Shiksha Nation
Skip to content

Exponential dispersion model

  • 1 min read
  • 3 chapters

Related on Lably

Through Statistics, Probability distribution and Statistical inference

Definition

Univariate case

There are two versions to formulate an exponential dispersion model.

Additive exponential dispersion model

In the univariate case, a real-valued random variable X belongs to the additive exponential dispersion model with canonical parameter θ and index parameter λ, X∼ED∗(θ,λ), if its probability density function can be written as

fX(x∣θ,λ)=h∗(λ,x)exp⁡(θx−λA(θ)).

Reproductive exponential dispersion model

The distribution of the transformed random variable Y=Xλ is called reproductive exponential dispersion model, Y∼ED(μ,σ2), and is given by

fY(y∣μ,σ2)=h(σ2,y)exp⁡(θy−A(θ)σ2),

with σ2=1λ and μ=A′(θ), implying θ=(A′)−1(μ). The terminology dispersion model stems from interpreting σ2 as dispersion parameter. For fixed parameter σ2, the ED(μ,σ2) is a natural exponential family.

Multivariate case

In the multivariate case, the n-dimensional random variable X has a probability density function of the following form

fX(x|θ,λ)=h(λ,x)exp⁡(λ(θ⊤x−A(θ))),

where the parameter θ has the same dimension as X.

Sign up free to read the rest

2 more chapters are free with a Lably account.

Already have an account?

More in Statistics