Correlation is Not Causation

Ajouté : 2026-05-10

[00:00:00]Across many summers, two things have been measured to rise together. Ice cream sales climb, and so do shark attacks. Plot them month by month, and you see the same upward drift, like two friends walking in step. So, a question almost asks itself. Does ice cream attract sharks?

[00:00:19]The mathematical version of they move together is correlation.

[00:00:24]We take pairs of measurements, drop them onto a scatter plot, and ask whether the cloud leans.

[00:00:30]If the dots tilt up and to the right, we get a positive correlation, summarized in a single number, our R close to one.

[00:00:38]For our example, R is about 0.85, which is what people mean by strong positive. But, notice what R is actually doing. It looks at the cloud and reports the lean. It does not look outside the cloud. Correlation only sees the cloud, not the cause.

[00:00:57]So, let's take the lean seriously and ask, is one of these things really pulling on the other? Picture ice cream over here, sharks over there, and the question mark on the arrow between them.

[00:01:09]The honest answer is neither of them is pulling. There is a third character standing above them, hot weather. Heat sends people to the beach, and people buy ice cream. Heat also sends people into the water, where sharks already are. The summer pulls both strings. Ice cream and sharks have no arrow between them at all, only a faint dashed shadow of an association.

[00:01:36]And here's the unsettling part. The exact same correlation, that same R near 0.85, can come from very different stories. X might cause Y, Y might cause X, or some hidden Z might cause both. The number on the page is identical. The The behind the number is not.

[00:01:57]So, how do we ever escape the cloud with a verb?

[00:02:01]Instead of just observing X, we intervene on it. We reach in and set X by hand, the way a randomized experiment assigns treatments by a coin flip. The moment we do that, the arrow from the lurking cause into X is cut. Whatever X does to Y now, it does on its own. And this gives us the central inequality of causal thinking. The probability of Y given that we observed X is not in general the same as the probability of Y given that we did X.

[00:02:34]The tiny change of verb is everything.

[00:02:37]Observation tells you what tends to come along with what.

[00:02:41]Intervention tells you what would happen if you reached in and changed the world.

[00:02:46]Correlation lives entirely in the first sentence.

[00:02:50]Causation needs the second.

[00:02:53]So, the next time something seems to predict something else, ask yourself the question that breaks the spell.

[00:03:00]What would happen if I did it on purpose?

[00:03:03]And that's basically it. If you found this helpful, hit that like button, subscribe for more, and drop a comment if you have any questions. See you in the next one. Bye-bye.