SCIENCE

Economic Methodology

Abstract. Economic methodology studies how economic knowledge is constructed, justified, criticized, transmitted, and applied. Its object extends from concepts of choice, value, institutions, and equilibrium to formal modeling, measurement, statistical identification, experiments, comparative history, ethical judgment, and the translation of research into policy. The discipline investigates what a model explains, which circumstances make an estimate causal, how measurement decisions shape empirical results, why a credible local effect may have limited general applicability, and what reproducibility contributes to scientific authority. This monograph reconstructs methodological disputes in their historical setting, examines the logical and empirical foundations of major research practices, and analyzes their interdependence across the economic research process. Its central interpretation is that the adequacy of an economic claim is relative to the question it answers, the assumptions needed to establish it, and the domain over which its conclusions are intended to travel. Formal validity, statistical precision, behavioral realism, institutional fit, empirical identification, and policy relevance are different properties that require separate examination. Case studies from minimum-wage research, local treatment effects, laboratory markets, macroeconomic policy evaluation, and institutional analysis demonstrate both the capabilities and limits of research strategies. The concluding synthesis specifies a disciplined procedure for relating explanation, identification, transportability, and decision-making, without advancing a new economic theory or claiming priority for a novel conceptual system.

Keywords: Economic methodology; philosophy of economics; economic explanation; economic models; abstraction; idealization; causality; identification; randomized experiments; quasi-experimental design; structural econometrics; measurement; counterfactuals; external validity; prediction; reproducibility; research ethics; institutions; normative economics; welfare; scientific pluralism

1. Subject, Questions, and Scope

Economic methodology is the systematic examination of the concepts, inferential procedures, evidentiary standards, and evaluative commitments through which economists produce knowledge. Economists study production, exchange, prices, incomes, institutions, finance, growth, and distribution. Methodologists ask which propositions about these phenomena are meaningful, what their models represent, how observational claims can be distinguished from causal claims, which counterfactuals are defensible, and what kinds of evidence bear on those propositions. The JEL classification of economic methodology locates B41 within the wider history-of-thought and methodology category B4. That classification recognizes a distinct field of inquiry, although its intellectual subject inevitably crosses the boundaries of the many specialized research practices it examines.

The unit of methodological evaluation is a research claim in context. Consider a statement that a minimum-wage increase changes employment. Several questions arise before its truth can be assessed. Does the statement concern a particular jurisdiction and period, a class of labor markets, or a general theoretical relation? Does employment mean headcount, hours, full-time equivalents, or payroll? Is the comparator a region without the policy, a modeled path that would otherwise have occurred, or a randomized assignment? Does the statement describe an average effect, a distribution of effects, a mechanism, or a prediction? A quantitative estimate acquires its meaning from these specifications. A policy recommendation adds judgments about distributional consequences, implementation constraints, and social objectives.

Economic methodology therefore connects the architecture of a claim to the evidence that can support it. Deductive analysis assesses what follows from premises; statistical analysis assesses what observed variation permits us to infer; historical analysis establishes temporal and institutional circumstances; behavioral research examines whether an account of choice fits observations; normative analysis evaluates consequences according to articulated criteria. The task of methodology is to make their relationships explicit. Specialist subjects such as econometrics or mathematical economics investigate techniques in greater technical detail. Economic methodology evaluates when those techniques answer the intended economic question and how their results should be interpreted.

Five distinct intellectual operations organize much of the field: conceptual specification, model construction, evidence design, uncertainty assessment, and practical interpretation. They can be repeated and revised as new findings emerge. A model might be internally coherent and useful for prediction while remaining insufficient for a particular intervention. A local experimental effect may be causally credible yet require further argument to be exported to another country. A historical reconstruction may identify a plausible mechanism while leaving its quantitative magnitude unknown. A normative assessment may be transparent and rigorous even when alternative value weights produce different social rankings. Methodological quality consists in knowing exactly which warrant supports which inference.

The scientific object also has a reflexive dimension. Economic explanations affect the expectations and choices of institutions and people who know those explanations. Government rules, financial arrangements, and measurements can change after becoming objects of analysis. This makes the circumstances under which a proposition was learned an integral part of its interpretation. It calls for carefully specified domains of applicability and for empirical openness when economic arrangements evolve.

2. Historical Formation of Methodological Debate

The methodological problems of economics developed with political economy. Classical writers examined the consequences of specialization, property, trade, population change, monetary institutions, and public policy by combining abstract propositions with historical comparison and observation. Nineteenth-century controversies over induction, deduction, and the status of economic laws arose because the mechanisms economists sought to explain operated within societies whose institutions varied. Mathematical reasoning promised clarity about dependence and choice; historical inquiry insisted on the significance of place, legal arrangements, and change. Later disputes over individual motivation, aggregation, empirical testing, and the appropriate role of experiment continued to transform these earlier questions.

An enduring distinction concerns positive propositions about economic processes and normative propositions about desirable outcomes. John Neville Keynes's late nineteenth-century account influenced a prominent twentieth-century discussion. Milton Friedman's methodology of positive economics, first issued in 1953, makes prediction and comparison with experience central to scientific appraisal. For Friedman, an economic theory may employ simplifying assumptions, and its significance depends on the quality of the empirically relevant implications it yields. His defense of abstraction encouraged a durable question: how far can predictive adequacy compensate for an unrealistic description of a mechanism?

That question has more than one answer because theories serve different purposes. A price model may isolate substitution effects to show why one institutional rule alters incentives. A forecasting system may combine variables that predict well without describing a stable intervention mechanism. A market-design study may need a sufficiently accurate description of strategic behavior to evaluate a proposed rule. Treating these purposes as equivalent obscures differences in what counts as a successful test. Friedman's essay remains a historical landmark precisely because its approach to testing still invites substantive disagreement about explanatory and causal adequacy.

Economics also absorbed competing philosophies of science. Logical analysis focused on the meaning and verification of statements; falsificationist approaches emphasized the vulnerability of hypotheses to adverse evidence; research-program perspectives assessed sequences of related models; historical accounts of scientific change emphasized communities, instruments, and the shifting organization of problems. These traditions influenced how economists described theory choice even as actual economic practice retained mathematical proof, simulation, calibration, historical interpretation, and statistical evidence. No single historical doctrine describes every activity economists call scientific.

The later twentieth century saw the parallel growth of formal general-equilibrium modeling, macroeconomic time-series analysis, structural econometrics, laboratory markets, and behavioral research. The subsequent design-based turn in empirical microeconomics made credible identification a leading criterion of applied work. Angrist and Pischke's account of the credibility revolution explicitly places research design at its center and relates the change to earlier calls for sensitivity analysis. This development altered the relative weight given to elaborate models and transparent sources of empirical variation. It also sharpened debates about whether research could answer important structural or policy questions when its strongest designs identified narrow effects.

