Back Matter
  • 1 https://isni.org/isni/0000000404811396, International Monetary Fund
  • | 2 https://isni.org/isni/0000000404811396, International Monetary Fund
  • | 3 https://isni.org/isni/0000000404811396, International Monetary Fund
  • | 4 https://isni.org/isni/0000000404811396, International Monetary Fund

Annex I. Data and Definitions of RTI and ICT Use Indices

The Organisation for Economic Co-operation and Development (OECD) has administered the Programme for the International Assessment of Adult Competencies (PIAAC) surveys in two rounds between 2011 and 2016. In our sample, we include 30 countries, for which data are available (refer to Table A1 for country coverage and sample sizes).37 The survey covers adults between the ages of 16 to 65 and collects detailed demographic and work information for each respondent. In addition, PIAAC assesses respondents’ numeracy, literacy, and problem-solving skills, which we use as proxies for workers’ ability. Variables describing the frequency at which a respondent performs a set of tasks at work are particularly relevant to the analysis.

Table A1.

Country Sample: Number of Observations

article image
Source: PIAAC survey; IALS Survey; and IMF staff calculations.

Table A2 lists the variables used for construction of the RTI and ICT use indices. We use the methodology in De La Rica and Gortazar (2016) to construct three components of the RTI index: abstract, manual, and routine. Most questionnaire items in Table A2 consist of five responses for indicating the frequency at which a task is performed: never, less than once a month, less than once a week but not every day, at least once a week but not every day, and every day.

Table A2.

Questionnaire Items Used to Construct RTI

article image
Source: PIAAC Survey; IMF Staff calculations.

The “abstract” component consists of analytical and interpersonal tasks like writing reports, solving complex problems, and negotiating with people. We consider two types of manual tasks—routine tasks involving hand and finger dexterity and nonroutine physical work associated with caregiving and operating construction-related equipment. Since PIAAC provides information on only two types of manual tasks, we classify them as routine and nonroutine based on previous work by Autor, Levy, and Murnane (2003) and De La Rica and Gortazar (2016). We also test the relationship between the manual tasks and our estimate of the probability of automation and find that hand and finger dexterity (performing physical work for long hours) is positively (negatively) associated with the probability of automation. The “routine” component consists of lack of flexibility and learning indicators, low values of which indicate repetitive nature of work and tasks that can be performed by following a set of rules, and thus easily codified. We perform a principal component analysis to derive an index for each RTI component – abstract, routine, and manual – and ICT use.38 Finally, we construct the RTI index by subtracting abstract and manual components from the routine component:

RTIi=RoutineiAbstractiManuali,(1)

and standardize the final index between zero and unity.

For the shift-share analysis in Figure 12, we use the International Adult Literacy Survey. This survey was conducted between 1994 and 1998 for 22 countries to measure adult skills and literacy. We link trend data from the IALS to the PIAAC.

Annex II. RTI, ICT Use and Wage Decomposition Method

We examine the contribution of individual and job characteristics to the observed gender RTI gap by estimating the following specification:

RTIi=β0+β1Femaleic+ΣβindXicind+ΣβabilityXicability+ΣβjobXicjob+α0+σs+τc+ϵic,(2)

in which β0 is a constant; Femaleic is an indicator for female; Xicind are individual controls including age, level of education, presence of a partner and children, and immigrant status; Xicability are numeracy and literacy test scores; Xicjob include experience, on-the-job training, and part-time status; α0 is occupation fixed effects; σs is sector fixed effects; τc is country fixed effects; and εic is a normally distributed error term. We cluster the standard errors at the country level.

PIAAC provides wage data for 24 countries in our sample (see Table A1 for country coverage and corresponding sample sizes). For this subset of countries, we evaluate the contribution of gender differences in routineness to the observed gender wage gap. We use a standard Mincer regression to pin down the explanatory power of the RTI index in accounting for the gender wage gap relative to demographic characteristics (age), education, and ability, as measured by literacy and numeracy. We augment our Mincer regression with controls for sectoral choice, occupational choice, part-time vs. full time work, motherhood, marital status which are known to affect gender wage differences (Blau and Kahn, 2017). In addition, we include controls that may affect wages in general such as immigration status and country fixed effects to account for institutional differences across countries. All variables are collected at an individual level in the PIAAC dataset.

