RESEARCH DOSSIER
Economic Agents and Units of Analysis
Abstract. Economic explanation depends on the correspondence between the entity to which an action is attributed and the entity for which an outcome is assessed. This research dossier studies persons, households, firms, public organizations, states and collectives as distinct economic objects, comparing their decision mechanisms, institutional and statistical boundaries, causal roles and aggregation properties. It reconstructs selected scholarly arguments concerning organizational choice, household allocation, economic coordination and representative agents alongside national accounting and statistical practice. Constructed cases demonstrate how consolidation, nonlinear taxation, changing group composition, internal allocation and interference can change the interpretation of an apparently common economic outcome. The analysis establishes the conditions under which a shift of unit preserves or alters a descriptive, causal or normative proposition. It provides an integrated methodological synthesis without introducing a new named theory or claiming independent empirical estimates.
Keywords: Economic agents; units of analysis; households; firms; institutional units; collective choice; aggregation; causal inference; statistical units; economic methodology
The Explanatory Object
An economic agent is a unit to which an economic action, decision or organized capacity for action is attributed within an explanation. A unit of analysis is the entity about which a particular descriptive, causal, predictive or normative proposition is made. The two can coincide, as when an individual's labor supply responds to a personal wage offer. They can also diverge: a bank's credit committee approves a loan, a legal bank contracts with the borrower, the borrower spends the proceeds, and employment in a region changes. Each proposition concerns a different entity, even though all belong to one sequence of events.
The identification of a unit requires a boundary, an attribution rule and a temporal interval. The boundary specifies membership and resources; the attribution rule specifies which actions or characteristics belong to the unit; the temporal interval specifies when the boundary and rule apply. A corporation that acquires a subsidiary changes its organizational perimeter. A household that splits changes its membership. A jurisdiction that delegates a service to another agency changes the location of authority without necessarily changing its territorial boundary. Stable names therefore need not imply stable units.
Six roles deserve separate treatment: the decision maker, the institution that owns or incurs obligations, the unit observed in data, the unit receiving or assigned an intervention, the unit bearing the outcome, and the aggregation unit. An individual worker may appear in a household survey, receive treatment through an employer, and contribute to an industry total. Assignment at the employer level does not imply that all workers experience the same exposure. An industry mean does not imply the existence of a chooser called the industry.
The scope of this dossier is the explanatory relation among persons, households, firms, states and collectives. It uses conceptual distinctions, historical reconstruction, elementary derivations, institutional comparisons and examples of research design. The numerical examples are constructed, with no implied empirical estimates. The evidence is a selective scholarly corpus; a systematic literature review or fresh econometric estimation is outside the present design. The inquiry evaluates when a change in analytical boundary changes the conclusion.
Historical Interpretations and Model Commitments
The familiar individual choice problem represents a decision by a feasible set, a ranking of outcomes and a rule for selecting an outcome. This representation permits rigorous conditional statements: when relative prices change, a specified decision rule yields a new choice. The historical development of behavioral rational choice, including Simon's (1955) A Behavioral Model of Rational Choice, widened the questions economists ask about decision procedures. For the present analysis, a behavioral rule can be represented explicitly without assuming that its output maximizes a complete preference ordering.
The organization of exchange alters the explanatory object. In The Nature of the Firm, Coase (1937) set entrepreneurial coordination alongside coordination through prices. His 1991 prize lecture revisited the comparative costs of these arrangements. The firm consequently becomes a boundary to be explained as well as a unit used to explain production. North's (1993) Economic Performance Through Time directs attention to institutional incentives and historically constituted beliefs. Taken together, these questions concern the feasibility of action as well as the process selecting it.
A further distinction arose in models of household behavior. Chiappori's (1992) Collective Labor Supply and Welfare models household choices using individual preferences and an efficiency condition. The collective outcome then has structure that an externally observed household budget does not disclose. In the theory of aggregation, Kirman (1992) challenged the representative individual reduction: heterogeneous individual optimizers need not behave collectively in a manner that justifies a single optimizing interpretation.