Modern economic methodology developed further under the demands of transparent computation, systematic replication, heterogeneous treatment effects, and machine learning. The history of the field is consequently best understood as the progressive differentiation of research questions and evidentiary techniques. Theoretical, experimental, econometric, historical, behavioral, and institutional traditions are continually re-evaluated in light of the claims they can support. The historical lesson is that scientific standards become more precise when the aims of inquiry are specified before techniques are selected.

The early analytical tradition also developed an explicit theory of economics as a distinct subject. John Neville Keynes's Scope and Method of Political Economy organized relations between positive analysis, normative judgment, and practical art. Lionel Robbins's Essay on the Nature and Significance, published in 1932, sharpened a scarcity-and-choice conception of economic subject matter and defended the importance of deductive reasoning. This historical vocabulary influenced economic curricula, but its status as a universal account of the field remains a methodological question. Empirical political economy, historical institutionalism, economic anthropology, and behavioral economics address economic processes that may require a broader analytical repertoire.

The dispute between abstract deduction and historical particularity remained visible in later discussions of methodology. Mark Blaug's Methodology of Economics reviews philosophies of science and their use in appraising economic research programs. McCloskey's Rhetoric of Economics redirected attention to the persuasive, metaphorical, argumentative, and professional practices through which economists make knowledge claims. This analysis of scientific communication matters because a polished formal result may derive rhetorical authority from conventions of publication that are distinct from empirical identification. The appropriate response is to make derivations, data, assumptions, and inferential steps visible to readers.

Institutional and evolutionary traditions offer a further historical perspective. Hodgson's Approach of Institutional Economics emphasizes habits and rules as conditions of economic activity and discusses the interaction of individuals with social structures. Ostrom's Governing the Commons uses comparative institutional inquiry to investigate collectively governed resource systems. Their methodological significance lies in the variety of observed organizational arrangements and the need to distinguish an abstract institutional possibility from an empirically grounded explanation of sustained collective action. These traditions broaden the available objects of explanation without exempting institutional claims from evidence and critical comparison.

3. Economic Explanation and Scientific Appraisal

An economic explanation identifies relationships that render an observed outcome intelligible. One explanation may show how a price emerges from supply and demand under specified conditions. Another may trace how a tax changes producer and consumer choices. A third may reconstruct how a banking regulation transformed liquidity management in an identifiable period. The explanatory burden differs across the cases. A derivation demonstrates the consequences of assumptions; a mechanism explains a process connecting conditions to outcomes; a causal estimate measures the consequences of a contrast; a historical account identifies sequences, institutional constraints, and actors. Good explanations can combine these functions while keeping the source of each inference visible.

The difference between explaining and predicting is significant. Accurate short-term forecasts can emerge from correlations that do not persist under intervention. An intervention model must specify how choices change when policies, information, and constraints change. Conversely, a model of equilibrium or bargaining can identify a causal possibility without yielding precise numerical forecasts for a particular time and place. Houseman's analysis in The Inexact and Separate Science of Economics examines the special status of generalizations that identify major causes while abstracting from disturbing factors. This idea captures a recurring methodological feature: the assessment of a conditional tendency requires an account of the conditions that permit its manifestation.

A scientific proposition should expose its logical commitments. A statement such as 'an increase in price reduces quantity demanded' normally belongs to a model with defined preferences, other variables, and a ceteris paribus restriction. It differs from an unconditional forecast that every observed price increase will coincide with an observed reduction in purchases. Simultaneous shifts in supply and demand, endogenous pricing, quality changes, and measurement errors can produce very different raw correlations. The methodological obligation is to distinguish a derived comparative-static statement from a directly observed empirical relation.

Explanatory assessment considers coherence, transparency of assumptions, scope, independent empirical content, and performance against appropriate rival accounts. Where competing explanations make different observable predictions, designed tests can discriminate among them. Where they are observationally equivalent, evidence may identify only an equivalence class of models. Alternative mechanisms can have the same reduced-form consequence while implying different responses under policies that alter the underlying constraints. Methodology should acknowledge that observational fit can coexist with structural ambiguity.

Economic knowledge advances through connected forms of criticism. Logical criticism finds contradictions or hidden assumptions. Statistical criticism questions sampling, uncertainty, and identification. Measurement criticism asks whether recorded quantities represent the intended concepts. Institutional criticism tests the relevance of legal and social context. Ethical criticism examines how an inquiry's selection of outcomes and populations affects interpretation. The discipline is strengthened when these criticisms are presented as answerable questions addressed to a specific claim, rather than as vague judgments on an entire school of thought.

4. Models, Abstraction, and Idealization

Models are deliberately constructed systems of relations that permit economists to reason about selected features of complex processes. A model can be a set of equations, a game, a simulation, a conceptual schema, an accounting identity, or a controlled experimental environment. Its essential operation is selection: some properties are represented, others are held constant, and still others are omitted. The act of abstraction has a purpose only in relation to the question the model is meant to illuminate. A theory of market clearing may isolate price adjustments, whereas a theory of unemployment persistence may need wages, contracts, expectations, and institutions that the simpler model sets aside.

Mary Morgan's The World in the Model analyzes how economists construct model worlds and use them as instruments of inquiry. The investigator can perform conceptual experiments inside a model and examine the consequences of a change before comparing them with a target economy. Robert Sugden's credible-worlds argument emphasizes the inferential gap between the internal world constructed by a model and the economic world outside it. His examples include models of quality uncertainty and segregation. Both contributions show why internal consistency is a condition for useful modeling but cannot alone establish that a result holds in an external population.

Idealization can simplify a property, assume it away, or introduce an intentionally stylized condition. Perfect information, infinitely divisible goods, representative consumers, frictionless transactions, and complete markets are familiar examples. Their methodological significance depends on sensitivity. If relaxing a strong assumption leaves the qualitative result intact, the model demonstrates a degree of robustness. If a prediction reverses when institutions or heterogeneous agents are introduced, the original assumption is substantively decisive. The relevant question becomes which omissions are harmless for a stated inference and which omissions determine it.

Model validation operates at several levels. Formal validation checks derivations and numerical algorithms. Construct validation asks whether a term such as 'capital,' 'market power,' or 'expectations' has been operationalized in a way that fits the phenomenon. Empirical validation compares implications with independent observations. Mechanism validation examines whether the pathways a model assumes correspond to the processes in the target setting. Policy validation evaluates the stability of those pathways under the contemplated intervention. These levels can yield different conclusions about the same model.

A useful illustration is the distinction between a consumer's utility function as a representation of consistent choice and as a psychological account of satisfaction. An ordinal utility representation may summarize observed rankings without committing to a measurable internal experience. If researchers use the representation to predict behavior in unfamiliar settings, additional stability assumptions enter. If it is employed in welfare evaluation, interpersonal and normative assumptions may be required. Methodology establishes the inferential permissions of the mathematical object and identifies the point at which the argument moves beyond them.

