CHAPTER 3

Discrete Random Variables

PART 1 Generalities
I Random Variables and Discrete Distributions

Let I be the sample space that compromises all the outcomes of a random
experiment It is often useful to assign to each outcome we r a numerical
value via a certain function This function is what we call a random variable

Def A Random Variable is simply a function

X r s IR
w a s X w

A random variable X is said to be discrete if the

range of
X
is finite or countably infinite
by R X r
denoted or

Example Suppose we throw a fair die I 21 2,3 4,5 65 We can

build many random variables on this space

 1 X X w w
simply gives the number shown on the die
2 Y 1 it w is even
0 if w is odd
3 Z rest of the Euclidean division of w
by 4

Now want to assign a probability to each of the values in the

we

range
R of X This assignment describes the distribution of the random
variable X For discrete RVs the function that assigns to each NER
a IP X N e Co I is the function
probability probability mass

Def The Probability Mass Function p.m f of a discrete RV X r R

is the function R C0,17 such that
p

Ca PLAN IPL wer NW n PL X ins

p

Note the range R E IR so for values in IR not in R o

p
 From the definition of the p.m f it is understood that the distribution
p
of X is inherited from the probability function IP on R

X
r s IR

p Po X p
h
0,1

Note that Une IR e Coil

pin

Example Here are the pints of the RVs X Y and Z from the last example

1 X a Yo Un 1 6 apt
p
f o o

2 Y p
n 42 An 1,2

i s is
 3 Z Z
maps 11 5 to 1 12,61 to 2 431 to 3 141 to 0

The p.m f of Z is thus given

by a pale

Yz
z 1 3
IP 2 27 YG of 43 1 4321 46 y
1 I 1 I 7 z
0 I 2 3

We notice an important property of probability mass functions

For X r R e IR discrete
Prop any
s
Erp n L

Proof Suppose for simplicity that the sample spacer itself is discrete

Er Palm
Er PKW NW RS EE wife a
IPC Lws

PEWS 1
In D
 In a the probability distribution of X dictates how the total
way
of all outcomes in R is distributed allocated to values
probability 1
in the R of X
range

Example Throw 2 fair dice and let X be the sum of the numbers

showing on each die see Chapter 2 The distribution of X is

in this table
given

N 2 3 4 5 6 7 8 9 10 11 12

3 4 5 6 3
Path 436 436 36 36 36 36 736 4 36 36 436 436

Or we can write it as a single function

a IPL Xen 6 17 M if 2 2,3 11,12

p
L 36

0 otherwise
Exercise We draw 7 cards from a deck of 52 cards without replacement
Let X be the random variable counting the number of spades in
the sample Compute the of X
p.m f
Exercise We draw 7 cards from a deck of 52 cards without replacement
Let X be the random variable counting the number of spades in
the sample Compute the of X
p.m f

Solution In 52 cards we have 13 spades and 39 other cards

We want to compute IPL X n An 0 7 The total number
of outcomes is 7 The favourable outcomes are counted as
follows
St 1 Pick cards out of 13
a
Ba

St 2 Pick 7 n cards out of 39 FL

1 In if
Hence the p.m f of X is plat Pl a n 2 0 17

Check ftp.lal 1
L y
O otherwise
Note This is a
special distribution known as the hypergeometric
distribution More on that in Part 2
I Cumulative Distribution Function

Another describe the distribution of discrete RV is

way to completely a

using a function that gives IP Xen Ane IR instead of Plan like

the p.m f does

Def The Cumulative Distribution Function c d f of a discrete random

variable X is the function
Fx R s 0,1
IPL Xen
ni s
En Alt
Example In the experiment of 2 dice throws and X their sum the edf is

N 2 3 4 5 6 7 8 9 10 11 12

3 6 10 15 21 26 33 35
Fx n 436 36 36 36 36 36 363036 36 36 1
 after
1

OF

o s
2 3 Y 5 6 I 8 to 11 12 N

CPF of X sum of 2
fair dice
The cd f for discrete RVs is a step function
 Properties of the c d f

