page 1 110SOR201(2002)

Chapter 1

Basic Probability Concepts

1.1 Sample and Event Spaces

1.1.1 Sample Space

A probabilistic (or statistical) experiment has the following characteristics:

(a) the set of all possible outcomes of the experiment can be described;

(b) the outcome of the experiment cannot be predicted with certainty prior to the performance
of the experiment.

The set of all possible outcomes (or sample points) of the experiment is called the sample space
and is denoted by S. For a given experiment it may be possible to define several sample spaces.
Example For the experiment of tossing a coin three times, we could define

(a) S = {HHH,HHT,HTH,THH,HTT,THT,TTH,TTT},
each outcome being an ordered sequence of results; or

(b) S = {0, 1, 2, 3},

each outcome being a possible value for the number of heads obtained.

If S consists of a list of outcomes (finite or infinite in number), S is a discrete sample space.

Examples

(i) Tossing a die: S = {1, 2, 3, 4, 5, 6}

(ii) Tossing a coin until the first head appears: S = {H,TH,TTH,TTTH, . . .}.

Otherwise S is an uncountable sample space. In particular, if S belongs to a Euclidean space

(e.g. real line, plane), S is a continuous sample space.
Example Lifetime of an electronic device: S = {t : 0 ≤ t < ∞}.
 page 2 110SOR201(2002)

1.1.2 Event Space

Events

A specified collection of outcomes in S is called an event: i.e., any subset of S (including S

itself) is an event. When the experiment is performed, an event A occurs if the outcome is a
member of A.
Example In tossing a die once, let the event A be the occurrence of an even number: i.e.,
A = {2, 4, 6}. If a 2 or 4 or 6 is obtained when the die is tossed, event A occurs.
The event S is called the certain event, since some member of S must occur. A single outcome
is called an elementary event. If an event contains no outcomes, it is called the impossible or
null event and is denoted by ∅.

Combination of events

Since events are sets, they may be combined using the notation of set theory: Venn diagrams
are useful for exhibiting definitions and results, and you should draw such a diagram for each
operation and identity introduced below.
[In the following, A, B, C, A1 , ..., An are events in the event space F (discussed below), and are
therefore subsets of the sample space S.]
The union of A and B, denoted by A ∪ B, is the event ‘either A or B, or both’.

The intersection of A and B, denoted by A ∩ B, is the event ‘both A and B’.

The union and intersection operations are commutative, i.e.

A ∪ B = B ∪ A, A ∩ B = B ∩ A, (1.1)
associative, i.e.
A ∪ (B ∪ C) = (A ∪ B) ∪ C = A ∪ B ∪ C, A ∩ (B ∩ C) = (A ∩ B) ∩ C = A ∩ B ∩ C, (1.2)
and distributive:
A ∩ (B ∪ C) = (A ∩ B) ∪ (A ∩ C), A ∪ (B ∩ C) = (A ∪ B) ∩ (A ∪ C). (1.3)

If A is a subset of B, denoted by A ⊂ B, then A ∪ B = B and A ∩ B = A.

The difference of A and B, denoted by A\B, is the event ‘A but not B’.

The complement of A, denoted by A, is the event ‘not A’.

The complement operation has the properties:

A ∪ A = S, A ∩ A = ∅, (A) = A, (1.4)
and A ∪ B = A ∩ B, A ∩ B = A ∪ B. (1.5)
Note also that A\B = A ∩ B. Hence use of the difference symbol can be avoided if desired
(and will be in our discussion).
 page 3 110SOR201(2002)

Two events A and B are termed mutually exclusive if A ∩ B = ∅.

Two events A and B are termed exhaustive if A ∪ B = S.