We estimate the following linear model for each worker i pooled across all countries and sectors:

wagei=β0+β1Femaleic+β2RTIic+ΣβindXicind+ΣβabilityXicability+ΣβjobXicjob+α0+σs+τc+ϵic,(3)

in which β0 is a constant; Femaleic is an indicator for female; RTIic is an individual RTI index described in Annex I; Xicind are individual controls including age, level of education, presence of a partner and children, and immigrant status; Xicability are numeracy and literacy test scores; Xicjob include experience, on-the-job training, and part-time status; α0 is occupation fixed effects; σs is sector fixed effects; τc is country fixed effects; εic is a normally distributed error term. We cluster the standard errors at the country level. For both decompositions, we first estimate the coefficient on the female indicator without additional controls to derive an unconditional gender gap in a given outcome variable – RTI index or wages. We then estimate the full model to pin down the conditional gender wage gap and examine how much of the change in the unconditional gender gap can be attributed to different control variables, using the decomposition method outlined in Gelbach (2012).

ANNEX III. Estimating Probability of Automation

We follow the method employed by Arntz, Gregory, and Zierahn (2017) to link occupation-based estimates of the probability of automation with the task composition and characteristics of individual workers and re-estimate the probability of automation at the level of each individual worker. The estimates for the probability of automation of occupational categories are drawn from Frey and Osborne (2017). Their work uses occupational classification and job task descriptions from O*NET, a database maintained by the US Department of Labor containing detailed standardized information across nearly a thousand occupations in the US economy. These task descriptions are used to determine the automatability of an occupation given ‘state of art computer-controlled equipment’ and the availability of big data.

To assign automatability, Frey and Osbourne (2017) use a two-stage process. In the first stage, they hand-label a subset of occupations from the data based on whether they are fully automatable, using the informed opinions of Machine Learning researchers. In the second stage, they use a probabilistic model to impute the probability of automation from the hand-labelled sample to the full sample of occupations using nine specific job task characteristics contained in the O*NET data that are deemed to constitute bottlenecks in automatability. The resulting dataset contains 702 occupations and their associated probabilities of automation on a continuous scale between 0 and 100 percent.

Since the Frey and Osbourne (2017) estimates for probability of automation are calculated at the level of occupations, Arntz, Gregory, and Zierahn (2017) further impute them to worker characteristics and job task descriptions as contained in the PIAAC dataset. Instead of using a small set of bottlenecks, this method relates the probabilities estimated by Frey and Osbourne (2017) to information regarding the worker such as their age, gender, education, competencies, training, income as well as a comprehensive set of job task characteristics. The full set of variables used in this estimation are contained in Table A4.

Table A3.

Correlations of RTI Components and Probability of Automation

article image
article image
Source: Authors calculations.Note: RTI = routine task intensity.
Table A4.

Estimation of Probability of Automation

article image
article image

To relate occupation level automation probabilities to individuals, individuals in the PIAAC data must be matched to the occupational codes in O*NET for which we have estimates of automation probability from Frey and Osbourne (2017). Since the PIAAC data only contains 2-digit ISCO codes for occupations, each individual can be mapped to multiple occupations in the Frey and Osbourne (2017) estimates. Therefore, in the spirit of Arntz, Gregory, and Zierahn (2017) we use the Expectation Maximization algorithm and estimate an individual-level regression:

Prob(Autom)ij=Σn=1NβnXin+ϵij(4)

in which i are individuals,j are duplicates of these individuals when multiple probabilities are associated with one individual, and Xin contains individual, job, and task characteristics. βn are parameters which capture the impact of the regressors on probability of automation, which is restricted to the interval 0 to 100 percent.

