Sunday, 19 October 2008

Correlation vs. Causation

Interesting post by Brian at

The study mentioned is (1) - Archives of Internal Medicine: Reduced Disability and Mortality Among Aging Runners.

Also see Crabsallover blog:

,,, regarding the numerous research studies that people (myself included) throw around to support why we should or should not eat Food A or do Exercise B. And that note is that correlation is not…repeat…is not the same as causation.

To start this discussion, we need to recognize that we’re all biased. I’m biased. You’re biased. Yes, some people have a more open mind than others, but for someone to say that they are completely unbiased is ludicrous. And it’s because of this bias that we have a tendency to blindly accept studies that reinforce what we already believe, and immediately dismiss or pick apart the ones that don’t. So, when we hear of a study(1) that, for instance, finds a correlation between exercise (in this case running) and longevity, we immediately accept that as Gospel. Why? Because that’s what we already believe to be true in our minds. Of course people who exercise live longer, better lives, right?!?

The conclusion that the authors of this particular study draw is that “Vigorous exercise (running) at middle and older ages is associated with reduced disability in later life and a notable survival advantage.” There’s nothing wrong with that statement - in fact, it’s 100% accurate - but it’s important to note that the authors said they found an “association” (correlation), not proof (causation). Reread their conclusion. Nowhere are they claiming they proved a link between exercise and longevity, but today’s mainstream media doesn’t make mention of that distinction when studies like this are blasted across the television, radio, and the internet. Saying something could, maybe, possibly, perhaps, we think help or hurt you doesn’t have the same impact as, “Want to avoid death? We’ll tell you how…tonight at 11:00″.

Now, let’s look at another example of correlation. Below is a famous, far-fetched example used in a letter to the Kansas School Board in January 2005 (2).

There is obviously a correlation between the two sets of data in that, as the number of pirates has decreased over the centuries, the global average temperature has increased. But I don’t think that anyone would believe that the decreasing number of pirates in the world has actually caused the increase in the global average temperature or that the increase in the temperature has caused a reduction in the number of pirates. Yet because a scientist or statistician somewhere in the world found a similar correlation between exercise and longevity we immediately think, “Ah-ha! Finally! Proof!”

The interesting thing about science is that we tend to define it as creating a hypothesis and then proving or disproving that hypothesis. In reality, hypotheses aren’t really proven. Experiments either support the data, or disprove the data. After enough support, we accept the hypothesis as fact, even though we have never proven it to be true. We just can’t find anything that disproves it (if you’re interested enough - and geeky enough - check out the graph of the results of the COBE Mission as this is probably the most amazing example of an experiment agreeing with a theory (3)).

So, as we look at the two sets of correlating data presented in this post, the first (that exercise is linked to longevity) is one that we’ll still walk away believing because we “know” that to be true, while the second (involving temperatures and pirates) we will continue to “know” is false even though neither study…proved…anything.


(1) - Archives of Internal Medicine: Reduced Disability and Mortality Among Aging Runners

(2) - Wikipedia: Flying Spaghetti Monster

(3) - Wikipedia: COBE Mission

No comments: