RESEARCH DOSSIER
Positive and Normative Propositions in Economic Theory
Abstract. Economic theory produces propositions with distinct logical functions. Descriptions characterize economic states, explanations identify mechanisms, forecasts specify prospective consequences, mathematical results establish implications of assumptions, and normative evaluations rank attainable states according to criteria of desirability. Policy recommendations combine these functions with institutional feasibility and the authority to select objectives. This monograph investigates the conditions under which each form of reasoning is warranted and the circumstances under which transitions between them are legitimate. It reconstructs the historical development of the positive–normative distinction, formalizes conditional economic arguments, examines empirical identification and measurement, and relates welfare economics and social choice to the distributional and intertemporal dimensions of public decisions. The analysis compares documented controversies over minimum wages and the design of education subsidies in Mexico with appraisal practices in the United Kingdom, historical regulatory guidance in the United States, and international environmental appraisal. Explicit numerical counterexamples demonstrate that identical consequences can support divergent evaluations and that identical value criteria can yield divergent decisions when causal beliefs differ. The original analytical contribution is an integrated interpretation of economic disagreement through the traceable separation of observations, conditional mechanisms, causal evidence, evaluative premises, and implementable choices. It establishes that the evidential warrant of a positive proposition and the justification of a normative proposition answer different questions, while economic policy analysis requires their disciplined combination. This is a conceptual and documentary investigation supported by illustrative calculations rather than an original estimation of national economic effects.
Keywords: Positive economics; normative economics; economic propositions; economic methodology; fact–value distinction; causal identification; welfare economics; social choice; Pareto efficiency; distributional evaluation; economic measurement; policy appraisal; uncertainty; economic theory
Foundations and Logical Status of Economic Propositions
The object of inquiry
Economic propositions are the units through which economic knowledge becomes explicit, testable, comparable, and available for criticism. A proposition may concern the production of goods, the distribution of income, the evolution of prices, the organization of exchange, the functioning of financial institutions, or the consequences of collective decisions. The scientific status of the proposition depends on its logical function and warrant. Statements about the same object can require entirely different forms of evidence.
Consider a jurisdiction contemplating a tax on a particular commodity. The observed tax base is a descriptive object; the predicted effect on expenditure is a conditional forecast; the estimated change attributable to the tax is a causal proposition; the equality of tax revenue to the applicable rate multiplied by the taxable base follows from a specified accounting or legal definition; the judgment that the tax is desirable invokes criteria of social evaluation. Enactment requires an additional examination of legal authority, enforcement, administrative burden, and alternatives. A coherent analysis connects all these judgments without allowing the validity of one to stand for the validity of the others.
A positive proposition characterizes, explains, predicts, or otherwise makes a claim about economic phenomena as they exist or would exist under specified conditions. Its empirical content is assessed through relevant evidence and the inferential conditions connecting evidence to the claim. A normative proposition evaluates a state, institution, distribution, action, or rule by reference to a criterion of desirability, obligation, justice, welfare, or rationality. Its justification requires both an intelligible criterion and a valid application of that criterion to its object. A mathematical implication occupies a distinct logical position: its validity is established by deduction from assumptions, while the empirical and normative interpretation of the assumptions and result requires separate justification.
Positive and normative therefore describe functions of propositions. They do not constitute permanent classifications of economists, methods, disciplines, or institutions. A single formal economic treatment may contain descriptive assumptions, proofs, hypotheses about behavior, and an explicit welfare objective. A statistical study may estimate the consequences of a policy that was selected for morally significant reasons. A normative inquiry may involve systematic measurement and rigorous deductions.
Description, explanation, prediction, and evaluation
Description identifies a pattern in a defined population or observation process. An explanation specifies what generates the pattern or why it occurs. A prediction extends a specified relation to events or observations that are prospective or unobserved. Causal analysis concerns the difference between outcomes induced by alternative actions or conditions. Evaluation assigns significance to those outcomes. Conditional mathematical analysis establishes the implications of formal premises and may subsequently serve any of these functions.
The statement that inflation measured by a particular price index was 4 percent over a specified interval has an operational definition, a period, a basket, and a procedure for constructing the index. The claim that monetary contraction caused a lower inflation rate has a counterfactual component: the relevant alternative is the path inflation would have followed without the contraction. A claim that 4 percent inflation is unacceptable presupposes a threshold, an objective, or a description of the harms considered unacceptable. All three propositions can refer to the same published statistic. They require different arguments.
An important corollary concerns the phrase "the economic effect." An intervention has a vector of effects that may include production, employment, income, risk, rights of access, environmental quality, and future fiscal capacity. Selecting one component for research is a decision about the object of the inquiry. Selecting one component as sufficient for a social verdict is a decision about the criterion of evaluation. The two selections must be distinguished.
Scope and research method
This dossier takes economic propositions as its scientific object. It reconstructs the logic of explanatory and evaluative claims, compares important historical texts and institutional documents, uses mathematical counterexamples, and examines contrasting empirical studies. The research question is how the evidence, assumptions, criteria, and inferences supporting an economic conclusion can be made independently inspectable.
The method is a structured conceptual and documentary review. The sources were selected for their direct relevance to the history of economic methodology, the formal treatment of welfare and collective choice, methods of causal identification, and specified applications of appraisal. Historical claims are assigned to dated works and editions. The numerical exercises are hypothetical and fully specified. The empirical illustrations use published study descriptions and results in their documented scope, with no new sampling, econometric estimation, pooled effect size, systematic-review completeness claim, or causal conclusion about unstudied economies. Full-text access was uneven: original passages from Hume, Keynes, Friedman, and Samuelson and scholarly theorem expositions were consulted; some empirical results are discussed at the level of author abstracts and institutional summaries. The accompanying evidence record identifies access levels and checked claims.
Historical Development of the Distinction
The logical inheritance of the is–ought problem
The historical significance of the fact–value distinction lies in its demand that a change in the logical character of a conclusion be accounted for. In Book III of A Treatise of Human Nature, Hume (1739–1740) examined the passage from statements of fact to obligation in the is–ought passage. His observation concerns the appearance of a new kind of relation in an argument. It supplies an enduring methodological question for economic inquiry: which additional premise permits an evaluation to follow from a description?
For example, consider the premises that an intervention lowers a group's real income and that no compensating increase occurs elsewhere. These statements describe consequences. The conclusion that the intervention should be rejected requires a criterion, such as a concern for income protection, an entitlement to the affected resource, or the aim of maximizing material well-being. An economist may make that premise explicit and examine its implications. Its presence changes the structure of the argument from a purely descriptive account to an evaluative inference.
This observation does not require agreement on every issue in moral philosophy. A researcher may accept objective moral facts, an attitude-based account of evaluation, a contractual account of obligation, or institutional authority as the basis of a decision. The methodological requirement is to disclose what warrants the evaluative step.
John Neville Keynes and the threefold division
John Neville Keynes (1891), in The Scope and Method of Political Economy, distinguished the study of existing relations, the inquiry into standards of evaluation, and practical knowledge about attaining ends. The threefold division is more informative for policy analysis than a two-part classification that places every practical statement under normative economics.
