The 18 patients who were wrong

Seventeen of eighteen were certain the weather controlled their arthritis. Fifteen months of measurements said otherwise. The study is from 1996 and is still uncomfortable.

The experiment bordered on rude. Two researchers found eighteen arthritis patients, and seventeen of them were convinced their joint pain followed the weather. So the researchers asked them which weather it was. The cold? The humidity? The pressure?

Then they measured for fifteen months.

The patients recorded their pain level twice a month. Doctors measured joint tenderness and function at the same time, so it didn't rest on self-report alone. Weather data were collected for the whole period. In the end, each patient's pain was compared against exactly the weather factor that patient had pointed to.

There was no statistically significant association. Not for a single one of them.

The second half of the experiment is worse

One of the researchers was Amos Tversky, who together with Daniel Kahneman had spent two decades mapping humans' systematic errors of judgement. So he didn't settle for establishing that the patients were wrong. He wanted to know why.

So they gave 97 students two number sequences and asked them to rate how strongly the two were related. The sequences were random. There was no connection at all.

The students found one anyway.

That's the whole explanation, and it isn't about arthritis. A person's intuitive sense of whether two things "go together" simply isn't the same quantity as the statistical one. We notice the days when it hurt and it rained. We don't count the days when it rained without hurting, and we don't count at all the days when neither happened. There are more of those last two kinds of days than any other, and they're what decides the matter.

What the study doesn't show

It's tempting to use the eighteen patients as a mic drop, and that would be wrong. The study is small. It's about arthritis, not headaches. It measured twice a month, which is coarse when the weather changes within hours. And since then, much larger studies have come along: when 2,658 Britons had 9,695 pain episodes measured against 81,727 control periods, there was an association with air pressure. It was just small.

So no, Redelmeier and Tversky did not prove that weather is innocent. They proved something more uncomfortable and far more useful:

Your conviction about your own correlation is not evidence for it.

It cuts both ways. Some of the eighteen may have had a perfectly real association that the experiment was too small and too coarse to catch. The point isn't that people are wrong. The point is that neither they nor anyone else can settle it by thinking back.

Why Aneroid asks when nothing is happening

That's the direct reason the app asks three times a day how you're feeling, with no pressure change having happened. Those questions feel pointless. They're the most important ones in the whole app.

Without them, you're left with only the days when both things happened — and then you're in exactly the same spot as the 97 students with their two random number sequences.