The above results may be generalised to combinations of n events: thus

A1 ∪ A2 ∪ ... ∪ An is the event ‘at least one of A1 , A2 , ..., An ’

A1 ∩ A2 ∩ ... ∩ An is the event ‘all of A1 , A2 , ..., An ’

A1 ∩ (A2 ∪ A3 ∪ ... ∪ An ) = (A1 ∩ A2 ) ∪ (A1 ∩ A3 ) ∪ ... ∪ (A1 ∩ An ) (1.6)

A1 ∪ (A2 ∩ A3 ∩ ... ∩ An ) = (A1 ∪ A2 ) ∩ (A1 ∪ A3 ) ∩ ... ∩ (A1 ∪ An ) (1.7)

Sn Tn
A1 ∪ A2 ∪ ... ∪ An = A1 ∩ A2 ∩ ... ∩ An or i=1 Ai = i=1 Ai (1.8)

Tn Sn
A1 ∩ A2 ∩ ... ∩ An = A1 ∪ A2 ∪ ... ∪ An or i=1 Ai = i=1 Ai (1.9)

(The last two results are known as de Morgan’s Laws - see e.g. Ross for proofs.)
The events A1 , A2 , ..., An are termed mutually exclusive if Ai ∩ Aj = ∅ for all i 6= j.

The events A1 , A2 , ..., An are termed exhaustive if A1 ∪ A2 ∪ ... ∪ An = S.

If the events A1 , A2 , ..., An are both mutually exclusive and exhaustive, they are called a
partition of S.

Event space

The collection of all subsets of S may be too large for probabilities to be assigned reasonably
to all its members. This suggests the concept of event space.
A collection F of subsets of the sample space is called an event space (or σ-field) if
(a) the certain event S and the impossible event ∅ belong to F;

(b) if A ∈ F, then A ∈ F;

(c) if A1 , A2 , . . . ∈ F, then A1 ∪ A2 ∪ . . . ∈ F, i.e. F is closed under the operation of taking

countable unions.
(Note that one of the statements in (a) is redundant, in view of (b).)
 page 4 110SOR201(2002)

∞ ∞
It is readily shown that, if A1 , A2 , . . . ∈ F, then Ai ∈ F. For (Ai ) ∈ F (invoking
T T S
Ai =
i=1 i i=1
properties (b) and (c) of F, and the result follows from property (b).
For a finite sample space, we normally use the collection of all subsets of S (the power set of
S) as the event space. For S = {−∞, ∞} (or a subset of the real line), the collection of sets
containing all one-point sets and all well-defined intervals is an event space.

1.2 Probability Spaces

1.2.1 Probability Axioms

Consideration of the more familiar ‘relative frequency’ interpretation of probability suggests

the following formal definition:
A function P defined on F is called a probability function (or probability measure) on (S, F ) if
(a) 0 ≤ P(A) ≤ 1 for all A ∈ F;

(b) P(∅) = 0 and P(S) = 1;

(c) if A1 , A2 , . . . is a collection of mutually exclusive events ∈ F, then

P(A1 ∪ A2 ∪ . . .) = P(A1 ) + P(A2 ) + · · · (1.10)

Note:

(i) There is redundancy in both axioms (a) and (b) - see Corollaries to the Complementarity
Rule (overleaf).

(ii) Axiom (c) states that P is countably additive. This is a more extensive property than
being finitely additive, i.e. the property that, if A 1 , A2 , ...An is a collection of n mutually
exclusive events, then
n
X
P(A1 ∪ A2 ∪ . . . An ) = P(Ai ). (1.11)
i=1

This is derived from axiom (c) by defining A n+1 , An+2 , ... to be ∅: only (1.11) is required in
the case of finite S and F. It can be shown that the property of being countably additive
is equivalent to the assertion that P is a continuous set function, i.e. given an increasing
∞
sequence of events A1 ⊆ A2 ⊆ ... and writing A = ∪ Ai = lim Ai , then P(A) = lim P(Ai )
i=1 i→∞ i→∞
(with a similar result for a decreasing sequence) - see e.g. Grimmett & Welsh §1.9

The triple (S, F, P) is called a probability space.

The function P(.) assigns a numerical value to each A ∈ F; however, the above axioms (a)-(c)
do not tell us how to assign the numerical value – this will depend on the description of, and
assumptions concerning, the experiment in question.
If P(A) = a/(a + b), the odds for A occurring are a to b and the odds against A occurring are
b to a.
 page 5 110SOR201(2002)

1.2.2 Derived Properties

From the probability axioms (a)-(c), many results follow. For example:

(i) Complementarity rule If A ∈ F, then

P(A) + P(A) = 1. (1.12)

Proof S = A ∪ A - the union of two disjoint (or mutually exclusive) events.

