Picture a café you walk past every morning: warm light, a queue out the door, a barista who knows the regulars by name. Someone asks you, “Will this place still be open in five years?” The story in front of you screams yes — it’s clearly lovely, clearly busy, clearly run by people who care. But that story is one café. The honest forecast starts somewhere much less romantic: of every new café that opens, only a minority survive five years. That dull, crowd-level number is your anchor — and ignoring it for the warm light in the window is one of the most common and most expensive thinking errors there is.
This lesson is about that anchor: where it comes from, how to find the right one, and why your brain is wired to throw it away exactly when a vivid case shows up.
Before you read — take a guess
Steve is described by a former neighbor: shy, withdrawn, meticulous, with a need for order and a passion for detail; little interest in people. Is Steve more likely to be a librarian or a farmer?
The outside view: start from the crowd
Before you study what’s special about a case, look at the crowd it belongs to. A base rate is how common something is in the relevant reference class — the plain frequency of an outcome across all the cases like yours — before you look at any of the specifics of your particular case. It’s the answer to “what usually happens to things like this?”
The idea comes straight from the work of psychologists Daniel Kahneman and Amos Tversky, who showed across dozens of experiments that people systematically ignore these background frequencies in favor of the vivid story. Kahneman’s shorthand for the discipline is the outside view: instead of asking “how will my project go?” (the inside view, which lives inside the details of your specific plan), you ask “how do projects of this type go, on average?” and start your estimate there.
Why start there? Because the base rate already contains the accumulated outcomes of thousands of cases that looked, to their owners, just as special as yours feels to you. The café owner who’s about to fail is also gazing at warm light and a queue. The crowd has already run the experiment you’re about to run, many times, and written down the answer.
Worked example. A friend pitches you on a startup: brilliant founder, hot market, real revenue. Should you expect it to become a billion-dollar “unicorn”? The inside view stacks up the impressive specifics and feels great. The outside view asks first: of all venture-funded startups, what fraction reach a billion-dollar valuation? It’s well under 1% — on the order of 1 in 100 or rarer, depending on how you count. That number is your starting estimate. The brilliant founder and hot market might nudge it up a bit, but they start from under one percent, not from zero-to-anything.
Drag the grid below to feel how small a “1 in 100” base rate actually is — and how the single vivid case in front of you is just one of those dots.
Base rate
Roughly 1 in 100
Drag the slider. Each dot is one funded startup. Light up how many you think become unicorns — then check it against reality.
1 in 100 — 1% — are startups that become a billion-dollar “unicorn”.
The one-line habit
Before you judge any single case, finish this sentence first: “Things like this usually turn out ___.” That blank is the base rate, and it’s your starting estimate — everything specific about the case is an adjustment to it, not a replacement for it.
The base-rate fallacy: when the vivid story wins
Here’s the failure mode that makes base rates worth a whole lesson. The base-rate fallacy (also called base-rate neglect) is the tendency to ignore the background frequency of an outcome when a specific, vivid, representative description is dangled in front of you. The description feels like better information because it’s concrete and detailed — so you reach for it and quietly drop the crowd-level number that should have anchored you.
Kahneman and Tversky traced this to a mental shortcut they named representativeness: we judge how likely something is by how much it resembles our mental prototype, instead of by how common it is. A tidy, detail-obsessed man resembles the “librarian” prototype, so we call him a librarian — never pausing to count how few librarians there are.
Worked numbers (the librarian problem). Farmers vastly outnumber librarians — call it 1 librarian for every 19 farmers among the relevant men. Take 1,000 men: about 950 farmers and 50 librarians (already generous to librarians). Now grant the stereotype real force: a meticulous, orderly man is four times as likely to be a librarian as a random man. So among librarians, maybe 40% fit the description → 20 men; among farmers, maybe 10% fit → 95 men. Of the men who match “shy and tidy,” 20 are librarians and 95 are farmers — so the matching man is still over four times more likely to be a farmer. The stereotype “won” on resemblance and lost on arithmetic, because the farmer crowd is so much bigger.
A second, sharper case — the airport-threat screen. A scanner flags a traveler who “fits the profile” of a threat. Even a very accurate system runs into a brutal base rate: of the millions of travelers a day, genuine threats are vanishingly rare — perhaps a handful in tens of millions. When the base rate of the thing you’re hunting is one in a million, a “profile match” or a machine alarm is, overwhelmingly, a false alarm. The vivid feeling (“this one looks dangerous”) is no match for how astronomically rare the real thing is.
A test for a rare disease is 99% accurate. The disease affects 1 in 1,000 people. Someone you don’t know tests positive. Roughly how likely is it that they actually have the disease?
Reference classes: choosing the right crowd
The base rate is only as good as the crowd you count. A reference class is the set of cases you treat as “like this one” — the group whose frequencies you borrow as your starting estimate. Choosing it well is most of the skill, because the same case can sit in many crowds with very different base rates.
This is the live tension between two views Kahneman named. The inside view looks at the unique details of your specific case and builds an estimate from them (“our team is great, our plan is solid, so we’ll finish in three months”). The outside view ignores the details at first, finds the right reference class, and reads off its base rate (“renovation projects like ours finish 40% over schedule on average, so budget for that”). The inside view feels more informed and is usually more wrong, because it has no way to account for the unknown unknowns that the base rate has already absorbed from real history.