Simulations add computational experimentation to model construction. They allow researchers to observe the implications of complex interactions, feedback, nonlinear transitions, or heterogeneous agents. Their outputs remain conditional on algorithmic rules, parameter choices, initial conditions, and numerical implementations. A simulated result is evidence about the modeled mechanism; its relevance to an economy depends on empirical links outside the simulation. Computational power increases the range of scenarios that can be examined, and it increases the need to document precisely which assumptions generate the result.

5. Agents, Institutions, Equilibrium, and Aggregation

Economic accounts differ in their choice of explanatory unit. Individual agents can represent consumers, firms, workers, investors, or officials. Organizations combine multiple individuals through governance and incentive arrangements. Markets coordinate interactions among buyers, sellers, intermediaries, and rule-makers. States and other collective institutions establish enforceable conditions and provide goods. The choice of unit is methodological because causal relations at one level need not survive aggregation or disaggregation unchanged.

Methodological individualism proposes that social outcomes be explained through individuals' actions and relations. In practice, economic models frequently attribute preferences or objectives to households, firms, and governments as if these entities acted as unified decision-makers. Such representations may be convenient approximations or may require explicit theories of bargaining and internal conflict. A household demand function can conceal differences among household members; a representative-agent model can eliminate distributional distinctions important to aggregate behavior. The issue is how the aggregation assumptions determine the conclusions drawn from the model.

Equilibrium is a family of consistency concepts. Competitive equilibrium describes the compatibility of choices and market clearing under specified prices and endowments. Game-theoretic equilibrium specifies mutual consistency of strategies under information and payoff assumptions. Dynamic equilibrium can incorporate expectations and transition paths. None of these concepts establishes that a particular economy must rapidly converge to the modeled state. To move from a consistency result to a claim about actual adjustment, an additional dynamic account is necessary. Comparative statics then describes relations between possible equilibria, while transition analysis examines the path between them.

Institutions enter at every level. Property rights, enforcement, customs, information networks, accounting systems, and political authority affect feasible choices and the costs of exchange. Douglass North's analysis of institutional change places rules and historical persistence at the center of economic performance. An explanation of market outcomes that leaves institutional circumstances implicit may describe a conditional pattern accurately in one setting while failing in another. A robust comparative account identifies which institutional features are essential mechanisms and which are incidental local characteristics.

Aggregation presents distinctive methodological difficulties. A statistically representative sample of firms does not automatically form a representative firm for macroeconomic purposes when nonlinear responses, market power, or balance-sheet linkages matter. Similarly, the effect of an intervention on an individual may differ from its general-equilibrium effect once wages, prices, migration, entry, and fiscal financing adjust. The contrast between partial-equilibrium and economy-wide analysis is therefore a contrast between counterfactual questions. Both can be well-posed, but their answers apply to different interventions and time horizons.

The general methodological requirement is to map the scope of an explanation to its level of analysis. A microeconomic experiment identifies responses within a specified environment. A macroeconomic model examines interaction and feedback under broader assumptions. Institutional history explains why the environment itself has particular features. Their results are most informative when their conditional relationships are compared explicitly and when a mismatch of units is treated as an inferential problem requiring argument.

6. Economic Measurement and the Construction of Data

Economic quantities are constructed through definitions and procedures. Gross domestic product, real wages, inflation, unemployment, productivity, wealth, poverty, and market concentration each depend on classification, coverage, units, time boundaries, and adjustment methods. An observed data series is a record produced by institutions, instruments, respondents, and coding decisions. Methodology asks which properties of the target concept are captured, how comparable measurements are across settings, and how uncertainty propagates into analysis.

Measurement begins with an estimand or descriptive target. For labor-market analysis, 'employment' may refer to people currently employed, jobs, hours, or a survey-specific status. A policy could leave employment headcount unchanged while altering weekly hours. A study measuring one of these outcomes cannot silently establish a result about the others. Likewise, a price index tracks the cost of a selected basket and adjustment procedures; its movement can diverge from the cost of living experienced by particular groups. Transparent definition is substantive research work, not merely metadata.

Data collection entails selection mechanisms. Administrative tax data can cover registered transactions thoroughly while omitting informal activity. Household surveys can reveal attitudes and activities not recorded administratively but suffer from nonresponse, recall error, or sampling limitations. Digital transactions offer extensive behavioral records but only for participants in a platform and under its changing rules. A large number of observations reduces sampling variation under suitable assumptions; it cannot by itself remove selection bias or construct invalidity. Precision must be evaluated in relation to the population and variable actually observed.

Measurement error has several forms. Classical independent additive error in an explanatory variable can attenuate a simple regression coefficient under restrictive conditions. Errors correlated with outcomes, covariates, or policy assignment can bias coefficients in different directions. Changes in survey questionnaires or administrative rules create discontinuities that may be mistaken for economic shocks. Missing values can be ignorable under explicit assumptions or informative about the mechanisms under study. Every handling decision needs to be recorded because alternative choices can alter a research finding.

Real variables introduce further interpretation. Deflation changes the nominal unit of account into a price-adjusted construct; seasonal adjustment separates patterns according to a statistical procedure; chain weighting and index-number methods can alter measured growth. Cross-country comparisons require conceptual equivalence as well as currency conversion. Where definitions evolve, a historical series may combine distinct measurement regimes. A methodologically sound analysis treats comparability as a question to be demonstrated rather than an automatic property of any published data table.

Measurement choices connect directly to identification. A researcher may identify an effect on recorded wages without identifying an effect on total labor compensation, or may find a change in registered firm creation that partly reflects changes in reporting incentives. The resulting inference remains valuable if the target is explicit. A good evidence ledger therefore records concept, operational definition, observation unit, source population, time span, cleaning rules, weighting, and transformations alongside estimation results.

7. Statistical Inference, Identification, and Counterfactuals

Statistical inference asks what population or process properties can be learned from available observations under specified probabilistic assumptions. Identification asks a prior logical question: whether those observations and assumptions imply a unique value for the target quantity. An estimator can be computed precisely while the intended causal quantity remains unidentified. Conversely, a parameter may be identified by a model but estimated imprecisely in a small or noisy sample. Distinguishing identification from estimation is a central achievement of modern econometric methodology.

In causal inference, an intervention is characterized by competing potential outcomes. For an individual unit i, the effect of a binary treatment is conventionally represented as Yᵢ(1) − Yᵢ(0). At a given point under ordinary exposure conditions, we observe at most one of the alternatives for that unit. Causal inference therefore depends on comparison, design, assumptions, and the definition of treatment. The propensity-score analysis of Rosenbaum and Rubin develops adjustment under an assumption of assignment conditional on observed covariates. Its logic highlights the gap between balancing observed variables and accounting for unmeasured confounding.

