CHAPTER 7

Discrete Probability Distributions

1. Working with Discrete Random Variables

To work with a discrete random variable, we need to know its probability distribution.
As we’ve seen, the probability distribution of a discrete random variable X is a list of the possible values
xi of X together with the probability P (X = xi ) that the variable X will take on that possible value xi .
A discrete probability distribution arises as part of a counting process and often1 describes a finite set
of possible occurrences.
For example, let X be the number of successful treatments out of 2 hospital patients. Then X is a
discrete random variable, because the random variable representing the number of successes can only
take the values 0, 1, or 2. The set of values that X can take is: X 2 {0, 1, 2} and the probabilities of
these possible values, namely, P (X = 0), P (X = 1), P (X = 2), make up the probability distribution for
this discrete random variable. Note that P (X = 1.5) cannot be calculated as X cannot take the value
1.5: half a patient cannot be treated successfully!
Some patterns of behaviour are well described by mathematical models. Many probability distribu-
tions are so important in theory or applications that they have been given specific names. There are
many families of discrete probability distributions. In this chapter, we will discuss two important and
commonly-occurring discrete random variables and their distributions: the Poisson and Binomial random
variables.

2. The Poisson Probability Distribution

In this section, we will deal with one of the most important discrete probability distributions: we will
focus on the Poisson distribution, which will describe a Poisson random variable X.
We are often interested in counting the number of occurrences of something in a given interval, e.g., the
number of arrivals in a queue in 1 minute, or the number of typos in a page of text. In the first of these,
it is a time interval; in the second it is a spatial interval.
The Poisson distribution is a discrete probability distribution that tells us the probability of a given
number of occurrences in a given interval of time or space, if these occur with a known constant rate
and independently of the time since the last event. This constant rate is denoted by , standing for the
mean number of occurrences per unit interval. Note: even though the number of occurrences is always
an integer (whole number), the mean need not be an integer.
The Poisson distribution is commonly used to identify probabilities about the number of occurrences
per unit time or the number of occurrences per unit length or area. Then the number of occurrences X
is a Poisson random variable and we write X ⇠ Poi( ).
The Poisson distribution assumes that:

(1) the occurrences are independent;

(2) the probability of occurrence is proportional to the length of the interval;
(3) there is a fixed mean or expected number of occurrences per unit interval (denoted by ).

1Though not always: e.g., the discrete Poisson distribution allows for a countably infinite set of values {1, 2, 3, 4, . . .}.

87
88 MIS10090

There is one parameter for the Poisson distribution: which is the mean rate of occurrence. Thus “the”
Poisson distribution is actually a family of distributions, one for each value of .
There is no upper limit on the Poisson random variable X standing for number of occurrences in a given
interval: it could be any number in {0, 1, 2, 3, 4, . . .}. However, there is a low probability of getting a
very large number of occurrences in the given interval. It can be shown that the probability that the
random variable X is equal to a particular nonnegative integer k of occurrences in the interval is given
by the Poisson probability function (also called Poisson probability mass function):
e k
P (X = k) = .
k!
where

• X = observed number of occurrences per unit interval i.e., random variable

• = expected number of occurrences per unit time (average for X)
• e = base of the natural logarithm (2.71828. . . ), a mathematical constant.

(Recall that the notation P (X = k) is read as “the probability that the random variable X has the
particular value k”.)
This defines the Poisson distribution, and it can be shown that the mean and variance of the Poisson
random variable described by this distribution are both :
2
E(X) = , Var(X) = X = .

Examples of random variables that can be described by the Poisson distribution include:

• The number of calls to a call centre in an hour: this value must be a whole number (you cannot
receive a fraction of a call). However, the number of calls can take on any integer value such as
0 calls in an hour, 10 calls in an hour.
• The number of cars that drive into a petrol station on a given day: the value must again be
a whole number as there is no such thing as half a car but there is no limit on the number of
occurrences (the number of cars driving into the petrol station).
• The number of bags lost in an airport per year;
• The number of flawed engines in a day’s production.

The shape of the Poisson distribution depends on the parameter, as shown in Figure 7.1 below.

Figure 7.1. Shape of Poisson distribution for = 0.5 (left) and = 3.0 (right)

For small values of , say < 2, the Poisson distribution is noticably right skewed. However, as
the parameter increases, the shape of the distribution becomes more and more like the bell-shaped
symmetrical normal distribution (see Chapter 8), as shown in Figure 7.2 below. Because of this, a normal
distribution is often used to approximate a Poisson distribution when is reasonably big.
 Data Analysis for Decision Makers 89

Figure 7.2. Shape of Poisson distribution becomes more symmetrical as increases.

