Confirmation Bias: Definition, Examples and How to Avoid It

Confirmation bias is the tendency to search for, interpret, favor and remember information in a way that confirms what one already believes. A simple example: someone convinced a new diet works notices every day they feel energetic and overlooks the days they do not. The term is usually traced to the British psychologist Peter Wason's reasoning experiments in the 1960s. It is one of the most robust findings in cognitive psychology and one of the most consequential, because it operates quietly across every domain of human thought — from how a scientist reads a study, to how a juror weighs evidence, to how a person interprets a partner's text message. It is not a flaw of stupid people or biased opponents; it is a default setting of normal human cognition.

Unlike a deliberate lie or a conscious choice to be partisan, confirmation bias works largely beneath awareness. We sincerely believe we are being objective even as we selectively read the headlines that flatter our view, ignore the data that contradict it, and remember our successful predictions more easily than our failed ones. Recognizing this pattern is the first step in any serious attempt at clear thinking — but, as we will see, recognition alone is rarely enough.

Understanding Confirmation Bias

A Precise Definition

Confirmation bias is the tendency to give privileged treatment to information that supports an existing belief and to discount, ignore, or reinterpret information that challenges it. The privilege can show up at any stage of cognition. We notice confirming evidence more quickly. We accept it with less scrutiny. We file it deeper into memory. And when asked later what we knew, we recall the confirming evidence first and the disconfirming evidence — if at all — only with effort.

Importantly, confirmation bias is not the same as having an opinion or being wrong. A person can be entirely correct and still process information in a confirmation-biased way. The bias is about the asymmetry of how we treat evidence, not the truth value of the underlying belief.

The Three Channels

Cognitive psychologists typically distinguish three components of the bias. Biased search means we look harder for evidence that supports our view than for evidence that contradicts it — typing the question into a search engine in a way that prejudges the answer, or reading the sources that already agree with us. Biased interpretation means that when ambiguous evidence arrives, we read it through the lens of our prior belief; the same study reads as compelling to people who already agreed with its conclusion and as flawed to those who did not. Biased memory means we more easily retrieve our hits than our misses, our correct predictions than our wrong ones.

Cold and Hot Versions

Some confirmation effects are "cold" — they reflect ordinary limits of attention, the structure of hypothesis testing, and how the brain handles ambiguity. Others are "hot" — driven by motivation, identity, and emotion. A scientist who falls in love with her own theory shows a mix of both. A partisan defending a politician shows mostly the hot version, sometimes called motivated reasoning. The line between the two is blurry, but the distinction matters when designing counters: cold biases respond to better procedures, while hot biases also require attention to identity and emotion.

Why It Is Not Stupidity

Confirmation bias has been found in physicists, judges, doctors, and intelligence analysts as readily as in undergraduates. Studies have repeatedly shown that higher cognitive ability does not reliably immunize a person; in some research, smarter people are actually better at constructing sophisticated defenses of their existing beliefs. The bias is a feature of how minds work, not a marker of how good a particular mind is.

The Research Foundation

Wason's 2-4-6 Task

The modern study of confirmation bias begins with the British psychologist Peter Wason. In a 1960 experiment, he told participants that the sequence 2-4-6 followed a rule he had in mind, and invited them to discover that rule by proposing further sequences. Most participants quickly formed a hypothesis — typically "ascending even numbers" or "adding two" — and then proposed sequences consistent with that hypothesis. They received "yes" answers, grew more confident, and announced the rule. Most were wrong. The actual rule was simply "any three increasing numbers." To discover it, they would have had to propose a sequence designed to fail under their current hypothesis (something like 3-7-9 or 1-2-100). Almost nobody did. The study became the first clean demonstration that people seek confirming, not disconfirming, evidence.

Later researchers added an important nuance. Joshua Klayman and Young-Won Ha argued in 1987 that testing cases you expect to fit your idea, which they called a positive test strategy, is often a sensible approach; it fails in the 2-4-6 task because the true rule is much broader than people's guesses. The deeper problem is not testing positive cases but failing to also try cases that could prove you wrong.

