Conflating correlation and causation is one of the most common mistakes in interpreting data. When we assume a relationship between two variables, thinking that one causes the other, it can lead to false conclusions and poor decisions. Social psychologist Jonathan Haidt gave two compelling examples in one of my favourite lectures of his on Two Incompatible Sacred Values in American Universities.
Not this one, the one further below.
Haidt’s first graph shows a correlating rise in autism and organic food sales. He asks sarcastically:
As autism has been going up in the 90s…so has organic food sales…! 🤨 What do you think? Do you think that autism is caused by organic food? Or do you think that it’s autistic people who buy organic food and that’s why they go up? Which is it?
The answer is, of course: “Probably neither.” Because the correlation between the two does not imply causation.
His second example points to the possible existence of a third variable. Consider this headline about a study saying that “people who have more sex make the most money”. Here’s Haidt again:
So what do you think? Do you think that if you’re currently in a relationship and you have more sex, your income will go up? Do you think that’s the way it works? Of course not. There’s a third variable.
In this case, the third variables are the Big 5 personality traits, extraversion and openness to experience. Meaning, “people who have that trait have more sex and make more money.” This is because those traits also happen to be associated with earning more.
Correlation does not imply causation. But it’s not always as obvious as in the cartoon from the beginning. The good news is that confusing the two is not lethal. Even though we’re often tempted to see causal relationships where there are none. In reality, causation is pretty difficult to prove. So if we find a correlation between two variables, as Haidt suggests, we should see it as an invitation to pause and look more closely to see if there is, in fact, a causation. And if there’s maybe a third variable the two have in common.