So P(S) = 1 = P(A) + P(A) by probability axiom (c). ♦
Corollaries
(a) Let A = S. Then P(S) + P(∅) = 1, so

P(S) = 1 ⇒ P(∅) = 0

so axiom (b) can be simplified to P(S) = 1.

(b) If A ⊂ S, then P(A) ≥ 0 ⇒ P(A) ≤ 1, so in axiom (a) only P(A) ≥ 0 is needed.
(ii)
If A ⊂ B , then P(A) ≤ P(B). (1.13)

Proof P(B) = P(A ∪ (B ∩ A)) = P(A) + P(B ∩ A) using axiom (c). The result follows
from the fact that P(B ∩ A) ≥ 0 (axiom (a)). ♦
(iii) Addition Law (for two events) If A, B ∈ F, then

P(A ∪ B) = P(A) + P(B) − P(A ∩ B). (1.14)

Proof We have that
A ∪ B = A ∪ (B ∩ A)
and B = (A ∩ B) ∪ (B ∩ A).
In each case, the r.h.s. is the union of two mutually exclusive events ∈ F . So by axiom
(c)
P(A ∪ B) = P(A) + P(B ∩ A)
and P(B) = P(A ∩ B) + P(B ∩ A).
Eliminating P(B ∩ A), the result follows. ♦
Corollaries Boole’s Inequalities If A, B ∈ F, then it follows from (1.14) and axiom
(a) that
(a) P(A ∪ B) ≤ P(A) + P(B); (1.15a)

(b) P(A ∩ B) ≥ P(A) + P(B) − 1. (1.15b)

(iv) Generalized Addition Law (for n events) If A 1 , A2 , . . . , An ∈ F, then the probability

that at least one of these events occurs is
n
X X
P(A1 ∪ A2 ∪ . . . ∪ An ) = P(Ai ) − P(Ai ∩ Aj )
i=1 1≤i<j≤n
X (1.16)
+ P(Ai ∩ Aj ∩ Ak ) − ..........
1≤i<j<k≤n
+(−1)n+1 P(A1 ∩ A2 ∩ . . . An ).
This is sometimes termed the inclusion-exclusion principle. One proof is by induction on
n: this is left as an exercise. For a non-inductive proof, see Ross.
 page 6 110SOR201(2002)

(v) Bonferroni’s Inequalities

If A1 , A2 , . . . , An ∈ F, then
n
X X n
X
(a) P(Ai )− P(Ai ∩ Aj ) ≤ P(A1 ∪ . . . ∪ An ) ≤ P(Ai ); (1.17a)
i=1 i<j i=1

n
X n
X
(b) P(A1 ∩ A2 ∩ . . . ∩ An ) ≥ P(Ai ) − (n − 1) = 1− P(Ai ). (1.17b)
i=1 i=1

Again, proofs are by induction on n (see Examples Sheet 1). [(1.17a) can be generalized.]

1.3 Solving Probability Problems

The axioms and derived properties of P provide a probability ‘calculus’, but they do not help in
setting up a probability model for an experiment or in assigning actual numerical values to the
probabilities of specific events. However, the properties of P may prove very useful in solving
a problem once these two steps have been completed.
An important special case (the ‘classical’ situation) is where S is finite and consists of N equally
likely outcomes E1 , E2 , . . . , EN : thus, E1 , . . . , EN are mutually exclusive and exhaustive, and
P(E1 ) = . . . P(EN ). Then, since

1 = P(S) = P(E1 ∪ . . . ∪ EN ) = P(E1 ) + · · · + P(EN ),

we have
1
P(Ei ) = , i = 1, . . . , N. (1.18)
N

Often there is more than one way of tackling a probability problem: on the other hand, an
approach which solves one problem may be inapplicable to another problem (or only solve it
with considerable difficulty). There are five general methods which are widely used:

(i) Listing the elements of the sample space S = {E 1 , . . . , EN }, where N is finite (and
fairly small), and identifying the ‘favourable’ ones. Thus, if the event of interest is A =
{Ei1 , . . . , EiN (A) }, then
P(A) = P(Ei1 ) + · · · + P(EiN (A) ).

