On the morning of February 5, 2003, Colin Powell stood before the United Nations Security Council and delivered what he called his "personal" guarantee that Iraq possessed active weapons of mass destruction. He cited intercepted communications, satellite imagery of suspected chemical weapons facilities, and informant testimony. "Every statement I make today is backed up by sources, solid sources," he told the assembled delegates. "These are not assertions. What we're giving you are facts and conclusions based on solid intelligence."
The intelligence was not solid. It was the product of one of the most consequential and well-documented episodes of confirmation bias in modern history — a systematic process in which analysts, policymakers, and senior officials interpreted ambiguous and contradictory evidence to mean what they already believed it meant. An internal CIA review later found that analysts had been under pressure to produce assessments consistent with the administration's position. A Defense Intelligence Agency report from 2002 had quietly noted that "there is no reliable information on whether Iraq is producing and stockpiling chemical weapons." That sentence was not read into the record on February 5.
The consequences were measured in hundreds of thousands of lives, trillions of dollars, and a regional destabilization whose effects are still unfolding. The failure was not primarily one of insufficient data. The failure was cognitive: a systematic tendency of the human mind to seek, notice, interpret, and remember evidence in ways that confirm what it already suspects.
What Confirmation Bias Is — and What It Is Not
Confirmation bias is the tendency to search for, interpret, favor, and recall information in a way that confirms or supports one's prior beliefs or values. The term was coined and systematically studied by British psychologist Peter Wason in the 1960s, though the phenomenon itself was first described by Francis Bacon in 1620. It operates not as a single cognitive error but as a cluster of related processes that occur at every stage of information handling: acquisition, interpretation, encoding, and retrieval.
Confirmation bias is often confused with related but distinct phenomena. The following table clarifies the differences:
| Concept | Definition | Core Mechanism | When It Fires |
|---|---|---|---|
| Confirmation bias | Tendency to seek and favor evidence that supports existing beliefs | Asymmetric evidence search and interpretation | Whenever a belief is held and new information is encountered |
| Motivated reasoning | Reasoning guided by a desired conclusion rather than the evidence | Emotion-driven logic that works backward from preferred outcome | When conclusions have personal stakes (identity, ego, reputation) |
| Belief perseverance | Maintaining a belief even after the evidence for it has been explicitly discredited | Compartmentalization of the original belief from its supporting evidence | After public correction or retraction |
| Myside bias | Evaluating arguments and evidence more favorably when they support one's own side | In-group evaluation standards applied to ideas | In adversarial or competitive reasoning contexts |
These phenomena often co-occur and reinforce each other, but they are mechanistically distinct. Confirmation bias does not require emotional investment; it can operate in purely cold cognition, influencing even well-intentioned, careful reasoners who believe they are being objective. This is what makes it so pervasive and so dangerous.
Why the Brain Does This: The Cognitive Architecture of Confirmation
The Wason Selection Task
In 1960, Peter Wason published what would become one of the most replicated and discussed experiments in the history of psychology. Participants were shown four cards laid flat on a table, each with a letter on one side and a number on the other. The visible faces showed: E, K, 4, and 7. Participants were told the rule: If a card has a vowel on one side, then it has an even number on the other side. They were asked which cards they needed to turn over to determine whether the rule was true or false.
The correct answer is E and 7. You must turn over E to check whether it has an even number on the back, and you must turn over 7 to check whether it has a vowel — because a vowel on the back of the 7 would violate the rule. The 4 is irrelevant: if it has a consonant on the back, the rule is unaffected; if it has a vowel, that confirms the rule but turning over 4 cannot disprove it.
Only about 10 percent of participants chose the correct combination. The overwhelming majority chose E and 4 — both cards that could only confirm the rule, never disprove it. Almost no one spontaneously chose the 7, the only card capable of falsifying the hypothesis. Wason's interpretation was stark: people do not naturally test hypotheses by looking for disconfirming evidence. They test hypotheses by looking for confirming evidence.
This finding has been replicated hundreds of times across cultures, education levels, and populations. Logicians, scientists, and trained statisticians perform only marginally better than chance on the original abstract version of the task. When the task is given a social framing — checking that people drinking in a bar are old enough — performance improves dramatically, because humans have evolved specialized reasoning circuits for detecting cheaters in social exchanges. The purely logical version exposes the default setting: seek confirmation.
