Confirmation bias describes systematic asymmetries in how people test ideas, choose evidence, interpret findings, and sometimes remember information after forming a belief or hypothesis. A person may ask questions that make the favored hypothesis seem likely, accept supportive evidence with relatively little scrutiny, and demand stronger proof from evidence that challenges it. The central scientific question is whether the same information would be processed differently if it supported a different initial belief. The phenomenon is studied through observable judgments and experimentally varied evidence; it is not a psychiatric diagnosis, a universal defect of human reasoning, or a claim that every search for confirmation is mistaken. Nickerson (1998) provides the foundational review.
What Confirmation Bias Means in Psychological Research
The term applies to a family of related patterns rather than to a single mental process. In hypothesis testing, it can mean selecting cases expected to satisfy a rule. In evidence evaluation, it can mean assigning greater credibility to congenial results than to equally strong inconvenient results. In information choice, it can mean preferring materials aligned with one's existing position. In confidence updating, it can mean using agreement more readily than disagreement to revise certainty. These patterns must be operationalized separately, and evidence for one does not establish that all the others occurred. Nickerson (1998); Kappes et al. (2020).
A working hypothesis is indispensable in science, diagnosis, and ordinary planning. The possible bias arises when the hypothesis changes standards of evidence or the value of additional tests in a way the task does not justify. A supportive observation can genuinely raise a hypothesis's probability; a contradictory observation can genuinely be unreliable. A rigorous assessment therefore specifies alternative hypotheses, the reliability of observations, their base rates, and the intended inferential benchmark before classifying a decision as biased.
Confirmation bias, selective exposure, and motivated reasoning
Selective exposure names a specific behavior: choosing what information to encounter. A person might select an article that endorses a preferred conclusion. Confirmation bias is the broader pattern that can also occur after information has been encountered, during credibility assessment, belief updating, or recall. The major meta-analysis of selective exposure found a moderate aggregate preference for congenial information, with important moderators. That finding should not be treated as a direct estimate of every form of confirmation bias. Hart et al. (2009).
Motivated reasoning asks how an accuracy goal or a desired conclusion affects thinking. A conclusion can be favored because it protects identity, status, emotion, or a practical interest. Such motivation is one route to selective evidence handling, but similar outward behavior can arise from an efficient testing strategy, prior knowledge, limited attention, or uncertainty about source quality. The causal explanation requires its own evidence. Kunda (1990); Klayman and Ha (1987).
Belief perseverance concerns a conclusion remaining influential after its original justification has been discredited. Hindsight bias concerns judgments of prior predictability after an outcome becomes known. An illusory truth effect concerns the influence of repetition on perceived accuracy. Those processes may interact with confirmation bias, but the experimental tests and defining outcomes differ. Likewise, a logical fallacy is a defect in an argument's form or inference, whereas confirmation bias is a psychological tendency in selecting or processing evidence. An incorrect statistical calculation is not evidence of a confirmation bias unless an asymmetric evidential process is demonstrated.
What counts as a valid benchmark?
For two hypotheses H1 and H2, a datum E favors H1 relative to H2 when the likelihood ratio P(E | H1) / P(E | H2) exceeds 1. A test yielding the same likely result under both hypotheses carries little power to distinguish them even if it produces an example compatible with H1. Bayesian updating combines that likelihood ratio with prior odds; it does not equate rationality with never looking for positive cases. If a task concerns deductive validity, the benchmark may instead be whether one counterexample logically falsifies a universal rule. The benchmark must fit the question being asked.
In practical life, individuals often face asymmetric costs, imperfect evidence, and constraints on time. A rational information policy may legitimately emphasize the most important risks or the most diagnostically useful cases. Researchers can claim confirmation bias when evidence treatment departs from a defensible, explicitly specified standard or when controlled contrasts reveal unjustified asymmetries. Without such a comparison, labeling disagreement as bias adds little scientific information.