The inquiry into the effectiveness of an instrument can proceed with an objective supplied by another institution. Suppose the goal is to achieve a given reduction in traffic deaths. An economic analysis can compare instruments in terms of costs per expected death prevented, reliability, feasibility, and administrative resources. The analysis remains conditional on the target. A further inquiry into whether that target deserves priority over other expenditure is evaluative. A legislative act authorizing funds supplies a separate institutional decision.
Practical competence connects means to ends. It requires knowledge of the initial environment, available instruments, likely responses, and uncertainty. An objective defined at too general a level often leaves unresolved trade-offs. "Improve economic security," for instance, could direct attention toward income stabilization, protection against catastrophic loss, sustainable employment, pension adequacy, or public fiscal resilience. The economist contributes to the specification of those objectives and presents their operational consequences.
Friedman and the scientific appraisal of predictions
In The Methodology of Positive Economics, Friedman (1953) defended testing positive economic hypotheses in relation to their predictive implications and recognized that policy judgments need predictions about alternatives. He explicitly acknowledged that agreement over the consequences of a measure can coexist with disagreement over their desirability. His view therefore separates predictive disagreement from evaluative disagreement even while emphasizing the practical importance of the former.
A stylized model may omit many features of real economies and remain useful for the purpose for which it is constructed. A demand model with a simple functional form may generate useful local forecasts over a limited price range. Its fit over that range does not establish that the model represents the structural responses to an unprecedented rationing regime. A model for assessing a policy change must be examined in relation to the contemplated intervention, the robustness of its mechanism, and its intended domain.
The inferential difference is substantive. Accurate prediction of aggregate sales can coexist with incorrect predictions for firms or household groups. A policy decision resting on distributional effects therefore requires a richer empirical characterization than an aggregate forecasting competition. Predictive performance must be judged at the level relevant to the question.
The integration of facts and values within economic inquiry
The practical relationship between the two classes of propositions is developed in Hausman's (2018) philosophy of economics. Research priorities, interpretations of rationality, the significance assigned to particular outcomes, and the use of preferences as indicators of well-being complicate a simple disciplinary division. An economist who investigates unemployment has selected a phenomenon that is scientifically and socially salient. That selection does not establish the truth of an employment-effect estimate.
The history of methodology thus points toward a distinction at the level of inferential roles. Value-guided research may generate empirically warranted statements; empirical discoveries may constrain or transform evaluative conclusions; a conditional normative theory may possess exact mathematical content. The possibility of interaction is compatible with maintaining standards appropriate to each proposition. The crucial question becomes whether the different components of an argument can be tracked through its assumptions and conclusion.
Formal Representation of Explanations and Evaluations
Conditional economic reasoning
Let a denote a feasible policy, s a state of the economy, and θ the parameters that determine the operation of a specified model M. Let Y denote the vector of outcomes relevant to the question. A deterministic representation is Y = M(a,s,θ). A stochastic representation uses a conditional distribution P(Y | a,s,θ). Both representations convey consequences conditional on the model, state, and parameters. The empirical task is to determine whether this mapping adequately represents the mechanism in the intended domain.
In causal analysis the conditional distribution under an observed action may differ from the distribution generated by an intervention. This distinction arises when the selection of an action depends on characteristics that also influence outcomes. People receiving a subsidy may be poorer before treatment than people who do not receive it. A difference in the observed outcomes of the two groups then combines the potential effect of the subsidy with differences relevant to participation. Appropriate research design supplies conditions under which the causal comparison is warranted.
A policy evaluation introduces a criterion V(Y), a feasible set A, and, under uncertainty, a decision rule. In a simple expected-value representation the recommended action is an element of
The notation makes the dependencies visible. Y(a) represents the potential consequences of each action, V expresses a selected evaluation, and A restricts the available actions. The expected-value rule introduces further assumptions about the treatment of risk and the comparability of values across possible states. The mathematics determines the maximum once those inputs are specified. Empirical investigation supports Y(a); institutional analysis supports A; normative argument supports V and the chosen treatment of uncertainty.
The independence of evaluative and empirical disagreement
An elementary counterexample demonstrates why a positive effect cannot, by itself, determine a social ordering. Assume a feasible baseline A gives two persons real resources of 50 and 50 units. Alternative B gives them 80 and 35 units. These allocations are hypothetical and certain. Total resources rise from 100 to 115 and the lower individual allocation falls from 50 to 35.
| Evaluation criterion | Baseline A | Alternative B | Preferred allocation |
|---|---|---|---|
| Total resources | 100 | 115 | B |
| Minimum individual resources | 50 | 35 | A |
| Total with weights 1 and 3 | 200 | 185 | A |
The weighted resource total for B is 80 + 3 × 35 = 185. Each comparison is calculated from the same two allocations. The disagreement arises from the criterion, not from a different account of what happens. The labels "total," "minimum," and "weighted" identify the evaluative construction; they are not independent descriptions of three economies.
A second counterexample holds the evaluative objective constant. Suppose both analysts favor minimizing a clearly specified adverse outcome under a fixed budget. Analyst X predicts a reduction of 20 units from intervention C and 10 from D; analyst Y predicts 5 from C and 15 from D. With equal costs and identical valuation, X selects C and Y selects D. Here the disagreement is empirical and may be narrowed by better causal evidence. It would be a category error to interpret this disagreement as proof that the analysts have different moral commitments.
The two examples also distinguish internal validity from soundness. Given the premises, the comparisons are correct. Whether the assumed effects accurately describe a real intervention and whether the evaluative criteria deserve adoption are separate questions. A proof of the maximum of a stipulated function therefore does not establish a universal duty to maximize it.
Explicit bridge premises
A bridge premise connects an economic consequence to an evaluative conclusion. It may state that a strict improvement for every affected person counts in favor of the intervention, that preventing severe deprivation merits particular weight, or that a legally protected entitlement restricts the set of admissible actions. The premise can be conditional, pluralistic, or lexicographic; it need not assign a dollar value to every consequence.
Suppose evidence indicates that policy P diverts resources from group G to group H. The words "diverts resources" describe a transfer relative to an explicitly defined baseline. The proposition that P is justified depends on at least the value assigned to the groups' gains and losses, the legitimacy of the initial entitlements, the effects of financing, and applicable constraints. An aggregate surplus may provide evidence relevant to a criterion of efficiency. It cannot establish the legitimacy of the initial distribution or the fairness of an uncompensated loss.
This point is especially important when terms such as efficient, optimal, rational, fair, affordable, viable, or excessive appear without a definition. "Optimal" is meaningful in relation to an objective function and constraints. "Affordable" is meaningful in relation to a budget, a time horizon, and the person or institution responsible for payment. "Excessive" requires a standard of comparison. Scientific language becomes more precise when such standards are supplied where the conclusion depends on them.
Identities, definitions, and theorems
An identity holds under the accounting or mathematical definitions that constitute it. A behavioral proposition describes a response that may fail to occur. A theorem follows from axioms and assumptions that need not hold empirically. These are three different forms of warrant.
