What Is Belief Perseverance?
Belief perseverance is the persistence of a judgment, impression, or explanatory belief after the original information supporting it has been discredited. In its classic experimental form, researchers induce a belief with apparently meaningful evidence, reveal that this evidence was fabricated or unrelated to the conclusion, and then ask whether the induced belief continues to influence judgment. The central result is a residual difference between people initially led toward competing conclusions despite their subsequent exposure to equivalent discrediting information. The phenomenon concerns belief updating after evidential invalidation; it is studied as a feature of ordinary cognition and social judgment, not as a psychiatric disorder. The foundational demonstration is Ross, Lepper, and Hubbard (1975).
The question is deceptively precise: when a person discovers that the very observation used to justify a belief was manufactured, why might the conclusion still seem plausible? One experimentally supported answer is that people generate interpretations, causal stories, and expectations while first processing a claim. Those mental products can remain accessible even when their initiating information is withdrawn. Yet perseverance is neither inevitable nor invariably strong. Beliefs also change when evidence is compelling and understood, and the magnitude of any residual effect depends on the task, measures, explanatory activity, correction, and population. A direct modern challenge to blanket claims of belief immutability comes from Anglin (2019).
A working example
Imagine two groups assessing whether firefighters who seek danger perform better or worse than cautious firefighters. One group reads two vivid case descriptions seemingly supporting a positive association; the other reads similarly persuasive examples supporting a negative association. Later both groups learn that the examples were invented and do not reveal the real association. If their subsequent estimates remain separated by the original assignment, the experiment demonstrates persistence after evidential discrediting. It does not establish that all firefighters are assessed unfairly in everyday life or that either theory is universally believed. This design is central to Anderson, Lepper, and Ross (1980) and Anderson (1983).
The Scientific Definition and Its Boundaries
Belief perseverance refers to an outcome relative to a prior evidential manipulation. A belief held for many years, or an opinion that survives disagreement, is not automatically an instance of the experimentally defined effect. To demonstrate perseverance, researchers must show that the evidence originally favoring a judgment no longer warrants that judgment and that the initial evidence condition still predicts later beliefs. This requires at least an original exposure, a credible correction or discrediting phase, and a subsequent measurement of the belief or behavior.
Belief perseverance versus confirmation bias
Confirmation bias concerns asymmetrical search, testing, interpretation, and sometimes recall in relation to an existing hypothesis. Belief perseverance focuses on the fate of a belief after a particular evidential foundation has been removed. They can occur together: selective examination may support a favored interpretation, and that interpretation may persist even after its source is exposed as invalid. Yet the experimental operations differ. A person can assimilate evidence selectively without ever encountering a decisive correction, and someone can preserve an initial belief after fully acknowledging that its source was false.
Belief perseverance versus motivated reasoning
Motivated reasoning refers to the influence of accuracy goals or desired conclusions on cognitive operations. Social identity, emotional comfort, reputation, and group commitments may shape the response to contradictory information. Belief perseverance can also emerge in low-stakes unfamiliar tasks where participants have no durable identity investment, as the firefighter and false-performance paradigms illustrate. Identity defense is one possible contributor, not a necessary ingredient. The distinction prevents treating any correction-resistant belief as evidence of an individual's hidden motivation.
Belief perseverance versus the continued influence effect
The continued influence effect is usually studied with narratives containing a misleading detail followed by a correction: even after accepting the correction, readers may use the initial misinformation in later explanations. Classic belief-perseverance studies often create a broader impression or causal theory with fabricated feedback or case evidence. Both concern incomplete updating, but outcome measures differ: belief ratings, explanations, causal inferences, and narrative reasoning are not interchangeable. Johnson and Seifert (1994), Lewandowsky and colleagues (2012), and Ecker and colleagues (2022) describe the related correction literature.
Related but distinct concepts
Hindsight bias concerns the reconstruction of an earlier estimate after learning an outcome; illusory truth concerns the increase in perceived credibility associated with repetition; selective exposure concerns choices about which information to encounter. A logical fallacy is identifiable through the structure of an argument, whereas a cognitive bias is identified through systematically varying judgments under specified conditions. Persevering in a true conclusion after one flawed argument is exposed can be entirely reasonable if independent evidence still supports that conclusion. The normative issue is how much rational support has actually been removed.
