FS1: Chapter 2, Poisson Distribution Siméon Denis

Poisson
1781-1840
Motivation:
Suppose when you’re playing a video game, you collect an average of 20 coins per hour.

After a while of playing, you think that you’ve improved.

During a one-hour window, you collect 25 coins and say that this is evidence you’ve
improved! But are you sure? Could this have just been a fluke?

A hospital A&E department admits on average 12 emergency patients per half hour.

After a local campaign advising people when it is appropriate to visit their GP instead
of going to A&E, the hospital measured only 16 people arriving in a one hour window.

Does this mean the local campaign worked?

Suppose there has been an average of 6 car accidents at a particular road

junction over a 1 month period. The local council installed a roundabout to
try and reduce the number of accidents.

In the following month, there were only 3 car accidents.

Is this evidence that the traffic lights have worked? Or could this month
have been a ‘one off’?
What is the Poisson Distribution?
Name Description Outcomes

Binomial The number of Number of successes

𝐵(𝑛, 𝑝) “successes” out of 𝑛 {0, 1, 2, … , 𝑛}
trials, each with a
probability 𝑝 of success.

How many events occur Number of events

Poisson
within some period of ℕ = {0, 1, 2, … }
𝑃𝑜(𝜆) (since technically any non-
time, given an average negative number of events
rate 𝜆 (“lambda”) at could occur)
which they occur.

Examples of things which are modelled by the Poisson distribution:

• The number of chocolate chips in a biscuit

• The number of radioactive particles being emitted by a particular source in a 3 minute period

• The number of phone calls arriving at a call centre in a one-hour period

 What actually is it?
We said that the Poisson Distribution allows us to calculate the count of a
number of events happening within some period, given an average rate.

Calculate the probability of 8 cars passing in the next hour, given that on average
5 pass an hour.

Suppose we wanted to see how we could do this with a Binomial Distribution…

2pm 3pm
We could divide the hour up into 10
equal time intervals of 6 minutes each.
✓ ✓ ✓ ✓ ✓ Each time slot is a trial, where “success”
is a car passing in that period.

Now if we expect 5 cars on average in the hour (i.e. 𝜆 = 5), and we’ve made 10 time slots (i.e.
𝑛 = 10), then the probability is 0.5, as half of the time slots are filled.

𝑝=
5
=
10 𝑛①𝜆

Then to answer the original question:

𝟏𝟎
𝑿~𝑩 𝟏𝟎, 𝟎. 𝟓 𝑷 𝑿=𝟖 = 𝟎. 𝟓𝟖 𝟎. 𝟓𝟐 = 𝟎. 𝟎𝟒𝟑𝟗
𝟖
 than
A problem with how we’ve modelled this is that more m the one car could pass in any given
time interval, and thus the count of successes won’t necessarily be the count of cars.

2pm 3pm

✓ ✓ ✓ ✓ ✓

I5 0.25
=
=

We can split into shorter time intervals so that we are less likely to have more than
one car in one time slot… But there’s a problem.

If there’s now 20 intervals of 3 minutes each (and still an average rate of 5 cars an
hour), let’s see the difference:

0.25
𝑿~𝑩 𝟐𝟎, 𝟎. 𝟐𝟓
z

𝑷 𝑿=𝟖 =
(2)0.2500.75"
= 0.0609/4dp)
 We could gradually increase the number 𝒏 of time intervals we’ve split the hour
interval into, until we get infinitely small slivers of time, where only one event can
occur in any time sliver, and thus it is acceptable to have “success” and “failure”, where
the count of successes will now indeed be the count of cars.

𝑛 𝑥 𝑛−𝑥
𝜆
𝑃 𝑋 = 𝑥 = lim 𝑝 1−𝑝 𝑤ℎ𝑒𝑟𝑒 𝑝 =
𝑛→∞ 𝑥 𝑛

57 5
=

−𝜆 𝑥
𝑒 𝜆 The proof for this can be found
=
𝑥! P0(5) on the next page but is not

↑
required.

