The rate of exploitation is the one magnitude in Marxist economics that promises to turn a theory of society into a number. It names the ratio between the value workers produce beyond the value of their own labour power and that value itself, surplus value divided by variable capital, and it carries the weight of the claim that profit rests on unpaid labour time. If the ratio can be measured for a real economy, the theory acquires an empirical leg it is often accused of lacking. If it cannot be measured, the concept risks remaining a piece of internal bookkeeping whose connection to observable magnitudes is asserted rather than demonstrated. The question is therefore not a technical footnote to the theory of exploitation. It is a test of whether that theory describes something in the world that statistics can reach.

A substantial literature has attempted the measurement, and its results are neither a triumph nor a mirage. Working from official national accounts, a line of researchers running from Joseph Gillman in the late 1950s through Anwar Shaikh and Ertugrul Ahmet Tonak, Fred Moseley, and Simon Mohun has produced estimates for real economies, above all for the postwar United States. Their shared finding, stated at the level of confidence the evidence supports, is directional: on most reconstructions the rate rose across the postwar decades, and that direction survives changes in method better than any particular level does. The levels themselves vary enormously, and the reason they vary is the real subject of this article. A handful of classification decisions, made before any data is touched, move the estimate more than the data ever does. Understanding those decisions is the difference between citing this literature and merely quoting it.

The rate of exploitation as a measurable claim

Marx states the rate in the ninth chapter of the first volume of Capital, where the working day is divided into two portions. In the first portion, necessary labour, the worker produces value equal to the value of labour power, the value of the wage goods and services that reproduce the capacity to work. In the remaining portion, surplus labour, the worker produces value that accrues to the employer. The rate of surplus value, which this series treats as identical in meaning to the rate of exploitation, is the ratio of the second portion to the first. Expressed in money terms for a whole economy, it becomes total surplus value divided by total variable capital, the aggregate wage bill of the workers whose labour produces that surplus. The concept article in this series, exploitation in Marxist theory, sets out what the ratio means and why it is a technical rather than a moral measure; the task here is narrower and harder, because a ratio defined on value magnitudes has to be recovered from statistics compiled for entirely different purposes.

What exactly is being measured when a study reports this rate?

A study reporting the rate is reporting a ratio between two reconstructed aggregates, not a figure read from any published table. The numerator estimates the surplus value produced in a year; the denominator estimates the variable capital advanced to the productive workforce. Both are built by reclassifying national accounts data under a stated, contestable scheme.

The translation problem deserves to be felt rather than merely acknowledged. National accounts, in the form maintained by bodies such as the Bureau of Economic Analysis in the United States, record compensation of employees, proprietors’ income, corporate profits, rental income, net interest, and taxes on production. None of these categories was designed with Marx’s distinctions in mind. Compensation of employees lumps together the wages of assembly workers, the salaries of senior managers, the pay of security guards, and the fees of advertising staff. Corporate profits exclude the large shares of the surplus that appear as interest, rent, and the salaries of those who organise and supervise production. Proprietors’ income mixes the earnings of a self-employed plumber’s labour with the return on the plumber’s van, tools, and goodwill in a single undifferentiated flow. A Marxian estimate begins only after each of these categories has been taken apart and reassembled, and the reassembly rules are where the argument lives.

Three features of the concept make the translation especially demanding. First, Marx’s categories are defined by function in the production of value, not by industry or occupation as statistical offices record them, so no published table corresponds to them even approximately. Second, the concept distinguishes labour that produces value and surplus value from labour that, however necessary and however hard, is paid out of surplus value already produced, and that distinction has no counterpart in standard accounting at all. Third, the rate is a ratio between flows over a year, and both flows are sensitive to how stocks such as fixed capital are depreciated in the source data. A reader who grasps these three features already understands why two honest studies can publish different numbers for the same economy in the same year, and why the interesting question is never the number alone but the construction behind it.

There is also a prior question about what kind of claim a measurement would support. A rising estimated rate does not by itself prove the labour theory of value, because the reconstruction uses categories the theory supplies and a sceptic can object that the exercise is circular. What a robust estimate does establish is more modest and still considerable: that the observable distribution of income, once reorganised along lines the theory specifies, behaves the way the theory leads one to expect, with the share of the working day performed beyond necessary labour growing over long stretches of the postwar period. Whether that reorganisation is the right lens is a theoretical dispute; whether the numbers move as reported, given the lens, is an empirical one. Keeping those two disputes separate is the first discipline of reading this literature well.

From the working day to the national accounts

The path from Marx’s chapter to a statistical estimate runs through three conversions, and each conversion is a place where reasonable researchers diverge. The first conversion is from labour time to money. In the text, the rate is a ratio of hours; in the accounts, it must be a ratio of money magnitudes. The standard defence of the substitution is that ratios are dimensionless: if money value added per hour of productive labour is roughly uniform across the productive sector in a given year, the money ratio and the time ratio coincide. That defence is serviceable for a single economy over time, and it weakens exactly where a reader might most want strength, in comparisons between economies with very different price structures and very different relationships between money magnitudes and labour time. Some researchers have attempted to work in labour time directly, using employment and hours data to build the numerator and denominator in hours, but hours data by function, distinguishing the productive from the unproductive within an industry, are rarely published at the required detail.

The second conversion is from the individual capital to the social aggregate. Marx’s examples concern a single firm or a stylised working day, while an estimate concerns an economy in which surplus value produced in one place is routinely realised, taxed, lent, and spent in others. Aggregation is not innocent. The surplus a manufacturing workforce produces may appear in the accounts as the profit of a retailer, the interest of a bank, the rent of a landlord, or the tax revenue of a state, and an estimator who looks for surplus only where it was produced will miss much of it, while an estimator who gathers it wherever it appears must be careful not to count the same surplus twice as it changes hands. The companion article on surplus value and profit explains why the pool of surplus value and the profit of any firm or sector are different objects; measurement inherits that distinction at national scale.

The third conversion is from the categories of the theory to the categories of the statistical office, and it is the conversion this article examines in most detail because it dominates the results. Six decisions recur in every serious estimate. The estimator must decide which labour counts as productive, how the wages of unproductive workers are to be treated, how the government sector enters the calculation, how self-employment income is to be split between a labour component and a surplus component, whether the calculation runs on gross or net magnitudes where depreciation is concerned, and whether the starting point is value added as the accounts define it or a reconstructed surplus product built from the production side. Each decision is defensible in more than one way, each has defenders in print, and the combination chosen fixes the level of the resulting series before the first datum is read.

A useful way to hold the whole procedure in mind is to see the estimate as the output of a small machine with six settings. The data fed into the machine are not in dispute; national statistical offices publish them, revise them, and document their construction in great detail. The settings are the dispute. This is why the literature’s most valuable product is arguably not any series it has published but the explicitness it has been forced into: a Marxian estimator must publish the settings, because without them the number cannot be interpreted at all. Mainstream distribution statistics rarely face the same compulsion, since their categories arrive pre-assembled by convention and their settings stay invisible. The Marxian literature, whatever its other faults, cannot hide its machinery.

The accounts as published and the adjustments they demand

Consider the anatomy of the United States national income and product accounts, the source nearly every major estimate uses, since the adjustments required there illustrate what any country’s accounts would demand. National income is presented as the sum of compensation of employees, proprietors’ income, rental income of persons, corporate profits, net interest and miscellaneous payments, taxes on production and imports less subsidies, and a few smaller lines. Compensation of employees is by far the largest component, and it is also the most misleading from a Marxian standpoint, because it treats every payment to every employee as the same kind of magnitude. Within that single line sit the wages that Marx’s scheme counts as variable capital, the salaries that his scheme counts as payments out of surplus value, and the remuneration of workers in trade, finance, and government whose labour the scheme classifies as unproductive. The first and largest adjustment is therefore a dissection of this line, industry by industry and, where the data allow, occupation by occupation.

Corporate profits require the opposite treatment: expansion rather than dissection. The profit line in the accounts is what remains to corporations after interest, and it excludes several flows that a Marxian reconstruction counts as parts of the surplus. Interest paid to lenders is a share of surplus value in Marx’s scheme, not a cost that reduces it. Rent on land likewise belongs to the surplus. Taxes on production and imports, levied before any profit is struck, are a claim by the state on the pool of value produced, and most reconstructions count them within the surplus rather than as a deduction from it. Corporate profit taxes are similarly a distribution of profit rather than a reduction of the surplus produced. Assembling the numerator of the Marxian ratio therefore means gathering a family of income lines that the accounts scatter across their tables precisely because the accounts were built to answer different questions about taxation, monetary policy, and the business cycle.

Proprietors’ income poses a subtler problem. The self-employed receive a single flow that remunerates their labour, compensates the capital they own, and rewards the risk they carry, and the accounts make no attempt to divide it. A Marxian estimator cannot leave the flow intact, because the division matters for both sides of the ratio: the labour component of a working proprietor’s income resembles variable capital or its self-employed analogue, while the remainder belongs in the surplus. The usual solutions impute a wage to proprietors, either by assuming their labour earns what comparable employees earn or by applying the economy-wide ratio between labour and total income within the relevant industry, and each solution imports an assumption that a critic can reasonably contest. As self-employment shrinks as a share of employment in advanced economies, the weight of this decision falls, but for earlier decades and for economies where small proprietorship remains extensive, it can shift the estimate appreciably.

Government enters the accounts in a way that flatly contradicts the Marxian scheme, and the contradiction must be handled by decision rather than by data. In the standard accounts, government output is valued at the cost of producing it, so government value added is essentially the compensation of government employees plus depreciation of government capital. No surplus is recorded because none is meant to be. For a Marxian reconstruction, the question is whether state employees produce value. The dominant answer in the empirical literature treats general government as unproductive activity financed from the surplus produced elsewhere, which places government compensation on the side of payments out of surplus value and removes government from the productive aggregate altogether. A minority approach seeks the subset of state activity that produces saleable output, in state enterprises and utilities, and classifies that subset as productive. Either way, the accounts as published cannot settle the matter, and the sheer size of government compensation in a modern economy makes the choice one of the largest single levers on the final number.

Depreciation, recorded in the accounts as consumption of fixed capital, creates a further fork between gross and net measurement. Marx’s value added in a year is a net concept: the new value created by living labour, after the value transferred from used-up means of production has been set aside. The accounts’ gross product includes depreciation as part of value added, and their net product subtracts an estimate of it built from assumed service lives and price indices for capital goods. An estimator working on a net basis inherits the statistical office’s depreciation model, with all its conventions; an estimator working gross avoids that inheritance at the price of departing from the concept. The fork returns in a later section, because its effect interacts with the rising capital intensity of production over the postwar period.

Behind all of these adjustments sits a distinction the accounts themselves half-acknowledge, between production and the activities that circulate, guard, finance, and administer what production yields. Input-output tables, which trace purchases between industries, give the estimator some purchase on that distinction, and the most thorough reconstructions lean on them heavily. Even so, the final classification of a whole industry’s workforce as productive or unproductive is a judgement informed by Marx’s texts, particularly the discussion of productive labour in the Theories of Surplus Value and the appendix to the first volume of Capital on the immediate process of production, where the criterion is not the material character of the work but its relation to capital: labour is productive, in the specifically capitalist sense, when it is exchanged against capital and produces surplus value for its purchaser. A transport worker moving goods and a commercial clerk moving paper may perform equally necessary tasks; the scheme sorts them differently because their employers stand in different relations to the production of value. Whether that sorting illuminates or distorts is debatable, and the debate is taken up below, but no estimate exists until the sorting is done.