We use a weighted Generalized Linear Model (GLM) for our estimation, with equal initial weights for all duplicates j for individual i . For each iteration of the regression, we compare the prediction from our estimated model with the actual probability of automation as given in the Frey and Osbourne (2017) data and recalculate the weights as per Ibrahim (1990) in which

wij=f(yhatyij|xin,βn)/Σf(yhatyij|xin,βn)(5)

and f(.) is the standard normal density. Once weights converge and best fit is achieved, we use the estimated parameters βn to calculate the predicted probabilities of automation based on individual worker and job task characteristics39. Table A4 shows estimates from the model estimated for the US data. We estimate 4 models with minor variations in the set of regressors contingent on data availability.40

REFERENCES

  • Acemoglu, Daron, and Pascual Restrepo. 2017. “The Race Between Machine and Man: Implications of Technology for Growth, Factor Shares, and Employment.” NBER Working Paper No. 22252, National Bureau of Economic Research, Cambridge, MA.

    • Search Google Scholar
    • Export Citation
  • Acemoglu, Daron, and Pascual Restrepo. 2018. “Artificial Intelligence, Automation, and Work.” NBER Working Paper No. 24196, National Bureau of Economic Research, Cambridge, MA.

    • Search Google Scholar
    • Export Citation
  • Altonji, Joseph, and Rebecca Blank. 1999. “Race and Gender in the Labor Market.” Handbook of Labor Economics 3: 3143259.

  • Arntz, Melanie, Terry Gregory, and Ulrich Zierahn. 2016. “The Risk of Automation for Jobs in OECD Countries: A Comparative Analysis”, OECD Social, Employment and Migration Working Papers, No. 189, OECD Publishing, Paris.

    • Search Google Scholar
    • Export Citation
  • Autor, David, and David Dorn. 2013. “The Growth of Low-Skill Service Jobs and the Polarization of the US Labor Market.” American Economic Review 103 (5): 155397.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Autor, David, Frank Levy, and Richard J. Murnan. 2003. “The Skill Content of Recent Technological Change: An Empirical Exploration.”The Quarterly Journal of Economics 118 (4): 1279333.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Bacolod, Marigee P., and Bernardo S. Blum. 2010. “Two Sides of the Same Coin US “Residual” Inequality and the Gender Gap.” Journal of Human Resources 45 (1): 197242.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Black, Sandra E., and Alexandra Spitz-Oener. 2010. “Explaining Women’s Success: Technological Change and the Skill Content of Women’s Work.” The Review of Economics and Statistics 92 (1): 18794.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Blau, Francine D., and Lawrence M. Kahn. 2017. “The Gender Wage Gap: Extent, Trends, and Explanations.” Journal of Economic Literature 55 (3): 789865.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Burtch, Gordon, Seth Carnahan, and Brad N. Greenwood. 2016. “Can You Gig It? An Empirical Examination of the Gig-Economy and Entrepreneurial Activity.” SSRN Electronic Journal.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Brussevich, Maria, Pooja Karnane, and Salma Khalid, 2018, “Is Technology Widening the Gender Gap? Automation and the Future of Female Employment,” IMF Working Paper (forthcoming), International Monetary Fund, Washington, DC.

    • Search Google Scholar
    • Export Citation
  • Brynjolfsson, Erik, Tom Mitchell, and Daniel Rock, 2018, “Economic Consequences of Artificial Intelligence and Robotics: What Can Machines Learn and What Does It Mean for Occupations and the Economy?AEA Papers and Proceedings 108: 4347.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Cuberes, David, Monique Newiak, and Marc Teignier. 2017. “Gender Inequality and Macroeconomic Performance.” In, Women, Work, and Economic Growth: Leveling the Playing Field 3148. Washington, DC: International Monetary Fund.

    • Search Google Scholar
    • Export Citation
  • Das, Mitali, and Benjamin Hilgenstock. 2018. “The Exposure to Routinization: Labor Market Implications for Developed and Developing Economies.” IMF Working Paper 18/135, International Monetary Fund, Washington, DC.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Dasgupta, Nilanjana, and Jane G. Stout. 2014. “Girls and Women in Science, Technology, Engineering, and Mathematics: STEMing the Tide and Broadening Participation in STEM Careers.”SAGE Journals 1 (1): 2129.

    • Search Google Scholar
    • Export Citation
  • Dasgupta, Nilanjana, and Tara C. Dennehy. 2017. “Female Peer Mentors Early in College Increase Women’s Positive Academic Experiences and Retention in Engineering.” National Academy of Sciences, Washington, DC.

