What Is Confirmation Bias? Why We See What We Expect to See
By the BrainSnail editorial team. How these articles are written and checked, and how to tell us when one is wrong.
Give people a rule to discover and they will test it by looking for examples that fit rather than ones that would break it, and they will do so even when told that breaking it is the only way to be sure. The tendency to gather, interpret and remember evidence in a way that supports what one already thinks was named confirmation bias by the psychologist Peter Wason in the 1960s, and it has since turned out to be the most pervasive distortion in human reasoning, present in scientists, judges, doctors and everyone who has ever read the news.
Wason's experiments
Wason gave subjects the sequence 2, 4, 6 and told them it obeyed a rule they were to discover by proposing further sequences, which he would mark as fitting the rule or not. Most guessed the rule was numbers rising by two and tested it by offering 8, 10, 12 and 20, 22, 24, each of which was confirmed. Very few tried 1, 2, 3 or 10, 9, 8, which would have been confirmed and refuted respectively, and so very few found the actual rule, which was any three rising numbers. They had tested their hypothesis only in ways that could not show it to be wrong.
A second experiment showed four cards, each with a letter on one side and a number on the other, showing A, K, 4 and 7, and asked which must be turned over to test the rule that a card with a vowel on one side has an even number on the other. Most people turn A and 4. The correct answer is A and 7, since only a vowel with an odd number behind it could break the rule, and the 4 can prove nothing whatever is behind it. Fewer than one person in ten gets it right.
Three forms
The bias operates at each stage of dealing with evidence:
- •Biased search: seeking out sources and asking questions likely to confirm, as when someone who suspects a friend is angry asks questions that assume it
- •Biased interpretation: reading ambiguous evidence as support, so that supporters of opposite views can watch the same debate and each conclude their side won
- •Biased memory: recalling the instances that fit and forgetting the ones that did not, which is how a belief in lucky charms or in a horoscope survives a lifetime of counter-examples
Why it exists
The bias is not stupidity, and it is not cured by intelligence; more able reasoners are better at constructing arguments for what they already believe. One explanation is that testing a belief by looking for confirming cases is a reasonable strategy when most hypotheses are wrong and the world is complicated, since a wrong belief will usually fail to be confirmed quickly enough. Another is that reasoning did not evolve to find truth in solitude but to persuade others and to defend oneself in argument, tasks at which a one-sided search for evidence is a strength. On that view confirmation bias is a feature of an individual mind that becomes a bug only when the mind reasons alone.
The bias grows with emotion. On questions where nothing is at stake people update readily; on questions tied to identity, politics, religion or a decision already made, the same people examine evidence against their view with a rigour they never apply to evidence for it, a pattern called motivated reasoning. Experiments in which people are shown a study supporting their position and a study opposing it, of equal quality, typically end with both sides more convinced than before, each having found the flaws in the other's study only.
Where it does damage
In medicine, a doctor who forms an early diagnosis tends to order tests that would confirm it and to discount results that do not, which is why diagnostic checklists force alternatives to be considered. In criminal investigation, a suspect identified early attracts the evidence and other lines go cold, and several notorious wrongful convictions followed exactly that path. In science, researchers find what they expected in noisy data, one reason replication fails so often and why the best trials are designed so that nobody analysing the results knows which group is which. In everyday life it is the reason a news feed that shows people what they already agree with is so comfortable and so narrowing.
What limits it
Knowing about the bias helps less than one would hope, since people readily see it in others and rarely in themselves. What does work is structure. Actively asking what evidence would change one's mind, and looking for it, is the direct antidote, and Wason's subjects who were prompted to seek disconfirmation found the rule. Writing down a prediction before seeing the result removes the room to reinterpret. Working in groups that contain disagreement uses the bias against itself, since each side searches hard for the flaws in the other's case, which is the principle behind adversarial trials, peer review and a good editorial meeting. The scientific method itself, with its insistence on the test that could fail, is the most successful institutional answer to a flaw nobody escapes individually.
Counting all four boxes
Many people believe that hospital emergency departments are busier on nights with a full moon. The belief feels well supported, because every busy full-moon night gets noticed and retold: of course it was chaos, look at the moon. But that is only one of four kinds of night. There are also quiet full-moon nights, busy nights without a full moon and quiet nights without one. The first kind is memorable and confirms the belief; the other three are either unremarkable or never linked to the moon at all, so they are quietly forgotten.
To test the idea you need all four counts, laid out as a simple table, and then a comparison of the proportion of busy nights with and without a full moon. When researchers have counted admissions and calls this way over long periods, the link has generally failed to appear. The same four-box check works on any belief of the form X goes with Y, from a lucky shirt and winning matches to a particular food and feeling unwell. The belief usually rests on one box; the answer lies in comparing it with the other three.
The same puzzle, made social
Wason's card puzzle has a twist that has kept psychologists arguing for decades. Change the content and keep the logic, and performance transforms. Imagine four cards, each showing a person's drink on one side and their age on the other: beer, cola, 25 and 16. The rule is that anyone drinking beer must be at least 18. Which cards must you turn? Most people quickly see that the answer is beer and 16, the only cases that could hide a rule-breaker. That is exactly the same structure as A and 7, the answer so few people find in the letters and numbers version.
Why the drinking version is so much easier is still debated. One view is that the rule is simply familiar, so people already know what a violation looks like. Another, put forward by the evolutionary psychologists Leda Cosmides and John Tooby, is that human reasoning is especially sharp at catching people who cheat on social agreements, taking a benefit without meeting the condition. A third notes that the drinking rule says what people are allowed to do rather than describing the world, and permissions seem to prompt a search for violations in a way that descriptions do not.
A fingerprint that fitted too well
In 2004, after the train bombings in Madrid, the FBI matched a partial fingerprint on a bag of detonators to Brandon Mayfield, a lawyer in Oregon. Several examiners confirmed the identification, and Mayfield was arrested and held for about two weeks before Spanish police matched the print to an Algerian man and the FBI withdrew its conclusion. A later official review found that once the first examiner had declared a match, the others tended to see the features they expected and to explain away differences between the two prints rather than treat them as grounds for doubt. A technique assumed to be objective had been shaped by expectation.
The psychologist Itiel Dror then tested the problem directly. He gave experienced fingerprint examiners pairs of prints that they themselves had judged to be matches years earlier, but told them the prints came from the Mayfield case, implying they were a famous mistake. Most of the experts now reached a different conclusion from the one they had reached before about the same prints. The remedy adopted in many forensic laboratories borrows the logic of blind trials: examiners analyse and document the crime scene mark first, before seeing the suspect's print, and are shielded from irrelevant case details, so that expectation has less room to shape what they see.
The takeaway
Confirmation bias is the tendency to search for, interpret and remember evidence in ways that support existing beliefs, demonstrated by Wason in the 1960s and found since in every field of judgement. It is strongest on questions bound up with identity, it is not reduced by intelligence, and it is best countered not by willpower but by structures that force the search for disconfirming evidence: predictions made in advance, blinded analysis, and disagreement built into the process.