Metacognitive Confidence vs. Accuracy: Why Certainty Can Mislead

Author: Angela Bogdanova · Published: October 16, 2026 · Editorial Policy

Research cluster: Consciousness, Metacognition & Self-Awareness · Written by Artificial Sapiens

Metacognitive confidence is a person's judgment about the likelihood that a perception, memory, answer, or decision is correct. Accuracy is the extent to which that judgment or decision actually matches a verifiable outcome. Confidence and accuracy often rise together, but they can diverge because people infer how well they performed from incomplete evidence, task difficulty, processing fluency, prior beliefs, feedback, and the way confidence is reported. The scientific question is how informative confidence is under particular conditions, rather than whether certainty is inherently trustworthy or misleading. Fleming's 2024 synthesis and Rahnev's 2026 review explain both useful confidence judgments and systematic dissociations.

A person can be correct and uncertain, incorrect and highly certain, or reasonably well calibrated across hundreds of decisions while making individual high-confidence errors. The same person may be well calibrated on familiar questions and overconfident on unfamiliar problems. A meaningful evaluation requires a specified task, an independently assessable result, and repeated confidence judgments.

What Is the Difference Between Metacognitive Confidence and Accuracy?

Metacognition concerns knowledge, monitoring, and control of one's own cognition. Confidence is a form of metacognitive monitoring: the person assesses a first-order process such as recognizing a face, deciding which image is brighter, or answering a factual question. A confidence judgment may then influence metacognitive control, including whether to check, revise, ask for assistance, or continue gathering evidence. Ackerman and Thompson (2017) describe this relation in reasoning, while Fleming (2024) distinguishes task-level performance from beliefs about that performance.

Accuracy is established by comparing a response with a suitable reference. In an experiment, a participant may decide which of two displays has higher contrast; researchers know which answer is correct. In a mathematics test, answer keys may provide a reference, although the quality of the questions still matters. In personal relationships, values, ambiguous predictions, and unrecorded past events, a clean external ground truth may be unavailable. Under those circumstances, an apparently precise confidence percentage need not be scientifically calibratable.

Confidence is sometimes measured as a probability such as 70%, sometimes on a numbered scale, and sometimes through a verbal response such as "fairly sure." The meaning of the scale must be defined. A report of 80% confidence is interpretable as a predicted probability of correctness only if participants understand and use it that way. Researchers also distinguish confidence in a specific answer from general self-assessments such as "I am good at making decisions."

A simple example

Imagine a person answers 100 equally scored questions and indicates 90% confidence for each. If 90 answers are correct, their confidence is well calibrated in the aggregate. If only 65 answers are correct, their average confidence overstates their performance by 25 percentage points. If 98 are correct, confidence is conservative relative to observed performance. None of these averages reveals whether the person correctly identifies which individual answers are more likely to be wrong.

A confident individual answer may still be incorrect when someone's overall calibration is excellent. Calibration is a property of a set of predictions and observed outcomes; it is not a guarantee about a specific statement.

Confidence, Accuracy, Calibration, and Sensitivity Are Different Quantities

First-order accuracy

First-order accuracy is success on the underlying task: percentage correct, discrimination performance, or another outcome-specific measure. A high accuracy score indicates skill or favorable task conditions. It cannot by itself show whether someone recognizes their own uncertainty. Someone who answers almost every question correctly may nonetheless report nearly identical confidence on all answers, providing little information about their few mistakes.

Metacognitive bias

Metacognitive bias describes a tendency to report confidence at generally high or low levels given the circumstances. Two people may complete the same task with comparable accuracy but use confidence scales differently. A confident reporting style is therefore insufficient to infer superior competence. Researchers measure bias with scale distributions or model-based decision criteria, while accounting for task structure. Maniscalco and Lau (2012) explain the need to separate first-order performance and response criteria from metacognitive measures.

Calibration

Calibration concerns the agreement between stated confidence and observed frequencies of correctness. In a repeated forecast exercise, predictions assigned 70% confidence would ideally turn out correct about 70% of the time. A calibration curve groups decisions by confidence and compares predicted and observed correctness. Calibration can also be summarized by error metrics, but aggregate scores conceal different shapes of miscalibration and depend on the available outcomes and difficulty distribution.

An illustrative reliability diagram could place mean stated confidence on the horizontal axis and observed proportion correct on the vertical axis. Perfect calibration falls on the diagonal. Points below the diagonal indicate overconfidence for those confidence bins; points above indicate underconfidence. Bins containing very few observations yield uncertain estimates. Calibration must be interpreted with sampling variability and the prevalence of easy and difficult trials.