How the Experimental Literature Developed
Ross, Lepper, and Hubbard: discredited performance feedback
In two influential 1975 experiments, participants received misleading feedback suggesting success or failure at an unfamiliar discrimination task. They subsequently learned that the apparent outcome had been predetermined rather than based on their ability. Actors, and in a second study observers who watched those actors, could nonetheless retain impressions influenced by the bogus outcome. The publication reports samples of 60 female high-school students in one experiment and 144 female undergraduates in another. These samples establish the specific laboratory observation; they are not representative prevalence estimates for the general population. Ross, Lepper, and Hubbard (1975).
The experiments also contrasted debriefing procedures. Telling participants that the performance information was invalid constituted an outcome-oriented correction. In the process-oriented condition, the debriefing additionally addressed how the original feedback could have shaped subsequent inferences. That more explicit intervention generally reduced or eliminated erroneous self-perceptions in this design. The result supports a mechanism-sensitive approach to correction, although its effectiveness cannot be assumed for political convictions, clinical beliefs, or long-established identities without separate tests.
Anderson, Lepper, and Ross: explanations that outlive evidence
The 1980 firefighter studies presented short case examples purportedly showing a positive or negative relationship between risk-taking and successful professional performance. Some participants were asked to explain the alleged association. After they learned that the cases were fictitious, beliefs often continued to reflect the initial relationship, and explanatory work was implicated in persistence. The experiments used a debriefing paradigm with 130 participants across two studies. They illuminate why an explanation produced from fabricated premises may later seem independently convincing. Anderson, Lepper, and Ross (1980).
This is an important distinction between the information that originally elicited a theory and the mental model subsequently used to defend it. Once a participant constructs a story about why a daring firefighter performs well, that story can recruit additional general assumptions about courage, speed, and leadership. Discrediting two anecdotes does not explicitly discredit every imagined intermediate step. The process can be studied without assuming dishonesty, pathology, or conscious refusal to accept evidence.
Concrete cases and causal elaboration
In a 1983 series, Anderson compared initial beliefs supported by concrete case histories with beliefs supported by abstract statistical summaries. Both sets were selected to elicit comparable initial belief strength. Perseverance appeared immediately and a week later, and concrete cases produced more persistence than abstract summaries. The reported interpretation emphasized the causal elaborations prompted by concrete material rather than simple survival of memory for the original data. These are task-specific results, not proof that quantitative information is always superior to narratives. Anderson (1983).
Beliefs about personal ability in classrooms
Lepper, Ross, and Lau used an educational setting in which apparent success or failure followed exposure to instructional materials of different quality. Even when participants recognized how instructional quality affected the result, inappropriate conclusions about their own aptitude sometimes persisted. Measures collected later suggested that these impressions could matter beyond the immediate debriefing moment. The work illustrates persistence of self-assessments, but the manipulation, age group, and educational context limit unqualified extrapolation to real diagnoses or long-term academic development. Lepper, Ross, and Lau (1986).
What Psychological Mechanisms Can Sustain a Discredited Belief?
Causal explanations become partly independent representations
The central explanatory account is that people transform observations into generalized causal models. The observation can later be withdrawn while the model remains familiar and plausible. A fabricated case about the relationship between risk-taking and competence may generate hypotheses about preparedness, teamwork, confidence, and speed. Those hypotheses, although prompted by the false examples, are stored or evaluated as separate possible explanations. Anderson, Lepper, and Ross (1980) and Anderson (1983) support this account within their experimental designs.
The mechanism is stronger when the reasoner actively generates explanations. Generation entails more elaboration, retrieval, and integration with existing knowledge than passive exposure does. It can also reduce the availability of opposing arguments at the moment of judgment. A later investigation of generated versus provided explanations found stronger persistence following self-generated explanations even when supplied explanations were judged better, with fewer opposing reasons produced afterward. This pattern concerns the experimental comparison, not a universal law that thinking deeply always increases error. Davies (1997).
Attribution after misleading feedback
A person may attribute an apparently successful test performance to stable ability even when subsequent information demonstrates that the feedback was arbitrary. Attributions are causal assessments that can affect self-image and expectations. Similarly, an observer can form an impression of another person's ability from the same fabricated feedback. If the feedback is invalidated but the causal attribution persists, later estimates remain contaminated. The key methodological advantage of the original debriefing experiments is that the source of the initial impressions was controlled by the researchers. Ross, Lepper, and Hubbard (1975).