These lines of inquiry are complementary when their assumptions and objects are made explicit. The same market can contain separately optimizing households, bargaining families, managerial firms and institutions that govern contractual exchange. The scientific task is to determine how these elements act upon one another. A reduction to one level becomes defensible when it preserves the property under investigation within the range of relevant counterfactuals. The mere convenience of representing a complex structure by a scalar objective is insufficient evidence of that preservation.
Persons, Decisions, and Constraints
An individual provides a natural starting point for questions about occupation, consumption, risk, time allocation and welfare. A standard consumer problem selects a bundle x from the feasible set defined by p · x ≤ m, where p is a vector of prices and m is available expenditure. If a utility representation u(x) is suitable, the choice can be written as maximization of u over that set. This statement is conditional on the preferences, the information available, and the relevant feasible set; it establishes a model of choice rather than a general assertion about actual motivation.
Income, credit, time, legal eligibility and information can constrain the same decision in different ways. A person may have sufficient income to pay for a course but be unable to attend because care obligations occupy the required hours. Another may have time but face a borrowing restriction. A policy that subsidizes tuition relaxes the price constraint, whereas a policy offering childcare relaxes a time constraint. If the response is measured only as expenditure, these mechanisms may appear similar. Attendance and time-use records can distinguish them.
An economic role is a component of the explanation. A person may simultaneously be a worker, consumer, debtor, landlord, taxpayer and caregiver. Each role has its own relationships and instruments. A change in an interest rate may raise income on savings and increase borrowing costs; net exposure depends on the person's balance sheet and contract terms. Assigning people exclusively to one role obscures the combination of channels.
Choice also depends on timing. Suppose a worker can choose training today at a cost of 20 units, then earn an additional 30 next period. Under discount factor d, with certain payment and no other consequences, the incremental present value is 30d − 20. Training is strictly preferred on these terms when d > 2/3. The conclusion changes if the worker cannot finance the initial cost, even with a high discount factor. The numerical inequality is a model implication, not a measurement of impatience. In empirical work, observed nonparticipation cannot identify the preference parameter without addressing the constraint and expectations.
The person as a bearer of well-being need not coincide with the person exercising effective economic control. A minor may benefit from a parent's choice; a patient may receive a service selected by a clinician or insurer. An account of welfare must therefore specify both the bearer of the outcome and those with authority to produce it. Interpersonal comparisons introduce additional assumptions beyond the ordinal ranking used to characterize a single person's choices.
Households and Internal Allocation
Households combine resources, obligations and recurring consumption. Housing, heating and some other services are shared; food, health expenditures, time and earnings may be pooled in differing proportions. A household-level budget can describe the feasible total without revealing who controls the resources or receives the benefits.
A unitary model treats a household as if it maximized one objective over an aggregate budget. That representation can be useful for aggregate expenditure predictions under stable conditions. A collective representation instead distinguishes the members' preferences and a mechanism of allocation. Chiappori (1992) gives a formal example in the cited collective labor-supply work. The assumption of efficient household choices is itself substantive: allocations can also reflect coercion, incomplete information, bargaining failure, administrative restrictions or commitments that prevent mutually beneficial changes.
Consider a constructed two-person household with a fixed total consumption budget of 100. Initially members A and B each receive 50. After an intervention the total reaches 120, with allocations (90, 30). The household mean increases from 50 to 60, while B loses 20 units. The change in total resources is 20. Equal division of the increase would instead produce (60, 60). Since these are feasible under the same aggregate budget, the sign of the effect on B is not identified from total expenditure alone.
A second example shows the role of nominal recipient identity. If two payments of equal size reach a household through different members, a model of complete pooling with an unchanged joint objective predicts the same choice set and outcome. When control rights, bargaining positions or commitments vary with the recipient, that prediction may fail. Duflo's (2000, published 2003) South African pension study supplies an empirical research setting for distinguishing transfer receipt from benefits within families. The magnitude and causal interpretation of particular reported estimates require full source inspection; none is imported here.
