Install Understanding people

Add the book to your home screen or desktop for a separate reading window.

If your browser has no install option, you can bookmark the book and read it here.

Search
Browse the book & library

Confirmation bias

Reading list

Listen to this lesson

Synthetic voice
Choose a section
  1. Read Give a belief a fair test
  2. Read Where information gets left out
  3. Read Why it can happen
  4. Read Notice an unequal standard
  5. Read Check fairly without dismissing a real concern
  6. Read What practice can change
  7. Read Choose one check

Andrew · Synthetic narration · Spoken lesson

A spoken adaptation. Full references and additional practice are on the written page. Audio streams when played; download the MP3 for offline listening.

On this page

Looking for or interpreting information in ways that favour what you already believe. The problem is how you check the belief—not simply whether you have one.

Remember: Before collecting another example that fits, decide what would count against your explanation.

Examples are fictional. The comparisons and practice steps are editorial suggestions, not a test that identifies bias in a person.

Listen

Detailed explanation

Why confirmation bias happens, how to notice an uneven test, and what can help. Audio loads when you choose to play or select a recording section.

10:30 · Sonia · Synthetic narration · Examples are fictional

Try the guided listening exercise · 3:50.

Recording sections, transcript and research

Select a time to listen from that section, or read its transcript below.

  1. Read what confirmation bias is
  2. Read where it enters your reasoning
  3. Read why it happens
  4. Read how to notice it in yourself
  5. Read give a belief a fair test
  6. Read can practice change the habit?
  7. Read use it to collaborate, not to label people
  8. Read what to take into everyday life

What confirmation bias is

Confirmation bias is a habit in how we handle evidence. We look for information, notice it, or explain it in ways that favour what we already believe. The belief can be about another person, a decision, or ourselves. The problem is not simply having a belief. We need working explanations to get through the day. The problem is giving that explanation an easier test than its alternatives.

Suppose you think a colleague dislikes your ideas. Their criticism stands out. Their positive comments barely register, or you dismiss them as politeness. Your conclusion might still be right. But you have made it harder to discover whether it is wrong.

Psychologist Raymond Nickerson's 1998 review brought together earlier research on this pattern across many kinds of reasoning. It was a review of research, not a test that can tell us whether a particular person is biased. In this explanation, the brief examples are fictional. The research sources are linked beside the transcript.

Research behind this section: Nickerson (1998). Confirmation bias: A ubiquitous phenomenon in many guises. A review of earlier research on selective seeking and interpretation of evidence, not a test of an individual’s reasoning.

Where it enters your reasoning

Confirmation bias can enter at several points.

First, what you look for. You suspect a friend is unreliable, so you search your messages for cancelled plans. You never check the plans they kept. A more useful comparison would include both, over the same period.

Second, what you notice and remember. A sharp comment is memorable. An ordinary helpful action may be easier to overlook. Memory is not a complete record. Forgetting something does not, by itself, establish confirmation bias. Ask whether your checking process systematically favours one account.

Third, how you interpret the same information. When someone you trust asks a question, you call it curiosity. When someone you distrust asks it, you call it an attack. Different histories can justify different expectations. But it is worth asking what evidence, beyond your expectation, supports the difference this time.

Finally, the standard you demand. A rumour that fits your belief seems convincing. A document that challenges it suddenly needs several independent witnesses. The useful comparison is whether you would accept the same quality of evidence if it pointed the other way.

Research behind this section: Nickerson (1998). Confirmation bias: A ubiquitous phenomenon in many guises. A review of earlier research on selective seeking and interpretation of evidence, not a test of an individual’s reasoning.

Why it happens

There is no single reason for confirmation bias.

One route is the way we test an idea. We often look where we expect a prediction to hold. If you think a child struggles with maths, you look closely at a difficult maths assignment. That can tell you something. But difficulty on that assignment could also reflect unfamiliar instructions or tiredness. If all three explanations predict difficulty, that result does little to distinguish them.

In their 1987 analysis of hypothesis testing, Joshua Klayman and Young Won Ha explained why looking for expected results can sometimes be useful. It becomes misleading when you treat a result that fits several explanations as proof of just one. A better test asks where the predictions differ. Does the child struggle with familiar questions too? Does a clearer instruction change what they can do? Those comparisons are more informative than collecting another difficult assignment.

Another route is wanting a particular answer. Admitting that a plan is failing might mean losing money, disappointing someone, or acknowledging your own mistake. Ziva Kunda's 1990 review of motivated reasoning describes how desired conclusions can influence the arguments people find acceptable. This need not be deliberate dishonesty. Someone can sincerely feel that their reasoning is fair.

