liveBook · Manning

Uniform distribution

As explained in Grokking Bayes

A probability distribution that assigns equal probability or density across its support.

Suppose you have several possible values for an unknown quantity and no reason to favor one. A uniform distribution gives each value in a finite set the same probability. If there are n possibilities, each gets 1/n. Used as a prior, this expresses initial indifference among those hypotheses—not a claim that the prior contains no assumptions. section 3.2 section 2.9

A quantity that can take any value between 0 and 1 needs a different treatment. There are infinitely many possible values, so assigning a positive probability to each would make the total exceed 1. A continuous uniform distribution instead uses a constant probability density. Probabilities belong to ranges: the interval from 0.1 to 0.2 occupies one tenth of the full range, so it has probability 10%. Any single exact value has probability zero. section 3.2 section 3.3

At least halfway

Imagine eleven equally likely candidate values: 0, 0.1, …, 1. Six are at least 0.5, giving a discrete probability of 6/11, or about 55%. If every value between 0 and 1 is possible under a continuous uniform distribution, the range from 0.5 to 1 occupies half the total range, giving 50%. section 3.3

Uniform is a useful prior when equal initial weight across a bounded set or range makes sense. It cannot serve as a uniform prior across an unbounded variable: the total probability would not remain one. A uniform distribution from 0 to 1, also written Beta(1, 1), can also represent one proposed model of a conversion rate when comparing models with a Bayes factor. section 2.9 section 11.5

How Grokking Bayes covers it

How deep it goes
Works from finite 1/n probabilities to continuous density and interval integration, then compares discrete and continuous answers.
Examples
Compares probabilities above 0.5 on an eleven-value grid and a continuous range, then considers a website conversion rate.
What it assumes
Introduces equal-probability finite hypotheses before using integration to explain continuous interval probabilities.
What you can do after
Evaluate/understand when a uniform prior is suitable and how to interpret its probabilities for finite sets versus continuous ranges.
How it connects
Connects prior distributions with probability density, integration, the Beta distribution, and Bayes factors.

Also explained in

Compare how 10 books explain uniform distribution →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Grokking Bayes. The section links point to the parts of the book it draws on.