Jump to content

Continuous binomial distribution

From Wikipedia, the free encyclopedia

Continuous binomial (cobin)
Parameters

θ∈ℝ (natural parameter)

λ∈{1,2,3,…} (inverse dispersion)
Support x∈[0,1] if λ=1, x∈(0,1) if λ≥2
PDF

f(x;θ,λ)=h(x;λ)exp⁡(λθx−λB(θ)),0≤x≤1,
with B(θ)={log⁡(eθ−1θ),θ≠0,0,θ=0, and

h(x;λ)=λ(λ−1)!∑k=0λ(−1)k(λk)max⁡(0,λx−k)λ−1.
Mean B′(θ)={eθeθ−1−1θ,θ≠0,12,θ=0,
Variance 1λB″(θ)={1λ(1θ2−eθ(eθ−1)2),θ≠0,112λ,θ=0.

In probability theory and statistics, the continuous binomial distribution (also called the cobin distribution) is a family of continuous probability distributions on the unit interval that belongs to an exponential dispersion family. The name continuous binomial (cobin) was adopted in 2026 in developing generalized linear models for continuous proportional data, as an alternative to beta regression [1]. The special case λ=1 coincides with the continuous Bernoulli distribution[2].

Definition

[edit]

A random variable X is said to follow a continuous binomial (cobin) distribution with natural parameter θ and inverse dispersion λ∈{1,2,…}, written X∼cobin(θ,λ−1), if it has density on [0,1] given by

f(x;θ,λ)=h(x;λ)exp⁡(λθx−λB(θ)),0≤x≤1,

where the log-partition function is

B(θ)={log⁡(eθ−1θ),θ≠0,0,θ=0,

and the base measure h(x;λ) is

h(x;λ)=λ(λ−1)!∑k=0λ(−1)k(λk)max⁡(0,λx−k)λ−1,

with h(x;1)=1. The function h(x;λ)/λ coincides with the probability density function of the Irwin–Hall distribution with parameter n=λ, evaluated at λx.

When λ is fixed, the cobin distribution belongs to a one-parameter natural exponential family in θ.

Earlier studies

[edit]

The corresponding two-parameter distributional family appeared in earlier statistical literature[1] without the name continuous binomial distribution. For example, Bates (1955)[3] studied the same distribution in the context of modeling time intervals between successive accidents and hypothesis testing against uniformity.

[edit]
  • Bates distribution: when θ=0, the density reduces to h(x;λ), corresponding to the distribution of the mean of λ independent Uniform(0,1) random variables (equivalently, a scaled Irwin–Hall distribution or Bates distribution).
  • If X1,…,Xλ are independent and identically distributed continuous Bernoulli random variables with common natural parameter θ, then
X¯=1λ∑i=1λXi∼cobin(θ,λ−1).

Properties

[edit]

Mean and variance

[edit]

The mean and variance of X∼cobin(θ,λ−1) can be expressed in terms of derivatives of B(θ):

  • E⁡(X)=B′(θ)=eθeθ−1−1θ, for θ≠0.
  • Var⁡(X)=1λB″(θ)=1λ(1θ2−eθ(eθ−1)2), for θ≠0.

If θ=0, then E⁡(X)=12 and Var⁡(X)=112λ.

Sufficient statistic for the mean

[edit]

If X1,…,Xn are independent and identically distributed continuous binomial random variables with common natural parameter θ and fixed inverse dispersion parameter λ, then the sample mean

X¯=1n∑i=1nXi

is a sufficient statistic for θ.

This is in contrast with the beta distribution: under a mean–precision parameterisation Xi∼Beta(μϕ,(1−μ)ϕ) with fixed ϕ, a sufficient statistic for the mean μ is

∑i=1nlog⁡(Xi1−Xi),

not the sample mean X¯.

Applications

[edit]

The cobin distribution has been proposed as a response distribution for generalized linear models of continuous proportional data, as an alternative to beta regression, including extensions with random effects.

References

[edit]
  1. 1 2 Lee, Changwoo J.; Dahl, Benjamin K.; Ovaskainen, Otso; Dunson, David B. (18 May 2026). "Scalable and robust regression models for continuous proportional data". Journal of the American Statistical Association. doi:10.1080/01621459.2026.2626081. ISSN 0162-1459. PMC 13188389. PMID 42169758.
  2. ↑ Loaiza-Ganem, Gabriel; Cunningham, John (2019). "The continuous Bernoulli: fixing a pervasive error in variational autoencoders". Advances in Neural Information Processing Systems. 32. Curran Associates, Inc.
  3. ↑ Bates, Grace E. (1955). "Joint Distributions of Time Intervals for the Occurrence of Successive Accidents in a Generalized Polya Scheme". The Annals of Mathematical Statistics. 26 (4): 705–720. doi:10.1214/aoms/1177728429.