Three questions determine a causal study's meaning. What intervention is being compared? Which population and outcome define the causal target? Which assumptions allow observed comparisons to stand in for missing counterfactuals? Random assignment can make treatment independent of potential outcomes under an implemented experimental design. In observational analysis, comparable conclusions require defensible conditional exchangeability, instruments, discontinuities, parallel trends, natural timing, or other identifying restrictions. An assumption may be partially testable, subject to falsification using auxiliary data, or irreducibly substantive.

Instrumental variables illustrate the power of explicit assumptions. A valid instrument must generate relevant variation in treatment and satisfy conditions connecting that variation to outcomes only through the specified channel. Under additional monotonicity conditions, Imbens and Angrist's local average treatment effect analysis identifies an average effect for units whose treatment status changes with the instrument. That population can differ from the full target population. Relevance, exclusion, monotonicity, and heterogeneity must be examined separately. An IV coefficient should not be presented as a universal treatment effect merely because the first-stage relation is statistically strong.

Identification depends on institutional detail. A legal threshold can support regression discontinuity if units cannot precisely manipulate placement around the cutoff and potential outcome relationships are suitably continuous. A policy introduced across regions may support differences-in-differences when untreated outcome paths would have been parallel under the relevant counterfactual and treatment is defined consistently. The serial-correlation critique of difference-in-differences by Bertrand, Duflo, and Mullainathan demonstrates that conventional standard errors can severely misstate uncertainty when treatment and outcomes are persistent. Research design and sampling uncertainty are joint components of credibility.

Partial identification is an important response when the available information cannot pin down a unique parameter. Rather than inserting strong assumptions solely to produce a point estimate, researchers can report a range of effects consistent with weaker restrictions. Sensitivity analysis can expose how conclusions change under plausible departures from exchangeability or functional-form assumptions. The choice is substantive: narrow conclusions under transparent assumptions may be more reliable than impressive estimates that hide uncertainty about the causal object.

The methodological synthesis is straightforward. An identified effect is conditional on its design; a general effect requires an argument about population and institutional scope; a policy prescription requires an evaluation of consequences and objectives. These layers should appear separately in research reporting. The modern applied econometrics assessment by Athey and Imbens emphasizes credible designs, robustness exercises, and the interpretation of heterogeneous effects, illustrating how formal estimation is inseparable from research design.

8. Experiments, Natural Experiments, and Design-Based Evidence

An experiment creates a controlled opportunity to observe economic behavior or intervention consequences under an explicitly arranged comparison. Randomized assignment can separate treatment from many determinants of an outcome in expectation, while laboratory control can make instructions, incentives, information, and transaction rules observable. Neither property automatically establishes a universal economic law. Investigators must specify the treatment, compliance, interference between units, subject selection, outcome measurement, and the meaning of the experimental environment. The experiment yields a credible answer to the question defined by its design.

Vernon Smith's experimental study of competitive markets demonstrates the methodological importance of constructing trading institutions in a laboratory to examine theoretical predictions. The laboratory supplies rules and information conditions that can be deliberately changed. Observing convergence or departures from a modeled outcome within this environment provides evidence about a particular market mechanism. Its inference to field markets requires comparison of the actual institutional and behavioral conditions, not merely an analogy between the word 'market' in two settings.

Randomized field experiments often address economic programs implemented among households, enterprises, schools, or communities. Randomization supports design-based identification of effects on measured outcomes for the eligible and assigned sample, subject to actual implementation. Noncompliance can shift the estimand from an intention-to-treat effect to a treatment-on-treated or complier effect when additional assumptions are imposed. Attrition can upset the initial comparability of groups; spillovers can cause control units to experience indirect treatment; selective reporting can make statistically significant findings more prominent than the full set of measurements. These are empirical questions about design integrity.

The intellectual debate over randomization concerns its contribution to understanding and policy. Deaton's critique of randomized and instrumental-variable evidence stresses heterogeneity, the meaning of exogeneity, and the need to study mechanisms rather than treating random assignment as a complete account of development. A carefully implemented experiment can establish an effect in the study setting, while the policy maker still needs to understand how the effect changes with prices, institutions, scale, and subsequent adaptation. The distinction is especially important when an intervention is expanded from a few trial sites to an entire economy.

Natural experiments exploit variation that arises outside the researcher's control. A judicial rule, policy threshold, staggered implementation, geographic border, or institutional accident may generate comparisons that approximate some properties of designed experiments. The credibility of such a study depends on the actual process producing that variation. Researchers must explain who was exposed and why, identify concurrent changes, and investigate whether the comparison is contaminated by sorting or spillovers. Calling an event 'natural' contributes no validity by itself; the identifying mechanism must withstand scrutiny.

The minimum-wage literature provides a canonical research-design example. Card and Krueger's New Jersey–Pennsylvania fast-food comparison, issued as a 1993 working paper and published in 1994, compares changes in employment following a minimum-wage increase with a neighboring-state comparison group. The study's findings challenged a simple prediction of inevitable employment decline in that setting. Methodological assessment concerns the comparability of locations, survey measurement, time horizon, choice of employment indicator, and the possibility of effects occurring through prices, turnover, or composition. The case illustrates how a specific empirical design can test an important proposition without establishing an unconditional sign for every labor market.

Replication and extension are especially valuable for experimental knowledge because implementation conditions are part of the evidence. Camerer and colleagues' replicability assessment of economic laboratory studies tested a defined collection of published experimental results using new experimental evidence. It contributes a direct empirical question about which findings recur under reported conditions; it does not imply that a single replication rate describes all economics or every research method. Experiments thus support cumulative knowledge when protocols, populations, incentive structures, and analysis choices are available for independent examination.

9. Structural Modeling and Macroeconomic Policy Evaluation

Structural economic analysis seeks relations whose parameters and mechanisms have interpretable connections to preferences, technologies, constraints, expectations, institutions, and policy environments. Reduced-form analysis summarizes observable associations or causal responses in a specified setting without always recovering every structural component. Both approaches can be rigorous. Their suitability depends on whether the research question asks for a local contrast under current arrangements, an account of underlying mechanisms, or the consequences of a policy that changes behavior and equilibrium.

The Lucas critique of policy evaluation argues that econometric relations estimated under one policy regime may change when policies change the decision problems and expectations of economic agents. A forecasting relation that tracks past inflation and unemployment can fail to describe the outcome of a novel monetary regime if households, firms, and institutions adapt. A policy analysis must justify which relationships remain stable under the intended intervention. Structural restrictions can address this problem, although their validity itself requires independent assessment.

