SCIENCE
Experimental Economics
Abstract. Experimental economics investigates economic behavior and institutions by deliberately varying decision environments and observing their consequences. Its methodological center is the controlled comparison: a researcher specifies an allocation rule, information structure, incentive system, interaction protocol, or policy treatment; documents how units encounter the intervention; and evaluates outcomes against explicit counterfactual and theoretical benchmarks. The discipline extends from individual choice under risk to bilateral bargaining, public goods, auctions, asset markets, coordination, organizational incentives, and large-scale field evaluations. This monograph reconstructs its historical formation, conceptual apparatus, principal research traditions, empirical findings, experimental design, identification strategies, statistical practice, ethical obligations, replication standards, and limits of generalization. It distinguishes evidence for a causal contrast from evidence for a behavioral mechanism and from predictions about other populations and institutions. Foundational market experiments demonstrate the importance of trading rules and experience; bargaining and cooperation experiments reveal structured heterogeneity in social behavior; randomized field studies identify consequential effects in defined contexts. The original synthesis integrates these findings through a transparent comparison of design fidelity, estimand, measurement, mechanism, uncertainty, and transportability. The synthesis is an interpretation of established scholarship, and introduces no new independent theory or doctrine.
Keywords: Experimental economics; laboratory experiments; field experiments; randomized controlled trials; induced value; incentive compatibility; economic games; experimental markets; auctions; bargaining; public goods; social preferences; causal inference; treatment effects; preregistration; replication; external validity; research ethics
1. Scope, Epistemic Aims, and Disciplinary Boundaries
Experimental economics is the systematic investigation of economic decisions, interactions, and institutions under research-designed variation. An experiment is defined by the intervention and assignment mechanism, the set of possible outcomes, the relevant units, the measurement protocol, and the intended comparison. Physical location is secondary. A computerized market at a university, a randomized pricing treatment in a retail platform, an interactive bargaining game conducted online, and a public program evaluated by lottery can all be economic experiments. The distinction rests on the investigator's leverage over the relevant variation and the credibility of the resulting counterfactual.
The discipline asks several different kinds of questions. Positive questions concern what people choose and how groups coordinate. Explanatory questions ask which informational, psychological, or institutional processes produce observed choices. Institutional questions concern the comparative performance of trading rules, matching systems, voting procedures, and market organizations. Policy questions ask how an intervention changes outcomes for a population whose exposure can be evaluated. Methodological questions address when a measured experimental effect is causally identified, precisely estimated, reproducible, ethically obtained, and relevant elsewhere. The Handbook of Experimental Economics documents the discipline's breadth across public goods, markets, coordination, bargaining, auctions, and individual choice.
Experimental economics is in continuous exchange with microeconomic theory, game theory, econometrics, behavioral economics, development economics, political economy, organizational economics, and psychology. It maintains a distinctive disciplinary problem: designing decision situations in which specific economic propositions acquire empirically discriminating consequences. A game-theoretic equilibrium may organize an experiment without uniquely predicting individual actions; a treatment effect may be identified without resolving the psychological process producing it. Both findings have scientific content when their evidential scope is stated accurately.
Three products of an experiment must be kept conceptually separate. The descriptive product records distributions, trajectories, and realized institutions. The causal product compares potential outcomes under alternatives, subject to assumptions about assignment and interference. The explanatory product links behavior to mechanisms, usually by manipulating additional channels, varying environments, or evaluating competing structural interpretations. The distinction prevents a statistically precise treatment effect from being treated as a complete theory of economic conduct.
2. The Historical Constitution of Experimental Economics
Formal experiment did not appear suddenly with modern behavioral economics. Intellectual precursors include the use of choice problems to examine utility and uncertainty, classroom and laboratory demonstrations of exchange, and experimental studies of strategic interaction. Maurice Allais's 1953 study of rational choice under risk challenged the empirical and normative reach of certain expected-utility axioms through structured choice comparisons. The article is a historical source for the study of elicited risky preferences; its challenge to an axiom system and the interpretation of any particular respondent's choice require separate analysis.
Edward Chamberlin's 1948 experimental imperfect market is a central episode in the experimental investigation of exchange. Its controlled trading environment brought market theory into observational contact with participant conduct and revealed the importance of trading organization. Vernon Smith's 1962 competitive market study subsequently demonstrated that repeated double-auction institutions could generate substantial convergence toward competitive predictions under specified conditions, even when individual participants had limited information. This result concerns organized exchange under the experiment's rules; it provides neither a universal efficiency theorem nor a claim that every market must converge.
Smith's 1976 induced value theory articulated how appropriate monetary rewards and controlled cost or value schedules could help separate the experimenter's target incentives from other influences. The methodology made it practical to construct laboratory demand and supply conditions rather than simply solicit stated opinions. Charles Plott's 1982 industrial organization survey and Alvin Roth's 1988 methodological overview document the transformation of experimentation from an occasional demonstration into a sustained research method addressing institutions, bargaining, and the evaluation of theory.
The late twentieth century expanded the subject from market clearing to negotiated allocations, cooperation, social preferences, information, and departures from simple expected-payoff predictions. The 1995 handbook consolidated diverse strands. Field experimentation subsequently increased the share of studies conducted in naturally occurring markets and policy environments. A central lesson from this history is cumulative: experimental rules generate observable consequences, while interpretation depends on explicit behavioral hypotheses and institutional constraints. The modern discipline inherits both the controlled-market tradition and the randomized evaluation of real-world interventions.
3. The Architecture of Economic Experiments
An experimental protocol consists of a research question, a theoretical contrast, a unit of randomization, a unit of observation, eligibility and recruitment rules, treatment conditions, instructions, incentives, timing, outcomes, statistical estimands, and an analysis plan. Each component can change the meaning of the result. In particular, the treatment must be operationally defined as the difference between conditions, while outcomes must be measured identically or with justified comparability across groups.
Between-subject designs assign different participants or independent groups to distinct conditions; within-subject designs expose the same participant to multiple conditions. The former minimizes certain carryover effects and the latter can gain precision from person-specific comparisons, but learning, anchoring, fatigue, preference construction, and order effects may change the estimand. Charness, Gneezy, and Kuhn's comparison of design strategies shows why the appropriate choice depends on the substantive mechanism and the experiment's practical implementation. Counterbalancing sequence can mitigate some order concerns; it does not automatically eliminate strategic memory.