poisson P(X 8)
2550
=
=

From Binomial to Poisson (desmos.com)

https://www.desmos.com/calculator/pegz5os4re 0.0653
=

! The Poisson Distribution is a distribution over the number of events

which occur within a period of time, given an average rate 𝜆.
𝑒 −𝜆 𝜆𝑥
𝑃 𝑋=𝑥 =
𝑥!
 𝑛 𝑥 𝑛−𝑥 𝑒 −𝜆 𝜆𝑥 𝜆
Prove that lim 𝑝 1−𝑝 = 𝑥! (where 𝑤ℎ𝑒𝑟𝑒 𝑝 = 𝑛)
𝑛→∞ 𝑥
𝑥 𝑛−𝑥
𝑛 𝜆 𝜆
lim 1−
𝑛→∞ 𝑥 𝑛 𝑛
𝑛−𝑥
𝑛 𝑛−1 𝑛−2 … 𝑛−𝑥+1 𝜆𝑥 𝜆
= lim 1 − ->0
𝑛→∞ 𝑥! o
𝑛𝑥 𝑛
xn n

⑦
1 2 𝑥−1 nxX...
𝑛−𝑥 1 - x -

1 1−𝑛 1−𝑛 … 1− 𝑛 𝜆
= lim 𝜆𝑥 1−
𝑛→∞ 𝑥! & 𝑛 zero
=>
1 2
But given 𝑛 → ∞, 1 − 𝑛 → 1, 1 − 𝑛 → 1 and so on.
The numerator therefore simplifies to 1.
1 -

os
Also, as 𝑛 → ∞, 𝑛 − 𝑥 → 𝑛 (i.e. infinity minus a finite number is still infinity!)
So:
𝜆o
𝑛−𝑥 𝑛
𝜆
lim 1 − = lim 1 − = 𝑒 −𝑘-7
𝑛→∞ 𝑛 𝑛→∞ 𝑛 7
pxz
-

= 1

/
(this is a standard result which we won’t prove here) x

𝑛 𝑥 𝑛−𝑥 𝑒 −𝜆 𝜆𝑥
So, we are left with lim 𝑝 1−𝑝 =
𝑛→∞ 𝑥 𝑥!
Here are some Poisson distributions plotted for various average rates.
Examples of recently studied phenomena obeying the Poisson distribution:

• the number of mutations on a given strand of DNA per time unit

• the number of bankruptcies that are filed in a month

• the number of arrivals at a car wash in one hour

• the number of network failures per day

• the number of file server virus infection at a data centre during a 24-hour period

• the number of Airbus 330 aircraft engine shutdowns per 100,000 flight hours

• the number of asthma patient arrivals in a given hour at a walk-in clinic

• the number of hungry persons entering McDonald's restaurant per day

• the number of work-related accidents over a given production time

• the number of birth, deaths, marriages, divorces, suicides, and homicides over a given period of time

• the number of customers who call to complain about a service problem per month

• the number of visitors to a web site per minute

• the number of calls to consumer hot line in a 5-minute period

• the number of telephone calls per minute in a small business

• the number of arrivals at a road tollbooth per minute between 3 A.M. and 4 A.M. in January on the Kansas Turnpike

Poisson Distribution | Brilliant Math & Science Wiki

Source: https://brilliant.org/wiki/poisson-distribution/
 Calculating probabilities using the Poisson Distribution
-7 4
=

𝑋~𝑃𝑜(4), find, both manually and using the distribution mode on your calculator
-

𝑃 𝑋=3 = e4 0.1954
=

P(X 4) P(X 5) 0.19536... 0.15629

= +

𝑃 3<𝑋≤5 =
=
=
+

0.3517 =

𝑃 𝑋≥1 = 1 -
p(X 0) 1 =

=
-
0.0183
8 1,2,3... = 0.9817
𝑃(𝑋 > 2) -

⑪33,4,5.... 1
=
-

&
0) p(X =
-
P(X =

1) -
P(X 2)
=

1-0.018315-0.07326..
= -
0.14652...

