Picture this: you’re lying on a sandy beach, working on your tan, listening to seagulls having a heated argument and the only thing between you and full happiness is ice cream. Cold, sweet, delicious vanilla and strawberry scoops.
Did you know that this innocent dessert can kill you? Eating ice cream is associated with deaths from drowning. Just look at the graph.
Ice cream sales and drownings are positively correlated, which means that when ice cream sales increase, the number of drownings increases as well. How is that possible? Let’s think of some possible explanations.
If you eat ice cream, you become heavier, so the gravitational force is more likely to take you to the bottom of the lake or sea. Or consuming cold desserts results in a sudden drop in your body temperature, which leads to slower reactions in the water. Or… ice cream doesn’t CAUSE people to drown. Neither the increase in drownings causes more ice cream sold.
What really happens is that these two variables are correlated with a third, lurking variable: summer, in particular high temperatures. When it’s hot, more people buy ice cream. And more people go swimming, which obviously leads to more water-related accidents.
CORRELATION DOESN’T IMPLY CAUSATION – it’s probably the most important phrase you’ve learned in your statistics class. The problem is, so many of us still haven’t done their homework and mistaking correlation with causation is the deadly sin of statistical analysis appearing in the media.
I like writing about climate change, so here’s one more thing to be warned against: raising CO2 levels cause obesity. We could think of possible explanations of this problem again… or we can just realise that since the 1950s, when the population started becoming richer, we’ve been producing more CO2 and eating more food.
To make things even trickier, there are many ways in which we can confuse correlation and causation.
- The causation might be reverse. One can say that using wheelchairs causes accidents, because there’s definitely a relationship between being on a wheelchair and having an accident. Obviously here the causality is reverse, it’s having an accident that is likely to lead to a need of a wheelchair.
- There might be a third, lurking variable underlying both phenomena. Did you know that bigger palms lead to a shorter life? Hint: who, on average, has bigger palms, women or men? And which gender has shorter life expentancy?
- The correlation might be just coincidental. To find examples for this case, I invite you to visit this excellent website and have fun playing around. Let me know what you found!
Does it mean that we cannot trust any studies that talk about causality? Of course it doesn’t! We know good methods of checking if in a particular case correlation indeed implies causation.
In many medical experiments researchers use a control group. For example, if they were about to check a new weight-loss method that requires to eat only red things, they could put one group of patients on such a diet, another group on only-blue-things diet (to make sure that it’s the red colour, not monochromaticity that has beneficial effects) and check if the patients in these groups really lose weight faster than the control group, so people who didn’t change anything in their eating habits.
Another example would be studies on identical twins used mostly in researching nature-vs-nurture questions. In this case we have a full control of the nature (because identical twins share DNA), but we can change some environmental factors.
Whenever you’re reading a breaking news in science, especially nutrition studies, think of yourself eating ice cream in a life-jacket. Does X really cause Y, or are they just correlated?