Path dependence

Working Definition

Path dependence is the property of a dynamic process in which the long-run outcome depends not only on current fundamentals and preferences, but on the specific sequence of earlier historical events, including small or contingent ones. In its technical economic formulation, pioneered by Paul A. David and W. Brian Arthur, it describes a stochastic process governed by increasing returns (self-reinforcement): each adoption of a given option lowers the cost or raises the payoff of subsequent adoptions until the system tips into lock-in on a single trajectory, even if an alternative path would have proved superior. In colloquial usage, the phrase degenerates into the truism that “history matters,” explaining nothing because it fits everything. A disciplined definition requires specifying both the mechanism of self-reinforcement (scale economies, learning effects, network externalities, complementary investments) and the threshold past which reversing course becomes costlier than staying on the path.

Genesis and Global Lineage

The concept emerged in economics in the mid-1980s from two parallel breakthroughs. In his six-page paper Clio and the Economics of QWERTY (1985), the economic historian Paul A. David used the 1870s typewriter keyboard layout—designed in part to keep mechanical typebars from jamming—to show how technical interrelatedness between hardware and touch-typists’ muscle memory, economies of scale, and quasi-irreversibility of investment can lock an industry into a standard long after the original engineering constraint has disappeared. Simultaneously, W. Brian Arthur, a mathematician and operations researcher working at IIASA, Stanford, and the Santa Fe Institute, built formal models of competing technologies under increasing returns using nonlinear Polya urn processes; his canonical article appeared in the Economic Journal in 1989 and his collected papers in Increasing Returns and Path Dependence in the Economy (1994). Both challenged neoclassical economics’ reliance on diminishing returns, which guaranteed a unique, history-independent equilibrium. In the 1990s and 2000s, Douglass North in institutional economics and Paul Pierson (Politics in Time, 2004) in political science extended increasing-returns path dependence to laws, welfare states, and party systems, pairing “critical junctures” with long stretches of institutional reproduction.

Central European Lineage

Central Europe’s contribution to the literature lies not in the biography of the concept’s coiners, but in providing two of its richest empirical laboratories. The first is the post-1989 transition from state socialism: in a classic 1992 study, the sociologist David Stark demonstrated that privatization strategies in Poland, Hungary, and Czechoslovakia did not implement “designer capitalism” on a tabula rasa, but followed distinct path-dependent trajectories shaped by how enterprises, worker councils, and ministries had bargained under late socialism. The second is the natural experiment of the Partitions of Poland (1772–1918), which divided Polish lands for over a century among the Russian, Prussian, and Austro-Hungarian empires before reunification in 1918. Using spatial regression discontinuity right across the vanished imperial borders, Irena Grosfeld and Ekaterina Zhuravskaya (2015) demonstrated a crucial refinement to naive theories of historical lock-in: differences in income, industrial output, education, and corruption across the former partition borders have completely disappeared under a century of unified state policy and market integration, whereas differences in church attendance, support for democratic parties, and physical railway density remain sharply visible along the old frontier lines. Path dependence is not a blanket curse of history; it endures only where a concrete transmission mechanism—sunk physical capital that is too costly to tear up, or intergenerational family socialization—keeps reproducing the initial divergence. Related legal scholarship (cited in the English Wikipedia entry on path dependence) has similarly traced how Austrian, Prussian, and Russian administrative-law traditions continue to differentiate judicial reasoning across Polish regional administrative courts.

Why Now

Between 2018 and 2026, path dependence returned to the center of three major policy battles. First, digital platform antitrust: regulatory actions by the European Commission, the EU Digital Markets Act (in force since 2023), and the U.S. Department of Justice’s landmark August 2024 federal court victory declaring Google an illegal monopoly in search all hinge on how default settings, data feedback loops, and switching costs create durable lock-in even at zero monetary price to consumers. Second, climate and energy infrastructure: the concept of carbon lock-in took on acute geopolitical urgency after 2022, as European states scrambling to replace Russian pipeline gas debated whether building permanent liquefied natural gas (LNG) terminals and gas-fired power plants as a “bridge fuel” would lock in fossil-fuel infrastructure and stranded assets for another forty years. Third, artificial intelligence architecture: the concentration of trillions of dollars of hardware fabrication, CUDA software libraries, and training recipes around the transformer architecture has prompted computer scientists (reviving Sara Hooker’s diagnosis of the “hardware lottery”) to ask whether the AI industry has locked itself into a narrow local optimum simply because graphics chips happened to favor matrix multiplication.

Explanatory Power

Path dependence severs the lazy inference from survival to optimality: the fact that an arrangement dominates the market does not prove that it won a fair engineering competition. Consider three examples. First, railway track gauges: nineteenth-century workshop choices remain frozen in steel because the cost of converting rolling stock and tunnels scales with the installed base. Second, legacy enterprise software: core banking systems and tax registries still run on 1960s COBOL mainframes not because COBOL is the best language, but because decades of regulatory patches layered atop the original code make a clean-slate rewrite riskier than another year of maintenance. Third, urban transport: post-war cities built around highway interchanges and low-density zoning cannot simply add a bus line and expect high ridership, because housing, retail, and tax bases have co-evolved around the car. In every case, path dependence reframes the policy question from “what is the ideal design?” to “how high is the switching barrier, and who will pay to lower it?”

Axes of Debate

Underappreciated Angle

Discussions of path dependence almost always focus on the material cost of switching hardware or institutions, while overlooking how lock-in systematically destroys epistemic knowledge of alternatives. When an industry or society settles onto a single path, what disappears is not just the rival product, but the engineers who knew how to build it, the empirical data on how it performed at scale, and the vocabulary needed to evaluate it fairly. After thirty years of cumulative investment in Path A, Path A is genuinely more efficient today than an unimproved 1990 prototype of Path B—not because Path A had a higher ceiling, but because Path A received thirty years of learning-by-doing while Path B was starved of R&D. Once a field converges entirely on a single paradigm, comparing it against the road not taken becomes empirically impossible, which is why maintaining live technological alternatives is a rational investment in knowledge rather than wasteful duplication.