Computer Automation’s Impact on Jobs and Skills

Tables and Figures Table 1. Share of Workers Using Computers at Work 1 Dependent variable: Routine “share” of tasks Share using computers 2 Share using computers 3 Annual change in share using computers 4 Share using computers -.230 (.030)** Routine tasks .015 (.003)** -.001 (.000)** Abstract tasks .064 (.002)** .002 (.000)** .000 (.003) Share with 4 or more years postsecondary education .235 (.052)** Mean log wage (1980) .450 (.019)** Log employment .015 (.005)** Adjusted R-squared .124 .358 .294 .352 N 1912 1912 452 1467 Note: By occupation, weighted least squares regressions. Data are from 1984, 1989, 1993, and   1997 October Current Population Surveys. Regressions pool the observation years and include year dummies (except for column 3). Dependent variable in columns 1, 2 and 4 is share of workers in occupation who use a computer at work and regression weights are sum of person weights for occupation. Dependent variable in column 3 is annual change in share of workers using computers at work estimated by regressing annual observations against year for each occupation (excluding occupations with fewer than three observations or with only one observation in any year); regression weights in column 3 are the inverse of the standard error of the regression for each occupation. Wage bill is the aggregate sample weight times the exponential of the mean log hourly wage in billions of dollars. Standard errors are in parentheses. * = significant at the 5% level, **=significant at the 1% level. Table 2. Employment Growth of Occupations, 1980-2013 Dependent variable: Annual growth rate (percent) of occupation hours worked Sample: Computer use 1 2 3 4 All All All All 1.18 (.38)** Offshorability index 3.34 (.47)** 2.23 (.24)** 2.32 (.15)** 5 Routine intensive 6 Mid wage 1.84 (.62) 2.08 (.14)** -1.96 (.24)** Use x capital skills -.33 (.52) Occupation group: Administrative support Service Production -1.58 (.06)** -1.61 (.03)** -1.54 (.19)** -1.48 (.05)** 1.37 (.07)** 1.38 (.06)** 2.12 (.18)** 1.59 (.00)** -1.56 (.06)** -1.56 (.06)** -1.53 (.15)** -1.23 (.02)** Adjusted R-squared .029 .189 .226 .227 .306 .147 N 317 301 317 314 154 145 Mean computer use .40 .39 .40 .40 .38 .41 Contribution of computer .47 (.15)** 1.29 (.18)** .88 (.10)** .92 (.06)** .70 (.24)** .85 (.06)** use to growth rate Note: Weighted least squares regressions of detailed occupations. Dependent variable is annual percentage growth in hours worked. Weighted by occupation hours worked. The offshorability index was developed by Jensen and Kletzer (2010). “Capital skills” refers to specific occupations that might be expected to be complementary with computers (see footnote 37). Standard errors are in parentheses. * = significant at the 5% level, **=significant at the 1% level. Errors in columns 3 through 6 are clustered by occupation group. Constant term not shown. Table 3. Employment Growth of Occupations, Full Model Estimates, 1980-2013 Dependent variable: Annual growth rate (percent) of occupation-industry hours worked Sample: 𝛼 𝛽 R-squared N 1 2 All All WLS IV-GMM 1.07 (.21)** -1.15 (.72) 3 Routine intensive 4 Low wage quartile 5 Mid wage quartiles 6 High wage quartile WLS WLS WLS WLS 1.82 (1.34) .30 (.57) -1.68 (1.00) -1.20 (.82) .49 (.91) .78 (.27)* -2.81 (1.11)* -1.61 (.50)** 1.52 (.74) .97 (1.01) .22 .07 .28 .30 .22 .34 10,837 10,809 4,954 2,279 6,208 2,350 Marginal effects of computer use Own occupation Industry .92 (.23)** -1.00 (.63) 1.61 (1.35) .20 (.58) .12 (.93) .58 (.27)* 1.64 (.76)* -1.42 (.44)** .84 (.88) -1.47 (.88) -1.10 (.75) -2.44 (.97)* Mean values of computer use Own occupation, U 41% 41% 39% 14% 42% 69% Industry, X 44% 44% 44% 32% 45% 55% Contribution of computer use to growth rate -.07 (.33) .01 (.71) -.41 (.42) -.84 (.38)* -.40 (.25) 