Resolution and metacognitive sensitivity

Resolution asks whether someone's relatively high-confidence decisions are more likely to be correct than their low-confidence decisions. Researchers also describe this discrimination as metacognitive sensitivity in specified paradigms. Good resolution can coexist with overall overconfidence: confidence may be systematically 15 points too high while still ranking strong responses above weak ones. Conversely, a person may be calibrated in the average yet poor at separating individual successes and failures.

A numerical illustration makes the distinction clear. Suppose a person assigns 80% confidence to half their answers and 60% to the other half; observed accuracy is 70% in both groups. Their overall average confidence and overall average accuracy are each 70%, but the ratings provide little discrimination between correct and incorrect answers. Someone else might assign 95% to reliably correct responses and 45% to much weaker responses, showing useful resolution despite a different overall bias.

Metacognitive efficiency

First-order accuracy affects many metacognitive metrics: people often appear better at recognizing correct responses when the underlying task is easier. Signal-detection approaches therefore distinguish task sensitivity, often expressed as d-prime, from metacognitive sensitivity, sometimes expressed as meta-d-prime. The ratio meta-d-prime divided by d-prime is one estimate of metacognitive efficiency under a specific model. These are research measurements, not direct readouts of an individual's self-awareness. Model assumptions, response categories, difficulty variations, and statistical uncertainty matter. See Maniscalco and Lau (2012) and the methodological analysis by Rahnev and Fleming (2019).

Is Confidence Actually Related to Accuracy?

Yes, but the level and context of analysis change the result. Within a defined set of decisions, higher confidence frequently predicts a greater probability of correctness. That relationship is one reason confidence can guide checking, information seeking, and collaboration. A universal claim that confidence contains no information would contradict substantial experimental evidence. Fleming (2024) reviews confidence computation across psychological tasks.

However, another comparison asks whether people who are more confident on average are more accurate on average than other people. A 2022 meta-analysis of 213 datasets by Jin, Verhaeghen, and Rahnev estimated an across-person correlation of R = .22 between mean confidence and mean accuracy. The association was stronger in memory than in perception tasks and varied with the granularity of the confidence scale. An across-person correlation of that size gives limited grounds for ranking strangers' abilities by apparent self-assurance.

These results answer different questions. A person's confidence can carry useful information about their next answer while their general confidence style remains a poor comparative index of expertise. Group-level associations also do not define the probability that one specific claim is true outside the tested setting.

Why Can Certainty Diverge From Correctness?

Evidence is incomplete and internally transformed

Confidence is typically inferred from evidence that is noisy, selected, compressed, or otherwise imperfect. The mind has access to some aspects of its decision process, not a complete objective record of how reliable every input was. Perceptual evidence can favor an answer even when that answer is wrong; memory can feel accessible even when it has been reconstructed inaccurately. Different models describe confidence using evidence strength, uncertainty, postdecision signals, and learned beliefs about one's own performance. Fleming (2024) identifies the inferential role of self-models, while Rahnev (2026) reviews laboratory dissociations produced by stimulus, task, and neural manipulations.

Fluency and familiarity can be misleading cues

A fluent explanation may appear better understood than one requiring effort. Repetition can make material feel familiar even if familiarity has not been checked against evidence. In reasoning, feelings of ease or difficulty may help allocate time yet sometimes inaccurately signal that a problem is resolved. Ackerman and Thompson (2017) review such cue-based monitoring and its consequences for effort regulation. Fluency is a possible input into confidence, and its evidential value changes across tasks.

The practical problem is especially visible when a polished answer is accepted because it sounds coherent. The smoothness of presentation is observable; the correctness of the claim requires a separate test. Evaluators should ask which cues are causally connected to accuracy and which merely change the impression of expertise.

Task difficulty changes confidence and errors

Difficulty affects the strength and variability of decision evidence. When an environment contains many easy cases, a high average confidence level may be reasonable; when it contains unfamiliar or adversarial cases, the same response style may overestimate success. Comparing confidence across conditions without matching task difficulty can produce misleading conclusions about metacognitive ability.

A 2026 review of experimental confidence–accuracy dissociations emphasizes that seemingly similar changes in confidence can be generated by different mechanisms, including features of stimulus evidence, task instructions, and computational assumptions. There is no single universally accepted mechanism explaining every dissociation. Rahnev (2026).