Macroeconomic data create particular challenges. Aggregate output, prices, credit, investment, and employment are jointly determined over time. Economic shocks are latent and often observationally similar in their short-run signatures. Models incorporate identities, dynamic responses, and expectations, but identifying a structural shock requires assumptions such as timing restrictions, exclusions, external instruments, sign restrictions, or explicit theoretical restrictions. Different credible restrictions can yield different impulse-response interpretations even when the underlying data and reduced-form fit agree.

The role of models in macroeconomics includes accounting consistency and scenario analysis. A stock-flow relation can be an exact identity within its definitions, whereas the behavioral response of consumption to income is an estimated or hypothesized mechanism. A calibrated parameter may reflect external studies, a target moment, or a normative scenario; these origins should not be conflated. When models compare policy scenarios, uncertainty arises from measurement, parameter estimation, model specification, expectations, and the possibility of new institutional arrangements. Reporting only a narrow parameter confidence interval understates this broader uncertainty.

Lars Hansen's generalized method of moments analysis shows how theoretical orthogonality conditions can be used for estimation and large-sample inference under specified assumptions. Such moment restrictions often express economically meaningful claims, including equilibrium conditions or restrictions on forecast errors. Their statistical estimation does not independently establish that the underlying economic theory is true. Model appraisal still requires scrutinizing the content of moment conditions, the relevance of instruments, the behavior of samples, and potential misspecification.

A structural model may answer policy questions beyond the range of observed historical variation, but its power comes with dependence on strong assumptions. A quasi-experiment may estimate a particularly credible historical contrast while telling us less about a new regime at a different scale. Researchers can compare the two by asking whether the model reproduces observed treatment effects and whether local evidence is informative about specific structural parameters. This is an empirical strategy for joint learning, rather than a hierarchy that awards universal superiority to one method.

10. Behavioral, Institutional, and Historical Methodologies

Behavioral economics tests and enriches the assumptions used to represent choice. It examines reference dependence, framing, limited attention, bounded rationality, social preferences, self-control, and responses to uncertainty. Kahneman and Tversky's prospect theory of decision under risk analyzes behavioral regularities in risky choice and presents an alternative descriptive representation to expected-utility theory. Methodologically, it illustrates how systematic empirical anomalies can motivate a revision of model primitives. Such revisions require careful attention to task framing, incentives, learning, domain specificity, and repeated-choice behavior.

The meaning of rationality itself is methodological. Internal consistency of preference rankings, instrumental effectiveness, informed reasoning, and ethical deliberation are different standards. Amartya Sen's Rational Fools critique challenges narrow reductions of human action to self-interested maximization and emphasizes commitment. A choice can have reasons that extend beyond a decision-maker's private consumption or narrowly defined payoff. The methodological implication is that an explanatory model must state what its behavioral primitives capture and what kinds of motivation its inference leaves open.

Institutional approaches investigate how formal rules, enforcement, organizations, informal norms, and political structures shape economic outcomes. North's institutional account of economic performance treats historical persistence and the costs of cooperation as fundamental explanatory considerations. An institutional explanation typically requires evidence about actual rules and their implementation. The written statute may differ from enforcement practice; a formal property right may coexist with an informal mechanism allocating access; the same nominal reform may operate differently under distinct administrative capacities.

Historical research adds chronological and contextual identification. Archival documents, legislation, commercial records, household accounts, prices, inventories, and historical national accounts can establish what occurred and when. Such records may have been collected for taxation or administration rather than designed for economic research. Their coverage and interpretation depend on historical institutions. Changes in currency definitions, territorial boundaries, reporting rules, and categories of labor or property may prevent direct numerical comparison. Methodological reconstruction requires transparent source provenance and explicit adjustment to the changing meaning of measured variables.

Comparative economic history seeks more than narrative sequence. It asks which institutional or technological differences plausibly explain divergence in outcomes across places and periods. Quantitative comparisons can discipline historical theories; narrative source criticism can reveal a confounder hidden in a statistical proxy. Event studies and instrumental variables may be useful where identification assumptions are plausible, but historical uniqueness often makes external validity especially demanding. The primary question is whether the research design permits the specific causal or explanatory inference that its headline proposes.

Institutional and behavioral insights also change the interpretation of economic experiments. Field participants may respond to norms, trust, and administrative arrangements that laboratory protocols only partly reproduce. A policy can alter aspirations, networks, or political responses as well as budget constraints. Heterogeneous responses may be expected rather than anomalous. A rich economic explanation frequently requires formal theory, observed behavior, and institution-specific evidence to be interpreted together while maintaining the individual evidentiary status of each component.

11. Prediction, Computation, and Heterogeneous Effects

Prediction estimates an unknown or future outcome using available information. A predictive model is judged in part by how it performs on data that did not determine its fitted parameters. Cross-validation, holdout samples, rolling-origin evaluation, and explicit forecast horizons reduce the risk of evaluating a model on observations used to construct it. Good predictive performance is a substantive achievement, but an accurate predictor may be unsuitable for estimating the consequences of intervening on its explanatory variables. A strong predictor of loan default, for example, does not automatically identify the effect of changing lending criteria on defaults.

High-dimensional and machine-learning methods expand the set of relationships that can be approximated using complex data. Regularization, decision trees, ensembles, and flexible functions can discover nonlinearity and interactions that conventional parametric models represent imperfectly. Their flexibility increases the importance of training and validation splits, leakage prevention, transparent preprocessing, and reliable calibration. Prediction can be evaluated empirically under a stable data-generating environment; transport to a new environment requires evidence that relevant mechanisms and measurement processes remain comparable.

Causal machine learning addresses a distinct task: estimating effects or conditional effects while preserving identification assumptions. An algorithm can estimate nuisance functions such as propensity scores or outcome regressions, but the legitimacy of the causal interpretation still rests on the study design. The survey of modern causal policy evaluation by Athey and Imbens discusses machine learning alongside identification, robustness, and effect heterogeneity. This juxtaposition is methodologically significant: computational sophistication cannot replace identification, and design credibility can coexist with sophisticated estimation when each performs its appropriate role.

Heterogeneous treatment effects ask how an intervention affects different units. These differences matter for targeting, distributional analysis, external validity, and the design of scalable programs. Yet discovering subgroups after inspecting many outcomes can overfit random variation. Researchers should distinguish prespecified subgroup questions from exploratory discovery, use honest evaluation where possible, and report uncertainty about subgroup estimates. The estimated conditional effect for a small sample cell may be far less precise than an overall average, and the observed covariates may capture only part of the relevant heterogeneity.

Distribution shift is a central frontier. The relationship learned from a population, time period, or institution can change as economic arrangements evolve. Labor demand after a technological transition, financial behavior after a regulatory change, or platform behavior after a design revision may depart from training data. Policy evaluation requires specifying whether the estimand is a historical effect, a forecast under a maintained regime, or a counterfactual response to a new regime. Each purpose calls for a different validation strategy.