    • Search Google Scholar
    • Export Citation
  • Dauth, Wolfgang, Sebastian Findeisen, Jens Südekum, and Nicole Wößner. 2017. “German Robots: The Impact of Industrial Robots on Workers.” IAB Discussion Paper 201730, Institute for Employment Research, Nuremberg.

    • Search Google Scholar
    • Export Citation
  • De La Rica, Sara, and Lucas Gortazar. 2016. “Differences in Job De-Routinization in OECD Countries: Evidence from PIAAC.” IZA Discussion Paper No. 9736. Institute for the Study of Labor, Bonn.

    • Search Google Scholar
    • Export Citation
  • Deloitte. 2017. Connecting Small Businesses in the US. Commissioned by Google.

  • Deloitte. 2018. 2018 Global Health Care Outlook: The Evolution of Smart Health Care.

  • Deming, David J. 2017. “The Growing Importance of Social Skills in the Labor Market.”The Quarterly Journal of Economics 132 (4): 1593640.

  • Demirguc-Kunt, Asli, and Leora Klapper. 2012. “Measuring Financial Inclusion: The Global Findex Database.” World Bank Policy Research Working Paper No. 6025, Washington, DC.

    • Search Google Scholar
    • Export Citation
  • Evers, Michiel, Ruud de Mooij, and Daniel Van Vuuren. 2008. “The Wage Elasticity of Labor Supply: A Synthesis of Empirical Estimates.” De Economist 156 (1): 2543.

    • Search Google Scholar
    • Export Citation
  • Frey, Carl B., and Michael A. Osborne. 2013. “The Future of Employment: How Susceptible are Jobs to Computerisation?” Unpublished Manuscript, Oxford University.

    • Search Google Scholar
    • Export Citation
  • Gelbach, Jonah B. 2012. “When Do Covariates Matter? And Which Ones, and How Much?Journal of Labor Economics 34 (2): 50943.

  • Goos, Maarten, Alan Manning, and Anna Salomons. 2014. “Explaining Job Polarization: Routine-Biased Technological Change and Offshoring.” American Economic Review 104 (8): 250926.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Grigoli, Francesco, Zsoka Koczan, and Petia Topalova. 2018. “Drivers of Labor Force Participation in Advanced Economies: Macro and Micro Evidence.” IMF Working Paper 18/150, International Monetary Fund, Washington, DC.

    • Search Google Scholar
    • Export Citation
  • Hawksworth, John, Richard Berriman, and Saloni Goel. 2018. Will Robots Really Steal our Jobs? An International Analysis of the Potential Long-Term Impact of Automation. PricewaterhouseCoopers.

    • Search Google Scholar
    • Export Citation
  • Hsieh, Chang-Tai, Erik Hurst, Charles I. Jones, and Peter J. Klenow. 2013. “The Allocation of Talent and U.S. Economic Growth.” NBER Working Paper No. 18693, National Bureau of Economic Research, Cambridge, MA.

    • Search Google Scholar
    • Export Citation
  • Hu, Luojia. 2003. “The Hiring Decisions and Compensation Structures of Large Firms.” Industrial and Labor Relations Review 56 (4): 66381.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • International Monetary Fund (IMF). 2018. G20 Report on Future of Work: Measurement and Policy Challenges. Washington, DC.

  • International Monetary Fund (IMF). 2017. “Gender Budgeting in G7 Countries.” IMF Policy Paper, Washington, DC.

  • International Monetary Fund (IMF). 2012. “Fiscal Policy and Employment in Advanced and Emerging Economies.” IMF Policy Paper, Washington, DC.

    • Search Google Scholar
    • Export Citation
  • International Telecommunication Union. 2017. ITU Gender Dashboard.

  • Kochhar, Kalpana, Sonali Jain-Chandra, and Monique Newiak (eds.). 2017. Women, Work, and Economic Growth: Leveling the Playing Field. Washington, DC: International Monetary Fund.

    • Search Google Scholar
    • Export Citation
  • McKinsey & Company. 2017. “Jobs Lost, Jobs Gained: Workforce Transitions in a Time of Automation.”