If the outcomes can be assumed to be equally likely, this reduces to

N (A)
P(A) = ; (1.19)
N
i.e.
number of outcomes in A
P(A) = .
total number of outcomes
In such calculations, symmetry arguments may play an important role.

(ii) Enumeration, when S is a finite sample space of equally likely outcomes. To avoid
listing, we calculate the numbers N (A) and N by combinatorial arguments, then again
use P(A) = N (A)/N .

(iii) Sequential (or Historical) method. Here we follow the history of the system (actual or
conceptual) and use the multiplication rules for probabilities (see later).
 page 7 110SOR201(2002)

(iv) Recursion. Here we express the problem in some recursive form, and solve it by recursion
or induction (see later).

(v) Define suitable random variables e.g. indicator random variables (see later).

In serious applications, we should also investigate the robustness (or sensitivity of the results)
with respect to small changes in the initially assumed probabilities and assumptions in the
model (e.g. assumed independence of successive trials).

1.4 Some Results in Combinatorial Analysis

As already observed, in many problems involving finite sample spaces with equally likely out-
comes, the calculation of a probability reduces to problems of combinatorial counting. Also, it
is common for problems which at first appear to be quite different to actually have identical
solutions: e.g. (see §1.6.2 below)

(i) calculate the probability that all students in a class of 30 have different birthdays;

(ii) calculate the probability that in a random sample of size 30, sampling with replacement,
from a set of 365 distinct objects, all objects are different.

In this course, we will not be doing many such examples, but from time to time combinatorial
arguments will be used, and you will be expected to be familiar with the more elementary
results in combinatorial analysis. A summary of the different contexts in which these can arise
is provided below.
All the following results follow from simple counting principles, and it is worth checking that
you can derive them.

Sampling

Suppose that a sample of size r is taken from a population of n distinct elements. Then

(i) If sampling is with replacement, there are n r different ordered samples.

(ii) If
sampling is without replacement, there are n!/(n − r)! different ordered samples and
n
r = n!/{r!(n − r)!} different non-ordered samples.

(A random sample of size r from a finite population is one which has been drawn in such a
way that all possible samples of size r have the same probability of being chosen. If a sample
is drawn by successively selecting elements from the population (with or without replacement)
in such a way that each element remaining has the same chance of being chosen at the next
selection, then the sample is a random sample.)

Arrangements

(i) The number of different arrangements of r objects chosen from n distinct objects is
n!/(n − r)!. In particular, the number of arrangements of n distinct objects is n!.

(ii) The number of different arrangements of n objects of which n 1 are of one kind, n2 of a
second kind,...,nk of a kth kind is n!/{n1 !n2 !...nk !}, where ki=1 ni = n
(known as the
P

multinomial coefficient). In particular, when k = 2, this reduces to nn1 , the binomial

coefficient.
 page 8 110SOR201(2002)

Sub-populations

(i) The number of ways in which a population of n distinct elements can be partitioned into
k sub-populations with n1 elements in the first, n2 elements in the second,..., nk elements
in the kth, so that ki=1 ni = n, is the multinomial coefficient n!/{n 1 !n2 !...nk !}.
P

(ii) Consider a population of n distinct elements which is partitioned into k sub-populations,

where the ith sub-population contains n i elements. Then the number of different non-
ordered samples of size r, sampling without replacement, such that the sample contains r 1
elements from the first sub-population, r 2 elements from the second,..., rk elements from
the kth, is
k k
! ! !
n1 n2 nk X X
··· , ni = n, ri = r
r1 r2 rk i=1 i=1

1.5 Matching and Collection Problems

1.5.1 Matching Problems

Example 1.1
Suppose n cards numbered 1, 2, ...n are laid out at random in a row. Define the event
Ai : card i appears in the ith position of the row.
This is termed a match (or coincidence or rencontre) in the ith position. What is the probability
of obtaining at least one match?
Solution First, some preliminary results:

(a) How many arrangements are there

in all? — n!.
with a match in position 1? — (n − 1)! (by ignoring position 1).
with a match in position i? — (n − 1)! (by symmetry).
with matches in positions 1 and 2? — (n − 2)!
with matches in positions i and j? — (n − 2)!
with matches in positions i, j, k? — (n − 3)!
etc.