Computational reproducibility is a condition for interpreting numerical evidence. Scripts, dependencies, seeds, software versions, model weights where permissible, input data provenance, and computational resources can affect results. A reproducible program establishes that the same procedure yields the stated outputs under specified conditions. Independent replication asks a further question: whether the substantive claim withstands a new dataset, implementation, or design. Distinguishing these achievements prevents an executable code package from being treated as direct proof of the research hypothesis.

12. Replication, Transparency, and Scientific Reliability

Scientific evidence is cumulative when other investigators can understand, interrogate, and where feasible reproduce the procedures by which claims were reached. The publication of a table alone provides limited information about data selection, sample exclusions, preprocessing, statistical specifications, and computational decisions. Transparent research documentation creates a traceable connection between observations, methods, estimates, and arguments. It also permits errors to be detected and corrected without obscuring the history of a claim.

The AEA Data and Code Availability Policy, in its February 2026 revision, requires empirical, simulation, and experimental papers in covered journals to provide data, code, and computational details sufficient to permit replication prior to acceptance, with specified exceptions and procedures for restricted information. Its documentation requirements include provenance statements, cleaning and analysis code, software environments, and README instructions. This professional standard is evidence of an institutional commitment to reproducibility, not a guarantee that every published substantive conclusion will replicate independently.

Three processes warrant separate labels. Computational reproducibility is the ability to recover reported numerical outputs using supplied materials and instructions. Replication is an independent attempt to assess the same substantive result, potentially with a new sample or implementation. Robustness analysis investigates whether a conclusion persists under credible variations in definitions, assumptions, estimators, or sample restrictions. A project may pass one process and fail another. For example, a completely reproducible regression can depend on a weak causal assumption, while a replication failure can reveal context-dependent effects rather than misconduct.

Publication selection and analytic flexibility challenge inference. Researchers may have multiple defensible outcomes, covariates, sample cuts, and statistical models; reporting only the most striking specifications creates a misleading distribution of published evidence. Ioannidis's analysis of published research reliability formalizes concerns about the interaction of power, bias, selection, and prior plausibility. The article is a methodological essay with assumptions that have been debated, so its title should not be interpreted as a measured proportion of false findings in economics. Its general contribution is to require attention to research practices beyond a single nominal significance threshold.

Preregistration and analysis plans may help separate planned tests from later exploration. Such plans are especially valuable when outcomes and designs can be specified before data inspection. They do not eliminate legitimate learning from unexpected evidence; they make the distinction between confirmatory and exploratory findings visible. Registered experiments also require actual adherence to implementation procedures, ethical oversight, and complete reporting of deviations. The AEA journal policies establish field-experiment registration and research-integrity requirements within their own publication scope.

Replication has substantive relevance to economic theory. If an effect varies across markets, institutions, or populations, repeated studies may reveal the conditions under which it appears. The economic laboratory replication project is important in this respect because it studies reproducibility as an empirical characteristic of a defined research corpus. Replication outcomes should be read in light of statistical power, original estimates, the comparability of protocols, and possible genuine heterogeneity. The goal is a more accurate map of effect distributions and mechanisms, rather than a ceremonial pass-fail judgment on a field.

An appropriate evidence record includes provenance, exact source citations, outcome definitions, identification assumptions, statistical uncertainty, corrections, contradictory findings, and the scope of verified access to primary research. The integrity of a scientific manuscript also depends on correct hyperlinking and bibliographic attribution. A DOI landing page can establish a work's bibliographic identity; it does not prove that a researcher has inspected every derivation or dataset. Scholarly audit should state precisely what was read or checked and what remains contingent on access or independent reproduction.

13. Welfare, Values, and Public Decisions

Economic research frequently informs decisions about taxes, regulation, social insurance, monetary policy, investment, and public goods. This creates a systematic connection between positive explanation and normative evaluation. A causal estimate can describe what an intervention did to employment, household income, health, or prices. Recommending the intervention requires specifying a decision objective, considering alternative uses of resources, and assessing consequences beyond the estimated endpoint. The methodological move from effect to recommendation must therefore be explicit.

Welfare economics commonly uses efficiency, distribution, and social choice criteria. Pareto improvement is a specific comparison under which at least one person is made better off and no person worse off, but many important policies distribute gains and losses among groups. Cost-benefit analysis aggregates monetized effects under valuation conventions; distributional analysis asks who receives benefits and bears burdens. Social-welfare functions introduce weights and assumptions about interpersonal comparison. A mathematical welfare ordering may be internally consistent while representing controversial choices about inclusion, measurement, discounting, and rights.

The positive–normative distinction discussed in Friedman's essay on positive economics is useful for identifying the type of proposition under examination. In practice, the selection of an outcome is already relevant to a decision: a study of employment numbers, wage income, hours, job quality, and profit margins provides different information. Transparent value premises improve the interpretation of empirical work because they explain why particular outcomes were prioritized and how evidence will enter a public decision.

Under uncertainty, policy decisions involve the distribution of possible consequences, reversibility, institutional capacity, and the costs of error. Statistical significance is an evidentiary convention and is not a welfare criterion. A statistically imprecise estimate could concern a high-stakes harm; a precisely estimated small effect could be irrelevant to a particular decision. A decision maker may rationally request better measurement or additional evidence when the expected value of information warrants it. This requires credible specifications of costs, benefits, uncertainty, and decision context rather than a mechanical demand for one universal confidence threshold.

Scale and general equilibrium further complicate policy inference. An intervention that benefits a treated sample in a pilot can alter markets once universally implemented. Prices, the behavior of untreated people, tax finance, and institutional enforcement may change. External validity requires argument about these pathways. Deaton's discussion of learning about development emphasizes the need to understand mechanisms and heterogeneous circumstances when translating research results into development policy. A responsible policy analysis can combine evidence from trials, observational studies, structural models, and historical experience while keeping their inferential contributions distinct.

Economics also faces ethical obligations in the conduct of research. Surveys and experiments should respect consent and appropriate review; administrative data analysis requires attention to legal access and privacy; public communication should avoid exaggerating findings or disguising conflicts of interest. The AEA's publication integrity policies include data legality, disclosures, and research requirements. These rules apply directly to the specified journal context, while the broader ethical rationale concerns the rights and interests of people who generate research data and may be affected by its use.

14. Cross-Disciplinary Connections and Methodological Pluralism

Economic methodology draws on philosophy of science to analyze explanation, models, counterfactuals, and evidentiary warrant. It draws on statistics to analyze uncertainty, identification, and learning from data. It draws on mathematics to represent logical consistency, optimization, dynamics, and equilibrium. It draws on history to reconstruct institutional change and source provenance. It draws on psychology to investigate cognition and behavior, on sociology and political science to study collective organization, and on law to understand enforceable rules and regulatory mechanisms. These relationships are part of the substantive architecture of economics because economic outcomes are produced by people acting through institutional arrangements and material constraints.

