Course contentsShow
← Mathematics
Lesson 3112 of 3,12160. Statistics — ModelingFree lesson

Prior, Likelihood, and Posterior: Bayes' Theorem Revisited

Understand how Bayes' theorem structures inference by updating prior beliefs with data to obtain posterior distributions.

Prior, Likelihood, and Posterior: Bayes' Theorem Revisited

You'll learn how Bayesian inference updates your initial beliefs with new data to form refined conclusions.

The Core Idea

Bayes' theorem provides a mathematical framework for learning from evidence. Unlike the hypothesis tests you've studied (z-tests, t-tests, chi-squared), which ask "Is the data extreme enough to reject a fixed hypothesis?", Bayesian inference asks "How should I update my beliefs given this new data?"

The process has three components:

Prior: Your initial belief about a parameter before seeing data. This could be based on previous studies, expert opinion, or even a "neutral" starting point. Think of it as your educated guess before the experiment.

Likelihood: How probable the observed data is under different parameter values. This is the same likelihood concept from maximum likelihood estimation—it quantifies what the data tells you.

Posterior: Your updated belief after combining the prior with the likelihood. This is the refined conclusion that incorporates both your initial knowledge and the new evidence.

The Formula Structure

Bayes' theorem combines these:

Posterior ∝ Prior × Likelihood

(More precisely: Posterior = Prior × Likelihood / Evidence, where Evidence normalizes everything to sum to 1)

A Simple Analogy

Imagine you're a doctor assessing if a patient has a rare disease. Your prior is the disease's base rate in the population (say, 1%). The likelihood is how consistent the patient's symptoms are with the disease versus other causes. The posterior is your updated probability that this specific patient has the disease after observing their symptoms—often very different from the base rate!

Key Takeaway: Bayesian inference systematically updates prior beliefs using data through the likelihood, producing posterior distributions that represent refined knowledge about parameters.