RESEARCH DOSSIER
Structural Change and Sectoral Reallocation
Abstract. This monograph investigates the economic mechanisms through which employment, productive resources and output change their distribution across industries. It develops exact additive growth-accounting identities, examines the interaction of heterogeneous productivity growth, nonhomothetic demand, relative prices, trade, production networks, labor mobility, capital accumulation and task allocation, and reviews selected international evidence. The analysis distinguishes realized compositional contributions from causal counterfactuals and marginal allocation returns. A worked numerical example establishes how alternative decomposition conventions assign interactions. The central analytical conclusion is that sector shares are endogenous outcomes of decisions made within a dynamic production system: their consequences depend on the productive capacities, market opportunities, adjustment costs and distributional conditions of the transition. The research is a theoretical and literature-based synthesis using checked bibliographic, abstract-level and institutional records, without claiming new econometric estimates or external peer review.
Keywords: Structural change; sectoral reallocation; multisector growth; labor productivity; relative prices; factor mobility; structural transformation; growth accounting; industrialization; services; economic development; causal identification
The economic object
Structural change is the evolution of the composition of economic activity. Sectoral reallocation is a process through which employment, hours, capital, intermediate inputs and newly established productive capacity change their distribution across industries. The object of this dossier is the relationship between the two: how decentralized decisions become aggregate compositional change, and what consequences follow for economic performance. Herrendorf, Rogerson, and Valentinyi (2014) organize growth and structural transformation around a multisector extension of aggregate growth analysis. That orientation makes the choice of aggregation a theoretical decision rather than a merely editorial convention.
Let the economic system contain industries indexed by i. Every industry has an output concept, a boundary, an input vector, a price system and an institutional environment. Production and consumption classifications differ because households consume products that industries jointly produce. Firms alter their product mix, buy services previously performed internally, and expand by establishing new plants. The boundaries of a statistical sector are therefore narrower than the boundaries of an economic activity. A factory that contracts its logistics to another company can reduce measured manufacturing employment without changing how goods are transported.
Three dimensions of composition require separate treatment: the allocation of inputs, the allocation of output and the distribution of incomes. A higher service employment share can coexist with faster manufacturing output growth and rising industrial profits. An economy can industrialize in volume terms while industrial employment contracts. The research question must specify which of these outcomes requires explanation. Otherwise an observation about employment may be mistaken for a claim about production or social welfare.
This study treats aggregate productivity as a useful but incomplete performance indicator. The allocation mechanism includes opportunities available to workers, the costs of moving, the capital required by receiving firms and the demand that sustains their production. Its performance is evaluated along output, time, distribution and feasibility dimensions. The empirical literature is interpreted at the level its designs warrant; the formal examples below are illustrative deductions, not historical observations.
Historical foundations and stages of development
The classical development problem concerns the coexistence of productive systems operating under different technical and social conditions. In Lewis's 1954 article, the availability of labor to a growing capitalist sector permits an expansion of productive employment before rising wages necessarily exhaust profit-financed accumulation. The article's historical importance lies in linking the labor supply facing modern production to capital accumulation. The model does not require every worker outside the expanding sector to have zero value, nor should a fixed subsistence wage be treated as an unconditional empirical fact.
The classical narrative gives a specific causal sequence: profitable investment expands labor demand, workers leave lower-income activities, and an expanding capital stock makes the process sustainable. The sequence becomes conditional once workers possess differentiated skills, households combine several forms of employment, and wage setting reflects urban living costs. Industrial employment may fail to expand if machinery substitutes for labor; profitable capital accumulation can occur in activities employing relatively few people. An assumption about elastic labor supply is consequently meaningful only alongside an assumption about the demand for that labor.
In contemporary multisector theory the same historical question is recast as a dynamic equilibrium problem. Production technologies evolve, household expenditures shift, and capital moves between uses. A growing economy can preserve some aggregate ratios while its occupational composition changes substantially. A broad phase description of agriculture, industry and services is useful for historical comparison; it is insufficient to establish a universal necessary sequence of development. Historical stages are descriptions at an aggregation level. They are not laws defining which sectors each economy must enter.
The boundary conditions of the Lewis mechanism are observable in principle: wage responses, vacancy growth, labor-force participation, rural earnings and the rate of capital formation. A sustained increase in outside wages may reflect the depletion of a labor reserve, rural productivity improvements, migration restrictions or demographic change. The observed wage movement alone does not identify which process dominates. The historical tradition is most productive when it generates distinctions that modern data can investigate.
Solow (1956), in A Contribution to the Theory of Economic Growth, establishes an aggregate growth framework in which factor substitution changes the stability implications of fixed production proportions. Its abstraction is productive for studying capital accumulation but leaves the internal composition of output undetermined. A multisector analysis must identify the production technologies, expenditure shares, relative prices, and factor mobility conditions that generate the aggregate trajectory. A steady aggregate ratio can be compatible with continuing changes in industrial composition.
National accounts, industries, and measurement
Gross value added is gross output less intermediate consumption. An industry can create a great deal of gross output using purchased inputs and still account for a modest amount of value added. Double-counting intermediate transactions makes gross-output shares unsuitable substitutes for GDP shares. A distinction is also needed between industries and institutional sectors such as households, corporations and government. The latter classify institutional units according to their economic role, whereas an industry classification concerns production activities.
The United Nations 2025 national accounts standard was adopted in 2025. Adoption of an international framework does not imply harmonized immediate implementation in all countries. Empirical studies must document the accounting vintage that governed the reported observations. Series revised under different benchmarks may show changes in level or coverage even when firms have not changed output. Revisions are part of the evidence, not a nuisance to be silently discarded.