The interplay of methods is clearest when researchers examine a difficult policy problem. A rent-control study may use administrative data to identify changes in tenancy, historical records to establish legislation and enforcement, theory to examine housing supply and mobility, and welfare analysis to assess the distribution of effects. A banking-crisis investigation may combine balance-sheet accounting, networks, expectations, macroeconomic counterfactuals, legal histories, and institutional comparison. A climate policy appraisal may combine physical models of emissions, economic behavior, discounting, distributional exposure, and political implementation. Each component answers a specific part of the larger question.

Pluralism is an epistemic practice when methods are selected on the basis of their relation to defined questions and their assumptions are made explicit. It is not a requirement to treat every inference as equally well supported. A randomized design can have a stronger claim about one local treatment contrast than a descriptive correlation. An archival reconstruction can have stronger evidence about the meaning of a historical institution than a present-day proxy. A formally identified equilibrium may be more reliable for its stated assumptions than an unstructured prediction about the same situation, while still remaining an incomplete basis for a particular policy. Standards remain rigorous because the research question and inferential burden guide the evaluation.

Cross-disciplinary integration also produces difficult translation problems. The word 'rationality' has different uses in decision theory, philosophy, psychology, and jurisprudence. 'Risk' may designate a probability distribution, a regulated financial exposure, or uncertainty arising from social vulnerability. 'Institution' may refer to a formal rule, an organization, or an informal convention. A comparative study should declare its terms rather than silently assume they coincide across disciplines. Conceptual precision is a prerequisite for useful synthesis.

Scholars using qualitative and quantitative methods can learn from each other's failure modes. Quantitative work can conceal institutional meaning behind numerical proxies; qualitative interpretation can understate uncertainty about the prevalence of a mechanism. Formal theory can make assumptions inspectable; historical inquiry can reveal why those assumptions matter. Experimentation can identify a causal effect; structural modeling can explore how the effect might change with market feedback. Methodological pluralism is strongest when these activities generate testable challenges to one another.

15. Analytical Synthesis: From Question to Warranted Claim

The preceding analysis supports an integrated interpretation of economic inquiry. A research result has a determinate epistemic status only when its target, assumptions, evidence, and intended domain are jointly specified. An economic explanation may be formally valid; a causal contrast may be identified; an estimate may be precise; a forecast may be accurate in a held-out sample; a recommendation may satisfy a declared welfare objective. Each statement answers a different question. Combining them requires a chain of arguments whose links are open to independent criticism.

The first stage of responsible inquiry is the articulation of the problem. The researcher defines the economic phenomenon, intended population, intervention or comparison, timing, outcome, and decision relevance. This prevents an attractive dataset or technique from quietly redefining the subject. A labor-market inquiry about the effect of minimum wages on household well-being is broader than an inquiry about employment headcount in a particular service industry. Both may be important, but they require different measurements and perhaps different research designs.

The second stage makes the explanatory representation explicit. The researcher identifies relevant agents, incentives, constraints, institutional arrangements, feedback, and time dynamics. Formal assumptions should be sufficiently clear to expose their empirical implications. At this stage, the conceptual model need not be maximally detailed. It must include the elements essential to the intended inference and acknowledge the idealizations on which it depends. Morgan's study of models as instruments and Sugden's account of credible model worlds illuminate the central tension between tractable representation and connection to the target domain.

The third stage determines what evidence can establish. In a design-based causal study this involves identifying assumptions, treatment allocation, comparison groups, population selection, and interference. In a structural study it involves the interpretation of parameters, constraints, moments, and equilibrium conditions. In historical work it involves provenance, chronology, measurement comparability, and rival explanations. Researchers should state the boundaries of identification before offering numerical results. A precisely measured association should remain an association until the assumptions necessary for causal interpretation have been justified.

The fourth stage evaluates uncertainty in all its relevant forms. Sampling uncertainty is only one dimension. The conclusion may also depend on choice of functional form, hidden confounding, missing data, measurement validity, researcher decisions, institutional evolution, or behavior under a new policy regime. Robustness exercises, partial-identification bounds, replications, sensitivity analyses, and explicit alternative mechanisms can reveal different facets of uncertainty. The investigator's obligation is to establish which conclusions persist under defensible variations and where the evidence becomes insufficient.

The fifth stage concerns transportability. Even a successfully identified local treatment effect requires further justification before application to another population or a different intervention scale. Relevant moderators may include baseline institutions, prices, participation, compliance, social networks, industrial structure, and enforcement. General-equilibrium changes may alter the intervention itself. Transport requires information about these mechanisms, comparison of domains, and often complementary evidence. An argument about generalizability is a substantive contribution and should be evaluated with the same seriousness as the original empirical identification.

The sixth stage translates evidence into decisions under declared values. A policy judgment requires explicit criteria for welfare, distribution, fairness, feasibility, and the treatment of uncertainty. This stage should preserve the provenance of each premise. An empirically estimated impact should be labeled as such; a forecast under a structural assumption should be identified as a scenario; a normative weight should appear as a value choice. When empirical or normative premises change, the recommendation may change without requiring the earlier research to have been incoherent.

This integrated account is an analytical synthesis of established methodological practices. It organizes the research process for critical interpretation and editorial clarity. It is not presented as a new named theory, mathematical model, priority claim, or doctrine. Its value lies in showing how longstanding methodological debates become tractable when the exact inferential task is kept visible. Disagreement over economic methods often reflects an unspoken shift between different intended claims: understanding a mechanism, predicting an event, identifying a local effect, extrapolating to a new setting, or recommending a social choice.

Consider one integrated example. An observational difference-in-differences analysis may credibly estimate the effect of a wage policy on measured employment among surveyed firms. The assumptions include a suitable counterfactual trend and correct handling of within-region dependence. A behavioral investigation may identify changes in worker search and employer expectations; a structural model may study wage and price adjustments when the policy becomes widespread; an institutional study may document enforcement and informal labor markets; a welfare analysis may weigh employment, income, prices, and distribution. The final assessment gains strength from explicit links between these pieces, while no one piece is made to carry a claim it cannot support.

Conclusion

Economic methodology establishes the conditions under which economic propositions deserve scientific and practical confidence. Its central questions concern the design of explanations, the role of models, the meaning of data, the logic of identification, the interpretation of uncertainty, and the passage from evidence to decisions. The history of economics supplies a rich plurality of methods whose validity is assessed by their logical structure, empirical discipline, substantive relevance, and declared scope.

