Can Achy Joints Predict the Weather? Grandpa's Knee, Tested
The short version: a prediction is tested by four numbers, not one. Over 100 example days, Grandpa's knee ached before 12 rainy days, but also on 28 dry ones, and missed 18 rainy days completely. It rained 30% of the time after an ache and 30% of the time without one. The ache is real. The forecast is not. Memory keeps the hits and throws away the rest.
Grandpa Frank, Uncle Gary's father, has a knee. The knee has opinions about the weather. "Rain tomorrow," he says, rubbing it, and he has been saying it for forty years. "Never been wrong."
After the weather app episode, Gary has switched forecasters. He now consults the knee every morning. It's free, it doesn't need updates, and it has never once shown him an advert.
Aunt Linda asked how many times the knee has been right. Grandpa said "every time". Linda asked how many times it had ached and then stayed dry. Grandpa looked at her the way people look at someone who has asked how many calories are in a birthday cake.
The four boxes
Every yes-or-no prediction can end four ways, not two. We'll use 100 example days: the knee aches on 40 of them, and it rains on 30.
| Rain next day | No rain | Total | |
|---|---|---|---|
| Knee aches | 12 (hit: "I told you!") | 28 (false alarm) | 40 |
| Knee is quiet | 18 (miss) | 42 (quiet and dry) | 60 |
| Total | 30 | 70 | 100 |
Grandpa remembers the 12 hits in vivid detail. The 28 false alarms don't feel like errors, because an ache with no rain is just a sore knee. The 18 misses don't feel like anything at all, because a knee that doesn't ache doesn't make a prediction. Only one box is memorable, and it's the one that agrees with him.
100 example days, sorted by the knee
Knee aches: 40 days · 30%
Knee is quiet: 60 days · 30%
Does the knee know anything?
Compare the rows, not the hits:
- After an ache, it rained on 12 of 40 days: 30%.
- After no ache, it rained on 18 of 60 days: 30%.
- On all days, it rained 30 times in 100: 30%.
In this example, the knee adds exactly nothing. Knowing whether it aches doesn't change the chance of rain by a single percentage point. And yet Grandpa can truthfully list a dozen occasions when it ached and then rained. Both of these are true at once. Only one of them is evidence.
Seeing a link where there isn't one, because you only remember the matching cases, is called an illusory correlation. It needs no lying and no stupidity. Just ordinary memory, and forty years.
To be fair to the knee
Our 100 days are an illustration, not Grandpa's real record. Real research on joint pain and weather has found mixed results: some studies find no link at all, others a small one with humidity and air pressure. So a real knee might know a little. The point is that nobody can tell from the hits. You can only tell from all four boxes, which is exactly what nobody in the family has ever written down.
And the ache itself is real. Nobody is arguing with the ache. We're only arguing with the forecast.
How to test any prediction
1. Write it down before you know the outcome. Afterwards, memory edits.
2. Fill all four boxes: hits, false alarms, misses, quiet days.
3. Compare the rows. Does the outcome happen more often after the signal than without it? If both rows show the same rate, the signal is decoration, however many hits you can list.
Where it costs you money
Every trading forum has a knee. "My indicator called the last three crashes." Fine. How many crashes did it call that never happened? How many real ones did it sleep through? A signal that goes off before every fall and also before many rises is not a crash detector. It's a nervous indicator. That's why we test every idea against a random choice on the same days, as in the golden cross test, instead of counting the times it looked clever.
The moral
Linda kept the four-box table for a month. She showed it to Grandpa. He studied it carefully, nodded, and said: "That was a bad month for the knee." Gary agreed. He has now combined both forecasters. When the app and the knee agree, he takes the umbrella. When they disagree, he follows whichever one says what he wanted. He has never been wrong.
The 100 days are an illustration chosen to show how the four boxes work, not a record of real weather or a real knee. Joint pain is a real medical issue; if it's getting worse, see a doctor, not a weather app. Educational content, not financial advice.