The Science of Judgment and Decision-Making
Cognitive biases, judgment, and decision-making form an interconnected scientific field concerned with how people interpret evidence, estimate uncertain events, assign value to possible outcomes, and choose courses of action. A cognitive bias is a systematic pattern in a judgment or decision that departs from a specified benchmark under identifiable conditions. The benchmark might be probability theory, formal logic, expected-value calculation, an experimentally controlled comparison, or the person's own stated preferences. The word systematic matters: a single mistaken estimate does not establish a cognitive bias, and the existence of a bias in one task does not establish that a person reasons poorly in every setting.
Judgment refers to the formation of estimates, evaluations, and beliefs: How likely is an event? How credible is a claim? What caused an outcome? Preference refers to the relative desirability assigned to alternatives. Decision-making refers to selecting an action or option under constraints. A person may estimate probabilities accurately yet prefer a risky option, or estimate probabilities poorly while choosing well because the consequences favor the selected action. Those are different scientific observations. Modern judgment and decision-making research accordingly combines analysis of decision problems, description of human responses, and evaluation of interventions designed to improve performance. Fischhoff and Broomell, 2020.
The field has established that presentation, reference points, sampling, uncertainty, expertise, incentives, emotion, and task structure can change measured responses. Its central methodological lesson is that a claim about rationality requires an explicit criterion and evidence that the criterion fits the problem being studied. A cognitive heuristic can produce a reliable judgment when it exploits the structure of an environment; the same strategy may fail when relevant base rates are obscured or information is systematically selected. This is why scientific accounts examine mechanisms, effect sizes, context, and alternative explanations rather than treating a list of named biases as a universal inventory of mental defects. Tversky and Kahneman, 1974; Gigerenzer and Gaissmaier, 2011.
What Counts as a Cognitive Bias?
A bias is usually identified by comparing responses with a defined standard across multiple trials, people, or conditions. If two groups receive mathematically equivalent descriptions but show different choice distributions, the design may reveal a framing effect. If confidence systematically exceeds observed correctness over many questions, the resulting pattern may indicate overconfidence under that scoring rule. If a probability judgment assigns a conjunction a greater probability than one of its constituent events, it violates the probability rule that P(A and B) cannot exceed P(A). These operational definitions allow competing explanations to be tested.
A heuristic is a rule or strategy for processing information, such as assessing frequency from remembered examples, sampling a small number of cues, or stopping a search when a satisfactory option appears. A heuristic can be efficient and accurate. A logical fallacy is a problem with the structure of an argument or inference. A statistical error may arise from inappropriate sampling or a mistaken model, without identifying an individual's psychological mechanism. A decision preference describes what a person values; a preference for security or risk is not itself a mathematical error. A behavioral tendency is an observed regularity without a settled causal account. A clinical symptom belongs to the assessment of a health condition; ordinary framing effects, anchoring, and overconfidence are not diagnoses.
The designation bias is therefore relational: bias relative to which benchmark, applied to which response, in which environment? The same judgment might be inaccurate in a well-defined numerical problem and reasonable in an uncertain real-world setting where the available probabilities are unreliable. Researchers distinguish errors produced by a participant from errors introduced by the experimenter's assumed model, ambiguous wording, selection of examples, or an inappropriate normative criterion. This distinction is central to theories of ecological rationality and to disputes about interpretation of classic tasks. Gigerenzer and Gaissmaier, 2011.
Normative, Descriptive, and Prescriptive Questions
Normative models specify how a judgment or choice should be made given particular axioms and goals. Probability theory constrains coherent probability assignments; Bayes' theorem specifies updating conditional on a prior and a likelihood; expected-utility theory represents preferences satisfying particular consistency axioms; and decision theory can identify a choice maximizing a stipulated objective. None of these benchmarks determines a person's goals, values, or access to information by itself.
Descriptive models seek to explain what people actually do, using mental representations, search strategies, reference points, learning, attention, emotional states, and computational limitations. Prospect theory, evidence-accumulation models, and fast-and-frugal heuristics operate at different descriptive levels. Prescriptive research tests whether methods such as statistical formats, structured forecasts, feedback, and decision aids improve accuracy, calibration, or outcomes. A successful prescriptive intervention needs a measurable criterion of improvement, not merely a change in the participant's answer. Fischhoff and Broomell, 2020.
Error, Variability, and Preference Reversal
Random error is unpredictable scatter; systematic bias is a directional pattern relative to a benchmark. Reliability measures whether repeated tasks produce stable results; validity measures whether the task actually captures the proposed construct. Preference reversal arises when equivalent or strategically related elicitation procedures produce inconsistent orderings. It can reveal context-dependent valuation, but the researcher must check whether the procedures genuinely elicit the same underlying preference. Some effects reflect changed attention or information rather than a fixed individual trait.
Historical Foundations and Research Programs
The study of judgment grew from psychophysics, statistics, decision theory, economics, cognitive psychology, and organizational research. Herbert Simon's account of bounded rationality emphasized decision-makers operating with limited information, time, and computation, and introduced the analysis of satisfactory rather than globally optimal choices. Work on subjective probability and Bayesian inference examined the gap between formal uncertainty and human estimates. Later experiments by Amos Tversky and Daniel Kahneman identified representativeness, availability, and anchoring as candidate organizing strategies and demonstrated systematic outcomes in specified judgment tasks. Simon, 1955; Tversky and Kahneman, 1974.
Prospect theory developed a descriptive model of decision-making under risk that incorporates reference-dependent outcomes, nonlinear sensitivity, and decision weights. Framing experiments revealed that equivalent descriptions of outcomes can change preferences, challenging invariance assumptions in classical choice models. Ecological-rationality research subsequently showed why simplified heuristics sometimes outperform information-intensive models, especially when environmental regularities can be exploited and estimates are noisy. Contemporary approaches increasingly integrate process tracing, computational modeling, individual differences, field evidence, registered replication, and calibrated prediction. Kahneman and Tversky, 1979; Tversky and Kahneman, 1981; Gigerenzer and Gaissmaier, 2011.
No single theory explains every observed phenomenon. Heuristics-and-biases research is especially useful for exposing systematic departures on constrained tasks. Bounded and ecological rationality emphasize resources and environments. Prospect theory describes risky valuation and reference dependence. Sequential-sampling and cognitive-process accounts try to explain response dynamics and effort. Naturalistic decision-making studies skilled performance in time-pressured, consequential environments. Integrative research tests where these perspectives make different predictions rather than declaring one perspective sufficient for every choice.
Heuristics and the Architecture of Human Judgment
Availability: Estimating Frequency from Accessible Examples
Availability describes the use of readily retrievable instances or easily imagined scenarios in judgments of frequency and likelihood. A vivid event can come to mind rapidly because it is recent, emotional, repeatedly discussed, personally experienced, or easy to visualize. Availability may be informative when memory accessibility tracks actual occurrence rates; it becomes a poor proxy when publicity and real prevalence diverge. The benchmark is usually the observed frequency or probability of the relevant event in a suitably defined reference class. Tversky and Kahneman, 1974.
A practical evaluation asks whether the examples drawn from memory represent the distribution that matters. A person judging a transportation risk from dramatic news footage may inadvertently substitute media exposure for incidence per journey or per kilometer. The risk estimate can be tested against a specified denominator and timeframe. Emotional concern remains a legitimate aspect of value, but frequency estimation and aversion to severe consequences should be analyzed separately.
