Baumol Effect

Baumol Effect and Why Labour-Intensive Health Care Grows Relatively More Expensive

Health care takes a rising share of national income in most high-income countries, and William Baumol offered an explanation that does not depend on ageing or new technology. His 1967 model of unbalanced growth, often called Baumol's cost disease, splits the economy into a progressive sector with cumulative productivity growth and a sector in which labour is itself the product, and asks what happens to their relative costs when a common wage links the two. The model can be shown with an illustrative twenty-year example, and it bears on NHS productivity measurement, health price indices and UK spending projections, as well as on the evidence for and against the effect in health care.

Baumol's 1967 model of unbalanced growth

Baumol set out the model in the American Economic Review in 1967, building on a hypothesis developed with William Bowen in a study of the performing arts. Its one indispensable premise is that some activities are technologically progressive, with innovation, capital accumulation and economies of scale raising output per hour cumulatively, while others permit only sporadic gains. The difference lies in the role of labour, which in manufacturing is an instrument for making the product but in some services is the product, so that quality is judged by the amount of labour supplied.

Baumol assumed that wages in the two sectors move together in the long run, because labour can move between them, and ignored non-labour costs. With a common wage, labour cost per unit of output in health care and in the rest of the economy is as follows.

$$c_{h,t} = \frac{w_t}{a_{h,t}}, \qquad c_{o,t} = \frac{w_t}{a_{o,t}}$$

where $c_{h,t}$ is the labour cost per unit of health care output in year $t$, $c_{o,t}$ is the labour cost per unit of other output, $w_t$ is the hourly wage common to both sectors, and $a_{h,t}$ and $a_{o,t}$ are output per hour in health care and in the rest of the economy. If output per hour grows at $g_h$ a year in health care and $g_o$ elsewhere, the ratio of unit costs evolves as follows.

$$\frac{c_{h,t}}{c_{o,t}} = \frac{c_{h,0}}{c_{o,0}} \left( \frac{1+g_o}{1+g_h} \right)^t$$

where $c_{h,0}/c_{o,0}$ is the cost ratio in the starting year and $t$ is the number of years elapsed. The wage cancels out, so health care becomes relatively more expensive whenever $g_o$ exceeds $g_h$, whatever happens to the general level of pay. In Baumol's own illustration, wages and productivity in the progressive sector both rise by 2 per cent a year, so costs there do not rise at all, while in a sector with constant productivity every wage rise feeds straight into costs.

The second consequence matters more for health systems. If the ratio of the two sectors' outputs is held constant, because demand is price inelastic or income elastic or because government finances the service, labour must keep shifting into the slow sector, as the expression below shows.

$$\frac{L_{h,t}}{L_{o,t}} = \frac{L_{h,0}}{L_{o,0}} \left( \frac{1+g_o}{1+g_h} \right)^t$$

where $L_{h,t}$ and $L_{o,t}$ are the hours of labour employed in health care and in the rest of the economy in year $t$, and the other terms are as defined above. Baumol showed that an ever larger share of the labour force then moves into the non-progressive sector and overall growth slows.

Why health care fits the slow-growth sector

Baumol named hospitals among the city services with very limited scope for cumulative productivity growth, and wrote that despite computers in medicine there is "no substitute for the personal attention of a physician". Consultations and nursing care are delivered by staff in real time, and shorter consultations or fewer nurses per patient are usually read as lower quality rather than higher productivity, much as larger classes are in education.

Baumol also stressed that the split between the sectors is one of degree, not a sharp division. Health care combines progressive elements, such as laboratory automation and medicines that replace inpatient stays, with stagnant ones, so the productivity growth of the sector depends on its mix. The labour-only simplification also fits health care less well than live performance, since medicines, equipment and buildings follow their own price paths.

Worked example of twenty years of unbalanced productivity growth

The figures in this example are illustrative. An economy employs 10 per cent of its working hours in health care and 90 per cent elsewhere. Output per hour grows by 2.0 per cent a year outside health care and 0.5 per cent a year in health care, wages in both sectors rise by 2.0 per cent a year, total hours are unchanged, and the volume of health care is held in a fixed ratio to other output, as a publicly financed service with income-elastic demand might be.

1. Relative cost growth. Each year the unit cost of health care rises relative to other output by a factor of $r = 1.02 / 1.005 = 1.014925$, about 1.49 per cent a year. Over twenty years wages rise by $W = 1.02^{20} = 1.4859$ and health productivity by $A = 1.005^{20} = 1.1049$, so the unit cost of health care rises by $R = 1.4859 / 1.1049 = 1.3449$ while the unit cost of other output is unchanged. Health care becomes about 34.5 per cent more expensive relative to other goods.