A decision unit may be an individual, a pair, a stable group, a trading market, a school, a village, or a platform cluster. In interactive games, actions of one participant affect others' outcomes, so isolated individual observations are frequently dependent. Matching protocols, partner rematching, anonymous identification, common information, feedback, and repeated rounds must be recorded because these arrangements shape learning and reputational incentives. For repeated markets, the independent replicates may be markets or sessions rather than every price quote or trade.
Factorial experiments cross multiple interventions and can identify interactions when sufficient independent assignment and power are present. A 2-by-2 design may vary, for example, information availability and communication. A causal interaction is the difference between treatment effects across the levels of the second factor. It is a property of the chosen outcome scale and distribution; numerical interaction on a linear scale need not persist on a nonlinear link scale. Sequential experiments change rules or introduce shocks over time, requiring identification of carryover, anticipation, and dynamic selection.
Pilot testing has a substantive scientific role: comprehension checks, usability, payment logistics, treatment differentiation, dropout, and the stability of data capture can expose flaws before the confirmatory sample is recruited. Pilot data may inform planning, provided that any reuse in a confirmatory test is declared and analyzed with an appropriate design. Transparent instructions, experimental scripts, assignment logs, and software versions are necessary for independent reproduction.
4. Causal Estimands, Randomization, and Identification
Let Y_i(1) be unit i's outcome under treatment and Y_i(0) its outcome under control. The unit-level causal effect is τ_i = Y_i(1) − Y_i(0). Only one potential outcome is observed for a given unit in an ordinary parallel-group experiment. For a stated target population P, the average treatment effect is E_P[Y(1) − Y(0)]. Under random assignment, valid measurement, and a stable version of the intervention, the difference of group means identifies an intention-to-treat contrast in expectation. This statement concerns assignment; it does not imply that observed covariates will be perfectly balanced in a finite sample.
Randomization breaks systematic dependence between assigned treatment and pretreatment characteristics in the design distribution. It can support exact randomization inference under sharp null hypotheses, design-based variance estimation, and transparent tests of the intervention's effect. Its power depends on the realized sample, outcome variability, assignment probabilities, and treatment saturation. The account in Understanding and Misunderstanding Randomized Controlled Trials emphasizes that randomization alone neither ensures balance in every realization nor guarantees relevance for new circumstances.
A basic difference-in-means estimator is τ̂_ITT = Ȳ_Z=1 − Ȳ_Z=0, where Z is randomized assignment. If some assigned subjects do not comply, τ̂_ITT remains a contrast of offers or assignments when follow-up is complete; it is not automatically the effect of receipt. With binary assignment Z, binary take-up D, a valid exclusion restriction, monotonicity, and other instrumental-variable assumptions, the Wald ratio [E(Y|Z=1)−E(Y|Z=0)]/[E(D|Z=1)−E(D|Z=0)] can identify an average causal effect for compliers. Angrist, Imbens, and Rubin's instrumental-variable framework establishes the importance of this estimand-specific interpretation. A numerator and denominator that are estimated imprecisely warrant appropriate weak-instrument and uncertainty analysis.
Randomization at the cluster level requires inference at that level. Spillovers and general-equilibrium responses may invalidate an assumption that each unit's potential outcome depends only on its own assignment. In social networks, villages, firms, schools, and markets, interference can be part of the substantive object of study; designs should then describe exposure mappings and assignment saturation rather than assume it away. Attrition that differs by treatment, intervention-induced survey participation, and selective outcome measurement can defeat a clean comparison even when the initial lottery was impeccable.
The methodological task begins by stating the estimand and its intended population, then justifying the variation that identifies it. Mechanism research adds further randomized contrasts and process measurements; policy transport adds assumptions about differences between experimental and target settings. These are distinct inferences requiring their own evidence.
5. Incentives, Valuation, and Revealed Choice
Economic experiments often pay participants according to their choices and outcomes so that the decision problem has real consequences. The purpose is to give observable behavior a direct connection to an explicitly described reward environment. In induced-value markets, controlled private costs and values allow a researcher to construct theoretical supply and demand schedules against which trades can be compared. The experimental design thereby isolates aspects of market institution performance while acknowledging that participants bring nonmonetary motivations and interpretive habits to the session.
An incentive-compatible mechanism aligns truthful revelation with a participant's optimal strategy under specified assumptions about preferences, beliefs, and the payment rule. Incentive compatibility must therefore be evaluated for the precise mechanism, including uncertainty, reference points, and social interaction. Random incentive selection, for instance, may be interpreted as isolating one randomly paid decision, but its validity depends on assumptions about compound lotteries, cross-task effects, and the structure of preferences. A payout is not a universal guarantee of truthful disclosure.
Holt and Laury's risk aversion and incentive study demonstrates why stake size and real payments matter for elicited risky choices. A multiple-price-list task provides ordered comparisons between lotteries, yet estimated risk parameters can be sensitive to comprehension, probabilities, background wealth, framing, and the assumed utility specification. A researcher should separate the observed switching pattern from the parameter that a model translates it into.
Experimental demand, willingness to pay, willingness to accept, labor supply, and effort depend on the exact feasible alternatives and the outside option. Subjects may respond to a cash reward, to the presence of an experimenter, to perceived fairness, or to information learned from the task. This is an argument for varied treatments and explicit checks rather than for rejecting incentivized evidence. The value of an experiment is strengthened when alternative incentive stories are formulated as testable rival explanations.
6. Markets, Price Formation, and Experimental Institutions
A double auction permits competing buyers and sellers to submit or accept offers. Within the controlled schedule of gains from trade, researchers can measure transaction prices, allocation efficiency, surplus distribution, convergence, and the path by which trading institutions approach or depart from equilibrium benchmarks. The seminal contrast between Chamberlin's trading institution and Smith's competitive experiments illustrates that the rule system can materially alter aggregate outcomes. Theoretical demand and supply schedules do not by themselves specify a unique dynamic path for each institutional arrangement.
A competitive price prediction typically rests on a schedule of reservation values and costs, while efficiency calculations depend on which trades create positive gains and which are actually executed. Measuring allocative efficiency as realized surplus divided by maximal feasible surplus can be informative when the denominator is positive and all relevant valuations are represented. The observed distribution of prices can be near a predicted equilibrium even if surplus allocation differs, and high efficiency can coexist with dispersion and unequal distribution. Experiments should therefore report multiple outcome dimensions rather than collapse market performance into a single success label.