Memory integration and event models
Correcting a false detail does not always update the entire event representation that depended on it. A reader who initially learned that flammable materials caused a fire may continue using that causal link after reading that the materials were not present. The correction has to compete with a coherent situation model and retrieval of an established explanation. Johnson and Seifert (1994) established a classic demonstration; Ecker and colleagues (2022) synthesize memory, explanatory coherence, cognitive, social, and affective contributors to misinformation correction.
A practical implication follows: successfully remembering that a source was discredited and successfully revising the inferences formed from it are separate outcomes. Research therefore needs both belief questions and reasoning questions. A respondent can accurately state that a claim has been retracted while continuing to reason as if it remained true. Conversely, someone might forget the precise correction yet form a more accurate judgment from other evidence. These outcomes require different measurements.
Familiarity, ease of processing, and source confusion
Familiar propositions can feel intuitively credible. Repeated statements sometimes receive higher truth ratings independently of their accuracy, a phenomenon extensively studied as the illusory truth effect. It can reinforce a belief that originated from unreliable material, but the causal role of repetition must be measured independently of explanatory belief perseverance. Fazio and colleagues (2015) show that repetition can influence truth judgments even when relevant knowledge is available. This supports a plausible interacting mechanism without reducing all perseverance to familiarity.
Identity and emotional investment
Beliefs may organize social affiliation, professional reputation, or personal meaning. Conflicting information can therefore affect a person's emotional and motivational goals, sometimes altering the attention devoted to evidence or the criteria for accepting it. The scope of this claim is limited: a person may oppose a correction because the correction is inaccurate, ambiguously communicated, or less trustworthy than independent evidence. Identity-based explanations should be supported by manipulation or measurement of identity relevance rather than inferred from the person's conclusion. Ecker and colleagues (2022) review the varied influences involved.
What Counts as a Rational Update?
Belief perseverance is usually assessed relative to the removal of causal evidence or to an experimentally equalized information state. Suppose an experiment assigns a participant to fabricated evidence E favoring hypothesis H, then fully and credibly discredits E. If no other evidence distinguishes the randomized conditions, the study's normative benchmark is that participants should no longer draw different conclusions merely because of the invalidated assignment. The comparison is between randomized groups after discrediting, rather than a claim that everyone's estimate should take one specific numerical value.
Bayesian updating and evidential relevance
Bayesian inference formalizes the normative question. Posterior odds for H are prior odds multiplied by a likelihood ratio associated with the observed evidence: P(H|E)/P(not H|E) = [P(H)/P(not H)] × [P(E|H)/P(E|not H)]. When the study establishes that purported evidence was fabricated independently of H, the experimenter-defined likelihood ratio of that manipulated information becomes one, and the fabricated evidence supplies no diagnostic reason to favor H over its alternative. The rational assessment must still incorporate legitimate prior information and any independently verified evidence. A correct conclusion is not rendered false merely because one earlier argument for it collapses.
This formalization has limits. Laboratory participants may doubt whether discrediting was complete, infer information about the experimenter, or treat their self-generated explanations as new hypothetical considerations. Researchers must assess the credibility and understanding of debriefing before describing residual differences as violations of an ideal evidence-updating rule. Bayesian models also depend on how hypotheses, prior beliefs, evidence reliability, and dependencies are specified. The experimental benchmark is strongest where the manipulated evidence is visibly and completely non-diagnostic.
Normative, descriptive, and ecological accounts
A normative account specifies how belief should change under defined evidence and utilities. A descriptive account explains observed updating using memory, explanation, attention, and motivation. An ecological account asks whether retaining a provisional theory could be adaptive when environments are noisy and evidence is incomplete. For example, a person should be cautious about abandoning a well-supported scientific conclusion because of one poor-quality contradiction. The laboratory effect is distinctive because discredited evidence was the actual basis for an experimentally induced difference. Retaining a belief can reflect either rational integration of independent evidence or continued dependence on an invalid source; distinguishing the two is the empirical task.
