Correlation CAN Imply Causation! | Statistics Misconceptions

This video is about how causal models (which use causal networks) allow us to infer causation from correlation, proving the common refrain not entirely accurate: statistics CAN be used to prove causality! Including: Reichenbach's principle, common causes, feedback, entanglement, EPR paradox, and so on.

REFERENCES:

MinutePhysics is on twitter - @minutephysics

Minute Physics provides an energetic and entertaining view of old and new problems in physics -- all in a minute!

Created by Henry Reich

[EpicDoughnut]

I still prefer Randall Munroe's "Correlation doesn't imply causation, but it does wiggle its eyebrows suggestively while mouthing 'look over there' "

[KundelX]

Minutephysics - because explaining everything with cats is an option.

[owbu]

"except maybe not in Quantum Mechanics" is probably what the book on Quantum Mechanics should be called.

[criticalapps]

"Correlation doesn't necessarily imply causation but it can if you analyse it with causal models, except maybe not in quantum mechanics?"

Catchy.

[extaxt]

Correlation implies causation if that correlation is the result of a trial (experiment) in which the variables are controlled and experimental units are randomly assigned and independent. What is being described in the video is an observational study: 'look at world and record data' and yes narrowing down causal relationships in this case has to be done more carefully and never assumed.

[Nightmare]

What about unknown unknowns? Spuriousness isn't just about the island, it can pertain to anything (star-sign, solar activity, elections, stock market etcetc). How do we decide in Cats (C) and Height (H) that its Island (I) that is the third factor and not some other factor X? Or Y? Or Z?

[EphraimAtkinson]

Wait - you didn't rule out the case of coincidence (the 20th variation)

[myopinionsarefacts]

Quantum mechanics:where the universe decided that laws are more like, suggestions

[tezzeret]

Good point, but in order to truly demonstrate causality in the method described, we would need to account for all possible influencing factors in the causal map. The lack of our ability to take literally everything into account epistemologically limits our knowledge to "best guesses" about causation.

[SableGear]

I always heard the saying as "Correlation does not EQUAL Causation." correlation implying causation is possible, but there's no guarantee it's a definite relationship.
Anyway, thank-you for the breakdown!

[omargoodman]

"Correlation doesn't prove causation, but it's certainly a hint."

[Zagardal]

"I know that sounds kinda nerdy" and that's supposed to be an issue? I mean, we're already here dude.

[thomasmichaels]

Yes, thank you. So many people hear this phrase "correlation doesn't imply causation", and then just repeat it without fully understanding it. This is much welcomed.

[Eta_Carinae__]

I think the short cut of "doesn't" is down to there being both a "doesn't necessarily" and "doesn't sufficiently" both being true.

[xystem]

As I've heard it, the saying is "Correlation does not prove causation", the important distinction being prove instead of imply. Which sort of nullifies the whole point you're trying to make, but I mean it's still a great point to show

[BrianCooperpiece]

It's funny that you came out with this video today. I actually had this thought, prior to watching this, where I was walking back to my car and noticed a spilled coffee cup next to the door of another car. It occurred to me that the possibilities exist where both of them have nothing to do with each other and were it was completely the their fault. I just think it's awesome how often this can be used in our daily lives!

[mrKreuzfeld]

problem is unknown causes. In the example, you assume that you know everything there is to know about the islands. In the real world, there is always some other unknown cause that also will correlate.

[mac]

I think the more apt way of putting this is that correlation doesn't imply causation, but does help us determine possible causation.

[OleksandrRabinovych]

There are more than 20 variations, actually. 3^3=27. So, 7 are missing: (C->H->I->C), (C->I->H->C), (C H->I), (C I->H), (C->I H), (I->C H), (C->I<-H). The first two are "cyclic" (ok, chains - see ftn. video), but it's still a possible variation which suggests that some time ago there was a 4th factor which triggered the chain

[chillsahoy]

Known knowns: your actual data.
Known unknowns: missing data, but you're aware that you don't have the information.
Unknown unknowns: the bane of every scientist and statistician. These are factors that affect your conclusions and results, but you don't know how, by how much, or even what the factors are; you don't even realize that they are there, so you can't properly account for them.