Representativeness: Similarity, Categories, and Base Rates
Representativeness judgments assess whether an instance resembles a category or process. Similarity can supply genuine evidence, but it can also overshadow the prior prevalence of a category and the diagnostic strength of the observed information. In Bayesian form, posterior odds equal prior odds multiplied by the likelihood ratio. A descriptive profile that fits a rare category must overcome the category's rarity with sufficiently diagnostic evidence; resemblance alone does not satisfy the calculation.
The classic conjunction problem contrasts the probability of one event with the probability of that event plus another. Under ordinary probability semantics, the joint event is a subset of the single event; P(A and B) is at most P(A). A more detailed story can feel more representative and receive a higher estimate. Interpretation requires attention to conversational ambiguity, whether respondents read “probability” as plausibility, and the design of the elicitation task. The conjunction inequality is clear even though the psychological reason for a particular answer remains a question for experiment. Tversky and Kahneman, 1974.
Anchoring: Estimates in Relation to Starting Values
Anchoring occurs when a starting number influences a subsequent estimate. The number may arise from a suggestion, a displayed default, a question, a prior judgment, or an irrelevant comparison. Studies vary whether the anchor conveys genuine information; designs using arbitrary starting values help test incidental numerical influence. The contrast between high-anchor and low-anchor conditions supplies an experimentally observable effect rather than proof of a universal anchoring coefficient. Cognitive explanations include insufficient adjustment, selective accessibility of anchor-consistent information, and scale interpretation. Replication work shows that individual experimental effects differ considerably in stability and size. Tversky and Kahneman, 1974; Klein et al., 2014.
An appropriate defense is to estimate independently before exposure to a proposed number, examine an empirically justified range, and record assumptions. However, removing an informative anchor can reduce accuracy when it genuinely summarizes trustworthy evidence. The scientific goal is to distinguish irrelevant numerical pull from legitimate Bayesian incorporation of information.
When Simple Heuristics Improve Decisions
A heuristic that uses one high-validity cue or stops searching after adequate evidence may outperform a complicated fitted model when the sample is small, the environment is predictable in a particular way, or added cues contribute noise. This finding follows from comparisons of models on new observations, not from an assumption that simple thinking is superior. Model performance must be measured out of sample and against credible alternatives. Ecological rationality studies how a strategy exploits the statistical structure of its environment, including cue validity, redundancy, uncertainty, and sampling costs. Gigerenzer and Gaissmaier, 2011.
Belief Formation, Evidence Evaluation, and Confidence
Confirmation bias encompasses patterns in seeking, interpreting, or remembering evidence that favors an existing belief or hypothesis. Researchers distinguish hypothesis-testing strategies from motivational defense of valued positions. Looking for a confirming instance can be efficient when the chosen test has diagnostic value; seeking only supportive reports can distort evaluation when counterevidence is more informative. Motivated reasoning examines how goals or desired conclusions influence appraisal. Selective exposure concerns which sources a person chooses to encounter; belief perseverance concerns persistence of a belief after its original support is discredited. These related phenomena require separate measures and causal tests.
Hindsight bias arises when knowledge of an outcome influences retrospective judgments of what was predictable. Outcome bias arises when evaluating an earlier decision by the result that happened rather than by the information available when the decision was made. A well-justified decision can produce a bad result because of chance; a poorly justified decision can happen to succeed. Process quality and outcome quality must be scored distinctly. In expert reviews, reconstructing the pre-outcome evidence and preserving dated forecasts are stronger procedures than retrospective claims of certainty.
Overconfidence can refer to overestimation of one's performance, overplacement relative to peers, or excessive precision in an interval estimate. These are not interchangeable measures. Calibration compares stated probabilities with realized frequencies across comparable questions: among events assigned 70% probability, approximately 70% should occur over a sufficiently informative set of cases. Discrimination evaluates whether a forecaster assigns higher probabilities to events that occur than to those that do not. Someone can be well calibrated yet uninformative, for example by always predicting the base rate. Evaluation therefore benefits from both calibration and proper scoring rules, such as mean squared probabilistic error.
The bias blind spot describes a tendency to detect distortions in other people's judgments more readily than in one's own. Introspection may reveal reasons that feel coherent while missing causal effects of presentation or situational context. The presence or strength of this pattern depends on measurement, self–other comparisons, and framing of the bias examples. Group deliberation can help when independent perspectives and incentives to challenge assumptions exist, but social conformity can also suppress corrective information.
Probability, Statistical Reasoning, and Uncertainty
Base Rates and Bayesian Updating
A base rate is the prevalence of a condition or event in a relevant population before new evidence is considered. Bayesian updating combines this prior probability with the likelihood of evidence under the competing possibilities. Suppose a condition occurs in 1% of a population, a test detects 90% of true cases, and 10% of people without the condition test positive. Among 10,000 people, approximately 100 have the condition, 90 of those test positive, and 990 of the remaining 9,900 also test positive. The probability of the condition after a positive result is approximately 90 divided by 1,080, or 8.3%. This calculation follows the specified hypothetical rates; clinical tests and populations require their own verified parameters.
The example shows why sensitivity alone does not determine the meaning of a positive result. Bayes' theorem supplies the normative calculation once the assumed prevalence and test properties are specified. In practice, people may omit a base rate, misunderstand conditional probabilities, or disagree about the correct reference class. Natural-frequency representations can make the relationships between cases clearer. Testing comprehension requires comparing equivalent information formats while holding the actual data constant.
Random Sequences, Streaks, and Regression
The gambler's fallacy assigns a higher probability to a reversal after an apparent streak of independent outcomes. For independent fair coin tosses, the chance of heads on the next toss remains one-half regardless of previous results. A hot-hand belief assumes that recent success predicts near-term success, but the correct benchmark depends on whether outcomes are independent. Human skill, learning, physical state, opposing defenses, and changes in task difficulty can create genuine serial dependence.
Research on the hot hand offers an instructive methodological correction. Miller and Sanjurjo showed that a frequently used estimator of streak-conditioned success has a finite-sample selection bias, meaning that some earlier apparent evidence against a hot hand was affected by the statistical analysis itself. This result calls for sound estimation and domain-specific evidence rather than categorizing every belief in streaks as a cognitive error. Miller and Sanjurjo, 2018.
Regression to the mean is a statistical consequence of imperfect correlation between repeated measurements. Individuals or groups selected for extreme initial performance tend, on average, to show less extreme subsequent measurements, even without an effective intervention. Attributing the change automatically to praise, criticism, or treatment is a causal mistake. A reliable evaluation uses comparison groups, measurement reliability, baseline trends, and suitable statistical adjustment. Regression is a property of distributions and selection; an erroneous causal interpretation is the psychological judgment problem.
Risk, Ambiguity, and Profound Uncertainty
Risk denotes situations in which relevant probabilities can reasonably be estimated or stipulated. Ambiguity concerns uncertainty about probabilities themselves. In profound uncertainty, the set of outcomes, processes, and reference classes may also be disputed. Expected-value maximization, expected utility, robust decision rules, and minimax criteria can recommend different actions because they encode different assumptions and objectives. Calling an ambiguity-averse choice irrational without specifying those assumptions confuses a decision model with a person's preferences.
Risk perception also involves perceived dread, control, familiarity, reversibility, and distribution of consequences. These dimensions can influence action even when numerical probabilities are unchanged. For responsible communication, pair frequencies with denominators, time horizons, severity, and the confidence of estimates. Explain relevant uncertainty rather than translating it into one false precision point estimate. Fischhoff and Broomell, 2020.