For example, a simplified budget identity states that total expenditure equals the sum of component expenditures over the same boundary and period. The identity cannot establish how households alter consumption in response to changes in disposable income. A behavioral estimate is needed. Likewise, a model of competitive equilibrium may establish properties of equilibria under its assumptions without demonstrating that a particular real market satisfies those assumptions.
Formal proofs are essential because they reveal exactly what conclusions follow from selected premises. Their empirical applicability and ethical significance must be established through further argument. In this sense, axiomatic economics can be normatively rich and empirically conditional at the same time.
Positive Economics and Empirical Identification
Observation and the counterfactual
Empirical economic analysis requires a defined outcome, unit, population, time, and comparison. An observed policy outcome often reflects multiple influences. A change in unemployment after a reform, for example, may reflect business-cycle developments, sectoral reallocation, population changes, or shifts in labor-force participation as well as the reform itself.
Define potential outcomes Y_i(1) and Y_i(0) for unit i under treatment and absence of treatment. The individual causal effect is Y_i(1) − Y_i(0). Because only the outcome associated with the realized treatment is ordinarily observed for a particular unit at a given time, inference about the unobserved outcome requires additional design or modeling assumptions.
The World Bank and Inter-American Development Bank's Impact Evaluation in Practice (Gertler et al., 2016) organizes evaluation around counterfactual reasoning and distinct designs, including randomized assignment, instruments, discontinuities, and differences-in-differences. The causal interpretation of each design depends on its own conditions. A well-designed randomized study balances confounding in expectation; attrition, interference and noncompliance can change the estimand or threaten the inference. A regression discontinuity design relies on conditions around a specified assignment threshold. An instrument requires a relationship with treatment and credible restrictions connecting it to outcomes. Such distinctions prevent the method's name from substituting for a substantive identification argument.
A transparent difference-in-differences exercise
Suppose average employment in a treated region increases from 100 to 108 measured units, while employment in a comparator region increases from 90 to 94. A simple difference-in-differences contrast is
Four units is the difference between the observed changes. A causal interpretation requires, among other things, that the untreated outcome in the treated region would have followed an appropriate parallel trend relative to the comparator and that the comparison is not contaminated by spillovers or simultaneous differential shocks. Differences in levels before treatment are compatible with the simple additive parallel-trends condition; differences in unobserved trends threaten it.
Now suppose the relevant unmeasured regional shock would have increased employment in the treated region by three units more than in the comparator even without the intervention. Under that hypothetical alternative, the simple estimate of four overstates the intervention's contribution by three, leaving an effect of one. If the shock were five units, the implied treatment effect would be negative one. The arithmetic is transparent, and the sensitivity is determined by the credibility of the assumptions.
The example illustrates a general point: empirical disagreement can arise from definitions of outcomes, comparison units, time windows, treatment timing, anticipation, measurement error, and identifying assumptions. Sign and magnitude can differ across studies without either study necessarily having committed an arithmetic error. Proper evaluation compares the estimands as well as the estimates.
Precision, accuracy, external validity, and forecasting
An estimate may be precise yet answer a poorly specified question. Sampling uncertainty summarizes variability conditional on a statistical model and research design; identification uncertainty concerns whether the relevant causal parameter is recoverable. Model uncertainty concerns the representation of mechanisms. Measurement uncertainty concerns the relationship between the variable and its intended construct. None of these is resolved merely by reporting more decimal places.
External validity adds a further problem. A locally identified effect can depend on the institutions, market structure, distribution of exposed persons, enforcement, and baseline values that generated it. Moving an estimate from one country to another requires evidence that the important response mechanisms and relevant margins are sufficiently comparable. If the effect of an employment policy depends on employer market power or compliance, a difference in those institutions can alter the expected response.
Explanations, forecasts and causal estimates interact. Mechanistic reasoning may identify where an observed association is likely to change. Forecasting may expose misspecification in models that perform well in historical samples. Experimental or quasi-experimental evidence may constrain plausible mechanisms but still leave aggregate equilibrium responses unresolved. A complete positive proposition states the target of inference and its domain rather than treating any empirical success as a license for unrestricted generalization.
The role of reproducibility
Reproducibility requires the data, definitions, transformations, code, model specification, and decision rules necessary to reconstruct the reported analysis. Replicability requires independent assessment of a finding, potentially with new data or alternative designs. Documentary verification of a publication's reported result is a third, narrower task. This monograph performs the last task for its empirical case studies and illustrates the mathematics of selected methods. It does not reproduce restricted datasets or estimate published coefficients.
A disciplined report distinguishes a result actually observed in the data from an inference dependent on unverified assumptions. It also reports when only an abstract, a publisher summary, or historical source passage was accessible. Such source-level distinctions permit readers to evaluate the strength of a conclusion and to identify where further direct investigation would be valuable.
Welfare Economics and the Evaluation of Allocations
Pareto efficiency and the limits of unanimity
The Pareto improvement relation compares feasible allocations by the individual rankings of the affected persons. If every person weakly prefers B to A and at least one strictly prefers B, B is a Pareto improvement over A, using individual preferences as the stated relation. A Pareto-efficient allocation admits no feasible Pareto improvement. The definitions are conditional on the feasible set, the population and the specified individual rankings.
The inability to compare many allocations is integral to the criterion. In a two-person society with a fixed stock of 100 divisible resource units, strictly increasing preferences over each person's own share, and no externalities, every allocation exhausting the total stock is Pareto efficient. Moving from (99,1) to (50,50) makes the first person worse off and the second better off in terms of the specified shares. Neither dominates the other. The equal allocation is efficient, and the extremely unequal allocation is efficient. Efficiency characterizes opportunities for unanimous improvement; it does not uniquely determine justice.
A geometric statement of efficiency therefore embodies a social interpretation only when paired with further evaluative commitments. Even the proposition that unanimous individual preference improvement is ethically decisive assumes that the preferences, the represented persons, and the described states supply an appropriate basis of evaluation. If an individual's preferences are ill informed or depend on controlling another person's life, the adequacy of that premise becomes an issue.
Individual preference and interpersonal comparisons
Ordinal utility represents a person's ordering of alternatives. A strictly increasing transformation of an ordinal utility index preserves the ranking. This property limits the conclusions that can be derived from numerical utility values across different persons without further structure.
Take two states, A and B, and suppose utility representations are u_1(A)=1, u_1(B)=3, u_2(A)=5, and u_2(B)=4. Summing the displayed numbers gives 6 for A and 7 for B. Transform the first person's utility representation by multiplying it by 0.1, leaving the second unchanged. The transformed sums are 5.1 for A and 4.3 for B. The individual rankings are identical before and after the transformation; the social judgment based on the sum reverses.
The example establishes that adding arbitrary ordinal utility representations does not provide an ordinally invariant social evaluation. It does not establish that interpersonal evaluation is impossible under every conception of welfare. A social welfare analysis may use comparable income resources, equivalent incomes, experienced well-being, or justified cardinal structures, with assumptions stated explicitly. The scientific issue is to determine the informational basis and the permissible transformations rather than to treat arbitrary numbers as interpersonally comparable quantities.
