Monday, November 1, 2010

Causation and correlation

Normally I don't link to conversations on Facebook, but Robert Wiblin's Facebook wall is open to the public to see his insightful musings. (If you use Facebook, I recommend friending him.) For instance, he says:

Correlation is not causation, but correlates with it.
My response:
Causation is not correlation, but causes it.
Then, someone asks whether all causation is "statistical." I give a whole book as an answer: Probabilistic Causality by Ellery Eells (1953-2006). As you might have guessed from the title, the author's answer is yes. Eells pointed out (to those of us who took his course on probabilistic causality at the University of Wisconsin - Madison) that, in addition to the ubiquitous refrain that correlation does not prove causation, there's also the less well-known fact that causation does not prove correlation. (For instance, if A causes both B and C, and C prevents B more strongly than A causes B, A won't be correlated with B.) That observation, along with Eells's whole theory, was the inspiration for my response to Wiblin.

Someone else links to this perfect XKCD comic: