Scribd
Upload
49 views

Poisson

The Poisson distribution is used to model count data and assumes the mean and variance of the response variable y are equal to λ. The expected value λ is linked to the linear predictor η such that λ(η) = E*exp(η), where E is a known constant. There are no hyperparameters for the Poisson distribution. It is specified using the family="poisson" argument and requires the response y and the offset E as arguments. An example simulation and model fitting using INLA is also shown.

Uploaded by

vsuarezf2732
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
49 views

Poisson

The Poisson distribution is used to model count data and assumes the mean and variance of the response variable y are equal to λ. The expected value λ is linked to the linear predictor η such that λ(η) = E*exp(η), where E is a known constant. There are no hyperparameters for the Poisson distribution. It is specified using the family="poisson" argument and requires the response y and the offset E as arguments. An example simulation and model fitting using INLA is also shown.

Uploaded by

vsuarezf2732
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 1

Poisson

Parametrisation
The Poisson distribution is
Prob(y) =

y
exp()
y!

for responses y = 0, 1, 2, . . ., where


: the expected value.

Link-function
The mean and variance of y are given as
=

and

2 =

and the mean is linked to the linear predictor by


() = E exp()
where E > 0 is a known constant (or log(E) is the offset of ).

Hyperparameters
None.

Specification
family = poisson
Required arguments: y and E

Example
In the following example we estimate the parameters in a simulated example with Poisson responses.
n=100
a = 1
b = 1
z = rnorm(n)
eta = a + b*z
E = sample(1:10, n, replace=TRUE)
lambda = E*exp(eta)
y = rpois(n, lambda = lambda)
data = list(y=y,z=z)
formula = y ~ 1+z
result = inla(formula, family = "poisson", data = data, E=E)
summary(result)

Notes
This likelihood also accept E = 0 and in this case log(E) is defined to be 0. Non-integer values
of y 0 is accepted although the normalising constant of the likelihood is then wrong (but its a
constant only).
1

You might also like