Compensation, distribution, and monetary valuation
Potential compensation tests seek to extend policy comparisons beyond unanimity. A simplified monetary surplus test considers whether the gains of some parties are large enough to cover the losses of others. In a quasilinear hypothetical environment where individual monetary changes faithfully represent welfare-relevant changes, a gain of 80 and loss of 60 imply a positive net surplus of 20. A transfer of exactly 60 from the gainer to the loser would leave the loser at the original position and the gainer with a positive gain of 20.
The existence of this transfer in an abstract feasible set and its actual delivery are different propositions. Administrative costs, legal entitlements, political restrictions, financing, and imperfect observability can make compensation unavailable. Moreover, willingness to pay can depend on wealth and the starting state. Compensation criteria that are well behaved in a simplified example may yield inconsistent rankings in richer settings. Fleurbaey's (2013) survey of economic justice discusses the development and limits of the Pareto and compensation approaches, including the problem of reversals and the role of unequal purchasing power.
When values are reported in currency, the analyst must explain whether the figures are accounting costs, market prices, willingness to pay, willingness to accept, shadow values, or resource opportunity costs. The distinction matters because these magnitudes answer different questions. A fiscal transfer may change the distribution of claims without directly consuming economic resources; collection and administration may create real costs, and behavioral responses may generate additional effects.
Collective consumption and optimality conditions
In The Pure Theory of Public Expenditure, Samuelson (1954) distinguishes individually allocable goods from jointly consumed goods, derives conditions for efficient collective expenditure, and supplies distinct interpersonal ethical conditions for choosing a particular optimal state. The paper presents a particularly clear case in which mathematical efficiency conditions do not by themselves select a unique social allocation.
For a simple formal illustration let G be a shared good, x_1 and x_2 private consumption, and let feasibility be x_1+x_2+cG=R, with c>0. Suppose u_i=x_i+v_i(G) and both v_i are differentiable and concave. Subject to interior feasibility and the stipulated quasilinear utilities, maximizing the unweighted sum reduces to maximizing R−cG+v_1(G)+v_2(G). The first-order condition is
The condition equates aggregated marginal willingness to substitute private resources for the shared good with its marginal resource cost. A weighted maximization or a different representation of preferences and feasibility may involve different distributional conditions; the general efficient public-good provision condition uses marginal rates of substitution, rather than arbitrary utility numbers.
To illustrate, let v_1'(G)=10−G, v_2'(G)=8−G on a domain where both marginal benefits remain nonnegative, and let c=6. Then 18−2G=6, so G=6 units. The second derivative of the reduced objective is −2, so this is its unique interior maximum on a feasible interval containing 6. This calculation assumes that enough resources exist to provide the public good and maintain the private allocations and that the represented marginal valuations are known. How those valuations are elicited, how the expenditure is financed, and what distribution is legitimate are distinct institutional and evaluative inquiries.
The formal condition identifies an economic trade-off. Its use in an actual budget requires the relevant beneficiaries, valuation procedure, feasible technology, and distributive premise. The condition does not constitute a positive forecast that a legislature will enact the efficient amount.
Measurement, Classification, and the Boundaries of Evidence
Economic variables as operational constructions
A statistical variable acquires meaning through its object, unit, boundary, and method of measurement. Gross income differs from disposable income; nominal income differs from inflation-adjusted income; household income differs from individual income. A national aggregate may omit differences in distribution that are decisive for welfare evaluation. A positive statistical proposition needs these definitions even where the measurement conventions reflect a research objective.
The OECD's GDP and Non-financial Accounts documentation describes gross domestic product as an output measure and identifies complementary indicators of household resources, consumption, services and financial behavior. The distinction is instructive: output, disposable income, and household consumption measure different objects. Using the growth of an output aggregate to conclude that welfare has improved requires assumptions about the distribution and significance of the resulting changes.
Consider hypothetical economies A and B, each with two households. Household income in A is (50,50), and in B is (90,20). Aggregate income increases from 100 to 110, but one household experiences a large decline and the lower income falls. A statement about the aggregate is warranted by the calculation. A welfare statement depends on further facts and the selected criterion.
Prices and quantity indices
Inflation statistics summarize changes in prices according to a specified index. Households with different consumption patterns can experience distinct price changes. Suppose good X increases in price by 10 percent while good Y remains unchanged. With fixed baseline expenditure weights of 0.8 for X and 0.2 for Y, the simple weighted price change is 8 percent. With weights of 0.2 and 0.8, it is 2 percent. No contradiction exists: two defined baskets produce different index values.
The example holds quantities and baseline weights fixed and omits substitution, quality adjustment, new goods, and index-number complications. Real price-index construction requires specific statistical methods. The question for economic reasoning is what an index measures and whether its population and conceptual scope match the conclusion.
A normative judgment about price stability may be concerned with the aggregate purchasing power of money, the incidence of essential-price increases, the cost of disinflation, or the distribution of burdens across fixed and variable income groups. Selecting a single index as the complete representation of those concerns introduces a further evaluative and practical premise.
Measurement and endogenous classifications
Eligibility for policies sometimes depends on measured variables. A poverty threshold, credit score, unemployment classification, or disability assessment may determine who receives a benefit. The classification then becomes part of the causal environment. Actors may adapt choices and reporting to the rules, and the observed distribution may change as a result.
An inquiry into whether a threshold correctly implements a policy's intended criterion is not identical to an inquiry into the causal consequences of adopting that threshold. The former requires a relationship between the criterion and the operational indicator; the latter requires evidence about responses and induced outcomes. Both can be tested in appropriate ways. A third inquiry concerns whether the underlying goal and eligibility rule are justifiable.
Scientific measurement should therefore expose the definition and construction of variables even when the primary research question is causal. The expression of a value criterion through an administrative measurement is itself an object of institutional analysis. This is especially important when a result will be transported to countries with different statistical capacities, legal categories, employment arrangements, or household structures.
Empirical Disagreement and the Minimum Wage
Competing mechanisms
Minimum-wage analysis provides a useful contrast between conditional theory, causal evidence, and evaluation. In a simple competitive labor-market model with a downward-sloping demand curve and a binding enforced wage floor, holding the demand schedule fixed, the quantity of labor demanded decreases relative to the unconstrained equilibrium. The prediction follows from the specified mechanism and conditions. It is not a theorem about all labor markets.
Under employer market power, employers may initially pay a wage below the competitive benchmark and choose employment according to a marginal expenditure schedule. In a stylized monopsony model an appropriately chosen wage floor can raise both the wage and employment over a range. At higher floors the employment response may become negative. Compliance, search frictions, adjustment of hours, productivity, price changes, firm entry and exit, and heterogeneous worker exposure can modify the aggregate relationship.
The empirical issue concerns which mechanisms and margins operate in a specified environment. The evaluation concerns how changes in employment, earnings, hours, prices, profits, poverty and the distribution of opportunities are to be assessed. Both issues are relevant to a recommendation but they answer different questions.