The principal conclusion of this monograph is that methodological rigor begins with precision about the claim. Formal proof derives consequences under assumptions. Statistical evidence supports inference under sampling and identification conditions. Experiments and quasi-experiments identify specified treatment contrasts when their designs work as intended. Structural models permit evaluation of mechanisms and hypothetical regimes when their invariant relationships are defensible. Historical and institutional inquiry establishes the circumstances that shape the operation of those mechanisms. Behavioral analysis tests the adequacy of assumptions about action. Normative economics supplies decision criteria that must be made explicit.

The scientific strength of an economic claim depends on the connection between these components. A result should state what was measured, how the relevant counterfactual was constructed, what uncertainty remains, and why its conclusion applies to the population or policy under consideration. Transparent sources and reproducible computational procedures help others inspect this connection. Independent studies and competing methods reveal the stability and scope of the finding. A bibliography and evidence ledger preserve the provenance of the underlying claims.

Economic methodology accordingly serves the entirety of economic science. It explains why research traditions can contribute different kinds of knowledge about a common economic world and identifies the additional reasoning needed when conclusions move between theory, observation, explanation, and action. The lasting standard is a traceable chain of inference, from a clearly formulated question to a conclusion whose warrant and domain have been made explicit.

References

  • American Economic Association. 2026. "AEA Journal Policies." Including data legality, conflicts disclosure, and field-experiment registration. Source

  • American Economic Association. 2026. "Data and Code Availability Policy." Updated February 2026. Source

  • American Economic Association. 2026. "JEL Classification System / EconLit Subject Descriptors." AEA, B4–B41. Source

  • Angrist, Joshua D., and Jörn-Steffen Pischke. 2010. "The Credibility Revolution in Empirical Economics: How Better Research Design Is Taking the Con out of Econometrics." Journal of Economic Perspectives 24 (2): 3–30. https://doi.org/10.1257/jep.24.2.3. Source

  • Athey, Susan, and Guido W. Imbens. 2017. "The State of Applied Econometrics: Causality and Policy Evaluation." Journal of Economic Perspectives 31 (2): 3–32. https://doi.org/10.1257/jep.31.2.3. Source

  • Bertrand, Marianne, Esther Duflo, and Sendhil Mullainathan. 2004. "How Much Should We Trust Differences-in-Differences Estimates?" Quarterly Journal of Economics 119 (1): 249–275. https://doi.org/10.1162/003355304772839588. Source

  • Blaug, Mark. 1992. The Methodology of Economics: Or, How Economists Explain. 2nd ed. Cambridge: Cambridge University Press. https://doi.org/10.1017/CBO9780511528224. Source

  • Camerer, Colin F., Anna Dreber, Eskil Forsell, Teck-Hua Ho, Jürgen Huber, Magnus Johannesson, Michael Kirchler, Johan Almenberg, Adam Altmejd, Taizan Chan, Emma Heikensten, Felix Holzmeister, Taisuke Imai, Siri Isaksson, Gideon Nave, Thomas Pfeiffer, Michael Razen, and Hang Wu. 2016. "Evaluating Replicability of Laboratory Experiments in Economics." Science 351 (6280): 1433–1436. https://doi.org/10.1126/science.aaf0918. Source

  • Card, David, and Alan B. Krueger. 1994. "Minimum Wages and Employment: A Case Study of the Fast-Food Industry in New Jersey and Pennsylvania." American Economic Review 84 (4): 772–793. Working-paper version: NBER Working Paper 4509 (1993), doi:10.3386/w4509. Source

  • Deaton, Angus. 2010. "Instruments, Randomization, and Learning about Development." Journal of Economic Literature 48 (2): 424–455. https://doi.org/10.1257/jel.48.2.424. Source

  • Friedman, Milton. 1953. "The Methodology of Positive Economics." In Essays in Positive Economics, 3–43. Chicago: University of Chicago Press. Source

  • Hansen, Lars Peter. 1982. "Large Sample Properties of Generalized Method of Moments Estimators." Econometrica 50 (4): 1029–1054. https://doi.org/10.2307/1912775. Source

  • Hausman, Daniel M. 1992. The Inexact and Separate Science of Economics. Cambridge: Cambridge University Press. https://doi.org/10.1017/CBO9780511752032. Source

  • Hodgson, Geoffrey M. 1998. "The Approach of Institutional Economics." Journal of Economic Literature 36 (1): 166–192. Source

  • Imbens, Guido W., and Joshua D. Angrist. 1994. "Identification and Estimation of Local Average Treatment Effects." Econometrica 62 (2): 467–475. Source

  • Ioannidis, John P. A. 2005. "Why Most Published Research Findings Are False." PLOS Medicine 2 (8): e124. https://doi.org/10.1371/journal.pmed.0020124. Source

  • Kahneman, Daniel, and Amos Tversky. 1979. "Prospect Theory: An Analysis of Decision under Risk." Econometrica 47 (2): 263–292. https://doi.org/10.2307/1914185. Source

  • Keynes, John Neville. 1891. The Scope and Method of Political Economy. London: Macmillan. Source

  • Lucas, Robert E., Jr. 1976. "Econometric Policy Evaluation: A Critique." Carnegie-Rochester Conference Series on Public Policy 1: 19–46. doi:10.1016/S0167-2231(76)80003-6. Source

  • McCloskey, Donald N. (Deirdre N. McCloskey). 1983. "The Rhetoric of Economics." Journal of Economic Literature 21 (2): 481–517. Source

  • Morgan, Mary S. 2012. The World in the Model: How Economists Work and Think. Cambridge: Cambridge University Press. https://doi.org/10.1017/CBO9781139026185. Source

  • North, Douglass C. 1990. Institutions, Institutional Change and Economic Performance. Cambridge: Cambridge University Press. https://doi.org/10.1017/CBO9780511808678. Source

  • Ostrom, Elinor. 1990. Governing the Commons: The Evolution of Institutions for Collective Action. Cambridge: Cambridge University Press. https://doi.org/10.1017/CBO9780511807763. Source

  • Robbins, Lionel. 1932. An Essay on the Nature and Significance of Economic Science. London: Macmillan. Source

  • Rosenbaum, Paul R., and Donald B. Rubin. 1983. "The Central Role of the Propensity Score in Observational Studies for Causal Effects." Biometrika 70 (1): 41–55. https://doi.org/10.1093/biomet/70.1.41. Source

  • Sen, Amartya K. 1977. "Rational Fools: A Critique of the Behavioural Foundations of Economic Theory." Philosophy & Public Affairs 6 (4): 317–344. Source

  • Smith, Vernon L. 1962. "An Experimental Study of Competitive Market Behavior." Journal of Political Economy 70 (2): 111–137. https://doi.org/10.1086/258609. Source

  • Sugden, Robert. 2000. "Credible Worlds: The Status of Theoretical Models in Economics." Journal of Economic Methodology 7 (1): 1–31. https://doi.org/10.1080/135017800362220. Source