A household may be a relatively stable accounting unit yet a changing welfare, bargaining or sampling unit. When members migrate, household composition and remittances change together. Researchers must distinguish outcomes for original members from outcomes for the current roster. Conditioning on membership after an intervention can select on its consequences. Longitudinal linkage and explicit rules for entries, exits and split households are essential to meaningful comparisons.
Economies of scale in shared consumption complicate the use of per capita resources. A household with two persons and resources of 100 has less housing duplication than two otherwise comparable single-person households each with 50. An equivalence scale introduces assumptions about needs and sharing; it does not identify actual consumption by each member. The appropriate statistic therefore depends on whether the object is household living standards, personal deprivation or demand for market goods.
Firms, Establishments, and Organizational Authority
A firm conducts recurring productive or commercial activity through an organization of resources, contracts and authority. Its explanatory boundary may be drawn around a decision-making organization, a legal person, a production process or an ownership group. The correct boundary depends on the question. An analysis of legal insolvency follows the entities bearing obligations; an account of plant productivity follows inputs and output at the relevant locations; a study of strategic price setting may require an ownership or control perimeter.
The conventional production model describes a feasible technology and chooses inputs and output using prices and an objective. Let output q=f(K,L) be produced with capital K and labor L, at prices p, r and w. A stylized profit problem maximizes p f(K,L) − rK − wL subject to technological and contractual constraints. This is a useful conditional representation for a specified decision. Its aggregate objective does not by itself identify the managers, owners or other parties whose actions implement the choice.
A firm that delegates decisions can face differing incentives across owners, managers, creditors and workers. Jensen and Meckling's (1976) agency and ownership analysis explicitly brings these relationships into the theory of the firm. For example, a manager compensated for sales may favor a project yielding revenue of 110 and cost of 108 over one yielding revenue of 90 and cost of 70. The former generates profit 2; the latter, 20. A revenue-based ranking and a profit-based ranking can diverge even though each project is evaluated using the same monetary data. Which ranking controls actual choice requires institutional evidence.
Firm boundaries also depend on whether transactions occur internally or through independent suppliers. Imagine a producer buying an intermediate component for 40 and selling final output for 100. The supplier incurs external input costs of 10. After integration, the internal transfer of 40 disappears from consolidated sales, but physical production can remain unchanged. Before integration, combined gross sales are 140; after integration, consolidated external sales are 100. Assuming no other intermediates, combined value added remains 90 in both arrangements. Statistics of gross sales are sensitive to the organization of the reporting perimeter; value added responds differently.
The European Union's statistical-unit definitions distinguish enterprises, groups, institutional units and geographically specified local units. Eurostat's structural business statistics metadata uses enterprise-level information for national business statistics and local units for regional data. A research file containing legal registration numbers is not automatically a file of economically independent enterprises. Mapping legal entities to control groups and establishments is a substantive measurement operation.
Firm heterogeneity also changes aggregation. Two independent firms can have identical production choices yet differ in debt maturity, ownership or exposure to a shock. Their sum is a valid total for a specified flow; a representative firm may be inadequate for default risk, especially when obligations or covenants belong to separate legal entities. An economic group cannot always use assets held by a subsidiary without respecting legal and contractual restrictions. Consolidation serves an accounting purpose; the ability to transfer resources is a separate question.
States, Public Organizations, and Collective Governance
A state is a territorial and political organization whose economic functions are exercised through legally constituted authorities, budgets, agencies and enterprises. In macroeconomic models, a government sector may be assigned a fiscal rule or objective. For analysis of implementation, distinct public bodies, legislatures, financial authorities and levels of jurisdiction often require separate treatment.