The Lord, Ross, and Lepper Experiment
In 1979, Charles Lord, Lee Ross, and Mark Lepper published a study that remains one of the most disturbing demonstrations of confirmation bias in operation. They recruited participants who either strongly supported or strongly opposed capital punishment, confirmed their views through a pre-test, and then gave all participants the same two-part portfolio of evidence about whether the death penalty deters crime.
One study in the portfolio appeared to support deterrence; the other appeared to undercut it. Both were described with identical methodological detail. The researchers measured how participants rated the quality and persuasiveness of each study and how their overall attitudes changed after reading both.
The results were striking. Participants on both sides rated the study that supported their prior view as more methodologically rigorous, more convincing, and more relevant. They rated the opposing study as flawed, poorly designed, or not applicable. Most significantly, after reading exactly the same mixed evidence, participants on both sides became more extreme in their views than before they started. The pro-capital punishment group became more pro-capital punishment; the anti-capital punishment group became more anti-capital punishment. This phenomenon — known as attitude polarization — means that exposure to balanced evidence does not reduce polarization when confirmation bias is operating. It can increase it.
The Asymmetry of Evidence Processing
The neuroscience and cognitive psychology of this asymmetry have become clearer over the past two decades. When we encounter evidence consistent with our existing beliefs, we tend to accept it at face value, processing it fluently and storing it efficiently. When we encounter disconfirming evidence, we tend to engage in what researchers call disfluency processing — scrutinizing the evidence more carefully, generating counterarguments, questioning methodology, and looking for reasons to discount it.
This asymmetry is not irrational in origin. It reflects a reasonable heuristic: well-established beliefs are usually well-established because they have survived prior encounters with evidence. New evidence that fits the model can be processed quickly; evidence that contradicts a stable model deserves more scrutiny. The problem is that this heuristic operates uniformly, even when the belief being protected was never well-founded in the first place — or when the new disconfirming evidence is actually more reliable than the old confirming evidence.
Motivated cognition overlays this structural bias with emotional energy. When beliefs are tied to identity, group membership, or professional reputation, the brain treats contradictory evidence as a threat to be neutralized rather than information to be integrated. Neuroimaging studies by Drew Westen and colleagues at Emory University (2006) showed that when partisans processed threatening political information, the brain's threat-response circuits activated and the reasoning circuits went quiet — and the threat circuits deactivated once a face-saving justification was found, triggering a dopamine-like reward response.
The result is that confirmation bias feels like good reasoning. The scrutiny applied to disconfirming evidence feels like rigor. The acceptance of confirming evidence feels like sound judgment. The person experiencing it has no phenomenological access to the asymmetry they are applying.
Case Studies: Confirmation Bias in Action
Case 1: The Iraq WMD Intelligence Failure (2003)
After the September 11 attacks, senior officials in the Bush administration held a strong prior belief that Saddam Hussein possessed weapons of mass destruction and was concealing them from inspectors. When the intelligence community was tasked with assessing this claim, the institutional structure amplified confirmation bias at every level.
Analysts who produced assessments consistent with the prior belief were rewarded with attention and access. The Curveball source — an Iraqi defector who claimed knowledge of mobile biological weapons laboratories — was assessed positively despite warnings from German intelligence that he was "not a stable, psychologically satisfactory person" and that his claims had not been independently verified. The Defense Intelligence Agency's doubts about chemical weapons were not featured in presentations to senior leadership. The aluminum tubes that Powell cited as evidence of a centrifuge program had been assessed by the Department of Energy's own weapons experts as likely intended for conventional rocket artillery — but the CIA's interpretation supporting the WMD hypothesis was the one that made it into the National Intelligence Estimate.
The disconfirming evidence was available. UN inspectors on the ground in Iraq had found nothing. The DIA memo existed. The DOE dissent was in the footnotes. Each piece of disconfirming evidence was individually explained away, while each piece of confirming evidence was accepted without similar scrutiny.
Case 2: Semmelweis and the Rejection of Handwashing (1840s-1865)
In 1847, Ignaz Semmelweis, a Hungarian physician working in Vienna's maternity wards, noticed that mortality rates from puerperal fever were dramatically higher in the ward staffed by medical students and physicians than in the ward staffed by midwives. The physicians went directly from performing autopsies to delivering babies without washing their hands. Semmelweis introduced mandatory handwashing with chlorinated lime solution; mortality in his ward dropped from around 10 percent to under 2 percent.