How Researchers Measure Belief Perseverance
Randomized evidence-discrediting designs
In the core paradigm, participants are randomized to opposing information conditions and asked to make a judgment. Investigators then administer a correction or complete debriefing, check whether participants recognize the invalidity of the initial information, and measure belief again. A particularly informative estimate is the post-discrediting difference between groups initially assigned to opposing fabricated findings, adjusted for baseline beliefs where available. Measures may use rating scales, confidence judgments, causal predictions, or choices. Pre-registration and transparent exclusion rules improve interpretability.
Separate belief persistence from disbelief of the correction
A participant who genuinely doubts the researcher or fails to understand the correction is different from one who accepts the invalidation but still finds the conclusion plausible. Comprehension checks and source-memory questions identify this distinction. Asking whether the original study was valid, and separately whether a hypothesized causal association is true, avoids collapsing factual knowledge about the source into the theory being evaluated. A credible correction can alter source beliefs more than general explanatory beliefs.
Distinguish immediate attitudes, delayed judgments, and action
Questionnaires administered seconds after debriefing capture immediate inference, whereas delayed tests measure retention, forgetting, and possible exposure to new information. Behavioral outcomes require their own validation. A change in a five-point belief scale does not directly quantify changes in voting, clinical adherence, purchasing, or real-world professional decisions. Where possible, researchers should report raw means, uncertainty intervals, and preregistered contrasts rather than treating statistical significance alone as practical importance.
Measuring alternative explanations
Researchers can code the number, plausibility, and variety of explanations participants generate in favor of each possible relationship. A causal-explanation manipulation offers evidence about mechanism when it precedes later judgment and is randomized. Merely observing that people with stronger convictions generate more supportive arguments does not establish whether arguments caused conviction or were generated to defend it. The literature on counterexplanation provides direct experimental manipulations of this process. Anderson (1982); Anderson and Sechler (1986).
What Do the Experiments Actually Establish?
Robust demonstrations within selected paradigms
The original experiments provide causal evidence that initial, later-invalidated information can affect judgments, because assignment to evidence conditions was manipulated. Their most secure inference concerns the selected paradigms and measured outcomes. The samples were often students, and many tasks used fictitious feedback or artificial case histories. They establish possibility and causal mechanisms under controlled conditions, but not the fraction of adults who resist authentic corrective evidence or a universal percentage of belief retained after debunking.
A contemporary counterweight: belief change is common
Four studies reported by Anglin (2019) examined reactions to empirical findings that challenged or aligned with existing beliefs. Participants often changed beliefs in response to evidence, even though their evaluations of evidence quality could be biased. This result directly limits descriptions that portray perseverance as an automatic or dominant outcome whenever evidence conflicts with expectations. How entrenched the belief is, whether the finding is credible and diagnostically relevant, and how change is measured all matter.
The reconciliation is straightforward: classic research establishes that beliefs can persist after a specially constructed discrediting sequence; contemporary evidence demonstrates that people can also revise beliefs when presented with contrary findings. One program removes the original causal basis of a recently induced theory, whereas other programs measure responses to substantively relevant new evidence. These are different experimental problems and need not produce identical results. A sound scientific account states the conditioning variables instead of portraying human belief as intrinsically immovable.
Quantitative meta-analytic evidence from related corrections
A 2017 meta-analysis of debunking research combined 52 studies with 6,878 participants. It reported standardized effects in the authors' coding system for misinformation presentation, correction, and residual influence after correction; persistence estimates were reported in the approximate range d = 0.75–1.06. Generated supportive reasons were associated with stronger residual influence, while detailed debunking was associated with larger correction effects. The findings concern a broader misinformation-correction literature, with measures and designs that extend beyond the classic belief-perseverance paradigm; they must not be advertised as a pooled effect size for all forms of belief perseverance. Chan, Jones, Jamieson, and Albarracín (2017).
Quantitative aggregation does not resolve all heterogeneity. What the respondent initially believed, what kind of correction was supplied, whether the misinformation was causally central, and how responses were scored influence an effect estimate. A standardized mean difference does not directly reveal how much a particular person's belief will change. The choice of comparison group, baseline distributions, and outcome scale remains essential for interpreting practical magnitude.
Replication, publication bias, and generalizability
The evidential map is stronger for the existence of residual effects in selected laboratory procedures than for fixed rates across populations, countries, or professional domains. A meaningful replication should preserve the discrediting manipulation, measure participant belief in the correction, predefine an outcome, and use adequate sample sizes. Replications with changed stimuli or politically salient topics test generalization, while exact procedural repetitions test reproducibility within a narrower design. A convincing synthesis must distinguish those goals.