The estimation decision tree

The adjustments just described can be organised into a single reference device, the estimation decision tree, which sets out each methodological decision an estimator must take, the options available at that point, the direction in which each option pushes the result, and the studies associated with each branch. Read across a row, the table shows how much of a published figure was decided before the data were consulted. Read down the final column, it shows that the major studies differ from one another less in their statistics than in their settings, which is precisely the classification dominance rule in tabular form: in Marxian empirical work the productive and unproductive classification moves the estimated rate of exploitation more than the data does, so any reported rate that does not state its classification rule is uninterpretable, and comparisons across studies that use different rules are meaningless.

Decision	Main options	Effect on the estimate	Studies and their branches
Classification of productive and unproductive labour	Broad production boundary including most services; narrow boundary restricted to goods production and value-adding services	A narrow boundary shrinks variable capital and inflates the surplus, raising the rate sharply	Shaikh and Tonak apply a detailed industry by industry boundary; Moseley works with a production sector defined against trade, finance, and government
Treatment of unproductive wages	Paid out of surplus value; treated as a cost deducted before surplus is struck	Counting unproductive wages inside the surplus raises the numerator and the measured rate; deducting them lowers both	Moseley treats the growth of unproductive wages as a deduction pressing on profit; Shaikh and Tonak carry them within the surplus aggregate
Government sector	Excluded as unproductive and tax financed; state enterprises only counted as productive	Exclusion enlarges the unproductive wage bill charged against the surplus and raises the rate	Nearly all reconstructions exclude general government; minority treatments isolate state trading enterprises
Self-employment income	Imputed wage at employee rates; split by the industry labour share; left whole outside the ratio	A generous imputed wage enlarges the denominator and lowers the rate; a lean one does the reverse	Most studies impute at average employee compensation for the industry; sensitivity to the choice is largest in early postwar data
Depreciation basis	Net value added after consumption of fixed capital; gross value added	A net basis tracks the concept of new value but inherits official depreciation models; a gross basis lifts both numerator and denominator	Shaikh and Tonak reconstruct net magnitudes; several shorter studies report gross based series alongside
Starting aggregate	Value added adjusted from the income side; surplus product rebuilt from the production side	The production side route ties the estimate to Marx’s reproduction schemes and exposes intermediate flows the income side hides	The Shaikh and Tonak framework is the fullest production side reconstruction; income side adjustments dominate shorter estimates
Supervisory and managerial pay	Counted as labour compensation; counted wholly or partly as a share of surplus	Reclassifying senior pay into the surplus raises the rate and steepens its postwar rise	Duménil and Lévy build their account around managerial pay as a surplus share; Mohun separates supervisory compensation in his United States series

Two lessons fall out of the table before any findings are considered. The first is that a Marxian rate of exploitation is not a statistic but the end of a chain of arguments, and citing it without the chain is like citing a court’s verdict without the reasoning. The second is that disagreement among estimators is unusually legible. When a mainstream labour share moves, the movement can be traced to data revisions, sectoral shifts, or definitional updates that require detective work to uncover. When two Marxian estimates diverge, the divergence can usually be assigned to named rows of the table within an afternoon, because the authors were obliged to publish their settings. Legibility does not make a measure correct, but it changes the character of the dispute from a quarrel about numbers into an argument about categories, which is where the theory wanted the argument to be all along.

Productive and unproductive labour: the boundary that decides the answer

Every row of the decision tree matters, but the first row matters most, and it deserves a section of its own. Marx inherited the distinction between productive and unproductive labour from Adam Smith, who had offered two definitions that do not coincide, and the Theories of Surplus Value spends long stretches untangling them. On the definition Marx finally endorses for capitalist production, the test is relational rather than material. Labour is productive when it is hired by capital to produce surplus value; it is unproductive, in this technical sense, when it is paid out of revenue, whether the revenue of a state, a firm buying circulation services, or a household buying domestic help. A singer employed by an impresario to fill a hall is productive for the impresario; the same singer giving a private recital for a fee paid from a patron’s income is not. The point of the distinction is not to rank occupations by usefulness. Marx is explicit that unproductive labour can be indispensable. The point is to separate the labour that expands the pool of surplus value from the labour that the pool must support.

Why does the productive boundary carry so much weight in the estimate?

The boundary moves both sides of the ratio in the same rotational direction. Reclassifying workers as unproductive removes their wages from variable capital, shrinking the denominator, and charges those wages to the surplus, swelling the numerator. A boundary drawn around goods production alone and one drawn around most market services can deliver rates differing by a factor of two.

The boundary’s leverage explains why the classification debate has consumed so much of the literature’s energy, and why the debate is genuinely difficult rather than merely scholastic. Consider the hard cases one by one, because each corresponds to a large and growing slice of employment in advanced economies. Transport and storage sit near the core of the classical case: Marx treats the transport industry as a sphere of production whose product is the change of location itself, so haulage and freight labour count as productive, while the purely commercial work of buying, selling, and bookkeeping does not. Maintenance and repair strain the line, since a mechanic restoring a machine preserves value that production would otherwise lose. Private security, cleaning, and catering supplied to firms are usually classified as unproductive overheads, yet they are conditions under which production proceeds. Health and education divide opinion most sharply: supplied as commodities by profit seeking firms, their staff appear productive on the relational test; supplied by the state from taxation, the same staff appear unproductive on it, and the boundary then tracks ownership form rather than anything about the work itself. Researchers who find that consequence embarrassing have proposed drawing the line by the character of the output instead, which rescues intuition at the price of departing from Marx’s criterion. Researchers who keep the relational test accept the consequence and reply that the embarrassment belongs to the economy, whose forms really do sort identical work into different economic categories.

Supervision deserves separate attention because it sits inside the factory gate, where the productive core is supposed to be unambiguous. Some supervisory labour is coordination that any complex production process requires, and some is the labour of discipline, exercised because the interests of employer and employed diverge within the working day. Marx himself remarks that the labour of supervision has this double character under capitalism. Empirically the two cannot be separated in payroll data, so estimators choose: some treat supervisory wages wholly as a charge on surplus, some split them by assumption, and some leave them in compensation. Because supervisory and managerial employment grew steadily across the postwar decades in the United States, the choice materially affects the measured trend and not only the level, which makes it one of the few classification decisions that the trend’s defenders and critics both have reason to fight over.

A further complication is that the boundary interacts with the structure of firms. When a manufacturer dismisses its own cleaners and buys cleaning from a contractor, the cleaners move between statistical categories without their work changing at all. Outsourcing across the postwar period shifted large numbers of workers from the payrolls of production firms to the payrolls of business service firms, and an industry based classification feels that shift as a change in the productive share even where the underlying division of labour barely moved. Occupation based classifications resist the distortion better but demand data that many countries publish only intermittently. The most careful reconstructions therefore work at the finest level the sources allow and report how the boundary was drawn industry by industry, which is one reason the fullest studies read as much like works of classification as works of economics.

Shaikh and Tonak and the rebuilding of the accounts

The methodological centre of the literature is the work of Anwar Shaikh and Ertugrul Ahmet Tonak, published in book form in 1994 as Measuring the Wealth of Nations: The Political Economy of National Accounts. The title announces the ambition. Rather than adjusting a few income lines, the authors set out to translate the entire apparatus of national accounting into the categories of Marx’s reproduction schemes, working from the production side as well as the income side, so that the resulting magnitudes for total value added, variable capital, and surplus value are internally consistent with the structure of the first and second volumes of Capital. The book’s empirical application covers the United States from 1948 to 1989, and its most cited result is a rate of surplus value that rises across that period, with the rise interrupted but not reversed by the disturbances of the 1970s. Earlier work in the same spirit had been attempted by Shane Mage in a 1963 doctoral dissertation on the United States economy and by Joseph Gillman, whose 1957 study of the falling rate of profit used official statistics to test Marxian magnitudes for the manufacturing sector; Shaikh and Tonak’s contribution was to give the translation a systematic and reproducible form.

Three features of their framework repay attention from any reader who wants to judge the whole tradition. The first is exhaustiveness about activities that produce no value in the Marxian scheme. Trade, finance, insurance, real estate services, private services of a personal kind, and general government are classified as unproductive sectors whose costs are met from the surplus generated in the productive sectors, and the book tracks the flows between the two great divisions with an accountant’s patience. The second feature is the treatment of the wages paid in unproductive activities, which figure as a redistribution of surplus value rather than as variable capital, so the growth of unproductive employment presses on the measured system from two directions at once: it raises the surplus aggregate as defined to include those wages, while the productive wage bill that forms the denominator grows more slowly. The third feature is candour about data limitations, including the interpolation of benchmark input-output tables between census years and the assumptions needed to split compensation where the published sources report only aggregates.

Their headline finding should be reported the way the tradition’s own standards demand, as a direction rather than a point. On their reconstruction, the share of the working day that the average productive worker spends beyond necessary labour grew substantially between the late 1940s and the end of the 1980s, so the rate of exploitation in the United States rose across the postwar era taken as a whole. The finding is not that every year exceeded the last; recessions dent the series, and the early 1970s show hesitation. Nor is the finding a level that can be quoted free standing, because the level is the joint product of the series and the classification published alongside it. What the study established, and what later work has repeatedly used, is a complete and inspectable machine: another researcher can take the same accounts, change one setting, and watch the result move, which is exactly the property that lets a disputed literature make cumulative progress.

Moseley, Mohun, and Duménil and Lévy: the postwar United States reworked

Fred Moseley’s study of the postwar United States economy, published in 1991, approaches the same accounts from a different question. Where Shaikh and Tonak ask what the rate of exploitation was, Moseley asks why the rate of profit fell across the postwar decades when, on Marxian reasoning, a rising rate of surplus value should have supported it. His answer turns on the distinction this article has been examining. Reconstructing the Marxian magnitudes from the national income accounts, Moseley found the rate of surplus value rising across the period while the share of unproductive labour in total employment and in the total wage bill also rose, so that an increasing portion of the growing surplus was absorbed by the wages and costs of trade, finance, supervision, and government. The rate of profit, whose denominator and deduction structure the companion article on the empirical evidence for the rate of profit examines, could therefore fall even as exploitation intensified, because the surplus was increasingly committed to sustaining activities that produce no surplus themselves. The argument reframes the postwar profit squeeze: not a failure of exploitation to rise, but a rising burden of unproductive expenditure on a rising surplus.

Moseley’s work also illustrates how the classification debate proceeds in practice, because his estimates drew detailed criticism and reply on exactly the decision tree’s rows: the placement of supervisory wages, the handling of taxes, the boundary around services. Simon Mohun’s estimates, constructed for the United States from the mid 1960s onward, enter the same conversation with yet another configuration. Mohun separates supervisory compensation explicitly, distinguishes the surplus available after various deductions, and reports the rate of surplus value rising across his period, with the increase concentrated after the early 1980s. His series and Moseley’s do not agree on levels, and were not built to, but their agreement on direction across the decades when the two overlap is the kind of convergence the literature leans on when pressed. Mohun has also been among the clearest writers on why the levels cannot be reconciled by better data alone: the residual differences are differences of framework, and framework differences are settled by argument about what the categories mean, not by waiting for a statistical office to publish a finer table.

