Automation and New Tasks - How Technology Displaces and Reinstates Labor Online Appendix Part 2

Change in task content of production, 1947−1987 20 L K A /A grows at a rate of 3% Reinstatement L K A /A grows at a rate of 1% 10 Change in task content L K A /A grows at a rate of 3% L K A /A grows at a rate of 1% 0 −10 L K A /A grows at a rate of 3% Displacement L K A /A grows at a rate of 1% −20 1947 1952 1957 1962 1967 1972 1977 1982 1987 Change in task content of production, 1987−2017 15 L K L K L K L K A /A grows at a rate of 2% Reinstatement L K A /A constant 10 5 0 −5 A /A grows at a rate of 2% Change in task content −10 A /A constant −15 A /A grows at a rate of 2% Displacement L K A /A constant −20 1987 1992 1997 2002 2007 2012 2017 Figure A1: Estimates of the displacement and reinstatement effects for K different assumed growth rates for AL i /Ai . Note: This figure presents our baseline estimates of the displacement and reinstatement K effects based on equation (A 13) for different values of the growth rate of AL i /Ai . The top panel is for 1947-1987, and as the baseline, assumes a growth rate for the relative labor-augmenting technological change of 2%. The bottom panel is for 1987-2017, and as the baseline, assumes a growth rate for the relative labor-augmenting technological change of 1.5%. Results for an elasticity of substitution σ = 0.8. Implied TFP growth, 1947−1987 200 Contribution of AL 150 100 Contribution of AK 50 Observed TFP 0 1947 1952 1957 1962 1967 1972 1977 1982 1987 Implied TFP growth, 1987−2017 200 Contribution of AL 150 100 Contribution of AK 50 Observed TFP 0 1987 1992 1997 2002 2007 2012 2017 Figure A2: Counterfactual TFP changes. Note: This figure presents the counterfactual TFP changes that would be implied if our estimates of the displacement and reinstatement effect in 1947-1987 and 1987-2017 were accounted for by industry-level changes in labor-augmenting and capital-augmenting technological changes alone, respectively, as derived in equations (A 14) and (A 15). For comparison, the figure also reports the observed increase in TFP for both periods. Manufacturing wage bill, 1947−1987 100 Productivity effect 80 60 Observed wage bill 40 20 Substitution effect Composition effect Change in task content 0 −20 Price effect −40 1947 1952 1957 1962 1967 1972 1977 1982 1987 Manufacturing wage bill, 1987−2017 60 40 Productivity effect 20 0 Substitution effect Composition effect Change in task content −20 Observed wage bill Price effect −40 1987 1992 1997 2002 2007 2012 2017 Figure A3: Labor demand in Manufacturing. Note: This figure presents the decomposition derived in equation (A3) for the manufacturing wage bill in 1947-1987 and 1987-2017. The top panel is for 1947-1987 and assumes a growth rate for the relative labor-augmenting technological change of 2%. The bottom panel is for 1987-2017 and assumes a growth rate for the relative labor-augmenting technological change of 1.5%. Results for an elasticity of substitution σ = 0.8. Change in task content of production, 1987−2017 50 Change in task content of production, 1987−2017 50 Farms Farms Forestry and fishing Forestry and fishing Apparel and leather and allied 0 Fabricated metal Furniture Textiles Plastics and rubber 0 Plastics and rubber Automotive Chemical −50 Mining, except oil and gas −50 Mining, except oil and gas Petroleum and coal −100 Petroleum and coal −100 0 10 20 30 40 0 20 Adjusted penetration of robots, 1993−2014 Change in task content of production, 1987−2007 60 80 Change in task content of production, 1987−2017 50 3641 50 40 Share of routine jobs in industry, 1990 3633 Farms Forestry and fishing 3861 Apparel and leather and allied Miscellaneous manufacturing 3764 0 0 3519 3489 Primary metals 3721 Computers 3561 −50 −50 Mining, except oil and gas Petroleum and coal −100 −100 3795 0 20 40 60 Share of firms using automation technologies, 1988−1993 80 0 2 4 6 8 10 Exposure to intermediate imports (% of total intermediates), 1993−2007 Figure A4: Automation technologies, offshoring, and changes in the task content of production. Note: Each panel presents the bivariate relationship at the industry level between change in task content and the indicated proxy for automation technologies or offshoring. Diamond markers designate manufacturing industries and circles non-manufacturing industries. The proxies are: adjusted penetration of robots, 1993-2014 (from Acemoglu and Restrepo, 2018b), share of employment in routine occupations in 1990 (Acemoglu and Autor, 2011), share of firms (weighted by employment) using automation technologies, from the 1988 and 1993 SMT, and