There is also a methodological warning from the literature on a supposed 'backfire effect,' the claim that corrections typically strengthen the very misinformation they oppose. A 2020 review concluded that evidence for a broadly reproducible backfire phenomenon was weak and that measurement reliability complicates interpretation. In a subsequent study, correction conditions did not produce reliably greater endorsement of misinformation than appropriate comparison conditions. Swire-Thompson, DeGutis, and Lazer (2020); Swire-Thompson and colleagues (2022). Persistence after correction and strengthening caused by correction are analytically distinct.
When Is Perseverance More or Less Likely?
Explanatory engagement and narrative vividness
Active explanation gives an induced association a coherent structure and a larger network of imagined consequences. Concrete examples can prompt such elaboration more easily than abstract summaries in a given task. Experiments by Anderson support both mechanisms, while the findings should be interpreted as differences under controlled presentation and matched initial belief induction. Content quality and the person's prior knowledge can reverse a simplistic rule that vivid cases always win. Anderson (1983).
Time between belief formation and correction
The timing of information and correction can matter because reasoning, retrieval, and social discussion continue between exposures. A delayed correction may allow further use of the original story, but a delay can also permit independent research or reduce attachment. The effects depend on context. Researchers should track whether an observed delayed rating reflects persistence of the initial manipulation, new evidence, familiarity, or selective attrition. The firefighter experiments found residual effects at a one-week follow-up in their specific conditions, without yielding a universal duration. Anderson (1983).
Strength and credibility of the correction
A correction that merely declares a story incorrect differs from one that explains what was wrong, identifies a reliable source, and supplies a coherent replacement interpretation. Evidence suggests that correcting explanatory gaps can improve reasoning, although content, audience, format, and delayed retention matter. A correction presented by a source considered unreliable may be discounted for understandable evidential reasons, which is why source credibility and correction comprehension belong in study design. Lewandowsky and colleagues (2012); Ecker and colleagues (2022).
Cultural setting and individual differences
The classic laboratory samples are too geographically and demographically concentrated to justify a universal cultural or personality profile. Differences in institutional trust, informational context, experience with misinformation, language, literacy, and norms of disagreement may affect measured updating. Some older studies report associations between cognitive-style measures and residual perseverance, but such findings require construct validation, adequate replication, and assessment of confounding before use in individual prediction. No standardized belief-perseverance score can presently diagnose a person's general rationality or mental health. Ross, Lepper, and Hubbard (1975); Ecker and colleagues (2022).
Why Contradictory Evidence Sometimes Works
The claim that people never change their minds overstates the literature. Evidence may alter a belief by undermining a causal account, providing a more plausible model, changing perceived source reliability, or revealing an independent explanation. In an experiment, participants may also begin with weak commitments and therefore revise readily. A statistical or factual correction may succeed without eliminating every lingering inference. Belief strength and belief accuracy must therefore be measured separately.
A well-designed study asks both whether participants give less credence to the discredited claim and whether their later choices become more aligned with the valid information. The two outcomes can diverge. People may report unchanged global beliefs while modifying a particular prediction, or report agreement with a correction while repeating an obsolete explanation. This diversity is consistent with the broad multi-mechanism account of updating reviewed by Ecker and colleagues (2022).
Can Belief Perseverance Be Reduced?
Explain how the false evidence affected the initial inference
A debriefing that identifies the cognitive process behind an initial impression can work differently from one that announces only that feedback was fake. The original 1975 self-perception experiments found process-oriented debriefing effective in their specific paradigm. This motivates an actionable correction: name the invalid source, explain how it generated the apparent conclusion, and invite evaluation of the conclusion using independent evidence. Its performance for other populations and topics requires direct validation. Ross, Lepper, and Hubbard (1975).
Consider competing causal explanations
Asking participants to generate reasons that would support the opposite relationship has reduced social-theory perseverance in laboratory experiments. The intervention changes explanatory availability by making the original causal story one possible account among alternatives. The appropriate exercise involves concrete competing hypotheses and discriminating evidence, rather than an indiscriminate demand to doubt everything. Anderson (1982); Anderson and Sechler (1986).