Wanting and expecting are different, though. You may hate the thought that a friend does not care about you, yet still interpret their behaviour through that expectation. Confirmation bias can protect a comforting story or keep a painful one going. You cannot identify the cause of a particular judgment merely by naming one of these possibilities.

Research behind this section: Klayman & Ha (1987). Confirmation, disconfirmation, and information in hypothesis testing. A theoretical analysis of hypothesis-testing strategies: looking where a prediction should hold can be informative, but is not always a fair test.; Kunda (1990). The case for motivated reasoning. Reviews how goals can influence reasoning; wanting a conclusion and expecting it are related but different processes.

How to notice it in yourself

Start with your own checking process. Four questions are useful.

What did I leave out? If you list only the times someone interrupted, have you also considered conversations where they listened? You do not need a perfect record. You need to know what your record leaves unknown.

What would change my mind? Try to answer before finding the next example. If every exception gets a new excuse, your belief may have become impossible to challenge.

Am I using the same standard? Imagine the information supported the opposite conclusion. Would you suddenly consider it good enough?

Does my question already contain my verdict? Asking, why do you never support me, assumes that support never occurs. Asking, which part of this proposal concerns you, leaves room for an answer you did not predict.

These questions identify places to check. They do not produce a bias score. Several supporting examples may be strong evidence. One exception may narrow a claim without overturning it.

Give a belief a fair test

Here is a practical routine. State the claim. Compare an alternative. Choose a useful check. Then keep, narrow or change the claim.

Take the belief, I always avoid difficult conversations. Start with something you could actually examine: I postponed two conversations about money this month. That describes occasions. It does not yet establish always.

Next, consider a plausible alternative. Perhaps you avoid every difficult conversation. Or perhaps the money conversations are difficult because you do not have the figures you need. The explanations could both play a part.

Now choose a comparison that would help. Were there other difficult topics you did discuss? Did postponement continue after the figures became available? Decide beforehand which results would strengthen or weaken each explanation. Use the same time period and include occasions that do not fit. Looking only at examples you selected because they went badly cannot tell you how typical the pattern is.

After checking, you might keep the concern, narrow it to a particular topic, or change it. A useful result could be: I do raise difficult issues, but I postpone money conversations when I feel unprepared. That suggests a more specific action: gather the figures and agree a time to talk.

This routine is our practical application of the research. Do not manufacture a confrontation or snoop to get data. Choose ordinary information you are entitled to use, or an honest question. If the answer remains uncertain, say so. You can still choose a reasonable next step.

Can practice change the habit?

Research gives some reason for optimism, with limits.

In two experiments published in 1984, Charles Lord and colleagues found that considering an opposing explanation reduced the judgment errors they studied. The instruction was more specific than simply telling people to be fair. It asked them to work through another possibility.

A 2015 study by Carey Morewedge and colleagues tested structured decision training using games and instructional videos. Improvements on tested judgment problems, including confirmation-bias tasks, were still present at least two months later. That does not show that a single recording will improve all your relationships. Nor does it mean someone becomes permanently unbiased.

A sensible aim is a more dependable checking habit. Look for evidence of progress in what you do: noticing a relevant exception, asking a question that could surprise you, or revising a claim when the evidence warrants it. Changing your mind every time is not the goal. Keeping a well-supported conclusion can also be good reasoning.

Research behind this section: Lord, Lepper & Preston (1984). Considering the opposite: A corrective strategy for social judgment. Two experiments supporting consideration of alternatives for the judgment errors studied, not validation of this worksheet.; Morewedge et al. (2015). Debiasing decisions: Improved decision making with a single training intervention. Two longitudinal experiments with games and instructional videos; gains on tested judgment problems persisted at least two months. This does not establish lasting improvement in family or workplace relationships from reading this entry.

Use it to collaborate, not to label people

Calling someone biased rarely supplies the missing evidence. Try making the disagreement more specific.

For a work decision, you might ask: what result would make each of us reconsider this plan? Agree on the same comparison before either side sees the result. For a family disagreement, you might say: we remember that differently; which part can we check, and which part remains uncertain? These are suggested ways to make a conversation more useful, not guarantees of agreement.

A fair hearing does not require treating every explanation as equally likely. One kind message does not cancel a repeated harmful pattern. You do not have to settle someone's motive before setting a limit on how they treat you. And if asking directly could expose you to harm or retaliation, seek a safer way to get support.

There is also a useful distinction from correspondence bias. Correspondence bias is explaining an action too readily through character: they declined, so they are selfish. Confirmation bias concerns what you do with the evidence after forming a belief. You could make an overconfident character judgment, then notice only the actions that appear to support it. Checking the circumstances and checking your treatment of evidence address different parts of the problem.