Noticethat for a finite IRI e x the value of Fx at the largest

range
value of the range and beyond is 1 Below is a generalization for

infinitely many values

1
Prop tiny Fx a and
kik Fx n O

We
skip the proof Also notice that the cd f is always non deceasing

Prop Hae b FxCa s Fx b

Proof
Fx lb Pl Xe b a PL X a PLACÉ 3 IPL x.ca Fla D

Equality above holds between the points of mass i.e on fei Nit where
Ute Ni Rin Px Ni 0
 Another important
property
is that for
RV X the c d f and p.m f
a

completely describe one another meaning that we only need one of them to

specify the distribution of X Indeed in the jumps in the graph of the

cdf are the probabilities at points ki where there is mass i.e where 0
p

Psni Fx Ni Fx ni

We formalize this In general for Fx a fiona Fx lt

Pla e Xe b Fx b Fx a

Hence
p
Ca Pl X a FxCal Fx a see on graph

Note We can define and study RVs using their distributions without reference to
an
underlying sample space r This is
what we do from now on
Example Example 3.13 in book the pint of a RV Y is
given
in this table

y I 2 4 8 16

ply 0.05 0.10 0.35 0.40 0.10

We compute the c d f
Fly
Ky 1 Fly O
F I P Ye 1
Hye 1,2 Fly 0.05
I
p
fye 12,4 Fly F 2 PLY a 2 1
p plz 0.15

bye 4,8 Fly FLY PLY 4 PID plz PCU 0.5

FLO PLY 83 pll tp12 0.9
Hye 8,16 Fly p 4 p8
16 Fly F1 6 PLY 163 1
bye x

O get
D 0.15 Leg Y
0.5
Key 8
0.9 8 eye 16
plot in Figure 3.5 book
s y 16
Exercise The RV X has the c d f below Compute the p.m f
O N C I

rant 0 72 3 ER 20

Solution Points of i.e where there are 21,3 205

mass jumps are
p 1 IPC X 1 Fx I 0.3

p 3 PIX 3 Fx 3 Fx i 0.72 0.3 0 42

IP X Fx Fx 1 0 72
p 120
203 20 3 0 28

We state one final property of the c d f without proof

Prop Fx is
right
continuous i.e Une IR

Flat FLA F x
fig
III Expected Value

In crude teems the mean expected value of a random variable X denoted

by pg or x weighted average of the values in the range of X

is a

with the weights being the probabilities of X taking these values

Def The Expected Value Mean of a discrete random variable X denoted

by 14
or EX is
given by

EX Plan
Eg x
p
Cal
Ern
that
granted Er IN pins ex

Intuitively if a random experiment is repeated a

large number of times the
average of the values X each time
obtained This is
by approximates 14
later formalized by the law of large Numbers
Example Let X be the RV from the exercise before whose
p.m f
is
given by
N 1 3 20
pyx 0.3 0.42 0.28 note EX does
not need to
Then EX I x 0.3 3 0.42 20 0.28 7 16 be a value
in the
range
Example In the fain die experiment with Nw w

X 1 2 3 4 5 6 21 3.5
I 6

Exercise In a lottery players pick 6 different integers between 1 and 49

the order being irrelevant The lottery committee then selects 6 integers at
random asthe winning numbers
If all 6 chosen numbers match the winning number then the player
wins 1,200 000
For 5 matches out of 6 the player wins 800
For 4 matches out of 6 theplayer wins 35
what is the expected value of the amount won in a
game
Solution Let X be the amount won in a in
game

PIX 1,200,0003 leg a 0.000 000 072

8
Pl X 800 I 0 000018

1
PL X 353 E Y 7 0 00097
Y
PIX 0 1 PIXEL35,800 1200000 0 999 011 928

EX I a
p
n 0.13

On average you win 13

cents
per game
This means that even if it costs
participate you lose on average 37 cents pergame
50 cents to
only
Playing 10,000 games leads to an approximate loss of 3700
 Random Variables with Infinite Range

A discrete RV can have a countably infinite range sequence In these cases

the expected value or may not exist If it does not exist all values in
may
the range have the same we write EX x or x
sign
As mentioned before EX exists if the series is absolutely convergent