The government budget identity connects expenditure, taxation, transfers and financing. A simplified period representation writes G + TR + iB = T + ΔB, where G denotes purchases, TR transfers, iB interest, T revenue and ΔB net debt issuance. This illustrative identity assumes no other financing items or valuation adjustments. It accounts for the uses and sources of funds; it does not specify a behavioral fiscal rule. Deriving a prediction about expenditure requires evidence about decision authority, budget law, political constraints and responses of other agents.
A public authority may define the terms under which private actors choose. A municipal regulator can allocate water-use permits, while separate suppliers decide how to deliver water and households decide how much to consume. The authority's choice is a characteristic of the institutional setting when explaining the household response; it is the dependent decision when explaining the regulator's allocation. Research can shift levels without changing the underlying economic system.
Collectives can organize activities through membership rules, voting, negotiation, monitoring and common ownership. Their behavior may be attributable to a formal association or to overlapping arrangements among users and authorities. Ostrom's (2010) polycentric governance research provides a historical and institutional reason to investigate arrangements extending beyond a simple market versus state classification. A local resource association may set a withdrawal quota, a regional body may monitor aggregate conditions, and national law may define enforcement powers. The effect of a quota depends on how these levels interact.
A group also needs an aggregation or decision rule where preferences conflict. Consider three equal-sized voting constituencies with rankings A > B > C, B > C > A, and C > A > B. Pairwise majority favors A over B, B over C, and C over A. No transitive social ranking reproduces all three majority choices. This constructed cycle shows how a collective decision mechanism can differ from a coherent individual preference order; it supplies no prediction of which outcome would be selected without an agenda, voting procedure or other rule.
Public ownership, authority and accounting classification are distinct. A market producer controlled by government and a ministry can belong to different institutional sectors even though both have public control. The policy effect of a subsidy or mandate may run through ownership, regulation, budget transfers and customer responses. A model grouping them together must explain which relations are being consolidated and which remain causally material.
Institutional Units, Statistical Boundaries, and Residence
National accounts classify entities for purposes of recording production, transactions, wealth and financial obligations. The 2008 SNA concepts in brief, chapter 4, associates institutional-unit status with capacity to transact, own assets and assume obligations. It groups resident units into households, financial corporations, nonfinancial corporations, general government and nonprofit institutions serving households. The United Nations Statistical Commission records in decision 56/101(b) the adoption of the 2025 System of National Accounts as the international statistical standard in 2025. The preceding definitions are explicitly grounded in the 2008 edition; comparisons during the transition must identify the edition and national implementation used.
A statistical household represents combined accounts for members in a manner suited to national accounting. This does not settle how the members negotiate or how individual welfare changes. Similarly, the classification of a business within the corporate sector supplies accounting context for its activities, not evidence that a single managerial objective governs them. Statistical concepts impose useful exhaustiveness and consistency on records; behavioral analysis adds assumptions about decision processes.
Residence and nationality describe different aspects of an entity. An enterprise operating across borders can be owned by investors who reside elsewhere. A statistic of domestic production, a statistic of income accruing to residents, and a statistic organized by ultimate ownership answer different questions. The analytical boundary must follow the outcome rather than assume that a country's political border identifies all relevant agents.
Consolidation cancels transactions and reciprocal claims inside a chosen perimeter. Suppose a group has a parent receivable of 30 against a subsidiary and a corresponding subsidiary payable of 30. In a consolidated presentation, the two internal positions cancel. This operation changes neither the original contracts nor the subsidiary's separate legal exposure. For a question about group exposure to outside parties, consolidation is useful. For a question about the subsidiary's ability to repay an outside creditor, individual accounts and transfer restrictions may be necessary.
The production unit and employment unit can also diverge. According to the ILO's 2023 statistical resolution, informality indicators distinguish the nature of a job from characteristics of the production unit. This makes it possible in principle to separate jobs without specified protections from employment located in informal production units. Applying such categories internationally requires country-specific survey design, legal context and a documented reference period.
