liveBook · Manning

Multinomial likelihood

As explained in Grokking Bayes

A multinomial likelihood is the likelihood function for counts across multiple mutually exclusive outcomes over a fixed number of trials.

When a trial can end in several mutually exclusive outcomes, the multinomial likelihood describes the observed counts across those outcomes. It generalizes the binomial likelihood beyond two outcomes. section 3.9

Three possible outcomes

Suppose three outcomes occur across n trials, with counts n₁, n₂, and n₃. A Dirichlet prior such as Dir(1, 2, 1) updates to Dir(n₁ + 1, n₂ + 2, n₃ + 1), adding each outcome’s observed count to its corresponding prior hyperparameter. section 3.9

The multinomial likelihood pairs naturally with the Dirichlet distribution, just as the binomial likelihood pairs with the beta distribution. This conjugate relationship makes the posterior update a direct count adjustment. section 3.9

How Grokking Bayes covers it

How deep it goes
Defines the likelihood through outcome counts and explains its Dirichlet conjugate update, without deriving the likelihood formula.
Examples
Uses three outcomes with counts n₁, n₂, and n₃ and a Dir(1, 2, 1) prior.
What you can do after
Evaluate/understand posterior count updates for multiple-outcome Bayesian models.
How it connects
Connects multinomial likelihood to binomial likelihood, Dirichlet distribution, beta distribution, and conjugate priors.

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.