Beta Distribution

Definition

The beta distribution is a two-parameter family of continuous probability distributions on the unit interval. For shape parameters \alpha,\beta>0, its density is

$$f(x;\alpha ,\beta )=\frac{1}{\mathrm{B}(\alpha ,\beta )}{x}^{\alpha -1}(1-x{)}^{\beta -1},0<x<1,$$

where \mathrm{B}(\alpha,\beta) is the beta function.

Why It Matters

It is one of the standard ways to represent uncertainty over probabilities and proportions. The family is flexible enough to express diffuse, skewed, concentrated, or U-shaped beliefs on [0,1], and it supports simple conjugate Bayesian updates for Bernoulli-style observations.

Formalism / Key Objects

• Support: X \in [0,1].
• Shape parameters: \alpha and \beta control the mass near 0, 1, and the interior.
• Mean and variance:

$$\mathbb{E}[X]=\frac{\alpha }{\alpha +\beta },\mathrm{Var}(X)=\frac{\alpha \beta }{(\alpha +\beta {)}^2(\alpha +\beta +1)}.$$

• Mean-concentration view: with \mu=\alpha/(\alpha+\beta) and \nu=\alpha+\beta, one can rewrite \alpha=\mu\nu and \beta=(1-\mu)\nu.
• Conjugacy: for Bernoulli or binomial data, a \mathrm{Beta}(\alpha,\beta) prior updates by incrementing the parameters with observed successes and failures.

Connections

• Lives under Probability and Statistics as a basic distribution family for bounded probabilities.
• Connects to Calibration because calibration concerns probabilities on [0,1], even though calibration itself is about forecast-frequency alignment rather than prior distributions.
• Related to Brier Score in the broader probabilistic-evaluation cluster, where uncertainty over event probabilities and evaluation of probabilistic forecasts often appear together.
• Useful whenever the wiki needs a lightweight Bayesian model for success probabilities rather than deterministic confidence values.

Common Confusions

• The beta distribution is a distribution over probabilities or proportions, not a scoring rule or calibration diagnostic.
• Large \alpha+\beta means higher concentration, not necessarily a larger sample size unless a specific Bayesian update story is fixed.
• A symmetric beta distribution is not always uniform; uniformity occurs only at \alpha=\beta=1.
• The beta family is limited to one bounded scalar; multicategory analogues require distributions such as the Dirichlet.

Key Sources

• Wikipedia2026 - Beta Distribution