The Wason Selection Task

A few years later, Wason designed a logic puzzle that has since become a staple of cognitive psychology courses. Four cards are placed on a table — say A, K, 4, 7 — and participants are told that each card has a letter on one side and a number on the other. They are then asked which cards to turn over to test the rule, "If a card has a vowel on one side, it has an even number on the other." The logically correct answer is to turn the A (to see if the back is even) and the 7 (to see if the back is a vowel). Most people turn the A and the 4, looking for confirmation rather than testing for disconfirmation. The error has been replicated across decades and cultures, although familiar real-world content can dramatically improve performance — which itself became a foundational finding in reasoning research. Researchers still debate how much of the error is confirmation bias as such; Jonathan Evans showed that many people simply pick the cards that match the words in the rule, a tendency called matching bias.

Biased Assimilation

In a landmark 1979 study, Lord, Ross, and Lepper gave proponents and opponents of capital punishment the same two studies — one suggesting the death penalty deters crime, one suggesting it does not. Both groups read both studies. After reading, both groups were more confident in their original position, and the gap between them had widened. They had assimilated the supporting study uncritically and pulled apart the methodology of the opposing one. This pattern, often called biased assimilation, has been replicated across topics from climate science to gun control. The polarization part is less certain: in the original study, polarization was measured by asking people how much their views had changed, and later studies that measured attitudes directly before and after, such as work by Deanna Kuhn and Joseph Lao in 1996, found that true polarization is less common than biased evaluation of the evidence. The robust lesson is that the same evidence does not reliably bring opposing minds together.

Myside Bias

Decades of research, much of it associated with the psychologist Keith Stanovich, has clarified that confirmation bias is at its strongest when a belief is connected to "my side" — my political party, my religion, my profession, my identity. Stanovich uses the term myside bias to describe this concentrated, identity-protective form of confirmation bias. His research suggests that myside bias is largely independent of cognitive ability: very intelligent people can be just as biased, and sometimes more articulate in their bias, than less intelligent people on the same question.

Identity-Protective Cognition

Dan Kahan and colleagues at Yale have studied what they call identity-protective cognition. In one 2017 study, participants with strong numeracy skills correctly interpreted a tricky data table about a skin cream, but when the same numbers were presented as evidence about gun control, numerate partisans became more likely to misread the data when the correct answer contradicted their political group's view. Findings like these suggest that, on identity-laden topics, reasoning skill can be recruited to defend a position rather than to test it.

Beyond the Lab

Confirmation effects have been documented across applied fields. Studies of medical diagnosis show that the first plausible diagnosis tends to dominate subsequent reasoning, even when later test results disconfirm it. Studies of intelligence analysis, motivated in part by post-mortems of strategic surprises, have shown analysts anchoring on initial frameworks. Studies of forensic examiners, including Itiel Dror and David Charlton's 2006 work with experienced fingerprint experts, found that contextual information, such as being told a suspect had confessed, influenced how ambiguous prints were judged, sometimes leading experts to contradict their own earlier conclusions about the same prints. These findings collectively suggest that confirmation bias is not a quirk of artificial tasks but a structural feature of how expertise meets uncertainty.

How It Works

Hypothesis Testing as Default

One reason confirmation bias is so universal is that "testing a hypothesis by looking for examples of it" is a reasonable-feeling cognitive strategy. If you suspect a friend is angry with you, scanning for confirming signs feels like investigation. The problem is that the strategy is logically incomplete. Any halfway-rich hypothesis can find supporting examples in the noise of daily life. What distinguishes good thinking is the willingness to ask, "What would I expect to see if my belief were wrong?" That counterfactual question is unnatural and effortful, while the confirming question is easy.