Experimental asset markets introduce additional complexities. Trading occurs over time and expectations about resale prices can dominate immediate dividends or fundamentals. Price trajectories, turnover, inventories, belief formation, and information revelation become joint objects of inquiry. Results from a laboratory market cannot automatically be extended to all financial systems because leverage, collateral, regulatory constraints, professional experience, and persistent heterogeneous beliefs differ. The controlled environment instead reveals which market features are sufficient to produce particular patterns and which institutional modifications change those patterns.
The study of experimental institutions is especially important when multiple mechanisms can implement similar nominal allocations. Posted prices, continuous double auctions, call markets, bilateral bargaining, and auction formats change who can communicate, the timing of action, and the strategic value of information. Experimental economics supplies comparative evidence on the rules through which decentralized preferences and resources are coordinated.
7. Auctions, Strategic Information, and Common Values
Auctions organize allocation and payment through rules that make strategic incentives explicit. Private-value auctions assign each bidder an individual valuation; common-value auctions expose bidders to noisy signals about an underlying value shared across bidders. Mixed-value environments include both components. The theoretical comparison of first-price, second-price, English, Dutch, and multiunit mechanisms depends on assumptions about risk attitudes, information, bidder symmetry, collusion, and the relevant equilibrium selection.
The winner's curse arises when winning is informative: the highest signal or most optimistic bid is disproportionately likely to belong to a bidder who overestimates the common value. Rational bidders should condition on winning. Kagel and Levin's common-value auction experiments document aggressive bidding and losses under some experimental conditions, with effects that depend on experience, the number of bidders, and public information. The study is evidence about these auction environments and the behavioral difficulty of conditional-value reasoning; it does not imply that all auction formats yield systematic bidder losses.
Experimenters can vary the number of bidders, quality and correlation of private signals, disclosure of public information, reserve prices, entry costs, repeated experience, and payment rules. Each manipulation addresses a different theoretical channel. Increasing entry may intensify competition but also change the informational value of the winning event. Public signals can reduce uncertainty while changing the strategic relation between bids. Empirical auction research becomes most informative when payoff feedback and learning are tracked over repeated rounds and when subjects' available information matches the model being tested.
Institutional design applications extend to procurement, spectrum allocation, online advertising, matching, and public-resource auctions. Laboratory tests can expose implementation bugs, unintended incentives, and fragile equilibria before a mechanism is deployed; field trials and structural evidence are still required to judge distributional, political, and operational performance at scale.
8. Bargaining, Negotiation, Trust, and Reciprocity
Bargaining experiments isolate the tension between strategic power and judgments about acceptable division. In an ultimatum game, one actor proposes a split and the other may accept or reject it; rejection destroys the relevant payoffs for both parties. Under a narrow model of purely monetary self-interest and common knowledge, positive offers should be accepted and proposers should offer the smallest positive amount allowed. Actual experimental responses vary with stakes, identities, framing, and local norms. Güth, Schmittberger, and Schwarze's ultimatum bargaining experiment became a foundational reference because offers and rejections make the social acceptability of allocation directly observable under a simple game structure. Kahneman, Knetsch, and Thaler's fairness and market-entitlements study examines judgments about acceptable market conduct, reinforcing the distinction between observed transaction incentives and normative perceptions.
Comparative experimental designs reveal that bargaining and competitive exchange respond differently to institutional and social context. The four-city study by Roth, Prasnikar, Okuno-Fujiwara, and Zamir, Bargaining and Market Behavior, reports stronger cross-country differences in bilateral bargaining than in the study's market settings. The inference is context-specific: participant pools and institutional details condition what the observed differences can establish, and national labels should not be treated as unchanging behavioral types.
Trust games investigate whether one participant makes a costly transfer that creates an opportunity for another to reciprocate. The trust, reciprocity, and social history experiments by Berg, Dickhaut, and McCabe analyze how transfers and returns respond to such a structure. A positive transfer may reflect trust, altruism, inequality concerns, beliefs about reciprocity, or risk attitudes. The amount returned may likewise reflect reciprocal response, distributional preferences, or reputational motives in variants where repeated interaction is possible. Identifying one mechanism requires targeted contrasts rather than reading it directly from a single allocation.
The bargaining literature establishes a general result about economic explanation: equilibrium conditions define strategic opportunities, while social evaluations, expectations, and institutional norms can influence how those opportunities are used. Experiments become cumulative when they retain comparable payoff structures while varying communication, anonymity, asymmetry, and outside options.
9. Public Goods, Cooperation, and Collective Action
A standard voluntary-contribution mechanism gives each group member an endowment and allows investment in a common account that benefits others. When the marginal private return to contribution is below its marginal cost but aggregate returns exceed costs, individual and group incentives diverge. This transparent conflict makes public goods games a powerful instrument for studying free riding, conditional cooperation, punishment, social norms, and institution design.
In Why Free Ride?, Andreoni distinguishes strategic behavior and learning as interpretations of contribution dynamics. In Are People Conditionally Cooperative?, Fischbacher, Gächter, and Fehr employ an elicitation design that reveals heterogeneity, including conditional cooperators and free riders in their experimental sample. The reported shares characterize that sample and procedure; they should not be exported as permanent population proportions across countries and institutions.
Fehr and Gächter's punishment and cooperation experiments demonstrate that allowing participants to impose costs on low contributors can influence cooperation in the studied games. Punishment itself consumes resources, so increased contributions and increased net welfare are separate empirical propositions. Punishment rules may also motivate retaliation, exclusion, or antisocial sanctions in other environments. Institutional evaluation therefore requires reporting contribution rates, enforcement costs, realized surplus, and distributive effects.
Public goods experiments connect to taxation, organizational teamwork, environmental stewardship, community infrastructure, and collective self-governance. Their key explanatory objects are conditional expectations about others, the visibility of action, the credibility of commitments, and the ability of groups to coordinate rules. The behavioral regularities are informative only alongside the relevant group composition, matching protocol, repetition, and local incentive structure.
10. Strategic Coordination, Cascades, and Expectations
Coordination games admit multiple mutually consistent outcomes, sometimes including equilibria with different efficiency and risk properties. In a setting where participants must select compatible actions, information about labels, previous choices, communication, and payoff asymmetry can become decisive. A focal point is an aspect of the situation that participants regard as mutually salient, yet salience does not guarantee coordinated choice when incentives conflict. Crawford, Gneezy, and Rottenstreich's focal-point experiments show that small payoff asymmetries can undermine coordination achieved by salient labels in otherwise symmetric conditions.