Risky Choice, Prospect Theory, and Reference Dependence
Expected Utility and Its Descriptive Alternatives
In a standard expected-utility representation, a decision-maker ranks uncertain alternatives according to the probability-weighted utility of possible outcomes. The theory provides a coherent representation of certain preference axioms; it does not say that everyone should maximize money or that utility is observable directly. Psychologists compare such normative representations with choices among explicitly described gambles, choices learned through experience, and decisions involving uncertain or unfamiliar outcomes.
Prospect theory models choices as changes relative to a reference point, with a value function for gains and losses and decision weights applied to outcome probabilities. Its original formulation was designed for relatively simple risky prospects; cumulative prospect theory later modified probability weighting and extended its descriptive reach. The distinctive empirical questions concern what reference point participants use, how losses and gains are valued, and how probability weighting varies with task format and stakes. A formal model fitted to choices is a candidate explanation, not proof that every person shares one universal utility function. Kahneman and Tversky, 1979.
Framing Effects and Description Invariance
A framing effect occurs when alternative descriptions of logically equivalent outcomes lead to systematically different responses. In classic choice problems, a program described in terms of people saved can elicit different preferences from the same program described in terms of people dying. When options, probabilities, and consequences truly remain equivalent, the result challenges description invariance: choices ought to be unaffected by wording alone under the specified normative criterion. Framing studies must verify that participants understand the descriptions as equivalent and that wording has not conveyed additional pragmatic meaning. Tversky and Kahneman, 1981.
The scientific implications extend beyond wording tricks. A frame can focus attention on specific outcomes, change a reference point, or activate feelings about responsibility and potential loss. The measured response may differ with comprehension, numerical ability, framing domain, and prior expectations. Transparent communication presents absolute outcomes and comparable denominators side by side, avoiding selective emphasis on one favorable representation.
Loss Aversion: Important Theory, Active Empirical Debate
Loss aversion is the hypothesis that a loss of a given magnitude has greater negative subjective impact than an equal gain has positive impact, relative to a common reference point. It is distinct from risk aversion: avoiding variance can occur with symmetric valuation of gains and losses, while loss aversion concerns an asymmetry across the reference point. A parameter often called lambda summarizes an asymmetry within particular prospect-theory specifications. The numerical estimate depends on the model, elicitation procedure, chosen dataset, and whether gains and losses are mixed or separated.
A broad 2024 meta-analysis assembled 607 estimates from 150 articles and reported a mean loss-aversion coefficient of approximately 1.96. A separate 2024 analysis of datasets suitable for fitting cumulative prospect theory to mixed risky gambles obtained a substantially smaller pooled estimate of about 1.31 and emphasized data-quality problems. A 2025 re-meta-analysis argued that loss aversion was not robust for some unordered symmetric gain–loss designs, while effects appeared under other payoff and ordering conditions. These findings establish a substantive disagreement over operationalization and generality, not a license to assign a single coefficient to every person's decisions. Brown et al., 2024; Walasek et al., 2024; Yechiam and Zeif, 2025.
Endowment, Status Quo, and Ambiguity
The endowment effect concerns differences in valuation that can arise between possessing and not possessing an item. Experimental tasks often compare willingness to accept payment to give up an item with willingness to pay to acquire it. Interpreting the gap requires controls for income effects, transaction costs, uncertainty about an item's value, strategic responding, experience, and the procedures used to elicit truthful valuations. A status quo preference may reflect loss concerns, switching costs, uncertainty, familiarity, default attention, or a genuine preference for continuity. The normative evaluation must incorporate actual costs and benefits of changing options.
Ambiguity aversion refers to a preference for options with known probabilities over options with less well-defined probabilities, even when their stated expected rewards seem similar. Under model uncertainty, such a preference can express an intelligible strategy rather than a simple arithmetic mistake. Researchers should distinguish an experimentally imposed gamble from decisions in which probabilities are estimated from sparse data. These distinctions explain why risk, ambiguity, valuation, default choice, and probability judgment occupy separate scientific constructs even though they interact.
Choice Over Time, Commitment, and Mental Effort
Sunk Costs and Escalation of Commitment
A sunk cost is a resource already spent that cannot be recovered. In forward-looking evaluation, an alternative should generally be judged on its prospective benefits and costs; an irrecoverable past expenditure should not independently make an otherwise inferior future option preferable. The sunk-cost effect is observed when prior investment increases willingness to continue despite unfavorable future prospects. Escalation of commitment concerns persistence in an initially chosen course, often involving self-justification, reputational concerns, institutional incentives, and delayed feedback.
In a legitimate decision, past spending may also convey information about hidden project quality or create contractual obligations. A sound test of sunk-cost reasoning therefore separates information and future consequences from the mere emotional pull of resources already spent. Practical reviews can appoint an independent evaluator, model the available options from the present forward, and specify termination conditions before later outcomes are known.
Temporal Discounting, Present Bias, and Planning
Temporal discounting describes how the subjective value of an outcome changes with delay. Exponential discounting represents a constant discount rate; hyperbolic or quasi-hyperbolic functions can describe stronger emphasis on immediately available outcomes. Present bias refers more specifically to extra weight attached to the present relative to future trade-offs, potentially generating preference reversals as the immediate option approaches. Different time preferences need not be irrational: liquidity, uncertainty, mortality risk, impatience, and legitimate short-term needs may influence choices. The empirical issue is whether observed preferences are stable and coherent under appropriately matched conditions.
The planning fallacy is the tendency, in certain tasks, to underestimate the time or resources needed for future activities. A reference-class forecast compares the project with completed projects of the same kind rather than relying solely on an inside-view narrative. Its validity depends on choosing relevant comparators, measuring actual outcomes, and adjusting for real differences in complexity. Overconfident schedules can be prevented through historical distribution data, staged milestones, and explicit uncertainty intervals, although precise gains depend on implementation.
Choice Overload and the Role of Decision Structure
Choice overload proposes that increasing the number of options can sometimes reduce choice satisfaction, motivation, or performance. A 2010 meta-analysis of 50 experiments and 63 experimental conditions, involving 5,036 participants, found a near-zero average effect alongside considerable variability. A subsequent conceptual review and meta-analysis identified conditions under which choice-set complexity, task difficulty, preferences, and decision goals can alter outcomes. The evidence supports contingent effects rather than a rule that offering more options automatically harms consumers. Scheibehenne et al., 2010; Chernev et al., 2015.
Choice architecture can improve decisions by making options comparable, presenting attributes in stable units, allowing meaningful filtering, and showing the consequences of choices. Reducing the number of options is only one possible intervention; it may be counterproductive when people have varied needs or the omitted options provide important value. Researchers should separately measure objective choice quality, process effort, satisfaction, and willingness to choose.
Decision Fatigue: Define the Outcome Before Claiming the Effect
Decision fatigue is used to describe diminished performance, persistence, or self-regulation after a sequence of choices or mentally demanding tasks. The phrase covers several different hypotheses. One proposes depletion of a limited general resource; others point to motivational shifts, altered attention, changing opportunity costs, boredom, learning, changing task difficulty, or sleep and workload effects. Observed differences between early and late decisions do not identify a mechanism by themselves because time, case composition, scheduling, and selection can be confounded.