Historical Foundations: From Rule Discovery to Belief Polarization
Wason's rule-discovery task
In a classic experiment, participants were shown the sequence 2–4–6 and asked to discover the rule generating it by proposing new number triples and receiving feedback. A natural conjecture was that the numbers increased by two. Repeated tests such as 8–10–12 could receive positive feedback but fail to distinguish that conjecture from a broader rule, such as any three ascending numbers. Some participants searched too narrowly and announced a rule prematurely. The original study included 29 participants and reported substantial difficulty in reaching the rule without initially incorrect proposals. The exact experimental setup matters: Wason did not measure the prevalence of confirmation bias in the general population. Wason (1960).
An informative test would place plausible rules in conflict. For example, a descending or unevenly spaced sequence can discriminate between the narrow “increase by two” rule and the broad “ascending order” rule. A researcher has to examine what participants believed the candidate rules were, which responses they anticipated, and how much information each test could provide. Merely counting positive examples cannot show whether the tests were inferior without reconstructing the hypothesis space.
Klayman and Ha's positive-test account
Klayman and Ha reexamined the assumption that testing an expected positive instance necessarily reflects defective reasoning. They proposed a positive-test strategy: ask about cases expected to possess the target property. Such testing may be highly informative when positive cases are uncommon and alternatives predict contrasting outcomes. Its efficiency changes with how common the target property is and how hypotheses overlap. The research shifted attention from the emotional appeal of confirmation to the diagnostic value of the chosen test. Klayman and Ha (1987).
Consider a product defect believed to occur only after a specific component overheats. Testing overheated units may quickly reveal whether those units are distinctive when overheating is uncommon. Conversely, observing many faulty units under that condition says little if nearly every unit also overheats. Information depends on the probability of each outcome under all serious competing accounts. The same positive-test strategy can be adaptive in one environment and misleading in another.
Biased assimilation of conflicting experiments
Lord, Ross, and Lepper presented students with purported research findings about the deterrent effect of capital punishment. Participants who entered with opposed views tended to evaluate evidence consistent with their starting position more favorably, and their attitudes polarized in the experimental setting. The original study involved 48 undergraduates and manipulated the presentation of putative evidence; it should not be read as proof that all encounters with disagreement cause polarization. It remains a canonical demonstration of how prior theories can organize criticism and evidential standards. Lord, Ross, and Lepper (1979).
This distinction matters because scientific criticism is itself desirable. A participant who notices a genuine methodological weakness in a study that happens to disagree with an existing view may be reasoning well. An experimentally detectable bias requires a double standard across comparably strong arguments, an interaction with prior attitude, or a credible measure of selective acceptance. Evaluation of research quality should never be replaced by a reflexive demand to treat all claims as equally plausible.
Four Locations Where Confirmation Bias Can Operate
1. Formulating questions and sampling cases
A hypothesis can determine which people, events, or measurements are examined. A manager convinced that a new workflow helps productivity may interview high-performing teams while overlooking teams that struggled. This produces selection effects before a formal judgment is made. Laboratory tasks can capture the process by tracking which questions participants ask or which candidate cases they choose under experimentally varied rules.
Sampling can be distorted even when each observed report is truthful. If evidence comes only from contexts where a prediction was expected to hold, the investigator loses the comparison needed to estimate false positives or counterfactual performance. Better inquiry requires cases selected for their ability to separate rival hypotheses, or a sampling design specified before the outcome is observed.
2. Choosing congenial sources
People can protect existing beliefs by allocating attention to agreeable news, advisers, or explanations. This component is measurable as source selection, information requests, click choices, or time spent with opposing views. The Hart et al. (2009) meta-analysis synthesized 91 studies from 67 reports, involving just under 8,000 participants, and found a mean congeniality preference of d = 0.36. That statistic summarizes a particular literature and its research conditions; it is neither the probability that any person exhibits confirmation bias nor an estimate of the size of every bias.
The review found that motivational and situational conditions moderated the preference. The tendency was weaker in some circumstances where accuracy motives were salient, attitude certainty was high, the issue was less identity-relevant, or the available information was poor in quality. These moderator patterns resist the simplified claim that people always avoid disagreement. Some seek counterattitudinal information precisely to become better informed or prepare a rebuttal.