Research behind this section: Gilbert & Malone (1995). The correspondence bias. A review describing several routes to underweighting circumstances when judging a person. It does not establish that every character judgment is mistaken.

What to take into everyday life

Keep this question available: what would I count as evidence against my explanation?

Before your next strong judgment, state the specific claim. Choose a plausible alternative. Decide what comparison would tell them apart. After checking, keep, narrow or change the claim, and choose a practical response.

To check your understanding now: if you find three examples that support a belief, does that prove confirmation bias? No. Ask how the examples were selected and what relevant information was left out. And if someone disagrees with you, does that show they are biased? No. Examine the evidence and the standard each of you is using.

The separate listening exercise gives you a short message to examine twice, with space to record what supports and challenges your first impression. Use it to practise the comparison. Then try one fair check in an everyday situation. Judge the value by whether it helps you understand more accurately and work together more constructively.

What it looks like

Suppose you believe a friend only contacts you when they need a favour. You remember two requests but overlook a message asking how you are. This fictional example illustrates selective attention to evidence. It does not settle whether the friendship is one-sided.

Confirmation bias can affect what you search for, what you notice and how you explain conflicting information. It can support a flattering belief too: “My team always agrees with me.” Confidence or disagreement alone does not identify it.

Research behind this explanation: Nickerson (1998)

Why does it happen?

There is no single cause. One route is testing only where your prediction seems likely to hold: if you suspect someone is selfish, you look for selfish acts. Klayman and Ha explain why this way of testing can sometimes be useful. It becomes misleading when the same evidence would also fit another explanation and you never make that comparison.

Another route is wanting a particular answer. Accepting a mistake might threaten your pride, a valued relationship or a group commitment. Kunda’s review describes how desired conclusions can influence which arguments people accept. You can also favour an explanation you dislike, simply because you already expect it.

These are research-based explanations of how biased reasoning can arise, not a way to know why you or another person overlooked a particular fact.

Research behind this explanation: Klayman & Ha (1987); Kunda (1990)

How can I recognise it?

Inspect the checking process. These prompts can reveal an uneven test; they are not a score or proof of bias.

A thought to examine A more useful check
I only collect examples that fit. Choose a defined period, such as the last month. Include ordinary contact and exceptions, not just memorable requests.
I explain away every exception. Before looking again, write what would genuinely change your view. Does each new exception face a new rule?
I demand stronger evidence for the answer I dislike. Would you accept the same quality of evidence if it supported your preferred conclusion?
I ask a question that assumes the answer. Replace “Why do you only call for favours?” with “Could we make time to catch up when neither of us needs anything?” Then observe what happens.

What is not enough to identify it?

Finding several supporting examples is not automatically biased. Strong evidence may favour one explanation. Nor does one friendly message cancel repeated requests or mistreatment. The useful question is whether you gave relevant contrary information a fair hearing.

With someone else, discuss the missing evidence rather than declaring, “You have confirmation bias.” Try, “Which observation would lead either of us to reconsider?” Refusing to agree with you is not itself evidence of a bias.

Can it change?

People can improve some judgment habits. Lord and colleagues found that considering an opposing explanation reduced the errors studied in two experiments. Morewedge and colleagues found improvements after structured training, including on confirmation-bias tasks, with effects still present at least two months later. That supports practice, not a promise that knowing the definition removes the bias.

Try a repeatable routine: state your explanation, choose a plausible alternative, decide what would distinguish them, and record what you learn. Judge progress by whether you notice relevant exceptions and revise claims when warranted. Always changing your mind would be another poor rule.

The routine below is an editorial application of the research. Its effect on your own relationships needs to be judged from what happens.

Research behind this explanation: Lord, Lepper & Preston (1984); Morewedge et al

How is this different from correspondence bias?

Confirmation bias concerns how you handle evidence. Correspondence bias concerns explaining an action too readily through someone’s character. Both can occur together: “They are selfish” may be an overconfident character judgment, followed by selective collection of supporting examples.

Read the other bias · Return to Chapter 1.

Put it into practice: give one belief a fair test

  1. Choose one modest claim about yourself or another person. Record what actually happened separately from the claim.
  2. Choose a relevant comparison and a check you can make without snooping or provoking someone. Write what would strengthen and weaken the claim before checking.
  3. After checking, keep, narrow or change your explanation. Note whether this helped you ask a clearer question or collaborate more fairly. A result that supports the original concern is allowed.
My practice notes: Give one belief a fair test

Notes save automatically in this browser when storage is available. Avoid names and sensitive details, especially on a shared browser. Download a text copy before clearing your browser data.

Optional guided listening exercise · 3:50

Hear the same fictional message twice. Compare the evidence you noticed and the evidence you dismissed.

