtained by projecting the intersection of the two curves onto the x-axis
and obtaining the value of the median.
II.3 QUARTILES Q1 AND Q3.
1) Definitions :

The quartiles also called quantiles designate the values of the character xi separating an
ordered series into 4 segments of equal size (i.e. containing the same number of values).
There are 3 quartiles noted Q1, Q2 and Q3, from lowest to highest. The size of each of the
4 segments represents ¼ (i.e. 25%) of the size of the series as shown in the diagram below:

 The first quartile, Q1, is the value that divides the sample by the proportions ¼ and
¾: 25% of the observations are below Q1, and 75% are above it.
 31
 The second quartile, Q2, is the value that divides the sample into two equal parts:
50% of the observations are below Q2, and 50% are above it. Q2 is therefore the
median.
 The last quartile, Q3, is the value that divides the sample by the proportions ¾ and
¼: 75% of the observations are below Q3, and 25% are above it.

Important note on quartiles

 The Q1 and Q3 quartiles are certainly indicators of position, but not of central
position.
 Only the second quartile, Q2, is an indicator of centrality and this is because it
corresponds to the median.
 The second quartile, Q2, can be considered as simply another name for the median.

Quartile status
We can find out where the quartiles are in the ordered series using the formulas below.
 Position of the first quartile Q1 : (n+1)/4th value
 Position of the second quartile (median) Q2 : (n+1)/2 nd value
 Position of the third quartile Q3: 3(n+1)/4 th value, where n is the number of
observations in the series.

2) Practical Determination of Quartiles :

1st N
If the total number N is a multiple of 4, then the quartile Q1 is the term of rank
4
3N
and the 3rd quartile q3 is the rank term i.e. q1 = x N and q3 = x 3 N .
4 4 4

If the total number N is not a multiple of 4, then the 1st quartile Q1 is the term of rank
N 3rd
immediately higher than and the quartile Q3 is the term of rank immediately
4
3N
higher than i.e. Q1 = x N +1and Q3 = x 3 N +1.
4 4 4

3) Application exercise :
Student A received 8 marks which are: x1 = 5; x2 = 5; x3 = 6; x4 = 9; x5 = 10; x6 = 12;
x7 = 13; x8 = 13.
Student B received 9 marks which are: x1 = 2; x2 = 3; x3 = 5; x4 = 6; x5 = 8; x6 = 9;
x7 = 9; x8 = 10; x9 = 10.
Solution
Let's find the quartiles of each student:
For student A (N= 8 =4x2): we have q1 = x N =x 8 =x 2= 5 and q3 = x 3 N = x 3 x 8 = x 6=¿ 12
4 4 4 4
Let q1 =5 and q3 = 12.
For student B (N= 9 =4x2 +1): we have q1 = x N +1=x 9 +1=x 3= 5 and q3 = x 3 N +1= x 3 x 9 +1=
4 4 4 4
x 7=¿ 9 Let q1 =5 and q3 = 9.
 32
a- Generalization: quantiles
The quartiles Q1, Q2, Q3 are the quantiles of order 25%, 50% and 75%. In the ordered
series, 25% of observations precede Q1 and 75% of observations follow Q1.
The deciles D1, D2 , ..., D9 are the quantiles of order 10%, 20%, ... 90%. In the ordered
series, 20% of observations precede D2 and 80% of observations follow D2.
The percentiles C1, C2, ..., C99 are the quantiles of order 1%, 2%, ..., 99%. In the ordered
series, 1% of observations precede C1 and 99% of observations follow C1.
In empirical studies, the median is a feature that many researchers prefer to the arithmetic
mean. This is because the median has a very concrete meaning and is not very sensitive to
outliers. However, when studying a discrete variable with a small number of possible
(distinct) values, the use of the median is not recommended. The same is true for quantiles.
Remark:
- The median is the quantile of order p = 1/2.
- We often use
x1/4 for the first quartile,
x3/4 for the third quartile,
x1/10 for the first decile,
x1/5 for the first quintile,
x4/5 for the fourth quintile,
x9/10 for the ninth decile,
x0.05 for the fifth percentile,
x0.95 for the ninety-fifth percentile.

5) The Arithmetic Mean m = x of a Statistical Series.

