Skip to content
mentua.org

What Is Confirmation Bias?

Line drawing of a magnifying glass over a small filled circle with three points inside it, all within a large dotted circle that also holds one small empty dotted ringLine drawing of a magnifying glass over a small filled circle with three points inside it, all within a large dotted circle that also holds one small empty dotted ring

5 min read≈ 1,180 words3 questions · up to 3 points

What this is about

In a famous experiment, most people tested their idea again and again, got a yes almost every time, and were still wrong. You learn what went wrong in their tests, how the same habit changes the way we judge studies we dislike, and one question that protects you against it. You also learn when looking for agreement is the sensible thing to do.

In a nutshell· skip ahead

Confirmation bias is the tendency to look for, notice and believe information that supports what you already think, and to overlook or discount information that speaks against it. It shapes which questions you ask and how strictly you judge the answers. The result is that a belief can feel well tested when it has never been put at risk.

In 1960 the psychologist Peter Wason showed 29 students three numbers, 2, 4 and 6, and told them that the numbers followed a simple rule. They could try any three numbers, he would say whether these fit the rule, and they should announce the rule only when they were sure. Most students tried only three or four triples, heard “yes” almost every time, and announced their guess with confidence, but only 6 of the 29 named the correct rule at their first attempt. How can every test agree with you while you are still wrong?

What is confirmation bias?

The students were showing confirmation bias, the tendency to look for, notice and believe information that supports what you already think, and to overlook or discount information against it. The bias needs no strong opinion: in Wason’s task nobody cared about the answer, and the students still tested their idea in a way that could only agree with it.

  1. The guess. A student sees 2, 4, 6 and thinks of “numbers that go up by two”, the most common first guess.
  2. The tests. The student tries 8, 10, 12 and hears “yes”, then 20, 22, 24 and hears “yes” again, and each answer feels like proof.
  3. The announcement. After a few hits the student announces “numbers going up by two” and is told that this is wrong.
  4. The real rule. The rule was “any three numbers in increasing order”. Every triple the student tried also fit this much wider rule, so the answers could never show the difference.
  5. The missing test. A triple such as 1, 2, 3, which the student’s own rule says should fail, would have helped, because a “yes” would have shown at once that the guess was too narrow.

The students used what researchers call a positive test strategy: they tested only the cases their idea predicted. These tests feel informative because they produce agreement, but here no answer could have shown the idea to be wrong.

Does it also change how we judge evidence?

It does. Choosing which questions to ask is one part of confirmation bias, and judging the answers is the other, which psychologists call biased assimilation: people accept evidence that fits their view easily and examine evidence against it much more strictly.

In 1979 Charles Lord, Lee Ross and Mark Lepper gave 48 Stanford students, half for the death penalty and half against, two invented studies on whether it deters murder. One study found a deterrent effect and one did not, and the methods were switched between students, so the method itself could not explain the ratings. Supporters still rated the pro-deterrence study as better conducted, and opponents rated the other one as better. When Andrew Guess and Alexander Coppock repeated the study exactly with online participants in 2014, they found almost the same pattern.

The 1979 authors also reported that both sides became more extreme, but that finding rested on people’s own reports of how much they had changed. When Guess and Coppock gave people different evidence at random and measured their views directly, supporters and opponents both moved a little toward whatever evidence they had read. People judge evidence unfairly, but they still learn from it, only less than they should.

What Is Cognitive Dissonance? explains why people defend a choice they have already made. Confirmation bias needs no earlier choice, as the 2-4-6 task shows, although dissonance can make it stronger.

What confirmation bias teaches you

The useful question is: “Which result would show that I am wrong, and have I looked for it?” A second question helps with any piece of evidence: would I also see this if my idea were false? Evidence that appears either way proves little, which is the logic of Bayes’ Theorem, Explained.

  • Health. If you search online for “can coffee cause headaches”, you will find pages that say yes. To test the idea, note your headaches on days without coffee as well, and compare.
  • People. Once you think a new colleague is unfriendly, every short email looks like proof. Count the friendly emails too, and ask how you would read the short ones from a colleague you like.
  • The news. When a study reaches a conclusion you dislike, you look for flaws in its method. Ask whether you would look just as hard if the result went your way.

Some professions build these questions into their methods. In Psychology of Intelligence Analysis (1999), the CIA’s Richards Heuer taught analysts to list all reasonable explanations side by side and to give little weight to evidence that fits all of them equally well. The same habit also helps to explain What Is the Anchoring Effect?, because checking whether an anchor could be right brings to mind mostly the reasons why it could.

Is looking for confirmation always a mistake?

