Bayesian Versus Frequentist
Introduction
Welcome back to Bayesian Reasoning in the Real World! This is lesson four of seven, so we are solidly past the halfway point. In the first three lessons, we built up practical Bayesian skills: we accumulated weak cues in a spam filter, corrected the prosecutor's fallacy in a courtroom, and learned to reason qualitatively about everyday beliefs without reaching for a calculator.
All of that work rested on a specific idea — that probability represents how confident you are in a claim, and that confidence updates as new evidence arrives. But this is not the only way to think about probability. There is another long-standing tradition that defines probability purely in terms of what happens when you repeat an experiment many, many times. In this lesson, we will place these two interpretations side by side, explore what each one does well, and come away with a clear sense of when to reach for which.
What Does "70% Chance of Rain" Mean?
Before we define anything formally, consider a statement you have probably heard dozens of times: "There is a 70% chance of rain tomorrow."
One way to read it: "If we could somehow replay tomorrow under identical atmospheric conditions a huge number of times, about 70% of those replays would produce rain." Another way to read it: "Given everything I know right now — the forecast models, the satellite images, the season — my confidence that it will rain tomorrow sits at 0.70."
Both readings feel sensible, yet they rest on very different ideas about what the number 0.70 actually is. The first treats probability as a fact about repeatable events. The second treats it as a snapshot of someone's informed belief. These two readings capture the essence of the frequentist and Bayesian interpretations, respectively — and understanding the difference is the goal of this lesson.
The Frequentist Interpretation
In the frequentist view, the probability of an event is the proportion of times that event occurs over a large (in principle, infinite) number of identical repetitions. The classic example is a fair coin: we say because, as the number of flips grows, the fraction of heads converges to one-half.
This interpretation ties probability directly to observable, repeatable outcomes. A manufacturing engineer who reports "our defect rate is 2%" means that across thousands of items on the production line, roughly 2 out of every 100 turn out defective. No personal opinion is involved; the number comes from counting real events across a long run of trials.
A key feature of the frequentist view is that it applies most naturally to events that can be repeated. It works beautifully for coin tosses, dice rolls, quality-control inspections, and clinical trial outcomes — all situations that generate large pools of data under similar conditions. Where it becomes awkward is with genuinely one-off events, because there is no long run to appeal to.