A data dictionary should therefore specify whether records refer to persons, jobs, establishments, enterprises, institutional units, groups or territories. It should also state coverage, rules for residency, identification keys, the observation interval, and changes in reporting perimeter. These details provide the foundation for comparisons across time and between countries. Where the source changes from one type of unit to another, the resulting break is a potential change in measurement even if the underlying activity is stable.
Aggregation, Heterogeneity, and Counterexamples
Aggregation is an operation that maps information about constituent units into a statistic or model variable. For a disjoint population with additive outcomes y_i, the sum Y=Σ_i y_i is mathematically well defined. The sum's existence does not establish the existence of a decision maker maximizing an objective over Y. The distinction matters most when the question concerns distribution, policy counterfactuals, interactions or nonlinear transformations.
Let labor income be 20 for one person and 80 for another. Income totals 100 and the mean is 50. An income tax defined by t(y)=max(0,0.2(y−50)) yields payments 0 and 6, for total tax 6. Applying the tax formula to a representative person at income 50, then multiplying by two, yields zero. The difference follows from the nonlinear tax schedule and dispersion in income. It does not indicate that the income mean was incorrectly measured; the mean was insufficient for the specified fiscal counterfactual.
With weights, a distinct problem arises. Two establishments have output 100 and 900, with employment 10 and 30, respectively. Their output per worker is 10 and 30. The simple mean is 20; aggregate output per worker is 1,000/40=25. The weighted mean using employment shares, (10/40)10+(30/40)30, equals 25. Equal establishment weighting answers a question about the typical establishment's ratio; employment weighting answers a question about production per worker across the combined establishments.
Composition changes can reverse a comparison. Suppose a high-productivity group improves from 10 to 11 and a lower-productivity group improves from 4 to 5. In period one the high group accounts for 80% of workers, so aggregate productivity is 8.8. In period two it accounts for 20%, giving aggregate productivity 6.2. Every group improves while the total falls. A comparison of aggregate outcomes must identify changes in within-group performance and composition before assigning a behavioral interpretation.
Coordination introduces additional structure. Consider two firms, each deciding whether to adopt a technology. Adoption yields 4 to each if both adopt, −1 to an isolated adopter, and 0 to a nonadopter. Under simultaneous choices with common knowledge of these payoffs, both adopting and neither adopting are pure-strategy Nash equilibria. The total payoff is 8 in the first and 0 in the second. Identical firm-level preferences and a common opportunity remain compatible with different collective outcomes. This example illustrates why Kirman's aggregation critique is relevant to coordination questions without claiming that representative-agent models are universally unusable.
The general consequence is conditional: whenever an aggregate representation is used to answer a micro-level or intervention question, the investigator must establish the relevant aggregation identity or invariance. A good fit to historical aggregate totals alone does not demonstrate that the representation preserves responses to a tax, price or change in the distribution of resources.
Causal Identification and the Unit of Observation
An observation is a recorded measurement; a unit of intervention is the entity at which treatment is defined or allocated; a unit of outcome is the entity for which a consequence is assessed. A causal estimand specifies the population, intervention contrast, outcome and aggregation rule. An analysis must align these elements with the process that generates the data.
A school-based financial literacy program may assign treatment to schools, deliver instruction to classes and record savings decisions of individuals. If treatment is randomized at school level, the independent assignment units are schools. A regression treating pupils as independent randomized units can underestimate uncertainty when their errors covary within schools. Cluster-based analysis, randomization inference or an appropriate hierarchical approach requires knowledge of the assignment process and cluster sizes. Merely clustering automatically at a convenient administrative level does not establish identification.
Even with random assignment, interference can matter: one person's outcome can depend on another person's treatment. Hudgens and Halloran's (2008) causal interference framework addresses causal analysis under this complication. In an illustrative workplace, participants who receive a training subsidy may teach their untreated colleagues. An untreated worker in a treated workplace does not necessarily represent the outcome of a worker in an entirely untreated workplace. A study must define the exposure contrast and, where relevant, estimate direct and spillover effects separately.