Gérard Duménil and Dominique Lévy complicate the picture from within, and their complication matters because it challenges a premise the other studies share. In their account of the transformation of the advanced economies, the income of managers and senior salaried staff is not well described either as a wage for labour or as profit in the classical sense. They treat the upper reaches of the salary scale as a share of the surplus appropriated by a managerial stratum whose position between capital owners and wage workers defines the class structure of the modern firm. On the measurement question, their stance implies that a large and growing slice of what the accounts record as compensation of employees belongs on the surplus side of the ledger, and that the postwar rise in measured exploitation is even steeper once senior pay is reassigned. Readers do not have to accept their class analysis to see the measurement moral: the single largest component of national income, employee compensation, contains a contested interior frontier, and where an estimator draws that frontier moves the result by amounts that dwarf most data revisions.

Taken together, the three bodies of work define the literature a reader must reckon with. They share the raw material of official statistics and the conviction that Marxian categories can be operationalised. They differ in the boundary they draw around production, in the fate they assign to unproductive and supervisory wages, and in the question they put first, whether the level of exploitation, its trend, or its bearing on profitability. A citation practice follows from the differences: a responsible writer names the study, names the period, and names the classification stance in the same breath as any figure, because the figure has no meaning detached from the other three. The next section states what, under that discipline, the estimates collectively support.

What the estimates find and what they cannot settle

The findings can be organised into three tiers of confidence, and the organisation is more informative than any summary number. In the highest tier sits a directional claim about the postwar United States: across reconstructions that differ in their boundaries and deductions, the rate of exploitation rose between the late 1940s and the end of the twentieth century, with the rise uneven, interrupted by cyclical reversals, and in several series steepest from the early 1980s onward. The claim survives because it is supported by different machines fed with the same data, and because a simpler observable points the same way: productivity in the productive sectors grew faster than the real compensation of productive workers across much of the period, which, under the theory’s own construction of necessary labour, is what a rising rate of exploitation looks like before any reclassification begins. A reader may reject the framework and still recognise that the directional finding is a fact about the literature and about the underlying distributional data, not a slogan.

The middle tier holds claims about magnitudes stated as ranges. Published levels for the postwar United States, expressed as the ratio of surplus value to variable capital, span a wide band: at the conservative end, reconstructions that deduct unproductive costs before striking the surplus and treat the boundary generously report rates in the neighbourhood of or below parity between surplus and productive wages in the early postwar decades, rising thereafter; at the expansive end, reconstructions that carry unproductive wages and senior salaries within the surplus report rates well above two to one for the later decades. The band is not measurement error in the statistician’s sense. It is the visible price of the decision tree, and any use of a single figure from inside the band, without its settings, misrepresents the literature more thoroughly than using no figure at all.

The lowest tier holds the claims the literature cannot support, and naming them protects the claims it can. The estimates do not establish a rate of exploitation for a whole economy in the way a census establishes a population, because the object measured is partly constituted by the measuring scheme. They do not license comparisons of levels between countries whose statistical systems and classification choices differ, a point developed below. They do not show that workers’ living standards fell; the value of labour power can decline, through cheapening of wage goods, while the physical basket the wage commands grows, so a rising rate is compatible with rising real wages, and several of the estimators insist on the compatibility precisely because the confusion between the ratio and the standard of life is the most common misreading their work attracts. Nor do the estimates, by themselves, settle the standing of the labour theory of value, since the reconstruction presupposes the theory’s categories; their evidential role is to test consistency and trend within the framework and to discipline loose talk about shares with an explicit construction.

One boundary of the findings deserves emphasis because summaries routinely cross it. The reconstructions speak with most confidence about the span from the late 1940s to the turn of the century, where the source data are richest and the studies overlap. Extensions beyond that span, backward into the interwar years on historical statistics or forward through the financial crisis and its aftermath, exist and broadly continue the story of an elevated and disturbed rate rather than a collapsing one, but they rest on thinner concordances and fewer rival constructions, so the convergence that underwrites the central finding thins with them. The honest summary of the whole literature is therefore a sentence with a date range inside it: for the postwar decades of the twentieth century, in the economy measured most thoroughly, the division of the working day shifted toward surplus labour on every serious reconstruction, and the size of the shift, though not its existence, is what the remaining argument is about.

The labour share error

No confusion does more damage in popular discussion of this subject than the substitution of the labour share for the rate of exploitation, and the substitution is worth dismantling carefully because it is made in good faith by writers who assume the two magnitudes are rough equivalents. The labour share is a mainstream statistic: employee compensation divided by national income, or by value added in a sector, expressing how much of the income generated accrues to employees rather than to profits, interest, and rent. Its movements are tracked by central banks and statistical offices, and its decline across many advanced economies over the closing decades of the twentieth century is a documented feature of the distributional data. A writer who reports that decline as a rising rate of exploitation has, however, changed the subject without noticing, because the two measures are built from different numerators, different denominators, and different ideas about what counts as labour’s portion.

Is the labour share the same as the rate of exploitation?

No. The labour share divides employee compensation by national income and counts unproductive and managerial salaries as labour. The Marxian rate divides surplus value by variable capital, which requires reclassifying the accounts first, rebuilding both aggregates around the productive boundary instead of reading the published lines as they stand. The two answer different questions and can move in opposite directions.

The divergences are concrete enough to enumerate in prose. Begin with the numerator of the labour share, compensation of employees, which includes the salaries of senior managers, finance professionals, supervisors, and sales staff. In the Marxian construction, much of that compensation is a distribution of surplus value rather than the price of labour power engaged in production, so a rise in executive pay raises the labour share while raising the Marxian surplus as well, pushing the two measures in opposite directions at exactly the point where popular commentary expects them to agree. Turn to the denominator. National income includes depreciation in some presentations and excludes it in others, includes the imputed income of owner occupiers, and counts government output at cost; variable capital is none of these things. Turn finally to coverage: the labour share is silent about the self-employed unless an imputation is bolted on, and studies differ in whether taxes and transfers enter the calculation at all. The mainstream literature on the labour share and its evidence documents these choices within its own tradition; the Marxian point is sharper, because even a perfectly measured labour share would still not be the rate of exploitation, since the concepts partition the same income along different lines.

None of this makes the labour share useless to the Marxian argument. A sustained fall in the labour share, on a consistent definition, is evidence about the distribution between employees as a whole and property incomes, and it constrains any Marxian reconstruction built from the same accounts. The error is one of identity, not of relevance: treating a neighbouring statistic as the thing itself, and then importing the statistic’s authority into a claim it cannot carry. The discipline the estimators modelled applies here with full force. State which measure a sentence is about, and the sentence usually becomes defensible; blur the two, and no subsequent care can rescue it.

A worked illustration: one stylised economy, three rates

Because the classification dominance rule can sound abstract, it is worth watching it operate on numbers small enough to hold in the head. The figures in this section are an arithmetic illustration with round values chosen for transparency; they describe no real economy and no published estimate, and their only purpose is to show how the settings move a ratio whose inputs nobody disputes. Imagine an economy whose accounts for one year record total employee compensation of 1,000 units, of which 620 are paid to workers in production industries broadly defined, 210 to workers in trade, finance, and business services, 90 to government employees, and 80 to supervisory and senior managerial staff inside production industries. Corporate profits after interest are 260, interest and rent paid out total 190, taxes on production and corporate income total 150, proprietors’ income is 120, and depreciation is 180.

Now build the Marxian ratio three ways. On the first construction, the generous boundary, treat all production industry compensation including supervisors as variable capital, split proprietors’ income equally between a labour component and a surplus component, and define the surplus as profits plus interest, rent, taxes, and the surplus half of proprietors’ income, with unproductive wages left as deductions already reflected in the profit figure. The denominator comes to 700 with the proprietors’ labour component added, and the numerator to 660, giving a rate a little below parity. On the second construction, the standard reconstruction in the Shaikh and Tonak spirit, remove the supervisory salaries from variable capital and charge them to the surplus, add the wages of trade, finance, and government workers to the surplus as sums it must support, and keep the proprietor split. The denominator falls to 620 and the numerator rises past 1,100 once unproductive compensation joins it, and the rate approaches one and four fifths. On the third construction, the expansive treatment associated with the managerial argument, reassign the whole of senior pay and a slice of professional salaries in services to the surplus and impute a leaner wage to proprietors; the rate climbs higher still, beyond two to one. Nothing in the accounts changed between the three readings. The same statistical office published the same lines, and every input is common property. What changed was the classification, and the ratio roughly doubled across defensible settings.

The illustration also shows why the trend can be sturdier than the level. Apply each construction to a second stylised year in which production compensation grows slowly, supervisory and service compensation grows quickly, and profits and interest grow fastest of all. Each construction then reports a higher rate than it reported for the first year, because the compositional shifts push every variant in the same direction even though they disagree about where the level sits. Agreement on direction amid disagreement on level is not a rhetorical refuge invented to save the literature from its spread of results. It is what the arithmetic of the decision tree leads one to expect whenever the underlying distributional shifts are broad, and it is the pattern the published studies in fact display.

Comparing rates of exploitation between countries

The framework has travelled. Researchers have applied reconstructions of the same kind to economies well beyond the United States: Edward Wolff’s study of Puerto Rico belongs to the earlier generation of the literature, Lefteris Tsoulfidis and co-authors have built long series for Greece, and research groups have produced estimates for economies across Europe, Asia, and Latin America, often adapting the Shaikh and Tonak architecture to national accounts of very different design. The existence of this wider literature tempts a use it cannot sustain, the league table: a ranking of countries by their rate of exploitation, quoted as though the figures were exchange rates. Resisting that temptation requires understanding what stands between two national estimates even when both are competently built.

The obstacles begin with the statistical raw material. Countries differ in how their accounts treat the financial sector, whose output is partly imputed rather than observed; in the size and form of government production; in the prevalence of self-employment and informal work, which swells the proprietors’ income problem from a nuisance into the central difficulty; and in the industrial detail available for drawing the productive boundary. A country where a third of employment is informal presents the estimator with flows the accounts capture partially or miss entirely, and the reconstruction then rests on survey imputations layered over the official lines. Two estimators facing two such systems, even with identical theoretical commitments, will build machines whose outputs are not commensurable at the level of precise values. Add that classification habits differ between research groups, that some national traditions draw the services boundary generously and others narrowly, and the honest conclusion is that cross-country level comparisons measure a compound of economies and conventions in unknown proportions.

What survives the obstacles is the comparison of directions and of structures. A researcher can defensibly report that reconstructions for several economies find rising rates across overlapping decades, and that the composition of the surplus differs instructively between them, with the weight of finance, of government employment, or of self-employment varying in ways the decision tree makes visible. A researcher can also compare a single country’s rate under alternative classifications to show which national features drive its level, using other countries as a repertoire of settings rather than as rivals in a ranking. Journalists and students should treat any published pairing of two countries’ rates, offered without a harmonisation exercise behind it, as a category error of the same family as the labour share substitution: two numbers that share a name and a formula while differing in construction. The remedy is identical in both cases. Ask for the settings, and decline the comparison until they match.

Depreciation, the state, and the self-employed

Three residual problems recur in every reconstruction and reward a slower look, because each conceals a theoretical choice inside what looks like a bookkeeping detail. Depreciation is the first. The concept of value added in Marx is net: the new value a year’s labour creates, after replacing the value of the means of production used up. The accounts offer both gross and net aggregates, but their net figures depend on estimated service lives for structures and equipment, on price indices for capital goods, and on geometric or straight line decay assumptions chosen by the statistical office. An estimator who adopts the official net product adopts those assumptions unexamined; an estimator who rebuilds depreciation from capital stock data must choose a method and defend it. The choice is not cosmetic. In economies where the stock of long lived equipment and structures grows relative to output, gross and net series can tell visibly different stories about the surplus available in a year, and the treatment interacts with the measured trend in the rate of exploitation through the numerator’s size.