Write V_i for nominal value added and Q_i for an additive fixed-price volume measure. Nominal output shares are V_i divided by total nominal value added; employment shares are L_i divided by total employment. The ratio V_i/L_i measures nominal value added per employed person. A comparable real labor-productivity measure uses Q_i/L_i under a consistent volume and classification system. Hours, persons and jobs are distinct denominators. A worker with two jobs may appear twice in an establishment count but once in a labor-force count. Hours per person can change because of part-time work, seasonality and shifting self-employment.
A fixed-price system has an advantage for exact additive identities but may be sensitive to its reference prices. Chain-linked series more readily accommodate changing relative prices; components may not add to a published aggregate. Research must use an officially consistent contribution method or construct a documented additive aggregate before applying identities requiring additivity. Comparing a nominal share with a volume series is appropriate when the purpose is to isolate valuation effects; substitution of the measures is inappropriate.
Nonmarket output, imputed rents, unpaid household production and agricultural subsistence production particularly complicate comparisons. Real estate value added can contain income attributable to the use of dwellings and need not vary proportionately with employment. A high ratio of sector output to persons employed can reflect measured capital services rather than extraordinary worker productivity. Sector definitions, valuation and the numerator-denominator match precede every substantive interpretation.
Exact productivity decomposition
Assume the specified real value-added series is additive and total employment is positive. Define sector productivity p_i = Q_i/L_i and employment share s_i = L_i/L, with L = sum_i L_i. For sectors with positive employment, aggregate productivity P = Q/L satisfies
P = sum_i s_i p_i.
Taking the difference between dates 0 and 1 and expanding the product gives an exact identity:
P_1 - P_0 = sum_i s_i0 (p_i1 - p_i0) + sum_i p_i0 (s_i1 - s_i0) + sum_i (s_i1 - s_i0)(p_i1 - p_i0).
The first term values changes in sector productivity using initial employment shares; the second values share changes using initial productivity; the third is an interaction. An equally exact two-term midpoint decomposition uses arithmetic means s_bar_i and p_bar_i:
P_1 - P_0 = sum_i s_bar_i (p_i1 - p_i0) + sum_i p_bar_i (s_i1 - s_i0).
This follows by subtracting the two products and collecting terms; half the interaction is assigned to each component. Because changes in the s_i sum to zero, a common scalar can be subtracted from all p_bar_i without affecting the reallocation component. A uniform increase in all sector productivity levels, with shares fixed, therefore creates within-sector change without a composition effect. Both decompositions are invariant to the choice of a common additive origin for the productivity levels in the allocation term.
The decomposition is an accounting identity. It reports the contribution of observed changes conditional on measured levels and a chosen convention. It cannot by itself describe what output would have been if workers had not moved. Productivity may respond to migration, capital deepening, job matching or changes in demand. The interaction term registers simultaneous changes; it does not identify an interaction coefficient in a structural equation.
When initial or final employment is zero, a sector productivity ratio may be undefined. A researcher can aggregate to a stable classification, define entry and exit contributions in levels, or use a transparent convention if compatible with the data. Inventing productivity at an empty sector creates a counterfactual parameter rather than an accounting observation. A defensible comparison should record the treatment of entrants, exits, industry reclassification and nonadditivity.
Worked example and alternative decompositions
Consider an artificial economy with agriculture, industry and services. Employment is 100 persons at both dates. Output units are additive at constant prices and the example contains no empirical statistics.
| Industry | Initial workers | Initial output per worker | Final workers | Final output per worker |
|---|---|---|---|---|
| Agriculture | 60 | 2 | 40 | 3 |
| Industry | 25 | 6 | 30 | 7 |
| Services | 15 | 4 | 30 | 5 |
Initial output is 120 + 150 + 60 = 330. Final output is 120 + 210 + 150 = 480. Aggregate output per worker changes from 3.3 to 4.8; the level gain is 1.5, or approximately 45.45 percent relative to the initial level.
Because every industry's productivity increases by one unit, the initial-share within component is 1.0. The initial-productivity-weighted share component is (-0.20)(2) + (0.05)(6) + (0.15)(4) = 0.50. The interaction is zero because all productivity changes equal one and total share changes sum to zero. Midpoint productivity weights also yield 0.50 as the compositional component. The direct total and both decompositions balance exactly.
Now let receiving service firms use substantial additional capital and suppose the observed service productivity rises to 8 rather than 5. The new final aggregate is 5.7. The initial-share within term becomes 1.45, the initial-productivity reallocation term remains 0.50, and the interaction becomes 0.45. Their sum is 2.40. The midpoint allocates half of the 0.45 interaction to each component, yielding 1.675 and 0.725. Neither accounting convention reveals how much of the final service productivity is attributable to the additional capital. A counterfactual intervention cannot be inferred by relabeling one of the decomposition terms.
The example also clarifies the role of a changing worker denominator. If final labor input were 120 persons rather than 100, the same relative shares and sector productivities would imply the same aggregate productivity but different total output. Growth in total output equals growth in labor input multiplied by growth in output per worker in gross growth-factor terms. The research design must separate extensive labor utilization from the intensive performance measure. All figures here are conceptual and reproducible using ordinary arithmetic.
Supply differences and technological growth
A sector's technology defines feasible input-output relationships. Under a simplified one-factor technology q_i = a_i l_i, where a_i is physical output per worker, perfectly competitive firms with freely mobile identical labor imply a common wage w and prices p_i = w/a_i. A sector's physical productivity advantage lowers its unit cost and equilibrium price relative to other sectors. It does not necessarily raise its nominal value added per worker in this setting: p_i a_i = w. This small example explains why physical productivity and nominal labor productivity cannot be used interchangeably.