3:50 · Sonia and Ryan · Synthetic narration · Examples are fictional

Imagine you suspect: “Alex only contacts me when they need something.” Listen before opening the transcript below. After the first message, write both the supporting details and the details that make the belief less certain. Leave those answers unchanged when adding your second-listen notes.

There are two 15-second pauses. Pause the player for as long as you need. Use the notes below or paper. If listening is difficult, read the transcript one section at a time.

My practice notes: What did I notice?

Notes save automatically in this browser when storage is available. Avoid names and sensitive details, especially on a shared browser. Download a text copy before clearing your browser data.

Exercise transcript and review — open after your first listen, or use instead of audio

Set up

Try a small listening experiment. Confirmation bias means looking for information, or reading it in a way, that supports what you already believe. This exercise lets you practise checking both sides. It is not a test of whether you are biased.

Imagine you suspect that your friend Alex only contacts you when they need something. That is a belief to examine, not a fact we have established. You are about to hear a fictional voice message. Listen once. Afterwards, write down the words that support your belief, and the words that make it less certain. Here is Alex.

Alex’s message

Hi. How did your appointment go? You sounded worried on Sunday, and I've been wondering how you got on. Also, could I borrow your drill this weekend? Mine has stopped working. If you need it, no problem. And are you free for a walk next week? I'd like to catch up properly. Tuesday or Thursday would work for me. Let me know about the appointment when you feel like talking.

First comparison

Before the replay, make two short notes. What supports your belief? What challenges it? Use Alex's actual words. If you remember nothing for one side, write that. Leave your first answers in place so you can compare them later. There will be fifteen seconds of quiet. Pause the recording if you need longer.

Pause for 15 seconds, or longer.

Listen again

Now listen to exactly the same message. Try to notice anything you missed, or explained away, the first time.

The same message, replayed

Hi. How did your appointment go? You sounded worried on Sunday, and I've been wondering how you got on. Also, could I borrow your drill this weekend? Mine has stopped working. If you need it, no problem. And are you free for a walk next week? I'd like to catch up properly. Tuesday or Thursday would work for me. Let me know about the appointment when you feel like talking.

Second comparison

What changed in what you noticed? Add a note without rewriting your first answers. Take another fifteen seconds, or pause for longer.

Pause for 15 seconds, or longer.

Review

The drill request supports the narrower statement: Alex is asking for a favour today. The appointment question and the invitation to walk challenge the word only. They do not prove Alex's sincerity or settle the history of your friendship.

Perhaps you remembered the request but dismissed the interest as politeness. Ask whether you would have accepted those same words as genuine interest from someone you trusted. That difference in how you judge evidence is worth examining. Missing a detail by itself could simply be a lapse of attention or memory.

Would you keep your original belief, narrow it, or change it? What would help you decide? You could compare several recent conversations: contact with and without requests, follow-through on invitations, and how each of you responds when the other needs help. Include exceptions.

The useful habit is not forcing a positive interpretation. It is giving information that challenges your belief a fair hearing. Next time you feel certain about someone's motive, ask: what have I noticed, and what might I have dismissed?

This is authored practice, not a validated assessment. Missing a detail alone does not demonstrate confirmation bias. Compare how fairly you judged supporting and challenging information.

Continue in the app

Get the private app download (repository access required). Open version 0.12.0 or later, choose Start → Explore the whole map, search for Confirmation bias, then select Read the full entry → Work through my example. You can also reach the entry through the map’s text list.

Record an observation, your answer and one check to try. Save it in My situations, then choose What happened when I tried? to record the outcome and how your explanation changed.

This web page cannot open a downloaded HTML file directly. Without app access, use the notes above. Book notes and app notes are separate; they do not sync automatically.

Sources and limits

Nickerson (1998). Confirmation bias: A ubiquitous phenomenon in many guises. A review of earlier research on selective seeking and interpretation of evidence, not a test of an individual’s reasoning.

Klayman & Ha (1987). Confirmation, disconfirmation, and information in hypothesis testing. A theoretical analysis of hypothesis-testing strategies: looking where a prediction should hold can be informative, but is not always a fair test.

Kunda (1990). The case for motivated reasoning. Reviews how goals can influence reasoning; wanting a conclusion and expecting it are related but different processes.

Lord, Lepper & Preston (1984). Considering the opposite: A corrective strategy for social judgment. Two experiments supporting consideration of alternatives for the judgment errors studied, not validation of this worksheet.

Morewedge et al. (2015). Debiasing decisions: Improved decision making with a single training intervention. Two longitudinal experiments with games and instructional videos; gains on tested judgment problems persisted at least two months. This does not establish lasting improvement in family or workplace relationships from reading this entry.

Go deeper in the reference library