Informational cascades concern sequential decisions based on private signals and visible predecessor actions. An early pattern of public decisions may become sufficiently informative to cause later participants to disregard their own signals. Anderson and Holt's laboratory cascade study experimentally illustrates conditions under which this inference can arise. The laboratory evidence supports a conditional mechanism for social learning, not a claim that cascades always express irrational herd behavior. In a Bayesian model, following predecessors can be individually optimal even when the aggregate informational outcome is fragile.
Dynamic interactions make the experiment's temporal structure essential. Feedback about others' actions may update beliefs, alter the value of experimentation, and create path dependence. Participants may learn the game, learn other players' types, or learn an institution's informal conventions. To distinguish these channels, experimenters can vary observability, information precision, payoff structure, and the possibility of revising commitments. An equilibrium comparison without the time path loses potentially important explanatory information.
Coordination and cascade experiments speak to technology adoption, bank runs, platform standards, network effects, public expectations, and financial-market sentiment. Translating them requires attention to what participants observe in actual institutions, whether agents can exit or delay, and how policy changes shared beliefs. Institutional design can sometimes improve coordination by changing the information structure even when private payoffs remain constant.
11. Risk, Uncertainty, Time, and Preference Elicitation
An experiment on risky choice specifies lotteries, probabilities, stakes, feedback, and a payment rule. A choice under ambiguity adds uncertainty about probabilities or the underlying state. A time-preference experiment compares outcomes delivered at different dates, often confounding impatience with trust in payment, liquidity needs, transaction costs, and perceived risk. Formal theories provide candidate behavioral restrictions, while observed decisions establish which restrictions survive under the experiment's conditions.
The historical importance of Allais's 1953 risk-choice critique lies in the use of carefully structured alternatives to test implications of an axiomatic representation. Observed preference patterns may contradict one representation while remaining compatible with several others. An experimental violation of independence, transitivity, or stationarity must be interpreted relative to the elicitation protocol and evidence for stable comprehension. Theory testing requires predictions specified at a level of precision sufficient to distinguish rival accounts.
Multiple-price-list tasks vary the relative attractiveness of risky options. Holt and Laury's incentive-effect evidence reinforces the importance of real consequences, but observed switching is not a complete psychological explanation. An estimated curvature parameter is conditional on the functional form, probability weighting, treatment of losses, background consumption, and consistency of responses. The same caution applies to discount factors estimated from deferred payments, strategic beliefs inferred from games, and social-preference weights recovered from allocations.
Preference elicitation can use choices, bids, certainty equivalents, probability reports, matching tasks, and real-effort decisions. Each instrument imposes response demands that may create method-specific behavior. A reliable program triangulates the same substantive construct with different incentive-compatible procedures where possible, reports within-person variation, and assesses out-of-sample predictive performance. The distinction between identifying a parameter within a model and identifying an enduring preference is central to responsible interpretation.
12. Laboratory, Field, Online, and Hybrid Environments
Laboratory research offers comparatively high control over instructions, information, matching, timing, and payment. Field research offers opportunities to recruit participants from relevant populations and observe behavior in familiar economic contexts. Harrison and List's field-experiment taxonomy differentiates settings through subject recruitment, task familiarity, commodities, rules, stakes, and environment, rather than through a simple laboratory-versus-field hierarchy. Laboratory and field methods address overlapping questions with distinct strengths and vulnerabilities.
Artefactual field experiments import structured laboratory tasks into field populations; framed field experiments embed tasks and context more deeply in real-world settings; natural field experiments may involve participants unaware that they are in a research experiment when legally and ethically permitted. Conventional laboratory experiments often make research participation explicit and experimentally control valuations. The appropriate design depends on whether the principal research objective is causal isolation, behavior in a target population, institutional realism, or some combination.
The comparative question is empirical. Levitt and List's laboratory-to-field assessment explains how scrutiny, selection, context, and task characteristics can alter social behavior across settings. Falk and Heckman's case for laboratory experiments emphasizes the distinctive explanatory benefits of controlled variation. These are complementary methodological insights: external validity cannot be assumed from naturalistic appearance, and causal clarity cannot be inferred solely from the controlled location of an experiment.
John List's market-experience study and subsequent exogenous-experience test demonstrate how related field environments and experimental variation can investigate whether accumulated market experience changes observed valuation anomalies. The latter addresses an important concern about endogenous sorting into experienced groups. Together, they show how an initially observed pattern can be refined by designs that identify the circumstances under which it appears.
Online experiments enlarge recruitment opportunities and support repeated decision tasks across locations, but introduce dropout, participant attention, device differences, comprehension challenges, multiple accounts, automated participation, and matching latency. Arechar, Gächter, and Molleman's interactive online experiment documents a case in which core public-goods behavior was reproduced online while addressing dropout and implementation challenges. Peer and colleagues' platform data-quality comparison shows why recruitment platform and screening practices are part of the evidence-generating process. These studies establish contextual feasibility and risks; they do not certify every contemporary platform or sample.
Experimental software has become a substantive element of reproducibility. The z-Tree research tool illustrates how standard implementations facilitate interactive laboratory studies. Modern online and in-person systems should document random seeds, session identifiers, matching histories, payments, event timestamps, code versions, and failed sessions. Reproducible code cannot rescue an ambiguously specified intervention, but it can make the implemented intervention auditable.
13. Randomized Evaluation and Economic Development
Field randomized evaluations investigate interventions whose consequences matter directly to households, organizations, governments, or markets. Random assignment can be implemented through lotteries, cluster allocation, phased rollouts, encouragement designs, or randomized offers. Treatment definitions may involve information, incentives, products, access to services, or institutional procedures. The methodological guide by Duflo, Glennerster, and Kremer, Using Randomization in Development Economics Research, explains assignment, stratification, sample size, noncompliance, externalities, and interpretation as connected design choices.
A policy intervention is rarely a simple pill-like object. Programs are delivered by organizations with their own incentives, capabilities, and errors. The treatment contrast may include monitoring, personnel, equipment, participant expectations, and implementation attention. Duflo, Hanna, and Ryan's teacher-attendance experiment combined attendance monitoring and pay incentives in Indian schools and reported a reduction of teacher absenteeism by 21 percentage points relative to controls and higher test scores in the study setting. Those numbers characterize a specific bundled intervention and context; isolating each component's separate effect or forecasting a national rollout requires additional evidence.