3. Evaluating and weighting evidence
Two equally informative observations can be assigned different credibility because of how they relate to an initial belief. A supportive study's limitations may be overlooked, while an inconvenient study's sampling or design is examined intensively. A controlled design can hold evidence quality constant, vary prior commitments, and ask participants to rate strength, choose a follow-up test, or revise probabilities. Lord, Ross, and Lepper (1979); Taber and Lodge (2006).
When evaluating disagreement, standards should be symmetrical: methodological requirements appropriate for one side should also apply to the other. Symmetry does not require identical posterior beliefs when parties possess different prior information. It requires transparency about how evidence and its reliability are being evaluated.
4. Updating confidence and reconstructing memory
Confirmation bias may remain after an evidentiary encounter if the strength of agreement changes confidence more than the strength of disagreement does. Kappes and colleagues found, in an experimental judgment paradigm, that participants made less use of the strength of others' discrepant opinions when updating confidence, along with a neural signal associated with that asymmetry. These findings concern specific tasks and neural measurements, and they do not identify a single universal brain location for confirmation bias. Kappes et al. (2020).
Recall offers an additional possible route, although memory accuracy should be studied under its own measurement design. Questions such as whether a person later remembers supportive facts more frequently, or reconstructs the meaning of mixed information toward an existing belief, require controls for prior exposure, emotional salience, repetitions, and baseline memorability. Observing selective recall alone cannot establish whether an earlier selection or encoding effect produced it.
Competing Explanations and Cognitive Mechanisms
Predictive processing and prior knowledge
A belief provides expectations that guide attention and interpretation. Under uncertainty, using prior information may improve predictions. The problem arises when the influence of that prior exceeds what the evidence warrants, or when new evidence is selected to keep a favored model insulated from informative challenge. Bayesian models offer one explicit reference for belief updating, but observed deviations do not by themselves tell researchers which psychological process generated them.
A new observation can be evidence for a hypothesis and still be less diagnostic than a different observation. The relevant question is how the observation differentiates the live alternatives. The prior probability of a condition, the sensitivity and specificity of a test, the independence of sources, and the likelihood of the reported outcome under each account all affect the correct update.
Attention and limited cognitive resources
Humans cannot examine every source or compute every competing likelihood. Time pressure, information overload, task complexity, and expertise influence which cues are prioritized. Relying on one working model can reduce search costs. Whether that shortcut becomes a systematic bias depends on task structure, incentives, and evidence quality. Limited processing capacity offers an explanation for some patterns but is not a sufficient explanation for all motivated asymmetries.
Goals, identity, and motivated skepticism
Kunda's account explains how a preferred conclusion can guide the retrieval and construction of plausible justifications, while the need for acceptable reasoning still constrains what one can believe. Taber and Lodge found asymmetrical reactions to congruent and incongruent political arguments under controlled conditions. The effects implicate motivated appraisal in particular contexts, especially when attitudes are consequential to identity or values. They do not establish that every political disagreement is psychologically irrational. Kunda (1990); Taber and Lodge (2006).
Emotional significance and social rewards
A piece of information can threaten an important commitment, professional identity, or social relationship. Seeking reassuring evidence may regulate anxiety or preserve relationships, whereas seeking corrective evidence can have immediate emotional and social costs. These influences are possible mechanisms to test, not automatic explanations for a person changing beliefs slowly. Observational data alone rarely separate accuracy concerns from identity protection, shared norms, or differential trust in information sources.
The limits of simple dual-process explanations
It is tempting to describe confirmation bias as intuitive thinking overriding analytic thinking. Such a formulation can be a useful prompt for experiments, but it does not identify an established one-to-one mapping between a bias and a single cognitive system. Careful, lengthy reasoning can be deployed in defense of a desired conclusion; quick recognition of a strong counterexample can be accurate. Investigators need measures of response processes, time, mental effort, and task performance before attributing the pattern to intuitive or reflective processing.