The New Jersey–Pennsylvania controversy
Card and Krueger (1994), in their New Jersey–Pennsylvania case study, investigated a rise in the hourly minimum wage from $4.25 to $5.05 in New Jersey on April 1, 1992. Their survey covered 410 fast-food restaurants in New Jersey and eastern Pennsylvania. The study compared employment changes before and after the reform with a neighboring state in which the floor was unchanged. Its reported findings did not indicate employment losses from the increase in the investigated setting.
Neumark and Wascher (2000) subsequently published a comment on the case study in the American Economic Review. The contrary finding from a payroll-record sample is described in the authors' response and related reanalysis. Card and Krueger's (1997) reanalysis with payroll data reported comparable or slightly higher employment growth in New Jersey in broader administrative data and identified sample composition and varying reporting intervals as relevant to the disagreement. These descriptions refer to the published authors' accounts. The present dossier has not independently inspected the underlying datasets or replicated either result.
The methodological importance of the controversy exceeds the initial sign of the reported effect. When a measured outcome depends on survey responses, payroll periods, differences between hours and head counts, establishment selection, and the choice of comparison population, the interpretation of a comparison must identify which employment variable the evidence represents. Testing alternative sources of data informs the positive inquiry. Policy preferences are not evidence for the validity of one measurement approach.
Broader evidence and carefully bounded inference
Cengiz, Dube, Lindner, and Zipperer (2019), in The Effect of Minimum Wages on Low-Wage Jobs, considered 138 prominent US state-level changes between 1979 and 2016. Their wage-bin analysis compares jobs disappearing below the new floor with jobs appearing around or above it. The authors reported an approximately unchanged total number of low-wage jobs over the subsequent five years, with some evidence of adverse employment effects in tradable sectors.
This study is relevant because it changes the analytical unit from a narrow establishment comparison to employment around points in the wage distribution across many policy episodes. Such a shift can expose reclassification and heterogeneity hidden by a single aggregate. It also changes the relevant empirical question and the conditions needed to interpret the result.
These studies do not establish a universal effect for arbitrary wage increases in every country. Differences in labor institutions, the magnitude of the floor relative to prevailing wages, sectoral structure, enforcement, price adjustment, and informality matter. A credible inference for a new jurisdiction would specify these conditions and the anticipated margins of adjustment.
From causal estimates to a policy verdict
Assume a fully hypothetical wage regulation affects two groups, each consisting of one hundred persons. It raises the annual disposable income of group X by an average of 200 units while reducing group Y's annual income by 100 units. Under equal monetary weights and assuming no other effects the total change is 100 × 200 − 100 × 100 = 10,000 units. Suppose the affected groups have very different initial incomes and that the evaluator assigns weight 1 to X and weight 3 to Y. The weighted change becomes 20,000−30,000=−10,000 units.
Neither figure describes an observed effect of any actual minimum-wage law. The example shows why a reported average earnings change does not settle a distributional evaluation. It also shows why an explicit valuation of exposed groups can reveal disagreement about who bears the burden.
A robust analysis would additionally examine uncertainty over the groups' changes, longer-run adjustment, administrative costs and affected people outside the two groups. A finding of a small or statistically indistinct average employment effect does not prove that all people are unharmed or that every associated policy criterion is satisfied. A finding of a negative employment effect does not settle the overall judgment in the absence of a stated treatment of other consequences.
Institutional Comparison and Public Policy Appraisal
Policy means, objectives, and administrative capacity
Public appraisal is a process through which anticipated effects are organized for institutional decisions. It may compare multiple interventions against a baseline, estimate benefits and costs, identify distributional consequences and risks, and disclose feasibility constraints. Different governmental and international institutions may adopt different reporting conventions and objectives.
The logical architecture persists across these settings: a positive claim concerns the anticipated or realized effects; an appraisal convention specifies how they are represented and compared; an evaluative premise supplies the relevant criterion; a governing institution authorizes action within its mandate. The same analyst may participate at more than one stage. Each stage leaves distinguishable evidence.
Mexico: education subsidies and program evaluation
In the evaluation of Mexico's Progresa program, Schultz (2004) examined school enrollment among children in poor rural communities, using comparisons involving communities assigned to the initial program phase and control communities. The author estimated short-run enrollment responses and extrapolated potential longer-term schooling and earnings consequences. The distinction between measured enrollment and projected lifetime returns is crucial for the positive inquiry.
Consider the decision to use a conditional transfer to encourage attendance rather than an unconditional transfer of the same budget. A positive comparison must estimate effects on child enrollment, attendance, learning, household expenditure, labor allocation, administrative compliance and access, among other outcomes relevant to the target. The value criterion may prioritize education, immediate income security, autonomy, or multiple objectives together. Equal program expenditure does not entail equal household welfare because the programs may impose different conditions and affect different persons.
Randomized assignment supports a particular comparison under its design assumptions. Its scientific role is to improve the credibility of causal estimates, not to choose the social purpose of education policy. Extrapolating a schooling effect to adult earnings introduces assumptions about future school completion, labor-market opportunities and returns to education. Those assumptions deserve separate scrutiny when projected benefits enter a decision.
The international implication is methodological. Program results may be transported with caution when eligibility, school supply, parental constraints and implementation conditions are understood. They cannot be treated as universal parameters of poverty policy solely because the original intervention used a strong identification design.
United Kingdom: the social value of public expenditure
HM Treasury's (2026) Green Book provides a concrete example of an appraisal convention that states methods for social valuation and discounting. Its treatment of transfers distinguishes resource valuation from distributional analysis. It describes welfare weighting based on household income and requires weighted estimates to be presented alongside unweighted estimates when weights are applied. The real social time preference rate specified for the first thirty years of appraisal is 3 percent.
These provisions supply an institutional practice, rather than an empirically universal coefficient. Applying an adopted weight to a monetary effect requires observed or estimated income data, the scope of the affected group, and a defined equivalization procedure. The decision to give an additional unit of income different social weight depending on income embodies a distributive premise; the empirical assessment of the affected income groups can improve the calculation without settling that premise.
The treatment of economic transfers supplies another illustration. If a transfer shifts 100 units from one domestic resident to another, there is a 100-unit loss to one and a 100-unit gain to the other. Under an unweighted accounting of the pure transfer the net amount is zero. With weights of 1 for the payer and 2 for the recipient the weighted transfer is +100. Real administrative costs and behavioral consequences must be accounted for separately. Public expenditure and social resource costs can therefore differ without any contradiction in the accounts.
United States: historical regulatory appraisal
The U.S. Office of Management and Budget's September 2003 Circular A-4 is a dated example of guidance that sought distinct presentation of efficiency and distributional effects. In its discussion of distribution, it recognizes that regulatory benefits and costs can fall on different groups and at different times. The document specifies a scope of analysis centered on US citizens and residents, with relevant international effects separately reported.
The historical document illustrates how institutional scope enters economic appraisal. A national regulatory mandate may require an analysis of domestic effects; a global assessment may use a broader population. Identical physical impacts can produce different reported monetary totals if the accounting boundaries differ. For cross-border harms, the selection of the population included in the aggregate may be particularly consequential.