  • McKinsey & Company. 2016. “Where Machines Could Replace Humans – and Where They Can’t (Yet).”

  • Ngai, L. Rachel, and Barbara Petrongolo. 2017. “Gender Gaps and the Rise of the Service Economy.” American Economic Journal: Macroeconomics 9 (4): 144.

    • Search Google Scholar
    • Export Citation
  • Organisation for Economic Co-operation and Development (OECD). 2018. “Empowering Women in the Digital Age. Where Do We Stand?Paris.

    • Search Google Scholar
    • Export Citation
  • Organisation for Economic Co-operation and Development (OECD). 2017. “Going Digital: The Future of Work for Women.” Policy Brief on the Future of Work, Paris.

    • Search Google Scholar
    • Export Citation
  • Organisation for Economic Co-operation and Development (OECD). 2016. Proceedings from the Conference on Improving Women’s Access to Leadership. Paris, March 8.

    • Search Google Scholar
    • Export Citation
  • Olivetti, Claudia, and Barbara Petrongolo. 2014. “Gender Gaps across Countries and Skills: The Industry Structure.” Review of Economic Dynamics 17 (4): 10942025.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Sahay, Ratna, Martin Čihák, and others. 2018 (forthcoming). “Women in Finance: A Case for Closing Gaps.” IMF Staff Discussion Note, International Monetary Fund, Washington, DC.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Weinberg, Bruce A. 2000. “Computer Use and the Demand for Female Workers.” Industrial and Labor Relations Review 53 (2) 290309.

  • West, Darrell M. 2015. “What Happens if Robots Take the Jobs? The Impact of Emerging Technologies on Employment and Public Policy.” Brookings Institution, Washington, DC.

    • Search Google Scholar
    • Export Citation
  • World Bank. Forthcoming. World Development Report 2019: The Changing Nature of Work. Washington, DC.

  • World Economic Forum (WEF). 2017a. The Global Gender Gap Report 2017. Geneva.

  • World Economic Forum (WEF). 2017b. “The Industry Gender Gap: Women and Work in the Fourth Industrial Revolution.” Geneva.

  • Zervas, Georgios, Davide Proserpio, and John W. Byers. 2017. “The Rise of the Sharing Economy: Estimating the Impact of Airbnb on the Hotel Industry.” Journal of Marketing Research 54 (5): 687705.

    • Crossref
    • Search Google Scholar
    • Export Citation

*

We would like to thank Paulo Mauro, Ruud de Mooij, Maurice Obstfeld and seminar participants at the IMF for their helpful comments. We are also grateful to Melanie Arntz, Terry Gregory and Ulrich Zierahn from the OECD for sharing their results.

1

Automation is hardly a novel phenomenon. Traditional sectors such as agriculture and manufacturing have already experienced large substitutions of labor with machine capital. But computerization of white-collar services in many advanced economies, such as logistics and tax preparation, has accelerated in recent years (Acemoglu and Restrepo, 2018). At the same time, progress in machine learning is further expanding the set of activities that can be performed more efficiently by computers than humans, such as image and speech recognition, natural language processing, and predictive analytics (Brynjolfsson, Mitchell, and Rock, 2018), suggesting a significantly broader scope for task automation over the medium-term.

2

See the routine vs. non-routine task typology of Autor, Levy, and Murnane (2003).

3

Using the occupational task data from the Dictionary of Occupational Titles, Autor, Levy, and Murnane (2003) show that routine-task intensity predicts workers’ exposure to computerization in the US. Goos, Manning, and Salomons (2014) extend the task-based approach to 16 western European countries to show that routine-biased technological change decreases employment mainly in the middle-skill occupations. Using data on occupational distribution of 85 countries, Das and Hilgenstock (2018) find that developing economies are significantly less exposed to routinization than advanced economies but the risks of routinization have risen globally over time.

4

In addition, the survey contains demographic information and measures of literacy, numeracy, and problem-solving skills for each respondent. Annex I contains further details on country and variable coverage.

5

We calculate the RTI index at the individual level using task composition of each survey respondent. This allows us to relax two important assumptions relative to the previous literature using US-based occupational routineness estimates: (1) workers perform identical tasks within occupations across countries; and (2) workers have access to the same technologies across countries.