Ngai and Pissarides (2007) investigate multisector balanced growth under heterogeneous technological progress. Their result addresses compatibility between aggregate growth regularities and a changing labor allocation under specified conditions. The economy-wide consequences of faster productivity growth depend on the response of demand and the use of the resulting output. With a common wage and demand that expands less rapidly than productivity, workers may leave the technologically improving industry even while its physical production expands.
Sector-specific technology may instead be complementary to labor. A new production process can increase the quality of output, lower its effective price and attract demand from previously unserved buyers. Capital goods whose supply improves can transmit productivity growth into industries that purchase those goods. Differentiated technologies have different capital requirements and skill compositions. Sectoral productivity growth therefore cannot be reduced to a single autonomous shifter.
The key comparison concerns output elasticity with respect to the cost reduction, rather than productivity growth by itself. At an unchanged demand for physical output, a doubling of output per hour halves the required hours. If the lower price more than doubles demanded output, the same innovation can expand employment. The aggregate implication depends on income effects and substitutions throughout the economy. That dependence motivates treating production and demand jointly.
Baumol (1967) formalizes unbalanced growth through activities with different rates of labor-productivity improvement and wages that move together. Let labor productivity in two activities be A and B. With a common wage w, unit labor costs are w/A and w/B, and their ratio is B/A. For a fixed positive ratio of physical outputs, relative labor requirements are inversely proportional to relative productivity. The slowly improving activity therefore requires a rising share of total labor if the output ratio is maintained. Changes in demand, labor supply, production methods, or wage-setting institutions can modify that result. The cost mechanism and employment response are analytically separate.
Demand, income, and relative prices
Household spending patterns change when incomes and relative prices change. Homothetic preferences permit expenditure growth without changing budget shares at fixed relative prices; nonhomothetic preferences allow households with different incomes to allocate spending differently. Food consumption may expand in quality and variety while the fraction of income spent on basic food declines. Services include everything from personal care to software, and their income responses need not be uniform.
A simple diagnostic is a two-good CES demand system with positive weights b_i and elasticity of substitution sigma. With price r_i and expenditure E, the demand for each good is proportional to b_i r_i raised to minus sigma; its expenditure share equals b_i r_i^(1-sigma) divided by the sum of those weighted price terms. For sigma below one, an improving industry's relative price decline reduces its relative expenditure share if the other price is unchanged. For sigma above one, the expenditure response reverses. These implications assume unchanged preferences, positive prices and an interior allocation.
With one-factor competitive production and a common wage, expenditure shares coincide with labor shares when all output is purchased for final consumption and no other factor or tax receives income. The price mechanism then gives a direct route from differential productivity growth to sectoral employment changes. Once capital, government and intermediates are introduced, the correspondence no longer follows. The consumption model identifies an instructive limiting case rather than an estimate of actual employment.
Comin, Lashkari, and Mestieri (2021) develop a model of income and price effects that incorporates nonhomothetic demand together with changing relative prices. The empirical question raised by their framework is whether composition changes reflect changing preferences over expenditure levels, changing technology and prices, or both. Identifying those contributions requires observations that distinguish movement along a demand curve from changes in demand. An industry time series alone generally cannot provide that distinction.
The analytical implication is that the development trajectory has a demand-side feasibility condition. Even technically productive investments require markets. A country with scarce domestic purchasing power may access foreign demand, while growing domestic demand can sustain a service expansion even without export participation. Market size and sector-specific input costs must be placed in the same explanation.
Labor mobility, human capital, and selection
Moving between industries requires time, information and resources. Skills are partly transferable and partly specific to occupations, production processes, languages and locations. The cost of changing work depends on distance, household care obligations, housing, certification and access to credit. Firms also incur costs when hiring and training. The observed wage differential can survive even if workers are actively responding to incentives.
The difference between average and marginal product is fundamental. An industry with a high output-to-employment ratio may have a large stock of capital, valuable natural resources, significant intangible assets or an unusually selected workforce. The output created by one additional worker may be much lower than the observed average. Conversely, complementary inputs could increase the worker's contribution above an extrapolation from current ratios. Measured averages are meaningful performance summaries but cannot directly rank the returns to every feasible transfer.
Gollin, Lagakos, and Waugh (2014) examine the agricultural productivity gap with adjustments to labor and output measurement. Their finding that a sizable gap persists after such adjustments motivates inquiry beyond simple measurement error. It does not establish that a worker randomly assigned to a different industry will reproduce the receiving industry's average output. An additional distinction comes from Lagakos and Waugh (2013), whose sector selection model links subsistence requirements to who chooses agricultural work. Selection is an economic mechanism that affects both productivity measures and worker flows.
Consider a worker whose agricultural contribution is three output units. Available service work may yield four units after training but require a one-time cost equivalent to five units and subsequent commuting costs. Comparing current sector averages of three and six overstates the private incentive if the worker expects a lower service productivity or bears a larger cost. The social return can differ again if training produces spillovers or migration strains infrastructure. A model of reallocation must specify whose costs and whose returns enter each decision.
The relevant empirical evidence includes worker-level employment transitions, wage changes, hours, employer characteristics, location and nonemployment spells. Selection and sector-specific prices are alternative explanations for productivity gaps; a credible design must discriminate among them rather than treating observed differences as proof of avoidable inefficiency.
Capital mobility and productive complementarities
Labor and capital reallocation are coupled processes. Manufacturing machinery, distribution facilities, transport networks and specialized buildings may be largely immobile after investment. Financial capital is easier to transfer than the productive asset it financed. Closing a factory releases workers and some financial claims but often destroys part of the existing capital's useful value. A loan for an expanding business does not ensure that equipment, infrastructure or trained workers become available on the same schedule.