The 2003 circular is used here as historical methodological evidence rather than as a claim about operative US regulatory requirements in November 2026. Institutional standards can be amended or superseded. Application to an actual contemporary regulatory proceeding requires verification of its applicable law and current guidance.
International environmental appraisal
The OECD's (2018) Cost-Benefit Analysis and the Environment examines environmental valuation, treatment of uncertainty and the implications of intergenerational discounting. Its discussion distinguishes the analytical role of appraisal from the political and institutional processes through which policies are selected.
Environmental evaluation often concerns consequences that occur outside the jurisdiction undertaking an intervention and extend over generations. A purely domestic accounting of damages can differ from a global accounting. The scientific investigation of emissions, ecological effects, probability distributions and the consequences of adaptation supplies important inputs. Their monetary valuation and aggregation across groups and generations require further choices.
A comparative study of institutional frameworks should therefore identify the purpose of the guidance, the units measured, the population covered, the decision authority, the handling of nonmonetary impacts, and the treatment of time and risk. The examples from Mexico, the United Kingdom, the United States and the OECD concern different objects: program evaluation, public expenditure appraisal, historical regulatory analysis, and international methodological guidance. They support comparison of inferential roles, not a ranking of national economic performance.
Uncertainty, Time, and Intergenerational Evaluation
The distinct sources of uncertainty
Economic decisions are often made before consequences are known. Empirical uncertainty may concern estimated causal effects, model parameters or future states. Evaluative uncertainty may concern the relative importance of outcomes or permissible trade-offs. Institutional uncertainty may concern enforcement, capacity, financing or future authority.
A complete analysis can represent these uncertainties separately. Suppose two feasible interventions have anticipated net material benefits, in units, under alternative states of the world:
| Action | State X | State Y |
|---|---|---|
| A | 120 | −40 |
| B | 50 | 50 |
With probabilities 0.6 for X and 0.4 for Y, the expected material benefit of A is 0.6 × 120 − 0.4 × 40 = 56. B yields 50. Under expected linear monetary benefit A ranks above B. Under a worst-case criterion B ranks above A because 50 exceeds −40.
The two criteria address the same probability and consequence specification. Their divergence reflects different treatment of uncertainty. If the probability assigned to X is p, the expected benefit of A is 160p−40. A weakly exceeds B under the linear expectation criterion when p is at least 90/160 = 0.5625. A change in the probability estimate across that threshold changes the expected-benefit ranking without changing the criterion.
This example separates empirical sensitivity from decision-rule sensitivity. A risk-averse valuation can also prefer B under suitable curvature and initial wealth. Neither the worst-case rule nor linear expectation is universally justified by the example. The appropriate standard depends on the interpretation of the outcomes, the stakes, and the authority of the decision.
Discounting and intertemporal comparisons
Discounting converts streams occurring at different times to units at a common reference date. Under a constant real discount rate r and real net benefits b_t, present value is
The rate may be based on an institutional convention, an opportunity-cost argument, a social time preference specification, or a relationship to a social welfare objective. The choice affects comparisons between alternatives with different temporal profiles. A positive forecast of future production or damages does not independently determine which rate is justified.
Suppose a hypothetical investment costs 100 units at time zero and produces a single real benefit of 180 units at year twenty. At 2 percent, its present net value is approximately 21.13; at 4 percent it is approximately −17.85. The break-even constant rate is (180/100)^(1/20)−1, or approximately 2.98 percent. All figures use consistent real units and are rounded. The ranking relative to doing nothing changes as the rate crosses the break-even value.
The example is especially relevant to environmental decisions. Future persons can be affected by present choices but cannot ordinarily express preferences in currently observed markets. Market evidence on time preference and returns can inform appraisal, while the ethical treatment of future welfare requires a separate justification. Alternative assumptions concerning future consumption, uncertainty, relative prices and the substitutability of resources may also change the calculated benefit stream itself.
Robustness and multidimensional decisions
Sensitivity analysis varies relevant inputs to reveal where conclusions change. Empirical sensitivity varies quantities such as treatment effects, elasticities and likelihoods. Normative sensitivity varies welfare weights, decision criteria and discounting conventions. Institutional sensitivity varies legal or resource constraints and implementation capacity.
A comparison is robust over an explicitly defined set of scenarios when the relevant ordering persists throughout that set. Robustness is therefore relative to the scope of plausible alternatives. A recommendation may be robust across a substantial interval of empirical parameters but reverse under a different distributive criterion. That is valuable information about the location of disagreement.
Some consequences resist reliable common-unit valuation. An appraisal can then report a multidimensional outcome vector, disclose thresholds or admissibility criteria, and identify dominance where it exists. A single summary number is useful only to the extent that its construction can be justified. Reporting a vector may preserve information that would otherwise disappear inside an aggregation rule.
Original Analytical Synthesis
The structure of an auditable economic conclusion
The examined literature and examples support an integrated interpretation of economic inference. An economic conclusion is strongest when its empirical, mathematical, evaluative and institutional components can each be traced to their own warrant. The task is not to isolate economics from questions of social purpose; it is to make the contribution of each kind of reasoning identifiable.
This interpretation follows from the combined difficulties encountered throughout the monograph. Hume's passage prompts inspection of the bridge premise. Keynes's classification separates explanation, evaluation and the practical pursuit of ends. Friedman's methodological treatment clarifies the role of prediction and recognizes evaluative disagreement. Welfare economics demonstrates the plurality of attainable efficient allocations. Arrow and Sen demonstrate limits on jointly satisfying selected collective decision properties. Empirical case studies reveal that evidence itself has a defined measurement and identification domain. Institutional appraisal documents show where bodies adopt conventions for combining consequences.
The synthesis concerns the organization of these results. It does not introduce an independent economic theory, a named model, or a new doctrine of social welfare. Its analytical contribution is to apply their joint implications to the evaluation of complete economic arguments.
Locating disagreement
A disagreement between economic conclusions can originate in several locations. Parties may disagree about observations because their measurement conventions or samples differ. They may accept the observations while disputing the causal mechanism. They may agree about a mechanism in the historical setting while disagreeing about its transfer to a new institutional environment. They may agree on predicted outcomes while preferring different distributions. They may agree on the criterion and consequences yet disagree about the actions legally or administratively available.
A simple diagnostic table makes these locations visible.
| Contested component | Question that resolves the relevant issue | Appropriate warrant |
|---|---|---|
| Observation and definition | What was measured, for whom, and when? | Source records, operational definitions, data auditing |
| Mathematical implication | What follows from the stipulated premises? | Derivation, proof, counterexample |
| Empirical mechanism | What changes under the intervention? | Research design, causal assumptions, evidence |
| Transfer and forecast | Under which conditions does the mechanism remain applicable? | Institutional comparison, model validation, sensitivity |
| Evaluation | Why does a given outcome count in favor of the action? | Explicit criterion and ethical or institutional justification |
| Feasibility and authority | Which actions can be implemented, and by whom? | Resource, legal, administrative and institutional evidence |
These are different research tasks. For instance, responding to an objection to the distributional assumptions of a welfare calculation by presenting a more precise causal estimate leaves the central objection unresolved. Conversely, responding to evidence that a program failed to achieve its target by reiterating the moral importance of that target leaves its effectiveness unresolved.