2. Labour needed to hold the output mix. The starting ratio of health hours to other hours is $x = 0.10 / 0.90 = 0.1111$, which becomes $y = 0.1111 \times 1.3449 = 0.1494$ after twenty years. Health care's share of total hours is then as follows.

$$s_{h,20} = \frac{0.1494}{1 + 0.1494} = 0.13$$

where $s_{h,20}$ is the share of total working hours employed in health care after twenty years. With equal wages and only labour costs, this is also health care's share of total spending, which rises from 10 per cent to 13 per cent.

3. Real output still grows in both sectors. Other output changes by a factor of $q = (0.87 / 0.90) \times 1.4859 = 1.4364$ and health output by $z = (0.13 / 0.10) \times 1.1049 = 1.4364$, so both are about 43.6 per cent higher than at the start. Both sectors deliver more, even though health care absorbs a larger share of spending and employment. This is the sense in which Baumol argued that rising health costs remain affordable when productivity grows elsewhere.

Measuring NHS productivity and what the statistics can and cannot show

Testing the effect needs a measure of health care productivity, which is hard to build because most NHS care has no market price. As described in its 2021 methods note, the Office for National Statistics (ONS) measures output with a cost-weighted activity index, weighting each activity by its unit cost, and divides the change in output volume by the change in the volume of labour, goods and services, and capital. Following the Atkinson Review, the note lists a quality adjustment based on post-operative survival, estimated health gain from procedures, waiting times, clinical measures from the Quality and Outcomes Framework and patient experience surveys; the ONS has since revised the healthcare quality adjustment under its Public Services Productivity Review.

For some components, such as care bought from non-NHS providers, output is measured by the same deflated spending used for inputs, so the ONS treats them as output equals inputs and they cannot change measured productivity. Nordhaus, testing Baumol's hypotheses on United States industry data for 1948 to 2001, warned that output measures for health and education are in reality measures of inputs. Slow measured growth can therefore reflect output that is hard to count as well as genuinely slow change.

The available figures suggest a gap, though a smaller one than a sector with no productivity growth would show. The Office for Budget Responsibility (OBR), using ONS data, reported quality-adjusted healthcare productivity growth averaging 0.7 per cent a year from 1995/96, against 1.2 per cent a year for the whole economy, although the ONS health figure is a multi-factor measure that the ONS says is not directly comparable with whole-economy labour productivity. The Centre for Health Economics at the University of York estimated NHS productivity growth of 1.11 per cent a year between 2004/05 and 2018/19, and 0.31 per cent a year over 2004/05 to 2022/23, a period that includes the pandemic.

Health price indices and the measured size of the effect

The Baumol effect is a relative price effect, so its estimated size depends on how health prices are measured. An index that prices the same procedure over time ignores gains in what the procedure achieves. Cutler and colleagues showed this for heart attack treatment in the United States between 1983 and 1994; more frequent reweighting and allowing for price discounts lowered measured price growth by three percentage points a year, and once quality change was included the quality-adjusted real cost of treating a heart attack fell by about 1 per cent a year.

Colombier designed his test of the cost disease to avoid medical price indices, whose construction he judged flawed. The issue also reaches routine economic evaluation, since the NICE manual (PMG36, section 4.4.12) states that costs from earlier years should be adjusted using inflation indices appropriate to the cost perspective, such as the NHS cost inflation index and the PSS pay and prices index, or the Office for National Statistics consumer price index. If staff-intensive services become relatively dearer, using a general measure of inflation such as the consumer price index to uprate their costs would understate their current cost.

The Baumol effect in UK health spending projections

The OBR divides real health spending growth into demographic change, an income effect and other cost pressures. In its 2024 Fiscal risks and sustainability report it set other cost pressures at 1 percentage point a year, drawing on OECD evidence, made up of 0.72 points from the Baumol effect, 0.15 from chronic conditions and a small residual (0.13 points) that partly reflects technology. The Baumol component multiplies projected economy-wide productivity growth by an OECD coefficient, as follows.

$$B = \beta \times g_{econ} = 0.48 \times 1.5 = 0.72$$

where $B$ is the Baumol contribution to annual real health spending growth in percentage points, $\beta$ is the Baumol coefficient (the percentage growth in real health spending associated with each 1 per cent of growth in economy-wide labour productivity), and $g_{econ}$ is projected economy-wide productivity growth, averaging 1.5 per cent a year. The coefficient matters. With a value of zero, health spending reached 10.7 per cent of GDP by the end of the 2024 projection, and with a value of 1 it reached 20.1 per cent of GDP in 2073/74, against a baseline of 14.5 per cent.

The July 2026 report kept these assumptions, projecting health spending rising from 8 per cent of GDP in 2030/31 to 13 per cent by 2075/76, and stressed that the future relative productivity of the health sector is very uncertain. Without the extra cost pressures, spending rose only to 9 per cent of GDP, and the OBR noted that artificial intelligence could raise health sector productivity and reduce cost growth.