5.1) Definition :
The mean is a value-based indicator of central tendency. More specifically, the mean is the
single value that all statistical units considered identical in a population (or sample) would
have to have in order for their total to remain unchanged. There are several types of
averages, in particular the arithmetic mean, the geometric mean, the quadratic mean and
the harmonic mean.
1- Arithmetic average

There are two types of arithmetic averages: the simple arithmetic average and the weighted
arithmetic average.
 Simple arithmetic average
 33
The simple arithmetic mean is used when each statistical unit has the same share in the
total value (this is known as equi-repartition). The average is the sum of the observed
values divided by their number, it is noted :
x 1+ x 2+ …+ xi+…+ xn 1 n
x= = ∑ xi
n n i=1
Example: The numbers of children in 8 families are 0, 0, 1, 1, 2, 3, 4.
0+0+1+1+1+2+3+ 4 12
The average is : x = = =1 ,5
8 8
 Weighted arithmetic average

The weighted arithmetic mean takes into account the number of individuals and the value xi of
the variable. The different xi values do not have the same importance (or weight); the
calculation of the average must therefore take into account these differences in weight,
hence the weighting:
n n
1
moyenne pondérée X= ∑ ¿ xi=∑ fi xi
n i=1 i=1
¿
NB: fi = n is called relative frequency or simply frequency.

Let be a quantitative statistical series (xi; ni) of frequencies fi.

Characters xi x1 x2 …… xp Total
p
Staffing and n1 n2 …… np
N =∑ ni
i=1
n f1 f2 …… fp p
Frequencies fi= i f =∑ f i= 1
N i=1

The average of this series is

∑ ni x i n1 x1 +n 2 x 2 +… ..+n p x p
x= i=1 =
N n1+ n2+ … ..+n p
p
where N =∑ ni=n1 +n 2+ …..+ n p.
i=1

2) Other Expression of the Arithmetic Average x :

p
ni
With f i= we show that : x=∑ f i x i=f 1 x1 + f 2 x 2 +… ..+ f p x p.
N i=1

3) Application exercises :
a) Exercise 1:
Let's complete the table below and calculate the average x :
Number of calls 0 1 2 3 4 5
xi
No. of minutes or 93 261 416 393 308 174
 34
ni 5,2 14,5 23,1 21,8 17,1 9,7
Frequency fi=
N
ni x i 0 261 832 1179 1232 870

Number of calls xi 6 7 8 Total

No. of minutes or 93 42 20 N=1800
n 5,2 2,3 1,1 100
Frequency fi= i
N
ni x i 558 294 160 p

∑ ni x i=¿5386
i=1
p

Therefore, we have
∑ ni x i 5386 = 2,992. Let x = 2,992.
x= i=1 =
N 1800

NB: If the modalities xi are classes, i.e. intervals, we use the centers ci of the classes.

When the statistical variable is continuous, the modalities being grouped in classes, by
convention, the centre of the classes is used for the calculation of the averages as the value
of the variable xi. The center of the class is the half-sum of the two limits of the interval.
xi+ xi+1
The center of the classes is defined by : Ci=
2
Note: when the observations are large numbers, to simplify the calculations, it is advisable
' xi−x 0
to perform a change of variables of the form : x i = .
a
xi is the variable under study; xi' is the auxiliary variable; a is the increase in the value of
the variable; x0 is a characteristic of the central tendency generally called the mode. This
change of variable is called change of origin and scale.
2- Geometric mean

The geometric mean, noted G, is widely used in the analysis of the evolution of a variable
over time. It is therefore used for everything that concerns the calculation of growth rates.
We can take the example of the growth rate of the turnover of a commercial company.
Like the arithmetic mean, a distinction is made between the simple geometric mean and the
weighted geometric mean. They are written respectively :
 Simple geometric mean.
G = √n ¿ ¿) = [(x 1 x 2 x 3 … .. x n )]1/n
 35
Note: The geometric mean is used, for example, when you want to calculate the average of
interest rates.
Example 1: The geometric mean of the numbers 8 and 12 is G = √2 8 x 12 = 9,8
Example 2. In a company X, the wage bill increases first by 10%, then by 15% and finally
by 30%. How much did it increase on average over the 3 periods?
Solution:
It is assumed that initially the base is 100.
1st
increase: 100 + 10% x100 = 1.1
2nd
increase: 100 + 15%x100 = 1.15
3rd
increase: 100 + 30%x100 = 1.3
G= √3 ( 1 ,1 ) ( 1 ,15 ) (1 ,3) = 1,1804. That is an average increase of 18.04% over the three
periods.
Note: the use of the arithmetic mean does not give a correct interpretation.
 Weighted geometric mean

It is defined by :
G = √ x 1n 1 +…+ xn nn = (x1n1 + x2n2 + ...+ xnnnn) 1/n = ¿ ni)1/n
n n
log G = 1/n log ¿ ) = 1/n ∑ log xi = 1/n ∑ ¿ log xi
ni ni

i=1 i=1
n
1
∑ xi log xi OR G = n
⇒ G=e n i=1 √ ¿ ¿ ¿ ¿)

