Linear transformation of Poisson

Distribution
Dr Richard Kenderdine

Kenderdine Maths Tutoring

www.kenderdinemathstutoring.com.au
The Poisson distribution for a random variable X with parameter λ has the probability function
λx ⅇ-λ
P (X = x) = (1)
x!
Now suppose we have another random variable Y defined by Y = aX + b ie a linear transformation
of X. The probability function for Y is given by
y-b
λ a ⅇ-λ
P (Y = y) = (2)
y-b
 a
!
It is important to recognise that while X can take on the values 0, 1, 2, .... ie non-negative integers,
the values that Y can take are determined by the transformation Y = aX + b. Hence to calculate the
expected value of Y, E(Y), we cannot use the usual expression for a Poisson distributed random
variable:
∞
E (Y ) = y P (Y = y) (3)
y=0

Instead we have to use only those values that Y can take under the transformation and we use
∞
E (Y ) = yi P (Y = yi ) (4)
i=1

Calculation of Expected Value and Variance

The calculation of the Expected Value and Variance of the transformed variable Y follow the usual
rules:

E (Y ) = E (a X + b) = a E (X ) + b = a λ + b
(5)

Var (Y ) = Var (a X + b) = a2 Var (X ) = a2 λ

(6)

We can also calculate

EY 2  = E  (aX + b)2  = Var (Y ) + ( E(Y ))2 = a2 λ + (aλ + b)2 (7)

2 Linear Transformation Poisson Distribution.nb

A function to calculate Expected Value and Variance

Here is a function in Mathematica to calculate the Expected Value and Variance for a general linear
transformation of the Poisson distribution:

lintrpoisson[λ_, a_, b_] :=

y = Table[a x + b, {x, 0, 99}];

y[[k]]-b
100
ⅇ-λ λ a
expect =  y[[k]] ;
y[[k]]-b
k=1  a
!

y[[k]]-b
100
ⅇ-λ λ a
expsq =  (y[[k]])2 ;
y[[k]]-b
k=1  a
!

Print"E(Y)= ", expect, " E(Y2 )= ",

expsq, " Var(Y)= ", expsq - expect2 

Example
Suppose we have the transformation Y = 0.25 X + 7.

The values of Y corresponding to X = 0, 1, 2, 3,..... are Y = 7, 7.25, 7.5, 7.75,......

We run the function lintrpoisson with the input parameters λ = 2, a = 0.25 and b = 7:

lintrpoisson[2, 0.25, 7]
E(Y)= 7.5 E(Y2)= 56.375 Var(Y)= 0.125

The calculated values agree with (5) and (6): E(Y) = 0.25×2 + 7 = 7.5 and Var(Y) = 0.25 2 ×2 = 0.125

Probability density plots for X and Y are shown below. The only difference is the scale on the horizon-
tal axis as defined by the transformation:
 Linear Transformation Poisson Distribution.nb 3

P(x)

0.25

0.20

0.15

0.10

0.05

x
2 4 6 8 10

P(y)

0.25

0.20

0.15

0.10

0.05

y
7.5 8.0 8.5 9.0

A conditional probability concerning two

Poisson processes
Suppose we have two independent random variables X and Z that have Poisson distributions with
parameters λ1 and λ2 respectively and we calculate Y = X + Z.

We want the probability that X = k given that Y = n ie P(X = k | Y = n ) with 0 ≤ k ≤ n.

P(A and B)
The standard result for conditional probability is P ( A | B ) = so here we have
P(B)

P (X = k and Y = n)
P (X = k Y = n) = (8)
P ( Y = n)
Now the sum of two Poisson distributions with parameters λ1 and λ2 is also a Poisson distribution
with parameter λ1 + λ2 , hence
(λ1 + λ2 )n ⅇ-(λ1 +λ2 )
P ( Y = n) = (9)
n!
If X = k and Y = n then Z = n - k and, since both X and Z are independent,

λ1k ⅇ-λ1 λ2 n-k ⅇ-λ2

P (X = k and Y = n) = P (X = k and Z = n - k ) = × (10)
k! (n - k) !
4 Linear Transformation Poisson Distribution.nb

Then
λ1 k ⅇ-λ1 λ2 n-k ⅇ-λ2
k!
× (n-k)!
P (X = k Y = n) =
(λ1 +λ2 )n ⅇ-(λ1 +λ2 )
n!
k n-k
n λ1 λ2 (11)
=
k (λ1 + λ2 )n

k n-k
n λ1 λ2
=
k λ1 + λ2 λ1 + λ2
λ1
which is a binomial probability function with p =
λ1 +λ2