Suppose two industries produce F_i(K_i,L_i) with diminishing marginal products. With no taxes, risks or adjustment costs, the allocation maximizing aggregate output valued at fixed prices equalizes the value of the marginal product of each mobile factor across interior uses. The first-order condition follows from the resource constraints sum_i K_i = K and sum_i L_i = L. If industry prices change with output, a complete welfare analysis instead requires the relevant marginal valuation of final goods and the full equilibrium response.
When capital is industry-specific, a labor transfer initially raises the ratio of labor to capital in the receiving industry and lowers it in the sending industry. The receiving marginal product can fall relative to its previous level. Investment then alters the next period's feasible allocation. Productivity can rise too slowly for workers to absorb the private costs of transition even if a long-run counterfactual predicts substantial social gains.
This reasoning makes the speed of factor accumulation part of the reallocation question. A growth episode financed by consumption credit or a real-estate boom can relocate workers without building the capital necessary to sustain diversified production. In the reverse direction, infrastructure investment may raise productivity in several industries without creating many direct jobs in the infrastructure provider. Investment and employment shares should therefore be interpreted across the production network.
A defensible policy analysis identifies the binding constraint. If capital is scarce or credit markets are distorted, relaxing a labor restriction alone may have a limited effect. If the barrier is certification, abundant financial capital may coexist with job vacancies. Equilibrium responses depend on which factor can adjust at which cost.
Firm selection and allocative efficiency
Sector averages aggregate heterogeneous establishments. Productivity can increase because continuing firms innovate, efficient firms expand, new producers enter or inefficient producers leave. A decomposition over industries can conceal these firm-level reallocations entirely. If all workers remain in manufacturing but the most efficient producers grow at the expense of inefficient plants, sector productivity rises through a within-industry process.
Hsieh and Klenow (2009) study manufacturing resource misallocation using establishment data for China, India and the United States. Their counterfactual calculations concern the potential manufacturing TFP gains under a modeled reduction in marginal-product dispersion. This is a different object from the shift-share contribution of agriculture-to-manufacturing employment movement. The study supplies an example of how an explicit allocation model produces a conditional efficiency estimate rather than an accounting component.
Differences in measured marginal revenue products may reflect distortions, markups, demand shocks, measurement error, product quality, risk or adjustment costs. The interpretation of dispersion depends on the maintained assumptions concerning technology and competition. An estimated reduction in dispersion is not automatically a feasible policy, since actual reforms may change firm entry, occupational selection and market demand.
The distinction matters when evaluating a sector's readiness to absorb labor. Expanding a measured high-productivity industry through low-performing entrants may decrease average productivity while increasing total output and employment. A rising number of firms could be a transitional investment in future capability rather than evidence of inefficient expansion. Conversely, a rise in sector productivity due to exit of low-productivity businesses can occur alongside falling employment and incomes of displaced workers.
The social evaluation must distinguish differences in technology from differences in allocation at an existing technology. Both can change observed productivity, but their policy implications differ. Data on firm entry, employment, output and capital structure are necessary for evaluating the underlying mechanism; an aggregate sector series cannot replace them.
Production networks and organizational boundaries
Sectoral output is jointly determined by final demand and demand for intermediate goods. A simple fixed-coefficient production network writes gross output vector x = A x + f, where A_ij represents intermediate input i required per unit of gross output j and f denotes final demand. If the spectral radius of A is below one, x = (I - A)^(-1) f. The inverse summarizes both direct production and successive rounds of input demand. This familiar input-output construction is an analytical device, not an empirical estimate here.
A shift in final demand from goods to services may raise manufacturing activity indirectly if service providers invest in computing equipment, vehicles or buildings. Faster manufacturing productivity can reduce the cost of a service provider's equipment. A rise in demand for construction can expand steel, cement, finance and transport. The employment generated per unit of final demand depends on the labor coefficients of the entire chain.
Outsourcing a service can change both the classification of jobs and the sector shares. In a stylized example, a manufacturer's in-house software workers move into a newly independent software firm and continue supplying the same factory. Manufacturing employment declines while services employment rises. Gross intermediate purchases become larger, while consolidated final output need not change. Sectoral organization has changed; physical tasks and aggregate productivity may initially be unchanged.
Input-output tables allow the researcher to ask whether an observed sector expansion is based on domestic supply networks or imported intermediates. They also reveal indirect dependence on industries with small final demand shares. Such comparisons require consistent valuation, industry definitions, and separate accounting for imports. In a fixed-coefficient calculation, a change in final demand is transmitted through a fixed A; technical substitution and capacity constraints require a different model.
A sector's strategic economic significance therefore exceeds its employment ratio. The role of a small infrastructure provider can be large if many industries depend on its output. Equally, a large employment sector can be predominantly a final-demand activity with limited input linkages. Neither pattern determines overall welfare independently of cost, productivity and consumption value.
Trade and international specialization
Open economies can allocate production differently from domestic consumption. Imports release domestic factors from some activities and supply firms with equipment or intermediates unavailable locally. Exports allow firms to reach larger markets and exploit specialized skills. Changes in trade costs and foreign demand alter opportunities for production independently of domestic household demand.
Uy, Yi, and Zhang (2013) examine open economy structural change in a model that includes productivity and trade-cost disturbances and is calibrated to South Korea from 1971 to 2005. The setting provides a concrete example of why the economy-wide allocation of factors must be analyzed with international markets rather than inferred from household expenditure alone. Its calibration is an internally coherent historical counterfactual exercise; transferring its parameters to unrelated countries would require further evidence.
