Unpaid Care Is Worth 9% of Global GDP and Counts as Zero

Start with the number that governs everything else. Data from 64 countries covering two-thirds of the world's working-age population show 16.4 billion hours of unpaid care work performed every day [2]. Valued at an hourly minimum wage, the most conservative method available, that work comes to about $11 trillion a year, or 9% of global output [2]. ✓ Established That is a sector larger than any national economy except those of the United States and China, recorded at zero in every set of national accounts on earth. The exclusion is not a local statistical quirk. It is written into the international standard that every statistical office follows.

The paid half of the sector is not small either. The ILO counts 381 million care workers worldwide, 11.5% of total global employment, of whom 249 million are women [2]. Education employs 123 million of them, health and social work 92 million, and domestic work another 70 million [2]. ✓ Established What makes care different from other large sectors is that its paid and unpaid halves are direct substitutes. Every hour a public system declines to fund is an hour a household absorbs, and every hour a household cannot absorb is an hour a market must price or a person goes without. The boundary between the two moves constantly, in both directions, and nothing in the statistical system records the movement.

The most carefully constructed national estimate comes from the United States. The AARP assessment published in March 2026 counts 59 million family caregivers providing 49.5 billion hours in 2025, valued at $20.41 an hour for a total of $1.01 trillion [1]. ✓ Established That is roughly 24 million full-time equivalent workers, about 17% of the American full-time workforce [1]. The total exceeds combined federal, state and local Medicaid spending and is close to double all out-of-pocket health spending in the country [1]. Britain produces the same shape of number: unpaid care in England and Wales was valued at 162 billion pounds a year, above the 156 billion pound NHS England budget for the same period [14]. The Bureau of Economic Analysis arrives from the opposite direction, putting household production at 16% of an extended measure of GDP in 2024 [22].

The consequence of that exclusion is fiscal rather than philosophical. A finance ministry projecting the cost of an ageing population counts the residential places it will have to fund and the pensions it will have to pay. It does not count the unpaid hours its own projection silently assumes will continue to be supplied. When those hours stop arriving, because the daughters who once supplied them are in paid work, live in another city, or were never born, the cost reappears as a service the state must buy or a family must purchase. It looks like a spending shock. It is nothing of the kind. It is a liability that was always present and was never written down.

The demand side is growing and changing shape at the same time. Some 2.1 billion people needed care in 2015: 1.9 billion children under 15 and 200 million older people [2]. The ILO expects 2.3 billion by 2030, with an additional 200 million older persons set against a shrinking cohort of children [2]. ✓ Established That substitution matters, because care for an older person is more hours-intensive, runs for longer and cannot be folded into a school day. Dementia alone absorbs 133 billion hours of informal care a year and cost the world economy $1.3 trillion in 2019, about half of it in the unpaid time of relatives [16]. ✓ Established

The rest of this report follows a single decision through its consequences. Because care output is not measured, care wages are set without reference to the value of what is produced. Because care labour is not counted, migration policy treats it as interchangeable with any other kind. Because unpaid hours sit on no balance sheet, fiscal projections omit the largest input to the systems they model. Those are not separate failures. They are one failure, expressed in wage-setting, in labour supply, in demography and in migration, and the evidence on each is now specific enough to be stated in numbers.

The rule that removes care from GDP has a name and a logic. National accounting distinguishes a general production boundary from the narrower boundary used to compute GDP, and unpaid services produced and consumed within the same household fall outside the second [30]. The test that defines production is the third-person criterion set out by Margaret Reid in 1934: an activity is productive if it could in principle be delegated to a paid worker [31]. Cooking, cleaning, childminding and bathing an elderly parent all pass that test comfortably. They are excluded anyway, by a separate decision about which productive activities the accounts will actually record.

Historians of national accounting have shown that the decision was contested when it was made and has been revisited repeatedly since [30]. The exclusion was defended on practical grounds: imputing a price to household services is difficult and the result is an estimate rather than a transaction. The difficulty is that the same objection applies to owner-occupied housing, which the accounts do impute, and to a large share of government output, which they also impute. The boundary is therefore not a line between the measurable and the unmeasurable. It is a line between activities the accounts judged worth the trouble and activities they did not.

Feminist economists made the structural argument early and it has not been answered [31]. A production boundary that excludes household services does not merely omit a category. It systematically omits the work of one sex more than the other, and it does so inside the single number governments use to judge whether policy is working. The ILO measured 76.2% of all unpaid care hours as performed by women [2], and UN Women puts the daily imbalance at 2.5 times more hours for women than for men [3]. ✓ Established When the omitted category is distributed that unevenly, the omission stops being neutral.

The gap between the commitments and the measurement is the striking part. Governments agreed in 1995 to devise statistical means to make unremunerated work visible [32]. Three decades later, 96 countries have reported the corresponding United Nations indicator at least once, 46 have reported it exactly once, and six have reported it often enough for anyone to compute a trend [28]. ✓ Established Time-use surveys are expensive, they compete with everything else in a statistical office's budget, and they produce a number that no fiscal rule refers to. The predictable result is that the data exist as a demonstration rather than as an input.

Where the measurement has been done properly, the results are consistent and large. The Bureau of Economic Analysis, which has tracked household production for nearly three decades, put its value added at 16% of an extended measure of GDP in 2024 and about a third of extended household consumption [22]. UN Women reports that in some countries unpaid care would exceed 40% of GDP if it were counted [3]. ◈ Strong Evidence These are not fringe estimates produced by campaign groups. They are official statistics, published by the same agencies that produce the headline number, and deliberately kept outside it.

The practical effect is a systematic bias in the direction of policy. A reform that moves care out of a household and into a paid service registers as growth even when the quantity of care is unchanged, because output that was invisible becomes visible. A reform that moves care the other way, by closing places, tightening eligibility or raising co-payments, registers as saving, because the cost is transferred into a sector the accounts do not observe. Governments therefore operate a measurement system that understates the cost of the second kind of reform. Sweden supplies the cleanest demonstration of what follows, and section four returns to it.

The wage evidence is unusually clean, because the sector is large, its occupations are well defined and its pay data are collected directly. The PHI assessment for 2025 puts median annual earnings for direct care workers in the United States at just under $26,000, with home care pay at a median of $16.77 an hour in 2024 against $13.07 in 2014 in inflation-adjusted terms [10]. ✓ Established A decade of nominal growth produced a real gain of under four dollars an hour. Thirty-six per cent of the workforce lives in or near poverty, defined as household income below 200% of the federal poverty level [10]. This is one of the largest growth occupations in the American labour market, and it pays wages that qualify its holders for the public benefits they help others obtain.

The explanation most often offered is skill, and it does not survive contact with the data. ILO analysis of United States data for 2015 to 2024 finds that a one percentage point increase in the female share of an occupation is associated with a 0.22% decline in wages for women in that occupation and a 0.20% decline for men [21]. ◈ Strong Evidence The penalty applies to men in feminised occupations as well as to women, which rules out any explanation resting on the characteristics of individual workers. It attaches to the job. Occupations with a relational or caring component pay less than occupations demanding comparable education and cognitive complexity without one [21].

Turnover is the mechanism that converts a wage level into a service-quality problem. Annual turnover in home care runs near three quarters of the workforce, and PHI projects 9.7 million total direct care job openings between 2024 and 2034 once occupational transfers and labour-force exits are counted [10]. ✓ Established An employer facing that rate cannot amortise training, cannot build supervisory depth and cannot offer continuity of carer, which is the attribute service users rank highest. A low wage does not simply transfer income from workers to purchasers. It destroys the asset, the trained and familiar worker, that the service consists of.

The pattern repeats across systems with entirely different financing. Women make up 87% of the long-term care workforce across the OECD [7]. Japan will need about 2.72 million care workers by fiscal 2040, roughly 570,000 more than in fiscal 2022, in an economy with the tightest labour market in the developed world [12]. In England the vacancy rate has fallen to 6.2%, its lowest since 2015-16, yet posts filled by British nationals dropped by 40,000 in a single year and by 130,000 since 2020-21 [13]. The English vacancy rate improved because migrants filled the gap, not because domestic workers returned to the sector.

The claim that better pay is unaffordable deserves a specific answer rather than a general one. Long-term care is labour, and labour-intensive services do not deliver the measured productivity growth that manufacturing does, so their relative cost rises over time whatever governments do [38]. ◈ Strong Evidence That mechanism is real and it is not a policy failure. What is a policy failure is the response: holding the relative cost down by suppressing the wage rather than by funding the price. The result is not a cheaper service. It is the same service delivered by a workforce that turns over three times in four years, financed by the unpaid overtime of relatives and by the wage suppression of migrants.

The private equity evidence shows what happens when the cost pressure is answered through ownership instead. Research covering United States nursing homes finds that private equity ownership raises short-term mortality by about 11%, alongside lower nurse staffing and reduced compliance with care standards [20]. ◈ Strong Evidence A systematic review of studies published between 2000 and 2024 finds higher deficiency counts, higher hospitalisation rates and higher mortality under that ownership model, with exceptions where capital was injected into distressed facilities [39]. In a labour-intensive service with a fixed reimbursement rate, the margin has to come from the labour.

The labour-supply effect is the part of the care economy that macroeconomic policy can see, because it appears in participation rates. UN Women puts 708 million working-age women outside the labour market for reasons of unpaid care, 45% of all working-age women, against 5% of men [3]. ✓ Established The earlier ILO count of 606 million women unavailable for paid work because of unpaid care, against 41 million men, produces the same ratio from different data [2]. No other single factor removes comparable numbers of people from the workforce in economies that describe themselves as short of workers.

The mechanism is a price. Child care in the United States averaged $13,184 a year in 2025, up 23% since 2021, taking 10% of median income in a two-parent household and 33% in a single-parent household [18]. ✓ Established A second earner facing a marginal tax rate, commuting costs and a childcare bill worth a third of household income is not making a cultural choice when they leave paid work. They are answering an arithmetic problem correctly. The ILO finds that family responsibilities account for up to 80% of the gender employment gap in high-income countries and 62% in low-income countries [4].

The effect persists long after the children are grown, which is the part most often missed. Care for older relatives falls on people at the peak of their earnings trajectory, and it is more likely to end a career than to interrupt it. In the United States, 59 million people provide unpaid care at an average of 27 hours a week, with 57% performing complex medical tasks alongside daily assistance [1]. ✓ Established A worker supplying 27 hours of care a week is supplying a second job. The labour market records the resulting reduction in hours as a preference for part-time work.

The distributional pattern inside the withdrawal is sharper than the aggregate. Swedish research on a system widely treated as the benchmark finds that tax-funded eldercare coverage has fallen substantially since 1980 and that family care rose to fill the gap, first among daughters with less education and later across social groups [27]. ◈ Strong Evidence Working-class daughters absorbed the largest share, and they are the group least able to buy a substitute. A universal system that contracts does not return to a neutral baseline. It returns to a familial one, and the family that absorbs it is not selected at random.

The employer-side cost is real and poorly measured, which is itself part of the pattern. Unpaid caregiving in the United States is equivalent to about 24 million full-time workers, some 17% of the country's full-time workforce [1]. ✓ Established Those hours are supplied disproportionately by people who also hold jobs, and they are supplied at the expense of overtime, promotion, training and, at the margin, of the job itself. Firms observe the consequence as absence, reduced availability and attrition among experienced mid-career staff. They rarely observe the cause, because no payroll system has a field for it.

There is a second-order effect that fiscal projections handle badly. A person who leaves paid work at 52 to care for a parent does not merely forgo current earnings. They forgo pension accrual, employer contributions and the compounding that makes late-career earnings disproportionately valuable, and they raise their own probability of needing publicly funded care later. The saving to the long-term care budget in year one is real. So is the cost to the pension and social-assistance budgets in year twenty. Only one of the two appears in the assessment that authorises the policy.

Japan is furthest along the curve and its numbers are therefore the most informative. The Ministry of Health, Labour and Welfare projects a requirement of about 2.40 million care workers by fiscal 2026 and about 2.72 million by fiscal 2040, an increase of roughly 570,000 on fiscal 2022 [12]. ✓ Established That is some 32,000 additional workers a year for eighteen years, drawn from a shrinking working-age population. Japan has operated a universal long-term care insurance scheme since 2000, so this is not a financing gap in the ordinary sense [37]. The entitlement exists and is funded. What is missing is the labour to discharge it.

Germany's projection is unusual in publishing its own uncertainty. The Federal Statistical Office expects demand for employed nursing staff to rise by a third, from 1.62 million in 2019 to 2.15 million in 2049 [11]. Under its trend variant, supply reaches 1.87 million and the shortfall is 90,000 by 2034 and 280,000 by 2049. Under its status quo variant, supply falls to 1.46 million and the shortfall is 350,000 by 2034 and 690,000 by 2049 [11]. ✓ Established The distance between the two scenarios exceeds the entire current care workforce of several European countries, and it is determined almost entirely by whether people keep entering the occupation.

England shows what the constraint looks like when it binds in real time. The adult social care vacancy rate fell to 6.2%, about 96,000 posts, in 2025-26, the lowest since 2015-16 [13]. That improvement was not produced by domestic recruitment. Posts filled by British nationals fell by 40,000 in a single year and by 130,000 since 2020-21, while international recruits held the total up [13]. Skills for Care estimates that England needs 410,000 additional posts by 2040 [13]. A vacancy statistic that improves while domestic labour supply contracts is a number waiting to reverse.

The OECD data set the scale of the mismatch. There are on average 5.0 long-term care workers for every 100 people aged 65 and over across the OECD, and 4.1 in the United States [7]. Every projection of demand assumes that ratio holds. Every honest projection of supply shows it falling, because the denominator grows faster than any plausible rate of recruitment. Public spending on long-term care averaged 1.8% of GDP across the OECD, ranging from 4.4% in the Netherlands and above 3% in Norway, Sweden and Denmark down to 1.3% in the United States [8]. ✓ Established

The United States has chosen to manage the mismatch by rationing rather than by price. Some 607,000 people sat on Medicaid waiting lists for home and community based services in 2025, across 41 states operating such lists [15]. Twenty-nine states reported an increase in the numbers waiting and 12 reported a decrease [15]. ✓ Established A waiting list is itself a form of unpaid care policy. The person waiting is not going without care. They are being cared for by a relative who is not counted, not paid and, in most states, not asked.

The technology answer is worth taking seriously, and the best evidence points the other way from the assumption behind it. A study of Japanese nursing homes using institution-level data found that adopting care robots was associated with a 28% increase in care workers and a 39% increase in nurses, together with a fall in retention problems [19]. ◈ Strong Evidence The robots did not displace labour. They made the job survivable enough that facilities could hold onto staff and expand, largely through flexible contracts. That is a valuable finding, and it is not the finding a workforce-substitution strategy requires.

The migration structure of the care economy is not incidental to it. Foreign-born workers are 26% of the long-term care workforce across the OECD, against 20% of all workers, and the concentration runs far higher in individual systems [7]. In the United States, immigrants are more than 30% of home care aides and more than 20% of nursing assistants, with roughly 820,000 immigrant workers across long-term care settings [34]. ◈ Strong Evidence This is not a temporary staffing expedient. It is the operating model, built over two decades, and it is the reason the sector's wage structure has held.

The sending side is where the arithmetic becomes uncomfortable. The ILO counts 75.6 million domestic workers worldwide, three quarters of them women, 81% in informal employment and 11 million working outside their country of origin [17]. ✓ Established The countries that supply care workers to ageing high-income economies are, with few exceptions, countries with worse care deficits and thinner health systems of their own. The WHO safeguards list of 2023 names 55 countries where active international recruitment is discouraged, on the criteria of a universal health coverage index below 50 and health worker density below the global median of 48.6 per 10,000 [25].

Japan built the most explicit version of the arrangement. Bilateral economic partnership agreements have brought nurse and certified care worker candidates from Indonesia, the Philippines and Vietnam since 2008, and Filipinos were the largest single nationality among foreign health and welfare workers in Japan, 14,704 of 57,788, as of October 2021 [35]. ✓ Established Successive residence statuses widened the route. The design is coherent and the training is real. Neither changes the accounting: a care deficit is closed by importing the solution from systems further from meeting their own needs.

Britain ran the entire cycle in five years and produced something close to a natural experiment. A dedicated visa route opened the sector to international recruitment, dependants were barred in March 2024, and new overseas recruitment into care worker roles closed altogether on 22 July 2025, with in-country transitional arrangements running to 2028 [33]. The effect was immediate. Recent migrants joining the workforce fell from 105,000 in 2023-24 to 44,000 in 2024-25 [36], and new international recruits fell to 30,000 in 2025-26, the lowest in four years [13]. Domestic recruitment did not replace them.

Korea's pilot shows the limit of the model when the receiving country will not pay the price. Seoul placed 86 Filipino caregivers with 143 households at 16,800 won an hour, about 1.46 million won a month for a part-time schedule, and the government subsequently said expansion would be difficult unless the cost problem were resolved [26]. The Bank of Korea had proposed exempting foreign care workers from the minimum wage [26]. ⚖ Contested That proposal states the logic of the whole arrangement with unusual candour: the service is affordable at the price the buyer wishes to pay only if the wage floor does not apply to the seller.

The United States is now testing the reverse case. Policy changes through 2025 and 2026 removed the protected-areas rule that had limited immigration enforcement in health care settings and withdrew status from workers who had been employed lawfully under temporary programmes [34]. In a sector where immigrants are close to a third of the home care workforce, that is not a marginal adjustment [34]. ◈ Strong Evidence It removes the input on which the sector's cost structure depends, without the increase in wages that replacing it domestically would require. The adjustment will be made by families.

Four families of instrument are in use and they are not interchangeable. Public insurance creates an entitlement and a financing stream, as in Japan since 2000 [37]. Direct service provision builds capacity, as in the Nordic systems that spend above 3% of GDP on long-term care [8]. Cash-for-care transfers give households money and leave the purchasing decision to them. Labour-market instruments, meaning wage floors, career structures, training subsidies and migration routes, act on supply. Each has an evidence base of different quality, and the quality of the evidence is close to inversely related to how often the instrument is proposed.

The strongest evidence supports direct service provision, and it is the least discussed. Countries that provide care as a public service rather than subsidising its purchase show both higher female employment and lower unmet need. The OECD figures make the association visible: the Netherlands at 4.4% of GDP, Norway, Sweden and Denmark above 3%, against an OECD average of 1.8% and a United States figure of 1.3% [8]. ✓ Established The causal claim needs care, and the Swedish evidence shows that provision can be withdrawn as readily as it was extended [27]. But no country has achieved low unmet need without paying for capacity.

Cash-for-care is the instrument most often chosen and the one with the weakest record against its stated objective. It is cheap to legislate, popular with recipients and administratively light. It also tends to be spent on the household's own labour, which converts a service entitlement into a modest income supplement for a relative who has already left paid work. The Bruegel assessment of the European position identifies heavy reliance on informal caregiving, alongside inadequate public investment and workforce shortages, as a root cause of the widening care gap rather than a mitigation of it [29]. ◈ Strong Evidence The transfer is defensible on distributional grounds. It should not be scored as workforce policy.

The investment case has been modelled carefully and its numbers should be read with their conditions attached. The ILO estimates that closing the gaps in care leave, childcare and long-term care would generate up to 299 million jobs by 2035 and return $3.76 in GDP for every dollar invested, with most of the jobs going to women [4]. ◈ Strong Evidence The same exercise puts the required annual investment at $5.4 trillion, about 4.2% of global GDP, partly offset by tax receipts on the additional employment [4]. UN Women adds that care investment creates two to three times as many jobs as an equivalent sum spent on construction [3]. These are model outputs rather than observed returns.