A social intervention also changes measurement and participation. If a formalization policy induces businesses to register, the post-policy register contains firms that would previously have been absent. Comparing average productivity among registered firms before and after the policy can combine behavioral change with selection into the database. A population-based survey, linked histories or a careful estimate of the changing registration margin can clarify the distinction. No data transformation alone repairs the absence of a counterfactual.
Ecological inference is a related hazard. A district with higher average earnings and higher educational attainment need not exhibit the same earnings-education relationship among individual residents. Group composition, labor demand and sorting can generate between-district correlations distinct from within-district relationships. The district coefficient describes an association at the district level unless a justified identification strategy supports an individual interpretation.
Panel data need consistent identity rules. When two firms merge, legal identifiers and establishment identifiers may persist differently. Treating the surviving entity's post-merger sales as the growth of the pre-merger firm confounds expansion with reassignment of productive units. Likewise, tracking households by the original reference person can alter their represented population as households split. The investigator should predefine whether the analysis tracks persons, an initial cohort, firms, establishments or stable economic activities.
Causal research has two distinct requirements: a credible counterfactual and adequate unit attribution. An experiment with credible assignment cannot answer a question about an outcome it never measures; detailed accounting cannot create exogenous variation. A persuasive design specifies the mapping between the unit assigned, the units exposed, the outcome recipients and the estimand's weighting convention.
Comparative Measurement and Institutional Variation
Cross-country research benefits from common classifications and from documentation of their local implementations. Comparing household savings rates, corporate profitability or employment requires consistent denominators, institutional coverage, price bases and periods. Harmonized labels are a starting point; differences in economic organization and reporting practice remain relevant.
The U.S. Bureau of Labor Statistics distinguishes the concepts underlying household and payroll employment surveys. One count concerns employed persons, the other payroll employment. A worker holding two payroll jobs can contribute one person to a person-based measure and two jobs to a job-based measure. Coverage differences also matter. Divergent series should be examined with their respective definitions before being attributed to changes in the economy.
The European statistical system supplies a complementary example of disaggregation by production location and enterprise. In a firm with plants in two regions, recording all employment at headquarters may distort a regional production analysis. A local-unit definition can address geographical attribution. Conversely, a group-wide strategy study may need consolidated decisions and ownership information, which location-specific counts cannot supply by themselves.
Households in migrant-sending and migrant-receiving communities may organize resources across borders. A household survey can miss members temporarily absent, while administrative records can miss informal transfers. In studying remittances, the sender, transaction, recipient, household consumption and regional effects require different records. An intervention that reduces transaction fees may affect the amount sent, the number of transfers or recipients' access; each represents a different unit-level outcome.
In a common-resource setting, an institutional analysis compares users' access rights, the ecological stock, monitoring arrangements and the authority charged with rule enforcement. A survey of fisher incomes provides evidence about outcomes for fishers, whereas biological stock measurements concern a resource system and enforcement records concern an institution. The combined study requires an explicit temporal and geographic correspondence. It cannot assume that a political jurisdiction coincides with a moving ecological population.
These examples establish a comparative design rather than a global ranking. They identify questions that can be investigated wherever appropriate microdata, administrative records and institutional descriptions exist. Extrapolation across jurisdictions requires evidence about the corresponding mechanism, not an assumption that a statistically equivalent category always behaves equivalently.
Normative Analysis and Original Analytical Synthesis
An economic explanation can correctly characterize a decision and still leave the distribution of its consequences unresolved. The unit authorized to allocate resources may differ from the units whose outcomes enter welfare evaluation. An organization can improve its measured performance while some members lose; a government can increase aggregate income while shifting costs among communities. The analytical consequence is that efficiency, incidence and accountability should be evaluated with explicit units and weighting rules.
Consider two policies with identical total benefits of 100. One allocates (50,50) to two groups and the other (90,10). The total-benefit ranking regards them as equal. A welfare criterion giving greater weight to the second group can prefer the first. The numerical result depends on the weights: using weights (1,2), the first has value 150 and the second 110. This is a conditional illustration of a normative decision, not an assertion that those are the correct social weights. The ethical or political grounds for weighting are part of the substantive argument.
