"Correlation doesn't equal causation" I've noticed that people tend to misinterpret this sentence. In a lot of discussions it shows that people think that there can be correlation while the things don't have _ANY_ relationship between eachother, as in the correlation would arise purely trough random chance.
That is blatantly false. Correlation implies at very minimum a causal link trough a third effect. The traditional example of drownings and ice cream consumption has that third link. Nice summer days. So those two are linked.
I think one should say "Correlation doesn't imply direct causation". That's closer to the truth.
Correlation is a statistical term that means that two random variables exhibit "similar behavior" across the probability space. Causation is a philosophical concept taken as a scientific axiom. Correlation does not imply causation at all -- neither direct or indirect -- though it may increase the chances of logical consequence that is sometimes equated (albeit occasionally through confusion) with causality.
Now, it is true that perfect correlation (which is rarely shown; more often than not in the business world, only "half" of the correlation is shown -- e.g. all successful startups have ping pong tables -- which is as good as nothing at all) in the real world usually implies some hidden causal link, but that link is not what the investigator has in mind. Often, the causation is actually quite direct but flows in the opposite direction from the common interpretation.
> That is blatantly false. Correlation implies at very minimum a causal link trough a third effect.
No, it doesn't.
Correlation can be coincidence; while a particular correlation may be unlikely on its own to be coincidence, there are lots of combinations of things in the universe that might be compared side-by-side, and so unlikely coincidental correlations are everywhere, and so if you troll through the universe looking for correlations, you are going to find a bunch that don't mean anything causally.
Sure. But with enough datapoints the probability of the correlation being a mere accident is going to be vanishingly small.
And by enough I am talking about hundreds and more of datapoints. If one looks at things like yearly values of something for within last 10 years like the page spurious correlations is doing one is bound to find some weird ones.
I thought the problem was with the statement was the "we've got a strong correlation here" bit, as if that were a meaningful thing. Correlation is weak evidence for a proposition (and sometimes weak evidence is all you've got) but the correlation being strong doesn't mean anything at all past the threshold of pure chance.
Someone who claims it confidently like that is well into the area of making shit up then rationalizing it.
That is blatantly false. Correlation implies at very minimum a causal link trough a third effect. The traditional example of drownings and ice cream consumption has that third link. Nice summer days. So those two are linked.
I think one should say "Correlation doesn't imply direct causation". That's closer to the truth.