The separation allows the economist to make conditional statements where consensus over ends does not exist. Such a statement can identify which policy would follow if a certain criterion and set of consequences were accepted. It can compare several defensible criteria on common evidence. It can also identify why available information fails to support a determinate ordering.
The relationship between empirical progress and evaluative judgment
Positive research can narrow the space of evaluative disagreement by eliminating impossible actions and false descriptions of consequences. If an evaluator favors a policy because it is believed to lower hardship, evidence that the intervention materially increases hardship under relevant conditions may undermine the case for selecting that particular means. The argument changes because one of its empirical premises changes.
Evidence can also reveal conflicts among values previously believed compatible. A conservation program may advance environmental objectives while restricting a community's income opportunities. A more accurate causal account may disclose the magnitude, incidence and persistence of the conflict. The evaluator must then examine how the competing objectives should be reconciled.
Normative inquiry can improve positive analysis by requiring measurement of previously omitted outcomes. A concern about the welfare of borrowers may motivate study of the distribution of debt distress rather than average loan volume alone. An interest in the long-run welfare of future populations can motivate research into long-horizon risks. These research choices affect what is observed and modeled. The validity of each resulting empirical claim still depends on appropriate evidence.
The relationship is therefore asymmetric in a useful sense: evaluations specify relevance, while empirical inquiry constrains the consequences and options to which evaluations apply. In practice the process is iterative. New facts alter the list of salient trade-offs; clearer criteria alter the choice of measurements; institutional alternatives change the available empirical questions.
Conditional consensus and the precision of recommendations
Many economic recommendations can be reformulated as conditional statements with explicit premises. If an authority gives priority to minimizing a specified harm within a fixed budget, if the estimated comparative effects are applicable to the affected population, and if both measures are feasible, then a cost-effectiveness comparison can inform the choice. The conclusion remains linked to these premises.
Conditionality does not make the statement scientifically weak. It identifies precisely what must hold for the recommendation to be warranted. Where empirical evidence is sufficiently strong, a conditional recommendation may be highly reliable within its domain. Where the evaluative premise has been authorized by a competent institution, the analyst may proceed on that basis while making the premise visible.
At the same time, institutional adoption cannot guarantee ethical truth or empirical success. An authorized objective may be subject to normative criticism. A policy selected under a legitimate process may fail because its causal forecast was incorrect. Distinguishing authorization, justification and performance makes it possible to evaluate decisions after implementation without confusing the original mandate with the subsequent evidence.
International implications
The international character of economic knowledge requires careful separation of the invariance of mathematical implications from the variability of institutions, measures and empirical relations. The arithmetic of discounting is independent of country names; the appropriate benefit stream, rate convention, population, resource constraint and legal authority may differ substantially. The formal aggregation difficulties of social choice remain relevant in diverse settings; actual decision rules and rights vary.
Economic comparison therefore becomes more informative when an analyst states which components are transferable and which need new evidence. Carrying a conclusion from one country to another without this decomposition can import a value criterion or institutional assumption under the appearance of empirical generalization. The reverse error is to treat context-specific evidence as incapable of informing other settings. Mechanisms can be compared conditionally when the relevant environments and margins are described.
This approach preserves the global applicability of economic reasoning while respecting the specific domain of each empirical claim. It is particularly useful for the relationship between economic theory, economic methodology, welfare economics, public finance, development economics and regulated institutions. Each field contributes distinctive evidence or analytical techniques to a coherent policy argument.
Open Questions and Research Priorities
The empirical study of value-sensitive measurement
An important research problem concerns how measurement decisions influence the substantive conclusions of evaluation. Analysts can compare alternative operational poverty thresholds, price indices, household equivalence scales, employment definitions, and welfare indicators under common underlying records. This research can quantify the sensitivity of reported distributional findings to measurement conventions.
A further question concerns how administrative classifications influence behavior. When an eligibility cutoff determines a real entitlement, it may alter observed choices and reporting. Empirical investigation should distinguish the effect of classification rules from the underlying distribution that the rules are intended to describe. Such inquiry requires institutional documentation and credible identification rather than merely comparing label frequencies.
Transportability and equilibrium adjustment
Localized causal evidence is indispensable in economic research. Its use in decisions about larger or institutionally different settings remains a continuing challenge. Scaling an intervention can change prices, wages, political incentives, resource costs and the behavior of people outside the originally treated group. Further studies should identify these adjustment margins and compare mechanisms across contexts.
For the empirical cases in this monograph, useful extensions would examine how minimum-wage effects vary with the relative position of the statutory floor, employer concentration, enforcement and informality, and how education-subsidy effects depend on school capacity and household constraints. Such research needs actual data and setting-specific designs. The cited studies provide starting evidence rather than a complete inventory of effects across the world.
Plural objectives and long-run uncertainty
Multidimensional evaluation requires further study of how decision procedures handle competing objectives, irreversible harms, incomplete preference information and long-run uncertainty. Comparative investigation can examine which conclusions persist under different admissible social weights and which depend on particular assumptions.
A useful research design would publish the full consequence vector for feasible actions, separately identify empirical uncertainty, and examine the stability of the ordering under a transparent range of evaluative criteria. Where different rights or mandates constrain admissible actions, the design should specify their effect on the feasible set.
The open scientific question is not whether a universal number can always be assigned to every economic consequence. It is how the relevant information can be retained and used when a collective decision must be made. The development of empirical methods and the analysis of normative commitments both contribute to that inquiry.
Conclusion
Positive and normative propositions perform distinct functions within economic theory. A positive proposition describes, explains or predicts an economic phenomenon, with its empirical warrant dependent on relevant observations, research design, assumptions, and domain of application. A normative proposition evaluates economic states or actions according to a criterion, and its warrant requires the justification and correct application of that criterion. Mathematical propositions establish implications within a stipulated formal environment; policy recommendations connect these forms of reasoning to feasible alternatives and institutional authority.
The historical distinction developed through debates over the transition from description to obligation, the organization of economic science, the role of prediction, and the ethical interpretation of economic analysis. Its contemporary importance lies in the ability to identify where an economic argument obtains its validity. Neither the existence of values in economic research nor the social significance of economic consequences eliminates the need to distinguish their inferential functions.
Welfare economics shows that efficiency alone is compatible with many different allocations. Social choice establishes formal limits on the simultaneous achievement of selected aggregation properties. Empirical identification reveals how strongly causal conclusions depend on measurement, comparison, assumptions and the relevant institutional environment. Appraisal practice demonstrates the operational significance of population boundaries, distributional weights, discount rates and legal constraints.