1.58 (.76)* Note: All regressions include a full set of dummy variables for major occupation group and for major industry group. Dependent variable is annual percentage growth in hours worked for detailed occupation-industry cell. The sample includes all cells where data are available on computer use. Weighted by occupation hours worked. Standard errors are in parentheses and are clustered by industry group. * = significant at the 5% level, **=significant at the 1% level. Column 2 instruments the computer use variable using the mean log wage of the occupation in 1980 and degree of automation rating from O*NET. The probability value of the Hansen J statistic testing the over-identifying restrictions is .269. Table 4. Change in Within-Occupation Wage Gaps Panel A. Change Between 90th and 50th Percentiles Share using computers 1 2 3 4 1980-2013 1990-2013 1990-2013 1990-2013 .21 (.08)** .53 (.09)** .57 (.09)** .51  (.09)**   Education, 2nd wage quartile -.03 (.04) -.08 (.04) -.11 (.04)** -.08 (.04) Education, 4th wage quartile .02 (.04) .06 (.04) .10 (.04)* .06 (.04) Superstar I -.16 (.05)** Superstar II .16 (.09) Adjusted R-squared .04 .16 .18 .17 N 310 310 310 310 Panel B. Change Between 50th and 10th Percentiles 1 2 Share using computers .40 (.09)** .29 (.10)** Education, 2nd wage quartile .06 (.04) -.02 (.05) Education, 4th wage quartile -.05 (.04) -.02 (.05) Adjusted R-squared .15 .03 N 317 317 Note: Weighted least squares regressions of detailed occupation data. Dependent variable is annual percentage change in the difference in log wages between the 90th (50th) and 50th (10th) percentiles in the upper (lower) panel. Top panel excludes 7 occupations where some topcoded wage observations fall below the 90th percentile. Constant term not shown. Weighted by occupation hours worked with standard errors reported in parentheses. * = significant at the 5% level, **=significant at the 1% level. Table 5. Change in Share of Workforce with College Education 1 Sample: Share using computers All .11 (.01)** 2 Few abstract tasks .11 (.01)** 3 College not required .12 (.01)** 4 College not required .12 (.01)** .34 (.07)** Computer use x wage growth -.03 (.03) Wage growth Adjusted R-squared .24 .25 .48 .57 N 317 242 171 171 Note: Weighted least squares regressions of detailed occupation data. Dependent variable is change in the share of hours worked by workers with four or more years of postsecondary education from 1980 to 2013. Constant term not shown. Weighted by occupation hours worked with standard errors reported in parentheses. * = significant at the 5% level, **=significant at the 1% level. Figure 1. Job Polarization: Employment Growth of Occupations by Computer Use Note: Shows smoothed weighted average of percentage growth in hours worked for 317 detailed occupations. Smoothing done with an Epanechnikov kernel with .3 bandwidth. Bottom panel shows occupations with above-median and below-median computer use separately. Dashed vertical lines are at the 25th and 75th percentiles in the occupational wage. Horizontal dotted line is total hours growth. Figure 2. Bank Tellers and Automated Teller Machines Note: Teller data from Census and ACS 1% samples. Fulltime equivalent workers calculated assuming 2080 hours per work year. Data on number of ATMs installed from the Bank for International Settlements. Figure 3. Changes Between Occupations Region Employment Change in Automated Occupation Employment Change in Other Occupation I + - II - - III - + IV + + Figure 4. Share of Workers Using Computers by Characteristics of Occupational Tasks Note: The first panel shows computer use for occupations with above-median and below-median rated importance rating of abstract tasks; the second panel shows above-median and below-median rated occupations on the importance of routine tasks.