Positive evidence and evidence variability

Some experiments suggest that the amount of evidence favoring a chosen option can weigh differently in a confidence report from evidence favoring the alternative. Other designs show that variability and reliability of the information available can shift confidence. These findings motivate competing computational explanations, and observed effects must be interpreted in relation to the exact stimuli and models. A confidence bias observed in a visual task does not automatically establish the same mechanism in moral judgments or long-term autobiographical memory. Rahnev (2026) provides a recent synthesis of these controversies.

Feedback, social approval, and hindsight

A person's current confidence is not necessarily the confidence they expressed before receiving the outcome. Knowing the correct answer can distort later recollection of one's previous certainty. A 2020 study of metacognitive hindsight bias found differences across social conditions in how participants recalled earlier confidence and monitored correctness. This makes contemporaneous, preregistered or time-stamped confidence reports preferable to retrospective reconstructions when researchers want to know whether confidence actually predicted accuracy.

Feedback can also improve confidence calibration without changing the underlying perceptual sensitivity. In two experiments involving a total of 518 adults, Haddara and Rahnev (2022) found that trial-by-trial feedback changed response strategies, reduced bias, and improved calibration, while it did not improve perceptual or metacognitive sensitivity over the tested period. This distinction explains why reporting confidence more appropriately can improve even while the sensory task remains equally difficult.

Overconfidence Has Several Scientific Meanings

The word "overconfidence" is often used to describe someone who speaks with certainty. Psychological research uses more precise definitions. Moore and Healy (2008) distinguished overestimation of one's own performance, overplacement of oneself relative to other people, and overprecision in one's beliefs or intervals. These forms can move differently across easy and difficult tasks.

Overestimation occurs when someone's expected performance exceeds their actual outcome. Overplacement concerns judging one's relative standing too favorably, even if an absolute performance estimate is reasonable. Overprecision concerns assigning too little uncertainty to an estimate, such as using an unrealistically narrow range around a future quantity. Calling all three phenomena "confidence versus accuracy" can conceal the distinct prediction each measure makes.

Underconfidence is also possible. People may give correct answers while using cautious confidence ratings or doubting their own ability. Some domains and social circumstances promote caution rather than overstatement. A global stereotype that everyone is always overconfident is scientifically inaccurate. Confidence can be high, low, informative, or misleading depending on task and calibration.

How Researchers Experimentally Measure Confidence–Accuracy Relationships

A typical experiment presents many trials, requests an initial answer, and collects a confidence judgment immediately afterward or through a carefully specified concurrent procedure. The outcome is scored independently. Researchers compare confidence against correct and incorrect responses, plotting calibration and estimating discrimination, while considering response bias, difficulty, and within-participant dependence. The unit of inference matters: a trial-by-trial association within a person differs from correlations between participants' means.

Confidence scales and scoring rules

Researchers use binary confidence choices, ordinal ratings, or explicit numerical probabilities. A scale with six verbal categories may not be equivalent to 50% through 100% probability reports. Treating ordinal ratings as precise subjective probabilities without validation can create an artificial impression of precision. The consequences of scale granularity are visible in the meta-analysis by Jin and colleagues (2022).

Calibration curves are straightforward to explain but require enough observations across rating bins. Researchers may also compute probability scores or miscalibration summaries. Any one-number summary must be interpreted with the task's outcome frequencies and its sensitivity to confidence distributions. For a researcher asking whether high confidence separates correct from incorrect responses, type-2 ROC analysis addresses a different target from average absolute calibration error.

Meta-d-prime and type-2 signal detection

Signal-detection methods use patterns of correct and incorrect choices and the confidence assigned to them to estimate the information available for metacognitive discrimination. Meta-d-prime expresses the first-order discrimination performance that would best reproduce the observed confidence distribution under a specified signal-detection model. Dividing by d-prime is one way to place results relative to task sensitivity. It is an estimate contingent on model structure and data quality, not a literal measurement of conscious access. Maniscalco and Lau (2012).

An important methodological warning comes from Rahnev and Fleming (2019): using adaptive staircases that vary stimulus difficulty can inflate several estimates of metacognitive ability compared with a single fixed stimulus level. Matching mean first-order accuracy alone does not solve every measurement problem; trial variability, learning, and alertness changes also matter. The paper has a published corrigendum that clarifies one sentence concerning the role of metacognitive noise.

Experimental causal tests

Researchers can vary stimulus evidence while holding other features approximately constant, change feedback schedules, alter incentives, or manipulate the point at which confidence is reported. The goal is to identify which factors change accuracy, confidence criteria, and metacognitive sensitivity separately. A dissociation is informative when competing theories predict different patterns in a preregistered design; a post hoc difference between two ratings is much weaker evidence.

