Skill-Biased Technological Change (SBTC)

Skill-biased technological change (SBTC) is technological progress that raises the productivity of skilled workers, and so the demand for them, more than it raises that of less-skilled workers. When it happens faster than the supply of skilled workers grows, the wage premium for skills, often measured as the pay gap between university graduates and people with only secondary schooling, widens.

The hypothesis became the leading explanation, in the 1990s, for rising wage inequality in the United States and other rich economies after 1980, a period when computers spread through offices and factories. It sits within labor economics and growth theory, alongside human capital theory, and competes or combines with other explanations such as globalization, the decline of trade unions and the falling real value of the minimum wage.

Picture an accounting office before and after spreadsheets. Before, armies of clerks added up columns by hand. After, a few analysts with software did the same work in minutes and spent the rest of their time on more complex tasks. The analysts became far more valuable; the clerks' jobs shrank or disappeared.

The new tool did not help everyone equally. It made people who could use it and interpret its results more productive, and it replaced people whose work it could copy. If schools and universities turn out more trained people fast enough, their pay advantage stays stable. If not, the gap between skilled and less-skilled pay grows.

Lawrence Katz and Kevin Murphy (1992) set out a simple model relating the skill premium to relative demand and supply:

• $w_S / w_U$ is the ratio of skilled to unskilled wages, the skill premium.
• $N_S / N_U$ is the relative supply of skilled to unskilled workers.
• $D$ is an index of relative demand for skilled labor, which SBTC pushes up over time.
• $\sigma$ is the elasticity of substitution between the two groups; Katz and Murphy estimated it at about 1.4.

Worked example: suppose $\sigma = 1.5$, relative demand grows by 3% a year and relative supply by 2%. The log skill premium then rises by $(0.03 - 0.02)/1.5 \approx 0.0067$ a year, or about 0.7%. If supply growth slows to 1%, the premium rises about 1.3% a year. This is the logic behind Claudia Goldin and Lawrence Katz's 2008 book The Race between Education and Technology, which argued that US inequality rose after 1980 mainly because growth in educational attainment slowed while technology kept raising demand for skills. Goldin received the 2023 Nobel Memorial Prize in Economic Sciences for her research on women's labor-market outcomes.

Studies in the 1990s, including work by Eli Berman, John Bound and Zvi Griliches, found that the shift towards skilled workers happened within industries, not mainly through shifts between them, which pointed to technology rather than trade. Industries that invested more in computers saw faster skill upgrading.

Technological change has not always favored the skilled. Nineteenth-century factory mass production and the division of labor replaced skilled artisans with less-skilled machine operators, so some historians describe that era as "deskilling". Whether technology complements or replaces a particular skill depends on the technology.

From the 2000s, data showed employment growing at both the top and bottom of the pay ladder while middle-paying jobs shrank, a pattern called job polarization. David Autor, Frank Levy and Richard Murnane (2003) proposed a task-based refinement, sometimes called routine-biased technological change: computers substitute for routine tasks that follow explicit rules, such as bookkeeping and assembly-line work, and complement non-routine analytical work, while having little effect on non-routine manual jobs such as cleaning and care. Later work by Daron Acemoglu and Pascual Restrepo models automation as replacing workers in specific tasks, with new tasks created by innovation partly offsetting the loss.

Generative artificial intelligence has reopened the debate. Some early studies, such as a study of customer-support agents by Erik Brynjolfsson and co-authors, found the largest productivity gains among less-experienced workers, which would be the opposite of classic SBTC. Evidence on economy-wide effects remained limited as of 2026.

• Timing. David Card and John DiNardo argued in 2002 that US inequality rose mostly in the early 1980s, while computer adoption continued strongly through the 1990s, a mismatch SBTC struggles to explain.
• Institutions. Many economists give more weight to the fall in unionization, a lower real minimum wage and changes in employer power, noting that countries with similar technology saw very different trends in economic inequality.
• Trade. Import competition and offshoring, examined in work on comparative advantage and trade shocks, hit less-skilled manufacturing workers directly.
• Top incomes. SBTC explains the gap between graduates and non-graduates better than the rise in the very highest incomes, which many link to finance, executive pay and winner-takes-all markets.

Policy responses usually focus on expanding education and retraining to raise the supply of skills, with some economists also favoring industrial policy or tax-and-transfer measures. The idea also bears on the lump of labor fallacy: technology changes which jobs exist more than the total amount of work, and the broader study of technological progress and total factor productivity.

• National Bureau of Economic Research (NBER)
• The Nobel Prize: Claudia Goldin, facts