What does depreciation do to a Marxian estimate of the rate?

Depreciation decides whether the estimate is struck on gross or net magnitudes. The surplus is calculated after the value of used up means of production has been replaced, and depreciation figures rest on assumed service lives and price indices, so a net estimate inherits those conventions while a gross estimate sidesteps them at the cost of drifting from the concept.

Government is the second problem, and it is double sided. The dominant treatment, described earlier, excludes general government from production and charges its wage bill to the surplus, which is coherent within the theory: state activity is financed by taxation, and taxation is a claim on value produced elsewhere. But the size of the modern state makes the treatment do heavy work. Where public employment is a fifth or more of all employment, the exclusion simultaneously removes a large wage bill from any denominator and adds it to the charges the surplus must bear, and critics reasonably ask whether a framework in which expanding public health provision mechanically registers as a change in the exploitation of factory workers is describing the economy or its own bookkeeping. Defenders answer that the registration is informative precisely because it shows how the surplus is committed, and that public services which reproduce labour power might better be handled through the value of labour power itself, a refinement some reconstructions attempt by treating parts of state provision as components of the social wage. The exchange has no purely statistical resolution, which is the point: the treatment of the state is a theoretical commitment wearing an accounting costume.

Self-employment is the third problem. The mixed income of proprietors must be divided, and every division rule embeds a view about what a proprietor is. Imputing the average employee wage assumes the proprietor’s labour is worth what hired labour costs in the same industry, which flatters the labour component where proprietors in fact work for less than employee rates and understates it where skilled practitioners command scarcity premiums. Splitting by the industry’s overall labour share imports the very distribution the estimate is trying to measure. Leaving proprietors outside the ratio altogether, as some studies do for the headline series while reporting the sensitivity separately, is the most transparent option and the least complete. In economies and decades where proprietorship is small, the choice barely matters; where a family farm and shop economy persists inside the statistics, it can dominate the result. The lesson extends well beyond these three cases: in this literature, the footnotes are where the findings are decided, and a reader who skips them has not read the study.

From values to prices: the transformation shadow over measurement

A deeper difficulty waits beneath the classification problem, and honest accounts of the literature name it rather than step around it. Marx’s rate is defined on values, on magnitudes of embodied labour time, while every statistic an estimator can touch is denominated in money prices. Between the two stands the transformation problem: in the third volume of Capital, competition redistributes surplus value so that prices of production deviate systematically from values wherever the composition of capital differs between sectors. If prices deviate from values sector by sector, then money aggregates built from the accounts are not transparent measures of value aggregates, and the measured ratio might track the price system rather than the value system it is meant to expose. An estimator who ignores the gap entirely is open to the charge of measuring a different object from the one the theory defines.

The literature has produced three broad bridges across the gap, and the choice among them is another setting on the machine, less visible than the decision tree’s rows because it operates before classification begins. The first bridge relies on aggregate equalities: whatever the deviations at sector level, total price equals total value and total profit equals total surplus value in Marx’s own construction, so economy wide ratios computed in money inherit a value interpretation at the aggregate even if no sector’s figures do. Critics of the bridge reply that the equalities hold only under contested readings of the transformation procedure, and that an aggregate ratio can be stable while its sectoral meaning dissolves. The second bridge is the reinterpretation associated with Gérard Duménil and Duncan Foley, often called the New Interpretation, which defines the monetary expression of labour time as the ratio of money value added to productive labour hours and treats the value of labour power as the wage share of that expression; on this construction, the money ratio and the value ratio are related by definition rather than by approximation, and several empirical studies adopt it, sometimes without advertising the adoption. The third bridge abandons the crossing and works in labour time throughout, building both aggregates from hours data and accepting the data limitations described earlier. Each bridge is defensible; none is compulsory; and a measured series states, implicitly, which bridge its author walked across. The wider assessment of how Marxist economics fares when it submits to empirical test, taken up in this series under testing Marxist economics, returns to this junction repeatedly, because so much of the empirical programme crosses the same river.

Intensity and skill complicate the time side of the ratio in ways money measures conceal. The denominator of the value rate is the value of labour power, and the numerator grows when the working day lengthens, when productivity in wage goods industries cheapens necessary labour, and when the intensity of labour rises so that more value is created in the same clock hour. National accounts cannot see intensity at all, and hours statistics see it only darkly, through output per hour movements that blend intensity with technology and effort norms. The reduction of skilled to simple labour poses a parallel problem: the theory counts an hour of complex labour as a multiple of simple labour, while payroll data record the multiple as a wage differential that may reflect scarcity, credentialing, or bargaining power rather than any value creating multiplier. Estimators in money terms absorb these difficulties silently, which is sometimes a virtue, since the wage differential is at least an observable magnitude, and sometimes a quiet change of subject. A reader comparing a money based series with an hours based series is therefore comparing two different strategies for surviving the same conceptual weather, and the weather, not the competence of the researchers, explains much of any divergence between them.

Turnover adds a final wrinkle that single year snapshots flatten. Variable capital is an advance: the wage fund a firm lays out returns as products are sold and is laid out again, so the capital advanced over a year is smaller than the sum of wages paid over the year whenever turnover is faster than annual. Marx distinguishes the rate of surplus value, computed on the variable capital employed in each cycle, from the annual rate computed against the advance. National accounts record annual wage flows, which suit the annual construction, but sectors differ in turnover speed, and changes in payment practices, from monthly to weekly wages or the reverse, alter the advance without altering the labour performed. Serious reconstructions acknowledge the wrinkle and generally proceed on flows, judging the distortion second order for trend purposes. The judgement is reasonable, and it is one more entry in the long list of places where the estimator’s discretion, openly exercised, stands between the reader and the number.

The strongest objection: a framework measuring itself

The objection that this literature must answer is not that its arithmetic is sloppy; by the standards of empirical economics the leading studies are careful, documented, and reproducible. The objection is that the care is beside the point, because the object being measured is manufactured by the measuring scheme. State it at full strength, in the form its most serious proponents would recognise. To produce a rate of exploitation, the analyst must reclassify the national accounts according to a theory whose categories, productive and unproductive labour above all, have no existence outside that theory. The reclassification involves discretionary judgements at every row of the decision tree: where the production boundary falls, whose salary counts as a deduction from surplus, how the state is treated, how mixed income is split. Different defensible judgements yield rates differing by a factor of two or more for the same economy and year. A measure whose value depends this heavily on the measurer’s prior commitments is not, the objection runs, a measurement of the economy at all. It is the framework contemplating its own reflection, and the procedure is unfalsifiable in practice: no imaginable configuration of the published accounts could return the verdict that exploitation was absent or falling under every defensible classification, because the analyst who disliked a result could always reach it by redrawing a boundary the theory leaves underdetermined.

The objection gains force from three supporting observations. The first is the absence, in most published studies, of systematic sensitivity analysis: authors report the series their preferred settings produce and rarely publish the fan of series that neighbouring settings would produce, so the reader cannot see how much of the headline result the settings own. The second is the direction of the discretion. The classification choices that raise the level and steepen the trend are, in several prominent cases, also the choices the theory’s sympathisers find most natural, and a sceptic is entitled to wonder whether the machinery was tuned, innocently and over years, toward the answers the tradition hoped to find. The third is the comparison with neighbouring empirical programmes. The price and value correlation literature, which tests whether labour contents predict market prices, at least exposes itself to a statistic computed by outsiders on definitions fixed in advance; the exploitation literature, by contrast, computes its central magnitude inside its own categories from first to last, so its results cannot be checked by anyone who does not first convert to its framework. On this account, the convergence of direction across studies proves less than defenders claim, because the studies share the framework’s deep settings even where their surface settings differ, and a shared lens can manufacture a shared picture.

There is a weaker companion objection that deserves naming because it is often confused with the strong one. It holds that the entire exercise inherits the fate of the labour theory of value, so that if embodied labour magnitudes lack explanatory standing, the rate of exploitation is a ratio between fictions, however carefully computed. That objection is really an objection to the theory rather than to its measurement, and it belongs to the wider argument about Marxist economics as a research programme. The strong objection is narrower and, for that reason, harder to evade: grant the theory its categories for the sake of argument, and the measurement still fails, because the categories cannot be applied to statistical reality without a discretion so wide that the result certifies the analyst rather than the economy. Any defence that does not meet the strong objection on its own ground, the ground of discretion and falsifiability, has not met the objection at all.

The reply, and the concession on levels

The reply begins by denying that discretion of this kind is unique to Marxian measurement, while accepting that its scale here is unusual. Every distributional statistic embodies classification judgements that could have gone otherwise: the treatment of imputed rent in income, of owner occupied housing in price indices, of the financial sector’s output in product accounts, of capital gains in measures of top incomes. The labour share itself moves by percentage points under alternative treatments of self-employment income, a discretion mainstream statisticians exercise quietly because convention, not nature, fixed their choice long ago. What distinguishes the Marxian literature is that its conventions are young enough to still be argued about in the open. Visibility of discretion is not the same as arbitrariness, and the decision tree’s rows are not free choices: each branch is defended by appeal to the theory’s texts and to the function the category performs in the theory’s explanations, and branches can be criticised, as they vigorously are within the literature itself, for betraying the texts or for blunting the category’s work. A discretion that must be argued for, in public, against named alternatives, is a different thing from a discretion that merely certifies its holder.

The stronger strand of the reply concerns convergence. The studies that find a rising postwar rate for the United States do not in fact share all their deep settings. They draw the production boundary at different places, treat supervisory pay differently, adopt different bridges between values and prices, and begin from different questions, in Moseley’s case from a puzzle about falling profitability that a rising rate of exploitation helps solve only in combination with the growth of unproductive costs. That the direction recurs across these configurations is evidence of the kind the objection demands: a result that survives the redrawing of boundaries is not manufactured by any single boundary. The underlying distributional arithmetic supports the same reading without Marxian categories at all, since output per hour in the productive sectors outgrew the real reward of the workers employed there across long postwar stretches, and any framework that partitions income between the producers’ consumption fund and the remainder will register the remainder’s share growing. The Marxian construction gives that remainder a name and a mechanism; the growth of the remainder is not the construction’s invention.

Then the concession, stated as plainly as the objection was. The level of the rate of exploitation for any real economy is not established, and on the evidence of the spread across competent studies it is not establishable within a narrow band by the methods in use. Anyone who quotes a single national rate as a fact about the world, rather than as the output of a named study with named settings, is claiming more than the literature contains, and defenders of the measurement programme do their cause no service by letting such quotations pass. Sensitivity analysis, the systematic publication of results under alternative classifications, remains rarer in this literature than its own standards require, and the objection scores a clean hit there: a field that knows its results are classification sensitive owes its readers the fan of alternatives as a matter of routine. The verdict that survives the exchange is therefore split. As a measurement of levels, the programme overpromises and should be read with the discount its spread imposes. As a measurement of direction, built on published settings and convergent across rival configurations, it delivers something no slogan in either direction can supply: a disciplined, inspectable claim about how the division of the working day changed across the postwar era, together with the machinery for anyone to challenge it setting by setting.

