Mathematics
Bayesian Statistics
Quick fact
Bayesian statistics is named after Thomas Bayes, an 18th-century Presbyterian minister, and his theorem was published posthumously by a friend. It remained controversial for centuries but now powers everything from spam filters to self-driving cars.
Why this is interesting
You've probably heard that 'taking new evidence into account' is wise. But how do you do it mathematically? Bayesian statistics provides a precise framework for updating your beliefs as you see new data.
Read the full explanation
Understanding Bayesian Statistics
Imagine you're a detective starting a case. You have an initial hunch (a 'prior') about who the culprit might be. As you gather clues (evidence), you update your suspicion. If a clue strongly points to someone else, you adjust accordingly. Bayesian statistics does this with math. You start with a prior probability representing your belief before seeing data. Then you collect data, and you calculate the 'likelihood'—how probable the data would be if your belief were true. Bayes' theorem combines these to give a posterior probability: your updated belief after considering the evidence. It's a principled way to learn from experience, quantifying uncertainty rather than ignoring it.
A deeper explanation
At its core, Bayesian statistics treats probability as a measure of belief, not just long-run frequencies. This philosophical shift allows you to incorporate prior knowledge and update it systematically as new data arrives. The mechanism is Bayes' theorem: P(A|B) = P(B|A) P(A) / P(B), where A is a hypothesis and B is data. The theorem follows logically from the axioms of probability, not as an arbitrary rule. Its power lies in iteration: today’s posterior becomes tomorrow’s prior, enabling continuous learning. This matters because it provides a coherent framework for decision-making under uncertainty. Applications range from A/B testing (update conversion rate beliefs) to spam filtering (compute probability an email is spam given word frequencies), medical diagnosis (update disease probability given test results), and even machine learning where Bayesian methods offer uncertainty estimates critical for safe AI.