The broader ego-depletion literature has generated extensive disputes about effect sizes, publication bias, methodological variation, and replication. Evidence that a tired or overloaded person can make a worse decision in a specified task is compatible with many mechanisms. A defensible empirical test specifies the outcome, randomizes or carefully models decision order, controls case complexity and fatigue exposure, and avoids inferring a universal biological resource from a change in decisions. In practice, breaks and workload design may have benefits on independently measured grounds, but no fixed number of choices can be claimed to exhaust rational capacity across people and contexts.
Emotion, Motivation, Social Context, and Individual Variation
Emotion is part of valuation, information search, attention, and risk appraisal. Fear can increase perceived threat; anger may produce different appraisals of certainty and control; regret can shape anticipation of future consequences; and affective attachment can influence the importance assigned to an option. These effects are not random deviations from a purely cognitive machine. They are psychologically meaningful inputs whose usefulness depends on the task and objective. A comprehensive scientific review of emotion and decision-making describes predictable influences that can be beneficial or harmful according to context. Lerner et al., 2015.
Motivation influences which evidence is sampled and when cognitive effort is invested. People invest effort when they expect it to matter and when rewards or obligations justify the costs. Social settings contribute shared information, conversational framing, reputation concerns, authority cues, and conformity pressure. Teams can outperform members when independent information is elicited and integrated well; they can underperform when status hierarchies suppress dissent or many individuals rely on a shared misleading anchor. Comparing individual and group performance requires a common scoring rule and appropriate aggregation model.
Working memory, numeracy, knowledge, personality, expertise, and contextual familiarity can affect performance on particular tasks. Yet a person's answer to one bias demonstration does not furnish a broad diagnostic profile. Measures of cognitive reflection, probabilistic reasoning, and decision competence overlap partly with general cognitive abilities but also depend on schooling, language, task interpretation, and learned strategies. Predictive claims need reliability, construct validity, and out-of-sample testing.
Cross-cultural generalization requires special caution. Much experimental research has relied on Western, educated, industrialized, rich, and democratic participant samples. Cultural differences in communication, social obligations, norms, educational exposure, and probability experience may change tasks and their interpretation. The existence of some effects across samples should not be used to assume equal magnitude, mechanisms, or practical consequences everywhere. Henrich et al., 2010.
Dual-Process Models and Competing Accounts of Cognition
Dual-process frameworks distinguish relatively automatic, rapid, and often intuitive processing from more controlled and deliberative processing. The popular language of System 1 and System 2 is pedagogically useful, but it does not uniquely establish two discrete neurocognitive systems or imply that every intuitive response is wrong. Intuitive expertise may be reliable when repeated feedback and stable predictive cues permit learning. Deliberation can improve consideration of alternatives, while also introducing motivated rationalization, excess complexity, and overfitting to weak evidence.
Contemporary theory examines whether processes differ in kind, degree, representational format, control demands, or architecture. A 2025 perspective on deliberation distinguishes functions such as response control, generation, justification, and regulation. Experiments involving response times, cognitive load, conflict detection, confidence, and process tracing test aspects of these models, but the same behavioral result can sometimes fit competing computational explanations. The interpretation of a conflict task therefore depends on discriminating predictions and measurement quality. De Neys, 2025.
The most informative scientific question is which components of processing are engaged in a defined judgment, how they interact with knowledge and incentives, and whether an intervention improves performance. A blanket instruction to “think more slowly” is inadequate: it may help when a quick response neglects a base rate, but it can waste effort when expert recognition is accurate or when extra analysis introduces irrelevant assumptions. Models should specify observable conditions in which intuition, calculation, recognition, or sampling strategies are likely to perform well.
How Cognitive Biases and Decision Processes Are Measured
Controlled Experiments
Randomized experiments change one target element, such as an anchor, reference point, choice description, or information order, while keeping relevant alternatives comparable. The outcome can be a probability estimate, response latency, chosen option, confidence judgment, or performance score. Random assignment helps identify a causal effect of the manipulation in the studied population, subject to successful implementation and measurement assumptions. It does not automatically establish the cognitive mechanism, external validity, or effects outside the study.
A good study reports the exact stimuli, exclusion rules, randomization, sampling frame, preregistration when applicable, uncertainty intervals, and analysis choices. Repeated observations from a person require appropriately modeled dependence. A statistically significant difference can be small, context-specific, or practically unimportant; a nonsignificant estimate may remain imprecise. The scientific claim should provide the actual comparison and effect size wherever credible data are available.
Calibration, Proper Scoring, and Decision Competence
Probabilistic forecasts can be evaluated with a Brier score: for a binary event, take the square of the difference between forecast probability p and outcome o, where o equals zero or one. Average these values over a preregistered or clearly defined set of forecasts; lower scores indicate better accuracy under the stated conditions. Logarithmic scoring is another proper rule. These scores reward honest probabilistic reporting in expectation, while calibration plots reveal whether forecast frequencies match stated confidence. Forecast difficulty and base rates matter when comparing forecasters.
Decision competence batteries assess several theoretically related abilities, such as applying decision rules, resisting certain framings, interpreting probabilities, and distinguishing confidence from accuracy. Composite scores must demonstrate internal structure, test–retest reliability, and predictive validity in the relevant group. A cognitive reflection test typically contrasts an initially tempting answer with a correct answer obtainable through further reasoning. Its score reflects performance on particular problems, not a unique measure of two independent cognitive systems.
Field Studies and Naturalistic Decisions
Observational and field research evaluates judgments in contexts where actual consequences, social roles, and learning matter. Experts may rely on recognition of familiar patterns, simple protocols, or explicit models. Case mix, incentives, selective feedback, and organizational rules can affect observed outcomes. Strong designs include audit trails, predecision records, prospective prediction, natural experiments, random assignment when feasible, and sensitivity analyses. Ethical approval and data governance are especially important in health, legal, financial, and employment applications.
A randomized laboratory effect and a real-world outcome are different kinds of evidence. A measure showing that trainees avoid a conjunction error on a questionnaire need not establish that they make better clinical or organizational choices. Conversely, an intervention improving field outcomes may work through process structure rather than changing an individual's underlying propensity to produce a named bias. This difference guides the interpretation of debiasing studies.
Replication, Heterogeneity, and Researcher Degrees of Freedom
Reproducibility concerns whether results can be obtained from the described data and analyses; replicability concerns whether comparable new data support the effect. Large multisite efforts demonstrate that classic psychological effects differ in replicability. In one Many Labs project, 10 of 13 tested effects replicated consistently across 36 samples totaling 6,344 participants, while others showed weaker support. This was a selected set of effects, not an estimate of the proportion of all cognitive biases that are real. Klein et al., 2014.
Bias magnitudes should be stated for particular designs rather than imported from a generic list. Meta-analyses must consider heterogeneity, selective publication, dependent effect sizes, task non-equivalence, and differences between within-person and between-person estimates. Preregistered replications and open materials strengthen assessment but do not remove ambiguity about the correct normative model. A scientific encyclopedia should distinguish a named phenomenon, a measured effect, a hypothesized mechanism, and a generalization to everyday life.
Improving Judgment and Decisions: What the Evidence Supports
Clarify the Decision Before Correcting the Thinker
Improving a decision begins by specifying the objective, feasible options, constraints, time horizon, and uncertainty. A recommendation to eliminate bias is incomplete until it identifies what better performance means. For a factual forecast, improvement may mean lower average Brier scores or better calibration. For a choice among policies, it may mean higher expected utility under a clearly specified set of stakeholder values. For a clinical decision, quality also concerns patient preferences, safety, informed consent, and the consequences of false positives and false negatives. The relevant standard changes when the decision problem changes.