G is equal to the exponential of the weighted arithmetic mean of the logarithms of

observations.
Example: a company's turnover increases by 5% in the first 2 years, 9% in the next 4 years
and 12% in the last year.
What is the annual increase in turnover for this company?
G = √ ( 1.05 )2 ¿ ¿ = 1.083 or 8.3% per year.
7

3- Harmonic average

The simple or weighted harmonic mean is used in the study of ratios such as speed, which
is expressed in kilometres travelled per hour. It is equal to the total number divided by the
average of the inverse of xi.
 Simple harmonic mean

It is the inverse of the mean of the observation inverses. It is defined by :

 36
1 n
n n
H= 1 =
∑
n i=1
1/ xi ∑ 1/ xi
i=1

Example : on a journey, we do 60 km/h on the way out and 30 km/h on the way back.
What is the average speed?
Solution:
1 1 1 1
= [ + ] → H = 40km/h
H 2 60 30
 Weighted harmonic average

1 n
n n
H= 1 =
∑
n i=1
¿∗1/ xi ∑ ¿/ xi
i=1

Note: It is wise to apply the harmonic mean to speeds.

Example: a reproduction workshop photocopied 1000 documents at a speed of 800
documents per hour and another workshop photocopied 2000 documents of the same
nature at a speed of 700 documents per hour.
What is the average speed for photocopying the 3000 documents?
n 3000
k
Solution: H = ni = 800 700 = 730.44 docs/h
∑ +
x i 1000 2000
i=1

4- Root mean square

It is described as the square root of the means of the squares of the observations. The
simple or weighted quadratic mean, noted Q, is very commonly used in physics (as is the
harmonic mean which we will define below).
 Simple quadratic mean

√ ∑x
2 2 2 2
x 1 + x +…+ x
Q= 2 = ( i=1 i )1/2
n
n
n
Example: let the series 1; 2; 3; 4; 5; 6 be;
Calculate its simple root mean square.

Q=
√ 1²+ 2 ²+3 ²+ 4 ²+5 ²+ 6 ²
6
=

 Weighted root mean square

 37

Q=
√ x 12 n1+ x22 n2 +…+ x k 2 nk
n
Example: let's say the following series.
xi 1 2 3 4 5 6 Calculate the root mean square of this series.
or 20 30 15 10 5 2
1
Q² = [(20x1²) + (30x2²) + (15x3²) + (10x4²) +
82
(5x5²) + (2x6²)] = 7.71
Q = 2,78
Remark: note that H ≤ G ≤ X ≤ Q. There is equality (H = G = X = Q) when xi is constant.
Comparison of Measures of Central Tendency
Regardless of the shape of the distribution of the statistical variable under study, it is
accepted that the following inequalities always hold: H ≤ G ≤ X ≤ Q.

III DISPERSION PARAMETERS.

Unlike central tendency indicators, which synthesize a large set of data, dispersion
indicators provide an objective answer to the question of the significance of the central
tendency indicator.
There are four main indicators of dispersion: range, interquartile ranges, variance and
absolute deviations. However, these indicators of dispersion are complementary.
III. 1 THE RANGE (E) OF A SERIES OR VARIATION INTERVAL
Definition:

The range is the difference between the extreme values of a statistical distribution, i.e.
the difference between the highest value in the distribution and the lowest value in the
distribution. It measures the spread of values around the median.
Consider a statistical series whose values are arranged in ascending order.

The difference between the extreme terms x1 and xn of the series, i.e. the number: e

= xn - x1, is called range e.

2) The Interquartilerange Q :
Definitions :
Consider a statistical series whose values are arranged in ascending order.
The interquartile range is the interval q1; q3 where q1= 1st
quartile and q3 = 3rd

quartile.
 38
The amplitude q of the interval q1; q3 is called the interquartile range, i.e. the
number q = Q3 - Q1.

Application exercise :
Student A received 8 marks which are: x1 = 5; x2 = 5; x3 = 6; x4 = 9; x5 = 10; x6 = 12; x7
= 13; x8 = 13.
Student B received 9 marks which are: x1 = 2; x2 = 3; x3 = 5; x4 = 6; x5 = 8; x6 = 9;
x7 = 9; x8 = 10; x9 = 10.

Solution:
Let's find the quartiles, interquartile range and interquartile range for each
student:
For student A (N= 8 =4x2): we have q1 = x N =x 8 =x 2= 5 and q3 = x 3 N = x 3 x 8 = x =¿ 12
6
4 4 4 4
Let q1 =5 and q3 = 12.

