Correlation, Causation and Internet Comments

Daniel Engber wrote a wonderful piece at Slate. Engber examines when you can hide behind the phrase correlation does not imply causation and when that may not be the best idea. "The correlation phrase has become so common and so irritating that a minor backlash has now ensued against the rhetoric if not the concept. No, correlation does not imply causation, but it sure as hell provides a hint." (thanks Rachel Scollon)


  1. He never claims that correlation proves causation, just that it might, which is the same as saying that it doesn’t.

    1. Which is to say it is a whiny article using lots of words to say nothing much beyond that the author is indignant.

    1.  Took me a while to note that there were sort of Easter egg comments on the xkcd toons. The one for this is sooo appropriate for this thread:  “Correlation doesn’t imply causation, but it does waggle its eyebrows suggestively and furtively while mouthing ‘look over there’.”

  2. I thought the phrase was: “Correlation is not causation”? And that you could elaborate by saying: “Sometimes, correlation implies causation”. The difference between “is” and “implies” is pretty substantial, no?

  3. As a marketing statistician, I concur. There’s an inclination to misuse the calling out of logical fallacies as a way of discrediting your opponent on a topic (perhaps better than ignoring logic altogether). If you adhere too strictly to logic you’ll get nowhere; eventually you have to use assumptions and empirical evidence to draw conclusions. 

    1.  As one of my favorite philosophy professors said when someone mentioned their logic 101 learning in the inductive logic class we were in, “Ahh yes, baby logic. We’re not dealing with baby logic here”

    2. I generally try to avoid using the names of fallacies in an argument because it’s a way of avoiding making an analysis. People rather regularly call out a straw man argument when it isn’t. If they took the time to describe the argument rather than using a label, they might notice that they were completely off. Or not. But there’s always hope.

  4. What I find vastly more fascinating is that causation doesn’t imply correlation.  It’s quite possible that A deterministically causes B, but, despite this fact, if you investigate A and B empirically, you will never find a correlation coefficient statistically different from zero.

    I’ll leave it as an exercise for the reader to come up with an example.

    1. Yes, initially I was thinking the same way. E.g. if A causes B, and we draw a plot of their variables that shows a sine-wave relationship, then the usual Pearson correlation coefficient will be 0.  However, I think “correlation” has a broader sense than this measure, which is inherently aimed at finding a linear relationship.

  5. “No, correlation does not imply causation, but it sure as hell provides a hint.”

    This refutation is irrelevant. Hints are no basis for sound debate or reporting. Only facts suffice.

    1. It works the other way too. Media are often too slow to accept conclusive evidence in favor of more sensationalistic ‘debates’ between a handful of holdouts and the entire scientific establishment. 

  6. One good point at the end is that the person doing the implying is usually also implying (or outright demanding that) Something Must Be Done about it. Then some hasty ill-thought out policy or legislation based on the assumed correlation is passed and just makes things worse (Hello California).

    But it can certainly be a starting point.

    1. Like how the state is always rushing into considering some tepid, half-hearted effort to control greenhouse emissions even though the evidence for global warming has only been considered conclusive for a little over a generation? Those crazy granola-munching hippies.

      1. Well that’s the other side of it – assuming the legislation accomplishes anything . I’m thinking of things like laws that mandate posting calorie counts on the menus even when you already know the only people who read them are the people who were already eating healthy. Lack of calorie counts clearly does not cause obesity, but we’d like to think it’s just a lack of education, so why not? Actually, it’s even worse. There’s not even a correlation, we just want there to be one, so we’re implying both the correlation and causation.

        We just love feel-good laws that make the Legislators feel like they’re accomplishing something (so they can avoid working on or thinking about the budget). And sometimes you might be. Or you just might be making it worse. The more economic impact there is the more pushback there is (the greenhouse gasses), so it’s usually on relatively trivial little stuff like this that it just sails through.

  7. correlation does not imply causation


    I usually say something like, “correlation does not necessarily imply causation” which has a much different meaning to me.

    1. The confusion is that “imply” in formal logic means something very different than “imply” in ordinary English. In the strict logical sense, correlation does not imply causation. That’s because A can be correlated with B without A causing B, or vice versa.  In ordinary English causation may imply causation, insofar as correlations may hint at causal relationships, even if they fall short of proving them.

Comments are closed.