Trend against level: what may safely be concluded

The distinction between trend and level, invoked throughout this article, deserves a precise statement because it carries the practical guidance a reader needs. A level claim assigns a value to the rate in a year: surplus value stood at such a multiple of variable capital. A trend claim assigns a direction and rough pace to its movement across a span of years. The two claims rest on different evidentiary footings. A level inherits every classification decision at full strength, since shifting a boundary shifts the level in the year of the shift and every other year alike, but it shifts them together, so the movement between years is governed more by the data and less by the setting than either level is. This is why the trend can be the more reliable object even for a reader who grants the objection most of its force: systematic construction error, held constant across a series, cancels in the comparison of its endpoints in a way it never cancels in the reading of a single point.

Can two careful studies of the same economy disagree without either being wrong?

Yes. The disagreement is usually traceable to named rows of the estimation decision tree rather than to error in either study. One reconstruction may deduct unproductive wages before striking the surplus while another carries them inside it, so the two series report different levels for every year while agreeing on the direction of movement.

What would count as evidence against a measured trend?

A reconstruction built on classification settings defended from the theory’s texts, in which the rate fails to rise across the period, or in which the apparent rise dissolves once a contested setting is corrected. Such a study would hold the data constant, vary the boundary or the treatment of supervisory pay, and show the direction reversing.

A reading protocol follows from everything established so far, and it can be stated as a sequence of questions to put to any estimate before using it. Which study produced the figure, for which economy, and for which years? What boundary separates productive from unproductive labour in the construction, and where do trade, finance, and government fall? Are unproductive wages carried inside the surplus or deducted before it is struck? How were supervisory salaries and the mixed income of the self-employed treated, and is the series gross or net of depreciation? Which bridge between values and prices does the construction assume, and does the author report any sensitivity of the result to the contested settings? An estimate that arrives with answers to these questions can be cited with confidence proportioned to the answers. An estimate that arrives without them, however striking the number, should be treated as an unfinished claim, and the professional literatures of journalism and teaching both have reason to insist on the difference, since a figure detached from its settings will be repeated by others as though it were a census count.

One further discipline concerns the uses to which a trend may be put. A rising estimated rate over the postwar decades is a finding about the division of value added between the consumption fund of productive workers and the surplus, within the theory’s partition of the economy. It is not a finding about poverty, about real wages, about the distribution among households after taxes and transfers, or about the moral character of any employer, and arguments that need those findings must go and get them from the appropriate evidence. The estimators themselves, at their best, model this restraint; the finding they defend is narrow, and its narrowness is what makes it defensible. Readers who honour the narrowness can use the literature with a clear conscience. Readers who inflate it into a verdict on the whole condition of a society have left the evidence behind, and critics who attack the inflated version are attacking a claim the careful studies never made.

Why the postwar United States became the laboratory

The concentration of the literature on one country and one era is not an accident of scholarly fashion, and understanding it clarifies both the strengths and the limits of the findings. The United States statistical system offers the estimator an unusual combination: national income accounts of great industrial detail extending back to the late 1940s, benchmark input-output tables at regular intervals, employment and hours series that can be cross-walked against the income data, and a single currency and institutional frame across the whole span. The postwar decades also pose the questions the tradition most wanted to answer. Profitability faltered from the mid 1960s, the distributional settlement of the early postwar years came apart after the early 1970s, and the restructuring that followed, in which finance and business services expanded enormously while manufacturing employment shrank, is almost a designed experiment for a framework built on the productive and unproductive distinction. A theory that predicts pressure on profit from the growth of unproductive activity could hardly ask for a better test bed than an economy that spent forty years expanding exactly those activities.

The concentration carries a cost that the literature’s critics and its careful practitioners both acknowledge. Findings established on United States data may travel badly. The country’s accounts embed its institutions: employer provided health insurance swelling the supplements component of compensation, a comparatively small state sector by the standards of Western Europe, an unusually large financial sector by most standards anywhere. The classification decisions that the American data force, and the quantitative weight of each row of the decision tree, are partly American facts. A reconstruction for a Scandinavian economy, where public employment is far larger and wage bargaining compresses the salary distribution, would load the government and supervisory rows with far greater weight and the finance row with less. This is a further reason the trend within a well studied economy commands more confidence than any comparison between economies: the laboratory result is real, and its generality is an open question rather than a presumption.

There is also a generational point about the data themselves. The estimators of the 1980s and 1990s worked from printed volumes and early machine readable files, reconstructing series that statistical agencies had revised repeatedly, so part of their craft was archival: reconciling vintages, bridging classification changes between industrial coding systems, and interpolating benchmark years. The findings that survived that craft have a robustness that a single download cannot confer, because inconsistencies in the source material had to be found and handled by hand. At the same time, revisions move the ground. When an agency redefines the treatment of software investment or of research and development, capitalising what was once expensed, the boundary between intermediate cost and value added shifts in the official data themselves, and a Marxian series spanning the revision must decide whether to accept the new convention, which changes measured depreciation and measured product together, or to rebuild on the old one and forfeit comparability with everything else the agency publishes. The estimates are thus historical artefacts in a double sense: measurements of a period, made with the statistical instruments the period and its aftermath supplied.

Compensation, supplements, and the social wage

A cluster of practical problems gathers around a deceptively simple question: what, exactly, is the wage that the denominator of the ratio should record? The accounts’ compensation of employees includes wages and salaries together with supplements, the employer contributions to pension funds, health insurance, and social insurance schemes. From the standpoint of the value of labour power, the two components are not obviously the same kind of thing. Cash wages buy the worker’s consumption directly. Pension contributions are deferred consumption, still plausibly part of what reproduces labour power across a life. Employer health insurance premiums, in the United States setting, purchase a service whose price has risen far faster than the general price level, so the supplements share of compensation climbed across the postwar decades for reasons connected to the medical sector’s cost dynamics rather than to any change in workers’ command over goods. An estimator who takes total compensation as variable capital imports that medical inflation into the denominator and thereby dampens the measured rise in the rate of exploitation; an estimator who works from wages and salaries alone must explain why deferred and in kind components are excluded from the reproduction fund when the theory’s own account of labour power’s value includes the costs of maintaining the worker and the worker’s family across time. The literature divides on the treatment, and the division matters quantitatively for the American case precisely because the supplements wedge grew so large there.

Taxation and public provision complicate the denominator from the other side, through what the literature calls the social wage. Workers pay taxes from their wages and receive public services, health care in many countries, education for their children, income support in unemployment and old age. Some reconstructions argue that the value of labour power should be measured net, after the tax wedge, with state provision counted as part of the reproduction fund the working class collectively receives; others keep the accounts’ gross magnitudes and treat the state’s whole operation as a charge on the surplus, warning that netting invites double counting when the same tax flows are also financing the unproductive employment already charged elsewhere. The dispute mirrors the treatment of government in the decision tree and interacts with it: an estimator cannot consistently count public health workers as unproductive labour financed from surplus while also counting the health care they provide as part of the value of labour power without specifying the flows with great care. The best studies specify them; the middling ones adopt a convention and flag it; the worst, as in every empirical field, adopt one silently. For the reader, the practical upshot is that the denominator of a published rate is a small theory of the welfare state in disguise, and cross study comparisons that ignore the disguise will misread institutional differences as differences in exploitation itself.

The productivity cross-check: an outside witness

A reconstruction that could not be checked against anything outside itself would deserve the circularity charge in full, so it matters that the estimators possess an independent witness: the official productivity statistics. Labour productivity series, output per hour in the nonfarm business sector or in manufacturing, are compiled by statistical offices for purposes unconnected with Marxian categories, and real compensation per hour is compiled beside them. When output per hour grows faster than real compensation per hour over a long span, the gap between the two lines is, in the theory’s vocabulary, the space in which a rising rate of exploitation must be located if the framework is coherent, since the value produced per hour is outrunning the value required to reproduce the labour that produces it. When the two lines grow together, the framework predicts a stable rate, and a reconstruction that reported a sharply rising one for such a span would indict its own settings. The postwar American record, on the official series, shows the two lines parting company across long stretches and most decisively after the early 1970s, which is the same territory where the reconstructions find their rising trend. The witness is not wholly independent, since both parties draw on overlapping source data, and the productivity lines carry their own well rehearsed problems of output measurement in services and of price deflators. But the categories of the productivity programme were not built to please Marxian conclusions, and their agreement in direction is the closest thing the measurement literature has to corroboration from outside the faith. A careful reader weights it accordingly: not as proof of the reconstruction, but as a reason the reconstruction’s central direction cannot be dismissed as an artefact manufactured inside the categories alone.

The cross-check has a popular cousin that should be distinguished from it, the productivity and pay gap tracked by policy institutes, which sets net productivity growth for the whole economy beside the hourly compensation of a typical worker and reports the wedge between them. That comparison is honest and informative about the decoupling of typical pay from average output, and it is frequently misread as a measure of exploitation. It is not one, for reasons this article has already assembled: its compensation line covers all employees rather than productive workers, its productivity line is an economy wide average rather than a surplus construction, and it makes no attempt to gather the interest, rent, and tax flows that the Marxian numerator includes. The gap and the rate are cousins, not twins. They point in the same direction across the postwar decades because both are downstream of the same distributional shift, the growing distance between what an hour of work produces and what the worker performing it receives, but each is computed on its own partition of the accounts, and neither can be substituted for the other in an argument that needs the other’s precision. The estimators who cite the productivity lines do so as corroboration of direction, which is the correct weight; writers who cite the gap as though it were the rate have repeated the labour share error in a new costume.

The estimators’ arguments among themselves

An outsider could be forgiven for imagining the measurement literature as a choir. From inside, it is an argument conducted over decades, and the argument’s contours are the fastest route to understanding what the estimates are worth. One axis of dispute concerns the services boundary. The classical classification, drawn strictly from the relational criterion, places most market services outside production, and the Shaikh and Tonak architecture follows that logic with detailed exceptions for activities such as transport and repair that continue the production process. Critics within the tradition object that a boundary which counts the labour of a cook in a factory canteen’s supplying firm differently from the same labour hired directly, and which treats enormous swathes of the modern workforce as producing nothing, purchases consistency at the price of describing an economy nobody recognises. Defenders reply that recognition is not the test; the categories were built to track the production and absorption of surplus, and an economy in which a growing share of employment is absorbed in circulation, guarding, and administration is exactly the economy the categories were built to describe. The dispute is live, technical, and conducted with data, which is more than can be said for most exchanges about Marxian categories.

A second axis concerns the status of the aggregate equalities used to bridge values and prices. The New Interpretation’s definitional bridge has the elegance of making the money rate and the value rate coincide at the aggregate, and its critics charge that the elegance is the problem: a bridge built by definition cannot be tested, and conclusions drawn across it inherit the definition’s immunity. The temporal single system interpretation, developed by Alan Freeman, Andrew Kliman, and others chiefly in debates over the rate of profit, proposes a different valuation of inputs and outputs across time and implies its own measurement conventions. A third axis, opened by Emmanuel Farjoun and Moshé Machover in their 1983 work on the laws of probability applied to political economy, questions whether deterministic value magnitudes are the right objects at all, proposing that labour contents and prices be treated as distributions; their stance has influenced the price correlation literature more than the exploitation estimates, but it stands in the background as a reminder that the measurand itself is contested inside the tradition. Through all three axes runs a common standard that the participants mostly honour: settings must be published, derivations must be reproducible from the cited statistical sources, and a rival is answered by rebuilding the series under the rival’s settings and showing where the difference arises. The literature’s failures are failures to meet its own standard, not the absence of one.

