Causation vs Correlation: How tell when someone's trying to feed you baloney

Don't fall for people misusing evidence to
tell you what you're doing is wrong.
Credit: Arlette Cifuentes Meneses via Flickr CC BY 2.0 licence.
by Gaia Cantelli, PhD

Last time that somebody told you that what you were doing was bad, did they offer evidence - and did you believe it? 

In an age obsessed with fake facts, hopefully you have an armour of skepticism forming already. But here's a chink in the armour you may not have thought of: perfectly true facts can also be warped to manipulate people. One of the most common ways to do this is by mixing up causation and correlation.

Causation and correlation are both extremely boring statistical words that underlie very simple concepts. Two events are correlated if they happen at the same place, or at the same time. For example, in children, shoe size is most likely correlated with the number of books they have read – older children have both read more books and have bigger feet. The two things do not cause each other but do happen at the same time. Causation, on the other hand, means that one of the two events is happening because of the other. For example, students who do better in exams also have a higher acceptance rate into college - because grades are one of the things that college considers when evaluating an applicant.

What does it mean that margarine consumption and divorce
were correlated in Maine? Nothing! Credit Sandra Cohen-Rose
and Colin Rose, via Flickr CC BY 2.0 licence.
It can be very difficult to separate causation and correlation. When two events happen at the same time or in the same place, it is quite difficult to determine whether one of the two is causing the other, or whether they are independent but happening together. Truly independent events, in fact, can be correlated simply by chance through what is known as spurious correlation. For example, in the years between 2000 and 2009, the rate of divorce in the state of Maine was almost perfectly correlated with how much margarine people consumed, although of course buttery spreads have nothing to do with divorce.

Also, events can be linked by some causation, but not in a straightforward way. Two correlated events can both be caused by the same thing but be independent of each other. That's what happens as children get bigger feet and read more books - both are caused by a third factor, their age. Another great example of this type of causation is health supplements. While scientists have shown that most nutritional supplements don't actually improve your health, people who take them tend to exercise more and eat well, which results them in also being healthier. Their health improvement is actually caused by their attention to their health, even though it's correlated with their use of supplements.

Some forms of oral contraception are correlated with a higher risk of cervical cancer - but they don't cause the cancer directly. This shows another way in which two correlated events can have a causal relationship, via a chain of cause-effect reactions. In this case, because women who take the pill often do not use any other form of contraception, they end up being more likely to contract a sexually-transmitted virus called HPV, which leads to increased cervical cancer risk. The best way for a woman to limit her cervical cancer risk is therefore to use barrier methods of contraception or to get an HPV vaccine and not, in fact, to stop taking the pill!

So how on earth can we differentiate causation and correlation when somebody is trying to sell us an idea? The best, most comprehensive thing to do is to find some authoritative sources on the subject and read around it. For example, the anti-vaccine movement maintains that vaccination leads to autism. It is very easy to find innumerable resources pointing at the fact that the scientific community is completely unanimous in agreeing that vaccines are perfectly safe. It's just that the age at which most children are diagnosed with autism is around the same age when they receive most of their shots. The two events are simply happening at the same time, even though they are just as related as margarine and divorces in New England. If you don't have time to do the digging, or if your digging is inconclusive, a good technique is to play devil's advocate. Is there any other way that these events could be related other than by a direct cause-and-effect reaction?

Is this post that popular? Hmm, that might be baloney.
Credit: frankieleon via Flickr CC BY 2.0 licence.
For instance, many people will tell you that couples who choose to live together before they are married have a 30% higher chance of divorce than those who wait. This is often brought forward as incontrovertible evidence that cohabitation causes divorce. However, it is very easy to think of other reasons for the link. People who live together before they marry, for example, are less likely to be religious than those who do not. Religious people tend to have much harsher views on divorce than their secular peers, and are likely to stay married when others would perhaps separate.

Another possibility is that people who live together may end up marrying because of pressure from their families, which means they commit to a partner they had not initially selected as a potential spouse. In this case, cohabitation causes forced marriage, which causes divorce. In other words, there are many reasonable ways in which divorce and pre-marital cohabitation can be related other than direct cause-effect. So next time you someone who quotes a shaky correlation at you, remember to ask about causation. Unless they can tell you how the correlation is caused, you'd be better off not letting them worry you.

Gaia Cantelli is a postdoctoral associate at Duke University, studying the mechanisms that regulate cancer cell metastasis to the bone and she regularly blogs over at scienceblog.com

No comments:

Post a Comment