This site is undergoing review. Some annotations were human-generated, some AI-generated — all are being verified.
Back to papersSummary Main Finding Notes NBER Working Paper 32872
[Claude classification]: This is a handbook chapter (revised March 2025) that synthesizes and extends prior work by the authors, particularly Acemoglu and Restrepo (2022). The paper introduces new empirical measures of new tasks and provides novel estimates of ripple effects via the propagation matrix. The authors identify two puzzles: a 'missing technology puzzle' (82% of demand shifts unexplained) and an 'incidence puzzle' (employment effects larger than predicted by competitive model), suggesting need for non-competitive labor market models. The paper assumes lambda=0.5 (task elasticity from Humlum 2020) and eta=0.3 (sectoral elasticity from Buera et al. 2021), and sets automation/new task cost savings at 30%.
[Claude classification]: This is a handbook chapter (revised March 2025) that synthesizes and extends prior work by the authors, particularly Acemoglu and Restrepo (2022). The paper introduces new empirical measures of new tasks and provides novel estimates of ripple effects via the propagation matrix. The authors identify two puzzles: a 'missing technology puzzle' (82% of demand shifts unexplained) and an 'incidence puzzle' (employment effects larger than predicted by competitive model), suggesting need for non-competitive labor market models. The paper assumes lambda=0.5 (task elasticity from Humlum 2020) and eta=0.3 (sectoral elasticity from Buera et al. 2021), and sets automation/new task cost savings at 30%.
[Claude classification]: This is a handbook chapter (revised March 2025) that synthesizes and extends prior work by the authors, particularly Acemoglu and Restrepo (2022). The paper introduces new empirical measures of new tasks and provides novel estimates of ripple effects via the propagation matrix. The authors identify two puzzles: a 'missing technology puzzle' (82% of demand shifts unexplained) and an 'incidence puzzle' (employment effects larger than predicted by competitive model), suggesting need for non-competitive labor market models. The paper assumes lambda=0.5 (task elasticity from Humlum 2020) and eta=0.3 (sectoral elasticity from Buera et al. 2021), and sets automation/new task cost savings at 30%.
[Claude classification]: This is a handbook chapter (revised March 2025) that synthesizes and extends prior work by the authors, particularly Acemoglu and Restrepo (2022). The paper introduces new empirical measures of new tasks and provides novel estimates of ripple effects via the propagation matrix. The authors identify two puzzles: a 'missing technology puzzle' (82% of demand shifts unexplained) and an 'incidence puzzle' (employment effects larger than predicted by competitive model), suggesting need for non-competitive labor market models. The paper assumes lambda=0.5 (task elasticity from Humlum 2020) and eta=0.3 (sectoral elasticity from Buera et al. 2021), and sets automation/new task cost savings at 30%.
[Claude classification]: This is a handbook chapter (revised March 2025) that synthesizes and extends prior work by the authors, particularly Acemoglu and Restrepo (2022). The paper introduces new empirical measures of new tasks and provides novel estimates of ripple effects via the propagation matrix. The authors identify two puzzles: a 'missing technology puzzle' (82% of demand shifts unexplained) and an 'incidence puzzle' (employment effects larger than predicted by competitive model), suggesting need for non-competitive labor market models. The paper assumes lambda=0.5 (task elasticity from Humlum 2020) and eta=0.3 (sectoral elasticity from Buera et al. 2021), and sets automation/new task cost savings at 30%.
[Claude classification]: This is a handbook chapter (revised March 2025) that synthesizes and extends prior work by the authors, particularly Acemoglu and Restrepo (2022). The paper introduces new empirical measures of new tasks and provides novel estimates of ripple effects via the propagation matrix. The authors identify two puzzles: a 'missing technology puzzle' (82% of demand shifts unexplained) and an 'incidence puzzle' (employment effects larger than predicted by competitive model), suggesting need for non-competitive labor market models. The paper assumes lambda=0.5 (task elasticity from Humlum 2020) and eta=0.3 (sectoral elasticity from Buera et al. 2021), and sets automation/new task cost savings at 30%.
[Claude classification]: This is a handbook chapter (revised March 2025) that synthesizes and extends prior work by the authors, particularly Acemoglu and Restrepo (2022). The paper introduces new empirical measures of new tasks and provides novel estimates of ripple effects via the propagation matrix. The authors identify two puzzles: a 'missing technology puzzle' (82% of demand shifts unexplained) and an 'incidence puzzle' (employment effects larger than predicted by competitive model), suggesting need for non-competitive labor market models. The paper assumes lambda=0.5 (task elasticity from Humlum 2020) and eta=0.3 (sectoral elasticity from Buera et al. 2021), and sets automation/new task cost savings at 30%.
[Claude classification]: This is a handbook chapter (revised March 2025) that synthesizes and extends prior work by the authors, particularly Acemoglu and Restrepo (2022). The paper introduces new empirical measures of new tasks and provides novel estimates of ripple effects via the propagation matrix. The authors identify two puzzles: a 'missing technology puzzle' (82% of demand shifts unexplained) and an 'incidence puzzle' (employment effects larger than predicted by competitive model), suggesting need for non-competitive labor market models. The paper assumes lambda=0.5 (task elasticity from Humlum 2020) and eta=0.3 (sectoral elasticity from Buera et al. 2021), and sets automation/new task cost savings at 30%.
Tasks At Work: Comparative Advantage, Technology and Labor Demand
Acemoglu, Kong, Restrepo
2024NBER Working Paper Series10 citations
Theoretical / conceptualCausalTheoretical model
Automation / RobotsAI (General)Routine task changeAugmentation vs. substitutionGeneral automation
Acemoglu, Kong, and Restrepo develop a comprehensive task-based model where production requires tasks assigned to workers with different skills or capital, showing how automation, new tasks, and factor-augmenting technologies have distinct effects on labor demand, factor shares, and productivity through both theoretical derivations and empirical analysis of US labor markets 1980-2016.
Automation and new tasks jointly explain 67-84% of between-group wage changes and 53-68% of employment changes in the US from 1980-2016, with a 10% task loss due to automation reducing relative wages by 12% and a 10% increase in new tasks raising wages by 8.5%; these extensive-margin changes have much larger effects than factor-augmenting technologies.
Primary Datasets
US Census 1980; American Community Survey 2014-2018; BEA-BLS Integrated Industry Accounts; Dictionary of Occupational Titles (1977, 1991); O*NET
Secondary Datasets
Current Population Survey; Compustat (for markups via Hubmer and Restrepo 2021); Robot penetration data (Acemoglu and Restrepo 2020a); BLS Detailed Capital Tables
- Key Methods
- The paper develops theoretical task model with propagation matrix capturing ripple effects, derives reduced-form estimating equations linking wage changes to task displacement and reinstatement, and implements structural GMM estimation combining theory with US Census/ACS data for 500 demographic groups across industries.
- Sample Period
- 1980-2016
- Geographic Coverage
- United States
- Sample Size
- 500 demographic groups defined by education (5 levels), gender, age (5 groups), ethnicity (4 categories), and native/foreign-born status
- Level of Analysis
- Individual, Task, Industry, Occupation
- Occupation Classification
- Census occupation codes mapped to O*NET; 300 detailed occupations tracked consistently; occupations aggregated into 6 broad groups for some analyses
- Industry Classification
- 50 industries from BEA-BLS Integrated Industry Accounts matched to Census industry codes; 16 aggregated industries used in some specifications