Interquartile range = Q3; Q1 = 5; 12 

Interquartile range = q = Q3 - Q1= 12 -5 =7. So q =7.
For student B ( N= 9 =4x2 +1): we have q1 = x N +1=x 9 +1=x 3= 5 and q3 = x 3 N +1=
4 4 4
x 3 x 9 = x =¿ 9 Let q1 =5 and q3 = 9.
+1 7
4
Interquartile range = Q3; Q1 = 5; 9 
Interquartile range = q = Q3 - Q1= 9 -5 =4. Let q =4.

III.2 VARIANCE AND STANDARD DEVIATION X OF A STATISTICAL SERIES

The variance and the mean, unlike other positional characteristics, are fundamental tools in
the study of the distribution of a statistical variable. We will present successively the
variance, the standard deviation and the coefficient of variation.
1- The variance

The variance is an indicator of the dispersion of a series with respect to its mean. It can be
defined as the sum of the squares of the deviations from the mean divided by the number of
observations. Its formula is the following:
The variance is the weighted sum of the squares of the deviations of the values in the
series from the mean.

Or
δx
=
√ V (x)
n n
1
V (X) = S2x = ∑ ¿ ¿ )2 = ∑ fi(xi−X ¿)¿ 2 KÖENIG's formula
n i=1 i=1

NB:
 39
The variance V(x) and the standard deviationx measure the dispersion of values
around the mean x .
The standard deviation x compares the dispersion of 2 series. Unlike the
interquartile range q and the interdecile range d which take into account the whole
population.

A series is said to be more homogeneous (i.e. its values are less dispersed) than the
other series if and only if its standard deviation is lower. In other words, a series that is
not very dispersed (with values grouped around the mean value) will have a rather low
standard deviation.

Ownership:

4) Coefficient of variation
The CV makes it possible to appreciate the representativeness of the average in relation to
all the observations. It gives a good idea of the degree of homogeneity of a series. It should
be as low as possible (<15% in practice).

IV. FORMAL CHARACTERISTICS

In addition to the position and dispersion characteristics, which are the most commonly
used in practice (and in theory as well), there are so-called shape characteristics that allow
the analysis of a statistical distribution to be refined. In practice, these skewness and
kurtosis characteristics are rarely used. It is still useful to know them.

IV.1 MEASUREMENT OF ASYMMETRY

A distribution is symmetrical if the observations identified by the frequencies are equally
dispersed on either side of a central value: the mode, the median or the mean, which are all
the same. Otherwise, the distribution is asymmetric. Several coefficients are used to
measure the symmetry of a distribution.

1- The first Pearson coefficient

X−Mo
It is denoted β1 and defined by: β1 =
σ
2- The second Pearson coefficient

(X ¿−Me )
It is defined by; β1 = 3 ¿
σ
3- The third Pearson coefficient
 40
It is defined by: β1 = μ32/ μ23with μ3 = ∑ f i( x i - x )3 ; μ2 = ∑ f i( x i - x )²

4- The Yule coefficient

Q1+Q 3−2 Me
It is defined by : S ¿
Q 3−Q 1
5- The coefficient of dissymmetry of YULE
The Yule coefficient of skewness, noted S, allows us to appreciate the symmetry or
skewness of the frequency polygon of a statistical distribution.

( Q3−M e )−(M e −Q1)

It is written: S =
( Q3−M e ) +( M e−Q 1)
Depending on the value of the dissymmetry coefficient S, three cases are possible:
 S = 0, the statistical distribution is symmetric.
 S > 0, the statistical distribution is spread to the right.
 S < 0, the statistical distribution is skewed to the left.

Note: S is only valid for distributions with low skewness

6- The Fisher coefficient

It is defined by: γ1 = μ3/ σ3
This coefficient is dimensionless and invariant to any change of origin.
 If the distribution is symmetric, μ3 = 0, and thus γ = 0.
 If the distribution is right-skewed (spread to the right), then γ > 0.
 If the distribution is left-skewed (left-spread), then γ < 0.

Note:
- If β1= S = γ1 = 0, the distribution is perfectly symmetric, then, M0 = Me = X
- If β1 > 0, S > 0, γ1 > 0, the distribution is right-spread, then, M0< Me< X
- If β1< 0, S < 0, γ1 < 0, the distribution is left-spread, then, X < Me< M0 or
M0> Me> X

Graphically, we have :
All skewness coefficients have the same properties, they are zero if the distribution is
symmetric, negative if the distribution is left asymmetric, and positive if the distribution is
right asymmetric as shown in the following figure:
Figure: Asymmetry of a distribution
 41
Note: Some variables are always very asymmetric on the right, such as income, firm size,
or municipality size. A simple method to make a variable symmetrical is to take the
logarithm of this variable.

IV.2 MEASUREMENT OF KURTOSIS

The degree of kurtosis of a frequency curve is measured in relation to the frequency of a
normal distribution of the same mean. The most commonly used coefficient is Fisher's
2
coefficient: γ2 = (μ4/ σ4) - 3 where σ 4 = μ2
- If γ2 = 0, the distribution is normal, we say it is mesokurtic
- If γ2 > 0, the distribution is less flattened than the normal distribution (the
distribution is hyper normal), it is said to be leptokurtic.
- If γ2 < 0, the distribution is flatter than the normal distribution (distribution is
hypo-normal), it is said to be platykurtic.

NB: one can also use the Pearson coefficient. The Pearson kurtosis is expressed by the
kurtosis of the dispersion. The kurtosis coefficient is equal to :
μ4 μ4
β2 = 2 = 4 with σ 4 = μ22
μ2 σ
 42
- If β2 = 0, then the distribution is normal or mesokurtic.
- If β2 > 0, then the distribution is sharper than the normal distribution, it is said to
be leptokurtic.
- If β2 < 0, then the distribution is flatter than the normal distribution, so it is
platykurtic.

Example: In the following figure, an example of two distributions with the same mean and
variance is shown. The sharper distribution is leptokurtic, the other is mesokurtic. The
leptokurtic distribution has a thicker tail.

Figure: Mesokurtic and Leptokurtic distributions

Application example:

Let be the following series: x i ∈ {0 ; 1 ; 2 ; 3}

xi 0 1 2 3
fi 0,216 0,432 0,288 0,064

1. Determine the Pearson and Ficher skewness and kurtosis coefficient.

2. Deduce the nature of the series and the shape of the frequency curves.

Solution:
fi f i xi f i xi
2
xi - x f i( x i - x )² f i( x i - x )3 f i( x i - x )4
xi
0 0,216 0 0 -1,2 0,31104 -0,373 0,44789
1 0,432 0,432 0,432 -0,2 0,01728 -0,00345 6,91x10-4
2 0,288 0,576 1,152 +0,8 0,18432 +0,1474 0,11796
3 0,064 0,192 0,576 +1,8 0,20736 +0,373 0,671846
Totals 1 x = 1,2 2,16 0,72 0,144 1,238

1. Calculation of Pearson's and Ficher's skewness coefficients.

2
μ3
β 1 (Pearson) = 3 with μ3 = ∑ f i( x i - x )3 ; μ2 = ∑ f i ( x i - x )²
μ2
 43
β 1 = 0,05
μ3
γ 1 (File) = 3/ 2 with μ3= 0,144 ; μ2= 0,72 ;
μ2
γ 1= 0,24
2. Calculation of Pearson and Ficher kurtosis coefficients

μ4
β 2 (Pearson) = 2 with μ4 = 1,238 ; μ2= 0,72
μ2
β 2= 2,39
γ 2 (File) = β 2 - 3 = 2,39 - 3 = -0,611

Interpretation:
- This series is skewed to the left in the case of the asymmetric Ficher coefficient.
- In the case of Ficher kurtosis coefficients, the frequency curve is hypo-normal.

V APPLICATION EXERCISES

EXERCISE 1:
In a high school, 32 students were randomly selected from the senior classes. They
were asked to specify the series in which they would sit the baccalaureate: Philosophy and
literature (A); Economics and social sciences (B); Mathematics and physical sciences (C);
Mathematics and biology (D).
The results are as follows:
A-D-D-A-A-D-D-A-D-D-D-C-D-D-A-A-A-A-B-A-A-D-D-D
- C - C - B.
For this statistical series (xi; ni),
(a) Give the population, the individual, the characteristic, the list of modalities and the
total number of individuals.
(b) Is it a qualitative or quantitative characteristic? Discrete or continuous? Why is it
so?
(c) Organize these data in a table showing modalities, numbers, CTAs and DCEs,
frequencies, CCFs and CDFs.

EXERCISE 2
The distances xi (in thousands of km) travelled by each of the 150 cars of a car rental
company are recorded.
The results of this survey are :
Distances xi 0 ; 5  5 ; 7  7 ; 9  9 ; 15  Total
Staffing and 15 78 36 21 150
1) Let's build the histogram, and the polygon of the numbers.
2) Calculate the centre, amplitude, cumulative decreasing number, density of each
class, median, mode, mean, variance and standard deviation of this distribution.
 44
EXERCISE 3
1) The number of children in 133 families is counted, the data is grouped as follows:
(a) Gross table :
001111111122222222223333333333
333333333333333333333333333333
333333333333444444444444444444
444444455555555555555666666666
6 6 6 6 6 6 6 6 7 7 9 9 10

(b) Ordered and counted table :

Number of 0 1 2 3 4 5 6 7 8 9 10 Total
children xi
Staffing and 2 8 10 52 25 14 17 2 0 2 1 133
ECC
EDC
n
Frequencies fi= i
N
FCC
FCD
Determine the mode, median, mean, range, variance and standard deviation of
this distribution.