How Researchers Measure Confirmation Bias
Hypothesis-testing tasks
A participant receives a candidate rule or explanation and chooses tests. Researchers code which tests could falsify competing interpretations, how frequently diagnostic questions are selected, and whether participants revise their rule after unexpected feedback. The correct comparison is typically the information value of the available options under the plausible hypothesis set. Wason's paradigm reveals a robust methodological tradition but also the danger of overinterpreting a single rule-discovery game. Wason (1960); Klayman and Ha (1987).
Balanced-evidence appraisal tasks
Researchers experimentally give participants evidence favoring and opposing a claim and measure ratings of credibility, methodological quality, and belief change. An adequate design verifies comparable evidence strength, reports which beliefs were measured before the manipulation, and distinguishes measured polarization from changes driven by genuinely unequal information. Political, health, and social-judgment stimuli can elicit different levels of personal involvement.
Selective information-exposure tasks
Participants are offered a choice between supportive and challenging materials while price, source credibility, headline attractiveness, or length are controlled. The outcomes include expressed preferences, actual reading, click-through, and attention duration. A stated desire to read agreeable information is not equivalent to real-world exposure or a final mistaken judgment. Hart et al. (2009).
Confidence-updating paradigms
Participants first make judgments and report their confidence, then receive another opinion or numerical estimate. The experiment separately varies whether new input agrees and how strong it is. The primary dependent variable is an update to confidence or the final probability estimate. Interpreting asymmetry requires a model of the information conveyed by the advice, including source reliability and the correlation between sources. Kappes et al. (2020).
Field studies and ecological measurement
Real institutions generate audit trails: competing hypotheses written before evidence collection, alternative explanations considered, patterns of sampling and requests for follow-up information, and corrections after outcomes become known. These measures connect cognitive theory with actual decisions but introduce confounding influences from law, organizational incentives, technical limitations, and unequal access to evidence. Stronger field designs preregister outcomes or randomize prompts, review protocols, or access to information wherever ethically possible.
Interpreting Effect Sizes, Individual Differences, and Replication
A standardized mean difference is a contrast between research conditions scaled by their variability. The selective-exposure synthesis reported d = 0.36 as an aggregate across its studies, which should not be converted into a universal risk or individual diagnosis. Many studies evaluate behavioral preferences rather than whether a participant ultimately reaches an objectively false belief. Hart et al. (2009).
The classic biased-assimilation experiment established that prior views can interact with evaluation of mixed evidence, but its sample was small and its materials and historical context were specific. Replications and newer studies must be evaluated individually for stimulus selection, measurement invariance, preregistration, power, effect-size uncertainty, and cultural context. Results from a highly polarized issue cannot simply be generalized to an unfamiliar laboratory rule, clinical consultation, or financial forecast.
There is no validated single scale that directly measures every version of confirmation bias in all situations. A person's performance may vary with expertise, conviction, incentives, cognitive reflection, information literacy, task wording, and the reliability of the evidence. Personality or demographic correlations should be described only with specific supporting datasets. General claims that one generation, occupation, nationality, or political group has uniquely strong confirmation bias are scientifically unwarranted without comparable tests.
The direction of a bias is also task-dependent. Some participants prefer congenial evidence, others actively inspect opposition; disagreement can trigger updating, strategic rebuttal, or stronger prior conviction. A negative result in one paradigm does not refute the entire family of constructs, and a positive result in one sample does not establish a universal trait. Research synthesis should say which behavior was measured and the uncertainty around that estimate.
Illustrative Examples and What Would Make Each Testable
Medical and scientific inquiry
A clinician who favors one diagnosis may ask follow-up questions predominantly about confirming symptoms. The appropriate comparator is an assessment that also weighs alternative diagnoses, symptom base rates, test sensitivity and specificity, and potentially dangerous missed conditions. A single clinical disagreement is insufficient evidence of a cognitive bias. A study might randomize preliminary diagnostic labels and test whether identical findings receive different weight. For actual medical decisions, evidence-based protocols and professional judgment remain necessary.