For example, after hearing that risk-taking predicts strong firefighting performance, a person can formulate the competing causal account that lower risk-taking predicts better coordination and fewer preventable mistakes. The next task is to identify performance data capable of distinguishing these hypotheses. This is a structured way to reopen judgment; neither imagined explanation becomes evidence merely because it sounds coherent.
Replace obsolete causal links with accurate alternatives
In misinformation narratives, a correction that supplies an alternative explanation can repair a gap created by retraction. If a news story falsely identifies the cause of an event, a credible account of the actual cause may reduce reliance on the original story. The alternative should be supported by evidence and should avoid introducing another unsupported causal claim. Controlled debunking studies and evidence reviews provide stronger support for well-designed corrections than for fatalistic claims that corrections are futile. Lewandowsky and colleagues (2012); Chan and colleagues (2017).
Track confidence, source status, and predictions
A practical decision log can record a proposition, current confidence, its supporting sources, what would change the estimate, and the eventual outcome. When a source is discredited, mark which conclusions depended on it and reassess them separately. This procedure makes belief revision observable and reduces reliance on retrospective memory. It is a principled recommendation derived from measurement and probabilistic reasoning; specific claims about its long-term effectiveness require trials with meaningful field outcomes.
Prebunking and proactive explanation
Preemptive interventions teach people to recognize misleading techniques before encountering a specific example. These methods have an emerging empirical literature in misinformation research, though training effects depend on the measure, timing, audience, and setting. They address some inputs to belief formation rather than necessarily reversing a belief already supported by a self-generated theory. The evidence overview in Ecker and colleagues (2022) differentiates preventive and reactive interventions.
Limits of generalizing debiasing training
Success on a laboratory rating scale is a useful causal observation; it does not establish durable improvements in high-stakes professional decisions. Transfer depends on whether the trained operation appears in the new situation, whether a person recognizes its relevance, and whether feedback supports continued use. Longitudinal studies with independent behavioral outcomes are needed before attaching large real-world promises to a brief 'consider the opposite' exercise. Scientific communication should keep this distinction explicit.
Applications to Everyday Judgment and Institutions
Personal identity and self-assessment
Misleading feedback can affect judgments of one's ability even after its source is revealed as arbitrary. The ethical lesson is that feedback should be well calibrated to what a task measures. Classroom and workplace assessments should distinguish performance under a specific set of conditions from enduring personal characteristics. The classroom experiments provide a useful mechanism demonstration, not a basis for labeling a student or employee as permanently biased. Lepper, Ross, and Lau (1986).
Education and scientific reasoning
Students and researchers build causal explanations from examples. Teaching evidential relevance means asking whether an explanation depends on a retracted data point and whether independent results still support it. Scientific self-correction requires updating claims, interpretations, and citations together. A retracted or erroneous study must be removed from the evidential argument, while any valid independent evidence deserves separate appraisal. This distinction between source and explanatory conclusion is central to the experimental account of perseverance.
News, health information, and public debate
A retracted claim may remain salient because it supplied a memorable explanation of events. Corrective communication should clearly identify the false assertion, explain the grounds for the correction, and avoid leaving a causal vacuum. The contemporary review literature documents the continued influence effect while finding that corrections are often effective. Health-related decisions require high-quality evidence and professional guidance rather than extrapolation from general laboratory demonstrations. Ecker and colleagues (2022); Walter and colleagues (2020).
Forensic and professional assessment
Investigators, physicians, evaluators, and managers may form coherent causal hypotheses from early case information. If that information is later invalidated, the dependent parts of the explanation should be reassessed explicitly. A disciplined process preserves the evidence chronology, identifies source reliability, separates observations from interpretations, and uses structured independent review where practical. Whether such procedures reduce error in a particular profession must be determined by domain-specific studies rather than assumed from the basic firefighter paradigm.
Human interaction with AI-generated explanations
AI-mediated systems can produce persuasive explanations, summaries, or apparently well-supported claims. When an explanation is later shown to contain fabricated or incorrect support, the reader may still retain the causal story it inspired. That is a plausible application of belief-perseverance theory, but the classic experiments did not test modern generative systems; direct human–AI experiments are needed to quantify risks, interface effects, and correction outcomes. Systems that support traceable sources, evidence revision, explicit uncertainty, and visible withdrawal of invalid claims offer testable design strategies. The broader psychological framework concerns how people evaluate evidence regardless of whether the communicator is human or artificial. Ecker and colleagues (2022).