Kremer and Miguel's deworming and public-goods study evaluates take-up, social spillovers, and attempts to sustain community provision under alternative payment arrangements. Its conclusions apply to the evaluated program and associated conditions. Broader inferences must also distinguish health externalities, household private benefits, and the institutional economics of delivery. The methodological contribution of such studies is that they make the content of actual policy interventions and the consequences of implementation empirically examinable.
A randomized policy evaluation can identify a local program impact while leaving open questions of budget constraints, market adjustment, political adoption, long-run equilibrium, and the consequences of scaling. The concept of cost-effectiveness requires a comparable outcome measure, a defined time horizon, an explicit perspective, and measurement of full implementation costs, including monitoring and administrative overhead. The lowest observed cost per unit of outcome in one study should not be treated as a universal policy ranking.
Field experiments have contributed to education, health, labor, development, energy demand, tax compliance, governance, and financial inclusion. The domain of inference changes when an intervention interacts with markets or public systems that were essentially unaffected at small experimental scale. A scientifically complete evaluation separates the effect in the research sample, its distribution across relevant groups, its resource costs, and its plausible behavior under altered scale or institutions.
14. Experimental Statistical Analysis and Design Integrity
Statistical inference follows the assignment mechanism. For individually randomized, independent units, the difference in observed means is a natural estimator of the assignment effect. Analysis of covariance using prespecified baseline variables can improve precision, particularly when these variables strongly predict outcomes. For matched pairs or blocked assignment, the analysis should respect the blocking. Cluster allocation requires cluster-appropriate uncertainty estimation, with caution for a small number of clusters. Repeated observations within a session are not independent experimental replications.
Power planning begins with a primary estimand, a minimally important effect, anticipated variability, allocation ratio, significance threshold, cluster correlation, and expected loss to follow-up. For an elementary two-arm comparison with equal-sized groups and independent observations, a useful normal approximation is n_per_arm ≈ 2σ²(z_{1−α/2}+z_{1−β})²/δ². This is a planning approximation under assumptions, not a general power formula for repeated games, many clusters, heterogeneous variances, or nonlinear outcomes. Small expected effects, rare events, and high within-cluster correlation call for correspondingly stronger design capacity.
A coherent statistical report includes the prespecified primary contrast, the total randomized, the analyzed sample, treatment receipt, missing outcomes, compliance, deviations from protocol, estimate, confidence interval, and relevant heterogeneity or sensitivity checks. Randomization checks should describe realized covariate imbalances; mechanical significance tests of baseline variables add limited information about whether an independently implemented lottery was valid. Exact and permutation tests can provide useful design-based checks when assignment is known.
Multiplicity arises when a study measures numerous outcomes, treatment arms, subgroups, stages, or specifications. A large set of exploratory tests increases the probability of apparently striking results under chance. A prespecified family of hypotheses, appropriate multiple-testing adjustments, and explicit designation of exploratory analysis help align claims with evidence. A nonsignificant result is not proof of zero effect; precise equivalence tests require an equivalence region and adequate information. Statistical significance, practical importance, cost-effectiveness, and scientific explanation answer different questions.
Missing data require explicit assumptions. Complete-case analysis may distort treatment comparisons if treatment changes attrition. Bounds, inverse probability weights, imputation under stated assumptions, and robust sensitivity analyses can illuminate how conclusions depend on missingness. Where incentives make strategic misreporting possible, measurement validation is as important as model choice. Unobserved spillovers, contaminated treatments, altered outcome collection, and treatment misclassification are design problems whose resolution cannot be delegated to a regression specification.
15. Replication, Publication Bias, and Open Research
The credibility of experimental economics depends on the relationship between reported results and the universe of studies attempted. Outcome reporting choices, selective exclusion, flexible stopping, multiple specifications, and nonpublication of null findings can yield exaggerated evidential confidence. Franco, Malhotra, and Simonovits's publication-bias investigation compared a known body of social science experiments with their publication histories and found differential reporting of statistically strong and null findings. This is direct evidence from that research pipeline; it should not be turned into a universal numerical correction for every experimental literature.
A replication may repeat a design closely in the same kind of environment, reproduce an analysis from existing data, or test whether a result persists in a materially different setting. These are distinct scientific exercises. Camerer and colleagues' 2016 economics replication project investigated a selected set of laboratory experiments, while their 2018 social-science replication project evaluated 21 systematically selected studies from Nature and Science published during 2010–2015. The latter found a significant effect in the original direction for 13 studies and smaller average effects in the replications. The 13/21 statistic belongs to that selected cross-disciplinary sample and is not the general replication rate of economics.
Preregistration records a research plan before outcome information can influence it. It can specify hypotheses, treatment definitions, outcome priorities, eligibility, stopping rules, statistical estimands, and planned exclusions. The AEA RCT Registry provides a durable record and explains journal registration requirements for field experiments. Preregistration improves the visibility of confirmatory commitments; appropriately labeled exploratory discovery remains valuable and may motivate subsequent confirmatory studies.
Transparent reproducibility also involves data provenance, documented code, versions of instructions and survey instruments, and the ability to reconstruct reported tables. The AEA's Data and Code Availability Policy describes replication materials, computational scripts, and instructions for original experiments. Privacy, contractual restrictions, and safety may limit access to raw data; in such cases researchers should provide lawful substitutes, access protocols, derived datasets, or detailed explanations of what can be independently checked.
A useful cumulative literature distinguishes replication of an estimate, replication of its qualitative direction, replication of a mechanism, and generalization of a policy conclusion. These criteria can diverge without logical contradiction. A robust scientific account reports the definition of success, the sampling of studies, expected precision, and the statistical uncertainty in its own meta-research claims.
16. External Validity, Cultural Variation, and Transport
External validity asks which causal and behavioral conclusions hold in populations, institutions, stakes, time periods, and environments different from the original experiment. It is not a property bestowed on a study by a location label. A natural field experiment with highly selected clients may generalize less widely than a carefully designed laboratory comparison whose mechanism is also established across multiple contexts. Transport depends on identifying moderators that differ between source and target situations.
Henrich, Heine, and Norenzayan's WEIRD-populations analysis highlights the risks of inferring universal behavior from populations concentrated in Western, educated, industrialized, rich, and democratic contexts. Diversity of samples is a research-design issue. Population weighting may support descriptive transport when the relevant characteristics and outcome models are known; it cannot correct for unobserved institutional differences by itself.
