What People Are Getting Wrong This Week (and Every Week): Correlation and Causation
Did you know you can set up Google to filter out junk? Follow these steps to improve your search results, including adding my work on Lifehacker as a preferred source .
Usually I focus on something that only a fraction of the population gets wrong so that the rest of us can feel smarter, but this week I’m going to go further and explain something that you, I, and everyone we’ve ever met has gotten wrong in the past, is wrong now, and will get wrong in the future: mistaking correlation for causation.
People have been repeating some variation of the “correlation is not causation” thesis since at least 1739, when David Hume formulated the concept in A Treatise of Human Nature . To paraphrase Hume: Just because two things happen at the same time doesn’t mean one causes the other. Every smart person already knows this, and it’s repeated all the time, but we still get it wrong.
Here are some examples:
-
The results of the last 20 years of “gut biome” research may be flawed, in part because scientists and the media have misinterpreted the relationships and causation. (I always had a hunch — get it? — that this research was suspect.)
-
For years, people have believed that moderate alcohol consumption is good for your health. But it’s not. In some populations, it’s correlated with better health , but it doesn’t cause it.
Proponents of the correlation/causation fallacy often collect data sets showing a nearly perfect correlation between, say, higher rates of autism and higher rates of vaccination, making it perfectly understandable to assume that one causes the other. But there is no evidence that vaccines cause autism , and the correlation is probably because children were given vaccines around the age when autism is typically diagnosed, and we have gotten better at both vaccinating children and diagnosing autism. But the “probably” in that sentence is a bit clever. While a causal link between vaccines and autism has never been proven, there could be any number of reasons why the numbers match.
Classic examples of correlation-causation refutations, such as the link between ice cream sales and shark attacks , typically offer a trivial explanation for causation—ice cream just seems to cause shark attacks, since both ice cream sales and ocean swimming increase as the weather warms—but even this potentially mistakes correlation for causation. It makes sense, but we don’t actually know why the two numbers match. And sometimes there’s simply no basis for a connection between the two disparate data points.
Take a look at the chart below. It proves that the ratings of the TV series Two and a Half Men directly correlate with the amount of jet fuel used in Serbia.
Or check out the exact relationship between people Googling “I just got scratched by a cat” and fruit consumption in the US.
The second example I did was on Tyler Vigen’s False Correlations site, using a tool that lets you make random connections all day long. What’s more, the site uses AI to create fake “research papers ” to explain the connection.
In the case of cats, ChatGPT offers the following possible explanation:
“Health-conscious families (dieting, buying fruit, etc.) are more likely to treat even minor injuries with more caution. A person who eats more fruit is not more likely to get scratched, but is more likely to Google “cat scratched me” to learn about the risks of infection or treatment options.”
Even though I know it’s bullshit, it still tracks . That’s why we can never stop being wrong in this particular sense. Our brains want to believe. A clearly articulated explanation, a neat diagram, a plausible-sounding theory—it feels so good . A simple “we don’t know” doesn’t match that certainty.
Mistaking correlation for causation, we adopt fad diets and believe we are healthy when we drink wine with dinner. It shapes health guidelines, public policy, and personal decisions that may actually harm people. The best we can do is try to recognize this: when we read a headline that says “X causes Y,” we should assume it is a “cat scratches cause fruit consumption” situation until evidence to the contrary emerges.