Natural exponential family
From Wikipedia, the free encyclopedia
In probability and statistics, a natural exponential family (NEF) is a class of probability distributions that is a special case of an exponential family (EF). Many common distributions are members of the natural exponential family, and the use of such distributions simplifies the theory and computation of generalized linear models.
Contents |
[edit] Definition
[edit] Univariate case
A univariate natural exponential family is a set of probability distributions whose probability density function (or probability mass function, for the case of a discrete distribution) can be expressed in the form
where h(x) and A(θ) are known functions, and θ is the parameter.
Note the similarity to an exponential family:
Thus a natural exponential family is an exponential family in which the natural parameter η and the natural statistic T are both the identity.
[edit] Multivariate case
Suppose that , then a natural exponential family of order p has density or mass function of the form:
where in this case the parameter
[edit] Examples
Examples of univariate cases include:
- Poisson distribution
- exponential distribution
- binomial distribution with known number of trials
- normal distribution with known variance
- gamma distribution with known shape parameter
- negative binomial distribution with known r
Note that most of these distributions will require a different parameterization than is listed in the above linked pages. For example in the Poisson case, the two parameterizations are related by θ = logλ, where λ is the mean parameter, and so that the density may be written as
for , so
- , and
An example of the multivariate case is the multinomial distribution (again with known number of trials).
[edit] Properties
[edit] Moment and cumulant generating function
A member of a natural exponential family will have moment generating function of the form:
and so cumulant generating function:
[edit] Mean and variance
From the cumulant generating function, it follows that in the univariate case, the distribution has mean and variance:
- and
In the multivariate case, the mean vector and covariance matrix are thus:
- and
where is the gradient and is the Hessian.
[edit] Additivity
The convolution densities from a natural exponential family are a natural exponential family. If are independent identically distributed from a natural exponential family, then:
is from a natural exponential family. This follows from the properties of the cumulant generating function.
[edit] Other
The variance function for random variables with an NEF distribution can be written in terms of the mean.
- Var(X) = V(μ)
The first two moments of a distribution uniquely characterize it.
- X˜NEF[μ,V(μ)]
[edit] Quadratic Variance Functions
Six NEFs have quadratic variance functions (QVF) in which the variance of the distribution can be written as a quadratic function of the mean. These are called NEF-QVF. The properties of these distributions was described by Carl Morris.
- Var(X) = V(μ) = ν0 + ν1μ + ν2μ2.
The six QVF-NEFs are:
1. A subset of all Normal distributions N(μ,σ2) has constant variance V(μ) = σ2.
2. Poisson Pois(μ) with linear V(μ) = μ.
3. Gamma Gam(r,λ) with μ = rλ and V(μ) = μ2 / r.
4. Binomial Bin(n,p) with μ = np and V(μ) = − μ2 / n + μ.
5. Negative binomial NegBin(n,p) with μ = np / (1 − p) and V(μ) = μ2 / n + μ.
6. The (not very famous) distribution generated by the generalized hyperbolic secant distribution (NEF-GHS) has V(μ) = μ2 / n + n and μ > 0.
One property of the NEF-QVFs is that they are closed under linear transformations.
[edit] References
- Morris C. "Natural exponential families", Encyclopedia of Statistical Sciences.
- Morris C. (1982) "Natural exponential families with quadratic variance functions". Ann. Statist., 10(1), 65--80.
- Morris C. (1982) Natural exponential families with quadratic variance functions: statistical theory. Dept of mathematics, Institute of Statistics, University of Texas, Austin.