0.7619
=

𝑒 −𝜆 𝜆𝑥
𝑃 𝑋=𝑥 =
𝑥!
-

𝑃 𝑋 > 𝑎 = 1 − 𝑃(𝑋 ≤ 𝑎)
Given 𝑋~𝑃𝑜 2.3 find
a) 𝑃 𝑋 = 4 = 0.1169
=>

1 -0.1003 0.8997
=

p(X 0)
=

1
=

b) 𝑃 𝑋 ≥ 1 = -

c) 𝑃 4<𝑋<6 = P(X 5) 0.0538

=
=

Your Turn
Given 𝑌~𝑃𝑜 0.35 find
a) 𝑃 𝑌 = 1 = 0.2466

b) 𝑃 𝑌 ≥ 1 = 1 -

p(X 0)
=
1
=
-
0.7047 0.2953
=

c) 𝑃 1 ≤ 𝑌 < 3 = P(Y 1) P(4 2)

= +
= =
0.24664 + 0.04316
1,2 0.2898
=

Given 𝑋~𝑃𝑜 3.6 find

a) 𝑃 𝑋 = 5 = 0.1377

b) 𝑃 3 < 𝑋 ≤ 6 = 4(X 4) P(X 5)

P(X 6) 0.4115
=
=
= +
+
=

4,5,6
c) 𝑃 𝑋<2 = P(X 0) =

P(X 1)
+
=
0.1257
=

0,1 Ex 2A
-
Cumulative Probabilities
As with the Binomial distribution, we can use tables for the Cumulative Distribution
Function of a Poisson distribution, i.e. the running total of the probability up to a particular
value.
You can also use your calculator; choose Poisson CD. For Graphics, you can always use CD,
even if you want to find a probability like 𝑃(𝑋 = 3). Just set the upper and lower to 3.

On average 8 cars come down a country road an hour. What’s the probability that:
-

a) Less than 5 cars pass in the next hour?

etx be the number
④
of
b) At least 3 cars pass? an hour.
cars
passing
c) Between 2 and 5 (inclusive) cars. X -
P0(8) 18 =

a) P(X < 5) p(X = 4)

= 0.0996
=

2:3 v=10000

b) P(X=3) 1
=
-
x2) 1
p(X- = -

0.0138

3,4,5,6...
= 0.9862
=

P(X15) P(X 41)

2)P(2=X(5)
-

= -

2,3,4,5 0.191236
=
-
0.003019 = 0.1882
-
Using tables backwards:
(the Classwiz calculators can’t unfortunately do discrete distributions backwards, but the graphics calculators can)

If 𝑋~𝑃𝑜(7.5) find 𝑎, 𝑏, 𝑐 if:

(a) 𝑃 𝑋 ≤ 𝑎 = 0.2414
(b) 𝑃 𝑋 < 𝑏 = 0.5246
(c) 𝑃 𝑋 ≥ 𝑐 = 0.3380
~
o
↑
a) P(x a) 0.2414
②
=

P(X(5) 0.2414

·
=

P(X =5) = 0.2414

Hence A= 5

b( P(X(7)
=
0.5246

b8=

P(X(8) P(X17) =

↑
I
P(X=C) 0.3380
2) 1-0.3380 0.6620
=

= 8) 0.6628 32ct1,c+" 1)
P(X 9) 0.3380/P(X
= 0.3 =
=

=
=
0.3388
0.6620 P(x 1) =
c -

0.2620 P(X= 8)
=

c9 =
B
2 1 8 c9 Ex 2A
=
=
=
- -
Properties of the Poisson Distribution
We saw with in order to model something with a Binomial Distribution, we had to
make some assumptions, e.g. each event was independent, and had two outcomes.
We have similar restrictions on events for a Poisson Distribution:

This means we can’t have multiple events occurring

! Events must occur: at once. We treat events as instantaneous.

• Singly in time. If an event occurred just a moment ago, another one

• Independently of each other. is no less likely to occur now than it a while later..
• At a constant rate in the sense that the mean number
of occurrences in the interval is proportional to the
length of the interval.
You can usually tell in an exam if a Poisson
Distribution is intended if the word ‘rate’ is used.
Jan 2012 Q4

Poisson (or 𝑷𝒐 𝟓 )

(From mark scheme) Hits occur singly in time

Hits are independent or hits occurs randomly
Hits occur at a constant rate
 Modelling with the Poisson Distribution
An internet service provider has a large number of users regularly connecting to the internet.
On average, 4 users every hour fail to connect to the internet at their first attempt.
-

a) Give two reasons why a Poisson distribution might be a suitable model for the number of failed
connections every hour. i)
occur
singlyin time ii) occur at a constant rate
b) Find the probability that in a randomly chosen hour:
i. 2 users fail to connect at their first attempt
[ ii.
< -

more than 6 users failed to connect at their first attempt

-

c) Find the probability that in a randomly chosen 90-minute

-
interval: Sometimes we need to
i. 5 users fail to connect at their first attempt scale the rate to a different
② time period/length.
ii. Fewer than 7 users fail to connect at their first attempt
b ⑨

to connecton their 1st attempt.