His colleagues did not celebrate. They dismissed the evidence. The dominant belief in Viennese and European medicine was that disease arose from miasmas — bad air — and from imbalances of humors. The handwashing hypothesis required accepting that physicians themselves were killing patients, an idea that was not only professionally threatening but incompatible with the existing explanatory framework. Semmelweis could not explain the mechanism (germ theory was still a decade away), and without a mechanism, the confirming evidence of his own ward's statistics could be attributed to other factors.
Critics pointed to the data that confirmed their priors — cases where death rates in midwife wards had fluctuated — and ignored the systematic pattern. Semmelweis died in a mental institution in 1865, still arguing his case. Germ theory, developed by Pasteur and Koch in the 1860s and 1870s, eventually vindicated him completely.
Case 3: The N-Ray Discovery That Was Never Made (1903-1904)
In 1903, the French physicist Rene Blondlot, a highly respected scientist at the University of Nancy, announced the discovery of a new form of radiation, which he called N-rays (after Nancy). He described their properties, demonstrated their detection, and published a series of papers in the prestigious Comptes Rendus de l'Academie des Sciences. Within months, dozens of French scientists had published papers confirming N-ray observations.
N-rays did not exist. The detections were subjective — involving small perceived changes in the brightness of a faintly glowing phosphorescent thread — and susceptible to expectation effects. Robert Wood, an American physicist from Johns Hopkins, visited Blondlot's laboratory in 1904 and, during a demonstration, surreptitiously removed the aluminum prism that was supposedly separating N-rays from a stream of radiation. Blondlot continued reporting the same observations. Wood published his account in Nature, and the N-ray literature collapsed within a year.
The case illustrates how a community of scientists, each hoping to make their own contribution to an exciting new phenomenon, could produce mutually reinforcing confirmatory reports without any deliberate fraud. Their observations were driven by expectation. Disconfirming observations — failure to reproduce results, anomalous readings — were attributed to improper technique or unsuitable subjects.
Case 4: The Theranos Clinical Testing Failure (2013-2016)
Elizabeth Holmes founded Theranos in 2003 on the claim that a single drop of blood from a fingertip prick could run hundreds of clinical tests accurately. By 2013, the company was valued at nearly ten billion dollars. Investors, board members, and media coverage collectively reinforced a story about revolutionary technology.
The disconfirming evidence was embedded in the technology itself. Theranos's proprietary Edison machines were running most tests on third-party analyzers, not their own technology, and were producing results significantly outside acceptable accuracy ranges. Several laboratory scientists flagged these problems; some resigned. None of this information reached investors or the board, partly because of Holmes's control of information flow, and partly because the investors and directors were not seeking it. They had already decided what Theranos was.
When journalist John Carreyrou began investigating for the Wall Street Journal in 2015, sources described a culture in which questioning the technology was treated as disloyalty — a perfect organizational expression of group-level confirmation bias.
Confirmation Bias Across Domains
Medical Diagnosis
Premature closure — settling on a diagnosis before all relevant evidence has been gathered — is one of the most common sources of diagnostic error documented in the medical literature. Once a physician has formed an initial hypothesis, subsequent observations tend to be filtered through that hypothesis. A study by Croskerry (2002) estimated that cognitive error, with premature closure as the most frequent type, accounted for 74 percent of malpractice claims in emergency medicine.
Investment and Financial Markets
George Soros famously described his trading approach as built on the assumption that his existing thesis was wrong — an explicit attempt to counteract the natural tendency to seek confirming evidence. Research on individual investor behavior consistently finds that people hold losing positions longer than winning ones, partly because selling a loss requires acknowledging the original thesis was incorrect. Investors seek financial news that supports their existing positions and avoid contrary views.
Organizational Management
Research on executive decision-making by Finkelstein, Whitehead, and Campbell (2008) documented a consistent pattern: senior executives surrounded themselves with advisors whose job, in practice, was to validate rather than challenge strategic decisions. The Groupthink phenomenon — described by Irving Janis in his 1972 analysis of the Bay of Pigs decision — is partly confirmation bias operating at the collective level: groups develop shared hypotheses that members reinforce rather than challenge.
Scientific Research and Publication Bias
The structural incentives of academic publishing create publication bias: studies with positive results are far more likely to be submitted and accepted than studies with null results. The replication crisis in psychology, which attracted serious attention around 2011, revealed that many classic findings failed to reproduce when investigators without a stake in the outcome ran the experiments again.
Social Media and Filter Bubbles
The algorithmic architecture of social media platforms optimizes for engagement, which correlates with content that confirms users' existing beliefs. Eli Pariser described this as the filter bubble in 2011: the personalized information environment that surrounds a user based on past clicks and preferences. The mechanism is structural confirmation bias — the information infrastructure itself performs the selecting.