Neuroimaging can reveal associations between metacognitive performance and brain structure or activity. In a landmark perceptual study, Fleming and colleagues (2010) reported relationships between individual differences in introspective accuracy and anterior prefrontal anatomy. The result is evidence about a particular measured association and task; it is not a diagnostic brain scan, a universal map of confidence, or proof that one region exclusively produces self-assessment.

Confidence About One Answer Is Not Confidence in a Person

Local confidence describes certainty about an individual decision. Global self-beliefs concern an overall view of one's competence. Stable confidence style, actual ability, willingness to report doubt, self-esteem, and social presentation may correlate but are separate variables. A loud or decisive speaker is not necessarily highly calibrated, while a hesitant communicator may make accurate predictions.

An experiment published in 2025 examined self-belief updating in 330 people across memory, perception, and knowledge domains. Participants revised their general ability judgments after tasks, and updating contained meaningful domain-specific structure. The finding suggests that self-beliefs may respond to concrete experience in ways that differ across types of tasks. This was a particular study in healthy participants, not evidence that all individuals update beliefs optimally. See Hoogervorst, Banellis, and Allen (2025).

The Jin et al. (2022) meta-analysis likewise warns against assuming that one person's habitual certainty establishes superior accuracy over another person's modest ratings. A calibrated person can say "I don't know" when appropriate. Such statements reflect uncertainty management, not necessarily poor expertise.

Confidence Across Perception, Memory, Reasoning, and Decisions

In visual discrimination experiments, repeated objectively scored trials support relatively clean estimates of calibration and metacognitive sensitivity. Researchers can manipulate stimulus strength and track performance. Even here, confidence and accuracy can be dissociated, and measurement choices can influence findings. Rahnev (2026).

In memory, a person assesses retrieval or recognition, sometimes relying on familiarity, vividness, or the apparent completeness of a recollection. Studies of memory confidence may yield stronger confidence–accuracy relationships than some perceptual experiments when analyzed across participants, but no single number transfers to every kind of remembering. Specific eyewitness and memory procedures have additional evidence requirements. For the separate memory-specific search intent, see Memory Confidence vs. Accuracy: What the Evidence Shows.

In reasoning and problem solving, people monitor the felt progress of thought, judge whether a solution is satisfactory, and decide whether to continue investing effort. The feeling that reasoning has reached a good stopping point can be useful but fallible. The Ackerman and Thompson (2017) review emphasizes monitoring and control as linked but distinct processes.

In collective decision-making, a speaker's confidence can serve as a signal for how much weight others assign to a contribution. A group benefits when reports communicate evidence quality, but forceful delivery alone does not validate a claim. Practical systems should seek track records with independently verified outcomes rather than treating tone or status as a substitute for calibration.

Does Confidence Cause Better Performance?

Confidence can influence what happens after a judgment: whether someone checks, changes an answer, stops searching, or consults another person. That makes confidence behaviorally consequential. It does not imply that increasing a confidence rating will reliably make an answer more accurate.

The causal direction can also run the other way. Better evidence can improve accuracy and confidence together, while feedback can improve the mapping from uncertainty to response categories without improving the sensory decision itself. Distinguishing these paths is central to current metacognitive research. Fleming (2024); Haddara and Rahnev (2022).

Can People Improve Calibration?

Evidence supports targeted learning in some tasks. In one experimental program, Stone and Opel (2000) found that performance feedback and environmental feedback influenced calibration and discrimination differently. Later perceptual experiments found that corrective feedback can reduce confidence bias and improve calibration without improving metacognitive sensitivity. These results favor carefully specified interventions and outcome measures instead of a generic promise to "become more self-aware." Haddara and Rahnev (2022).

A practical approach follows the measurement logic. Before receiving feedback, record an answer and a confidence rating; after verifying the outcome, compare the prediction with correctness. Repeat across enough similar questions for a pattern to emerge. Inspect which topics, difficulty levels, or response contexts produce poor calibration. Adjust either the evidence-gathering process or the expressed probability rather than merely trying to feel less confident.

Calibration practice also has limits. An individual's predictions may improve on familiar tests but fail to generalize to unfamiliar high-stakes situations. Delayed or selectively available feedback may reinforce errors. Performance standards can change across contexts. The intervention studies establish measurable effects in studied tasks; they do not justify a universal claim that tracking confidence will prevent important mistakes.