(b) How many terms are there in the summations

X X
bij , cijk , ....?
1≤i<j≤n 1≤i<j<k≤n

The easiest argument is as follows. In the case of the first summation, take a sample
of size 2, {r,s} say, from the population {1, 2, . . . , n}, sampling without
! replacement: let
n
i = min(r, s), j = max(r, s). The number of such samples is , so this is the number
2
of terms in the first summation.
!
n
Similarly, the number of terms in the second summation is – and so on.
3
 page 9 110SOR201(2002)

Now for the main problem. We have that

P(at least one match) = P(Ai ∪ A2 ∪ . . . ∪ An )

n
X X
= P(Ai )− P(Ai ∩ Aj ) + · · · + (−1)n+1 P(A1 ∩ · · · ∩ An )
i=1 i<j

by the generalized addition law (1.16). From (a) above,

(n − 1)!
P(Ai ) = , i = 1, . . . , n
n!
(n − 2)!
P(Ai ∪ Aj ) = , 1≤i<j≤n
n!
and more generally
(n − r)!
P(Ai1 ∩ Ai2 ∩ · · · ∩ Air ) = , 1 ≤ i1 < i2 < · · · < ir ≤ n.
n!
From preliminary result (b) above, the number of terms in the summations are
! ! ! !
n n n n
n, , ,..., ,..., =1
2 3 r n

respectively. Hence

! ! !
(n − 1)! n (n − 2)! n (n − 3)! n 0!
P(A1 ∪ · · · ∪ An ) = n. − . + . − · · · + (−1)n+1
n! 2 n! 3 n! n n!
1 1 1
= 1 − + − · · · + (−1)n+1 . .
2! 3! n!
(1.20)
Now
∞
X xr x x2 x3
ex = =1+ + + + ··· for all real x.
r=0
r! 1! 2! 3!
So ∞
X (−1)r 1 1 1
e−1 = =1−1+ − + − ···,
r=0
r! 2! 3! 4!
and therefore

P(A1 ∪ . . . ∪ An ) ≈ 1 − e−1 = 0.632121 when n is large. (1.21)

(The approximation may be regarded as adequate for n ≥ 6: for n = 6, the exact result is
0.631944.) We deduce also that

P(no match) ≈ e−1 = 0.367879. (1.22)

An outcome in which there is no match is termed a derangement.

We shall return to this problem from time to time. The problem can be extended to finding
the probability of exactly k matches (0 ≤ k ≤ n) – see Examples Sheet 1.
This matching problem is typical of many probability problems in that it can appear in a variety
of guises: e.g.
matching of two packs of playing cards (consider one pack to be in a fixed order);
envelopes and letters are mixed up by accident and then paired randomly;
n drivers put their car keys into a basket, then each driver draws a key at random.
It is important to be able to recognize the essential equivalence of such problems.
 page 10 110SOR201(2002)

1.5.2 Collecting Problems

Example 1.2 A Token-Collecting Problem (or Adding Spice to Breakfast)

In a promotional campaign, a company making a breakfast cereal places one of a set of 4
cards in each packet of cereal. Anyone who collects a complete set of cards can claim a prize.
Assuming that the cards are distributed at random into the packets, calculate the probability
that a family which purchases 10 packets will be able to claim at least one prize.
Solution Define
Ai : family does not find card i (i = 1, . . . , 4) in its purchase of 10 packets.
Then the required probability is

P(A1 ∩ A2 ∩ A3 ∩ A4 ) = 1 − P(A1 ∩ A2 ∩ A3 ∩ A4 ) = 1 − P(A1 ∪ A2 ∪ A3 ∪ A4 ).

Again invoking the generalized addition law (1.16), we have

4
X X X
P(A1 ∪A2 ∪A3 ∪A4 ) = P(Ai )− P(Ai ∩Aj )+ P(Ai ∩Aj ∩Ak )−P(A1 ∩A2 ∩A3 ∩A4 ).
i=1 i<j i<j<k

Now  10
3
P(A1 ) = P(no card 1 in 10 packets) =
4
3 × 3 × ... × 3
(by combinatorial argument or multiplication law or binomial distribution (see
4 × 4 × ... × 4
later)). By symmetry
 10
3
P(Ai ) = , i = 1, . . . , 4.
4
Similarly, we have that
 10
2
P(Ai ∩ Aj ) = , i 6= j,
 4 10
1
P(Ai ∩ Aj ∩ Ak ) = , i 6= j 6= k,
4
P(A1 ∩ A2 ∩ A3 ∩ A4 ) = 0.

