On most reconstructions, a preindustrial European consumed somewhere between 15 and 30 gigajoules of energy a year. Most of it never passed through a market. It was the food that powered human muscle, the fodder eaten by draught animals, and the firewood gathered from hedgerows and commons. By 1900, per capita consumption in Britain had risen several-fold, and its composition had changed almost beyond recognition.
This is a central quantitative puzzle of modern economic history. Energy consumption multiplied within a few generations. Yet in standard growth accounting, energy appears as a minor input, because its cost share rarely exceeds a few percent of national income. There are two possibilities. Either the energy transition was a sideshow to the real drivers of growth, such as institutions, human capital, and technical knowledge. Or our accounting frameworks systematically understate its role.
Resolving that tension requires three things. The first is credible estimates of how much energy past societies used, built from sources never designed to answer the question. The second is a measure of how efficiently that energy was converted into useful work. The third is a framework that links energy inputs to output growth without assuming the answer in advance. What follows examines each in turn, noting where the numbers are robust and where they rest on heroic assumptions.
Reconstructing Energy Use from Fragmentary Sources
Systematic energy statistics are a recent invention. Britain's official coal output figures begin only in 1854, with Robert Hunt's Mineral Statistics, and most European series start later still. For earlier periods, historians build consumption estimates from sources compiled for other purposes. These include coastal shipping records of coal moving from Newcastle to London, municipal coal duties, colliery accounts, and estate records of woodland sales. Each source captures a fraction of the whole, and the skill lies in knowing which fraction.
Fossil fuels were traded and taxed, so they are comparatively easy to measure. Biomass is the harder problem. Firewood was often gathered rather than bought, so its consumption must be inferred indirectly. One route multiplies woodland area by sustainable yield per hectare. Another scales up household fuel allowances recorded in institutional accounts by population. Food energy is estimated from population and plausible caloric requirements, and fodder from livestock counts multiplied by feed needs per animal. Each step adds assumptions, and the errors compound.
Converting these different carriers into a common unit forces methodological choices that materially affect the results. Should human food count as an energy input to the economy, or as consumption? Should sunlight captured by crops be included? In Power to the People, Astrid Kander, Paolo Malanima, and Paul Warde count the energy content of food, fodder, firewood, and modern carriers at the point of primary supply. The convention is defensible, but alternatives shift preindustrial baselines substantially.
Applied consistently, the method reveals a striking divergence. Warde's reconstructions for England and Wales show coal overtaking firewood as the dominant fuel by the seventeenth century, an extraordinarily early transition. Across most of continental Europe, traditional carriers remained dominant well into the nineteenth century, and per capita consumption in places such as Italy stayed broadly flat for centuries. England's per capita energy use, by contrast, was climbing long before the classic dates of industrialisation.
Tony Wrigley offered the most memorable way to grasp the scale. Suppose coal output had to be replaced by woodland managed for a sustainable yield of the same energy. By the mid-nineteenth century, Britain would have needed forests larger than its entire land surface. Biomass estimates may carry errors of thirty or even fifty percent, but sensitivity analysis shows the qualitative conclusion holds. No plausible revision closes a gap of that size. The uncertainty is real, but it does not change the answer.
TakeawayWhen a conclusion survives errors of fifty percent in the underlying estimates, the uncertainty is not the story. The magnitude is.
Multiplying Useful Work: The Efficiency Revolution
Primary energy measures fuel burned. What matters economically is useful energy: the heat, motion, and light actually delivered to an end use. The ratio between the two is conversion efficiency, and it varied enormously across technologies and over time. Roger Fouquet's work on energy services reframes the question in these terms. Instead of tracking tons of coal, he tracks the cost of heating a room, moving a ton of freight one kilometre, or lighting a space for an hour.
The steam engine provides the best-documented series, partly because engine performance was itself a commercial variable. From 1811, Cornish mine owners published monthly engine duty figures in Lean's Engine Reporter, creating one of the earliest systematic efficiency datasets. The long-run trajectory is clear. Newcomen engines converted perhaps one percent of fuel energy into work. Watt's separate condenser roughly tripled that. High-pressure compound engines of the late nineteenth century then cut the coal needed per horsepower-hour by another large multiple.