6

Das and Hilgenstock (2018) show that exposure to routinization is driven by the declining price of investment goods, the structure of employment (high manual-intensive agriculture versus high routine-intensive clerical work) and job offshoring, in advanced economies. IMF (2018) finds significant cross-country variation in ICT prices and rapid growth of the gig economy, both of which may affect our estimates of job routineness levels.

7

The gender ICT gap is defined as the ratio of the male ICT use index to the female ICT use index. See Annex I for details on the construction of the ICT index.

8

Occupation categories are based on the International Standard Classification of Occupations (ISCO-08). Sector categories are based on the International Standard Industrial Classification (ISIC rev. 4).

9

Our results are consistent with the existing literature, which attributes differences in earnings to gender disparities along both the extensive margin of FLFP—share of women in the labor force (Olivetti and Petrongolo, 2014)—and intensive margin—uneven distribution of women across sectors and occupations (Hsieh and others, 2013; Ngai and Petrongolo, 2017). Differences in educational attainment have also been found to be large drivers of documented gender wage gaps (Altonjii and Blank, 1999).

10

The impact of RTI differences on the wage, and therefore the wage gap, is linked to the prevailing structure of production and whether it favors skill sets that are relatively unequally distributed between the genders. For instance, Black and Spitz-Oener (2010) find that the demand shift towards non-routine tasks in the labor market and women’s increasing share of non-routine tasks in the workplace has narrowed gender wage inequality in Germany. Bacolod and Blum (2010) show that the gender wage gap in the US narrowed owing to increasing returns to cognitive and interpersonal skills, with women having higher participation in jobs requiring these skills.

11

This is calculated as the present value of 5 percent of the average US pay gap (assuming the gap is $171 per week, $9 approx. is attributable to RTI), applied over a 20-year working period and applying a 4 percent annual rate of return.

12

Our approach follows Arntz, Gregory, and Zierahn (2017). Hawksworth, Berriman, and Goel (2018) use a similar framework, but focus less on the gender dimension.

13

Most studies assume that all individuals within an occupation are assigned identical tasks. In such a setup, differences between men and women’s exposure to automation can only arise from dissimilar modes of participation in occupations that face different risks of automation. This assumption obscures the fact that gender differences in exposure to automation can also arise from variation in task assignment at the workplace. Additionally, estimating automation probabilities at the occupational level assumes that all tasks within an occupation are automatable. This is likely to overstate automation probabilities, given that many occupations will entail a mix of tasks, not all of which can be automated at the current level of technology.

14

As foreshadowed in the previous section, the likelihood of automation increases with the RTI level. Our estimates for the probability of automation line up with Autor, Levy, and Murnane’s (2003) task framework, with the probability of automation having a strong, statistically significant, positive relationship with the degree of routineness of work tasks and a statistically significant negative relationship with the abstract and manual components of the RTI index (See Table A3 in Annex III).

15

IMF (2018) finds that labor force participation in the 15–24 age bracket has fallen over the past few decades owing to higher returns from schooling.

16

Dauth and others (2017) show that, for the manufacturing sector in Germany, greater use of robots entrenches older workers into their jobs and lowers employment by reducing new-job creation.

17

Gender differences among older cohorts may reflect larger gender disparities in educational attainment in older cohorts.

18

The The Global Gender Gap Report 2017 (WEF, 2017) from the World Economic Forum shows that in only 27 of the 144 sample countries, the gender gap in education at the primary, secondary, and tertiary level is fully closed. However, there has been consistent progress globally in narrowing the gap over the past decades. In OECD countries, the gender gap in secondary and college education has, in fact, been reversed (OECD, 2015).

19

Deloitte (2017), for instance, finds that 80 percent of small businesses in the US are not fully utilizing digital technologies, and the biggest reason cited for their lack of use is not resource constraints, but a perceived irrelevance of technology to their work.

20

Theory and evidence indicates that large firms conduct more intensive searches for employees and provide more firm-specific human capital, which may result in less worker substitutability (Hu, 2003).