2) The age at admission of 100 patients to the hospital is recorded, the data are
grouped as follows:
(a) Gross table :
10 22 24 42 37 77 89 85 28 63 9 10 7 51 2 1 52 7 48 54
32 29 2 15 46 48 39 6 72 14 36 69 40 61 12 21 54 53 58 32
27 31 1 25 22 6 81 11 56 5 63 53 88 48 52 87 71 51 53 33
46 33 85 22 5 87 28 2 85 61 16 42 69 7 10 53 33 3 85 8
51 60 58 9 14 74 24 87 7 81 30 76 7 6 27 18 17 53 70 49

(b) Ordered and counted table :

Number of 0 ; 10  10 ; 20  20 ; 30  30 ; 40  40 ; 50 
children xi
Staffing and
ECC
EDC
n
Frequencies fi= i
N
FCC
FCD

Number of 50 ; 60  60 ; 70  70 ; 80  80 ; 90  Total

children xi
Staffing and
ECC
 45
EDC
ni
Frequencies fi=
N
FCC
FCD
TAF: Complete the above tables

CHAPTER 3: BIVARIATE DESCRIPTIVE STATISTS

INTRODUCTION TO THE CHAPTER :

(I) INTRODUCTION.
1) Position of the Problem.
2) Solution to the Problem: Find the Correlation between xi and yj.
II) THE NUAGE OF POINTS Mij (xi; yj).
1) Definitions.
2) The Mean Point G of a Two Variable Statistical Series.
3) Representation of a Scatterplot Mi (xi; yi).

III) THE xy COVARIANCE OF A DOUBLE SERIES (x ; y).

1) Definition.
2) Other Expression of ❑xy .
IV) THE AFFINE ADJUSTMENT OF A DOUBLE SERIES (x ; y).
1) Adjustment.
2) Least Squares Refined Fitting.
V) THE COEFFICIENT r OF LINEAR CORRELATION OF A DOUBLE SERIES
(x; y).
1) Definition.
2) Expression of the linear correlation coefficient r.
3) Properties of the linear correlation coefficient r.
 46
VI) EXERCISES.

Objectives:

The general objective of this chapter is to characterize bivariate descriptive statistics.

Specifically, this chapter should allow us to better understand the following concepts:
 Marginal laws and conditional laws ;
 Independence and functional links;
 Correlations.
I- Bivariate statistical series
We are interested in two variables x and y. These two variables are measured on the n units
of observation. For each unit, we obtain two measurements. The statistical series is then a
sequence of n pairs of values taken by the two variables on each individual:
(x1, y1), . . . (xi, yi), . . . (xn, yn).
Each of the two variables can be either quantitative or qualitative. Two cases are examined.
- Both variables are quantitative.
- Both variables are qualitative.

1- Two quantitative variables

1.1- Graphical representation of two variables

The two-character statistical series

( x i , y j )1≤i≤n is represented in a plane where the values of
x are plotted on the x-axis and those of y on the y-axis. This representation constitutes the
cloud of points of the sample. In this case, each pair is composed of two numerical values.
A pair of numbers (integer or real) can always be represented as a point in a plane (x1, y1),
. (xi, yi), . . . (xn, yn).

Example: We measure the weight Y and height X of 20 individuals

yi xi yi xi
60 155 75 180
61 162 76 175
64 157 78 173
67 170 80 175
68 164 85 179
 47
69 162 90 175
70 169 96 180
70 170 96 185
72 178 98 189
73 173 101 187

Figure: the point cloud

The problem with bivariate statistical series is mainly that of the relationship or otherwise
between the variables.

I - Position of the problem. Vocabulary

For the sake of clarity, this chapter is based on the following example:

Exercise: The following table shows the evolution of the membership of a rugby club from
2001 to 2006.

The aim is to study this two-variable statistical series (rank and membership) in order to
predict the evolution of membership in the following years.

I.1 Point cloud

The first step is to make a graph that translates the two statistical series above.
 48
Definition 1

Let X and Y be two numerical statistical variables observed on n individuals. In an

orthogonal reference frame (O;i;j ), the set of n points with coordinates (xi, yi) forms the
point cloud associated with this statistical series.

