What is confirmation bias? Seeing what you expect to see

Confirmation bias is the tendency to search for, interpret and remember information in ways that support what one already believes, and to give less weight to information that cuts against it. It works quietly: people rarely ignore contrary evidence on purpose. They just ask questions whose answers can only agree, read ambiguous facts as support, and recall the hits better than the misses. It is one of the most studied of all cognitive biases.

In short

• Confirmation bias shows up in three places: which evidence people look for, how they read it, and what they remember.
• In Peter Wason's 2-4-6 task, people test only cases their guess predicts will fit, and so never find out that the guess is wrong.
• It affects experts as well as everyone else, and the best defenses are procedures: look for the case that would prove you wrong, and fix your test before you see the results.

What does confirmation bias look like?

It works at three stages, and the cognitive bias can enter at any of them.

Search. People look for evidence where they expect to find agreement. Someone convinced a supplement works searches "benefits of" the supplement, not "does it work", and the search engine obliges. Interpretation. The same evidence gets read differently depending on the reader's starting point. In a well-known 1979 study at Stanford, people who supported and people who opposed the death penalty read the same two studies, one favoring each side; each group rated the study that agreed with them as more convincing, and both said they came away more sure of their original view, as the Wikipedia account of the research describes. Memory. Hits are easier to recall than misses. A person who believes a full moon makes hospital wards busy remembers the hectic full-moon nights and forgets the quiet ones.

The psychologist Raymond Nickerson, reviewing decades of research in 1998, called it "a ubiquitous phenomenon in many guises". It is closely related to motivated reasoning, in which a person's wishes shape the conclusion, but it doesn't need a motive. People show it even about rules they have no stake in, as Wason's experiment showed.

What is Wason's 2-4-6 task? A worked example

In 1960, the English psychologist Peter Wason gave people a puzzle. He told them that the three numbers 2, 4, 6 fit a rule he had in mind. Their job was to find the rule by proposing other sets of three numbers; he would say "yes" or "no" to each. When they were confident, they announced the rule.

A typical session goes like this:

Guess in mind	Triple tested	Answer	What the test could teach
"Even numbers going up by 2"	8, 10, 12	Yes	Nothing new: the guess predicted yes
Same	20, 22, 24	Yes	Nothing new
Same	100, 102, 104	Yes	Nothing new; confidence rises anyway
Announces "even numbers going up by 2"		Wrong
"Going up by 2"	1, 3, 5	Yes	Rules out "even"
Same	1, 2, 3	Yes	Surprise: the step needn't be 2
Same	3, 10, 50	Yes	The gaps needn't be equal
Same	6, 4, 2	No	Order matters

The rule was simply "any three numbers in ascending order". The first three tests are the trap. Every one was chosen because the guess predicted "yes", so a "yes" could never show the guess was wrong; many other rules, including the true one, predict "yes" for them too. Only a test that your idea says should fail can tell you your idea is wrong. That is falsifiability, applied to everyday reasoning. Wason found that people had great difficulty finding the rule and often announced far more specific ones, because they tested almost only triples that fit their own guess.

Still argued

Whether the 2-4-6 results show a bias or a sensible strategy is debated. In 1987 the psychologists Joshua Klayman and Young-Won Ha argued that testing cases your hypothesis predicts, which they called the positive test strategy, is often an efficient way to learn; it fails in Wason's task because the true rule happens to be much broader than most guesses. On their view, the error lies in not noticing when a positive test can't catch you out, not in testing positively as such. Others reply that in daily life, where beliefs come with feelings and stakes attached, the one-sided search Wason found is common and costly.

Does confirmation bias affect experts?

Yes, and training doesn't make anyone immune. Two cases show how.

In May 2004, the FBI arrested Brandon Mayfield, a lawyer in Oregon, after its fingerprint examiners matched his print to one found on a bag of detonators linked to the Madrid train bombings of March that year. Other FBI examiners verified the match, and an independent expert appointed by the court agreed. About two weeks after the arrest, Spanish police identified the print as belonging to an Algerian man, and the FBI withdrew its identification. The Justice Department's Inspector General, reviewing the case in 2006, listed among the main causes "bias from the exemplar prints (circular reasoning)": having Mayfield's print in front of them, examiners read features of it into the blurry print from the bag (Office of the Inspector General).

Science has the same weakness, which is why its fixes are procedural. Since the US registry ClinicalTrials.gov opened in 2000, large clinical trials have increasingly been required to register their main outcome in advance, so researchers can't choose afterwards the measure that looks best. A 2015 study in PLOS ONE of large heart-disease trials funded by the National Heart, Lung, and Blood Institute found that 17 of 30 published before 2000 (57%) reported a significant benefit, against 2 of 25 published afterwards (8%), and that registering in advance was strongly linked to the drop. The authors suggested that declaring outcomes in advance may have contributed to the drop: with the measure fixed beforehand, there is less room to settle on whichever one looks best. The same flexibility, used on data after the fact, is how researchers can talk themselves into a small p-value.

Why do people have confirmation bias?

There are two kinds of explanation, and both probably play a part. The cognitive one says that testing an idea fairly is hard work. It is easier to hold one hypothesis in mind and check whether the world fits it than to keep several open and work out which evidence would separate them. The motivational one says that being wrong feels bad, so people steer away from evidence that would force a change of mind, which links it to cognitive dissonance, the discomfort of holding beliefs that clash.

Confirmation bias also feeds itself. Each confirming case raises confidence, and confidence makes the next contrary case easier to dismiss. That is why people on opposite sides of a question can look at the same news and both come away more certain.

How can you counter confirmation bias?

Knowing about the bias helps little on its own. Habits that change the procedure help more:

• Ask what would prove you wrong, before looking. Then look for that, as in the second half of the 2-4-6 table.
• Write the test down first. Fix the question, the measure and what counts as success before seeing results, as preregistered trials do.
• Search both ways. Search "does X work" and "X doesn't work", not only "benefits of X".
• Argue the other side. State the strongest case against your view well enough that someone who holds it would agree with your summary.
• Count the misses. Keep a written record of predictions, so memory can't quietly keep only the hits.
• Get a fresh second opinion. A reviewer who doesn't know the first answer can't be steered by it; in the Mayfield case the FBI's verifier already knew a colleague had made the match, and the Inspector General urged fully independent checks.

These are the working tools of critical thinking, and they rest on a clear idea of what counts as evidence in the first place.

Check yourself

• Without looking back, explain why testing 20, 22, 24 told the guesser nothing about whether the rule was "going up by 2".
• Name the three stages where confirmation bias can enter, with an example of each.
• Why does registering a trial's main outcome in advance protect against confirmation bias?

Sources

• On the failure to eliminate hypotheses in a conceptual task — Peter Wason, Quarterly Journal of Experimental Psychology, 1960
• Confirmation bias: A ubiquitous phenomenon in many guises — Raymond Nickerson, Review of General Psychology, 1998
• A Review of the FBI's Handling of the Brandon Mayfield Case — U.S. Department of Justice, Office of the Inspector General, 2006
• Likelihood of Null Effects of Large NHLBI Clinical Trials Has Increased over Time — Robert Kaplan and Veronica Irvin, PLOS ONE, 2015
• Confirmation bias — Wikipedia