mindjump
estimation

Base rates first

Published October 29, 2026 · revised October 30, 2026 · 3 min read

When judging a single case — a project, a test result, a plan — it is natural to start from its details. The details are vivid and specific. But the most useful number is often a boring one: how often do things of this kind turn out a certain way in general?

The idea

That general frequency is the base rate. In the 1970s the psychologists Daniel Kahneman and Amos Tversky showed that people tend to neglect it. Given a detailed description, we judge by how well the case fits a story and forget how common or rare the outcome is to begin with.

Starting from the base rate is sometimes called taking the outside view. The details still matter, but they adjust a sensible starting point instead of replacing it.

Case study · Quality control

The inspection that is mostly wrong

A factory makes 1,000 parts a day, and on average 1% are defective. An automated camera inspection catches 90% of defective parts and wrongly flags only 5% of good ones. When the camera flags a part, how likely is it to be defective?

Most people guess something close to 90%. The real answer is about 15%.

9 defective, flagged1 defective, missed50 good, flagged anyway940 good, passedFlagged: 59 · actually defective: 99 / 59 ≈ 15%
One day of production, one square per part.

The reason is the base rate. There are only 10 defective parts, so even a good camera finds just 9 of them. But there are 990 good parts, and 5% of those is about 50 false alarms. Of the 59 flagged parts, only 9 are really defective. The same arithmetic applies to spam filters, fraud alerts and screening tests of all kinds: when the thing you are looking for is rare, most alarms are false, even from an accurate detector.

That does not make the camera useless. It shrinks the pile to examine from 1,000 parts to 59. It just means a flagged part should be checked again rather than scrapped.

Base rates in estimates

The same habit improves schedules and budgets. Before estimating how long your project will take, look at how long similar projects actually took. Studies of large construction and software projects have repeatedly found that most exceed their original budgets and schedules, often by a wide margin. A plan that assumes it will be the exception needs a specific reason to believe so.

0%25%50%75%100%Base rate 30%similar projects on timeAdjusted for specifics: 40%Gut feeling 80%from the project’s own story
Start from the base rate and move only as far as the specific evidence justifies (illustrative numbers).
Try it yourselfHow often is the alarm right?Change the numbers and watch the 1,000 parts re-sort.

How to use it

  1. Name the reference class: what is this a case of?
  2. Find the typical outcome for that class — a success rate, a usual duration, a usual cost.
  3. Start your estimate there.
  4. Adjust for evidence that is genuinely specific to this case, and adjust less than feels natural.
  5. Try two or three reference classes. If they disagree a lot, your uncertainty is larger than any single number suggests.

Where it misleads

Choosing the reference class is itself a judgement, and a poorly chosen class gives a confident but wrong anchor. Base rates also drift: what was typical ten years ago may not be typical now. And small samples make noisy base rates — five similar projects are a hint, not a statistic.

In one line: before the story, ask how often stories like this one end well.

Next modelReversible and irreversible decisionsMove fast through doors that open both ways; slow down at the ones that lock.
mindjump · small thinking tools, written down so I actually use them.