Exposure to imports can cause persistent local adjustment burdens. Autor, Dorn, and Hanson (2013) study local import competition in the United States over 1990–2007. Their design uses differences in regional industrial specialization and external import variation to estimate local responses. Reported worsening labor-market outcomes in exposed places illustrate why national consumption gains and displaced workers' local opportunities must be kept analytically distinct.
The geographical boundary changes the measured adjustment. A community may lose manufacturing jobs without attracting the industries that expand nationally. National employment totals can recover while earnings losses remain concentrated in affected regions. Geographic migration, commuting and occupational retraining are different channels and cannot be presumed equally available.
Trade-induced reallocation also differs from a resource boom. An export price increase in a capital-intensive industry can shift relative prices and wages without directly employing many people. Expansion of domestic nontradable activities may occur at the same time as contraction of other tradable industries. Measurement at current prices can amplify the apparent transformation when commodity prices fluctuate. Comparative analysis requires price-adjusted measures, separately identified demand and supply shocks and careful treatment of exchange-rate movements.
International comparative evidence
International comparisons are informative because the same broad composition processes occur under different distributions of technology, resources and institutions. McMillan, Rodrik, and Verduzco-Gallo (2014) investigate structural change and productivity growth across developing regions. Their reported comparison distinguishes a weaker or adverse aggregate contribution of reallocation in parts of Africa and Latin America after 1990 from more favorable trajectories in Asia, with a more positive African contribution in the period after 2000. Table 3 reports average annual structural-change contributions over 1990–2005 of −0.88 percentage points for Latin America, −1.27 for Africa, +0.57 for Asia, and −0.09 for high-income countries (unweighted group means). The published decomposition weights within-sector productivity changes by initial employment shares and employment-share changes by final productivity levels. These rounded, sample-specific accounting results do not independently identify the welfare effect of mobility. The table and substantive discussion clarify the direction of reallocation, notwithstanding an inconsistent directional phrase in the abstract.
A receiving industry can provide expanding employment but low productivity growth. A sending industry can contract in employment yet contribute strongly through technical improvement. Distinguishing the industry composition of growth from its total rate helps interpret why economies with comparable GDP growth have different labor-market outcomes. Regional averages may conceal strikingly different national experiences.
Duarte and Restuccia (2010) study the role of structural transformation in productivity convergence using model-based sectoral comparisons. Their results emphasize heterogeneity in productivity levels and catch-up across industries, with particularly important consequences of service-sector performance for aggregate trajectories. This evidence makes an argument for examining productivity conditional on the new economic composition, rather than assuming a move out of agriculture completes the development process.
The competing interpretations are compatible in principle. Positive compositional contributions can coexist with weak within-sector improvement in a receiving industry. Strong technological catch-up can occur while short-run labor mobility reduces the measured composition contribution. The decomposition evaluates realized changes; the model identifies mechanisms under assumptions. The two approaches answer related but different questions.
A historical comparison should therefore define the period before observing the result, keep a consistent set of countries, document exclusions and publish country-level as well as regional summaries. Comparisons weighted by employment describe the experience of the average worker; equal-country averages describe the experience of the average observed country. Neither weighting is intrinsically preferable. The scientific question determines the appropriate statistic.
Deindustrialization and the service transition
Manufacturing's employment share can peak and fall when demand growth becomes weaker than labor productivity growth, when trade changes the location of production, or when firms outsource tasks. A decline in industrial employment is not necessarily a decline in industrial volume or productive capability. The economic consequences depend on what expands in its place.
Rodrik (2016) analyzes premature deindustrialization as a tendency for manufacturing peaks to occur at lower income or employment levels than historical industrializers experienced. This raises a development problem when manufacturing offered employment expansion together with productivity improvements and international market access. The relevant test is whether alternative activities can provide those functions under the new conditions.
Kruse, Mensah, Sen, and de Vries (2021), using a broader developing-country source base, examine manufacturing employment trends and report a reversal of widespread deindustrialization in several cases after the early 2000s. The existence of differentiated paths constrains any universal claim about the end of industrialization. It also illustrates how the reference period and coverage of data may alter interpretations of the same historical phenomenon.
Services are heterogeneous in task intensity, scalability, capital requirements, tradability and skill demands. Selling standardized software across borders differs markedly from a local repair business; both are services statistically. The World Bank research agenda represented by Nayyar, Hallward-Driemeier, and Davies (2021), services-led development, provides a useful point of departure for investigating service-sector opportunities. No numerical finding is attributed here to the presentation record alone.
In an economy with growing digital services and shrinking industrial employment, a central question is the absorption of workers previously employed in routine jobs. A high-output service industry can be too small or skill-intensive to absorb many workers. Expansion of lower-productivity local services may consequently coexist with internationally competitive advanced services. The service employment share alone cannot show which development path is operating.
Automation, tasks, and industry reallocation
An industry is a bundle of tasks organized into production. Automation changes the allocation of tasks between labor, machines and software. The employment impact cannot be inferred merely from the classification of the industry whose tasks are automated. Lower production costs affect prices, demand, investment and income distribution; complementary jobs may be created in the same or other industries.
Acemoglu and Restrepo (2019) analyze automation and new tasks through a task-centered framework that distinguishes displacement from reinstatement. The development of new labor-intensive tasks can offset a decline in the labor content of existing production. The framework also explains why measured productivity growth and the demand for workers need not move together.
The sectoral implications depend on whether new tasks are carried out inside an existing firm, purchased from outside providers or organized into a new industry. A manufacturer using external computing infrastructure can transfer measured employment toward business services while retaining its product line. A medical provider implementing software may change its labor composition without a change in sectoral employment share. A reduction in routine work can therefore be visible in occupational data long before industry data reveal a shift.