In our example, if we place the rank on the abscissa, and the number of members on the
ordinate, we can represent each value by a point. We thus obtain a succession of points,
whose coordinates are (1; 70), (2; 90), ... (6; 220), form a point cloud.

Question 1

In a plane with an orthogonal reference frame and graphical units: 2 cm for one year on the
x-axis and 1 cm for 20 members on the y-axis, represent the cloud of points associated with
the series (xi; yi).

I.2 The problem of adjustment

 49
The scatterplot associated with a bivariate statistical series therefore immediately gives
qualitative information.

In order to obtain more quantitative information, we need to pose the problem of

adjustment. The plot highlights the possibility of "recognizing" graphically the possibility
of a functional relationship between the two observed quantities (here rank and number of
members). The problem of establishing a functional relationship between the two series is
the problem of adjustment.

I.3 Average point

Definition 2

Let be a statistical series with two variables, X and Y, whose values are pairs (xi; yi). We
call the mean point of the series the point G of coordinates.

Question 2
Determine the coordinates of the following mean points:
 G1 for the years 2001 to 2003,
 G2 for the years 2004 to 2006,
 G, mean point of the entire point cloud.

Solution:

G1 (2 ; 91, 7)
G2 (5 ; 176, 7)
G (3,5 ; 134, 2)
 50
II Adjustments
II.1 Adjustment to the rule

The aim is to use the results obtained to make forecasts for the years to come. One way to
do this is to draw a line (D) as close as possible to the points of the cloud and to find the
equation of the type y = ax + b.

II.2 Mayer's method

This adjustment consists in determining the line passing through two average points of the
point cloud.

Question 3

Determine the equation of the line D1 which passes through the mean points G1and G2
and plot it on the previous graph. The line D1 is not parallel to the ordinate axis, so its
equation is y = ax + b with :

Solution:

yG2− yG1
a= =28 , 3
xG2−xG2

yG1= axG1+ b ⇒ 91, 7 = 28.3 x 2 + b ⇒ b = 35, 1.

Moreover, it passes through the point G1 (2; 91.7) from which :

Conclusion : D1 : y = 28.3x + 35.1.

To draw D1, simply place G1 and G2 and draw the line that connects them.

II.3 Least squares method

The aim is to obtain a line that is equidistant from the points on either side of it. To achieve
this, we try to minimize the sum of the distances of the points to the squared line. We
consider a statistical series with two variables represented by a cloud justifying an affine
adjustment.

Definition 3
 51
In the plane provided with an orthogonal reference frame, we consider a cloud of n points
of coordinates (xi; yi). The line D of equation y = ax + b is called the regression line from
y to x of the statistical series if the following quantity is minimal:

Note 1
It would be just as well to look at the line D/ which minimizes the quantity :

This line is called the regression line from x to y.

Definition 4
The covariance of the dual statistical series of variables x and y is the real number :
 52
Property 1
The regression line D from y to x has the equation y = ax + b where :

Remark 3 : The real numbers a and b are given by the calculator.

Property 2 : The mean point G of the cloud always belongs to the regression line from y
to x.

Question 4

Determine an equation of the least squares fit line D2 of y to x and plot it on the previous
graph.
The calculator gives D2: y = ax + b with a = 29 and b = 32.7

II.4 Comparison

With the last three adjustments, we can assess what will happen later, let's compare them:
Question 8
Assuming that the adjustments remain valid for subsequent years, give an estimate of the
number of members in 2007 using all three methods. In all cases, calculate y when x
corresponds to the year 2007, i.e. rank 7.

 Mayer's method: y = 28.3 (7) + 35.1 = 233.2, i.e. approximately 233 members.
 Affine adjustment: y = 29 (7) + 32.7 = 235.7 or approximately 236 members.

III- Linear correlation coefficient

Definition 5
 53
The linear correlation coefficient of a statistical series of variables x and y is the number r
defined by :

This coefficient is used to measure the quality of an affine fit.

Graphical interpretation :
The closer the linear regression coefficient is to 1 in absolute value, the better the linear fit.
When r = ±1, the regression line passes through all the points of the cloud, which are
therefore aligned.
Question 10
Determine the linear correlation coefficient for the affine fit (between x and y) and then the
exponential fit (between x and z). Which fit is the most accurate?

Using the calculator, we find successively r2 = 0.987 and r3 = 0.999. This is consistent
with what we had deduced earlier, namely that the exponential fit is more reliable for this
case

Property 3
The linear correlation coefficient r checks -1 < r <1.

Property4 : Relationship between the regression coefficients and the linear correlation
coefficient
This relationship is as follows:
δx
r=a
δy
' δy
r=a
r2=a a ⇒|r|=√ aa '
δx
'

The sign of r depends on the sign of the covariance of the variables X and Y (Cov(X,Y))
III.1) The Distribution or Marginal Laws.