Let↑be the number of users
failing in a (hour period.
X P0(4)i)P(X 2)
046534dp.
=
=
-

ii) p(X > 6) 0.1107

=

-
in
t be
Let number
the of users failing
2) 4-7 hour
P0(6)
↓X.5 Y
period.
-

a 90-min
x(51
6 -
90 mins P (45) 0.1606
= =

-
4dp.
p(Y(7) 06063 =
 A shop sells radios at a rate of 2.5 per week.
~

a) Find the probability that in a two-week period the shop sells at least 7 radios.
-> -

b) Deliveries of these radios come every 4 weeks. Find the probability of selling fewer
than 12 radios in a four-week period.
c) The manager wishes to make sure that the probability of the shop running out of
radios during a four-week period is less than 0.01. Find the smallest number of
radios the manager should have in stock immediately after the delivery.

radios sold in
LetX number
be the of a 2 week
period.
a) X -
p0(5) 7) 0.3937
P(X= -
=

radios sold in 4 week period.

3) Letb e the number of a

--P0(10) P(Y 12) 0.9799

=

<
-
stock
the number of radios in
2) If a is

a)<0.01 he should
P(y >
Hence a =
18,
18 stock.
1 -
p(y(a)<0.01 least
have at in

<p(Y (a)
3(Y
0.9
= 18) 0.9928
=
 Tip: Think carefully about whether we wish to use Binomial or Poisson.
May 2011 Q5
-

-
Po

B(n,p)
I

=
p Y -

->

a)
LetXbe the
number
of defects in 100cm planch a

P0(5)P(x13) 0.2650(4dp)
=

x -
~

6) Let Y be the
number of plants with atmost 3 defects.
p(y <2) p(47))
B(6,0.2650)
=

Y -

04987(4dp)
=

Ex 2C
 Adding Poisson Distributions
If we have two Poisson distributions and want to represent the total number of events
across the same time interval, we can form a new Poisson distribution in which we simply
add the rates.

e.g. If 5 cars pass per hour in road A and 8 cars pass per hour in road B, then 13 cars pass
per hour in roads A and B combined

!If 𝑋~𝑃𝑜 𝜆1 and 𝑌~𝑃𝑜 𝜆2 then

𝑋 + 𝑌~𝑃𝑜 𝜆1 + 𝜆2 .
For 𝑋 + 𝑌 to be meaningful, 𝑋 and 𝑌 must
represent the same time interval.

If 𝑋~𝑃𝑜 3.6 and 𝑌~𝑃𝑜 4.4 find:

- -
+
x y n
p0(8)
7) 0.1396/4dp)
(a) 𝑃 𝑋 + 𝑌 = 7
a) P(x y
=
+

(b) 𝑃(𝑋 + 𝑌 ≤ 5)
=

b) P(x y25) 0.19/2/4dp)

=
+

-
 The number of cars passing an observation point in a 5-minute interval is modelled by a
Poisson distribution with mean 2. The number of other vehicles passing the observation
point in a 15-minute interval is modelled by a Poisson distribution with mean 3. Find the
probability that:
(a) exactly 5 vehicles, of any type, pass the observation point in a 10-minute interval
(b) more than 8 vehicles, of any type, pass the observation point in a 15-minute interval

Let
④
xbe the number
of vehicles
passing
in a 10-min interval

x -
P0(4 +

2)p(X 5) = 0.1606(4dp)
=

x w
p0(6)
15-min into
of vehicles passing
in a

b) Let Y be the number

YwP0(9) p(y > 0) 1

=
p(4-8)
-

=
0.5443/4dp)
-

Ex 2D
Mean and Variance of Poisson distribution
We know a Poisson distribution is the number of events that occur within a
given time interval, where events occur at a rate of 𝜆 per that interval.
Given this, is should be clear the expected value 𝑬 𝑿 is 𝜆
i.e. the expected number of events that occur