A prospective analysis should distinguish substitution for particular tasks, creation of complementary tasks, changes in output demand and the movement of people between employers. A scenario with accelerating automation but stationary aggregate employment is internally possible; so is employment expansion alongside declining labor income shares if the value of other factors grows faster. Current or future impacts of specific AI technologies require evidence of adoption, task coverage and market responses. No forecast of a particular technology is implied by these deductions.
Policy assessment in this context requires attention to the time path. An immediate displacement effect, an investment period and later creation of tasks need not coincide for the same worker or region. Transitional unemployment and income losses matter even when the aggregate long-run outcome is favorable.
Distribution, institutions, and transition costs
Economic composition affects earnings because industries employ different workers, operate under different institutions and distribute income differently between labor and capital. Even at unchanged aggregate productivity, an expansion of an industry with concentrated capital ownership may alter the distribution of national income. At unchanged output per worker, increasing employment can raise household earnings while reducing average wages if new participants enter relatively low-paid jobs. These are distinct social outcomes.
Institutions shape the resource-allocation mechanism by determining the enforceability of contracts, access to credit, the mobility of workers, and the capacity of firms to enter markets. Social insurance can reduce the private risk of leaving an existing job. Affordable housing and transport can increase the feasible set of occupations without a direct industrial subsidy. Education can make a higher-return activity accessible to workers otherwise excluded by qualification requirements.
The output consequence of a transfer should be evaluated over a feasible time horizon. If an investment restores employment only after several years, interim income losses and the cost of retraining require inclusion in welfare analysis. A discounted present value can compare resources and benefits under explicit assumptions about risk, the discount rate and the allocation of gains. It does not automatically measure distributional fairness. A social evaluation may attach different weights to gains and losses according to the explicitly adopted welfare criterion.
Informality introduces a further distinction. A worker can move from household production into unpaid family work, into informal urban activity or into formal wage employment; statistics may register these transitions differently. Employment growth without access to stable earnings or social protection can improve some outcomes while leaving significant insecurity. Survey evidence about earnings, hours and working conditions is needed in addition to national accounts.
The normative conclusion is conditional: expanding aggregate output is valuable, but the evaluation of reallocation should include the costs and benefits incurred by actual participants. Institutions designed to facilitate mobility should be judged on whether they enlarge productive choices rather than merely whether they alter sector shares.
Empirical design and international data
The Economic Transformation Database provides an operational starting point for empirical reconstruction. Its documented 2021 release covers 1990–2018, includes 51 economies, and groups production into 12 industries. It reports current and constant 2015 price value added in national currency millions and employment in thousands. The ETD technical note by de Vries and colleagues (2021) describes its statistical construction and country coverage. These records establish the data specification; this dossier does not claim to have downloaded or estimated the complete panel.
A reproducible study would first fix the dataset version, obtain the underlying series and preserve the original extraction. It would check that all 12 industry codes have a one-to-one mapping across dates, verify that employment and output refer to comparable economic activities, and document missing observations. Agricultural subsistence, real estate imputation and government services require explicit treatment. The official database page records September 2023 corrections affecting a Bangladesh employment series and Myanmar service deflators, demonstrating why date-stamped source versions matter.
For within-country decomposition, fixed-price national currency output is sufficient when additive and consistently defined. The unit conversion from millions of currency units divided by thousands of employed persons yields thousands of currency units per person. Cross-country comparisons of levels require a common price and purchasing-power framework appropriate to the sectors; a market exchange rate alone cannot establish physical output comparability. Aggregate PPP conversion need not reproduce sector-specific relative prices.
A research workflow would retain both endpoint and annual decompositions. Long endpoints summarize accumulated change but can conceal business cycles and reversals. Summing annual midpoint contributions defines a chain of accounting contributions; it may allocate changes differently from one long interval because weights vary along the path. Report both approaches and specify the classification, deflator, time horizon and units.
Additional evidence may include labor-force surveys, enterprise surveys, trade data, input-output accounts and worker-employer panels. Each source has a distinct sampling frame and unit of observation. Joining them without a validated bridge can misattribute production to occupations or employment to industries. Replication should preserve source files, code, data dictionaries and all exclusion decisions. This bibliographic study supplies the methodology; it does not replace a separately executed empirical estimate.
From description to causal identification
An increase in sector share and aggregate productivity provides a descriptive association. Several explanations may fit: workers move because receiving industries become productive; sector productivity rises because high-ability workers move; demand shifts toward an industry, inducing both investment and hiring; or institutions change several of these variables simultaneously. An identifying design specifies what intervention is being evaluated and under which conditions the observed variation can approximate it.
Suppose an analyst regresses productivity growth on the change in manufacturing employment share. Reverse causality is plausible because technical progress can attract labor, yet labor-saving technology can reduce manufacturing employment. A third variable such as export demand may affect both. Conditioning on contemporaneous sector productivity can introduce bias when productivity is an intermediate consequence of the very shock being studied. A coefficient from this regression is not a general causal productivity effect of manufacturing.
A shift-share instrument uses initial exposure weights combined with external industry shocks. It can be valuable when the shock variation is credibly exogenous to local potential outcomes and the design accounts for correlated exposures. It requires examination of the shares, shock origin and identifying restrictions. The Autor–Dorn–Hanson study provides a concrete example of an import-exposure strategy; its local estimates cannot be directly substituted for an economy-wide welfare calculation.
Panel event studies can track employment and wages before and after plausibly external changes. Pre-event patterns, anticipation, changes in industry coverage and simultaneous shocks must be assessed. When treatment begins at different dates, heterogeneous effects complicate simple two-way fixed-effects interpretations. A structural equilibrium model offers a different approach: specify preferences, technologies, market structure and mobility costs, calibrate or estimate parameters, and solve a counterfactual. Its credibility depends on independent validation and sensitivity to assumptions.