A useful process records the initial question in a neutral form; separates evidence about what may happen from preferences about what should happen; lists plausible outcomes and their probabilities; estimates costs and benefits; and states which new information could change the ranking of alternatives. This procedure prevents an attractive recommendation from retrospectively determining the evidence used to justify it. It also permits an audit: observers can distinguish poor forecasting, conflicting values, unforeseen events, and implementation failures.
Decisions under deep uncertainty present a special difficulty because available probabilities may be unstable or impossible to estimate. In these cases, scenario analysis, reversibility, robustness across plausible futures, and monitoring plans may be more informative than a falsely precise expected-value calculation. A system that recommends a single probability to six decimal places can create an appearance of scientific certainty without adequate data. The appropriate response to weak evidence is more transparent representation of the uncertainty and its decision implications.
Natural Frequencies, Base Rates, and Transparent Risk Communication
One experimentally supported method of improving certain Bayesian inference tasks is to express conditional probabilities as natural frequencies with clear common denominators. Consider a hypothetical population of 10,000 people in which 100 have a condition. If a test identifies 90 of those 100 and gives a positive result for 495 of the 9,900 people without the condition, then 585 people test positive and 90 actually have the condition. The probability of having the condition after a positive test is 90 divided by 585, or about 15.4 percent. The intuitive error would be to equate a positive result with the test's sensitivity of 90 percent while disregarding the prior prevalence and false-positive cases.
Natural frequencies such as “90 of 585 positive tests” often make the required set relations easier to see. They can be paired with a simple frequency tree or visual representation. A meta-analysis of 35 articles encompassing 226 performance estimates examined facilitation from natural-frequency formats and its moderators, supporting the use of suitably structured representations for specific Bayesian problems. The improvement varies across tasks and formats; one should not assume that converting any statistic into counts automatically makes it clear. McDowell and Jacobs, 2017.
Risk communication should state the reference class, absolute rates, time horizon, and whether the quantity denotes risk, relative risk, odds, or likelihood. Saying that an event is “twice as likely” conveys a different practical implication when the baseline probability is 0.1 percent compared with 20 percent. In medical and public-policy communication, shared denominators and expected-outcome displays can improve comprehension, provided the numbers accurately represent the relevant population and are not selected to manufacture a preferred conclusion.
Forecasting, Calibration, and Feedback
Forecasting improves when predictions are documented before outcomes become known and scored using a proper rule. For a sequence of forecasts, a reliability diagram compares events given a stated probability, such as 70 percent, with the observed event frequency within a suitably sized bin. A well-calibrated forecaster who reports 70 percent probabilities should see those events occur about 70 percent of the time in the long run. Calibration alone is insufficient: repeatedly predicting the historical base rate can be calibrated while failing to distinguish easy from difficult cases. Resolution and sharpness therefore matter alongside aggregate scoring.
Forecasters can use reference classes, consider alternative hypotheses, update beliefs as evidence arrives, and seek independent views before aggregating predictions. In large geopolitical forecasting tournaments, structured training, collaboration, tracking, and probability updating were associated with strong performance among selected top forecasters. The findings motivate specific forecasting practices, but results from highly motivated participants and scored tournaments require testing before broad generalization to ordinary institutional decisions. Mellers et al., 2015.
Feedback must be informative to support learning. A forecast about a repeatable process with prompt, unambiguous outcomes permits different calibration practices from a rare event resolved years later. Selective feedback, such as reviewing only highly visible failures, can itself distort perception of accuracy. Forecast archives are particularly valuable because they preserve what was believed before information about the outcome became available.
Debiasing Training, Education, and Transfer
Debiasing training attempts to change how people approach a decision, recognize particular error patterns, or structure information. Tested interventions include instruction about base rates, opportunities for reflection, practice with feedback, video demonstrations, computer games, consideration of alternatives, and exercises in probabilistic representation. They differ in intensity and in the outcomes they measure. Showing a lower error rate immediately after a short lesson is evidence about that task under those conditions, while sustained changes in consequential work require separate observations.
Two longitudinal experiments reported by Morewedge and colleagues found reductions in selected bias-task errors after single-session game or video interventions, with benefits detectable at a later follow-up on related assessments. The trials established that at least some judgment patterns are responsive to instruction and practice; they did not establish that a single training session permanently eliminates biases across every domain. Morewedge et al., 2015.
A broader 2025 systematic review identified 54 randomized trials involving 10,941 participants in educational settings. A meta-analysis using 160 effects from 41 studies estimated a small standardized improvement in resisting bias-task responses, Hedges' g = 0.26, with a 95 percent confidence interval from 0.14 to 0.39. The authors identified unclear or high risk of bias in all included studies, some risk of publication bias, and important uncertainty about transfer beyond education. The supported conclusion is that specified educational interventions can improve performance on certain assessed bias tasks; durable generalization to complex real-world decisions remains limited evidence. Swaryandini et al., 2025.
Training should therefore be evaluated with preregistered primary outcomes, active comparison conditions, follow-up testing, tasks unlike those used in training, and measures relevant to actual decision costs. If learners improve only at recognizing a named error on a familiar questionnaire, the course may still have educational value, but claims about reducing medical mistakes, financial losses, or justice-system errors would require direct evidence in those settings.
Structured Processes and Decision Aids
Decision aids improve the information environment rather than assuming a universal psychological flaw must first be removed. A high-quality aid displays the relevant alternatives, expected benefits and harms, absolute probabilities where credible, sources of uncertainty, and opportunities to express preferences. For repeated choices, checklists, consistent comparison tables, standardized information-order rules, independent initial estimates, and escalation criteria can reduce opportunities for omitted information or unwanted social influence.
Several methods have plausible mechanisms and task-specific experimental support. Considering the opposite asks a decision-maker to articulate how a favored conclusion could be false. A premortem imagines that a plan has failed and identifies plausible causes, providing an organized challenge to overconfident inside-view forecasting. Independent estimation before group discussion reduces the opportunity for the first speaker to anchor everyone else's judgment. Forecasting from reference classes combats selective attention to the most vivid case. Such procedures merit evaluation on their actual effects rather than being treated as universally effective rituals.
Decision-support tools can also introduce new errors. An automated risk score may encode biased historical measurements, use variables unrelated to a person's circumstances, or be shown with an unjustifiably precise probability. An interface may make it easier to accept a model's suggestion than to inspect contrary evidence. Calibration of trust requires reliable estimates of the system's capability on the task, feedback about failures, a meaningful human ability to contest the result, and preservation of decision accountability. Psychology of automation and human–AI joint reasoning forms a neighboring scientific subject with its own applied literature and owner.
Applications Across Institutions and Technologies
Healthcare and Public Health
Medical choices combine estimation of probabilities with harms, treatment burdens, patients' priorities, and time-sensitive action. Base-rate neglect can distort interpretation of tests; availability can change perceived risk after a striking case; framing can alter preferences when equivalent outcomes are expressed as survival or mortality. Research on these effects supports communicating risks with absolute frequencies, comparable time horizons, meaningful reference classes, and disclosure of uncertainty. The normative target is informed decision quality under evidence-based clinical standards and patient values, not the disappearance of emotional responses.