1) The Staffing Table :

(a) Definition :
Let x1, x2, ..., xm be the values taken by the character x and y1, y2, ..., yp the values taken
by the character y of a double series
 54

The following double entry or contingency table is called a staffing table:

y
x y1 y2 ……… yj ……… yp

x1 n11 n12 ……… n1j ……… n1p

x2 n21 n22 ……… n2j ……… n2p

. . . . . . .
. . . . . . .
. . . . . . .

xi ni1 ni2 ……… nij ……… nip

. . . . . . .
. . . . . . .
. . . . . . .

xm nm1 nm2 ……… nmj ……… nmp

Where nij = number of employees in the pair (xi; yj) with i = row number and j = column
number.
b) Example:
Consider the following double series:
y= ages
x= (in years) y1=14 y2=15 y3=16 y4=17 y5=18
class centres
sizes (in m)

x1=1,5 1 2 1 0 1

x2=1,6 5 81 50 40 4

x3=1,7 10 120 161 130 6

x4=1,8 0 1 8 10 9

For example, n23 = 50 means that there are 50 16-year-olds between 1.55m and 1.65m tall.
 55
2) The Marginal Distribution :
(a) Definitions :
In a table of the numbers of 2 characters x (in rows) and y (in columns) :
A table margin is a completed row or column giving the total of the row or column,
also called the "total" row or "total" column.
The marginal distribution according to characteristic x is the "total" column: L1,
L2, ..., Lm.
The marginal distribution according to characteristic y is the "total" line: ,
C1 C2 , ...,
Cp.
NB:
The marginal distribution allows to study only one of the 2 characters x and y, the other
not being involved.
This amounts to a one-variable statistical series.
y
x y1 y …… yj …… yp Total" column
2

x1 n11 n12 …… n1j …… n1p L1

x2 n21 n22 …… n2j …… n2p L2

. . . . . . . .
. . . . . . . .
. . . . . . . .

xi ni1 ni2 …… nij …… nip Li

. . . . . . . .
. . . . . . . .
. . . . . . . .

xm nm1 nm …… nmj …… nm Lm
2 p

Total" C1 C2 …… Cj …… Cp N= Total
line number

Where Li = ni1 + ni2 +...+ nip; Cj = n1j + n2j +...+ nmj and N = C1 + C2 +...+ Cp = L1 + L2 +...+
Lm = Total population.
 56

b) Example: Let's take the following double series:

y= ages
x= (in years) y1=14 y2=15 y3=16 y4=17 y5=18 Total"
class centres column
taiies (in m)
x1=1,5 1 2 1 0 1 L1= 5

x2=1,6 5 81 50 40 4 L2= 180

x3=1,7 10 120 161 130 6 L3= 427

x4=1,8 0 1 8 10 9
L4= 28

Total" line C1=16 C2=204 C3=220 C4=180 C5=20 N=640

III.2) Marginal frequencies.

1) Frequency fij of a pair (xi; yj).
nij
a) Definition: We have fij = .
N
b) Example: Let's complete the table below:
y= ages
x= (in years) y1=14 y2=15 y3=16 y4=17 y5=18 Total"
class centres column
taiies (in m)
x1=1,5 1 2 1 0 1 L1= 5
f11=… f12=…. f13=… f14=… f15=….
. . .
x2=1,6 5 81 50 40 4 L2= 180
f21=… f22=…. f23=… f24=… f25=….
. . .
x3=1,7 10 120 161 130 6 L3= 427
f31=… f32=…. f33=… f34=… f35=….
. . .
x4=1,8 0 1 8 10 9
f41=… f42=…. f43=… f44=… f45=…. L4= 28
. . .

Total" line C1=16 C2=204 C3=220 C4=180 C5=20 N=640

2) The Marginal Frequencies.

 57
(a) Definitions :
Li
The marginal frequency fxi of the variable x is the number: fxi = . .
N
Cj
The marginal frequency fyj of the variable y is the number: fyj = .
N
b) Example: Let's complete the table below
y Marginal
x y1=1 y2=15 y3=16 y4=17 y5=18 Total" frequency
4 column L
fxi = i
N
x1=1,5 1 2 1 0 1 L1= 5

x2=1,6 5 81 50 40 4 L2= 180

x3=1,7 10 120 161 130 6 L3= 427

x4=1,8 0 1 8 10 9
L4= 28
Total" line C1=16 C2=204 C3=220 C4=180 C5=20 N=640

Marginal Total
frequency frequency
C f=1
fyj = j
N