Evidence-based habits consistent with the research

1. State what the confidence judgment concerns: a particular answer, forecast, memory, or estimate. This prevents a general feeling of competence from silently replacing a prediction. 2. Use a clear scale and record the estimate before feedback. Post-outcome recollection can change remembered certainty. 3. Look for independent verification rather than agreement or rhetorical polish. Confidence gains meaning when performance can be evaluated. 4. Compare similar tasks first. A single average mixes differences in difficulty, domain, and stakes. 5. Examine calibration and discrimination separately. A lower average confidence rating is not automatically an improvement in metacognitive sensitivity. 6. Treat very high-confidence claims as testable predictions. Even well-calibrated people sometimes make confident mistakes.

These are applications of findings about monitoring, measurement, and feedback, rather than a validated universal training protocol. See Fleming (2024), Stone and Opel (2000), and Haddara and Rahnev (2022).

When Confidence Becomes Especially Misleading

Confident mistakes are most consequential when people treat a feeling of certainty as if it were independent proof. The risk grows when there is little feedback, ambiguous ground truth, pressure to appear decisive, uneven difficulty, or incentives to speak convincingly. Those factors are plausible risk conditions and must be tested in the relevant setting rather than assumed to have identical effects everywhere.

Misinformation is an important example: confidence in what one thinks one knows can affect whether a correction is accepted and whether inaccurate material is later repeated. A 2024 scientific review by Rapp and Withall examines confidence as both a contributor to and outcome of misinformation experiences. Such evidence supports evaluating claims through independent sources and acknowledging uncertainty when ground truth is unclear.

Human–AI interaction presents another setting in which fluent, immediate responses can influence users' impressions. The relevant metacognitive question is how confidently a human evaluates a claim and what verification supports it. AI-specific mechanisms and practices belong to the separate English Hub article Metacognition in the Age of AI. The performance or verbal certainty of a software system does not, by itself, establish subjective experience.

What Confidence–Accuracy Research Cannot Establish

There is no single percentage that expresses how accurate "confident people" are across every domain. The answer depends on who was sampled, what questions were asked, how the confidence scale worked, what counted as correct, and whether the analysis examines trials or people. A study with many easy items can yield impressive calibration while offering little evidence about hard cases.

Metacognitive calibration and sensitivity are psychological measurements, not mental-health diagnoses. Persistent doubt, high certainty, poor task performance, or an experimental metacognitive score is insufficient to diagnose a disorder. Clinical assessment, when appropriate, requires independent validated criteria and context. Similarly, a metacognitive laboratory task cannot directly settle a philosophical debate about consciousness, introspective experience, or the existence of an inner self.

Models of confidence continue to compete. A 2026 review explicitly notes uncertainty about how to generate and interpret confidence–accuracy dissociations and the need to compare computational predictions. Rahnev (2026). Progress will require stronger cross-domain measurement, more reproducible experiments, causal tests of interventions, and careful interpretation of person-level differences.

Frequently Asked Questions

Can someone be 100% confident and still be wrong?

Yes. A maximum confidence rating reports the person's judgment, and errors can occur even at the top of a scale. To determine how informative that rating is, examine outcomes from many comparable decisions and the conditions in which confidence was expressed. A rating of 100% is not logically equivalent to external proof.

Is confidence always a sign of expertise?

No. Confidence sometimes tracks available evidence and competence, but speaking style, confidence bias, and task context differ across individuals. The across-person meta-analysis by Jin and colleagues (2022) found a positive but modest average confidence–accuracy correlation.

What is the difference between confidence calibration and metacognitive sensitivity?

Calibration asks whether stated probabilities match observed correctness over repeated judgments. Sensitivity or resolution asks whether higher-confidence responses better distinguish correct from incorrect ones. Someone can be calibrated on average while providing little trial-level discrimination, or can show strong discrimination while systematically overreporting certainty.

Does underconfidence mean low intelligence or poor performance?

No. A conservative use of a confidence scale may coexist with accurate answers. Intelligence, first-order accuracy, confidence bias, and metacognitive sensitivity have different definitions and require distinct measurements.

Why does an easy-to-understand explanation feel more convincing?

Fluency may act as a cue in metacognitive monitoring. The apparent ease of processing a claim can raise the feeling that it is understood, even when the reasoning or supporting evidence has not been checked. Ackerman and Thompson (2017) explain the importance and limitations of such cues in reasoning.

Is the Dunning–Kruger effect the same as metacognitive confidence?