Cross-cultural evidence is more convincing when experimental instructions, stakes relative to local income, translation, comprehension, recruitment, and experimenter effects are transparently compared. The four-city bargaining study illustrates how market outcomes and bargaining outcomes can display different levels of variation under controlled comparisons. Culture is an empirical shorthand for social histories and institutions that require further explanation; the country of recruitment is not itself a uniquely identified causal mechanism.
Deaton and Cartwright's RCT interpretation analysis makes clear that extrapolation requires evidence beyond random assignment. Researchers can strengthen external inference by conducting multisite replications, varying meaningful institutional parameters, combining experiments with observational population data, testing measured mechanisms, and identifying whether equilibria change when interventions become common. A transport claim should name the target setting and specify which causal relations are assumed stable across settings.
Economic experiments are particularly vulnerable to scale-dependent mechanisms. An individual subsidy may alter a person's behavior without affecting market prices at small scale, yet full implementation can change supply, wages, entry, or political responses. Interventions may also erode after novelty wears off or when implementing organizations face different constraints. Therefore, experimental evidence should be integrated with institutional knowledge and economic modeling when extrapolating across scale and time.
17. Research Ethics, Consent, and Governance
Experimental economics involves human participants whose time, resources, income, opportunities, and expectations may be affected by research design. Ethics begins with a defensible research purpose, proportionate risks, fair recruitment and compensation, comprehensible information, and protection of confidential data. Institutional review and local legal requirements apply according to the actual jurisdiction, study design, and participant population. Economic incentives can create additional pressures when participants have limited resources or unequal bargaining power.
Informed consent should describe what participation entails, foreseeable costs, data handling, compensation, withdrawal options, and any justified limits on advance disclosure. Some natural field designs use ordinary commercial interactions or deception to identify behavior under genuine conditions. Such choices require a setting-specific ethical and legal justification, proportionate safeguards, and appropriate review; scientific realism alone is insufficient. Debriefing may help but does not retroactively eliminate harm or privacy risk.
Group experiments require particular attention to strategic exposure. An invitation to punish, exclude, reveal private information, or make socially visible decisions can change relationships and vulnerabilities. Field interventions involving schools, healthcare, employers, political institutions, or financial products may affect persons who never consented to individual participation. Researchers must consider potential spillovers, procedural fairness, and the distribution of benefits and harms across participants and nonparticipants.
Digital experimentation creates additional governance duties involving data minimization, secure storage, retention periods, participant identification, third-party recruitment platforms, automated screening, and geographic restrictions. The presence of an identifier in a dataset can turn behavioral observations into personally sensitive information. Open-science commitments should be implemented in forms compatible with legitimate privacy and safety protections, rather than assuming unrestricted public release of microdata.
A comprehensive ethics record can include the approval basis, consent wording, participant support routes, adverse events, exclusions, data retention, recruitment payment, conflicts of interest, and the treatment of incidental disclosures. These matters are part of the reliability and accountability of scientific inference, because coercive recruitment, unequal understanding, and concealed selection can alter observed behavior as well as violate obligations to participants.
18. Connections to Economic Theory and Other Disciplines
Experimental economics contributes to economic theory by making theoretical models vulnerable to observable alternatives. A theory may make an exact prediction about prices, contributions, strategies, or distributions. An experiment determines whether the relevant design assumptions hold sufficiently well, whether competing mechanisms imply different observations, and how learning changes initial behavior. A failed prediction can identify a missing preference component, informational assumption, equilibrium selection rule, or institutional condition, but a single experimental discrepancy rarely determines one unique model revision.
Game theory supplies the strategic language for auctions, public goods, trust, coordination, and bargaining. Behavioral economics contributes hypotheses concerning reference dependence, social preferences, attention, beliefs, and bounded reasoning. Econometrics supplies identification, estimation, and uncertainty quantification. Psychology contributes elicitation methods, experimental manipulation, measurement of cognitive processes, and attention to demand effects. Sociology, political science, anthropology, and law contribute substantive accounts of norms, institutions, power, and cultural variation.
Labor and organizational economics study incentive pay, monitoring, effort, peer effects, teams, and contracts. Development and public economics use field trials to test program mechanisms and fiscal or administrative trade-offs. Environmental economics investigates cooperation, conservation, energy consumption, and collective action. Finance experiments examine trading institutions, information, risk, and speculative expectations. Experimental economic history can use reconstructed rules and institutions to evaluate mechanisms, while retaining the distinction between a modern simulation and historical observation.
The relationship to computer science has grown through digital platforms, algorithmic matching, mechanism-design software, and reproducible experimental infrastructure. An algorithmically administered experiment should be audited as a software system and an economic institution: the assignment implementation, information shown, strategic possibilities, and potential technical failures all influence the observed outcome. Automated agents and machine-generated participant responses also create fresh questions about construct validity and whether the sampled unit remains the human decision process the investigator intended to study.
19. An Integrated Assessment of Experimental Evidence
The central analytical conclusion of this monograph is that the evidential force of an experiment depends on a chain of well-specified relationships connecting intervention, observation, comparison, interpretation, and target use. These relationships can be examined separately without converting the synthesis into a new named theory. A strong causal result is valuable within its estimand even if behavior is difficult to explain; a convincing mechanism is valuable even if its population frequency remains unknown. Scientific interpretation is improved by making the scope of each claim explicit.
First, design fidelity concerns whether participants encountered the intervention specified in the protocol. In computerized interaction this includes actual code behavior, feedback, matching, timestamps, and payments. Second, identification concerns whether assignment and measurement support the stated causal contrast. Third, behavioral interpretation concerns which preferences, beliefs, learning processes, or institutions could produce the pattern; identifying one process requires discriminating evidence. Fourth, precision concerns the uncertainty appropriate to the design, not merely a favorable p-value. Fifth, reproducibility concerns whether data, code, protocols, and results can be inspected and rerun. Sixth, transport concerns which mechanisms and populations resemble the intended setting. Seventh, ethical accountability concerns the legitimacy of the experiment's conduct and its impacts on people.
This assessment resolves several recurrent interpretive confusions. Experimental price convergence in a double auction supports the performance of a rule system under controlled conditions; it does not entail universal efficiency. Rejection of an unequal bargaining offer is a behavioral observation; it is not uniquely an estimate of inequity aversion. Higher contributions with punishment identify the effect of making sanctioning possible in a given game; the welfare conclusion additionally requires counting enforcement costs. A program effect identified by randomization estimates the evaluated intervention's effect; it is not a national cost-benefit estimate. A successful replication strengthens the credibility of a specific pattern; it does not establish universal transport.