The preceding chapters support a connected interpretation of economic agency. An actor's capacity to choose depends on feasible actions and institutional authority. An accounting boundary records responsibilities and transactions. An observation boundary determines what the researcher can see. A causal design determines which contrast can be supported. An aggregation rule converts constituent measurements into a statement about a wider population. Each of these operations preserves certain properties of the underlying system while potentially discarding others.
From this interpretation follow three concrete tests for analytical practice. First, trace the mechanism to the entity that can act: a policy transmitted through corporate lending decisions cannot be explained completely by describing borrower preferences alone. Second, test whether the measure changes under a plausible boundary rearrangement with activity held fixed: the integration example shows why gross sales can move even when value added remains constant. Third, compare an aggregate conclusion with its constituent outcomes: the intrahousehold example and nonlinear tax example establish precisely where totals fail to determine incidence. These are methodological judgments derived from the analysis, not a claim to a novel formal theory.
When two studies report apparently conflicting results, their units can explain some of the disagreement. A rise in payroll employment with unchanged employed-person count may signify more jobs per person or other coverage differences. Rising enterprise profits and falling wages refer to different claimants within an economic organization. A fall in average household expenditure with stable incomes among original household members may reflect changing membership. These are hypotheses for reconciliation; determining the actual cause requires data.
This dossier's authorial contribution is the integrated comparison of attribution roles, accounting perimeters, decision mechanisms and causal estimands, supported by independently constructed counterexamples. Its strongest conclusion is about the logic of inference: the information required to defend a proposition may exceed the information needed to compute the aggregate it mentions. The value of the distinction lies in the ability to diagnose concrete measurement and inference errors, and to choose data and designs suited to the question.
Research Limits and Open Questions
Several questions merit further inquiry. The first concerns dynamic boundaries in economic systems. Cooperation, marriage, migration, merger, entry and exit may change units as a consequence of the processes under investigation. Research designs should explicitly examine how alternative identity rules alter estimated treatment effects or growth measures.
The second concerns digital platforms and delegated decision systems. A transaction can be selected by software, executed on behalf of a business, charged to a household and governed by platform rules. The appropriate economic unit depends on the locus of contractual liability, control, information and outcome. A comparison of automated decision arrangements requires documentation of actual authorization mechanisms rather than attribution of human characteristics to software.
The third concerns internationally comparable microdata connecting people, households, jobs, establishments, enterprises and ownership networks. Administrative linkage can strengthen measurement, but coverage, consent and confidentiality limit permissible use. A study must document how it handles unlinked records, informal activities and historically changing registration rules.
The fourth concerns the limits of representation. Researchers can test whether a representative unit preserves a model's relevant predictions by comparing simulations under alternative distributions and interactions. Good agreement over one intervention range does not establish invariance under a different policy or institutional regime. Identifying the conditions of valid simplification is a scientific task in its own right.
The study uses selected academic and statistical material and contains no new empirical estimation. The 1991 Coase lecture is consulted through the authorized reprint; the adoption of SNA 2025 is verified against the Statistical Commission’s decision, and the distinction between job and production-unit informality against the 2023 ILO statistical resolution. Institutional-unit definitions remain grounded in the World Bank’s 2008 SNA summary. Relevant passages of that summary were inspected in two directly accessible official-hosted text extractions; page-image retrieval remains unavailable. Other academic sources are used at their documented text or abstract access levels. These limits define the scope of the present analytical synthesis, including its selective evidence base, absence of original empirical estimates, and restriction to the reported adoption of the new national-accounts standard rather than its implementation.