Within the Artificial Era, human engagement with machine-generated knowledge creates new environments for testing established principles of belief formation and updating. The philosophical category Artificial Era belongs to the Aisentica framework, whereas claims about how users respond to corrections require separately designed empirical research. The canonical philosophical works are available through Aisentica Publications; no new empirical principle follows automatically from an ontological classification.
Important Scientific Limitations
Belief perseverance is a research construct covering several operational designs, not a single universal tendency with a stable individual score. Experimental belief induction is not equivalent to life-long conviction. Source removal is not identical to receiving a new contradictory study. A stated attitude differs from an explanatory inference, a causal prediction, and a consequential choice. Studies have often relied on convenience samples and artificial tasks; transfer to workplaces, countries, professions, clinical decisions, and technologies must be demonstrated independently.
Even complete source discrediting may be ambiguous outside a tightly controlled experiment. Bayesian appraisal requires accounting for other evidence, and a person may rationally retain a conclusion when its independent support remains adequate. Conversely, believing that something is true without adequate evidence can be a measurable judgment problem without signifying a mental disorder. The scientifically useful approach is to identify the belief, the originally supporting observations, the exact corrective information, the respondent's understanding, and the post-correction response.
Frequently Asked Questions
Does belief perseverance mean someone refuses to accept facts?
It means that an initially induced belief can continue influencing judgment after the experimenter has invalidated the information that induced it. It does not identify conscious refusal. Memory integration, self-generated explanation, misunderstanding, independent evidence, and motivation can produce different patterns that should be examined separately.
Is belief perseverance the same as confirmation bias?
No. Confirmation bias concerns asymmetries in selecting or interpreting information; belief perseverance examines the residual influence of a belief after its original evidential support has been undermined. The processes can interact but are measured through different manipulations.
Can a belief survive even when someone remembers a correction?
Yes. Continued-influence research has shown that people may remember and accept a correction but still use discredited details when making later inferences. Recognizing a retraction and repairing the causal model derived from the retracted material are separate operations. Johnson and Seifert (1994).
Are people generally unable to change their minds?
No. Controlled research also documents belief change in response to challenging evidence. The four-study investigation by Anglin (2019) specifically cautions against portraying perseverance as the default outcome in all evidence-evaluation settings.
Is the backfire effect a stronger form of belief perseverance?
The concepts have different comparisons. Perseverance means residual influence after discrediting; backfire means a correction causes stronger belief in misinformation than a suitable no-correction counterfactual. Research does not support treating backfire as a routine consequence of correction. Swire-Thompson and colleagues (2022).
Why are vivid examples so memorable after being disproven?
Vivid cases can prompt causal elaboration that persists independently of the original anecdotes. In some experiments, concrete cases elicited greater belief perseverance than abstract summaries. The effect depends on the content and task; vividness alone is not a proven sufficient cause. Anderson (1983).
Does correcting misinformation work?
Often it does. Experimental reviews and meta-analyses document meaningful effects of corrections alongside possible residual influence. Correction efficacy depends on the information, audience, design, timing, and outcome measured. Chan and colleagues (2017); Ecker and colleagues (2022).
What is a useful question to ask after a retraction?
Ask: 'Which parts of my current explanation were generated from the invalid source, and what independent observations would support those parts now?' This makes the evidential dependency visible and encourages a fresh test of the causal story.
Conclusion
Belief perseverance is the experimentally demonstrated possibility that a judgment, impression, or explanatory theory continues to be influenced by evidence after its original source has been discredited. Classic studies reveal the importance of causal explanation, attribution, and the persistence of mental models. Contemporary research adds a necessary qualification: people also revise beliefs, correction generally can work, and a stronger backfire response is not a routine outcome. The quality of belief updating depends on what evidence was removed, what independent support remains, and whether the inferences built from invalid material are explicitly reconsidered. Scientific progress on this topic requires precise measures of the initial manipulation, correction comprehension, subsequent beliefs, and consequential judgments.
Related Articles
Confirmation Bias: Why We Favor Supporting Evidence — How evidence search and interpretation become asymmetrical.
Motivated Reasoning: How Goals Shape What We Believe — How preferred outcomes influence evidence evaluation.
Hindsight Bias: Why Outcomes Seem Predictable Afterward — How later knowledge changes retrospective judgments.