A research team may favor a theory by changing exclusion rules after seeing results, searching only confirming analyses, or setting an unusually strict evidentiary threshold for contradictory findings. Preregistration, blinded analysis, transparent data, and independent replication help identify such asymmetries. Selective reporting is a research-practice problem with statistical and institutional components; its existence should not be equated automatically with a particular scientist's psychology.
Hiring, leadership, and organizational decisions
When an interviewer forms an early favorable impression, later questions may become easier or more affirming than those given to an otherwise equivalent candidate. An audit could randomize prior summary information and compare the questions asked, scoring criteria, and ratings of identically qualified candidates. Structured interviews, consistent behavioral criteria, and independent judgments can make the process more transparent, though their specific effects depend on design and implementation.
In organizational planning, a project sponsor may ask only whether milestones can be met instead of asking which observations would show the plan has failed. A concrete decision protocol specifies failure indicators, outside-view comparisons, and independent review before resources are committed. The procedure supports better falsifiability even when the final decision remains unchanged.
Politics, media, and public disagreement
A person can judge two equally well-designed studies differently when the conclusions point in opposite ideological directions. Taber and Lodge's experiments illustrate motivated skepticism in policy contexts, while Hart and colleagues' meta-analysis concerns the separate problem of which materials are chosen. Neither makes it legitimate to dismiss sincere disagreement as merely a bias. The substantive quality of evidence remains decisive. Taber and Lodge (2006); Hart et al. (2009).
Simply forcing exposure to disagreeable content is not a guaranteed corrective. In a field experiment involving Twitter users, Bail and colleagues found an increase in conservative positions among the Republican participants assigned to follow an opposing-view bot; the corresponding estimated effect among Democrats was not statistically significant. It was one platform, period, population, and intervention, and it did not establish that all balanced information exposure backfires. Bail et al. (2018).
Forensic inquiry and high-stakes evidence
An investigator's early suspect theory can influence which leads are followed and how ambiguous evidence is interpreted. Scientific practice requires comparing rival explanations and controlling exposure to task-irrelevant contextual cues where feasible. The evidence and procedural duties are specialized, so an application in justice belongs with the separate forensic investigation article, rather than being treated as proof that every case result reflects generic confirmation bias.
Everyday recommendations and AI systems
An algorithmic recommender can repeatedly surface content similar to past clicks, increasing the chance that a person encounters agreeable information. To establish a causal effect on belief, a study would need to distinguish personal selection, platform ranking, prior attitudes, message credibility, and downstream updating. Algorithmic amplification and human confirmation bias are analytically distinct even when they reinforce one another. Human–AI trust and automation effects have additional mechanisms, illustrated by the specialized AI as Authority article.
Conversational AI can produce fluent arguments that appear to support a user's initial theory, particularly when the request itself supplies a favored conclusion. A useful decision-support interaction asks for independent alternatives, disconfirming evidence, and source reliability before generating a recommendation. Assertions about the amount of harm or improvement from any particular AI product require product-specific controlled studies. A behavioral hypothesis about users is separate from any philosophical claim about the AI system's subjective status.
Improving Evidence Evaluation: What Has Experimental Support?
Consider the opposite
One of the best-known interventions asks people to generate plausible reasons why an initially rejected explanation might be true. Lord, Lepper, and Preston examined this approach in two experiments spanning biased evidence assimilation and personality-hypothesis testing. It can redirect attention toward counterexamples and expose asymmetric standards within the studied tasks. It is not an independently established cure for every manifestation of confirmation bias. Lord, Lepper, and Preston (1984).
A practical version is to write the favored hypothesis, a competing hypothesis, and two observations that would favor the competitor. Then decide in advance what evidence would justify changing confidence. This turns an exhortation to be objective into a decision rule that can be audited later.
Precommit to evidence standards
Before reading results, define what counts as a high-quality source, how alternatives will be compared, and what magnitude of evidence triggers an update. Apply those standards symmetrically to congenial and incongruent findings. The method follows from the logic of diagnostic testing and research transparency; the effectiveness of a particular checklist depends on the field, implementation, and direct assessment of outcomes. Klayman and Ha (1987).