exposure to imports of intermediate goods, from Feenstra and Hanson (1999). See text for details. Change in task content of production, 1987−2017 50 Change in task content of production, 1987−2017 50 Farms Farms Forestry and fishing Forestry and fishing Amusement andcare recreation Ambulatory health services Live entertainment Social assistance Educational services Hospitals Computer systems design 0 Educational services Hospitals Legal services 0 Data services and internet −50 −50 Mining, except oil and gas Petroleum and coal Mining, except oil and gas Petroleum and coal −100 −100 5 10 15 20 1 Share of new job titles, based on 1991 DOT and 1990 employment by occupation Change in task content of production, 1987−2017 50 3 4 5 Change in task content of production, 1987−2017 50 Farms Farms Forestry and fishing Forestry and fishing Professional services Live entertainment Management of companies Live entertainment Computer systemsactivities design Other transportation and support Administrative and support services 0 2 Number of emerging tasks, based on 1990 employment by occupation 0 Data services and internet Pipeline transportation −50 −50 Mining, except oil and gas Petroleum and coal Mining, except oil and gas −100 Petroleum and coal −100 0 5 10 Share of growth between 1990−2016 in occupations not in industry in 1990 15 −40 −20 0 20 Percent increase in number of occupations represented in industry Figure A5: New tasks and change in task content of production. Note: Each panel presents the bivariate relationship at the industry level between change in task content and the indicated proxy for new tasks. Diamond markers designate manufacturing industries and circles non-manufacturing industries. The proxies are: share of new job titles (from Linn, 2011), number of emerging tasks (from ONET), share employment growth between 1990 and 2016 in “new occupations”—those that were not present in the industry in 1990—, and the percent increase in the number of occupations present in the industry between 1990 and 2016. See text for details. Wage bill, 1947−1987 (alternative ordering) 100 Observed wage bill Productivity effect 80 60 40 20 Substitution effect Composition effect Change in task content 0 1947 1952 1957 1962 1967 1972 1977 1982 1987 Wage bill, 1987−2017 (alternative ordering) Productivity effect 40 Observed wage bill 30 20 10 Composition effect Substitution effect 0 Change in task content −10 1987 1992 1997 2002 2007 2012 2017 Figure A6: Alternative Ordering of the Wage Bill Decomposition. Note: The two panels present decompositions of changes in wage bill using the alternating ordering in equation (A 17). The top panel presents the decomposition of labor demand (wage bill) between 1947 and 1987. The bottom panel presents the decomposition of labor demand (wage bill) between 1987 and 2017. Results for an elasticity of substitution σ = 0.8 and relative labor-augmenting technological change at the rate of 2% per year (for 1947-1987) and 1.5% a year (for 1987-2017). Wage bill, 1987−2017, BLS data Productivity effect 40 Observed wage bill 30 20 10 Composition effect 0 Substitution effect Change in task content −10 1987 1992 1997 2002 2007 2012 2017 Change in task content of production, 1987−2017, BLS data 10 Reinstatement 5 0 −5 −10 Change in task content −15 −20 1987 Displacement 1992 1997 2002 2007 2012 2017 Manufacturing task content of production, 1987−2017, BLS data 10 Reinstatement 0 −10 Change in task content −20 −30 1987 Displacement 1992 1997 2002 2007 2012 2017 Figure A7: Sources of changes in labor demand, 1987-2017. Note: The top panel presents the decomposition of labor demand (wage bill) between 1987 and 2017 using BLS data. The middle and bottom panels present our estimates of the displacement and reinstatement effects for the entire economy and the manufacturing sector, respectively. Results for an elasticity of substitution σ = 0.8 and relative labor-augmenting technological change at the rate of 1.5% a year. Change in task content of production, 1947−1987 Reinstatement, yearly 40 20 Reinstatement Change in task content 0 Displacement −20 −40 Displacement, yearly 1947 1952 1957 1962 1967 1972 1977 1982 1987 Change in task content of production, 1987−2017 30 Reinstatement, yearly 20 Reinstatement 10 0 Change in task content −10 −20 Displacement −30 Displacement, yearly −40 1987 1992 1997 2002 2007 2012 2017 Figure A8: Estimates of the displacement and reinstatement effects, yearly and five-year changes. Note: This figure presents our baseline estimates of the displacement and reinstatement effects based on equation (A 13) and additional