Robots and Jobs: Evidence from US labor markets Daron Acemoglu 1 Pascual Restrepo 2 1 MIT: daron@mit.edu 2 Boston University pascual@bu.edu January 2018
Industrial Robots Automatic and multipurpose machines with several axis suitable for industrial applications. Salient example of class of techs that compete task-by-task against labor. Large increase between 1993 and 2014: in the US, fivefold increase from 2 to 10 robots per thousand industry workers.
Exposure to Robots Exposure to robots for a commuting zone, c: Exposure to robots in c = industryi Employment share industry i in c Adoption of robots in i. Employment share industry i in c: is the czone share of employment in each industry in 1990. Adoption of robots industry i: use of robots in three-digit manufacturing industries and two-digit nonmanufacturing industries. Data from the International Federation of Robotics covering the period from 1993-2007. We use adoption among European industries as a proxy for the technological possibilities for the U.S.
Adoption of Robots Over Time Figure: Trends in the use of industrial robots per thousand workers. Data: IFR.
Exposed Industries Figure: Increase use of robots in US and European industries. Data from the IFR. Both axis are in a logarithmic scale.
Exposed Industries Automotive Plastic and chemicals Metal products Electronics Metal machinery Textiles Other manufacturing (1) (2) (3) (4) (1) (2) (3) (4) (1) (2) (3) (4) (1) (2) (3) (4) (1) (2) (3) (4) (1) (2) (3) (4) (1) (2) (3) (4) Normalized indices 0.5 1 (1)Exposure to robots (3)Offshoring of intermediate goods (2)Exposure to chinese imports (4)ICT capital growth Figure: Increase in robots and other trends among selected US industries.
Exposed Regions A. Exogenous exposure to robots from 1993 to 2007 Figure: Exposure to robots across U.S. commuting zones. Robots per thousand workers 1.00 4.87 0.74 1.00 0.56 0.74 0.45 0.56 0.30 0.45 0.12 0.30
Exposed Regions A. Exogenous exposure to robots from 1993 to 2007 C. Exposure to Chinese imports from 1990 to 2007 Robots per thousand workers 1.00 0.74 0.56 0.45 0.30 0.12 4.87 1.00 0.74 0.56 0.45 0.30 Imports per worker ($1,000) 5.82 3.71 2.57 1.48 0.51 0.00 49.00 5.82 3.71 2.57 1.48 0.51 Figure: Exposure to robots across U.S. commuting zones. Comparison to the China shock.
Robot-related Activities in Exposed Regions log of 1+number of integrators 0 1 2 3 4 5 Detroit city, MI Muncie city, IN Defiance city, OH Lorain city, OH Lansing city, MI Racine city, WI Saginaw city, MI Wilmington city, DE 0 1 2 3 4 5 Exposure to robots from 1993 to 2007 Figure: Exposure to robots predicts the location of robot integrators in the US. Data on integrators: Leigh and Kraft (2016)
Reduced-form Impact on Employment Rate Change in Census employment from 1990 to 2007-5 0 5 10 Defiance city, OH Racine city, WI Lorain Saginaw city, OH city, MI Wilmington city, DE Muncie city, IN Lansing city, MI 0 1 2 3 4 5 Exposure to robots from 1993 to 2007 Detroit city, MI Figure: Regression plots for the impact of exposure to robots on the Census employment to population rate (in p.p.). IV estimates: one robot reduces employment by 6 jobs.
Reduced-form Impact on Wages Change in log of hourly wages from 1990 to 2007-15 -10-5 0 5 10 Muncie city, IN Racine city, WI Lorain city, OH Wilmington city, DE Defiance Lansing city, OHcity, MI Saginaw city, MI 0 1 2 3 4 5 Exposure to robots from 1993 to 2007 Detroit city, MI Figure: Regression plots for the impact of exposure to robots on the log of hourly wages (in log points). IV estimates: one robot per thousand workers reduces wages by 0.75%.
Placebo Exercise Change in Census employment from 1970 to 1990 -.05 0.05.1.15.2 0 1 2 3 4 5 Exposure to robots from 1993 to 2007 Figure: Placebo test of whether exposure to robots is correlated with past trends in employment and wages.