The Quebec evidence deserves to be stated at full strength rather than filed as a caveat. Baker, Gruber and Milligan found that the province's universal childcare programme was associated with worse non-cognitive outcomes in early childhood that persisted into adolescence and young adulthood, with worse self-reported health, lower life satisfaction and higher crime rates, concentrated among boys and among children who already had elevated behavioural problems [23]. ⚖ Contested They found no consistent gain in test scores. The most plausible reading concerns quality: children moved from adequate home care into low-quality group care do worse. That is a finding about programme design and staffing, and it is what the wage and turnover evidence would predict.

The regulatory response has moved faster than the fiscal one, and it is not nothing. The European Union adopted a care strategy in September 2022, with Council Recommendations on childcare targets and on access to affordable long-term care in December of the same year [24]. The International Labour Conference adopted the first tripartite resolution on decent work and the care economy on 14 June 2024, followed by a plan of action running to 2030 [6] [5]. ✓ Established These instruments establish standards and reporting obligations. They do not appropriate money and they do not train anyone. Their value lies in creating the categories in which the sector can later be argued about.

The central finding of this file is that the care economy is not undercounted by accident and will not be made visible by argument. It is being made visible by arithmetic. The people who supplied unpaid care in the twentieth century did so under conditions that no longer hold in any high-income country and are ceasing to hold in most middle-income ones: low female labour-force participation, large families, geographic stability and a shorter old age. The 16.4 billion daily hours were never free [2]. They were paid for out of the foregone earnings, pensions and health of the people who supplied them, in a currency the accounts do not record.

The second finding is that the wage is the mechanism rather than a symptom. Care work pays below occupations of comparable difficulty because it is care work, and the penalty falls on the men and women in those occupations alike [21]. ◈ Strong Evidence That single fact explains the turnover, the turnover explains the quality, the quality explains the reluctance to fund expansion, and the reluctance to fund expansion returns the work to households. Every part of that loop is measurable. Every part of it is currently managed as though it were a recruitment problem.

The third finding is that the international transfer is running out of room. Foreign-born workers are 26% of the OECD long-term care workforce and close to a third of some national ones [7] [34]. ✓ Established The countries supplying them are, on the WHO's own criteria, countries whose health systems cannot afford the loss [25]. Britain's closure of its recruitment route in July 2025 was the first large test of what happens when a receiving country stops [33], and the answer so far is that domestic supply does not replace it. Posts filled by British nationals fell by 40,000 in a year [13].

The instruments with the best evidence are the least fashionable. Direct public provision reduces unmet need where it is funded, and no country has achieved low unmet need without paying for capacity [8]. Raising the care wage is expensive, slow and effective, because turnover is the binding constraint on quality and turnover responds to pay [10]. Both are unattractive to finance ministries for the same reason. They cost money in the current period and deliver returns that appear in other departments' accounts, over horizons longer than an electoral cycle.

The instruments with the weakest evidence are the most widely adopted. Cash transfers to households substitute for services and are absorbed by family labour that was already being supplied [29]. Technology substitution has been tested most seriously in Japan, where care robot adoption was associated with more care workers and more nurses rather than fewer [19]. ◈ Strong Evidence Recruitment abroad works until either the sending system or the receiving politics gives way, and both have now given way in at least one large market [33] [26]. None of these is useless. None is a substitute for capacity.

The last observation is the one the data support most strongly and the discourse accommodates least. There is no version of the next twenty years in which the unpaid share of care holds at its current level. It will fall, because the people who supply it are in paid work, are fewer in number, and are themselves ageing. Whatever it stops supplying will be supplied by a public system, by a market, or not at all, and the third option is not hypothetical. It is a waiting list with 607,000 people on it [15]. The reports that matter will not be about whether the transfer happened. They will be about who was asked to pay for it, and whether anyone was required to write it down first.