The Productivity Paradox: What Changed, and How Slowly

"Revolution" suggests something abrupt. The classic account described British industry accelerating sharply from the 1780s, and that account rested on output indices assembled in the mid-twentieth century.

Those indices were rebuilt from the 1980s onward, principally by Nicholas Crafts and Knick Harley, and the picture changed substantially. Harley showed that the earlier work had over-weighted cotton and iron, the two fastest-growing industries, in a dataset that covered only about half of industrial output. Weight the whole sector properly and the acceleration flattens.

The revised figures put per-capita growth over the core decades in the region of a third to under one percent per year, a rate that would today be described as sluggish.

Key idea: The revision did not find that the Industrial Revolution was unimportant. It found that its importance is not visible as a growth rate at the time, which is a different and more interesting claim.

The reconciliation is that a small number of industries changed almost beyond recognition while the rest of the economy did not change much at all, and an aggregate is a weighted average of both.

Cotton output rose by orders of magnitude. Iron was transformed by coke smelting. Coal output climbed steadily. Meanwhile agriculture, construction, domestic service, transport before the railways, and most retail and craft trades carried on with broadly the pre-industrial technology and organisation.

Since the transformed sectors were a small share of the economy at the start, spectacular growth in them moves the national average very little. Only when they had grown large enough to matter did their growth start to dominate, and that took decades.

Gotcha: This is a general property of aggregates, not a quirk of the eighteenth century. A sector growing at 20 percent a year from a 2 percent base adds 0.4 points to the total, which disappears into measurement error. The same arithmetic governs every argument about whether a new technology is showing up in productivity statistics today.

The sharper finding concerns who received the output. Comparing two well-known series over the same six decades:

Output per worker rose 46 percent. The real wage index rose 12 percent. Over sixty years, that second figure is close to stagnation: roughly one fifth of one percent per year, well inside the error bars of the reconstruction.

Robert Allen named this period, roughly 1790 to 1840, Engels' pause, after Friedrich Engels' 1845 account of conditions in Manchester. The naming is deliberate: Engels was describing what the wage series shows, at the moment it was happening.

That interpretation is one reading of the numbers rather than a settled conclusion, and the alternatives are worth knowing: some historians emphasise population growth holding wages down, some the Napoleonic Wars and the grain prices they produced, some the specific weaknesses of the wage series itself.

Key idea: The measured facts and their explanation sit at different confidence levels. That output per worker outran wages for roughly half a century is about as well established as anything in the field. Why it did is argued.

Steam is the clearest case of a technology whose measured effect lags its invention by a very long way, and there is a well-studied parallel that explains the mechanism.

Paul David's 1990 paper The Dynamo and the Computer examined electrification. The lightbulb was patented in 1880, and at the turn of the century electric motors still accounted for less than 5 percent of factory mechanical drive. Productivity gains took around forty years to appear.

The reason is precisely the line shaft from the previous lesson. Early electrified factories replaced the steam engine with a dynamo and kept the shafts and belts, which captured almost none of the benefit. The gain required rebuilding around unit drive, one motor per machine, which allowed machines to be arranged by workflow rather than by proximity to a shaft. That happened in the 1920s, and it needed new buildings.

Key idea: A general-purpose technology pays off only after the organisation around it is rebuilt, and the old organisation is usually still working and already paid for. The delay is not slow adoption of the device; it is the cost of everything that has to change around it.

Growth rates and wages are contested. Several other measures are not, and they are the ones that carry the claim that something genuinely different happened.

The last row is the one that matters most and is easiest to miss. Before this period, output per head rose and fell around a rough ceiling: good times raised population until living standards fell back. Afterwards, output per head simply kept climbing.

Key idea: The break was not the rate of growth but the fact that growth stopped reverting. A slow rate that persists for two centuries compounds into a change of kind, and that is what the word revolution should be taken to name.

Real wages measure purchasing power, and they are only one dimension of how people lived. Other evidence from the same decades points in different directions, which is why the standard-of-living debate has run for so long.

• Height. Anthropometric studies of army and prison records find average heights stagnating or falling for cohorts born during the early industrial decades, which is generally read as a signal of nutrition and disease burden in childhood.
• Urban mortality. Cities were considerably more lethal than the countryside before sanitary reform, so moving to higher-paid urban work could raise money wages and lower life expectancy at the same time.
• Hours and autonomy. A real wage index cannot capture the shift from task-paced work to clock-paced work, or the loss of control over when to work at all.
• Consumption. Some goods, notably cotton cloth, tea and sugar, became dramatically cheaper and more widely available, which a general price index may understate.

Gotcha: "Did living standards rise?" is not one question. Purchasing power, health, working conditions and access to goods moved differently and sometimes in opposite directions, and an argument that only cites one of them is answering a narrower question than it appears to.

Every number in this lesson is a reconstruction from incomplete records, and knowing how they were built tells you how far to trust them.

In practice: The price basket is the most consequential of these. A real wage index is a money wage divided by the cost of a bundle of goods, and which goods go in the bundle changes the answer. Choose a basket heavy in bread and the wars raise the cost of living; choose one that captures cheap cotton cloth and it falls. Several long-running disagreements in this literature are disagreements about the basket rather than about the wages.

Putting the four lessons together, the transition has a consistent shape, and it is not the one the word revolution suggests.

1. Adoption followed local economics. Machines were built where the wages they saved exceeded the fuel and capital they consumed, which was a narrow condition met in few places.
2. The technology was poor for a long time. Half a percent efficiency, unreliable power looms, engines that could not turn a shaft. Each was improved by attacking whatever constraint currently bound.
3. The organisation changed as much as the machinery. The factory was a reorganisation of monitoring, payment and time, and only one of its four functions was technical.
4. The measured payoff lagged by decades. Sectoral concentration hid it in the aggregates, and the complementary rebuilding took a generation.
5. The gains were distributed late. Output per worker outran wages for roughly fifty years before the relationship reversed.

Key idea: Every one of these five is a general property of a major technological transition rather than a fact about the eighteenth century. That is the reason this period is still argued about: it is the best-documented instance of a pattern people keep needing to reason about, and it shows that the pattern is slow, uneven, and only obvious afterwards.

A closing set of checks, useful whenever you meet a confident statement about the Industrial Revolution or about technological change generally.

1. Which decades? "The Industrial Revolution" spans at least 1760 to 1840, and claims that hold for the 1830s often fail for the 1780s. A statement without a date range is usually smuggling one in.
2. Which sector? Cotton is not the economy. A great deal of writing generalises from the fastest-changing industry to the whole country.
3. Which measure? Output, wages, height and mortality moved differently. Naming the measure is what makes a claim checkable.
4. Measured or inferred? Output per worker rising faster than wages is measured. Why it happened is inferred, and the inference is disputed.
5. Compared with what? "Conditions were terrible" needs a comparison: terrible against the rural poverty people were leaving, or against what was affordable at the time, are different claims with different answers.

In practice: These five questions dissolve most of the apparent contradictions in popular accounts. Two writers saying opposite things are very often measuring different quantities over different decades in different sectors, and both are reporting their data correctly.