So  10  10  10

3 2 1
P(A1 ∪ A2 ∪ A3 ∪ A4 ) = 4 −6 +4 = 0.2194.
4 4 4
Hence the required probability is 1 − 0.2194 ≈ 0.78. ♦
 page 11 110SOR201(2002)

1.6 Further Examples

Example 1.3 A Sampling Problem

A bag contains 12 balls numbered 1 to 12. A random sample of 5 balls is drawn (without
replacement). What is the probability that

(a) the largest number chosen is 8;

(b) the median number in the sample is 8?

!
12
Solution There are points in the sample space. Then:
5

(a) If the largest number chosen is 8, the other

! 4 are chosen from the numbers 1 to 7 and the
7
number of ways this can be done is ; so the required probability is
4
!
7
4
!.
12
5

(b) If the median is 8, two of the chosen numbers come from {1, 2, 3, 4, 5, 6, 7} and 2 from
{9, 10, 11, 12}; so the required probability is
! !
7 4
·
2 2
! . u
t
12
5

Example 1.4 The Birthday Problem

If there are n people in a room, what is the probability that at least two of them have the same
birthday?
Solution Recording the n birthdays is equivalent to taking a sample of size n with replacement
from a population of size 365 (leap-years will be ignored). The number of samples is (365) n
and we assume that each is equally likely (corresponds to the assumption that birth rates are
constant throughout the year). The number of samples in which there are no common birthdays
is clearly
365 × 364 × ... × (365 − (n − 1)).
So the probability that there are no common birthdays is

(365)(364)...(365 − (n − 1)) 1 2 n−1

    
n
= 1− 1− ... 1 − . (1.23)
365 365 365 365
1
The required probability is 1 - this expression: it increases rapidly with n, first exceeding 2 at
n = 23 (value 0.5073). ♦
 page 12 110SOR201(2002)

[Some people may find this result surprising, thinking that the critical value of n should be
much larger; but this is perhaps because they are confusing this problem with another one -
what is the probability that at least one person in a group of n people has a specific birthday
364 n
e.g. 1st January. The answer to this is clearly 1 − ( 365 ) : this rises more slowly with n and
1
does not exceed 2 until n = 253.]
Example 1.5 A Seating Problem
If n married couples are seated at random at a round table, find the probability that no wife
sits next to her husband.
Solution Let Ai : couple i sit next to each other. Then the required probability is

1 − P(A1 ∪ A2 ∪ ... ∪ An )
n
X X
= 1− P(Ai ) − · · · − (−1)r+1 P(Ai1 ∩ Ai2 ∩ · · · ∩ Air ) − · · ·
i=1 1≤i1 <i2 <...<ir ≤n
−(−1)n+1 P(A 1 ∩ · · · ∩ An ).

To compute the generic term

P(Ai1 ∩ Ai2 ∩ · · · ∩ Air )
we proceed as follows. There are (2n − 1)! ways of seating 2n people at a round table (put
the first person on some seat, then arrange the other (2n − 1) around them). The number of
arrangements in which a specified set of r men, i 1 , i2 , ..., ir , sit next to their wives is (2n−r −1)!
(regard each specified couple as a single entity - then we have to arrange (2n−2r +r) = (2n−r)
entities (the r specified couples and the other people) around the table). Also, each of the r
specified couples can be seated together in two ways. So the number of arrangements in which
a specified set of r men sit next to their wives is 2 r (2n − r − 1)!. Hence

2r (2n − r − 1)!
P(Ai1 ∩ Ai2 ∩ · · · ∩ Air ) = .
(2n − 1)!

It follows that the required probability is

n
!
X
r+1 n r
1− (−1) 2 (2n − r − 1)!/(2n − 1)!. u
t
r=1
r

(Note: there is another version of this problem in which one imposes the constraint that men
and women must alternate around the table).