The two approaches are complementary. Reduced-form estimates identify particular responses under stated designs; models can organize indirect effects outside the observed variation. Neither automatically establishes external validity. A convincing interpretation joins carefully measured descriptive patterns, mechanism-specific identification and an explicit account of general-equilibrium consequences.
Counterexamples and robustness
Consider three hypothetical economic histories. In the first, agricultural productivity rises while agricultural employment declines, leaving output stable and freeing workers for other production. Aggregate productivity grows through both within-sector improvement and reallocation. In the second, an urban boom draws workers into low-productivity services, agriculture contracts and aggregate productivity grows weakly. In the third, an export industry automates rapidly, contracts employment and raises its volume output; displaced workers experience lengthy unemployment. The same descriptive fact of declining employment in the historically dominant activity is compatible with all three sequences.
A change in measured productivity can also reflect a composition shift within the unit of observation. A hospital system serving more complex patients can report lower throughput per worker even if quality improves. A highly automated extraction project can report high output per worker even if its employment multipliers are limited. Measured value added includes valuation conventions, and quality-adjusted measures may differ from output counts. Robustness requires multiple indicators and an explicit definition of what is being evaluated.
The first robustness test is definitional: compare output, employment and where possible hours shares; distinguish nominal and real measures. The second is temporal: compare predeclared subperiods and exclude a single recession or commodity spike from driving the entire interpretation. The third is classificatory: aggregate to broader categories and evaluate whether the conclusion depends on one imputed or unusually capital-intensive industry.
A fourth test varies the decomposition convention. If the initial-share and midpoint results differ markedly, simultaneous changes in sector composition and productivity are material. That information should be reported rather than choosing the convention that gives the preferred sign. A fifth test concerns plausible capacity constraints: can the receiving sector actually employ additional workers at observed output per worker? An affirmative answer requires support from capital and labor-demand evidence.
These counterexamples show why a defensible general claim must be conditional. They also generate concrete empirical tasks: compare sector classifications, distinguish hours and workers, test price measures, track individual transitions and investigate receiving-sector expansion. Robustness is a substantive scientific practice rather than a mechanical requirement.
Original analytical synthesis
The relationship between structural change and economic performance can be organized through a sequence of distinct questions. What makes the economy's distribution of demand and production change? Which workers and assets can respond? What productive capacities are created or lost as the response proceeds? What aggregate and distributional consequences result? These questions assemble existing economic mechanisms into an interpretation of the evidence; they do not establish a new named theory.
The first conclusion is that composition is an endogenous state variable. Employment shares respond to productivity, relative prices, household demand, trade and institutions. Since the composition changes the aggregate weighting of sector outcomes, sectoral development affects the future path of aggregate productivity. This feedback requires a dynamic explanation. It also explains why the historical success of one industry cannot establish its future importance in a changed economic environment.
The second conclusion concerns the economic meaning of a productivity advantage. A gap in average output per worker reveals a difference worth investigating, not a quantified gain from transferring particular workers. A meaningful reallocation counterfactual needs marginal returns, workforce characteristics, available capital, effects on prices, network demands and the timing of adjustments. The gap is evidence about recorded production; the return to an intervention is a separate economic object.
The third conclusion concerns demand and capacity. A sustainable expansion requires a market for additional production and sufficient complementary factors. Supply-side innovation can release workers into new activities, yet a receiving industry has to create viable positions. Demand growth may expand employment even in industries with limited measured productivity growth. The performance of structural change therefore depends on the coordination between the production actually released, the products buyers want and the productive capacity available in receiving industries.
The fourth conclusion concerns welfare and time. Positive aggregate allocation effects can coexist with losses among displaced workers, especially where specialization is geographically concentrated. Transitional costs and distribution can change the ranking of two alternative paths with identical long-run output. A careful evaluation should report these consequences independently of a productivity decomposition.
Together these deductions explain why the objective of economic policy is not a predetermined employment composition. An analytically defensible objective is to improve productive opportunities and feasible transitions while evaluating output, earnings, resilience, cost and distribution. Evidence about the likely consequences of an intervention must precede the selection of particular instruments. The original contribution of this dossier is this interpretive integration, its exact worked decomposition and its identification diagnostics; no claim of priority or independent empirical discovery is made.
Research boundaries and unresolved scientific questions
This dossier is a literature-based theoretical and methodological inquiry. It uses bibliographic records, publisher abstracts, author-affiliated repositories, and official database and standard descriptions. It does not represent a full-text audit of every cited paper, a new estimation of a cross-country panel, or replication of proprietary establishment observations. Source descriptions are deliberately limited to results supported at the level of access achieved. The formal examples are deductions with stated assumptions and contain no historical data.
The principal open scientific question is how to estimate the productive potential of workers in alternative activities while separating ability selection from the effects of sector-specific capital and technology. Worker-level longitudinal designs can improve the analysis, but individual mobility choices are endogenous. Another question concerns the interaction between technological change and demand when new products emerge and historical price comparisons become less informative. Measures of quality and network output require special attention.
Further research should also identify how local infrastructure capacity limits the expansion of a high-demand industry, how different forms of finance alter the transition path, and how publicly provided services affect the productivity of other sectors. These are research opportunities rather than facts left unverified in the present argument. A separately specified empirical project would need an archived dataset, a precise identification strategy, preregistered or transparently reported sensitivity analysis, and a reproducible method for computing estimates.
The ability of economies to support diversified employment under rapid automation remains an open problem. Projecting an exact future allocation requires uncertainty about technology, changing preferences and institutional responses to be made explicit. A useful forward-looking analysis would compare feasible scenarios rather than project historical sector shares mechanically. The present monograph provides conceptual distinctions and a research method for that inquiry.
