The saying should be: “statistical dependence does not imply causality.” Correlation is a particular measure of a linear relationship. A lack of correlation can happily coexist with statistical dependence if variables are related in a complicated non-linear way. This “correlation” business needlessly emphasizes linear models (prevalent in Stat at the time of Pearson et al.) See also this: http://en.wikipedia.org/wiki/Correlation_and_dependence
Also, this is true: “lack of statistical dependence does not imply lack of causality” (due to effect cancellation).
Agreed. It’s just a pet peeve because the concept of “correlation” does not cut at the seams here. I was certainly not faulting you (as your post was likely a response to the slate article, which used the term “correlation” as well).
I was going to mention that “correlation does not imply causation” sounds snappier, but the more I play around with it, “association does not imply causation” seems somewhat more aesthetically appealing.
doesnotimply.com is also free, shorter, less loaded (in the way you describe above), and has an obvious logo to go with it (=/=>). If I go ahead with it, I might use that instead.
Somehow, the “correlation does not imply causation (but it furtively suggests it, etc)” idea is linked in my brain with the “absence of evidence is not evidence of absence (but it is if you’re a Bayesian)” idea.
At the risk of diluting the original good idea, maybe doesnotimply.com could incorporate the latter also.
They can also be completely independant and have causation, but that’s not something that would happen by chance. The only time I know of where something like that will happen is if the cause is designed to regulate whatever it’s independant of. For example, the temperature doesn’t correlate to the power going through the heater or air conditioner, since it’s always constant, which is because the heater and air conditioner keep it constant.
Pet peeve:
The saying should be: “statistical dependence does not imply causality.” Correlation is a particular measure of a linear relationship. A lack of correlation can happily coexist with statistical dependence if variables are related in a complicated non-linear way. This “correlation” business needlessly emphasizes linear models (prevalent in Stat at the time of Pearson et al.) See also this: http://en.wikipedia.org/wiki/Correlation_and_dependence
Also, this is true: “lack of statistical dependence does not imply lack of causality” (due to effect cancellation).
I had intended on tackling this, but the original still works as a Steel Man of the general case.
Agreed. It’s just a pet peeve because the concept of “correlation” does not cut at the seams here. I was certainly not faulting you (as your post was likely a response to the slate article, which used the term “correlation” as well).
I was going to mention that “correlation does not imply causation” sounds snappier, but the more I play around with it, “association does not imply causation” seems somewhat more aesthetically appealing.
doesnotimply.com is also free, shorter, less loaded (in the way you describe above), and has an obvious logo to go with it (=/=>). If I go ahead with it, I might use that instead.
My vote goes to doesnotimp.ly
Somehow, the “correlation does not imply causation (but it furtively suggests it, etc)” idea is linked in my brain with the “absence of evidence is not evidence of absence (but it is if you’re a Bayesian)” idea.
At the risk of diluting the original good idea, maybe doesnotimply.com could incorporate the latter also.
They can also be completely independant and have causation, but that’s not something that would happen by chance. The only time I know of where something like that will happen is if the cause is designed to regulate whatever it’s independant of. For example, the temperature doesn’t correlate to the power going through the heater or air conditioner, since it’s always constant, which is because the heater and air conditioner keep it constant.
I agree, but this regulation business seems important and occurs a lot in nature.