The discipline's two historical emphases reinforce each other. Controlled games reveal the mechanism-bearing consequences of rules and incentives. Field randomized experiments reveal what happens when actual institutions deploy interventions. A research program that moves between the two can discover patterns, isolate explanatory features, and test whether those features retain causal force in settings where they matter. Multiple designs, rather than allegiance to one preferred setting, provide the basis for cumulative knowledge.
The comparative findings from the verified literature can be summarized in prose. Market studies show that pricing outcomes respond to institution and experience. Auction studies reveal the informational and strategic challenge of conditional valuation. Bargaining and trust studies expose deviations from exclusively monetary-payoff predictions that have multiple potential interpretations. Public-goods research documents conditional cooperation and sanctions under specific rules. Coordination and cascades research shows how expectation, observability, and asymmetry shape the collective path. Field studies demonstrate causally identifiable, sometimes substantively large program effects within actual delivery contexts. Metascience shows why selective reporting, replication, and explicit uncertainty are essential to the field's long-term credibility.
The synthesis therefore supports a demanding definition of an experimentally established economic result: an observed contrast is scientifically established to the degree that its intervention, assignment, measurement, uncertainty, theoretical interpretation, and relevant domain of application are disclosed and independently assessable. Different results may meet different degrees of this demanding standard. The appropriate scientific practice is to report those differences rather than attach an undifferentiated label of proof.
20. Contemporary Research Agenda and Unresolved Questions
Contemporary experimental economics faces the problem of constructing evidence across rapidly changing decision environments. Digital platforms combine economic incentives with algorithmic presentation, personalized information, and adaptive interaction. The assignment of a price or default may simultaneously change recommendation ranking and informational salience. Designs need to capture the whole treatment bundle, or build factorial and sequential interventions that identify its components. Interactive online experiments additionally require controls for differential dropout, repeat participation, and automatic or assisted responses.
A second problem concerns dynamic and general-equilibrium adaptation. Many experiments identify short-run responses to a marginal intervention, while actual policy or institutional reform alters incentives for entry, investment, organizational behavior, and social expectations. Integrating randomized pilots with economic theory, long-run follow-up, and comparisons across implementation intensity can clarify when local effects persist or change with scale. The question is substantive: interventions can transform their own comparison environment.
A third problem is the measurement and interpretation of heterogeneity. Individuals differ in prior experience, wealth, institutions, information, comprehension, and expectations. An average can mask effects of opposite sign or substantially different mechanisms. Credible heterogeneity requires prespecified moderators or appropriately cautious exploratory models, sufficient effective sample size, and external comparison. Explanations grounded in fixed demographic labels alone are generally underidentified when institutional context remains unmeasured.
A fourth problem is the relationship between behavioral regularity and welfare. Revealed choices depend on information, framing, commitment constraints, and social norms. Experimental evidence can identify how a choice changes under an intervention; evaluating whether that change improves welfare requires an explicit welfare criterion and attention to other affected persons. Researchers should document whose preferences and opportunities enter the welfare calculation and how the distribution of benefits and costs changes.
A fifth problem is the durability of research records. Experimental software, platforms, data protection requirements, and scholarly workflows evolve. Maintaining usable code, task definitions, protocol histories, permitted data access, and consistent bibliographic identifiers is a continuing task. Replication is more effective when reproducibility is planned before data collection rather than reconstructed after publication. Transparent, exact source-to-claim attribution in a scholarly synthesis serves the same goal at the level of interpretation.
Finally, interdisciplinary experimental work needs a precise vocabulary of what was varied, who responded, which mechanism is proposed, which estimand was identified, and where the result is expected to apply. Experimental economics has matured by joining theory with controlled evidence. Its continued progress depends on extending that relation to broader populations, changing institutions, and complex systems without losing the interpretive clarity that controlled comparison provides.
21. Conclusion
Experimental economics is a foundational method of modern economic science. It organizes economic questions as observable comparisons between specified incentive structures, information arrangements, trading institutions, and policy conditions. Its evidence spans the controlled market, strategic game, incentivized choice task, online interaction, and randomized real-world evaluation. The discipline's achievements include systematic testing of market theories, new evidence on strategic and social conduct, improved understanding of institutional performance, and rigorous evaluation of concrete interventions.
The historical record establishes an intellectual progression from early experiments on exchange and risky decisions to the comparative study of institutions, field experimentation, and reproducibility-oriented methods. The literature shows that individual incentives matter, rules matter, information matters, experience matters, and social context matters. It also shows why no single experiment can identify every mechanism or transfer a result to every population. Causal identification, theoretical interpretation, and external application are related scientific tasks with distinct evidential requirements.
The original analytical synthesis advanced here is a disciplined reading of the existing record: experiments become most informative when the complete path from assigned intervention to measured behavior, causal comparison, explanatory claim, and intended application is explicit. Reliable experimental economics integrates formal theory, sound research design, reproducible statistics, contextual knowledge, ethical governance, and cumulative replication. Its permanent contribution is an empirical discipline of economic explanation in which institutional and behavioral claims can be exposed to structured observation and revised in light of evidence.
References
Allais, M. (1953). Le comportement de l'homme rationnel devant le risque: Critique des postulats et axiomes de l'école américaine. Econometrica, 21(4), 503–546. Bibliothèque nationale de France record.
Anderson, L. R., & Holt, C. A. (1997). Information cascades in the laboratory. American Economic Review, 87(5), 847–862. Bibliographic record.