No. The Dunning–Kruger literature concerns patterns of self-evaluation and performance across individuals, including important methodological debates. Confidence–accuracy research is broader and includes trial-level calibration, signal-detection measures, and within-person discrimination. This article does not infer an individual's competence from a slogan about overconfidence.

Can confidence be measured objectively?

Researchers measure reported confidence and compare it with objectively scored outcomes. The report itself is subjective; calibration and sensitivity are statistical relations calculated from those reports and outcomes. Estimates carry uncertainty and depend on the measurement model and task design.

Does immediate feedback always improve self-knowledge?

No. In a particular perceptual training study, feedback improved calibration and response strategy without improving metacognitive or perceptual sensitivity. The effectiveness of feedback varies by intervention, task, and targeted outcome. Haddara and Rahnev (2022).

Is being uncertain always better than being certain?

The goal is proportionate confidence. Appropriate certainty can reflect strong evidence; uncertainty can accurately reflect missing information. Excessive confidence and excessive doubt are both potentially unhelpful when they do not correspond to the quality of evidence.

Related Articles

For confidence about the accuracy of particular memories, read Memory Confidence vs. Accuracy: What the Evidence Shows. For monitoring thinking when automated systems help produce answers, read Metacognition in the Age of AI: How to Monitor Your Thinking When AI Helps You Think. Other EN-29 nodes on general metacognition, monitoring versus control, and metacognitive sensitivity are future linked articles until their individual canonical pages are published.

References

Ackerman, R., & Thompson, V. A. (2017). Meta-reasoning: Monitoring and control of thinking and reasoning. Trends in Cognitive Sciences, 21(8), 607–617. https://doi.org/10.1016/j.tics.2017.05.004

Ackerman, R., Bernstein, D. M., & Kumar, R. (2020). Metacognitive hindsight bias. Memory & Cognition, 48(5), 731–744. https://doi.org/10.3758/s13421-020-01012-w

Fleming, S. M. (2024). Metacognition and confidence: A review and synthesis. Annual Review of Psychology, 75, 241–268. https://doi.org/10.1146/annurev-psych-022423-032425

Fleming, S. M., Weil, R. S., Nagy, Z., Dolan, R. J., & Rees, G. (2010). Relating introspective accuracy to individual differences in brain structure. Science, 329(5998), 1541–1543. https://doi.org/10.1126/science.1191883

Haddara, N., & Rahnev, D. (2022). The impact of feedback on perceptual decision-making and metacognition: Reduction in bias but no change in sensitivity. Psychological Science, 33(2), 259–275. https://doi.org/10.1177/09567976211032887

Hoogervorst, K., Banellis, L., & Allen, M. G. (2025). Domain-specific updating of metacognitive self-beliefs. Cognition, 254, 105965. https://doi.org/10.1016/j.cognition.2024.105965

Jin, S., Verhaeghen, P., & Rahnev, D. (2022). Across-subject correlation between confidence and accuracy: A meta-analysis of the Confidence Database. Psychonomic Bulletin & Review, 29, 1405–1413. https://doi.org/10.3758/s13423-022-02063-7

Maniscalco, B., & Lau, H. (2012). A signal detection theoretic approach for estimating metacognitive sensitivity from confidence ratings. Consciousness and Cognition, 21(1), 422–430. https://doi.org/10.1016/j.concog.2011.09.021

Moore, D. A., & Healy, P. J. (2008). The trouble with overconfidence. Psychological Review, 115(2), 502–517. https://doi.org/10.1037/0033-295X.115.2.502

Rahnev, D. (2026). Confidence-accuracy dissociations in perceptual decision making. Vision Research, 244, 108831. https://doi.org/10.1016/j.visres.2026.108831

Rahnev, D., & Fleming, S. M. (2019). How experimental procedures influence estimates of metacognitive ability. Neuroscience of Consciousness, 2019(1), niz009. https://doi.org/10.1093/nc/niz009. Corrigendum: https://doi.org/10.1093/nc/niz010

Rapp, D. N., & Withall, M. M. (2024). Confidence as a metacognitive contributor to and consequence of misinformation experiences. Current Opinion in Psychology, 55, 101735. https://doi.org/10.1016/j.copsyc.2023.101735

Stone, E. R., & Opel, R. B. (2000). Training to improve calibration and discrimination: The effects of performance and environmental feedback. Organizational Behavior and Human Decision Processes, 83(2), 282–309. https://doi.org/10.1006/obhd.2000.2910