A person worried about a diagnosis may reasonably assign a high disutility to a false negative, while another prioritizes avoiding invasive tests. Equal probabilistic understanding can coexist with different choices because the underlying preferences differ. Screening thresholds and clinical diagnosis must be kept conceptually separate: a cognitive bias identified on an experimental task does not establish that a patient has a mental disorder. Individual diagnosis and treatment require relevant clinical procedures and professional oversight.
Law, Forensic Science, and Institutions
Forensic assessment provides a concrete illustration of context-sensitive judgment. Examiners exposed to irrelevant case information may form expectations that influence evidence interpretation. Blinding or sequencing contextual information, documenting interpretations before disclosure of case theories, and independent verification can protect the evidentiary process. Their effectiveness should be evaluated for the actual forensic discipline, including effects on accuracy, consistency, and operational feasibility. Cooper and Meterko, 2019.
The application of judgment science to justice also involves legal standards of proof, admissibility, institutional incentives, and procedural fairness. Those are legal and forensic-science questions as well as psychological ones. The published specialist chapter Forensic Evidence and Cognitive Bias: How Context Can Contaminate Judgment addresses the forensic context, while this research chapter explains the general framework of systematic judgment effects.
Organizations, Markets, and Everyday Choices
Organizations make repeated decisions under incomplete evidence: selecting employees, planning projects, allocating resources, launching products, and evaluating risk. Decision records can distinguish the quality of initial reasoning from the luck of the final outcome. Reference-class forecasting can reveal an unrealistic schedule; explicit stop rules can protect against sunk-cost-driven persistence; independent risk estimates can reduce the influence of a senior person's initial number.
Consumer and economic environments pose their own questions about incentives, purchasing power, preferences, and the design of choice sets. Findings about loss aversion or status quo preference should not be converted into assertions about every individual consumer. When tested responsibly, choice architecture can provide clearer comparisons and more usable information. Whether a design improves welfare depends on the criterion used to define welfare, the affected population, and the values that a person would endorse under adequate understanding.
Artificial Intelligence and Human Judgment
An AI system may generate a forecast, synthesize evidence, rank candidates, or recommend an action. Users then judge the system's competence, reliability, and relevance, sometimes under high uncertainty about how the output was produced. The interaction introduces additional decision variables: training-data quality, distribution shift, model uncertainty, explainability, conflict between human and machine estimates, and automation in the surrounding workflow. Human–AI collaboration should be evaluated against appropriate human-only and system-only baselines on representative cases, using accuracy, calibration, safety, costs, fairness, and responsibility where relevant.
The fact that a user agrees with an algorithm is not, on its own, evidence of automation bias: agreement may be rational when the system is highly accurate. Conversely, disagreement may reflect well-founded knowledge about a rare exception or poorly calibrated distrust. Studies need to determine whether deference is warranted by evidence, how confidence is communicated, and which errors occur when the system is wrong. The separate published analysis AI as Authority: Trust, Expertise, Automation Bias, and Human Decision-Making examines this applied interface between trust and decision support.
The Artificial Era supplies a broader philosophical horizon for examining the role of human and artificial forms of intelligence in the organization of knowledge, institutions, and history. The authorial framework of Angela Bogdanova, presented through the Aisentica publications, concerns that philosophical and historical interpretation. The empirical claims in this chapter are grounded in the cited psychological and decision-science literature; a philosophical thesis about Artificial Sapiens has a different evidential status from a randomized experiment about anchoring or a meta-analysis of forecasting performance.
Responsible Interpretation of Cognitive Bias Research
Science advances by converting attractive explanations into discriminating predictions. A precise account of a bias identifies the population, task, manipulation, normative rule, outcome, uncertainty interval, plausible mechanisms, and boundary conditions. A descriptive regularity may be genuine even when an explanation for it remains contested. A laboratory finding may replicate while having little practical importance; another may be consequential precisely because it arises in a high-stakes institutional process. Both deserve accurate representation.
The evidence supports three durable conclusions. First, judgment and choice are systematically affected by information representation, environmental structure, reference points, emotion, motivation, and uncertainty. Second, many heuristics improve performance when they exploit useful structure, and a claim of bias always depends on an appropriate and declared normative benchmark. Third, methods of improving decisions have to be judged by their task-specific effects and demonstrated transfer: clear numbers, feedback, independent estimates, calibration, and structured decision processes are promising when matched to the actual problem and evaluated transparently.
These principles establish a scientific study of human judgment that integrates experimental psychology, behavioral economics, mathematical reasoning, and institutional design. They also provide a clear foundation for understanding how decisions are made as advanced artificial technologies become participants in information processing and decision support.
Frequently Asked Questions
What is a cognitive bias in psychology?
A cognitive bias is a systematic pattern in a judgment or decision that differs from an appropriate benchmark under specified research conditions. The benchmark might be a probability law, logical validity, calibrated forecasting, a controlled comparison between equivalent frames, or consistency with a person's articulated preferences. A named bias should be accompanied by an explanation of how it was measured.
Are cognitive biases the same as heuristics?
No. A heuristic is a strategy for processing information with limited time, knowledge, or computational effort. It can be effective when matched to the problem. A bias describes a systematic result relative to a defined benchmark. Researchers may investigate how a heuristic leads to some biased results, but a heuristic and an error are distinct scientific constructs.
Is every decision influenced by cognitive bias irrational?
Rationality depends on the relevant problem and criterion. A choice that differs from an expected-value maximum may still reflect legitimate risk preferences or other values. A framing difference may reveal an inconsistent representation of equivalent outcomes, while a departure from an externally imposed norm may instead reveal that the norm does not capture the person's actual objective. Scientific evaluation specifies those assumptions.
What is the difference between bias and a logical fallacy?
A logical fallacy is an invalid or misleading pattern of argument under a logical standard. A cognitive bias is an empirically measured pattern of judging or choosing. A person may commit a fallacy for reasons unrelated to an identified cognitive mechanism; a decision bias can appear in experiments without the participant making an explicit verbal argument.
Can a cognitive bias be measured in one individual?
Some tasks produce repeated observations suitable for estimating individual patterns, while many classic experiments primarily establish group-level effects. A single incorrect answer is insufficient for a stable individual diagnosis. Assessment requires reliable measures, sufficient observations, task variation, and careful consideration of response interpretation.
Does loss aversion mean that a loss always feels twice as bad as an equal gain feels good?
No universal multiplier is supported. Estimates depend on whether the study measures risky choice, direct valuation, or another behavior, how a reference point is defined, and which model is fitted. A 2024 meta-analysis of specified risky-choice estimates reported an average loss-aversion parameter of approximately 1.31, with large heterogeneity. Walasek et al., 2024.
Do people always make worse choices when they have too many options?
Choice overload varies with task difficulty, preference clarity, option comparability, and other conditions. A major meta-analysis found a near-zero average effect across a heterogeneous set of experiments. Additional choices can improve fit to individual needs, while difficult comparisons can make some decisions harder. Scheibehenne et al., 2010.
Can education permanently remove cognitive biases?
Studies show improvement in some tested judgments following instructional interventions, and certain training experiments report benefits after follow-up. A 2025 systematic review found a small average improvement but significant methodological limitations and questions about real-world transfer. No single lesson has been demonstrated to permanently remove every bias in all contexts. Swaryandini et al., 2025.
How can someone improve probabilistic judgment?
