Learning by Doing in Economic Growth

Productivity does not fall from the sky. Learning by doing in economic growth refers to the mechanism by which firms and workers become more efficient simply by producing—each new unit made teaches them something, raising output per worker without requiring new investment in formal education or external technology. Kenneth Arrow’s landmark 1962 model showed how this process alone can sustain perpetual growth, making experience itself a form of capital.

How Production Experience Becomes Productivity

The basic insight is intuitive: repetition works. An aircraft assembly line that has built 500 planes runs faster than one that has built 50. A software team that has shipped 10 releases catches bugs more quickly on the 11th. This is not magic—workers internalize procedures, managers discover bottlenecks, supply chains synchronize.

Arrow formalized this as a learnable, measurable process. He proposed that the growth rate of productivity is proportional to the rate of cumulative gross investment (output). In other words, the more an economy produces in total, the more it learns. Learning happens involuntarily, as a side effect of production itself, so it cannot be hoarded or hidden—knowledge about efficient layouts, material handling, or troubleshooting spills over to competitors and other firms in the supply chain.

This is fundamentally different from neoclassical growth theory, which treats technological progress as an exogenous gift: steady-state growth simply happens at a rate determined by outside factors. Learning by doing makes growth endogenous—it emerges from the economy’s own activity.

The Arrow Model: Productivity Linked to Capital Stock

In Arrow’s setup, productivity grows with the economy’s cumulative capital stock. If firms have invested 10 trillion dollars cumulatively, they know more than if they had invested 1 trillion. The production function captures this: output per worker depends not just on current capital per worker but on the entire history of past capital investment.

Formally, the productivity level (the “knowledge stock”) evolves as:

$$\dot{A} = \lambda K$$

where $A$ is the productivity level, $K$ is the capital stock, and $\lambda$ is a learning parameter. Capital accumulation teaches the economy.

Crucially, this means that economies with larger installed capital bases learn faster. A mature manufacturer producing millions of units annually gains more experience per year than a startup. Countries that have sustained investment for decades have steeper learning curves than those just beginning industrialization. This creates a feedback loop: high growth today → large capital stock → faster learning → high growth tomorrow.

Mechanisms: Where Learning Actually Happens

Learning by doing operates through several channels:

Within-firm efficiency. Management learns to coordinate labor and machines better. Workers develop skill and speed through repetition. Supply chains synchronize. A factory’s first 1,000 units are chaotic; the 100,000th unit is routine.

Spillovers and imitation. Workers move to competitors and teach them what they learned. Suppliers watch successful production techniques and adopt them. Published case studies and industry conferences spread knowledge. This means one firm’s learning benefit is not private for long—it seeps into the broader economy.

Design refinement. Each production run reveals which designs are actually manufacturable at scale. Engineers use production data to improve blueprints. A product that worked in the prototype lab may require dozens of iterations to run smoothly in a factory. Learning by doing captures this product-process coevolution.

Organizational learning. Firms learn which organizational structures work at different scales, which hiring profiles suit which roles, and how to incentivize quality. This is neither obvious nor fast—it emerges from lived experience.

Empirical Evidence: The Learning Curve

The learning curve—also called experience curve—is an empirical regularity: cumulative output and unit cost follow a predictable, often logarithmic relationship. Semiconductor manufacturing, commercial aircraft, and solar panels all show this pattern. Cost per watt for solar modules has fallen roughly 20% for every doubling of cumulative production. This is not because solar cells became cheaper to buy; it is because factories learned to make them more efficiently.

Industry-level data also shows that sectors with rapid output growth tend to have faster productivity gains. Economies that sustained investment during 1950–1990 entered the 21st century with steeper learning curves than latecomer industries. This empirical fact aligns with Arrow’s theory.

However, the evidence also reveals limits. Learning curves do eventually flatten. Costs for a mature technology stop falling once best practices are near-universal. Moreover, learning spillovers are imperfect—a firm’s proprietary processes stay secret, and workers do not freely share all their knowledge. Some sectors have stronger within-firm learning (chemicals, semiconductors) than others (retail, agriculture).

Growth Implications and the Steady State

In the Arrow model, learning by doing alone can sustain perpetual growth without external technological progress. If an economy keeps accumulating capital, productivity keeps rising, so per-capita output can grow indefinitely. This avoids the “steady-state trap” of the Solow model, where growth eventually stops unless exogenous technological progress arrives.

The growth rate depends on how fast capital stock is accumulating and how strong the learning parameter is. A country with a 20% investment rate and strong spillovers will sustain faster growth than one investing 5% with knowledge leaks to rivals. This gives policy—subsidizing investment or reducing knowledge barriers—real long-run bite.

However, learning by doing is not a silver bullet. Many sectors exhibit weak spillovers: a bank’s proprietary trading algorithms, a retailer’s supply chain network, or a pharmaceutical firm’s manufacturing secrets stay contained. If knowledge cannot spill over broadly, learning becomes a private advantage, not a public good, and the economy-wide growth boost is smaller.

Policy Angles: Subsidies, Education, and Openness

If learning by doing drives growth, then policies that accelerate accumulation or deepen spillovers have outsized payoff. This is the case for:

• Young-industry protection. Temporary tariffs or subsidies for infant industries increase their cumulative output, speed up their learning, and allow them to compete at mature cost levels. The mechanism is not that they are naturally viable; it is that early protection lets them climb their learning curve faster.

• Education and skill complementarity. Learning by doing works faster when workers can absorb lessons. Schooling creates a base of adaptability. Sectors with higher education levels show steeper learning curves and stronger spillovers.

• Openness and technology transfer. Free trade and foreign direct investment expose domestic firms to practices used abroad. A joint venture with a global competitor accelerates learning. Restrictions that block contact slow learning.

• Public investment and subsidized R&D. Knowledge spillovers work best when discovery is shared. Public funding of basic research increases the productivity data available to the whole sector.

Conversely, policies that slow accumulation—high capital taxes, financial repression, or trade barriers—slow learning by doing and thus long-run growth.

Limitations and Extensions

Learning by doing is not costless. Producing to learn can be inefficient in the short run: a firm producing slowly to manage quality may learn more per unit than one racing to volume, but it sacrifices scale economies. There is a trade-off between learning speed and current efficiency.

Also, Arrow’s model assumes all learning is productive. In reality, some repetition can entrench bad habits. Manufacturing cultures that have worked one way for decades may be slower to adopt radically new methods. Path dependence can limit learning’s upside.

Modern growth theory has extended learning by doing to endogenous growth models where R&D effort, human capital investment, and production learning all interact. Learning by doing is often one engine among several—not the only source of growth, but a crucial one alongside formal innovation and human capital formation.

See also