No, and this is the strongest objection to the usual story. In 1987 Joshua Klayman and Young-Won Ha showed that testing the cases your idea predicts is often a good strategy. It fails mainly when your guess is narrower than the truth, as in the 2-4-6 task, because then only the tests your guess says will fail can reveal the error. When the thing you are looking for is rare, testing where you expect it is often the better way to find out that you are wrong. Wason’s task was built for exactly the case in which the habit fails.

A second misunderstanding is that confirmation bias makes people reject every fact that does not suit them. The research above suggests otherwise: people discount such facts but still move toward them, and the idea of a general “backfire”, in which contrary evidence makes beliefs stronger, has not held up well. The bias is also different from building a case on purpose. A lawyer who presents only the evidence for a client knows what she is doing, while confirmation bias works without our noticing it.

Learn more

Confirmation bias has no single discoverer. In 1620 the English philosopher Francis Bacon described it in his Novum Organum. He told of a man who was shown pictures in a temple of sailors who had made vows to the gods and escaped shipwreck, and who asked where the pictures were of those who had made vows and drowned anyway.

The human understanding when it has once adopted an opinion (either as being the received opinion or as being agreeable to itself) draws all things else to support and agree with it.

Francis Bacon, Novum Organum (1620), Book I, Aphorism 46, tr. James Spedding, Robert Ellis and Douglas Heath

Peter Wason turned the idea into experiments in the 1960s. Raymond Nickerson’s review of 1998 collected decades of studies under the single term, and Klayman and Ha showed when the habit helps and when it hurts.

To see the bias at work together with self-justification, read “Mistakes Were Made (but Not by Me)” by Carol Tavris and Elliot Aronson: Summary and Review. The book shows how the two keep therapists, police officers and couples from seeing their own mistakes.

Quiz

3 questions · up to 3 points

Open a question and pick one answer. Only your first answer counts: +1 if it is right, −1 if it is wrong.

  1. In Wason's 2-4-6 task, why did most people announce a wrong rule although the triples they tested kept getting a yes?
    Show answer

    They tested only triples their own rule predicted, and these also fit the wider real rule. Triples like 8, 10, 12 fit both 'goes up by two' and 'any increasing numbers', so the yes answers could not tell the two rules apart.

  2. You believe a new diet gives you more energy. Which check would tell you most about whether you are right?
    Show answer

    Stop the diet for two weeks and note whether your energy really drops. This is a result that could prove the belief wrong. If your energy stays the same without the diet, the diet was probably not the cause.

  3. When supporters and opponents of the death penalty read the same mixed evidence, what does the later research suggest happens?
    Show answer

    Both sides rate the study that agrees with them as better, but both still shift a little toward the evidence. The biased judgment of the studies replicated well, but when people were randomly given different evidence, both sides moved in the direction of what they read.

The takeaway

Confirmation bias is the tendency to search for and accept evidence that agrees with what you already believe, and to discount evidence against it. In Wason's 2-4-6 task, most people tested only number triples their own rule predicted, so almost every answer was yes and their wrong rule felt proven. People also rate the same study design as better when its results suit them, but they still move toward the evidence. Looking for agreement is not always a mistake, so before trusting a belief, ask which result would prove it wrong and look for that.

References

  1. On the Failure to Eliminate Hypotheses in a Conceptual Task. P. C. Wason, Quarterly Journal of Experimental Psychology 12 (3), 129–140, via the Internet Archive, 1960

    The original 2-4-6 experiment with 29 students, including the guesses they announced and the triples they tried.

  2. Confirmation Bias: A Ubiquitous Phenomenon in Many Guises. Raymond S. Nickerson, Review of General Psychology 2 (2), 175–220, 1998

    The standard review: the many forms of the bias, examples from medicine, law and science, and a discussion of when it might be useful.

  3. Confirmation, Disconfirmation, and Information in Hypothesis Testing. Joshua Klayman and Young-Won Ha, Psychological Review 94 (2), 211–228, 1987

    The strongest objection: testing the cases your idea predicts is often a good strategy, and fails mainly when your idea is narrower than the truth.

  4. Does Counter-Attitudinal Information Cause Backlash? Online Appendix. Andrew Guess and Alexander Coppock, British Journal of Political Science 50 (4), 1497–1515, 2020

    Appendix C repeats the 1979 death penalty study exactly: the biased judgment of the studies returns, but random assignment shows people still move toward the evidence.

confirmation-biascognitive-biasbiased-assimilationpositive-test-strategywason-2-4-6-taskpeter-wasonhypothesis-testingmotivated-reasoningfalsificationpolarizationfrancis-bacondecision-makingcritical-thinkingsocial-psychologyreplication

Bayes' Theorem, Explained

Why a positive test result often means less than it seems. Bayes' theorem worked through with 1,000 women and a mammogram, step by step.

8 min read≈ 1,770 words3 questions