Saturday, January 9, 2010

Correlation does not imply causality

I think this mantra is one that is not chanted enough!

I feel I read too many papers concerning material properties that say something like "we measured both properties A and B on material 1 and material 2. Properties A and B were correlated (e.g. they both increased). Therefore A causes B."

We should do better. We should teach our students to do better.

What is more likely is that there is some other property C which causes both A and B to change. For a concrete example of this, consider the oxygen isotope effect in cuprate superconductors. The fact that Tc and other superconducting properties such as the penetration depth varies with the oxygen isotope does not imply that superconductivity is caused by electron-phonon coupling.

Perhaps, a point of even greater concern should be that this mantra is overlooked in many public discussions of social issues.
Should a major goal of high school science education be to help students learn why they should believe this mantra?

A really good book which I recommend is not a "science" book but Freakonomics. I thank my colleague, Ben Powell for giving my 15-year old son a copy. After he read it, I did. In the book, the authors several times emphasise this point in their analysis of "economic" data and show how the data can be analysed in a way to tease out true causality.


  1. A great place to read more about this and related themes is Ben Goldacre's blog and his book of the same name. Today's entry was particularly good on the subject of lies, damn lies and statistics -

  2. Absolutely agreed, Ross - I think this is perhaps the most important lesson that could be taught in high school science, followed closely by a discussion of selection bias.

    Oh, and of course xkcd has a relevant comic:

    Makes a great t-shirt, too!