A wealth of evidence demonstrates how people suffer from confirmation bias, but most of it is on ordinary people. Surely intelligence is a cure? Smarter cookies might better appreciate the logic in a counterargument, and notice defects in data even if supports their viewpoint. If so, then confirmation bias might not be a big problem — it affects the person on the street, but not the leaders of countries and companies. You yourself might think that your intelligence will protect you from confirmation bias, so you’d never fall for misinformation.
Unfortunately, intelligence can make the problem worse. A study gave students one of the following tables on how a skin cream affects a rash, and asked them whether it makes it better or worse:
|
Rash got better |
Rash got worse |
|
|
Users of skin cream |
223 |
75 |
|
Non-users of skin cream |
107 |
21 |
|
Rash got worse |
Rash got better |
|
|
Users of skin cream |
223 |
75 |
|
Non-users of skin cream |
107 |
21 |
If you saw the first table, you might think the cream improved the rash — 223 got better and only 75 got worse. But that’s incorrect, because they might have improved anyway. You instead need to compare the better-worse ratio for users (223/75 = 3.0) to non-users (107/21 = 5.1). Since the latter is higher, the cream actually aggravated the ailment.
Since the answer isn’t obvious, it’s not surprising that many people got it wrong. It’s equally unsurprising that students with higher numeracy — intelligence in understanding data — did better. What was unexpected was how a separate set of students performed when given the same data, but on how gun control affects crime (rather than how skin cream affects a rash).
|
Decreases in crime |
Increases in crime |
|
|
Cities that banned concealed handguns in public |
223 |
75 |
|
Cities that did not ban concealed handguns in public |
107 |
21 |
|
Increases in crime |
Decreases in crime |
|
|
Cities that banned concealed handguns in public |
223 |
75 |
|
Cities that did not ban concealed handguns in public |
107 |
21 |
Even though the data is identical, students made far more mistakes when it related to gun control — particularly when the data contradicted their own views on the issue. Democrats were more likely to claim that gun control cut crime when given the first table, despite the data showing the opposite; Republicans argued that it increased crime when given the second, when that’s also wrong. Strikingly, the more numerate students were, the more polarised their views became — the greater the difference in how Democrats and Republicans interpreted the same table. Numeracy made their biases worse.
Here’s why. A numerate Republican given the first table would start by comparing the 223 and 75, which suggests that gun control reduced crime. He didn’t want this result, so used his maths skills to realise that you also need to consider the 107 and 21, leading to the opposite conclusion. But if a numerate Democrat saw the exact same numbers, the simple comparison of 223 and 75 implies that gun control worked. Since she liked this conclusion, she stopped there and didn’t bother to deploy her arithmetic ability. Intelligence helps you get the right answer — but only when it suits you.