Emp In ex

Exercise Let X be the number of interviews a student has to go through prior

to landing a of X is given
The
job p.m f by

kn if 2 1 2,3 ne N't
pate
L o otherwise

1 Compute k
2 Does EX exist
 Solution
1 To compute k recall that 1 In this
Er pin case

É Ma k
E I L

We know that EI it k 6 2

Harmonic Series

2 EX É m
p la É m
En f E I x diverges

EX does notexist Ex o We say the distribution has heavy trails

Example In a certain population children are borneither boys with
as

probability p or girls with probability l p for some pelo 1 The births

are all independent of one another
Let X number of children born and including the first boy
up to

let us compute IP X 4 Think of it as ordered sampling with replacement

Pl X 4 PL 3 girls then I boy A p p

In general
p n Pl X n
p 1
p
if l 2,3
n

O otherwise
Note that the distribution of X depends on a parameter here This is
p
common For the f above X is said to follow a geometric
very p.m
distribution with parameter e lo 1 We write Xn geom p
p

We start
by checking that p is a legitimate p.m f i.e
Epix 1
 E p tat p EY Ci p p IE d p p I up
Plp 1
verified

Now we compute X if it exists We use a smart trick notice

I p n l
p
Ip
pE l
Efx Ee xp l
p dap 11
p

fp Ep E CI p f p top Ip 1 Pipe Yp

Hence EX Yp If
p Ya we an
equally likely to get a boy or a

girl we can expect to see the first boy on the second birth

If 1 then decreases to see the first boy

I
so we expect
p
sooner which makes sense
 Expected Value of a Function of a RV

Def The nthmoment of a random variable or of its distribution is XD

In order to compute the moments of a distribution it is useful to have a

formula for Eth X for real valued functions h We don't need

to compute the p.m f of h X in order to
get ECh x

Prop let h R R let X be a discrete random variable Then

Echl x Er hat pain

let X be RV with the following f n 1 3 20

Example a
p.m
The second moment is thus PY 0.3 0.42 0.28
by given
2
EX 1 0.3 330.42 2070.28 116.08
 An important special case what if h is linear actually affine let
heat an tb Then 4Th x can be directly computed in terms of ELN

Prop The Expected Value of a RV X is linear i e Ya be IR we have

all b a EA b

Proof let h IR IR hat an tb

Flax b heal p n
E at b Pala
a n b a EX Ib a EX b
up pain
D

Example In the example above we have

2
3 7 3 X 7 3 116.08 7 355.24
Exercise The
daily number of sales S of distributed follows
an
employee is as

s I 2 3 4 5 6

PCS 0.3 0.3 0.2 0 I 0.05 0.05

The daily bonus depends on the number of sales S and it given is

by
5750 but it is capped at 70 Compute the expected bonus
Exercise The
daily number of sales S of distributed follows
an
employee is as

s I 2 3 4 5 6

PCS D 0.3 0.3 0.2 0 I 0.05 0.05

The daily bonus depends on the number of sales S and it given is

by
5750 but it is capped at 70 Compute the expected bonus

Solution let B be the RV giving the bonus of an

employee
B h S min 5 50,70

S I 2 3 4 5 6

b his 51 54 59 66 70 70

ECB 4Th S 0.3 51 54 0 2159 0 1 66 o 05 70 70

56.9
 I Variana and Standard Deviation

The Variance Val x of a random variable measures the average squared

deviation from the mean It indicates how spread the possible values
p
the RV can take ale

Def The Variance Var X of a discrete RV X with and pmf

p is
meany

Vai X
Er n
p pen

The Standard Deviation Tx is Tx Tx

The variance is thus always positive Vax o for all RVs X

The following is an easier formula for the variance that uses the first
2 moments of X
Prop Ven X x2 Ex

Proof Veux E n'pin 23 my pint Ept

E X 214 pet 142 E X D
y

let X be a RV with the following p.m.f n 1 3 20

Example
PEX 0.3 0.42 0.28

We computed HX 7.16 and X 3 116.08 before

Var X EX X 116.08 7.162 64.8144
and Ty 8.05
 Variance of a function of a RV

Prop let h R R let X be a discrete random variable Then

VenCh X
Er hea EL NXD pain

For h affine there is a simpler formula for Vai ha that only uses Vai x
However unlike the mean the variance is not linear

Prop For X a discrete RV and all a bEIR we have

Ven ax b a Var X and Jax tb la Tx

Proof Left as an exercise

In the second part of the chapter we study special examples of discrete

distributions that arise very frequently in modelling