Econometric evidence for and against the cost disease in health

Empirical tests ask whether health spending grows faster when economy-wide wages outpace productivity. Hartwig, using a panel of 19 OECD countries, found that health care expenditure is driven by wage increases in excess of productivity growth. Bates and Santerre applied this type of test to the 50 US states from 1980 to 2009 and found that health care costs grow more rapidly when economy-wide wage increases exceed productivity gains. Across the whole US economy, Nordhaus found that industries with slow productivity growth had rising relative prices and falling relative real output.

Later work suggests the effect is partial or weaker. Colombier, using 20 OECD countries from 1970 to 2010, estimated that the cost disease exerts between 15 and 40 per cent of its potential full effect on health expenditure, leaving policymakers more room to restrain spending growth than earlier studies implied. Atanda and colleagues tested predictions of a two-sector model, including that health care's employment share and relative price should rise with economy-wide productivity, and found little support in data for 28 OECD countries from 1995 to 2016 and 14 US industry groups from 1947 to 2015 once common trends were handled.

What the Baumol effect does and does not imply

The model is often cited as a reason for alarm, but its message concerns relative prices, not a loss of real resources. Baumol argued in 1967 that such cost rises occur for reasons largely beyond the control of those running the services, so rising unit costs are not on their own evidence of inefficiency. In his 2012 book and an accompanying article he argued that society can afford rising medical bills because productivity growth elsewhere raises incomes, as the worked example shows, and that the belief that such care is unaffordable is the main threat.

The Baumol effect concerns the price of a given volume of care. It is separate from the income effect, through which richer populations demand more care, and from cost growth driven by technologies that change what care can do, although estimates often blend them, and the OBR assumes that most technology-driven growth is captured by a mix of its Baumol and income effects. It also depends on the wage link. Pay that lags the wider labour market suspends the effect for a time but, in Baumol's argument, cannot do so indefinitely, because labour moves between sectors in the long run.

For economic evaluation and health technology assessment, the practical lesson concerns prices. A long-horizon model that holds unit costs at base-year relative prices implicitly assumes that staff-intensive care will not become relatively dearer, and a cost-effectiveness threshold fixed in money terms buys progressively less staff-intensive care if the effect operates.

Sources

• Arabadzhyan A, Castelli A, Gaughan J, Nazerai A, Song W. Productivity of the English National Health Service: 2022/23 update. CHE Research Paper 198. York: Centre for Health Economics, University of York; 2025.
• Atanda A, Menclova AK, Reed WR. Is health care infected by Baumol's cost disease? Test of a new model. Health Economics. 2018;27(5):832-849.
• Bates LJ, Santerre RE. Does the U.S. health care sector suffer from Baumol's cost disease? Evidence from the 50 states. Journal of Health Economics. 2013;32(2):386-391.
• Baumol WJ. Macroeconomics of unbalanced growth: the anatomy of urban crisis. American Economic Review. 1967;57(3):415-426.
• Baumol WJ, with contributions by de Ferranti D, Malach M, Pablos-Méndez A, Tabish H, Wu LG. The Cost Disease: Why Computers Get Cheaper and Health Care Doesn't. New Haven: Yale University Press; 2012.
• Baumol WJ. We can have it all: why health care will still be affordable. Policy Options. 1 November 2012.
• Colombier C. Drivers of health-care expenditure: what role does Baumol's cost disease play? Social Science Quarterly. 2017;98(5):1603-1621.
• Cutler DM, McClellan M, Newhouse JP, Remler D. Are medical prices declining? Evidence from heart attack treatments. Quarterly Journal of Economics. 1998;113(4):991-1024.
• Hartwig J. What drives health care expenditure? Baumol's model of 'unbalanced growth' revisited. Journal of Health Economics. 2008;27(3):603-623.
• National Institute for Health and Care Excellence. NICE technology appraisal and highly specialised technologies guidance: the manual (PMG36). London: NICE; 2022, last updated 13 October 2026. Section 4.4.
• Nordhaus WD. Baumol's diseases: a macroeconomic perspective. NBER Working Paper 12218. Cambridge, MA: National Bureau of Economic Research; 2006.
• Office for Budget Responsibility. Fiscal risks and sustainability (CP 1142). London: OBR; September 2024. Chapter 3.
• Office for Budget Responsibility. Fiscal risks and sustainability. London: OBR; July 2026. Chapter 3.
• Office for National Statistics. Public service productivity estimates: healthcare QMI. Newport: ONS; 2021.
• Office for National Statistics. Improved methods for total public service productivity: total, UK, 2021. Newport: ONS; 2024.