The Cost-Benefit Logic of the Brain

From a resource-management perspective, the brain is conservative. Maintaining a belief is metabolically cheap; revising it is expensive, especially if it is tied to other beliefs, plans, and relationships. When new information arrives, the brain runs an implicit triage: How costly would it be to change my mind? How threatening is this evidence? Is there a way to read it that fits what I already think? The path of least resistance is to keep the existing belief and shape the new data to fit. This is not laziness so much as efficient default behavior in a brain that cannot afford to constantly re-examine its commitments.

Identity Protection

When a belief is part of identity, disconfirming evidence threatens not just the belief but the self. People who define themselves as scientifically literate, as religious, as politically progressive, or as fans of a particular team have a stake in the truth of beliefs associated with those identities. Research on "identity-protective cognition" has shown that on emotionally loaded topics, more knowledgeable individuals can become more polarized, not less, because they have the cognitive resources to mount sophisticated defenses of their group's position.

Memory's Quiet Selection

Memory does much of the work of confirmation bias without our noticing. We do not remember everything we ever read or heard about a topic; we remember a thin, selected subset. Items consistent with our current view are easier to retrieve, partly because they were attended to more carefully when encoded and partly because they have been reactivated more often. The result is that the past, as we recall it, looks more supportive of our present view than it actually was.

Social Reinforcement

Confirmation bias is not only individual. Social networks — both offline friendship groups and online platforms — sort people into clusters that share beliefs. Within those clusters, confirming information arrives more frequently, is praised, and is rewarded with attention and status. Disconfirming information arrives less often, and when it does, voicing it can cost social standing. The bias inside the head is reinforced by the bias inside the network, and the loop tightens.

Everyday Examples

Reading the News

Two people read the same article about an economic policy. The reader who already supports the policy notices the favorable statistics, accepts the expert quotes at face value, and dismisses the critical voices as biased. The opposing reader notices the same article's gaps, treats the favorable statistics with suspicion, and finds the critics persuasive. Each leaves the article more confident than before. They will swear, accurately, that they "read the whole thing."

Hiring Decisions

A hiring manager forms a quick impression of a candidate within the first minute of the interview. From that point on, the interview functions less like an investigation and more like a confirmation exercise. Friendly questions are asked of candidates who seem promising; harder, more skeptical questions are asked of those who don't. Answers from the favored candidate are heard generously; equivalent answers from the disfavored candidate are heard with doubt. Studies of structured versus unstructured interviews show that unstructured interviews — where this confirmation dynamic runs free — are notoriously poor predictors of job performance.

Relationships

If you have decided that your partner is selfish, every time they take the last slice of pizza becomes evidence. If you have decided that they are generous, the same action becomes a small oversight. Long-term relationships often run on these accumulating interpretive habits, and once a narrative is set, it can be remarkably resistant to revision. Therapy often works in part by interrupting these narratives and inviting fresh interpretations of the same behavior.

Investing

An investor who buys a stock develops an emotional commitment to it. Once the position is opened, news that supports the thesis is read closely; news that contradicts it is rationalized away or filed under "the market doesn't understand yet." Earnings disappointments become temporary noise; positive surprises become confirmation of a long-term thesis. This pattern helps explain why investors so often hold losing positions too long.

Searching Health Information Online

A person who suspects they have a particular condition types the condition name into a search engine, finds a list of symptoms, and recognizes many of them in their own experience. They do not search for "symptoms I have that are inconsistent with this condition," and the search engine does not volunteer that question. This is one reason online symptom lists are a poor substitute for assessment by a qualified clinician, who is trained to consider alternative explanations.

Parenting

A parent who has labeled one child "the responsible one" and another "the wild one" tends to notice — and remember — behavior that fits the label. Counterexamples get treated as exceptions. Over years, this can shape how each child is talked to, expected of, and trusted with responsibility, with effects on the child's own self-concept.

Where It Shows Up

Medicine

The phenomenon of "diagnosis momentum" is a clinical face of confirmation bias. Once a diagnosis enters the chart, subsequent clinicians often treat it as the starting point of their own thinking rather than as a hypothesis to be retested. Symptoms that fit are weighted; those that do not are written off as atypical. Diagnostic checklists, second opinions, and the deliberate practice of asking "What else could this be?" are designed in part to interrupt this momentum.