Andreoni, J. (1988). Why free ride? Strategies and learning in public goods experiments. Journal of Public Economics, 37(3), 291–304. https://www.sciencedirect.com/science/article/pii/0047272788900436
Angrist, J. D., Imbens, G. W., & Rubin, D. B. (1996). Identification of causal effects using instrumental variables. Journal of the American Statistical Association, 91(434), 444–455. https://doi.org/10.1080/01621459.1996.10476902
Arechar, A. A., Gächter, S., & Molleman, L. (2018). Conducting interactive experiments online. Experimental Economics, 21(1), 99–131. https://doi.org/10.1007/s10683-017-9527-2
Berg, J., Dickhaut, J., & McCabe, K. (1995). Trust, reciprocity, and social history. Games and Economic Behavior, 10(1), 122–142. https://doi.org/10.1006/game.1995.1027
Camerer, C. F., Dreber, A., Forsell, E., et al. (2016). Evaluating replicability of laboratory experiments in economics. Science, 351(6280), 1433–1436. https://doi.org/10.1126/science.aaf0918
Camerer, C. F., Dreber, A., Holzmeister, F., et al. (2018). Evaluating the replicability of social science experiments in Nature and Science between 2010 and 2015. Nature Human Behaviour, 2, 637–644. https://doi.org/10.1038/s41562-018-0399-z
Chamberlin, E. H. (1948). An experimental imperfect market. Journal of Political Economy, 56(2), 95–108. https://doi.org/10.1086/256654
Charness, G., Gneezy, U., & Kuhn, M. A. (2012). Experimental methods: Between-subject and within-subject design. Journal of Economic Behavior & Organization, 81(1), 1–8. https://doi.org/10.1016/j.jebo.2011.08.009
Crawford, V. P., Gneezy, U., & Rottenstreich, Y. (2008). The power of focal points is limited: Even minute payoff asymmetry may yield large coordination failures. American Economic Review, 98(4), 1443–1458. https://doi.org/10.1257/aer.98.4.1443
Deaton, A., & Cartwright, N. (2018). Understanding and misunderstanding randomized controlled trials. Social Science & Medicine, 210, 2–21. https://doi.org/10.1016/j.socscimed.2017.12.005
Duflo, E., Glennerster, R., & Kremer, M. (2007). Using randomization in development economics research: A toolkit. In T. P. Schultz & J. A. Strauss (Eds.), Handbook of Development Economics (Vol. 4, pp. 3895–3962). Elsevier. https://www.sciencedirect.com/science/article/pii/S1573447107040612
Duflo, E., Hanna, R., & Ryan, S. P. (2012). Incentives work: Getting teachers to come to school. American Economic Review, 102(4), 1241–1278. https://doi.org/10.1257/aer.102.4.1241
Falk, A., & Heckman, J. J. (2009). Lab experiments are a major source of knowledge in the social sciences. Science, 326(5952), 535–538. https://doi.org/10.1126/science.1168244
Fehr, E., & Gächter, S. (2000). Cooperation and punishment in public goods experiments. American Economic Review, 90(4), 980–994. https://doi.org/10.1257/aer.90.4.980
Fischbacher, U. (2007). z-Tree: Zurich toolbox for ready-made economic experiments. Experimental Economics, 10, 171–178. https://doi.org/10.1007/s10683-006-9159-4
Fischbacher, U., Gächter, S., & Fehr, E. (2001). Are people conditionally cooperative? Evidence from a public goods experiment. Economics Letters, 71(3), 397–404. https://www.sciencedirect.com/science/article/pii/S0165176501003949
Franco, A., Malhotra, N., & Simonovits, G. (2014). Publication bias in the social sciences: Unlocking the file drawer. Science, 345(6203), 1502–1505. https://doi.org/10.1126/science.1255484
Güth, W., Schmittberger, R., & Schwarze, B. (1982). An experimental analysis of ultimatum bargaining. Journal of Economic Behavior & Organization, 3(4), 367–388. https://www.sciencedirect.com/science/article/pii/0167268182900117
Harrison, G. W., & List, J. A. (2004). Field experiments. Journal of Economic Literature, 42(4), 1009–1055. https://doi.org/10.1257/0022051043004577
Henrich, J., Heine, S. J., & Norenzayan, A. (2010). The weirdest people in the world? Behavioral and Brain Sciences, 33(2–3), 61–83. https://doi.org/10.1017/S0140525X0999152X
Holt, C. A., & Laury, S. K. (2002). Risk aversion and incentive effects. American Economic Review, 92(5), 1644–1655. https://doi.org/10.1257/000282802762024700
Kagel, J. H., & Levin, D. (1986). The winner's curse and public information in common value auctions. American Economic Review, 76(5), 894–920. Bibliographic record.
Kagel, J. H., & Roth, A. E. (Eds.). (1995). The handbook of experimental economics. Princeton University Press. Publisher edition record.
Kahneman, D., Knetsch, J. L., & Thaler, R. H. (1986). Fairness as a constraint on profit seeking: Entitlements in the market. American Economic Review, 76(4), 728–741. Author-hosted article.
Kremer, M., & Miguel, E. (2007). The illusion of sustainability. Quarterly Journal of Economics, 122(3), 1007–1065. https://doi.org/10.1162/qjec.122.3.1007
Levitt, S. D., & List, J. A. (2007). What do laboratory experiments measuring social preferences reveal about the real world? Journal of Economic Perspectives, 21(2), 153–174. https://doi.org/10.1257/jep.21.2.153
List, J. A. (2003). Does market experience eliminate market anomalies? Quarterly Journal of Economics, 118(1), 41–71. https://doi.org/10.1162/00335530360535144
List, J. A. (2011). Does market experience eliminate market anomalies? The case of exogenous market experience. American Economic Review, 101(3), 313–317. https://doi.org/10.1257/aer.101.3.313
Peer, E., Rothschild, D., Gordon, A., Evernden, Z., & Damer, E. (2022). Data quality of platforms and panels for online behavioral research. Behavior Research Methods, 54, 1643–1662. https://doi.org/10.3758/s13428-021-01694-3
Plott, C. R. (1982). Industrial organization theory and experimental economics. Journal of Economic Literature, 20(4), 1485–1527. Bibliographic record.
Roth, A. E. (1988). Laboratory experimentation in economics: A methodological overview. Economic Journal, 98(393), 974–1031. Author-hosted manuscript.
Roth, A. E., Prasnikar, V., Okuno-Fujiwara, M., & Zamir, S. (1991). Bargaining and market behavior in Jerusalem, Ljubljana, Pittsburgh, and Tokyo: An experimental study. American Economic Review, 81(5), 1068–1095. Bibliographic record.
Smith, V. L. (1962). An experimental study of competitive market behavior. Journal of Political Economy, 70(2), 111–137. https://doi.org/10.1086/258609
Smith, V. L. (1976). Experimental economics: Induced value theory. American Economic Review, 66(2), 274–279. Bibliographic record.
American Economic Association. (n.d.). AEA RCT Registry: Registration and journal policy. Registry policy.
American Economic Association. (n.d.). Data and Code Availability Policy. Publisher policy.