Effects on Different Industries CBP manuf. Manufacturing Highly robotized Other manuf. Transportation Agri. and mining Construction Services & retail Business services Wholesale Finance Public sector (1) (2) (3) (1) (2) (3) (1) (2) (3) (1) (2) (3) (1) (2) (3) (1) (2) (3) (1) (2) (3) (1) (2) (3) (1) (2) (3) (1) (2) (3) (1) (2) (3) (1) (2) (3) Point estimate -.8 -.6 -.4 -.2 0.2 (1)-Long-differences (2)-Downweights outliers (3)-Stacked differences Figure: The figure plots estimates of the impact of exposure to robots on the change in employment rates separated by industry. Heavily robotized: cars, plastic and chemicals, pharmaceuticals, electronics, metal products, and food production. Non-robotized manufacturing: recycling, toys, basic metals, textiles, paper, furniture and transportation equipment (not cars).
Effects on Different Occupations Routine manual Blue collar Assembly Mech. and transp. Farmers and miners Services Clerical Retail Professionals Management Financiers (1) (2) (3) (1) (2) (3) (1) (2) (3) (1) (2) (3) (1) (2) (3) (1) (2) (3) (1) (2) (3) (1) (2) (3) (1) (2) (3) (1) (2) (3) (1) (2) (3) Point estimate -.8 -.6 -.4 -.2 0.2 (1)-Long-differences (2)-Downweights outliers (3)-Stacked differences Figure: The figure plots estimates of the impact of exposure to robots on the change in employment rates separated by occupation.
Where did the workers go? Workers do not seem to be reallocating to other jobs. Workers do not seem to be moving to other regions. Maybe this will happen, but the adjustment seems sluggish and costly. Workers are dropping out of the labor force.
Adding Up The Numbers Robots could have reduced employment by 360,000-660,000. Moderate estimates. No robocalypse. But the reason why aggregate effects are small is because industrial robots are not adopted widely yet. US expected to add 600,000 robots in the next 10 years: 2 million manufacturing jobs. 2 million additional jobs due to multiplier.
Robots in other countries Full sample of countries Increase in robots per thousand workers from 1993 to 2014 (annualized) 0.2.4.6.8 PAK EGY ARG DNK ESP TUR IRL GRC SWE CZE BGR BEL ISL NZL PRT ISR HUN USA BRA EST AUT MDA MYS ROM VEN UKR SVK MAC VNM MAR COL FRA PER CHL CHN IDN IND AUS CHE PHL NOR POL GBR THA NLDITA SVN FIN DEU KOR SGP HKG 0.2.4.6 Aging from 1990 to 2025 Figure: Relationship between aging (change in the ratio of workers above 56 to workers between 21 and 55 between 1990 and 2025) and the increase in the number of industrial robots per thousand workers from 1993 to 2014.
Future of labor Future of labor depends on whether new technologies favoring labor are created and counteracting forces offset losses. The productivity effect. Capital accumulation and investment. Deepening of automation. Investment in skills. New tasks and jobs.
History of automation British industrial revolution: factory system and automation in textiles. Stagnant wages for over 60 years. But eventually, large increase in capital, and improvement of installed machinery. New jobs: engineers, machinists, repairmen, conductors, back-office workers and managers. The mechanization of agriculture: horse-powered reaper, harvesters, and threshing machines. Large migration away from rural areas. But eventually, large increase in capital, and improvement of installed machinery (horses to tractors). New jobs and sectors: industry and farm equipment. Large investment in high school education.
Is this time different? Counteracting forces are lagging behind automation: The productivity effect: so-so technologies? Capital accumulation and investment: much more to do. Investment in skills: severe mismatch? New tasks and jobs: geographic gap?
New Job Titles Employment growth (annualized) 10% 5% 0 5% 10% 15% From 1980 to 2015 From 1990 to 2015 From 2000 to 2015 0 0.2 0.4 0.6 0.8 Share of new job titles at the beginning of each period Figure: Employment growth by occupation over different time periods (annualized), plotted against the share of new job titles at the beginning of each period in each occupation. Data from Jeffrey Lin (2011).
New Job Titles Figure: Commuting zones with the largest share of workers doing new work in 2007. Geographic gap: zero correlation with exposure to automation.
New Job Titles Share of workers in "new jobs" in 2007.02.04.06.08.1 0 1 2 3 4 5 Exposure to robots from 1990 to 2007 Figure: Correlation between exposure to robots and the location of new work.