Science

Science was designed, in part, to compensate for confirmation bias in individual researchers. Peer review, replication, registered reports, and adversarial collaboration all aim to subject hypotheses to the disconfirmation that individual scientists tend to avoid. The replication crisis of the 2010s, which affected several fields including social psychology, was partly a story about how research practices that allowed selective reporting and post-hoc theorizing let confirmation bias accumulate at scale.

Politics and Media

Partisan media ecosystems and algorithmically curated feeds offer a steady supply of confirming evidence and a steady down-ranking of challenging evidence. The result is not necessarily that people believe false things — sometimes their views are reasonable — but that the gap between groups widens because each side's information diet is different. Cross-cutting exposure, when it does occur, often provokes defensive reactions rather than reflection.

Finance

Beyond individual investors, confirmation bias appears in analysts' models, in management's interpretation of business results, and in regulators' assessments of risks. Long bull markets generate confirming evidence for optimistic stories that, in hindsight, look hopelessly thin. Risk officers who flag concerns during such periods often find their analyses dismissed not because they are wrong but because they are inconvenient.

Justice

Investigators who form an early suspicion may unconsciously direct evidence collection toward that suspect. Forensic examiners who know contextual information can be influenced by it when interpreting ambiguous physical evidence. Jurors arrive with prior beliefs about the kind of person who commits the kind of crime in question. Blinding protocols, structured decision aids, and explicit consideration of alternative suspects are among the tools developed in response.

Education and Hiring

Teachers' expectations can shape how they read student work — a phenomenon studied since Rosenthal's early research on expectancy effects. Hiring panels that share information freely before independent evaluation tend to converge on the first opinion expressed. Structured rubrics, independent ratings, and blind review are common counters in both domains.

Real-World Consequences

Costly Persistence in Error

When confirmation bias is allowed to operate unchecked, it produces a characteristic failure mode: persistence in error long after warning signs have appeared. Corporate strategies that fail in foreseeable ways, policy interventions that ignore mounting counter-evidence, and personal projects that limp along well past their expiration date often share this profile. The actors are not stupid; they are filtering out signals that would force a hard reassessment.

Polarization and Tribal Truth

At a societal level, confirmation bias plus social sorting plus algorithmic curation plus partisan media produce environments in which different groups can hold incompatible beliefs about the same events with comparable confidence. Disagreements that should be resolvable by evidence resist resolution because each side has access to a curated supply of supporting evidence and treats the other's supply as compromised.

Damage to Relationships

Within families, friendships, and workplaces, confirmation bias hardens interpretive habits. Once a person is filed under a particular character, future behavior is read through that filter. This shows up in long-running conflicts where each party has assembled an impressive case for the other's bad faith — assembled, that is, by selectively retaining incidents that fit.

Cost to Decision Quality

In domains where decisions can be tracked over time — investing, sports forecasting, medical diagnosis — research shows that people who actively look for disconfirming evidence outperform those who do not. The cost of unchecked confirmation bias is not abstract; it can be measured in worse predictions and worse outcomes.

Erosion of Trust in Institutions

When organizations become known for ignoring inconvenient findings, the cost extends beyond the immediate decision. Trust in scientific institutions, governments, and businesses is partly a function of their visible willingness to be moved by evidence. Confirmation-biased institutions, even when their underlying conclusions are right, project an unattractive certainty that breeds suspicion among everyone outside their bubble.

How to Recognize It in Yourself

The Surprise Test

If you cannot remember the last time evidence changed your mind on a topic, that is a clue. Reality is complicated, and beliefs that are sensitive to evidence should occasionally update. A belief that has been impervious to a decade of new information is either an extraordinarily lucky guess at the truth or, more often, the product of a closed loop.

Notice Emotional Heat

When you read a study, a news article, or a colleague's analysis, notice your emotional reaction. A flash of satisfaction at evidence that supports your view, or a flash of irritation at evidence that contradicts it, is not in itself proof of bias — but it is a flag worth attending to. The strongest confirmation effects ride on the strongest emotional reactions.