21

About 7 percent of the female workforce is employed in accommodation and food services relative to 5 percent of the male workforce. Retail trade employs 14 percent of the female and male workforce; it is the second-largest employer of males and females overall.

22

About 20 percent of the overall female workforce falls under this occupation, relative to 13 percent of the male workforce.

23

In combination with the PIAAC data, we use the International Adult Literacy Survey (IALS) survey, conducted between 1994 and 1998. We compute occupational shares for men and women in 17 OECD countries, for which IALS is available (see Annex I for country coverage).

24

The ILO produces employment estimates based on labor force surveys, using ISIC classifications that can be matched with sectors in the PIAAC data set. We employ the most recent estimates for 2017. We use the average probability of automation at the sectoral level in our sample and apply it to sectoral population estimates from the ILO in order to arrive at total jobs at risk across all sectors globally.

25

McKinsey (2016) points out that while the skill content of the work of bookkeepers and accountants is higher than a cook, the cost of automating the tasks of the former are significantly lower than the latter.

27

Burtch, Carnahan, and Greenwood (2018) show that the entry of Uber into local markets reduces low-quality entrepreneurial activity by providing alternate employment.

28

For instance, Zervas, Proserpio, and Byers (2017) find that the entry of Airbnb leads to revenue declines in the hotel industry, which will have implications for labor demand in the industry.

29

Emerging market economies show encouraging counter trends: with more than 260,000 female tertiary ICT graduates in 2015, India is the country closest to gender parity in this field, followed by Indonesia (OECD, 2018).

30

The Bureau of Labor Statistics in the US predicts that health care industries will account for the largest share of new jobs created during 2016–26.

31

See discussion and references in Kochhar, Jain-Chandra, and Newiak (2017).

32

Caregiver credits—pension credits provided to individuals who have spent time out of the workforce (e.g., in many EU countries)—are also increasingly used to bolster retirement security for women. However, care should be exercised in the use of benefits that can lower incentives to remain attached to the labor force, such as prolonged paid parental leave which can deteriorate future employment opportunities (IMF, 2012).

33

Some countries offer an option for joint or individual taxation (e.g., Germany; United States; IMF, 2017).

34

Gender gap in STEM fields grows with age: 15-year-old girls are two times less likely to aspire to a career as an engineer, a scientist, or an architect (OECD, 2018). Dasgupta and Stout (2014) and Dasgupta and Dennehy (2017) find that peer mentors, especially female mentors, are critically important in raising the retention rate of women in STEM fields. A growing number of countries are introducing explicit support for women in their education or child care to prevent them from dropping out of demanding careers in science and technology (e.g., Australia, Germany, Italy, Japa,) and tackling stereotypes in education (e.g., France).

35

IMF research finds that women accounted for less than 2 percent of financial institutions’ chief executive officers, and less than 20 percent of executive board members are women (Sahay, Čihák, and others, 2018). The presence of women as well as a higher share of women on bank boards, however, is associated with greater financial resilience.

36

Future work will delve deeper into differences between age groups in exposure to routine tasks and the threat of displacement on a sectoral, occupational, and country level.

37

We exclude the Russian sample because it is not representative for Moscow and the Moscow region. For Germany, we separately obtain wage data from the GESIS Leibniz Institute for the Social Sciences.

38

For routine component of RTI, we first perform a principal component analysis on variables describing flexibility and learning on the job separately. Using an inverse of the resulting flexibility and learning indices, along with the routine manual component, we perform principal component analysis again to construct a composite routine component.

39

The EM estimation is carried out for the US sample only, since Frey and Osbourne’s (2017) probabilities are calculated for US occupational descriptions. Once the model converges on the US sample, the estimated parameters are used to construct individual probabilities for the full sample, using their own individual sample characteristics

40

Austria, Ireland, and Singapore do not have information on payment schemes; Cyprus, France, Italy, and Spain do not collect data on problem solving skills, and Canada does not collect information on payment scheme or amount of experience needed for the job.

Gender, Technology, and the Future of Work
Author: Mariya Brussevich, Ms. Era Dabla-Norris, Christine Kamunge, Pooja Karnane, Salma Khalid, and Ms. Kalpana Kochhar