Conclusion
The unit of analysis is a substantive component of an economic proposition. Identifying the decision maker, the owner of resources and obligations, the entity recorded in data, the subject of intervention and the bearer of consequences establishes the structure within which the proposition can be evaluated. Persons exercise choices under constraints; households organize shared resources and internal allocation; firms combine production with organizational authority and contractual relationships; states and collectives operate through institutions and decision processes. Their boundaries depend on economic functions and the object of research.
A correctly calculated aggregate can leave the incidence, mechanism and counterfactual unresolved. The constructed examples show that group composition can reverse a productivity trend, nonlinear rules can make the average inadequate, household resource gains can coexist with individual losses, integration can change reported sales without altering production, and collective decision rules can yield outcomes that differ from coherent individual rankings. These are logical results under stated assumptions, and they identify questions for empirical investigation rather than empirical findings themselves.
Economic research gains explanatory precision when its accounting, statistical, behavioral and normative units are connected explicitly. The selection of the unit becomes part of the explanation and the research design, subject to evidence and appropriate limits of inference.
References
Coase, Ronald H. (1937). The Nature of the Firm. Economica, 4(16), 386–405. https://doi.org/10.1111/j.1468-0335.1937.tb00002.x. Source
Coase, Ronald H. (1991). The Institutional Structure of Production. Nobel Prize lecture, December 9, 1991. Authorized open-access reprint with minor authorial changes, Handbook of New Institutional Economics. doi:10.1007/978-3-031-50810-3_3. Source
Chiappori, Pierre-André. (1992). Collective Labor Supply and Welfare. Journal of Political Economy, 100(3), 437–467. https://doi.org/10.1086/261825. Source
Kirman, Alan P. (1992). Whom or What Does the Representative Individual Represent? Journal of Economic Perspectives, 6(2), 117–136. https://doi.org/10.1257/jep.6.2.117. Source
World Bank. (2014, UN catalog release). The 2008 SNA – Concepts in Brief: A Complement to the System of National Accounts 2008. Prepared for the World Bank by Anne Harrison, chapter 4. Source
United Nations Statistical Commission. (2025). Report on the fifty-sixth session (4–7 March 2025). E/2025/24–E/CN.3/2025/37, decision 56/101(b), p. 7. Source
European Union. (n.d.). EU Production System — Statistical Units. EUR-Lex summary of Council Regulation (EEC) No 696/93. Source
Eurostat. (n.d.). Structural Business Statistics (SBS): Reference Metadata in Euro SDMX Metadata Structure. Source
Jensen, Michael C., and William H. Meckling. (1976). Theory of the Firm: Managerial Behavior, Agency Costs and Ownership Structure. Journal of Financial Economics, 3(4), 305–360. https://doi.org/10.1016/0304-405X(76)90026-X. Source
North, Douglass C. (1993). Economic Performance Through Time. Nobel Prize lecture, December 9, 1993. Source
Ostrom, Elinor. (2010). Beyond Markets and States: Polycentric Governance of Complex Economic Systems. American Economic Review, 100(3), 641–672. https://doi.org/10.1257/aer.100.3.641. Source
Duflo, Esther. (2000). Grandmothers and Granddaughters: Old Age Pension and Intra-household Allocation in South Africa. NBER Working Paper 8061. https://doi.org/10.3386/w8061. Published version (2003), World Bank Economic Review, 17(1), 1–25. https://doi.org/10.1093/wber/lhg013. Source
Hudgens, Michael G., and M. Elizabeth Halloran. (2008). Toward Causal Inference With Interference. Journal of the American Statistical Association, 103(482), 832–842. https://doi.org/10.1198/016214508000000292. Source
U.S. Bureau of Labor Statistics. (n.d.). Comparing Employment from the BLS Household and Payroll Surveys. Source
International Labour Organization. (2023). Resolution concerning statistics on the informal economy. 21st International Conference of Labour Statisticians, ICLS/21/2023/RES. I, October 2023, paras. 1–2. Source
Simon, Herbert A. (1955). A Behavioral Model of Rational Choice. Quarterly Journal of Economics, 69(1), 99–118. https://doi.org/10.2307/1884852. Source