The Intellectual Lineage
The phenomenon was named before it was studied. In 1620, Francis Bacon wrote in Novum Organum: "The human understanding when it has once adopted an opinion draws all things else to support and agree with it. And though there be a greater number and weight of instances to be found on the other side, yet these it either neglects and despises, or else by some distinction sets aside and rejects." Bacon identified confirmation bias as the central obstacle to the empirical method he was advocating — and his remedy was systematic: force the mind to attend equally to negative instances.
Wason's 1960 selection task gave confirmation bias its first experimental operationalization. Lord, Ross, and Lepper's 1979 study extended the finding to real-world belief domains and showed that balanced evidence could increase rather than decrease polarization when confirmation bias was operating.
Raymond Nickerson's 1998 comprehensive review in the Review of General Psychology — "Confirmation Bias: A Ubiquitous Phenomenon in Many Guises" — synthesized three decades of research and concluded that confirmation bias appeared across virtually every domain studied. Daniel Kahneman's Thinking, Fast and Slow (2011) integrated confirmation bias into his System 1 / System 2 framework: System 1 generates hypotheses and seeks confirming evidence efficiently; System 2 is capable of seeking disconfirming evidence, but is frequently enlisted instead to rationalize conclusions that System 1 has already reached.
The Empirical Record
The selection task has been administered to tens of thousands of participants across dozens of countries, with performance on the abstract version consistently near 10 percent. Wason (1968) followed up showing that performance did not improve significantly when participants were explicitly told some cards would disprove the rule — the problem is a genuine tendency to test hypotheses confirmationally, not just ignorance of the logical structure.
Lord, Ross, and Lepper (1979) showed that after reading the same mixed evidence, both pro- and anti-capital punishment groups became more extreme in their views — attitude polarization from balanced evidence exposure. Darley and Gross (1983) demonstrated that prior expectations about a child's social class led participants to interpret the same ambiguous academic performance differently. Mynatt, Doherty, and Tweney (1977) found that even when participants were given explicit instructions to test hypotheses systematically in a simulated scientific discovery task, they overwhelmingly generated confirming rather than disconfirming predictions.
Nickerson's 1998 review covered over 2,000 studies and found confirmation bias present across legal reasoning, medical diagnosis, scientific hypothesis testing, personnel selection, eyewitness testimony, social impression formation, and political judgment.
The Limits of Debiasing
Some degree of confirmatory reasoning is not only normal but adaptive. Bayesian reasoning — the statistically optimal approach to updating beliefs — requires using prior probabilities. The problem is not that humans use priors; the problem is that they weight them asymmetrically against disconfirming evidence.
The most effective single debiasing technique identified in the literature is the consider-the-opposite strategy: explicitly instructing people to generate reasons why their current hypothesis might be wrong. Lord, Lepper, and Preston (1984) showed this instruction significantly reduced attitude polarization in a follow-up to the 1979 capital punishment study. The effect was real but modest — and it required explicit instruction; it did not occur spontaneously.
Red teams — groups specifically tasked with attacking a plan or hypothesis — represent an organizational implementation of this strategy. Evidence on their effectiveness is mixed; when the red team's conclusions are systematically ignored or explained away — which is what confirmation bias predicts — the institutional structure has not addressed the cognitive problem.
The deepest problem with debiasing is that the bias does not feel like bias. It feels like reasoning. The scrutiny applied to disconfirming evidence feels like rigor. Overcoming confirmation bias requires not just knowing it exists, but building structural environments — checklists, adversarial roles, blind review processes, pre-specified stopping rules — that make the asymmetry visible and costly to maintain. Good intentions, on their own, are not enough.
Colin Powell later called his UN presentation "a blot" on his record. He was not a dishonest man. He was a man who had encountered a body of evidence that had been pre-filtered by a system organized to produce the conclusion it had already reached. No one at any stage had a method in place to ask, systematically and with genuine curiosity, which cards needed to be turned over to prove the hypothesis wrong.
References
Bacon, F. (1620). Novum Organum. London: John Bill.
Wason, P. C. (1960). On the failure to eliminate hypotheses in a conceptual task. Quarterly Journal of Experimental Psychology, 12(3), 129-140.
Wason, P. C. (1968). Reasoning about a rule. Quarterly Journal of Experimental Psychology, 20(3), 273-281.
