The brain finds patterns in pure randomness

97 students were given two number sequences with no connection between them and asked to find one. They did.

Two researchers gave 97 students two rows of numbers. The rows were random — no connection, no structure, just noise. The task was to rate how strongly the two rows were related. The students found a connection anyway, and they were in no doubt about it.

That was the half of Redelmeier and Amos Tversky's 1996 experiment that wasn't about patients, but about the rest of us. The other half — eighteen arthritis patients convinced their pain followed the weather — you can read about elsewhere. Here it's about why. Why a person who has never seen a particular dataset before can find a pattern in it within minutes, even when the pattern isn't there.

The sum you never do

Picture four kinds of days. Days when it both rains and hurts. Days when it rains without hurting. Days when it hurts without rain. And days when neither happens.

To know whether rain and pain actually go together, you have to count all four. No one does. The brain notices the first kind of day — the coincidence — and forgets the other three, because they aren't interesting to remember. The researchers behind the experiment themselves point out that a person's intuitive notion of correlation diverges from the statistical one, and that this divergence helps create the belief that pain is governed by the weather.

Can the blind spot itself be measured?

Redelmeier and Tversky's experiment was the first, but not the last. A laboratory study from 2020 set out to test the same ability more directly: how well can a person even read the degree of correlation between a given trigger and headache, when the subject is guided by experimental headache calendars rather than their own everyday life?

That's a different question from "are the patients wrong". It's the question: can a person, with the best will in the world, even read the degree of a correlation correctly from their own experience? Or is the tool itself — memory, pattern recognition — too coarse for the task, no matter how attentive you are?

A belief can be learned

A follow-up study from the same research group took a different route in 2023. It examined how people learn a belief in a trigger — through repeated pairings between a possible trigger and an attack. Not whether the belief is true. Whether it can arise simply because two things happened together a few times.

That's an important distinction. A conviction that "X triggers my headache" can be completely genuine, completely unreflective, and completely out of step with what's actually going on — at the same time. That's not lying. It's how an associative system works when it gets no outside help counting correctly.

What the app does instead

That's why Aneroid doesn't ask you to remember. The app logs the pressure when it happens, and asks you when it happens — and just as important: it also asks when nothing is happening, so the days when nothing happened get counted too. That's the boring part of the sum that the brain otherwise drops.