Track Your Predictions

Most of us remember our hits and forget our misses. Writing down predictions — in advance, with dates and specifics — and checking them later is a humbling exercise. Forecasters who do this systematically tend to be less confident, more accurate, and more receptive to updating their models than those who don't.

Audit Your Information Diet

If the sources you trust agree with each other on every contested topic, that is information about the structure of your diet, not necessarily about the truth of the topics. Ask: where does my information come from? When was the last time I read a thoughtful article by someone whose conclusions I expect to dislike? Did I read it carefully or skim it for flaws?

The "How Would I Know I'm Wrong?" Question

For any belief you hold strongly, try to articulate the specific observation or evidence that would lead you to revise it. If you cannot specify what would change your mind, your belief is functioning as unfalsifiable, which is a feature of confirmation-protected positions, not of empirical ones.

How to Counter It

Consider the Opposite

The most robust debiasing intervention in the cognitive psychology literature is also one of the simplest: explicitly ask yourself, "How could this be wrong? What is the best case for the opposite view?" Charles Lord, Mark Lepper and Elizabeth Preston showed in 1984 that asking people to consider whether they would have rated a study the same way if it had produced the opposite result reduced biased assimilation, and later studies found similar benefits for overconfidence and anchoring. The instruction works not because it removes bias but because it forces the cognitive system to do, by deliberate effort, what it would not do by default — generate the disconfirming alternatives.

Adversarial Collaboration

When two researchers or analysts disagree, the temptation is to publish dueling papers. Adversarial collaboration — pioneered by psychologists including Daniel Kahneman — instead invites them to design a study together that both sides agree would settle (or at least bear meaningfully on) the disagreement. The discipline of co-designing forces each side to articulate what evidence would actually change their view, which is exactly the move confirmation bias prevents.

Pre-Registration and Pre-Mortems

In science, pre-registration commits researchers to a hypothesis, analysis plan, and primary measures before seeing data, narrowing the space for post-hoc confirmation. In organizational decision-making, the pre-mortem — imagining that a project has failed and writing the story of why — surfaces the disconfirming considerations that ordinary planning suppresses. Both techniques outsource skepticism to a procedure rather than relying on it from individual willpower.

Structured Analytic Techniques

Intelligence agencies and forensic units have developed structured methods to push against confirmation bias: analysis of competing hypotheses (where multiple hypotheses are evaluated against the same evidence), key assumptions checks, and devil's advocate roles. None of these are silver bullets, but they make disconfirmation a required part of the workflow.

Blinding

Where possible, removing identifying context from evidence reduces confirmation pressure. Blinded peer review, blinded grading, and blinded interpretation of forensic samples all aim to prevent the analyst's prior beliefs about the source from shaping the evaluation.

Designated Skeptics and Red Teams

Group decisions are particularly vulnerable to confirmation bias because once the group leans one way, dissent becomes socially costly. Assigning a person or sub-team the explicit role of arguing against the majority position — a "red team" — legitimizes disconfirmation and increases the chance that real problems are surfaced. The role works best when the position rotates and when leadership visibly rewards substantive disagreement.

Calibrated Forecasting

Forecasting tournaments have shown that systematic practice — making explicit probabilistic predictions, tracking accuracy, and breaking complex questions into smaller estimable parts — improves judgment over time. Part of the benefit is structural: writing down a forecast in advance makes confirmation bias's selective memory less able to rewrite history.

Limits of Debiasing

Individual Effort Is Not Enough

Debiasing research has been chastening. Interventions that produce measurable effects in the lab often fade outside it, and effects that are real on average can be small or unstable for any given person. Confirmation bias is not a habit you can resolve to drop the way you might resolve to stop smoking. It is a default of cognition that returns the moment attention slips.