Lord, C. G., Ross, L., & Lepper, M. R. (1979). Biased assimilation and attitude polarization: The effects of prior theories on subsequently considered evidence. Journal of Personality and Social Psychology, 37(11), 2098-2109.
Mynatt, C. R., Doherty, M. E., & Tweney, R. D. (1977). Confirmation bias in a simulated research environment. Quarterly Journal of Experimental Psychology, 29(1), 85-95.
Darley, J. M., & Gross, P. H. (1983). A hypothesis-confirming bias in labeling effects. Journal of Personality and Social Psychology, 44(1), 20-33.
Lord, C. G., Lepper, M. R., & Preston, E. (1984). Considering the opposite: A corrective strategy for social judgment. Journal of Personality and Social Psychology, 47(6), 1231-1243.
Nickerson, R. S. (1998). Confirmation bias: A ubiquitous phenomenon in many guises. Review of General Psychology, 2(2), 175-220.
Westen, D., Blagov, P. S., Harenski, K., Kilts, C., & Hamann, S. (2006). Neural bases of motivated reasoning. Journal of Cognitive Neuroscience, 18(11), 1947-1958.
Kahneman, D. (2011). Thinking, Fast and Slow. New York: Farrar, Straus and Giroux.
Croskerry, P. (2002). Achieving quality in clinical decision making: Cognitive strategies and detection of bias. Academic Emergency Medicine, 9(11), 1184-1204.
Wood, R. W. (1904). The n-rays. Nature, 70(1822), 530-531.
Frequently Asked Questions
What is confirmation bias?
Confirmation bias is the tendency to search for, interpret, favor, and recall information in a way that confirms or supports prior beliefs or values. Coined by Peter Wason in the 1960s, it operates at every stage of information handling: acquisition, interpretation, encoding, and retrieval. It does not require emotional investment — it operates in purely cold cognition, influencing even careful reasoners who believe they are being objective.
What is the Wason selection task?
The Wason selection task (1960) is the founding experiment in confirmation bias research. Participants see four cards (E, K, 4, 7) and must identify which 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 correct answer is E and 7 — only 10 percent of participants find it. Most choose E and 4, which can only confirm the rule, never disprove it. The task demonstrates that people naturally test hypotheses by seeking confirming rather than disconfirming evidence.
What is attitude polarization?
Attitude polarization, demonstrated by Lord, Ross, and Lepper (1979), is the phenomenon whereby exposure to balanced evidence makes people more extreme in their existing views, not less. In their capital punishment study, both pro- and anti-capital punishment groups read the same mixed evidence and rated the study supporting their view as more rigorous. After reading identical information, both groups became more extreme. This means confirmation bias can make balanced information counterproductive.
How does confirmation bias differ from motivated reasoning?
Confirmation bias is the asymmetric search for and interpretation of evidence that supports existing beliefs — it operates even without emotional stakes. Motivated reasoning is reasoning guided by a desired conclusion, driven by emotional investment in the outcome. Confirmation bias can operate in cold cognition; motivated reasoning requires that conclusions are emotionally or personally significant. Both often co-occur, and motivated reasoning amplifies the distortions of confirmation bias.
What is the most effective way to counter confirmation bias?
The most effective single technique identified in the literature is the consider-the-opposite strategy: explicitly generating reasons why your current hypothesis might be wrong. Lord, Lepper, and Preston (1984) showed this instruction significantly reduced attitude polarization. Structurally, red teams, adversarial collaboration, pre-registered hypotheses, and blind review processes are the most robust institutional solutions. Good intentions alone are insufficient — the bias does not feel like bias; it feels like rigor.
Is all confirmatory reasoning a bias?
No. Bayesian reasoning — the statistically optimal approach — requires using prior probabilities. The problem is not that humans use priors; it is that they weight them asymmetrically: updating strongly on confirming evidence and weakly on disconfirming evidence. A true Bayesian asks how much more likely is this evidence if the belief is correct than if it is incorrect. Confirmation bias short-circuits this calculation by treating consistent evidence as highly diagnostic and inconsistent evidence as suspect.
How did confirmation bias contribute to the Iraq WMD failure?
After 9/11, the Bush administration held a strong prior belief that Iraq had WMDs. The intelligence process amplified confirmation bias at every level: analysts rewarded for consistent assessments, the Curveball source accepted despite German intelligence warnings, the DIA memo doubting chemical weapons not featured in presentations, DOE dissent about aluminum tubes buried in footnotes. UN inspectors on the ground had found nothing. Each piece of disconfirming evidence was individually explained away; confirming evidence was accepted without similar scrutiny.