estimates using yearly changes (rather than five-year windows). The top panel is for 1947-1987 and assumes a growth rate for the relative labor-augmenting technological change of 2%. The bottom panel is for 19872017 and assumes a growth rate for the relative labor-augmenting technological change of 1.5%. Results for an elasticity of substitution σ = 0.8. Wage bill, 1947−1987 , sigma=0.6 Wage bill, 1947−1987 , sigma=0.8 100 Observed wage bill Productivity effect 100 80 80 60 60 40 40 20 20 Substitution effect Composition effect Change in task content 0 1947 1952 1957 1962 1967 1972 1977 1982 1987 1947 1957 1962 1967 1972 1977 1982 100 80 80 60 60 40 40 20 20 Change in task content Composition effect Substitution effect 0 1957 1952 1987 Wage bill, 1947−1987 , sigma=1.2 Observed wage bill Productivity effect 1952 Substitution effect Composition effect Change in task content 0 Wage bill, 1947−1987 , sigma=1 100 1947 Observed wage bill Productivity effect 1962 1967 1972 1977 1982 1987 Observed wage bill Productivity effect Change in task content Composition effect Substitution effect 0 1947 1952 1957 1962 1967 1972 1977 1982 1987 Figure A9: Sources of changes in labor demand for the entire economy, 1947-1987, for different values of σ. Note: This figure presents the decomposition of labor demand (wage bill) between 1987 and 2017 based on equation (A 8) in the text. The panels present the results for the vauels of σ indicated in their headers. In all panels, we assume relative labor-augmenting technological change at the rate of 2% a year. Change in task content of production, 1947−1987 , sigma=0.6 Change in task content of production, 1947−1987 , sigma=0.8 20 20 Reinstatement Reinstatement 10 10 0 Change in task content 0 Change in task content −10 −10 Displacement −20 Displacement −20 1947 1952 1957 1962 1967 1972 1977 1982 1987 1947 Change in task content of production, 1947−1987 , sigma=1 1952 1957 1962 1967 1972 1977 1982 1987 Change in task content of production, 1947−1987 , sigma=1.2 Reinstatement Reinstatement 20 20 10 10 Change in task content Change in task content 0 0 −10 −10 −20 −20 Displacement 1947 1952 1957 1962 1967 1972 1977 1982 1987 Displacement 1947 1952 1957 1962 1967 1972 1977 1982 1987 Figure A10: Estimates of the displacement and reinstatement effects for the entire economy, 1947-1987, for different values of σ. Note: This figure presents our baseline estimates of the displacement and reinstatement effects based on equation (A 13) in the text. The panels present the results for the values of σ indicated in their headers. In all panels, we assume relative labor-augmenting technological change at the rate of 2% a year. Manufacturing task content of production, 1947−1987 , sigma=0.6 Manufacturing task content of production, 1947−1987 , sigma=0.8 20 20 Reinstatement Reinstatement 10 10 Change in task content 0 0 Change in task content −10 −10 Displacement −20 1947 1952 1957 1962 1967 1972 1977 1982 1987 −20 Displacement 1947 Manufacturing task content of production, 1947−1987 , sigma=1 1952 1957 1962 1967 1972 1977 1982 1987 Manufacturing task content of production, 1947−1987 , sigma=1.2 Reinstatement Reinstatement 20 20 10 Change in task content 0 10 Change in task content 0 −10 −10 −20 −20 Displacement 1947 1952 1957 1962 1967 1972 1977 1982 1987 Displacement 1947 1952 1957 1962 1967 1972 1977 1982 1987 Figure A11: Estimates of the displacement and reinstatement effects for manufacturing, 1947-1987, for different values of σ. Note: This figure presents our baseline estimates of the displacement and reinstatement effects based on equation (A 13) in the text. The panels present the results for the values of σ indicated in their headers. In all panels, we assume relative labor-augmenting technological change at the rate of 2% a year. Wage bill, 1987−2017 , sigma=0.6 Wage bill, 1987−2017 , sigma=0.8 Productivity effect 40 Observed wage bill Productivity effect 40 30 30 20 20 10 Observed wage bill 10 Composition effect 0 Substitution effect Composition effect Substitution effect 0 Change in task content −10 Change in task content −10 1987 1992 1997 2002 2007 2012 2017 1987 Wage bill, 1987−2017 , sigma=1 1992 1997 2002 2007 2012 Wage bill, 1987−2017 , sigma=1.2 Productivity effect 40 Observed wage bill Productivity effect 40 30 30 20 20 10 10 Composition effect Substitution effect 0 Observed wage bill Composition effect Substitution effect 0 Change in task content −10 1987 2017 Change in task content −10 1992 1997 2002 2007 2012 2017 1987 1992 1997 2002 2007 2012 2017 Figure A12: Sources of changes in labor demand for the entire economy, 