Increasing Returns and Path Dependence in the Economy

Brian Arthur collects his papers from the 1980s and early 1990s arguing that positive feedback governs important parts of the economy and that standard equilibrium analysis, built on diminishing returns, cannot describe those parts. Where returns increase with adoption, small early events determine which of several possible outcomes the system settles into, and the outcome that wins need not be the most efficient.

Arthur's formal contribution is the application of non-linear Polya urn processes to technology adoption. He shows that when each adoption raises the probability of further adoptions, the process converges to one of multiple possible limits, and which limit it reaches depends on the random early sequence. The technology examples are QWERTY, the internal combustion engine over steam, VHS over Betamax, and nuclear reactor designs. He also applies the framework to industrial location, arguing that agglomeration produces regional concentration that historical accident selects.

The economic content became the theoretical basis for how technology investors talk about network effects and lock-in, and the book is the primary source for that vocabulary. The empirical claims have been contested: Liebowitz and Margolis published detailed rebuttals arguing that QWERTY was not inferior to Dvorak and that VHS won on merit, and the general question of whether inefficient lock-in actually occurs in practice remains open. The papers are technical in places and repetitive as a collection, since they were written separately.