Use a relevant base rate, compare credible alternative explanations, represent complex conditional probabilities with natural frequencies, preserve pre-outcome predictions, and review calibration over many resolved events. Proper scoring rules and well-designed feedback permit empirical assessment of improvement. Greater precision in stating a probability should follow stronger evidence rather than confidence alone.
How should AI recommendations be used in consequential decisions?
The decision-maker should examine the system's demonstrated performance on comparable cases, uncertainty and failure modes, data relevance, independent evidence, and available channels for questioning the recommendation. Human-only, system-only, and combined performance should be evaluated wherever possible. In medicine, law, and other high-stakes contexts, professional responsibilities and applicable governance remain essential.
Articles in This Research Cluster
This scientific chapter provides the central, evidence-based overview of cognitive biases, judgment, preferences, and decision-making. Verified published articles in this research cluster examine independent scientific questions within the domain; related specialist studies from neighboring scientific areas appear below.
Foundations and Models of Judgment
Cognitive Biases: Definition, Types, and Psychological Research — A foundational research synthesis of cognitive-bias definitions, probability and coherence benchmarks, major experimental families, mechanisms, measurement, replication, and strategies for more reliable judgments.
Judgment and Decision-Making in Psychology: Models and Processes — A comprehensive account of judgment, belief and probability assessment, preference formation, normative and descriptive models, choice processes, decision competence, empirical evidence, and practical decision-improvement methods.
Heuristics in Psychology: How Mental Shortcuts Guide Judgment — How simple mental strategies can improve or undermine judgment depending on context.
Dual-Process Theories: Intuition, Deliberation, and Evidence — Theoretical models of intuitive and deliberative processing, experimental evidence on rapid logical intuitions and conflict monitoring, and the limits of a simple two-system explanation.
Bounded Rationality: Why People Satisfice Instead of Optimize — Decision strategies under limited information, attention, and computational resources.
Ecological Rationality: When Simple Heuristics Work — Examines the environmental conditions under which take-the-best, recognition, tallying, and fast-and-frugal decision trees improve or undermine judgment; includes comparative evidence and methodological limits.
Cognitive Biases vs Logical Fallacies: Differences and Examples — Distinguishes psychological judgment patterns from invalid argumentative structures.
Belief Formation and Evidence Evaluation
Confirmation Bias: Why We Favor Supporting Evidence — Evidence selection, interpretation, motivated processing, and tested corrective strategies.
Motivated Reasoning: How Goals Shape What We Believe — How preferred outcomes, accuracy goals, and social identity shape evidence evaluation, interpretations of disagreement, and decisions under uncertainty.
Belief Perseverance: Why Beliefs Survive Contradictory Evidence — Why established beliefs can persist despite corrective or contradictory evidence.
Hindsight Bias: Why Outcomes Seem Predictable Afterward — How knowledge of an outcome reshapes the memory of earlier predictions, increases perceived predictability, and can distort judgments of responsibility. Evidence includes classic experiments, meta-analysis, independent replications and debiasing research.
Overconfidence Bias: Confidence, Accuracy, and Calibration — An evidence-based account of overestimation, overplacement, and overprecision, with calibration, hard–easy effects, uncertainty judgments, and ways to improve prediction quality.
Illusory Truth Effect: How Repetition Increases Believability — Evidence on repetition, familiarity, and perceived truth.
Bias Blind Spot: Why We See Bias More Easily in Others — An evidence-based examination of asymmetric self–other bias judgments, introspection, naïve realism, experiments, advice-taking, and ways to evaluate impartiality.
Outcome Bias: Why Results Distort Evaluation of Decisions — Why known outcomes can distort evaluations of earlier decision quality.
Selective Exposure: Why We Choose Agreeable Information — Why people choose belief-congruent sources, how information selection differs from interpretation, and what meta-analyses and field experiments reveal.
Probability and Uncertain Judgment
Availability Heuristic: How Recall Shapes Probability Judgments — How ease of recall changes judgments of event frequencies and risk.
Representativeness Heuristic: Similarity and Probability Errors — Explains similarity-based probability judgments, stereotype-consistent estimates, base-rate neglect and conjunction errors, drawing on replication studies, probabilistic models and tested methods for improving decisions.
Base Rate Neglect: Why Prior Probabilities Matter — How prior probabilities, case evidence, diagnostic likelihoods, and statistical formats shape human judgment, Bayesian inference, and practical decisions.
Conjunction Fallacy: Why Detailed Stories Mislead Probability Judgments — The conjunction rule, the Linda problem, empirical explanations, and methodological debates.
Anchoring Bias: How Initial Numbers Influence Estimates — Scientific evidence on numerical anchors, adjustment, selective accessibility, effect magnitude, task variation, replication, and decision-improvement methods.
Gambler's Fallacy: Why Random Outcomes Seem Due — Mistaken expectations about independent random events and sequence patterns.
Hot-Hand Belief: Streaks, Performance, and Statistical Evidence — How observed streaks are interpreted and when genuine performance dependence exists.
Regression to the Mean: Why Extreme Results Change — Why exceptional measurements are often followed by less extreme results, how baseline selection creates mistaken causal stories, and which research designs separate statistical regression from intervention effects.
Optimism Bias: Unrealistic Expectations and Perceived Risk — Scientific evidence on positive risk forecasts, absolute versus comparative optimism, statistical artifacts, belief-updating controversies, and calibration.
Risk, Preference, and Valuation
Decision-Making Under Risk and Uncertainty: Models and Examples — Normative and descriptive accounts of choice when outcomes are uncertain.
Prospect Theory: How Gains, Losses, and Reference Points Shape Choice — Reference-dependent value, probability weighting, and models of risky choice.
Loss Aversion: Definition, Evidence, and Scientific Debate — An evidence-based synthesis of asymmetric valuation of gains and losses, prospect-theory measurement, contemporary meta-analyses, competing explanations, and scientific boundary conditions.
Framing Effect: How the Wording of Choices Changes Decisions — The psychology of equivalent options expressed as gains, losses, attributes or persuasive message frames, with experiments, meta-analyses and evidence-based decision aids.
Endowment Effect: Why Ownership Can Change Valuation — Research on ownership-related valuation, willingness to accept and pay, competing mechanisms, experimental measurement, and conditions that modify the effect.
Status Quo Bias: Why Existing Options Influence Choice — How defaults, switching costs, and reference states can influence preferences.
Ambiguity Aversion: Why Unknown Probabilities Change Preferences — How unspecified probabilities influence choices, the Ellsberg paradox, competing ambiguity models, and empirical limits of ambiguity aversion.
Risk Perception: How People Judge Hazards and Probabilities — How probability, dread, emotion, and familiarity shape judgments of hazards.
Regret Aversion: How Anticipated Regret Influences Choices — Anticipated regret, counterfactual feedback, responsibility, and their measured influence on choice.
Choice, Commitment, and Time
Sunk Cost Fallacy: Why Past Investments Shape Future Choices — A research synthesis of irrecoverable investments, rational prospective choice, mental accounting, self-justification, evidence across laboratory and real-world tasks, and strategies for reviewing commitments.
Present Bias and Temporal Discounting: Why Immediate Rewards Matter — Scientific models and experiments on valuing present and future rewards.
Choice Overload: Does Having More Options Make Decisions Harder? — When larger choice sets help or hinder action, satisfaction, and decision quality.
