Base Rates and Reference-Class Forecasting

A base rate is the background frequency of an outcome across a whole class of similar cases, and reference-class forecasting is the discipline of starting from that frequency instead of the vivid details of the single case in front of you.

Ask how likely a startup is to still exist in five years and most people reason from the founder’s energy and the pitch deck. A forecaster asks a duller question first: of all startups like this one, what share were still around after five years? That number is the base rate, and it is the anchor almost every good forecast is built on.

The outside view versus the inside view

There are two ways to size up any single case. The inside view builds a story from the specifics of that case: this founder, this product, this market, this plan. The outside view ignores the story at first and treats the case as one more member of a class, asking how members of that class usually turn out. The base rate is the headline number the outside view returns.

The inside view feels richer and more insightful, which is exactly why it misleads. A compelling narrative is persuasive whether or not it is likely, and our confidence tracks the story’s vividness rather than its odds. The outside view is duller and usually far more accurate, because it is grounded in what actually happened to comparable cases rather than in how good the plan sounds.

Daniel Kahneman popularized this framing and the base-rate anchor it depends on in “Thinking, Fast and Slow”, arguing that the outside view should be your default starting point and the inside view a set of adjustments to it.

Base-rate neglect: why we drop the anchor

People routinely throw the base rate away the moment they get a specific detail. In the classic experiment, participants judged whether a described person was more likely to be a lawyer or an engineer. Told the sample was mostly lawyers, they should have leaned lawyer; instead they seized on a personality sketch that sounded engineer-ish and ignored the proportions entirely. The vivid description overrode the frequency.

Amos Tversky and Daniel Kahneman named this pattern base-rate neglect and traced it to the representativeness heuristic, our habit of judging probability by resemblance rather than by frequency, in their foundational paper “Judgment under Uncertainty: Heuristics and Biases” (Science, 1974).

The fix is not to discard the specifics, but to reverse the order of operations. Start from the base rate, treat it as your first-draft forecast, and only then let the details move you away from it, and move less than instinct demands. A detail that sounds diagnostic often is not, and a base rate that sounds boring usually carries most of the information.

The planning fallacy

The most familiar form of base-rate neglect is the planning fallacy: our chronic tendency to expect projects to finish faster and cheaper than they ever do. Everyone knows renovations run over, software slips, and infrastructure blows its budget, yet each new plan is forecast as if it will be the exception. That is the inside view winning: the detailed schedule feels achievable even though the class of similar projects almost never hits its schedule.

Bent Flyvbjerg turned the cure into a formal method for large projects, using the track record of comparable past projects to de-bias each new estimate, in “From Nobel Prize to Project Management: Getting Risks Right” (Project Management Journal, 2006), the paper that put practical reference-class forecasting on the map.

The lesson generalizes well beyond construction. Whenever you forecast your own case, from a product launch to a diet, your inside view is systematically optimistic. Anchoring on what happened to everyone else who tried the same thing is the cheapest correction available.

How to build a reference class

Reference-class forecasting is a repeatable procedure, not a talent. Here is the loop a good forecaster runs, in order.

1. Define the outcome precisely. Decide exactly what counts as the event happening (survives five years, ships on time, wins the seat) before you go looking for cases.
2. Pick a reference class. Gather past cases that resemble yours on the features that matter, broad enough to have real data, narrow enough to stay genuinely comparable.
3. Count the outcomes. Work out how often the event actually happened across that class. That share is your base rate.
4. Anchor on it. Write the base rate down as your first-draft forecast before you consider anything special about your own case.
5. Adjust for defensible differences only. Move off the base rate for factors you can justify with evidence, and move less than your gut wants to.
6. Record your reasoning. Note each adjustment so you can check later whether it helped or hurt, and tighten the process next time.

That final adjustment step, updating a base-rate prior with case-specific evidence, is Bayesian reasoning in everyday clothes. For the mechanics of updating, see Bayesian thinking for forecasters.

When the base rate is not the whole story

The outside view is your anchor, not your answer. A base rate can mislead when the class is chosen badly or the world has genuinely changed.

• Too broad a class hides the case: “startups” is less useful than “venture-backed B2B startups at seed stage.”
• Too narrow a class leaves no data: slice it thin enough and you have a class of one, which is just the inside view again.
• Structural change breaks the past: if the rules genuinely shifted, older cases may no longer be comparable, so weight recent ones more.
• Reference-class shopping is a trap: do not pick the class that gives the number you already wanted.

Balancing a stable base rate against fresh, case-specific signals is the core skill of accurate forecasting. For the wider toolkit, see how to get better at predicting the news.

Put base rates to work

Reading about base rates is easy; the habit only sticks when you forecast real questions and get scored on them. Every time you anchor on the class first and adjust second, you can watch whether the discipline actually improved your calibration.

Clutch lets you predict real news and sports with in-app credits and keeps score, so you can practice the outside view on live questions and see your accuracy add up. Get the app and start with a base rate.