Bayes and base rates · free, no sign-up

Bayesian intuition test

Six base-rate problems, half in counts and half in percentages, with an icon array after each answer and a result that compares the two formats.

In short

The Bayesian intuition test gives six problems about a check that sometimes raises false alarms, such as a spam filter or an inspection camera, and asks what share of flagged cases are real. Three problems state the population as counts and three give only percentages, so the result shows whether the wording changed how you did. After every answer an icon array draws the same numbers as cases.

How the test works

Each problem is generated from a seed: a scenario, a base rate, a hit rate and a false-alarm rate, on the smallest population in which every count is a whole number. The four options are the right answer and three of the classic errors, such as the hit rate, which answers the reverse question, or the base rate on its own. If you pick an error, the feedback names which one it was.

What the result shows

The result reports your score on the count problems and on the percentage problems separately. Gigerenzer and Hoffrage found that the count format made correct answers much more common. Bar-Hillel's earlier work showed how easily the base rate is ignored when a specific piece of case evidence is available, which is the error the option built from the hit rate represents.

What it cannot tell you

Six problems cannot rank you against anyone, and three of each format is too few to prove a format effect in one person. The numbers in every scenario are invented, including the one about a health screening; nothing here interprets a real test result and none of it is medical advice.

Sources

Text, questions and code on this page are original to MyTestAtlas. The cited studies describe the classic procedure; none of their items is reproduced.