1.3 - Correlation Does Not Imply Causation and Why

In this part of the Introduction to Causal Inference course, we walk through why correlation does not imply causation. You have probably heard this mantra before, but we go a bit more in depth here. Please post questions in the YouTube comments section.

[janetkipova]

This channel is amazing. Thank you for doing this!

[JGP]

thanks for your work. I am a data scientist at IAC and this is helpful in my research to determine causality revenue changes in app updates

[yasamanasgari]

Thanks for your great explanation. It is 10000 out of 10.

[LeahBarrett-dqud]

The issue for me here is... No shoe wearing didn't cause headaches but one could say "drinking before bed causes headaches and I'd hear the same comment - correlation does not imply causation. - It doesn't ever make sense because yes at some point it DOES! This is like saying cause doesn't have an effect or there is not an equal reaction.

[ClassicJukeboxBand]

A causal association would be something like "hitting yourself in the head with a hammer tends to cause headaches."

[vicpinto]

I always knew Cage was a bastard. This video confirms it. 😃

[soumyasarkar]

this is some brilliant explainations and slide making... any good resources on the topic you think is a must read ?

[MrSammer]

Co2 correlation to temperature increase

[ToddJohnson]

With respect to Nicolas Cage and pool drownings, I am reminded of the opening narrator in Magnolia: It is in the humble opinion of this narrator that this is not just "something that happened." This cannot be "one of those things"... This, please, cannot be that. And for what I would like to say, I can't. This was not just a matter of chance. ... These strange things happen all the time. ;-)

[anzhang]

Thanks so much for the example. Do you have any mathematical examples to explain the casual and confounding association?

[kevinkorb]

This presentation of the relation between correlation and causation is misleading. Yes, neither prediction nor correlation imply direct causation between the two variables in question. They both, however, imply the two variables are connected in a causal nexus, whether direct, indirect or via a common ancestor. That's the basis of all causal discovery (causal machine learning).

Even the "cognitive bias" of inferring a direct causal connection is not simply something that should be dismissed as a kind of fallacy. The inference isn't deductive, but abductive, a kind of inference to the best explanation, meaning something often worth using as a working hypothesis. By not mentioning these points, Neal is doing his audience a disservice.

If you want to know more, look up Hans Reichenbach's Principle of the Common Cause, Peirce on abduction, Pearl's "The Book of Why". The statisticians' mantra of "Correlation doesn't imply causation" is long past its use-by date.