The rate across the cycle and across sectors

Two further patterns in the estimates deserve report, because both constrain interpretation. The first is cyclical. In reconstructions for the United States, the measured rate does not glide upward smoothly; it rises and falls with the business cycle, typically strengthening through expansions as output per worker climbs faster than real compensation, and faltering in recessions when output collapses against a wage bill that adjusts more slowly. The cyclical signature matters for two reasons. It warns against reading any short span, a single expansion or a single crisis, as evidence about the long trend, since the cycle can imitate a trend for years at a time. It also connects the measurement literature to crisis theory, because the behaviour of the rate near cyclical peaks, and the question of whether a squeezed or an engorged surplus precedes downturns, is one of the points where the estimates feed back into the theory’s account of instability rather than merely illustrating its distribution chapters.

The second pattern is sectoral dispersion. Where studies construct rates for broad sectors, the manufacturing and goods producing core generally shows a higher measured rate than the economy wide figure the headlines report, partly because the productive boundary is least contested there and partly because the surplus generated in production is partly realised and recorded in trade and finance, sectors whose own accounts show margins rather than production of surplus. Sectoral figures therefore answer a different question from the national figure: not how the whole economy divides its new value, but where in the circuit the surplus originates and where it surfaces. Confusing the two questions produces a characteristic error in secondary writing, in which a high manufacturing rate is quoted as the national rate, or a low margin in retail is quoted as proof that retail workers are lightly exploited, when the framework itself insists that the retail margin is a share of surplus produced elsewhere and that the exploitation of retail workers, in the technical sense, turns on the contested classification of their labour rather than on the margin their employer reports. Keeping origin and surface apart is elementary within the theory and routinely lost outside it, which is why the estimators’ habit of stating the aggregate they are reporting deserves imitation.

The estimator’s workbench: sources as they are actually used

Abstract talk of national accounts can hide how concrete the estimator’s materials are, so it is worth naming the actual instruments. For the United States, the income side comes from the national income and product accounts maintained by the Bureau of Economic Analysis, whose underlying detail tables give compensation, proprietors’ income, and profits by industry at a fineness the headline releases never show. The production side leans on the Bureau’s benchmark input-output tables, constructed at intervals of roughly five years from economic census material, and on the integrated industry accounts produced jointly with the Bureau of Labor Statistics, which reconcile output, intermediate purchases, and employment industry by industry. Capital stock and depreciation come from the Bureau’s fixed assets estimates, built by the perpetual inventory method from investment flows and assumed service lives. Hours and employment by industry come from labour statistics programmes whose industry classifications never quite match the income accounts’, so a nontrivial share of any reconstruction is concordance work: mapping one agency’s categories onto another’s across coding revisions that break naive continuity. Internationally, the corresponding instruments are the OECD’s structural analysis databases and inter-country input-output tables, the World Input-Output Database assembled by a European research consortium, and the Penn World Table’s capital and labour share series, each with its own conventions that an estimator must absorb or correct.

Two practical morals emerge from the workbench. The first is that reproducibility in this literature is a genuine achievement rather than a slogan, because the path from published sources to a Marxian series passes through dozens of documented concordances and imputations, and a study that publishes them, as the best do in appendices and data supplements, exposes itself to exact replication and therefore to exact refutation. The second is that data revisions are a standing condition of the field. Benchmark revisions rewrite history in the source accounts every few years; a rate estimated on one vintage of the accounts will differ in detail from the same method applied to a later vintage, and honest reporting attaches the vintage to the result. Neither moral is peculiar to Marxian work, but both are sharpened there, because the reconstruction multiplies the number of places where a revision or a concordance choice can enter, and because the field’s critics are watching precisely those places.

The boundary of the measurable: unpaid labour and reproduction

Every measurement scheme draws an outer boundary around what it attempts to count, and for this literature the most instructive boundary dispute concerns the unpaid labour performed in households. Cooking, cleaning, childcare, and the care of the elderly reproduce labour power daily and generationally; without them the wage labour force whose exploitation the rate measures would not exist. Yet in Marx’s construction this labour produces no value, because its products are not commodities exchanged on a market, and it therefore enters no numerator and no denominator in the standard estimates. The wages for housework debate of the 1970s, associated with Mariarosa Dalla Costa, Selma James, and later Silvia Federici, pressed exactly here, arguing in different registers that the exclusion renders the wage relation’s foundation invisible and understates the total labour on which accumulation rests. Whatever verdict a reader reaches on that argument’s theoretical claims, its measurement implication is precise and should be stated plainly: the rate of exploitation as estimated is a rate within the sphere of value production, and it is silent, by construction, about the unpaid labour that sphere presupposes.

The silence has a quantitative shadow that feminist economists have worked to illuminate through time use surveys and satellite accounts, which estimate the scale of unpaid household production relative to measured output. Those exercises are not Marxian estimates and do not yield a rate of exploitation in the technical sense, but they discipline interpretation in two ways. They remind the reader that a rising measured rate within the market sphere can coincide with a shifting boundary between paid and unpaid work, as services once performed in households are commodified and move inside the accounts, changing both the productive aggregate and the recorded wage bill for reasons that are partly bookkeeping. And they mark the limit of what the instrument can testify about: a society’s total extraction of unpaid labour time, across market and household together, is a larger question than the one the estimators set out to answer, and their silence on it is a scope condition, not a finding. Precision about scope is the recurring virtue of this literature at its best, and the household boundary is where readers most often need it supplied.

Safe sentences and unsafe sentences

The practical yield of the whole discussion can be compressed into the contrast between sentences a writer may responsibly publish and sentences that outrun the evidence, and the contrast is worth drawing explicitly because this subject attracts overclaiming from every direction. A safe sentence names the study, the economy, the period, and the direction, and lets the level go: Reconstructions of the United States accounts by Shaikh and Tonak and by Moseley, using different classifications, both find the rate of exploitation rising across the postwar decades. Every element of that sentence can be defended in a footnote, and its caution costs it nothing in force. An unsafe sentence takes the same material and strips the qualifiers: American workers are now exploited at a rate of 180 percent. The figure might match one study’s output for one year under one classification, but presented bare it asserts a settled fact about a nation, and no such fact exists in the literature. Between the two lies the large territory of sentences that are defensible with their settings attached: a writer may quote a level from a named study for a named year, provided the classification stance travels with the number and the spread of rival estimates is acknowledged where the argument depends on precision.

Two further unsafe patterns recur often enough to name. The first converts the trend into a welfare claim, writing that a rising rate shows workers being ground down or living standards collapsing; as established earlier, the ratio can rise while real wages rise, because productivity growth in wage goods industries cheapens the value of labour power even as the basket it buys expands. The second converts a sectoral or partial figure into a national one, or moves a figure across a border, quoting an American estimate in an argument about a European economy whose state sector, wage structure, and statistical system would each shift the construction. Both patterns share the same root as the labour share error: a number is treated as a portable fact rather than as the output of a situated construction. The remedy is unglamorous and complete. Keep the number attached to its study, its settings, its economy, and its years, and the literature will support the weight placed on it; detach any of the four, and the citation becomes decoration wearing the costume of evidence.

The first generation: Gillman, Mage, and the manufacturing studies

The reconstruction tradition did not begin with Shaikh and Tonak, and its earlier history explains some of its habits. Joseph Gillman’s study of the falling rate of profit, published in 1957, worked with the statistical material available for United States manufacturing and asked whether Marx’s law could be seen in the record of the century’s first half. Gillman’s interest centred on the rate of profit and the composition of capital, but the exercise required him to build the rate of surplus value as an intermediate magnitude, and his finding set the pattern the later literature would repeat: the rate of surplus value in his reconstruction rose across the period he examined, while the rate of profit was pressed downward by the growing mass of capital relative to the labour it set in motion. The postwar debate his book opened, about whether the tendency he identified was a law, a period pattern, or an artefact of his price measures, rehearsed nearly every methodological quarrel the mature literature would later formalise, including the dependence of the result on how constant capital is valued.

Shane Mage’s doctoral dissertation of 1963, written at Columbia University, carried the estimation to the whole economy and to a longer span, and it remains a reference point because Mage confronted the productive and unproductive distinction head on with the data of his day. His reconstruction found the rate of surplus value rising across the decades he covered, and his careful separation of production workers from the growing ranks of circulation and administrative employees anticipated the central mechanism of Moseley’s later account by a generation. The dissertation circulated in the samizdat fashion of unpublished theses, cited by those who could obtain it, and its influence ran ahead of its availability. Between Gillman and Mage on one side and the 1990s rebuilders on the other stands a scatter of partial studies, many confined to manufacturing because the census of manufactures offered the firmest ground, which shared a limitation the later work was designed to escape: a rate estimated for one sector cannot speak for an economy in which the surplus is realised, taxed, and spent across all the others. The move from sectoral to whole economy reconstruction, with the surplus gathered wherever it surfaces in the accounts, is the methodological signature of the mature literature, and it is what makes its estimates answers to the question the theory actually poses.

What would settle more: the data the field wishes it had

It clarifies the state of the evidence to ask what additional material would genuinely narrow the disputes, as opposed to merely multiplying estimates. Four wishes recur in the methodological writing. The first is compensation data split simultaneously by industry and by occupation, over long spans, so that supervisory, sales, and clerical components could be separated from production employment inside each industry rather than allocated by assumption; some statistical systems approach this in benchmark years, none sustains it across the postwar span. The second is hours data classified by function, which would let the labour time construction be built with the same care as the money construction and would discipline the bridge between them empirically instead of by convention. The third is linked employer and employee microdata of the kind several national statistical offices now maintain under restricted access, which would allow the surplus and the wage bill to be reconstructed from firm records upward and compared against the accounts based aggregates, exposing where the concordances bend the result. The fourth is a longer and more consistent run of input-output tables for more economies, since the production side reconstruction is only as strong as its benchmarks and interpolation between them is one of the quieter sources of divergence among studies.

A neighbouring research programme shows both the promise and the limit of such material. The distributional national accounts project associated with Thomas Piketty, Emmanuel Saez, and Gabriel Zucman also reclassifies official statistics, allocating the whole of national income to individuals and groups, and it has demonstrated how much public and scholarly traction a well documented reconstruction can achieve. Its categories, however, are distributional rather than functional in Marx’s sense: it can say who receives income without sorting the labour that produced value from the labour paid out of it. The Marxian programme’s distinctive question would survive an abundance of microdata, because the productive boundary is a theoretical line that no dataset draws by itself. Better data would narrow the disputes that are about the world, the sizes of flows and the weights of sectors, and would leave untouched the disputes that are about the framework, which is another way of stating the article’s central finding. The classification dominance rule is not a complaint that the data are poor. It is the observation that, in this field, the argument about categories is the argument, and it will continue under any feasible improvement in the statistics.

For students and exam candidates

Cite this literature the way its own practitioners write it, and an examiner will see the discipline immediately. A strong answer states that the rate of exploitation has been estimated from national accounts, names a study such as Shaikh and Tonak or Moseley, reports the direction found for the postwar United States rather than a bare level, and identifies the classification of productive and unproductive labour as the decision the result depends on. A weak answer quotes a percentage with no source, or offers the falling labour share as though it were the Marxian rate, or concludes that a rising rate proves living standards fell. Each of those errors is avoidable in a single sentence of care. Where a question invites evaluation, the unfalsifiability objection and the convergence reply are the two moves that earn credit, provided the concession on levels is included: markers reward the candidate who can say both what the estimates show and what they cannot.

