6 ms·
Interestingly, the example he uses about correlation isn't true: > Probably the shortest true statement that can be made about causality and correlation is "Em
by rvilj 5y ago
Interestingly, the example he uses about correlation isn't true:
> Probably the shortest true statement that can be made about causality and correlation is "Empirically observed covariation is a necessary but not sufficient condition for causality."
Correlation isn't necessary for causation, so ironically the supposed mutilation used to fit on a PP slide - "Correlation is not causation" - is actually far more correct.
The same mistake is also made in Kahneman, Sibony and Sunstein's new book [1].
[1] https://twitter.com/economeager/status/1395791301627596806 https://twitter.com/economeager/status/1395791301627596806
- djoldman 5y agoThis depends on the exact definition of causation. Arguably, causation is merely a human word for "A always happens when B happens." A causes B or vice versa is one of those things where it's true until it isn't. Causation is frequently used to mean that B will always occur when A occurs. The problem is that no probability is truly 100%.
- rvilj 5y agoEven with that definition, you won't necessarily observe correlation between A and B. See the example used in the twitter thread I linked: > Imagine driving a car, reaching a hill and pumping the gas as you begin to go up so that your speed is constant. The correlation between pressing on the gas and the speed of the car is zero but they're obviously causally related, it's that the agent is optimizing speed!
- djoldman 5y agoWell in that scenario, the only way your speed would be constant (assuming a constant angle on the hill) is if your engine thrust was constant. So "pumping" the gas just achieves a constant thrust. So your speed stays the same and the thrust stays the same. So there is perfect correlation and causation (by some definition of causation). But more to the point, there isn't a good definition of causation. If you use the "but for A, B would not occur" there are problems. Other definitions also have problems. This is a deep epistemological problem and I think not as simple as it may seem.
- goatlover 5y agoLaws of in fundamental physics might be causal. Humean causation has the problem that everything is contingent. It just so happens that energy always equals mass times the speed of light squared. But there’s no reason for it. This seems to run counter to the goal of physics and cosmology, where some fundamental explanation is sought.