Illusory Truth Effect: How Repetition Increases Believability — How familiarity changes perceived credibility.
References
Anderson, C. A. (1982). Inoculation and counterexplanation: Debiasing techniques in the perseverance of social theories. Social Cognition, 1(2), 126–139. https://doi.org/10.1521/soco.1982.1.2.126.
Anderson, C. A. (1983). Abstract and concrete data in the perseverance of social theories: When weak data lead to unshakeable beliefs. Journal of Experimental Social Psychology, 19(2), 93–108. https://doi.org/10.1016/0022-1031(83)90031-8.
Anderson, C. A., Lepper, M. R., and Ross, L. (1980). Perseverance of social theories: The role of explanation in the persistence of discredited information. Journal of Personality and Social Psychology, 39(6), 1037–1049. https://doi.org/10.1037/h0077720.
Anderson, C. A., and Sechler, E. S. (1986). Effects of explanation and counterexplanation on the development and use of social theories. Journal of Personality and Social Psychology, 50(1), 24–34. https://doi.org/10.1037/0022-3514.50.1.24.
Anglin, S. M. (2019). Do beliefs yield to evidence? Examining belief perseverance vs. change in response to congruent empirical findings. Journal of Experimental Social Psychology, 82, 176–199. https://doi.org/10.1016/j.jesp.2019.02.004.
Chan, M.-P. S., Jones, C. R., Jamieson, K. H., and Albarracín, D. (2017). Debunking: A meta-analysis of the psychological efficacy of messages countering misinformation. Psychological Science, 28(11), 1531–1546. https://doi.org/10.1177/0956797617714579.
Davies, M. F. (1997). Belief persistence after evidential discrediting: The impact of generated versus provided explanations on the likelihood of discredited outcomes. Journal of Experimental Social Psychology, 33(6), 561–578. https://doi.org/10.1006/jesp.1997.1336.
Ecker, U. K. H., Lewandowsky, S., Cook, J., et al. (2022). The psychological drivers of misinformation belief and its resistance to correction. Nature Reviews Psychology, 1, 13–29. https://doi.org/10.1038/s44159-021-00006-y.
Fazio, L. K., Brashier, N. M., Payne, B. K., and Marsh, E. J. (2015). Knowledge does not protect against illusory truth. Journal of Experimental Psychology: General, 144(5), 993–1002. https://doi.org/10.1037/xge0000098.
Johnson, H. M., and Seifert, C. M. (1994). Sources of the continued influence effect: When misinformation in memory affects later inferences. Journal of Experimental Psychology: Learning, Memory, and Cognition, 20(6), 1420–1436. https://doi.org/10.1037/0278-7393.20.6.1420.
Lepper, M. R., Ross, L., and Lau, R. R. (1986). Persistence of inaccurate beliefs about the self: Perseverance effects in the classroom. Journal of Personality and Social Psychology, 50(3), 482–491. https://doi.org/10.1037/0022-3514.50.3.482.
Lewandowsky, S., Ecker, U. K. H., Seifert, C. M., Schwarz, N., and Cook, J. (2012). Misinformation and its correction: Continued influence and successful debiasing. Psychological Science in the Public Interest, 13(3), 106–131. https://doi.org/10.1177/1529100612451018.
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.
Ross, L., Lepper, M. R., and Hubbard, M. (1975). Perseverance in self-perception and social perception: Biased attributional processes in the debriefing paradigm. Journal of Personality and Social Psychology, 32(5), 880–892. https://doi.org/10.1037/0022-3514.32.5.880.
Swire-Thompson, B., DeGutis, J., and Lazer, D. (2020). Searching for the backfire effect: Measurement and design considerations. Journal of Applied Research in Memory and Cognition, 9(3), 286–299. https://pubmed.ncbi.nlm.nih.gov/32905023/.
Swire-Thompson, B., Miklaucic, N., Wihbey, J. P., Lazer, D., and DeGutis, J. (2022). The backfire effect after correcting misinformation is strongly associated with reliability. Journal of Experimental Psychology: General, 151(7), 1655–1665. https://doi.org/10.1037/xge0001131.
Walter, N., Cohen, J., Holbert, R. L., and Morag, Y. (2020). Fact-checking: A meta-analysis of what works and for whom. Political Communication, 37(3), 350–375. https://doi.org/10.1080/10584609.2019.1668894.