5 ms·

On the other hand, people repeating "correlation != causation" ad nauseam is, in my opinion, one of the great success stories of science communication. Proving

On the other hand, people repeating "correlation != causation" ad nauseam is, in my opinion, one of the great success stories of science communication. Proving causation is the singular most difficult general problem across many scientific disciplines, and at the same time our brains are hardwired to conflate correlation with causation. I think that it's simply awesome that so many people are aware of this issue nowadays.

nathias 5y ago
People mindlessly parroting anything is not a great success of anything, especially not of science communication. There is also no hardwired conflation of correlation with causation, it's just a general jumping to conclusions error.
- cornel_io 5y ago
  Social science, health, and psychological research is 90% bad because of the correlation = causation fallacy, especially articles that claim to have "controlled for" X, Y, and Z. Even after controlling most of these studies still incorrectly suggest causation. So by default, people are correct to mindlessly parrot this, even if they don't understand when they're wrong or misapplying the maxim.
  - nathias 5y ago
    ah yes, if they could only remember that correlation !== causation, 90% of social science, health and psychological research would become good instead of bad also the irony is amazing
- funklute 5y ago
  > There is also no hardwired conflation of correlation with causation, it's just a general jumping to conclusions error. There is plenty of literature in the neuroscience and psychology fields that show we are biased towards finding patterns where there are none. Evolutionarily speaking, this is easy to understand. Better to run away one extra time, than getting eaten by a lion. The latter is less likely to pass on their genes. > People mindlessly parroting anything is not a great success of anything, especially not of science communication. I respectfully disagree. Appreciating the subtleties of causality is beyond what you can expect from most lay people. Even many of my machine learning colleagues can get tripped up by this stuff. So even if it is simplistic, I still think it is useful that people at least understand that causality is difficult.
  - rosetremiere 5y ago
    Isn't "finding patterns where there are none" more akin to finding correlation in the first place, than to making the correlation->causation jump?
    - funklute 5y ago
      Well, both really. But in practice, erroneously identifying correlations doesn't tend to be quite so big a problem as conflating correlation with causation. The former is easy to address with strict data recording, whereas the latter is more of a conceptual thing, and requires some deeper thinking that often goes against intuition.
  - nathias 5y ago
    Excess pattern recognition isn't conflating correlation for causation.
    - funklute 5y ago
      No, but conflating correlation for causation _is_ excess pattern recognition.
      - nathias 5y ago
        not at all, pattern recognition has to do with raw data, correlation/causation about highly conceptual interpretation of the data
        funklute 5y ago
        > pattern recognition has to do with raw data This statement makes no sense at all. Pattern recognition applies to any data. That's not even contested terminology. (besides, just redefine your processed data to be your new raw data, and you're back to square one...and in many cases it wouldn't even be clear how to precisely define unprocessed data)
deltaonefour 5y ago
It's not that difficult. It is "difficult" in the sense that many of the experiments to prove causation are cost prohibitive. But coming up with the experimental parameters needed to verify causation are trivial. In fact, ALL clinical trials for all medicine are designed to verify causation. It's that straightforward. It just happens to cost a lot of money. Think about that. Why are clinical trials much more expensive than say the experiment correlating grip strength and life span? What is the differentiating factor? This is the key. If you don't understand why/how clinical trials verify causation and what exactly is the differentiator between experiments that only correlate things then you don't truly understand the dichotomy between correlation and causation. Most people don't understand it. Even the programming crowd. I even met data scientists who don't know about it.

nathias 5y ago

People mindlessly parroting anything is not a great success of anything, especially not of science communication. There is also no hardwired conflation of correlation with causation, it's just a general jumping to conclusions error.

cornel_io 5y ago
Social science, health, and psychological research is 90% bad because of the correlation = causation fallacy, especially articles that claim to have "controlled for" X, Y, and Z. Even after controlling most of these studies still incorrectly suggest causation. So by default, people are correct to mindlessly parrot this, even if they don't understand when they're wrong or misapplying the maxim.
- nathias 5y ago
  ah yes, if they could only remember that correlation !== causation, 90% of social science, health and psychological research would become good instead of bad also the irony is amazing
funklute 5y ago
> There is also no hardwired conflation of correlation with causation, it's just a general jumping to conclusions error. There is plenty of literature in the neuroscience and psychology fields that show we are biased towards finding patterns where there are none. Evolutionarily speaking, this is easy to understand. Better to run away one extra time, than getting eaten by a lion. The latter is less likely to pass on their genes. > People mindlessly parroting anything is not a great success of anything, especially not of science communication. I respectfully disagree. Appreciating the subtleties of causality is beyond what you can expect from most lay people. Even many of my machine learning colleagues can get tripped up by this stuff. So even if it is simplistic, I still think it is useful that people at least understand that causality is difficult.
- rosetremiere 5y ago
  Isn't "finding patterns where there are none" more akin to finding correlation in the first place, than to making the correlation->causation jump?
  - funklute 5y ago
    Well, both really. But in practice, erroneously identifying correlations doesn't tend to be quite so big a problem as conflating correlation with causation. The former is easy to address with strict data recording, whereas the latter is more of a conceptual thing, and requires some deeper thinking that often goes against intuition.
- nathias 5y ago
  Excess pattern recognition isn't conflating correlation for causation.
  - funklute 5y ago
    No, but conflating correlation for causation _is_ excess pattern recognition.
    - nathias 5y ago
      not at all, pattern recognition has to do with raw data, correlation/causation about highly conceptual interpretation of the data
      - funklute 5y ago
        > pattern recognition has to do with raw data This statement makes no sense at all. Pattern recognition applies to any data. That's not even contested terminology. (besides, just redefine your processed data to be your new raw data, and you're back to square one...and in many cases it wouldn't even be clear how to precisely define unprocessed data)

deltaonefour 5y ago

It's not that difficult. It is "difficult" in the sense that many of the experiments to prove causation are cost prohibitive. But coming up with the experimental parameters needed to verify causation are trivial. In fact, ALL clinical trials for all medicine are designed to verify causation. It's that straightforward. It just happens to cost a lot of money. Think about that. Why are clinical trials much more expensive than say the experiment correlating grip strength and life span? What is the differentiating factor? This is the key. If you don't understand why/how clinical trials verify causation and what exactly is the differentiator between experiments that only correlate things then you don't truly understand the dichotomy between correlation and causation. Most people don't understand it. Even the programming crowd. I even met data scientists who don't know about it.