Seek diagnostic information, not merely opposition
The most useful question is often: Which outcome would be different if the alternative explanation were true? Consider a lab test that is positive among 80% of people with a condition but also among 75% without it. A positive test weakly discriminates the alternatives despite seeming to confirm the suspected condition. The likelihood ratio is 0.80 / 0.75, about 1.07; base rates remain essential. This numerical illustration is a constructed example, not a published estimate for a real test.
A test that a favored theory predicts correctly can be valuable. A test that competing theories all predict correctly adds far less discriminating information. The goal is diagnostic power, a principle distinct from reflexively searching for negative anecdotes.
Encourage independent and structured review
Before a group discussion, members can record their judgments and the evidence supporting them. Structured review can require at least one rival explanation, identification of uncertain assumptions, and a record of why a claim was accepted or rejected. Blind review of irrelevant context can prevent some pathways for expectation effects. These are defensible practices that require domain-specific effectiveness studies; they should not be advertised as universally proven clinical or organizational treatments.
Accuracy-oriented prompts and misinformation
A synthesis of 20 experiments involving 26,863 participants found that asking social-media users to attend to accuracy could improve the quality of subsequent information-sharing judgments in the studied conditions. The outcome was sharing discernment, not a direct laboratory assay of every component of confirmation bias; it is an adjacent intervention rather than proof of broad debiasing. Pennycook and Rand (2022).
What the latest educational evidence actually shows
A 2025 systematic review assembled 54 randomized trials with 10,941 participants and reported a meta-analytic estimate of g = 0.26 (95% CI 0.14 to 0.39) for the subset of bias-reduction outcomes across 41 studies. Its authors identified unclear or high risk of bias across included trials, possible publication bias, and uncertainty about transfer beyond educational tasks. The aggregated effect spans different biases and interventions; it must not be described as the confirmed effect size of a specific confirmation-bias exercise. Swaryandini et al. (2025).
A 2026 randomized study of a brief educational video reported differences between teachers and learners and between near and broader outcome tasks. Its relevance is to the conditions under which short interventions may generalize, rather than to a settled universal correction. Interventions should be evaluated by durable behavior and real decisions whenever possible. Baruh, Cemalcilar, and Stoycheff (2026).
A Decision Procedure for Evidence Under Uncertainty
First, state the decision and the specific claim being evaluated. “Our intervention works” is too broad; “the intervention improves the prespecified outcome compared with a credible control after four weeks” creates a testable target. Identify what constitutes relevant evidence and which rival explanation would generate a different prediction.
Second, list prior reasons to believe each account and the conditions under which a new observation could favor it. For quantitative problems, write approximate prior odds, expected test accuracy, and potential dependence between sources. For qualitative questions, specify the range of plausible interpretations and why some observations would be especially informative.
Third, collect evidence in a manner that does not systematically hide counterexamples. Seek high-quality contradictory information along with supporting data, and distinguish a genuinely diagnostic counterexample from an unrelated objection. Track sources and their independence; five reports all repeating one original claim are not five independent confirmations.
Fourth, evaluate reliability using common criteria before comparing conclusions. If a research design lacks a control group, this weakness matters whether its result is welcome or inconvenient. If a source has demonstrated domain expertise or a better measurement instrument, its greater weight may be legitimate, regardless of agreement.
Fifth, update confidence explicitly. Record the original estimate, the information received, and the revised estimate, along with the reason for the change. Decide what future evidence would reverse the assessment. A change in probability need not be a binary conversion from belief to disbelief; modest evidence often warrants a modest update.
Finally, revisit outcomes with a record of the information actually available at decision time. Correct errors in forecasts or evidence standards without treating every unfortunate outcome as proof of cognitive bias. Longitudinal calibration and external review can show whether a procedure improves judgments, but outcomes should be analyzed under the conditions in which the procedure was deployed.