The formal counterexamples support two complementary conclusions. Agreement about economic consequences can coexist with disagreement about evaluation; agreement about evaluative objectives can coexist with disagreement about economic consequences. The appropriate response to each disagreement depends on its source. Better evidence, clearer axioms, explicit value premises and more accurate institutional descriptions are complementary forms of intellectual progress.
The original analytical synthesis of this dossier is the reconstruction of a complete economic argument through its independently examinable components. Economic research acquires practical force by establishing what is known about consequences, what assumptions sustain the inference, what criteria determine the ordering of alternatives, and what actions are feasible. The precision of these connections makes economic knowledge available for rational criticism and responsible institutional application.
References
Arrow, Kenneth J. 1951 [2012]. Social Choice and Individual Values. 3rd ed., foreword by Eric S. Maskin. New Haven: Yale University Press.
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.
Card, David, and Alan B. Krueger. 1997. A Reanalysis of the Effect of the New Jersey Minimum Wage Increase on the Fast-Food Industry with Representative Payroll Data. Industrial Relations Section Working Paper 393. Princeton University. Subsequently published as the 2000 reply, American Economic Review 90 (5): 1397–1420.
Cengiz, Doruk, Arindrajit Dube, Attila Lindner, and Ben Zipperer. 2019. The Effect of Minimum Wages on Low-Wage Jobs. Quarterly Journal of Economics 134 (3): 1405–1454. DOI: 10.1093/qje/qjz014.
Fleurbaey, Marc. 2013. Economics and Economic Justice. In The Stanford Encyclopedia of Philosophy, Summer 2013 Edition. Stanford University.
Friedman, Milton. 1953. The Methodology of Positive Economics. In Essays in Positive Economics, 3–43. Chicago: University of Chicago Press.
Gertler, Paul J., Sebastian Martinez, Patrick Premand, Laura B. Rawlings, and Christel M. J. Vermeersch. 2016. Impact Evaluation in Practice. 2nd ed. Washington, DC: World Bank and Inter-American Development Bank. DOI: 10.1596/978-1-4648-0779-4.
Hausman, Daniel M. 2018. Philosophy of Economics. In The Stanford Encyclopedia of Philosophy, substantive revision September 4, 2018. Stanford University.
HM Treasury. 2026. The Green Book: Central Government Guidance on Appraisal and Evaluation. London: HM Treasury.
Hume, David. 1739–1740. A Treatise of Human Nature. London: John Noon and Thomas Longman. Book III, part I, section I, paragraph 27.
Keynes, John Neville. 1891. The Scope and Method of Political Economy. London and New York: Macmillan, p. 34.
List, Christian. 2022. Social Choice Theory. In The Stanford Encyclopedia of Philosophy, Winter 2022 Edition. Stanford University.
Neumark, David, and William Wascher. 2000. Minimum Wages and Employment: A Case Study of the Fast-Food Industry in New Jersey and Pennsylvania: Comment. American Economic Review 90 (5): 1362–1396. DOI: 10.1257/aer.90.5.1362.
OECD. 2018. Cost-Benefit Analysis and the Environment: Further Developments and Policy Use. Paris: OECD Publishing. DOI: 10.1787/9789264085169-en.
OECD. n.d. GDP and Non-financial Accounts. OECD Data, dataset description. Accessed November 17, 2026.
Samuelson, Paul A. 1954. The Pure Theory of Public Expenditure. Review of Economics and Statistics 36 (4): 387–389. DOI: 10.2307/1925895.
Schultz, T. Paul. 2004. School Subsidies for the Poor: Evaluating the Mexican Progresa Poverty Program. Journal of Development Economics 74 (1): 199–250. DOI: 10.1016/j.jdeveco.2003.12.009.
Sen, Amartya. 1970. The Impossibility of a Paretian Liberal. Journal of Political Economy 78 (1): 152–157. DOI: 10.1086/259614.
U.S. Office of Management and Budget. 2003. Circular A-4: Regulatory Analysis. September 17, 2003. Washington, DC: Executive Office of the President.
Social Choice and the Limits of Aggregation
Collective rankings
Collective decisions often concern situations that different persons rank differently. A preference aggregation rule converts a profile of individual orderings into a social ordering or a collective choice. Observations about individual choices can supply data for this process. The aggregation rule expresses a further institutional or normative commitment.
Consider three persons ranking three alternatives A, B, and C:
Pairwise majority voting gives A over B by two votes to one, B over C by two votes to one, and C over A by two votes to one. All individual rankings are complete and transitive; the induced strict majority relation is cyclical. Its failure of transitivity arises from the rule applied to the preference profile, rather than from erroneous preference data.
A procedure could fix an agenda, use scores, restrict admissible preferences, select an alternative without constructing a complete social ordering, or employ a different decision arrangement. Each changes a component of the collective decision procedure. Choosing among them requires criteria beyond the observation that each member of society has preferences.
Arrow's impossibility theorem
Arrow's (1951) Social Choice and Individual Values provides a formal generalization of the aggregation problem. The theorem's exact scope is set out in List's (2022) social choice theory survey: for a finite society of at least two persons and a set of at least three alternatives, no social welfare function satisfies unrestricted preference domain, completeness and transitivity of the social ordering, weak Pareto unanimity, independence of irrelevant alternatives, and nondictatorship simultaneously.
The conditions express distinct desiderata. Unrestricted domain requires that admissible complete and transitive individual orderings can be accommodated. A coherent social ordering supplies the required output structure. Weak Pareto requires respect for unanimous strict individual preference. Independence restricts the social comparison of two alternatives to individual rankings of that pair. Nondictatorship excludes the particular kind of unilateral decisiveness specified by the theorem.
The result is a conditional impossibility statement. It is valid within the formal domain. A policy analyst cannot conclude from it that social decision-making itself is impossible, that democracy has no value, or that one specified aggregation mechanism must be adopted. The theorem identifies the need to examine which properties a particular social choice arrangement can and should satisfy under its intended interpretation.
A further consequence concerns the relationship between positive preference information and normative choice. Even perfectly measured individual rankings do not supply a uniquely justified collective ranking. Institutional rules determine the way those rankings are combined, and those rules are themselves objects of positive investigation and normative assessment.
Individual liberty and welfare evaluation
Sen (1970), in The Impossibility of a Paretian Liberal, identifies another conflict among conditions imposed on social preferences. As reconstructed in List (2022), universal domain, acyclicity, weak Pareto, and a minimal liberalism condition giving at least two distinct individuals decisiveness over specified pairs of alternatives cannot all be satisfied.
The significance for economic methodology is that an apparently unobjectionable social evaluation principle can conflict with another attractive principle. The particular mathematical interpretation of individual decisiveness matters. The result is a limitation on the joint satisfaction of formal conditions, not a demonstration that every real-world right conflicts with every improvement in well-being.
In applied economic appraisal, rights and constraints may be represented in the feasible set rather than monetized in an objective. For instance, an action that violates a protected entitlement may be screened out before costs and benefits are compared. A numerical objective can still guide the choice among admissible actions. The difference between treating a right as a constraint and treating it as an ordinary cost is normative and can affect the set of eligible policies even when causal predictions are unchanged.