An economic historian looks at the computer productivity paradox — and finds it had a near-identical rehearsal a century earlier, when electricity was the new general-purpose engine.
“We see the computers everywhere but in the productivity statistics.”
01 The paradox
By the late 1980s, economists raised to read total-factor-productivity growth as a thermometer for “technical progress” faced something strange: the residual had gone flat at the exact moment microelectronics, fiber optics and the microprocessor were transforming the toolkit. Firms were pouring money into data-processing gear; the statistics showed almost nothing.
David’s move is to refuse the word anomaly. Read through the lens of the economic history of large technical systems — and on the right time-scale — the paradox stops looking unprecedented. It has a mirror, just under a century back.
“We see the dynamos everywhere but in the productivity statistics.”
A contemporary observer of the “Electrical Age” could have said exactly the same thing. Two decades after Edison’s first power stations, electricity barely registered in the aggregate numbers.
“We see the computers everywhere but in the productivity statistics.”
Solow’s quip. David’s claim: it is the same sentence, one general-purpose engine later. The shape of the disappointment is a recurring feature of how these technologies arrive.
02 The general-purpose engine
Both are general-purpose engines: key functional components, embodied in hardware, that become modular building blocks for an enormous variety of specific processes. When such a technology matures, you find it everywhere inside the system. Getting there is slow, and the reasons are structural — not accidental.
The clocks, aligned at year zero
David’s sharpest line: in 1990 we stood the same distance from the computer’s breakthroughs (Intel’s memory chip, 1969; the microprocessor, 1970) as a 1900 observer stood from electricity’s (Edison & Swan’s lamp, 1879; the first central stations, 1881). Set both clocks to zero and the parallel is exact — we were still early.
03 The long, slow S-curve
In 1899 electric lighting reached just 3% of US homes; electric motors were under 5% of factory mechanical drive. It took roughly two decades to cross 50% — and the impact on manufacturing productivity didn’t really land until the early 1920s. Drag along the curve to feel how late the turn comes.
Drag · tap · or focus the chart and use ← →
04 The signature mechanism · from shafts to wires
Swapping a steam engine for an electric one changed little. The real gains came only when engineers stopped bolting motors onto old line-shaft systems (“group drive”) and rebuilt the factory around a motor on every machine (“unit drive”). That redesign is the hidden work behind the lag. Drag the slider to rebuild the factory.
No overhead shafting to brace for → far cheaper construction.
No need to stack floors to shorten shafts → sprawling one-level plants.
Machines placed for material flow, reconfigured at will.
Service one section without shutting the whole mill down.
05 Why forty years?
The delay wasn’t blindness — engineers saw the potential clearly by 1900. It was a stack of rational, structural frictions that no single firm could shortcut.
It made no sense to scrap serviceable steam- and water-powered plants. Best-practice electric factories appeared first where demand was booming — tobacco, fabricated metals, transport equipment — and elsewhere only as old buildings physically wore out.
Through the 1900s–1910s the “group drive” retrofit added motors on top of existing shafting. The old belts and engines stayed in place as idle capacity, raising the capital–output ratio — which actively suppressed measured productivity, especially when energy and quality gains went uncounted.
The big savings — light buildings, single storeys, flexible layouts, modular maintenance — only arrived with ground-up “unit drive” factories. That meant new buildings, not new motors, on a schedule set by depreciation and the building cycle.
Implementing this everywhere required a cadre of factory architects and electrical engineers who’d worked it out in practice, site by site — a decentralised, inherently slow learning process tied to the volume of new construction.
The factory-building industry was fragmented, with high turnover of firms and skilled staff. When knowledge leaks freely among competitive suppliers, no one captures the returns to learning — so adoption runs slower than is socially optimal. (Cheaper utility rates after 1914–17 finally tipped the economics.)
06 What the statistics missed
Beyond the lag, conventional productivity measures systematically undercount an emergent general-purpose technology: the quality of brand-new goods, and whole categories of value that the national accounts never recorded. Tap each to see what slipped through — and its modern echo.
07 The payoff, at last
Once cheap power, depreciation and the building boom converged, the redesigned factory spread — and the residual finally moved. Across industries, the rise in (energy-adjusted) TFP growth in the 1920s tracked how much secondary electric-motor capacity each had installed.
08 But computers are not dynamos
David ends with caution against taking the parallel too literally. The disanalogies don’t make the computer’s payoff easier to see; if anything they suggest its frictions run deeper than the dynamo’s did.
Designing efficient interfaces between people and computers is far more subtle and complex than wiring up lighting and power ever was.
It lacks super-additivity and has near-zero marginal cost of transfer — which makes measuring its production and allocation, and relying on ordinary markets, genuinely problematic.
Cheap distribution encourages indiscriminate broadcasting; screening the flood is costly. Resources spent coping with overload can displace activities the accounts actually record.
A firm’s data and processes are sunk costs that don’t physically depreciate — so time alone never forces the redesign. Expect a strong inertial drag on information-intensive reorganisation.
09 The not-too-distant mirror
A general-purpose engine pays off only after a long, costly reorganisation — and the statistics are the last to know.
Judge a general-purpose technology over decades, not quarters. The dynamo’s residual was flat for ~40 years before it surged.
Value comes from redesigning organisations and capital around the engine — not from the engine itself. That work is slow, decentralised, and easy to underrate.
History guards against naive impatience (“it’s failed”) and naive hype (“it’ll be instant”). Slow statistics are compatible with a real revolution underway.