𝐸 𝑋 =𝜆
𝑉𝑎𝑟 𝑋 = 𝜆
The proof that 𝑉𝑎𝑟 𝑋 = 𝜆 is tricky, but will be easier once you do FS1 Ch7 ‘Probability generating functions’

The fact that 𝐸(𝑋) and 𝑉𝑎𝑟(𝑋) are equal can be used to assess
whether a Poisson distribution is a suitable model.
 A botanist counts the number of daisies, 𝑥, in each of 80 randomly selected squares within a field.
The results are summarised below.
Σ𝑥 = 295, Σ𝑥 2 = 1386
(a) Calculate the mean and the variance of the number of daisies per square for the 80 squares.
Give your answers to 2 decimal places.
(b) Explain how the answers from part a support the choice of a Poisson distribution as a model.
(c) Using a suitable value for 𝜆, estimate the probability that exactly 3 daises will be found in a
- -

randomly selected square.

3.6875b)Asii a
a)i x 5
=

=
= r

N
Poisson distribution is

f' = 2x2 -
3.68752 suitable.
-

=
386 -
368753 3.727...
=

so
use 1 3.7
3.7 to ldp, so
=

2) is and as both round to

daisies found in
of square
a
X be the number
Let
X - P0(3.7) p(X 3) 0.2087(4dp)
=
=

Ex 2E
Mean and Variance of Binomial distribution
I want to know amongst 200 people the probability of a certain number being left handed,
where the probability any given person is left handed is 0.1.

𝑋~𝐵 200, 0.1 gives a binomial distribution over the possible numbers of left-handed
people.

Based on what you learnt in lower school, how many people would you expect on average
to be left-handed?
0.1x200 20
=

people up
200x0.1

The variance is much harder to work out and we won’t give the proof here.

!If 𝑋~𝐵 𝑛, 𝑝
𝐸 𝑋 = 𝜇 = 𝑛𝑝 Tip: It’s perhaps easier to
𝑉𝑎𝑟 𝑋 = 𝜎 2 = 𝑛𝑝 1 − 𝑝 remember variance as 𝑛𝑝𝑞
(where 𝑞 = 1 − 𝑝 is the
probability of failure, although
these are given the formula book)
 A fair, 4-sided die has the numbers 1, 2, 3, 4 on its faces. The die is rolled 20 times. The random

=
-

variable 𝑋 represents the number of 4s obtained.

Find the mean and variance of 𝑋.

~
(20,0.25)
Find 𝑃 𝑋 < 𝜇 − 𝜎
575) P(X(3.06..)

1
0 25 P(x < 5
=
- =

20 +
-
=

u up
=

p(X 13)
=

z2 np(1 p) 2070 25 x 0.75

= -
-
=

0.2252
=

3.75
=

⑭dp)
A company produces a certain type of delicate component. The probability of any one
component being defective is 𝑝. The probability of obtaining at least one defective component
-> -

in a sample of 4 is 0.3439.
-

The company produces 600 components in a day.

Find the mean and variance of the number of defective components produced per day.

LetXbe the
number of defective components. X-B(4,p)
0.3439 - B(4,0.1)
P(X=1)
X

1 -
p(X 0) 0.3439
= =

n 600x0
=
60 1
=

->

60x0.9
p(X 0) r' np(1 p)
=
-

0.6561
=
=
=

54
p)4
=

0.6561 (1 =
-
-
-> 0.1
=

0.9 1 -p p>
=
 p 0.35
=

David believes that 35% of people in a certain town will vote for him in the next election and he
<

commissions a survey. Find the minimum number of people the survey should ask to have a
-
mean number of more than 100 voting for David.
-

>100
up
100 a must be 286,
n + 0.35>

should ask 286 as a minimum

n > [00 he
0.35

n > 285.7...