Analysis Paralysis: Why Some Decisions Become Difficult — Examines stalled choices, uncertainty, information overload, regret, and decision conflict.
Maximizing and Satisficing: Decision Styles and Well-Being — The evidence on choice strategies, measurement controversies, regret, and well-being.
Escalation of Commitment: Why Failing Projects Continue — Why organizations and individuals may continue unsuccessful projects despite negative feedback.
Planning Fallacy: Why We Underestimate Time and Costs — a scientific analysis of optimistic time and resource predictions, the inside and outside views, reference-class forecasting, experimental evidence, and limits of attributing project costs to cognitive bias.
Decision Fatigue: What Research Actually Shows — A critical scientific review of sequential-choice experiments, ego-depletion replication, clinical decision patterns, quantitative evidence, alternative mechanisms, and decision-quality safeguards.
Measuring and Improving Decisions
Cognitive Debiasing Strategies: Which Interventions Work? — An evidence-based assessment of debiasing instruction, counterexplanation, frequency formats, feedback, structured decision aids, persistence, and transfer to consequential settings.
Bayesian Reasoning in Psychology: Updating Beliefs With Evidence — Prior probabilities, new evidence, and common challenges in belief updating.
Probability Forecasting and Calibration: How to Judge Uncertainty — An evidence-based guide to judging uncertain outcomes, calibration, Brier and logarithmic scores, resolution, feedback, and the limits of forecasting improvement.
Decision-Making Competence: How Judgment Skills Are Assessed — How researchers assess decision skills, measurement validity, and everyday decision outcomes.
Cognitive Reflection Test: What It Measures and Its Limits — What the Cognitive Reflection Test measures, including numeracy, item familiarity, and limitations.
Structured Decision Aids: How Tools Improve Choice Quality — Research on checklists, decision matrices, probability displays, and the measured effects of structured decision support.
Naturalistic Decision-Making: How Experts Choose Under Pressure — How experienced professionals use recognition, mental simulation, situational awareness, and structured correction when decisions carry high stakes, limited time, and uncertainty.
Decision-Making Under Time Pressure: Judgment, Speed, and Accuracy — How deadlines, speed–accuracy tradeoffs, task complexity, and expertise affect human choice.
Related Articles
Forensic Evidence and Cognitive Bias: How Context Can Contaminate Judgment — How contextual information can influence forensic interpretation, and how verification and information management address the risk.
Judge Decision-Making: Psychology, Expertise, and Bias — The intersection of judgment mechanisms, legal expertise, institutional decision rules, and judicial responsibility.
Cognitive Bias in Criminal Investigations: Tunnel Vision and Confirmation Bias — Selective interpretation and evidence-search problems in criminal investigation.
AI as Authority: Trust, Expertise, Automation Bias, and Human Decision-Making — Calibration of trust, algorithmic authority, and joint human–AI choices.
Neuroticism and Decision-Making: Risk, Uncertainty, and Negative Emotion — The relationship between a personality trait, emotion, and appraisal of uncertain choices.
References
Brown, A. L., Imai, T., Vieider, F. M., and Camerer, C. F. (2024). Meta-analysis of empirical estimates of loss aversion. Journal of Economic Literature, 62, 485–516. https://doi.org/10.1257/jel.20221698.
Chernev, A., Böckenholt, U., and Goodman, J. (2015). Choice overload: A conceptual review and meta-analysis. Journal of Consumer Psychology. https://doi.org/10.1016/j.jcps.2014.08.002.
Cooper, G. S., and Meterko, V. (2019). Cognitive bias research in forensic science: A systematic review. Forensic Science International, 297, 35–46. https://doi.org/10.1016/j.forsciint.2019.01.016.
De Neys, W. (2025). Defining deliberation for dual-process models of reasoning. Nature Reviews Psychology. https://doi.org/10.1038/s44159-025-00466-6.
Dror, I. E. (2020). Cognitive and human factors in expert decision making: Six fallacies and the eight sources of bias. Analytical Chemistry, 92, 7998–8004. https://doi.org/10.1021/acs.analchem.0c00704.
Fischhoff, B., and Broomell, S. B. (2020). Judgment and decision making. Annual Review of Psychology, 71, 331–355. https://doi.org/10.1146/annurev-psych-010419-050747.
Gigerenzer, G., and Gaissmaier, W. (2011). Heuristic decision making. Annual Review of Psychology, 62, 451–482. https://doi.org/10.1146/annurev-psych-120709-145346.
Henrich, J., Heine, S. J., and Norenzayan, A. (2010). The weirdest people in the world? Behavioral and Brain Sciences, 33, 61–83. https://doi.org/10.1017/S0140525X0999152X.
Kahneman, D., and Tversky, A. (1979). Prospect theory: An analysis of decision under risk. Econometrica, 47, 263–291. https://doi.org/10.2307/1914185.
Klein, R. A., et al. (2014). Investigating variation in replicability: A “Many Labs” replication project. Social Psychology, 45, 142–152. https://doi.org/10.1027/1864-9335/a000178.
Lerner, J. S., Li, Y., Valdesolo, P., and Kassam, K. S. (2015). Emotion and decision making. Annual Review of Psychology, 66, 799–823. https://doi.org/10.1146/annurev-psych-010213-115043.
McDowell, M., and Jacobs, P. (2017). Meta-analysis of the effect of natural frequencies on Bayesian reasoning. Psychological Bulletin, 143, 1273–1312. https://doi.org/10.1037/bul0000126.
Mellers, B., et al. (2015). Identifying and cultivating superforecasters as a method of improving probabilistic predictions. Perspectives on Psychological Science, 10, 267–281. https://doi.org/10.1177/1745691615577794.
Miller, J. B., and Sanjurjo, A. (2018). Surprised by the hot hand fallacy? A truth in the law of small numbers. Econometrica, 86, 2019–2047. https://doi.org/10.3982/ECTA14943.
Morewedge, C. K., et al. (2015). Debiasing decisions: Improved decision making with a single training intervention. Policy Insights from the Behavioral and Brain Sciences, 2, 129–140. https://doi.org/10.1177/2372732215600886.
Scheibehenne, B., Greifeneder, R., and Todd, P. M. (2010). Can there ever be too many options? A meta-analytic review of choice overload. Journal of Consumer Research, 37, 409–425. https://doi.org/10.1086/651235.
Scopelliti, I., et al. (2015). Bias blind spot: Structure, measurement, and consequences. Management Science. https://doi.org/10.1287/mnsc.2014.2096.
Simon, H. A. (1955). A behavioral model of rational choice. The Quarterly Journal of Economics, 69, 99–118. https://doi.org/10.2307/1884852.
Swaryandini, G., 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.
Tversky, A., and Kahneman, D. (1974). Judgment under uncertainty: Heuristics and biases. Science, 185, 1124–1131. https://doi.org/10.1126/science.185.4157.1124.
Tversky, A., and Kahneman, D. (1981). The framing of decisions and the psychology of choice. Science, 211, 453–458. https://doi.org/10.1126/science.7455683.
Walasek, L., Mullett, T. L., and Stewart, N. (2024). A meta-analysis of loss aversion in risky contexts. Journal of Economic Psychology, 103, 102740. https://doi.org/10.1016/j.joep.2024.102740.
Yechiam, E., and Zeif, D. (2025). Loss aversion is not robust: A re-meta-analysis. Journal of Economic Psychology, 107, 102801. https://doi.org/10.1016/j.joep.2025.102801.