This site is a work in progress and has not been widely shared. Content may contain errors. Feedback is welcome.
This site is undergoing review. Some annotations were human-generated, some AI-generated — all are being verified.
Back to papers

Beyond the Race between Education and Technology

Katz

2025NBER Working Paper Series
Theoretical / conceptualTheoretical model
AI (General)Training / upskillingRoutine task changeAugmentation vs. substitutionGeneral automation
DOI: 10.3386/w26705
Summary

Autor, Goldin, and Katz extend the race between education and technology framework using Census/CPS data from the 1800s-2017 to study how changes in educational wage differentials and skill supply growth explain U.S. wage inequality patterns across two centuries.

Main Finding

Educational wage differentials explain 75 percent of U.S. wage inequality increase from 1980-2000 but only 38 percent from 2000-2017; the slowdown in relative supply growth of college workers (from 3.1% to 2.13% per annum) accounts for 62% of the post-1979 surge in the college wage premium, with implied demand shifts explaining only 38%.

Primary Datasets

CPS

Labor outcomes

Census/CPS (19th century-2023)

Secondary Datasets

Occupational wage data from Goldin and Katz (2008); Katz and Margo (2014) clerical/production worker series

Key Methods
CES production function framework with regression analysis of college wage premium on relative supply of college workers, combined with counterfactual simulations using Mincerian earnings regressions
Sample Period
1800s-2023
Geographic Coverage
US
Sample Size
Multiple samples spanning two centuries; CPS MORG samples include all wage and salary workers aged 18-64 with 0-39 years potential experience
Level of Analysis
Occupation, Country
Occupation Classification
None
Industry Classification
None
Notes
Harvard WP; Kansas City Fed Conference; education explains only 33% of inequality change 2000-2023 (vs 75% for 1980-2000) [Claude classification]: This is a theoretical/conceptual paper that extends the canonical RBET framework. The paper mentions AI, robots, and computerization but only in passing as examples of skill-biased technological change; it does not empirically study AI impacts. The framework is calibrated/estimated with historical data. Education explains only 38% of inequality change 2000-2017 vs 75% for 1980-2000. Paper uses CES production function with college/high school workers, estimates σ_SU ≈ 1.62. Finds puzzling slowdown in trend demand growth for college workers post-1992 despite rapid technological change. [Claude classification]: This is a theoretical/conceptual paper that extends the canonical RBET framework. The paper mentions AI, robots, and computerization but only in passing as examples of skill-biased technological change; it does not empirically study AI impacts. The framework is calibrated/estimated with historical data. Education explains only 38% of inequality change 2000-2017 vs 75% for 1980-2000. Paper uses CES production function with college/high school workers, estimates σ_SU ≈ 1.62. Finds puzzling slowdown in trend demand growth for college workers post-1992 despite rapid technological change. [Claude classification]: This is a theoretical/conceptual paper that extends the canonical RBET framework. The paper mentions AI, robots, and computerization but only in passing as examples of skill-biased technological change; it does not empirically study AI impacts. The framework is calibrated/estimated with historical data. Education explains only 38% of inequality change 2000-2017 vs 75% for 1980-2000. Paper uses CES production function with college/high school workers, estimates σ_SU ≈ 1.62. Finds puzzling slowdown in trend demand growth for college workers post-1992 despite rapid technological change. [Claude classification]: This is a theoretical/conceptual paper that extends the canonical RBET framework. The paper mentions AI, robots, and computerization but only in passing as examples of skill-biased technological change; it does not empirically study AI impacts. The framework is calibrated/estimated with historical data. Education explains only 38% of inequality change 2000-2017 vs 75% for 1980-2000. Paper uses CES production function with college/high school workers, estimates σ_SU ≈ 1.62. Finds puzzling slowdown in trend demand growth for college workers post-1992 despite rapid technological change. [Claude classification]: This is a theoretical/conceptual paper that extends the canonical RBET framework. The paper mentions AI, robots, and computerization but only in passing as examples of skill-biased technological change; it does not empirically study AI impacts. The framework is calibrated/estimated with historical data. Education explains only 38% of inequality change 2000-2017 vs 75% for 1980-2000. Paper uses CES production function with college/high school workers, estimates σ_SU ≈ 1.62. Finds puzzling slowdown in trend demand growth for college workers post-1992 despite rapid technological change. [Claude classification]: This is a theoretical/conceptual paper that extends the canonical RBET framework. The paper mentions AI, robots, and computerization but only in passing as examples of skill-biased technological change; it does not empirically study AI impacts. The framework is calibrated/estimated with historical data. Education explains only 38% of inequality change 2000-2017 vs 75% for 1980-2000. Paper uses CES production function with college/high school workers, estimates σ_SU ≈ 1.62. Finds puzzling slowdown in trend demand growth for college workers post-1992 despite rapid technological change. [Claude classification]: This is a theoretical/conceptual paper that extends the canonical RBET framework. The paper mentions AI, robots, and computerization but only in passing as examples of skill-biased technological change; it does not empirically study AI impacts. The framework is calibrated/estimated with historical data. Education explains only 38% of inequality change 2000-2017 vs 75% for 1980-2000. Paper uses CES production function with college/high school workers, estimates σ_SU ≈ 1.62. Finds puzzling slowdown in trend demand growth for college workers post-1992 despite rapid technological change. [Claude classification]: This is a theoretical/conceptual paper that extends the canonical RBET framework. The paper mentions AI, robots, and computerization but only in passing as examples of skill-biased technological change; it does not empirically study AI impacts. The framework is calibrated/estimated with historical data. Education explains only 38% of inequality change 2000-2017 vs 75% for 1980-2000. Paper uses CES production function with college/high school workers, estimates σ_SU ≈ 1.62. Finds puzzling slowdown in trend demand growth for college workers post-1992 despite rapid technological change. [Claude classification]: This is a theoretical/conceptual paper that extends the canonical RBET framework. The paper mentions AI, robots, and computerization but only in passing as examples of skill-biased technological change; it does not empirically study AI impacts. The framework is calibrated/estimated with historical data. Education explains only 38% of inequality change 2000-2017 vs 75% for 1980-2000. Paper uses CES production function with college/high school workers, estimates σ_SU ≈ 1.62. Finds puzzling slowdown in trend demand growth for college workers post-1992 despite rapid technological change. [Claude classification]: This is a theoretical/conceptual paper that extends the canonical RBET framework. The paper mentions AI, robots, and computerization but only in passing as examples of skill-biased technological change; it does not empirically study AI impacts. The framework is calibrated/estimated with historical data. Education explains only 38% of inequality change 2000-2017 vs 75% for 1980-2000. Paper uses CES production function with college/high school workers, estimates σ_SU ≈ 1.62. Finds puzzling slowdown in trend demand growth for college workers post-1992 despite rapid technological change. [Claude classification]: This is a theoretical/conceptual paper that extends the canonical RBET framework. The paper mentions AI, robots, and computerization but only in passing as examples of skill-biased technological change; it does not empirically study AI impacts. The framework is calibrated/estimated with historical data. Education explains only 38% of inequality change 2000-2017 vs 75% for 1980-2000. Paper uses CES production function with college/high school workers, estimates σ_SU ≈ 1.62. Finds puzzling slowdown in trend demand growth for college workers post-1992 despite rapid technological change.