An examiner is trying to design a multiple choice test. For students answering the test at
random, he requires that the mean score on the test should be 20 and standard deviation at
- ->

least 4. Find how many choices each question should have and the number of questions there
should be. Number of choices P
- - n
is fixed across questions and should be as small as possible.
5 4- 82 16
=
=

20 np(1 p)=16 If
p
0.2
=

#
-

n
16
20(1 p) = -

up
=
20

p>0.8 0.2n 20
=

1 -

n 10 108
P
=

0.2]
=

0-2

choices, (p 0.2) Ex 2F
=

5 100 questions
Approximating a Binomial using a Poisson
Earlier we saw how we could solve problems involving rates using a Binomial
Distribution, where we’d split up the time period into intervals where in each the event
𝜆
either happened or it didn’t, where 𝑝 = 𝑛.

2pm 3pm

✓ ✓ ✓ ✓ ✓

We saw that the Poisson Distribution is obtained when we divided time into more and
more chunks (i.e. where 𝑛 became infinitely large). Note also that 𝑝 is small because in a
very short time interval, the probability of the event occurring is low. This naturally leads
us to:
X B(n,p)
-

! If 𝑋~𝐵(𝑛, 𝑝) and: mean= np pissmall,

if
▪ 𝑛 is large
-
variance up(1
p) =
-
-
1 p=1 -

▪ 𝑝 is small
=np mean variance
Then 𝑋 can be approximated by 𝑃𝑜 𝑛𝑝
Generally if 𝑛𝑝 ≤ 10 then a Poisson is suitable enough approximation, but in an
exam, use the original Binomial unless instructed to approximate.
Why would we want to approximate?

𝑋~𝐵 𝑛, 𝑝 𝑋~𝑃𝑜(𝑛𝑝)
𝑛 𝑥
𝑃 𝑋=𝑥 = 𝑝 1−𝑝 𝑛−𝑥 𝑒 −𝜆 𝜆𝑥
𝑥 𝑃 𝑋=𝑥 =
𝑥!

𝑛 𝑛!
For the Binomial Distribution, the probability involves =
𝑥 𝑥! 𝑛−𝑥 !
When 𝑛 is really large, calculating 𝒏! is really horrid.

However, the probability function for the Poisson Distribution doesn’t involve
𝑛! so avoids the problem. Note also that 𝜆 = 𝑛𝑝 is not too large (as 𝑝 is small)
and so 𝑒 −𝜆 is not too difficult to compute.
State whether you would use a Poisson Approximation for each Binomial
(recall: 𝑛𝑝 ≤ 10), and state the distribution used as the approximation.
-

𝑋~𝐵(100, 0.1) Po (10)

𝑋~𝐵(50, 0.5) X Notsuitable
25
up
=

𝑋~𝐵(40, 0.02) Po (0.8)

0.8
np
=

𝑋~𝐵(300, 0.2)
--
* Not suitable
up 60
=
 ~

X number
=

of people buying magazine

X-1(100,0.05)
X-P0(5)P(X > 10) = 0.0137 (4dp)
=

X number
=

of letters to wrong house

B(1000, 0.01) -> X p0(10)
-
X-
P(X=4) 0.9897
=

·
 -

a) X number
=

of reds

0.8474 (4dp)
B(30,0.15) P(X = 6)
=

X - ②

b) n 60
=

up 9
=

0.15
p
=

X-4.(9)P(X(13) 08758(4dp)
=

Ex 2G
Exam Questions
AS 2020

X number
=

of call outs

2.453 = 7.2x-P0(7.2), P(X 7) 0.1486(4dp)

= =

&

2.4x8 19.2
=

X-P0 (19.2), P(X >28) 0.02199

=

0.022
=
(3dp)
->

p108 0.022
=

x
22 days
=

+-p) 0.002) 167(sdp)

=

X- p0(2.2)P(x2b) 049(4dp) =
 AS SAMs
A
X number of hives for
=

Y number
=

of hives for I

X-P0(2.6), Y-P0(.2). P(X 2)=

x P(Y 2)
=
0.25104469
= 0.21685983
x
= 0.0544
⑤

=3) 0.2046(4dp) (4dp)

W number total
= in W-Po (3.8), P(W
-
=

W =80(7 -

b),P(W<9) =
082(4dp)
->
p 0.004
=

n 608
=

up 2.4
=

up(1 p) 2.3904
- =

D number defectD-Po (2.4),

with P(D = 4) 8904) (4dp)
=

ii) up=npl1-p) i.e mean and various are roughly equal, i is large
and
small.
is