Graphic from Wikipedia

Example 7.1 (Poisson Distribution). The number of complaints to the customer service desk of a large
shopping centre is known to follow a Poisson distribution with an average of 3.4 per day. What is the
probability that on any given day, the number of complaints is at least two?
Solution. We will work to precision of four decimal places. In this case, = 3.4. We are asked to
find P (X 2) but to do this directly would mean working out infinitely many expressions, as X can
take any of the values 0, 1, 2, 3, . . . . So, we use an indirect approach, invoking the complement rule:
P (X 2) = 1 P (X  1).
Now P (X  1) = P (X = 0) + P (X = 1), so to calculate P (X  1) we must calculate P (X = 0) and
P (X = 1) and then add them together. First,
e 0 e 3.4 3.40
P (X = 0) = = .
0! 0!
Since 0! = 1 and any nonzero number raised to the power of 0 is 1, we have 3.40 = 1 so
e 3.4
P (X = 0) = = 0.0334.
1
Also,
e 1 e 3.4 3.41
P (X = 1) = = = 0.1135.
1! 1!
Then
P (X 2) = 1 P (X  1) = 1 (P (X = 0) + P (X = 1)) = 1 0.0334 0.1135 = 0.8531.
It is useful to draw the probability distribution to visualise this example: see Figure 7.3.
Notice that the probability of “at least 2 complaints” is the complement of “at most 1 complaint”: that
is, P (at least 2 complaints) = 1 P (at most 1 complaint). }
Exercise 7.2. Find the mean and standard deviation of the Poisson random variable in the last example.
Exercise 7.3 (Poisson Distribution). A toll-free phone number is available from 9 a.m. to 9 p.m. to
customers who want to register a complaint about a product purchased from your company. Past history
indicates that, on average, 0.4 calls are received per hour.
What is the probability that during a 1-hour period:

(a) zero calls will be received?

(b) one phone call will be received?
(c) two phone calls will be received?
(d ) three or more phone calls will be received?

Hint for (d ): what is the sum of the probabilities from (a), (b) and (c)?
90 MIS10090

Figure 7.3. Poisson distribution of complaints: “at most 1 complaint” in green; “at
least 2 complaints” in blue

Note 7.4. You may be given the rate for a given interval size (e.g., told = 3 arrivals per minute)
but asked something like “what is the probability of 20 arrivals in an hour”. Make sure you use the same
unit of interval size throughout, converting if necessary! In this example, convert to = 180 arrivals per
hour.

3. The Binomial Probability Distribution

Definition 7.5. A Bernoulli trial is a trial (experiment) with only two possible outcomes, often called
success (1) and failure (0). That is, 0 and 1 are the two mutually exclusive and collectively exhaustive
outcomes.

We let p denote the probability of success: thus, the probability of failure is 1 p.

Definition 7.6. A binomial random variable X is a DRV which describes the number of successes if we
carry out n independent Bernoulli trials, where each Bernoulli trial has the same probability p of success.

A binomial random variable follows a binomial distribution, with two parameters: n and p. We write
X ⇠ Bin(n, p).
Clearly the least possible number of successes in n trials is 0, while the greatest possible number of
successes is n, and any number of successes between 0 and n inclusive is possible.
The shape of the Poisson distribution depends on the n and p parameters, as shown in Figure 7.4 below.

The binomial distribution is a useful model when the DRV is the number of events of interest in a
sample of n observations where there are only two possible outcomes for each observation: success (with
probability p) and failure (with probability 1 p).
To summarise: the binomial distribution has four basic properties:

(1) The sample contains a fixed number of observations n;

(2) There are two mutually exclusive and collectively exhaustive outcomes (success and failure);
(3) The probability of success, p, is constant for each observation;
(4) The outcomes for each trial are independent.
 Data Analysis for Decision Makers 91

Figure 7.4. Shape of Binomial distribution for n = 20 and p = 0.1 (left) and p = 0.4 (right)

Samples are selected with replacement from a finite population (or without replacement for an infinite
population).
It can be shown that the probability of getting k successes from n trials is
✓ ◆
n k
P (X = k) = p (1 p)n k .
k
This defines the binomial distribution, and it can be shown that the mean and variance of the binomial
distribution are
2
E(X) = np, Var(X) = X = np(1 p).

If n is big enough, then the binomial distribution is not very skewed. In this case, a good approximation
to Bin(n, p) is the normal distribution N (np, np(1 p)), that is, a normal distribution with the same
mean and variance as the binomial distribution.
There are various rules of thumb to decide if n is big enough, and if p is far enough away from the
extreme values of 0 or 1. An example rule is that both of the values np and n(1 p) must be 5 (a
stronger rule is 9).
Example 7.7. A call centre routinely monitors 10% of all incoming calls for quality control purposes.
In a random sample of 10 calls, we wish to know the probability that exactly two of these ten calls were
monitored.
Solution: Let X stand for the number of calls monitored. Think of each call as a Bernoulli trial with
“success” as meaning “call monitored”: then we know there is a 10% chance of a call being monitored,
i.e., probability of success is p = 0.1. Our sample is 10 calls, so n = 10, and we assume they are all
independent. Thus X is a binomial random variable.
In this case, we are asked for k = 2 so we wish to quantify P (X = 2):
✓ ◆
10
P (X = 2) = 0.12 0.98 since 1 p = 1 0.1 = 0.9.
2
From Chapter 4, §5.3, we know the formula for the binomial coefficient 10
2 is
✓ ◆
10 10! 10! 10 ⇥ 9
= = = = 45
2 2!(10 2)! 2!8! 2⇥1
since the 8! above the line cancels the 8! below the line (notice that 10! = 10 ⇥ 9 ⇥ 8!). Thus
P (X = 2) = 45 ⇥ 0.12 0.98 = 45 ⇥ 0.01 ⇥ 0.4304671 = 0.1937102445
so the probability that exactly two of these ten calls were monitored is a little less than 20%. }
Exercise 7.8. Continuing on from the last example, find:
(a) P (X = 0), the probability that no calls were monitored;
(b) P (X = 1), the probability that exactly one call was monitored;
(c) P (X 3), the probability that at least three calls were monitored.
92 MIS10090

Hint for (c): use the work from (a), (b) and the last example; and think about the complement.
Exercise 7.9. Find for yourselves the mean and variance of the binomial random variable in the last
example.