1987-2017, for different values of σ. Note: This figure presents the decomposition of labor demand (wage bill) between 1987 and 2017 based on equation (A 8) in the text. The panels present the results for the values of σ indicated in their headers. In all panels, we assume relative labor-augmenting technological change at the rate of 1.5% a year. Change in task content of production, 1987−2017 , sigma=0.6 Change in task content of production, 1987−2017 , sigma=0.8 Reinstatement 10 Reinstatement 10 5 5 0 0 −5 −5 −10 Change in task content −10 Change in task content −15 −15 Displacement −20 1987 1992 1997 2002 2007 2012 2017 −20 Displacement 1987 Change in task content of production, 1987−2017 , sigma=1 1992 1997 2002 2007 2012 2017 Change in task content of production, 1987−2017 , sigma=1.2 Reinstatement Reinstatement 10 10 5 5 0 0 −5 −5 −10 Change in task content −10 Change in task content −15 −15 −20 −20 Displacement 1987 1992 1997 2002 2007 2012 2017 Displacement 1987 1992 1997 2002 2007 2012 2017 Figure A13: Estimates of the displacement and reinstatement effects for the entire economy, 1987-2017, for different values of σ. Note: This figure presents our baseline estimates of the displacement and reinstatement effects based on equation (A 13) in the text. The panels present the results for the values of σ indicated in their headers. In all panels, we assume relative labor-augmenting technological change at the rate of 1.5% a year. Manufacturing task content of production, 1987−2017 , sigma=0.6 Manufacturing task content of production, 1987−2017 , sigma=0.8 Reinstatement 10 Reinstatement 10 0 0 −10 −10 Change in task content Change in task content −20 −20 −30 Displacement −30 1987 1992 1997 2002 2007 2012 2017 Displacement 1987 Manufacturing task content of production, 1987−2017 , sigma=1 1992 1997 2002 2007 2012 2017 Manufacturing task content of production, 1987−2017 , sigma=1.2 Reinstatement 10 Reinstatement 10 0 0 −10 −10 Change in task content −20 −20 Change in task content −30 −30 Displacement 1987 1992 1997 2002 2007 2012 2017 Displacement 1987 1992 1997 2002 2007 2012 2017 Figure A14: Estimates of the displacement and reinstatement effects for manufacturing, 1987-2017, for different values of σ. Note: This figure presents our baseline estimates of the displacement and reinstatement effects based on equation (A 13) in the text. The panels present the results for the values of σ indicated in their headers. In all panels, we assume relative labor-augmenting technological change at the rate of 1.5% a year. Labor share, 1850−1910 60 Industry 50 Overall 40 30 20 Agriculture 1850 1860 1870 1880 1890 1900 1910 Share GDP, 1850−1910 Industry 80 70 60 50 40 30 20 1850 Agriculture 1860 1870 1880 1890 1900 1910 Figure A15: Labor share and sectoral evolutions during the mechanization of agriculture, 1850-1910. Note: The top panel shows the labor share in value added in industry (services and manufacturing) and agriculture between 1850-1910, while the bottom panel shows the share of value added in these sectors relative to GDP. Data from Budd (1960). Table A 1: Relationship between gross change in task content of production, quantities produced, TFP, and skill intensity of industries. log change quantity, 1987-2016 Gross change in task content of production Change skill intensity, 1990-2016 (1) (2) (3) (4) (5) (6) 0.799 (0.262) 0.321 (0.152) 0.161 (0.206) 0.233 (0.141) -0.357 (0.407) 17.930 (1.219) -0.187 (0.086) 0.096 (0.037) -0.286 (0.427) 0.581 (0.235) -3.463 (1.541) 40.957 (2.726) -0.884 (0.162) -0.000 (0.019) 0.096 (0.037) 0.324 (0.149) 0.593 (0.219) -0.037 (0.018) 61 0.08 61 0.61 61 0.12 61 0.60 61 0.16 61 0.23 Chinese import competititon Offshoring of intermediates Manufacturing log change TFP, 1987-2016 Computer industry Observations R-squared Note: The table reports estimates between gross changes in task content of production and the change in quantities, TFP, and skill intensity of industries. The gross change in task content is defined as the sum of the absolute values of the displacement and reinstatement effects computed in equation (A 13). Columns 1-2 present results for the change in quantities produced (from the BEA-KLEMS). Columns 3-4 present results for the change in TFP (from the BEA-KLEMS). Columns 5-6 present results for the change in skill requirements, measured by the share of college educated workers in each industry (from the 1990 US Census and the pooled 2012-2016 ACS). All regressions are for the 61 industries used in or analysis of the 1987-2017 period. Standard errors robust against heteroskedasticity are in parenthesis.