Scientific Boundaries, Generalizability, and Open Questions
Confirmation-bias findings depend on the population, the task's hypothesis space, the credibility of sources, incentives, personal involvement, and how a researcher defines a correct answer. Findings from U.S. student samples or political controversy cannot by themselves establish the same magnitude in a multilingual clinical team, a forensic laboratory, or routine engineering decisions. Cross-cultural research should use comparable tasks and test whether translated stimuli carry equivalent social meaning before comparing effect sizes.
A major methodological challenge is distinguishing bias from rational use of priors. Another is separating cognitive mechanisms: selective search, biased evaluation, miscalibrated confidence, and post-event reconstruction can lead to similar final opinions by different routes. A third challenge is measuring consequential outcomes. A participant who selects one congenial abstract in a lab may make a different choice when accuracy is rewarded, reputation is at stake, or expert feedback is available.
Causal studies must also separate belief consistency from truth. A correct belief can be maintained for poor reasons; an incorrect belief can persist because available evidence genuinely remains ambiguous. Research on judgment should describe how standards, incentives, evidence quality, and uncertainty were defined. The scientifically useful conclusion is about specific processing asymmetries, not a blanket judgment of a person's intelligence or integrity.
Frequently Asked Questions
Is confirmation bias the same as being wrong?
No. It concerns how evidence is sought, interpreted, or weighted relative to an existing belief. A person may hold a correct conclusion and still use a poor testing strategy; someone may hold an incorrect conclusion after responsibly processing incomplete evidence. Identifying a bias requires a comparison of procedures or responses, not just the final answer.
Does confirmation bias always involve deliberate dishonesty?
No. It may emerge from ordinary hypothesis testing, attention, unequal trust, or motivated appraisal. A study that detects asymmetric information weighting does not by itself establish awareness or intent. Conscious deception is a separate behavior requiring different evidence.
What is the difference between confirmation bias and motivated reasoning?
Confirmation bias describes patterns of evidence handling relative to an existing belief. Motivated reasoning investigates whether desired conclusions or goals help produce those patterns. A person can use confirmatory tests with little emotional stake, and strong motivation can influence reasoning in ways other than selecting confirming evidence. Kunda (1990).
Do intelligent or scientifically trained people avoid it?
Expertise can improve the choice and interpretation of evidence within a familiar domain, while prior commitments and specialized expectations can still shape what is tested. No single study establishes that intelligence or professional training eliminates all forms of confirmation bias. Appropriate evaluation examines performance on specific tasks and independent review procedures.
Why is seeking evidence for a hypothesis sometimes rational?
If a positive case is rare and strongly distinguishes competing hypotheses, looking for it can be an efficient strategy. Positive testing becomes uninformative when competing explanations predict the same positive result. The crucial quantity is diagnostic information, not whether a test feels supportive. Klayman and Ha (1987).
Does showing opposing information automatically change a belief?
No. Effects depend on credibility, context, identity, and the form of exposure. An experiment that supplied opposing political content through social-media bots found differential responses by political group, but it should not be generalized into a universal backfire rule. Bail et al. (2018).
Can a checklist permanently remove confirmation bias?
Checklists, consideration of alternatives, and preregistered evaluation standards may improve particular decisions. Evidence about durable transfer across contexts is more limited. The 2025 educational meta-analysis reports a small aggregate improvement on studied outcomes and important methodological risks. Swaryandini et al. (2025).
Is confirmation bias a mental illness or a symptom of one?
Confirmation bias is a construct in cognitive and social psychology and is not itself a mental disorder. It can be investigated in both clinical and nonclinical settings. Diagnostic assessment must rely on clinically validated symptoms, duration, impairment, and appropriate professional standards, rather than a general observation about evidential preference.
How can I challenge a strong belief without abandoning evidence-based reasoning?
Specify the best competing explanation and what observation would distinguish it; judge the quality of confirming and disconfirming sources by the same criteria; and record how far the new evidence should shift confidence. Critically evaluating an alternative is compatible with retaining a well-supported belief. The goal is a justified conclusion reached through transparent inquiry.
Related Articles
The Cognitive Biases, Judgment and Decision-Making research chapter situates evidence evaluation within the broader science of uncertainty, heuristics, rationality, and choice.