The classic casualty is the planning fallacy: people predict their own projects will go faster and cheaper than nearly identical past projects did, because they reason from the rosy inside view of this plan instead of the track record of all such plans. The fix is mechanical — find a reference class of completed similar projects, take their real timelines as your anchor, and only then adjust.
Worked example. You’re estimating a kitchen renovation. Inside view: “Three weeks — the contractor said so, the design is simple.” Outside view: pull the reference class “kitchen renovations like mine” and you’ll find a strong tendency to overrun. If similar jobs took five weeks on average, five weeks is your anchor, and “but ours is simpler” is a small downward nudge — not a license to believe the three-week story. The reference class did the forecasting; your optimism just edits the margins.
The catch: pick a class that’s too narrow and you have no data (“renovations of my exact kitchen” is a sample of one); too broad and the number is meaningless (“all home projects ever”). Aim for the tightest class that still has enough real cases to give a stable frequency.
You want a base rate for how long your specific software project will take. Which reference class is the BEST starting point?
Combining the base rate with the specifics
Starting from the base rate doesn’t mean stopping there. The crowd gives you the anchor; the specifics of your case move you off it — up if your evidence is genuinely favorable, down if it’s unfavorable. The base rate is where the estimate begins, not where it ends.
The discipline is the order of operations: anchor first, adjust second. A first-rate founder doesn’t turn a 1-in-100 unicorn shot into a coin flip, but it might double your estimate to 2 in 100. A glowing café in a wealthy, under-served neighborhood doesn’t guarantee five-year survival, but it might lift you above the dismal average. The mistake the base-rate fallacy describes is skipping straight to the specifics and forgetting there was ever an anchor at all.
How much you should move off the base rate for a given piece of evidence — that’s a precise question with a precise answer, and it has a name: Bayesian updating, the engine for combining a prior (your base rate) with new evidence (the specifics) into a sharper estimate. That machinery gets a full course of its own later. For now, plant just the shape of it: start at the base rate, then move toward the specifics in proportion to how strong the evidence really is — strong evidence moves you a lot, weak or vivid-but-uninformative evidence moves you barely at all. The disease test above is the whole idea in miniature: a positive result did move you (from 0.1% up to ~9%), just nowhere near as far as your gut wanted, because the base rate held most of the weight.
Base rate
Even after a positive test
Imagine 100 people who all tested positive for a rare disease. Drag to set how many you think truly have it.
3 in 100 — 3% — are who actually have the rare disease.
Where base rates lie to you
Base rates are powerful, not infallible. Three ways the outside view can mislead — each a reason to check the anchor, never to abandon it.
The wrong or too-broad reference class. If your crowd doesn’t actually match your case, its frequency is just a confident-looking irrelevance. “Restaurants fail at rate X” is the wrong anchor for a 30-year-old family institution with a paid-off building; “all drivers crash at rate Y” is the wrong anchor for a professional with a million accident-free miles. Garbage class in, garbage base rate out. The fix isn’t to drop the base rate — it’s to find a tighter, truer class.
Base rates that have genuinely changed. A reference class built from history assumes the world hasn’t moved. Sometimes it has. The base rate of “letters lost in the mail” tells you nothing about email; the survival rate of pre-internet retailers misforecasts an online one. When the underlying conditions shift structurally, old frequencies can mislead — so ask whether the past cases were generated by the same world your case lives in.
Truly unique events with no reference class. Some cases have no honest crowd: a genuinely first-of-its-kind technology, a singular historical turning point. Here the outside view runs out of data, and you’re forced back toward careful inside-view reasoning — but the danger is the reverse of base-rate neglect: people declare their case “unique” precisely to escape an inconvenient base rate. The café owner insists their café is special; the over-budget project insists it’s different this time. Most cases that feel unique aren’t.
“But this one is different” is the trap’s favorite disguise
The single most common way smart people fall back into base-rate neglect is by ruling their case exceptional — special enough that the crowd’s frequency “doesn’t apply.” Sometimes that’s genuinely true. Far more often it’s the inside view talking, dressing up ordinary optimism as a reason to ignore an uncomfortable number. Make the case earn the exemption: name specifically what makes it differ from its reference class, and check that the difference is real, not just flattering.
When to reach for it
Reach for base rates every single time you forecast or judge one specific case — a hire, an investment, a diagnosis, a project deadline, a relationship, a bet. The trigger is any moment you catch yourself reasoning from a vivid, detailed story about this one. Before you let the story drive, stop and ask: what usually happens to things like this? Find the crowd, read its frequency, make that your anchor — and only then let the specifics adjust it. It costs ten seconds and routinely saves you from the most confident wrong answers you’ll ever produce.
Which of these is the clearest, correct use of base-rate thinking?
Recap
Check yourself: base rates
What is a base rate?
Check your answer to continue.
Where this goes next
You can now do the move that separates calibrated forecasters from confident guessers: before you judge any single case, find the crowd it belongs to and start from how often things like it turn out the way you’re asking about. Anchor on the base rate, then let the specifics adjust it — never the other way around. Next, in “Expected Value,” we put numbers on the outcomes themselves: once you know how likely each result is, expected value tells you what a choice is worth on average, so probabilities stop being trivia and start driving decisions.