The Base Rate Fallacy: Why Good Forecasters Start With Frequency

Ask most people to estimate the chance of a dramatic event and they reason from the story in front of them — the vivid details, the recent headline, the compelling narrative. Ask a good forecaster the same question and they start somewhere far less exciting: how often does this kind of thing actually happen? That starting point is called the base rate, and ignoring it is one of the most common and costly mistakes in prediction. It even has a name: the base rate fallacy.

What a Base Rate Is

A base rate is the underlying frequency of an event before you know anything specific about the case in front of you. What share of startups fail? How often does the higher-ranked team win? How frequently does a favourite lose in a single match? Those long-run frequencies are the anchor a forecast should begin from. They encode everything the world already knows about how often the outcome tends to occur, stripped of the particular story.

Base rates are unglamorous precisely because they are averages. They do not care about the details of your specific case. And that is exactly why people abandon them too quickly.

The Fallacy: Getting Swept Up in the Specifics

The base rate fallacy is the tendency to discard the underlying frequency and reason almost entirely from case-specific detail — especially detail that is vivid, recent, or emotionally charged. A single dramatic anecdote overrides the fact that the event is rare. A confident description of one company overrides the fact that most companies like it do not succeed.

The mistake is not using the specifics — good forecasting absolutely uses them. The mistake is letting them completely replace the base rate instead of adjusting away from it. When you throw the base rate out, your forecast becomes a story, and stories are systematically overconfident.

A Classic Example

The textbook illustration: a test for a rare condition is 99% accurate, and someone tests positive. How worried should they be? Intuition screams "99%." But if the condition affects only 1 in 10,000 people, the base rate is so low that most positive results are false alarms — the true probability of actually having the condition can be under 1%. The specific evidence (the positive test) matters, but only as an adjustment to the base rate, not a replacement for it. Drop the base rate and you are off by a factor of a hundred.

The same trap appears everywhere forecasts are made: the base rate is boring, the specific case is compelling, and the compelling thing wins unless you consciously stop it.

How Forecasters Actually Use Base Rates

The disciplined method is a two-step move. First, find the base rate: how often does this outcome happen in the reference class of similar situations? Start your estimate there. Second, adjust — up or down — based on the specific, credible evidence for this particular case, in proportion to how strong that evidence really is. Strong, reliable evidence justifies a big move away from the base rate; weak or anecdotal evidence justifies almost none.

This is just Bayesian reasoning in everyday clothes: the base rate is your prior, the specifics are your evidence, and the forecast is the two combined honestly rather than one of them shouting down the other.

Base Rates in Our Experiment

Anchoring to frequency is quietly built into how we forecast. Our football model, for instance, starts from documented scoring rates rather than from whatever narrative surrounds a match, and only then adjusts. It keeps us from being swept up by a hot streak or a marquee name — the exact pull the base rate fallacy exploits. Whether it makes us right is a separate question, and we answer it in public: every call is locked before the event, Bitcoin-timestamped, and scored against the market afterward, with the market currently ahead of our model. You can read the full record at neuportal.ai/experiment.