Lighting shows the effect at its most extreme. Fouquet and Peter Pearson estimate that between 1800 and 2000, the price of light services in Britain fell by a factor in the thousands, as tallow candles gave way to gas, kerosene, and electric light. William Nordhaus reached a related conclusion for the United States. Conventional price indices track the cost of fuel rather than the service delivered, so they overstated inflation in lighting and understated real income growth.
Efficiency gains did not reduce fuel use. In 1865, William Stanley Jevons observed that more efficient steam engines had increased Britain's coal consumption, because cheaper power opened up new applications. The historical series bear him out. Demand for energy services proved highly elastic as their costs fell, so useful energy consumption grew far faster than primary supply. This rebound was not a side effect of the transition. In many sectors, it was the transition.
This has direct consequences for the familiar energy intensity curve. Series that count only modern carriers show an inverted U: energy per unit of GDP rises during industrialisation, then falls. Kander and her colleagues show that once traditional biomass is included, the hump largely disappears in many European cases, and aggregate energy intensity declines over the long run. The apparent rise was partly a substitution artefact. Fossil fuels were replacing unmeasured biomass, rather than economies becoming genuinely more energy hungry.
TakeawayEfficiency gains did not save energy. They made energy services so cheap that consumption exploded, which means the quantity that matters for growth is work done, not fuel burned.
Energy in Growth Accounting: The Cost Share Problem
Standard growth accounting splits output growth into contributions from each input, weighted by its output elasticity. Under competitive assumptions, those elasticities equal factor cost shares. Energy spending typically accounts for a small fraction of national income, often around five percent. The arithmetic follows directly: even large increases in energy input produce modest contributions to growth. Robert Fogel applied the same logic to American railroads. He estimated their social savings at below three percent of GNP and concluded that no single innovation was indispensable.
Nicholas Crafts applied the framework to steam power, with similarly deflating results. His estimates show that steam's contribution to British labour productivity growth was negligible before 1830. It peaked only after 1850, more than a century after Newcomen's first engine. Diffusion was slow, early engines were costly and inefficient, and water power remained competitive for decades. On this evidence, the classic Industrial Revolution was not, in growth accounting terms, a steam-powered event.
Gregory Clark and David Jacks reached a parallel conclusion about coal itself. Examining coal prices and output, they found supply elastic enough that doing without cheap coal would have cost only a modest share of national income. Robert Allen reads similar data differently. In his account, cheap coal combined with high British wages made labour-saving, energy-using techniques profitable. Coal's significance then lies in the direction of technical change it encouraged, not in its weight within an accounting identity.
A more fundamental critique targets the cost share assumption itself. Robert Ayres, Benjamin Warr, and Reiner Kümmel estimate output elasticities for energy far above its cost share, though their econometric methods remain contested. The deeper issue is substitutability. The cost share theorem holds at the margin, under unconstrained conditions. Wrigley's organic economy, however, was defined by a binding land constraint. If energy and other inputs are poor substitutes, energy's share stays small while energy is abundant and rises sharply when it is scarce, as the 1970s oil shocks briefly showed.
The disagreement is therefore partly about which question is being asked. Growth accounting answers a marginal question: what did additional energy contribute, given everything else? The energy transition raises an inframarginal one: what was it worth to remove the ceiling on total energy supply? Kander, Malanima, and Warde propose the concept of development blocks, clusters of complementary technologies built around new energy carriers, to bridge the two. The concept gains interpretive reach at the cost of statistical precision.
TakeawayGrowth accounting measures what a little more energy would have bought, not what the absence of fossil fuels would have cost. How much energy mattered depends on which of those questions you are asking.
The quantitative record settles several questions decisively. Per capita energy consumption in the leading economies rose several-fold. England's shift to coal began centuries before industrialisation. Efficiency gains multiplied useful work far faster than primary supply grew. These findings survive generous allowances for error in biomass estimates.
What remains unsettled is causal weight. Small cost shares imply a modest contribution, while a binding land constraint implies an indispensable one. Reconciling the two depends on how easily energy could substitute for other inputs in organic economies. That parameter has almost never been estimated directly from historical data.
This points to clear research priorities. Comparable biomass reconstructions for China, India, and Japan would sharpen Great Divergence debates. Cross-national useful-energy series would let efficiency enter productivity analysis directly. Episodes of regional fuel scarcity, such as early modern wood price spikes, offer natural experiments for estimating substitutability. Until such work is done, our standard accounting framework will struggle to capture the energy transition, despite its consequences.