Conclusion
Structural change is an integral dimension of economic growth because the relative distribution of activities affects aggregate outcomes and responds to the same forces that generate growth. The arithmetic of reallocation is exact under additive output measurement and consistent labor denominators. Its causal interpretation requires a model or credible research design.
Differential technological progress, the structure of demand, production linkages, international specialization and factor mobility jointly generate changes in employment and output composition. The direction of an employment shift cannot establish its contribution to welfare or its economic cause. Receiving industries differ in their capacity to absorb workers, generate income and expand output; sending industries differ in their ability to release resources without sacrificing productive capacity.
International evidence supports examination of heterogeneous historical trajectories rather than reliance on a single necessary sequence of sector shares. Productivity gaps motivate investigation of labor and capital allocation, but marginal returns depend on individual selection, technology, complementary resources and institutions. The proper scientific object is the feasible transition, including its costs, timing and distribution. Research that separates accounting from identification and integrates the economic mechanisms behind observed sector shares is best equipped to explain the development process.
References
Acemoglu, Daron, and Pascual Restrepo. 2019. “Automation and New Tasks: How Technology Displaces and Reinstates Labor.” Journal of Economic Perspectives 33 (2): 3–30. DOI: 10.1257/jep.33.2.3.
Autor, David H., David Dorn, and Gordon H. Hanson. 2013. “The China Syndrome: Local Labor Market Effects of Import Competition in the United States.” American Economic Review 103 (6): 2121–2168. DOI: 10.1257/aer.103.6.2121.
Baumol, William J. 1967. “Macroeconomics of Unbalanced Growth: The Anatomy of Urban Crisis.” American Economic Review 57 (3): 415–426.
Comin, Diego, Danial Lashkari, and Martí Mestieri. 2021. “Structural Change With Long-Run Income and Price Effects.” Econometrica 89 (1): 311–374. DOI: 10.3982/ECTA16317.
de Vries, Gaaitzen, Linda Arfelt, Dorothea Drees, Mareike Godemann, Calumn Hamilton, Bente Jessen-Thiesen, Ahmet Ihsan Kaya, Hagen Kruse, Emmanuel Mensah, and Pieter Woltjer. 2021. The Economic Transformation Database (ETD): Content, Sources, and Methods. WIDER Technical Note 2/2021. UNU-WIDER. DOI: 10.35188/UNU-WIDER/WTN/2021-2.
Duarte, Margarida, and Diego Restuccia. 2010. “The Role of the Structural Transformation in Aggregate Productivity.” Quarterly Journal of Economics 125 (1): 129–173. DOI: 10.1162/qjec.2010.125.1.129.
Gollin, Douglas, David Lagakos, and Michael E. Waugh. 2014. “The Agricultural Productivity Gap.” Quarterly Journal of Economics 129 (2): 939–993. DOI: 10.1093/qje/qjt056.
Groningen Growth and Development Centre and UNU-WIDER. 2021. Economic Transformation Database. Dataset, with September 2023 corrections. Groningen and Helsinki.
Herrendorf, Berthold, Richard Rogerson, and Ákos Valentinyi. 2014. “Growth and Structural Transformation.” In Handbook of Economic Growth, vol. 2, 855–941. Elsevier. DOI: 10.1016/B978-0-444-53540-5.00006-9.
Hsieh, Chang-Tai, and Peter J. Klenow. 2009. “Misallocation and Manufacturing TFP in China and India.” Quarterly Journal of Economics 124 (4): 1403–1448. DOI: 10.1162/qjec.2009.124.4.1403.
Kruse, Hagen, Emmanuel Mensah, Kunal Sen, and Gaaitzen de Vries. 2021. A Manufacturing Renaissance? Industrialization Trends in the Developing World. WIDER Working Paper 2021/28. UNU-WIDER. DOI: 10.35188/UNU-WIDER/2021/966-2.
Lagakos, David, and Michael E. Waugh. 2013. “Selection, Agriculture, and Cross-Country Productivity Differences.” American Economic Review 103 (2): 948–980. DOI: 10.1257/aer.103.2.948.
Lewis, W. Arthur. 1954. “Economic Development with Unlimited Supplies of Labour.” Manchester School 22 (2): 139–191. DOI: 10.1111/j.1467-9957.1954.tb00021.x.
McMillan, Margaret, Dani Rodrik, and Íñigo Verduzco-Gallo. 2014. “Globalization, Structural Change, and Productivity Growth, with an Update on Africa.” World Development 63: 11–32. DOI: 10.1016/j.worlddev.2013.10.012.
Nayyar, Gaurav, Mary Hallward-Driemeier, and Elwyn Davies. 2021. At Your Service? The Promise of Services-Led Development. Presentation, December 3. World Bank.
Ngai, L. Rachel, and Christopher A. Pissarides. 2007. “Structural Change in a Multisector Model of Growth.” American Economic Review 97 (1): 429–443. DOI: 10.1257/aer.97.1.429.
Rodrik, Dani. 2016. “Premature Deindustrialization.” Journal of Economic Growth 21 (1): 1–33. DOI: 10.1007/s10887-015-9122-3.
Solow, Robert M. 1956. “A Contribution to the Theory of Economic Growth.” Quarterly Journal of Economics 70 (1): 65–94. DOI: 10.2307/1884513.
United Nations Statistics Division. 2025. System of National Accounts 2025. International statistical standard, adopted at the fifty-sixth session of the United Nations Statistical Commission.
Uy, Timothy, Kei-Mu Yi, and Jing Zhang. 2013. Structural Change in an Open Economy. Working Paper 2013-09. Federal Reserve Bank of Chicago.