Forensic Evidence and Cognitive Bias: How Context Can Contaminate Judgment examines how expectations and contextual information can alter specialized evidence assessments.
Cognitive Bias in Criminal Investigations: Tunnel Vision and Confirmation Bias explores investigative hypothesis testing and case-level narrowing.
Bias in Forensic Psychology: How Evaluators Can Reduce Error covers assessment practices and safeguards in forensic evaluation.
Judge Decision-Making: Psychology, Expertise, and Bias examines evidentiary judgment in legal decision-making.
AI as Authority: Trust, Expertise, Automation Bias, and Human Decision-Making addresses authority cues and confidence in human–AI interactions.
Belief Perseverance: Why Beliefs Survive Contradictory Evidence — Studies why beliefs can persist after contrary evidence.
Outcome Bias: Why Results Distort Evaluation of Decisions — Examines how known results alter evaluations of preceding decisions.
Selective Exposure: Why We Choose Agreeable Information — Examines information selection before belief evaluation.
References
Bail, C. A., Argyle, L. P., Brown, T. W., et al. (2018). Exposure to opposing views on social media can increase political polarization. Proceedings of the National Academy of Sciences, 115(37), 9216–9221. https://doi.org/10.1073/pnas.1804840115
Baruh, L., Cemalcilar, Z., and Stoycheff, A. (2026). Mitigating cognitive biases in higher education contexts through a brief video-based intervention: A randomized controlled trial. Computers and Education, 245, 105535. https://doi.org/10.1016/j.compedu.2025.105535
Hart, W., Albarracín, D., Eagly, A. H., Brechan, I., Lindberg, M. J., and Merrill, L. (2009). Feeling validated versus being correct: A meta-analysis of selective exposure to information. Psychological Bulletin, 135(4), 555–588. https://doi.org/10.1037/a0015701
Kappes, A., Harvey, A. H., Lohrenz, T., Montague, P. R., and Sharot, T. (2020). Confirmation bias in the utilization of others' opinion strength. Nature Neuroscience, 23, 130–137. https://doi.org/10.1038/s41593-019-0549-2
Klayman, J., and Ha, Y.-w. (1987). Confirmation, disconfirmation, and information in hypothesis testing. Psychological Review, 94(2), 211–228. https://doi.org/10.1037/0033-295X.94.2.211
Kunda, Z. (1990). The case for motivated reasoning. Psychological Bulletin, 108(3), 480–498. https://doi.org/10.1037/0033-2909.108.3.480
Lord, C. G., Lepper, M. R., and Preston, E. (1984). Considering the opposite: A corrective strategy for social judgment. Journal of Personality and Social Psychology, 47(6), 1231–1243. https://pubmed.ncbi.nlm.nih.gov/6527215/
Lord, C. G., Ross, L., and 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. https://doi.org/10.1037/0022-3514.37.11.2098
Nickerson, R. S. (1998). Confirmation bias: A ubiquitous phenomenon in many guises. Review of General Psychology, 2(2), 175–220. https://doi.org/10.1037/1089-2680.2.2.175
Pennycook, G., and Rand, D. G. (2022). Accuracy prompts are a replicable and generalizable approach for reducing the spread of misinformation. Nature Communications, 13, 2333. https://doi.org/10.1038/s41467-022-30073-5
Swaryandini, G., Graham, J., Griffith, S., et al. (2025). Systematic review and meta-analysis of educational approaches to reduce cognitive biases among students. Nature Human Behaviour, 9, 2510–2538. https://doi.org/10.1038/s41562-025-02253-y
Taber, C. S., and Lodge, M. (2006). Motivated skepticism in the evaluation of political beliefs. American Journal of Political Science, 50(3), 755–769. https://doi.org/10.1111/j.1540-5907.2006.00214.x
Wason, P. C. (1960). On the failure to eliminate hypotheses in a conceptual task. Quarterly Journal of Experimental Psychology, 12(3), 129–140. https://doi.org/10.1080/17470216008416717