For teachers and seminar leaders

This topic teaches best when students rebuild a small estimate themselves before they are shown a finished one. Give a seminar a single year of simplified accounts and let groups draw the productive boundary in different places; the spread of rates they produce, from identical data, teaches the classification dominance rule more durably than any lecture summary. Follow with the published studies and ask whose settings each group reinvented, then set the labour share alongside the reconstructions and have the class enumerate why the two diverge. The predictable misconceptions to surface early are that the rate is a published statistic, that unproductive means useless, and that a rising rate implies falling real wages; each dissolves under the arithmetic the students have just done. Assessment should credit the naming of settings and the trend level distinction, and should treat any unattributed national figure, however confidently presented, as the error it is. A closing exercise worth the time asks students to take one newspaper paragraph that quotes a rate and rewrite it to the standard of the safe sentences discussed in this article, settings attached and direction distinguished from level; the rewrite is usually shorter than the original, which makes its own point about how much of the original was decoration.

For researchers, historians, and journalists

Researchers extending this literature face three standing obligations. Publish the full classification, industry by industry, with the treatment of supervisory pay, mixed income, government, and depreciation stated beside the headline series; deposit the concordances so the reconstruction can be replicated from the cited vintage of the accounts; and report sensitivity to the contested settings as a matter of course rather than under referee pressure, since the field’s credibility rests on showing the fan of results and not only the preferred line. Historians using the estimates should remember that they are artefacts of two histories at once, the economy measured and the statistical system that measured it, and that revisions to the accounts rewrite the earlier series in ways that must be checked before long comparisons are drawn. Journalists have the narrower duty set out in the safe sentences section: name the study, the economy, and the period, report the direction, and let no level travel without its settings. Keeping a working notebook of classification decisions, study by study, is the practical habit that makes all of this manageable, and the Marxism study notes tool is built for exactly that kind of structured record, where each estimate can be filed with its boundary choices and deductions attached rather than as a naked number.

What is settled and what remains disputed

Settled, within the limits this article has drawn: the rate of exploitation is estimable in the sense that complete, documented, reproducible reconstructions of it exist for real economies; the productive and unproductive classification is the dominant determinant of the levels those reconstructions report; the labour share is a different magnitude and no proxy for the rate; and the direction of movement in the postwar United States, upward across the decades on the balance of rival constructions, is as robust as anything in this literature. Disputed, and likely to remain so: the level of the rate in any economy and year; the correct placement of the services boundary and of supervisory pay; the proper bridge between value and price magnitudes; and the standing charge that the whole procedure, for all its internal discipline, measures the framework’s commitments as faithfully as it measures the economy. A reader who can hold both lists at once, and can say which list a given claim belongs to, has what this article set out to provide. The estimates neither crown the theory nor collapse under their critics. They mark the point where an argument about capitalism becomes an argument about evidence, with the machinery of the argument left open on the table for anyone to inspect.

An industry estimate in miniature

The national construction can be hard to picture, so it helps to shrink the problem to a single industry and watch the same decisions recur at smaller scale. Suppose a researcher sets out to estimate the rate of exploitation for motor vehicle manufacturing in one country and year. The statistical raw material is richer than the national case in one respect and poorer in another. Richer, because production censuses report the industry’s shipments, materials consumed, and production worker wages with a directness the aggregate accounts lack, and the value added of the industry can be built from the output side rather than inferred. Poorer, because the industry’s boundary in the statistics is an administrative artefact: the census establishment that assembles vehicles may also house the design studio, the captive finance office, and the parts warehouse, and its reported wage bill mixes the assembly line with the showroom’s accountants. The researcher’s first task is the familiar one of dissection, separating production workers from the industry’s own unproductive staff using occupational detail where the census supplies it and ratios where it does not.

The numerator poses the industry specific version of the realisation problem. The surplus produced by the industry’s productive workers is not the industry’s recorded profit. Some of it leaves as interest on the firm’s borrowings, some as rent on leased plants, some as the margins of the dealer network that sells the vehicles, and some returns as the fees of the advertising agencies and logistics firms the manufacturer buys from, each of which books its own apparent profit from value the assembly process created. An industry estimate must therefore decide whether it is measuring the surplus produced within the industry’s walls, traced outward through these channels, or merely the surplus retained under the industry’s own name in the accounts. The two objects differ widely, and published industry figures are not always candid about which they report. The denominator has its own local puzzles: the industry’s production workforce includes maintenance crews whose classification the theory debates, apprentices whose wages sit below the value of their labour power as conventionally measured, and agency workers supplied by employment services firms, whose wages appear in the accounts of an entirely different industry while their labour is performed on the assembly line. Each puzzle is small; together they decide the second digit of any result, which is one more reason the literature reports industry rates as illustrations of structure rather than as league positions.

Two ratios, two literatures: keeping exploitation and profit apart

Because the rate of exploitation and the rate of profit share a numerator in Marx’s scheme, their measurement literatures are entangled, and a reader benefits from seeing exactly where they separate. The rate of profit divides surplus value by the total capital advanced, constant and variable together, so its estimation requires everything the exploitation estimate requires plus a valuation of the capital stock, with all the further decisions that valuation drags behind it: historical cost or replacement cost, the treatment of capacity utilisation, the dating of inputs and outputs. The exploitation rate is spared the capital stock problem entirely, which is one reason its estimates command a measure of agreement the profit rate literature does not. When Moseley argues that a rising rate of surplus value coexisted with a falling rate of profit in the postwar United States, the claim’s two halves rest on different evidentiary footings, and the half about exploitation is the firmer. Critics of the profit calculations, including the temporal single system writers who dispute the valuation conventions of their rivals, can accept the exploitation series while rejecting the profit series built beside it, and some do.

The separation also disciplines a common inference in the other direction. A measured rise in the rate of exploitation does not predict a rising rate of profit, because the denominator of the second ratio grows with accumulation and the deductions for unproductive activity grow with the surplus itself. Journalists and students who move from an exploitation headline to a profitability conclusion have crossed between the two literatures without noticing the border. The estimators who work on both magnitudes are the most careful about the crossing, precisely because their own results display the divergence: the postwar United States, on the reconstructions surveyed here, is an economy that became more exploitative by the Marxian measure while becoming less profitable by the same scheme’s other measure, and any summary that reports only one of those movements has told half of a two part story.

Rival measures: markups, factor shares, and the exploitation rate

The rate of exploitation is not the only distributional magnitude competing for the same headlines, and part of measuring it well is knowing what its rivals measure instead. The labour share has already been dealt with. Two others deserve the same treatment, because journalists reach for them interchangeably and they answer genuinely different questions. The first is the markup, the ratio of price to marginal cost, whose estimation from production data became a prominent research programme through the work of Jan De Loecker and Jan Eeckhout and their collaborators, who reported rising average markups for United States firms across the closing decades of the twentieth century. A markup measures market power in product markets: the firm’s ability to price above the cost of an additional unit. It is silent about how the resulting revenue divides between the firm’s workers and its owners, silent about which workers produced value, and compatible with any rate of exploitation whatsoever, since a competitive firm with no markup can still organise production in which surplus labour is performed, and a monopolist’s markup can be absorbed by the salaries of its own unproductive apparatus. The two literatures occasionally borrow each other’s rhetoric, and the borrowing is always a confusion: monopoly rent and surplus value are different objects, extracted at different points of the circuit, and evidence for one is not evidence about the other.

The second rival is the family of factor share and top income statistics, which partition national income among labour, capital, and the highest earners. Those statistics have transformed public knowledge of distribution, and the Marxian estimator can learn from their craft, particularly their treatment of tax data and their transparency about imputation. Yet the partition they draw runs between persons and income types, not between functions in production. A chief executive’s salary and a cleaner’s wage enter the same labour share; a landlord’s rent and a manufacturer’s profit enter the same capital share; and no rearrangement of the published shares will recover the Marxian numerator, because the information about which labour was productive is not contained in who received what. This is the deepest reason the reconstruction literature exists at all. If the rate of exploitation could be read off distributional statistics, however refined, there would be nothing to reconstruct. The categories cross cut, and any honest measurement has to begin by taking the accounts apart along the theory’s lines rather than along the lines of income receipt. Readers who keep the three families of measures distinct, exploitation rates, markups, and factor shares, will find that much public argument about distribution dissolves into parties citing different instruments as though they were thermometers in the same room.

Beyond the laboratory: Greece and the wider reconstruction literature

The Greek reconstructions repay attention precisely because Greece violates the laboratory conditions that make the United States tractable, and watching the method cope is instructive. Lefteris Tsoulfidis, Persefoni Tsaliki, and their collaborators have built long run estimates for the Greek economy across the postwar decades, working with national accounts that must accommodate a large agricultural sector, extensive self-employment in small family enterprises, a substantial state, and successive statistical revisions through the country’s European integration. Every row of the decision tree carries extra weight in such a setting. The proprietor split is not a refinement but a central determinant, because a workforce in which a third or more of employment long sat outside the wage relation cannot be partitioned by a rule of thumb without the rule deciding the result. The boundary between production and services must be drawn through an economy where tourism mixes transport, accommodation, and commerce inside single establishments. The researchers’ response follows the tradition’s best practice: an explicit classification argued from the structure of the Greek accounts, headline series accompanied by variants, and findings stated as movements across decades rather than as levels to be set beside American ones. Their central finding, a rate of surplus value that rose across much of the postwar period with marked interruptions around the country’s political and economic ruptures, is reported by its authors with the classification attached, which is what allows a reader to use it.

Wolff’s earlier study of Puerto Rico makes a complementary point from a different direction, since the island’s accounts were shaped by its fiscal relationship with the United States and by the weight of American owned manufacturing operating under tax concessions. There the realisation problem dominates: surplus produced by Puerto Rican workers surfaced substantially in the profit reports of mainland corporations, so an estimate confined to income recorded on the island would understate the numerator by construction, and the study’s interest lies in how it traced the flows across the accounting border. Read together, the Greek and Puerto Rican cases show the two ways an estimate can be bent by its setting, through the composition of employment at home and through the geography of ownership abroad, and they explain why the wider literature’s comparative value lies in its variety of difficulties rather than in any table of levels. A researcher entering a new national case inherits the decision tree and a gallery of worked examples of what each row costs in different conditions; that inheritance, more than any single series, is the tradition’s cumulative product.

Translating the rate back into the working day

The money ratio has one interpretive virtue that should not be surrendered in the technical thickets: under the construction the estimators use, it translates back into the division of the working day from which the concept began. A rate of 100 percent means the two portions are equal, so that of an eight hour day, four hours replace the value of labour power and four produce surplus. A rate of 150 percent divides the same day into roughly three hours and twelve minutes of necessary labour against four hours and forty eight minutes of surplus labour. The translation is legitimate because the ratio is dimensionless: whatever the money magnitudes, the proportions carry over, provided the bridge between values and prices is granted at the aggregate and provided intensity is held in the background. This is the form in which the estimates speak most directly to the theory’s founding image, and it explains the hold the concept keeps on readers who will never open an input-output table. The abstraction of the accounts resolves, at the end of the procedure, into a statement about time: how much of the social working day is spent reproducing the workers, and how much is spent producing the surplus that the rest of the social machinery consumes.

