Scribd
Upload
24 views2 pages

Gaussian: Parametrisation

The Gaussian distribution is used to model continuously valued responses y in a generalized linear model. It has two key parameters: the mean μ, which is linked to a linear predictor η, and the precision τ. The precision τ is assigned a log-gamma prior and estimated as the hyperparameter θ = log(τ). An example is provided where Gaussian responses y are simulated with mean η = a + b*z and precision τ = 100. These data are then modeled using INLA with a log-gamma prior on the precision θ.

Uploaded by

Trilce
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
24 views2 pages

Gaussian: Parametrisation

The Gaussian distribution is used to model continuously valued responses y in a generalized linear model. It has two key parameters: the mean μ, which is linked to a linear predictor η, and the precision τ. The precision τ is assigned a log-gamma prior and estimated as the hyperparameter θ = log(τ). An example is provided where Gaussian responses y are simulated with mean η = a + b*z and precision τ = 100. These data are then modeled using INLA with a log-gamma prior on the precision θ.

Uploaded by

Trilce
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/ 2

Gaussian

Parametrisation
The Gaussian distribution is



s
1
2
f (y) = exp s (y )
2
2
for continuously responses y where
: is the the mean
: is the precision
s: is a fixed scaling, s > 0.

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

2 =

and

and the mean is linked to the linear predictor by


=

Hyperparameters
The precision is represented as
= log
and the prior is defined on .

Specification
family = gaussian
Required arguments: y and s (argument scale)
The scalings have default value 1.
Hyperparameter spesification and default values
hyper
theta
name log precision
short.name prec
initial 4
fixed FALSE
prior loggamma
param 1 5e-05
to.theta function(x) log(x)
from.theta function(x) exp(x)
1

1
s

survival FALSE
discrete FALSE
link default identity logit log
pdf gaussian

Example
In the following example we estimate the parameters in a simulated example with Gaussian responses,
giving a Gamma-prior with parameters (1, 0.01) and initial value (for the optimisations) of exp(2.0).
n=100
a = 1
b = 1
z = rnorm(n)
eta = a + b*z
tau = 100
scale = exp(rnorm(n))
prec = scale*tau
y = rnorm(n, mean = eta, sd = 1/sqrt(prec))
data = list(y=y, z=z)
formula = y ~ 1+z
result = inla(formula, family = "gaussian", data = data,
control.family = list(hyper = list(
prec = list(
prior = "loggamma",
param = c(1.0,0.01),
initial = 2))),
scale=scale, keep=TRUE)
summary(result)

Notes
None.

You might also like