Backfire and Identity

When a challenged belief is tied to identity, confronting people with disconfirming evidence sometimes produces a strengthening of the belief — a backfire effect — though the size and reliability of this effect is debated in current literature. What is clearer is that approaches that respect the person's identity and that make space for them to update without losing face tend to outperform approaches that frame the update as an ideological surrender.

The Role of Structure

Because individual willpower is unreliable, the most durable counters to confirmation bias are structural. Institutions and procedures that build in disconfirmation — peer review, registered reports, blinded analysis, red teams, devil's advocacy, decision audits — accomplish at the system level what individuals struggle to do alone. The bias does not vanish; it is constrained by an environment that does not let it run unchecked.

Living With It

Perhaps the most realistic stance is one of epistemic humility: you cannot eliminate confirmation bias from your thinking, but you can hold your beliefs more loosely, design your information diet more thoughtfully, and seek out the people, procedures, and habits that subject your conclusions to disconfirmation you would not otherwise pursue. The goal is not to become unbiased — no one is — but to be biased a little less than you would otherwise be, and to know that.

Test Yourself: Questions on Confirmation Bias

These practice questions, similar to those used in psychology courses and exams, check whether you can recognize confirmation bias and tell it apart from related ideas. Try each one before reading the answer.

1. A student believes she is "bad at math." After a test, she remembers the two questions she got wrong but not the eight she got right. Which component of confirmation bias is this?

Answer: Biased memory (selective recall). She retrieves the evidence that fits her belief more easily than the evidence that contradicts it.

2. In Wason's 2-4-6 task, why did most participants fail to find the rule?

Answer: They only proposed sequences that fit their own hypothesis (such as 8-10-12), so every answer came back "yes." They rarely tried a sequence that could have shown them they were wrong, such as 1-2-3 or 10-6-2.

3. In the selection task with cards A, K, 4 and 7 and the rule "if a vowel, then an even number," which cards must be turned over?

Answer: A and 7. The A could have an odd number on the back, and the 7 could have a vowel on the back; either would break the rule. Turning the 4 cannot disprove it.

4. Two groups with opposite views read the same mixed evidence on a policy, and each rates the study supporting its own side as better designed. What is this called, and who first demonstrated it?

Answer: Biased assimilation, demonstrated by Charles Lord, Lee Ross and Mark Lepper in their 1979 study of attitudes to capital punishment.

5. How does confirmation bias differ from the self-fulfilling prophecy?

Answer: Confirmation bias distorts how we gather and interpret evidence, so we wrongly believe our expectation was confirmed. In a self-fulfilling prophecy, the expectation changes behavior in a way that actually makes it come true.

6. Name one research-supported technique for reducing confirmation bias.

Answer: "Consider the opposite": deliberately generating reasons your belief could be wrong. Others include pre-registration, blinding, red teams and analysis of competing hypotheses.

For a broader set of practice items covering other biases, try our cognitive bias test.

Conclusion

Confirmation bias is one of the deepest patterns in human cognition. It is not a quirk to be embarrassed about; it is a feature of how minds save effort, protect identity, and weave new information into existing structures. Understanding the bias does not abolish it, but it does change what you can do about it — both individually, through habits like consider-the-opposite and prediction tracking, and collectively, through the institutional architectures that make disconfirmation a required step rather than an optional one.

What separates the more accurate thinker from the less accurate one is rarely the absence of bias. It is the presence of a system — internal habits and external structures — that exposes beliefs to evidence that would, in the absence of effort, never have been encountered. Science at its best is such a system. Good investigative journalism, well-run organizations, and serious deliberative groups can be such systems too. So can a personal practice of writing down predictions, articulating disconfirming conditions, and reading sources you expect to disagree with.

The temptation, when learning about confirmation bias, is to feel relieved that you are now aware of it — and then to apply that awareness mostly to the failings of others. That move is itself a form of the bias. The serious application is inward. The beliefs most worth examining are the ones you hold most strongly and have re-examined least recently. Confirmation bias is what makes that examination feel unnecessary; clear thinking begins where that feeling is treated as a clue.