The translation also marks its own limits with unusual clarity. It describes an average over millions of working days of different lengths and intensities, in employments the classification sorts into different relations to value, and no individual worker’s day divides at the average point. A nurse in a public hospital does not appear in the productive aggregate at all under the dominant treatment; a machinist’s day divides at a point set by the industry’s conditions, not the nation’s. The average is a property of the construction, informative about the system’s proportions and mute about any biography inside it. Writers who move from the translated figure to a claim about what a particular worker’s afternoon contains commit the ecological fallacy in period dress. Kept at its own level, though, the translation earns its place as the literature’s most quotable product, and unlike the bare level it survives the classification disputes in rough form: rival reconstructions that place the level far apart still agree that the surplus portion of the productive working day grew across the postwar decades, which is the trend finding stated in the theory’s native units.

Frequently Asked Questions

Q: How do you measure the rate of exploitation in a country?

You measure it by reconstruction, not by looking it up. The researcher takes the national accounts, reclassifies their categories along Marxian lines, and builds two aggregates: an estimate of the surplus value produced in a year and an estimate of the variable capital paid to the productive workforce. The rate is the first aggregate divided by the second. Every step requires decisions the accounts cannot make themselves, above all which labour counts as productive, how the wages of unproductive workers are treated, how government and self-employment are handled, and whether the calculation runs gross or net of depreciation. Because the decisions drive the result, a competent study publishes them alongside the figure, and a competent reader asks for them before using the figure for anything.

Q: Which national accounts categories have to be adjusted?

Nearly all of the major ones. Compensation of employees must be dissected, because it mixes the wages of productive workers with the salaries of managers, supervisors, sales staff, and government employees, which the Marxian scheme treats very differently. Corporate profits must be expanded, since interest, rent, and taxes on production are shares of the surplus in this framework rather than deductions from it. Proprietors’ income must be split between a labour component and a surplus component, because the accounts record it as a single mixed flow. Government output, which the accounts value at cost, must be reclassified according to the study’s treatment of the state. Depreciation, recorded as consumption of fixed capital, must be accepted on the statistical office’s assumptions or rebuilt. The adjustments are the method; there is no unadjusted version of the estimate.

Q: Has the rate of exploitation risen or fallen?

For the postwar United States, the economy studied most thoroughly, the balance of the reconstruction literature finds that it rose. Studies using different classifications of productive labour, different treatments of supervisory pay, and different starting aggregates, from Shaikh and Tonak through Moseley to Mohun, converge on an upward movement between the late 1940s and the end of the twentieth century, uneven across the decades and interrupted by recessions. That convergence supports a claim about direction, not about any precise level in any year, and it should not be extended to other countries without their own reconstructions. Nor does a rising rate imply falling living standards, since the value of labour power can fall through cheaper wage goods while the real wage basket grows. The safe statement is the directional one, with the study and the period named.

Q: Is the labour share the same as the rate of exploitation?

No, and the difference is structural rather than a matter of refinement. The labour share divides all employee compensation by national income, so its numerator counts the salaries of senior managers, finance professionals, and unproductive staff as labour’s portion, while its denominator is an accounting aggregate with no relation to variable capital. The Marxian rate divides a reconstructed surplus by the wages of productive labour only, after the accounts have been reclassified around the productive boundary. A rise in executive pay can therefore lift the labour share and the Marxian surplus at the same time, sending the two measures in opposite directions. The labour share is a valuable statistic about the distribution between employees as a whole and property income, and it constrains any reconstruction built from the same accounts, but substituting it for the rate of exploitation changes the subject.

Q: Why do estimates of the rate of exploitation differ so much?

Because the estimate is the output of a chain of classification decisions, and competent researchers set the chain differently. The largest single source of divergence is the boundary between productive and unproductive labour, which moves both sides of the ratio at once: narrowing the boundary shrinks variable capital and swells the surplus that must carry the excluded wages. Behind it come the treatment of supervisory and managerial pay, the handling of government employment, the split of self-employment income, the choice between gross and net magnitudes, and the bridge assumed between money prices and values. Two studies can share every datum and differ by a factor of two in the reported level while agreeing on the trend. The divergence is therefore information about the framework’s sensitivities, not evidence that one study must be fraudulent or incompetent.

Q: How does the productive and unproductive split change the result?

It changes the result more than any other decision, which is why this article names the pattern the classification dominance rule. Draw the boundary narrowly, around goods production and a short list of value adding services, and most service, trade, finance, and government wages leave the denominator and reappear as charges on the surplus, so the measured rate climbs steeply. Draw it generously, and the same economy returns a far lower figure. The split also shapes the trend, because unproductive employment grew across the postwar decades in the advanced economies, so a narrow boundary registers that growth as upward pressure on the rate while a generous one absorbs it quietly. No data revision in the source accounts has an effect of comparable size, which is why debates about the estimates are always, at bottom, debates about where the boundary belongs.

Q: Can you compare rates of exploitation between countries?

Only with great care, and usually not at the level of the headline figure. Two countries’ estimates are comparable only if the reconstructions behind them use the same classification rules, and even then the statistical raw material differs: countries vary in the size of the state sector, the prevalence of self-employment and informal work, the treatment of the financial sector in their accounts, and the industrial detail available for drawing the boundary. A pairing of national rates quoted without a harmonisation exercise behind it compounds differences of economy with differences of convention in unknown proportions. What can be compared defensibly is direction, where several national literatures find rates rising across overlapping decades, and structure, where the decision tree shows which national features load most heavily on the result. The league table of countries by rate is a use the evidence does not support.

Q: Which studies have estimated the rate of exploitation?

The tradition runs from Joseph Gillman’s 1957 study of United States manufacturing and Shane Mage’s 1963 doctoral reconstruction of the whole economy, through the systematic framework of Anwar Shaikh and Ertugrul Ahmet Tonak, published in 1994 as Measuring the Wealth of Nations and applied to the United States from 1948 to 1989. Fred Moseley’s 1991 study of the postwar United States economy used the reconstruction to explain falling profitability through the growth of unproductive labour, and Simon Mohun’s estimates from the mid 1960s onward refined the treatment of supervisory pay. Gérard Duménil and Dominique Lévy developed the related argument that senior managerial income is a share of the surplus. Beyond the United States, Edward Wolff estimated the rate for Puerto Rico, and Lefteris Tsoulfidis and collaborators built long series for Greece, with further applications across Europe, Asia, and Latin America.

Q: Is the measured trend more reliable than the measured level?

Yes, and the reason is mechanical rather than rhetorical. A classification decision held constant across a series shifts every year’s level in the same direction, so it distorts the level in each year while leaving the movement between years comparatively untouched. The trend is therefore governed more by the underlying distributional data, above all by the race between productivity growth in the productive sectors and the growth of productive workers’ real compensation, and less by the estimator’s settings. This is why rival studies whose levels sit far apart can converge on direction, and why the convergence counts as evidence. The reliability is relative, not absolute: a trend can still be bent where a contested sector grows fast enough for its treatment to matter, as with supervisory employment, which is why the trend claim is strongest for long spans and weakest for short ones.

Q: What are the main criticisms of these estimates?

The strongest criticism is that the reclassification the estimates require involves so much discretionary judgement that the resulting series measures the analyst’s framework rather than the economy, and that the procedure is unfalsifiable in practice, since a committed analyst can redraw a boundary to reach a preferred result. Supporting criticisms point to the rarity of published sensitivity analysis, to the dependence on official depreciation and imputation conventions that the estimator inherits unexamined, and to the gap between value magnitudes and the money prices the statistics record. A weaker criticism, that the exercise inherits whatever fate awaits the labour theory of value itself, targets the theory rather than its measurement. The defenders’ reply rests on the convergence of direction across studies built with different settings, and on the fact that the discretion involved must be published and argued for, unlike the quiet conventions inside mainstream statistics.

Q: What data would you need to estimate the rate of exploitation for a single industry?

Start with the industry’s output and intermediate purchases, from a production census or input-output table, so that value added can be built from the production side. You then need the wage bill split by function, separating production workers from the industry’s own supervisory, sales, and administrative staff, which usually requires occupational detail the headline sources lack. Hours data matter if the construction is to be checked in labour time. You also need the industry’s payments of interest, rent, and taxes, since the surplus produced in the industry surfaces partly in those flows and partly in the margins of the dealers and service firms around it, and a decision about whether the estimate traces the surplus produced or merely the profit retained. Finally you need a ruling on awkward categories the industry contains, agency workers booked to another industry, maintenance crews, apprentices, before any ratio is struck.

Q: How do researchers treat government employees when estimating the rate of exploitation?

The dominant treatment classifies general government as unproductive activity financed from the surplus, so government compensation is excluded from variable capital and counted among the charges that surplus value must support. The reasoning is that state employees are paid from taxation, and taxation is a claim on value produced in the productive sectors, whatever the social usefulness of the work. A minority approach separates state enterprises that sell their output, utilities and trading bodies, and treats their workforces as productive while keeping general administration outside. The choice carries unusual weight because public employment is a large share of all employment in advanced economies, so the treatment moves the level of the estimate substantially and shapes cross country comparisons. Some reconstructions add a refinement, treating part of state provision as a social wage that enters the reproduction of labour power, which must then be specified carefully to avoid counting the same flows twice.

Q: Why does depreciation complicate Marxian measurement from national accounts?

Because the Marxian concept is net while the accounts offer depreciation only as a modelled estimate. Value added in Marx’s sense is the new value created in a year, after the value of used up means of production has been replaced, so the surplus must be struck after depreciation is deducted. The accounts do publish a deduction, consumption of fixed capital, but it is built from assumed service lives for buildings and equipment, price indices for capital goods, and decay patterns chosen by the statistical office. An estimator who adopts the official net figures inherits all of those conventions; an estimator who works gross avoids the inheritance and drifts from the concept. The complication grows with capital intensity, since a rising stock of long lived assets makes the depreciation line larger relative to value added, so the choice between gross and net treatment can bend the measured trend and not only the level, which is why the decision appears as a row of the estimation decision tree.

Q: What should a journalist ask before quoting a single rate of exploitation figure?

Ask four questions and decline the figure until they are answered. Which study produced it, using which classification of productive and unproductive labour? For which economy and which years does it hold, and is the number a level from one construction or a direction supported by several? How were supervisory pay, government employment, and self-employment income treated, and is the series gross or net of depreciation? What do rival reconstructions of the same economy report, and how wide is the spread? A figure that survives those questions can be quoted with its settings attached, in a sentence that names the study and the period. A figure that arrives without answers is not evidence but decoration, and quoting it bare, as a settled fact about a nation, misrepresents a literature whose own practitioners state their findings as directions and ranges.

Q: How does self-employment income distort estimates of the rate of exploitation?

Self-employment income arrives in the accounts as a single mixed flow, proprietors’ income, which pays for the proprietor’s labour, compensates the capital the proprietor owns, and rewards risk, all in one undifferentiated sum. A Marxian estimate cannot leave the flow whole, because its labour component belongs near the denominator of the ratio while the remainder belongs in the surplus. Every division rule embeds an assumption: imputing the average employee wage for the industry may overstate the labour component where proprietors work for less than hired rates and understate it where skilled practitioners command premiums, while splitting by the industry’s labour share imports the distribution the estimate is trying to measure. The distortion is modest where proprietorship is a small share of employment and can dominate the result in economies, or earlier decades, where family farms and shops employ a large fraction of the workforce.