## EXECUTIVE SUMMARY

## Source details

**Canonical URL:** [EXECUTIVE SUMMARY](https://www.imf.org/-/media/files/publications/sdn/2018/sdn1807.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/sdn/2018/sdn1807.pdf.md)
- [Structured JSON version](/-/media/files/publications/sdn/2018/sdn1807.pdf.json)

---

### New landscape of work
- New technologies—digitalization, artificial intelligence, and machine learning—are changing the way work gets done at an unprecedented rate.
- Key questions addressed:
  - gender implications of changing work;
  - vulnerability of women’s jobs to displacement by technology;
  - policies to ensure technological change narrows rather than widens gender gaps.

### Main findings on routineness and gender gaps
- Data and methodology:
  - Individual-level task-composition data from the Program for the International Assessment of Adult Competencies (PIAAC) survey for adults 16 to 65 years of age, covering 28 OECD member countries, Cyprus, and Singapore.
  - Constructed a Routine Task Intensity (RTI) index at the individual level (index ranges from zero to one) following Autor and Dorn (2013) as modified by De La Rica and Gortazar (2016).
- Stylized facts:
  - Women, on average, perform more routine or codifiable tasks than men across all sectors and occupations—tasks more prone to automation.
  - The RTI index, on average, is 13 percent higher for female workers across the sample of 30 countries.
  - Women perform fewer tasks requiring analytical and interpersonal skills and more repetitive, low-flexibility tasks.
  - Cross-country heterogeneity: RTI index level for female workers ranges from 0.45 to 0.61 across countries, with exposure highest in eastern and southern European countries and lowest in Scandinavian and central European countries.
  - Selection of women into specific sectors and occupations explains most of the gender differences in RTI.
  - The gender gap in RTI and ICT use is lower in sectors and countries where female labor force participation (FLFP) is higher.

### Quantified risk of automation (current technology, next two decades)
- Sample of 30 countries (28 OECD member countries, Cyprus, and Singapore):
  - 10 percent of the male and female workforce (54 million workers) are at high risk (facing higher than 70 percent likelihood of being automated) of being displaced within the next two decades.
  - 11 percent of the female workforce versus 9 percent of the male workforce are at high risk.
  - 26 million female jobs in these 30 countries are potentially at stake.
  - Disproportionately exposed groups: less well-educated female workers; older female workers (aged 40 and older); female workers in low-skill clerical, service, and sales positions.
  - Extrapolation: about 180 million female jobs are at high risk of being displaced by automation globally.

### Sectoral and occupational patterns
- Sectors with largest exposure: accommodation and food services, retail trade, and transportation.
- Retail trade and accommodation and food services employ roughly similar proportions of male and female workers; transport is male-dominated.
- Women are overrepresented in education, health, and social services (34 percent of the female workforce relative to 11 percent of the male workforce).
- Within-sector gender gaps vary; for example, despite large female shares in education, women face higher likelihood of being automated than men in that sector.
- Elementary occupations are most exposed: nearly 40 percent of workers are at high risk of being displaced.
- Among professionals, women are twice as likely as men to be at high risk for automation.
- ICT sector: share of women at high-risk of displacement in ICT is significantly greater than share of men; 87 percent of men versus 72 percent of women occupy managerial and professional positions in ICT; 26 percent of women versus 7 percent of men are employed as clerks or service workers in ICT.
- Health and social services: lower likelihood of automation driven by smaller gender gaps in managerial and professional positions and a reversed manual index where women perform more non-routine manual work (e.g., nursing).

### Demographic and firm characteristics associated with automation risk
- Age:
  - 48 percent of women in the 16–19 age group fall into the high risk of automation category.
  - Older cohorts (older than 40) of working women are at significantly higher risk for automation than men in the same age cohorts.
- Education:
  - Risk of automation is decreasing in education, numeracy and literacy skills.
  - Risk is less than 1 percent among workers who have a bachelor’s degree or higher.
  - Nearly 50 percent of women with lower secondary education or less are at high risk, relative to less than 40 percent of men with the same education level.
- Firm size:
  - Workers in small and medium enterprises (SMEs) face a much higher risk of being automated compared to workers in large enterprises (greater than 1000 workers).
  - Female workers are significantly more likely than male workers to face risk of automation in SMEs, but significantly less likely to face risk of automation in large firms relative to male counterparts.

### Distributional magnitudes and cross-country heterogeneity
- Average probability of automation:
  - Women: 40 percent.
  - Men: 38 percent.
  - Difference: 2 percent higher for women (statistically significant).
- High-risk shares and counts:
  - 10 percent of all male and female jobs in the sample are at high risk (about 54 million workers in the sample).
  - 11 percent of female workers (or 26 million jobs) are at high risk.
  - 9 percent of male workers (or 28 million jobs) are at high risk.
- Cross-country examples:
  - Japan: 4 percent of the male workforce is at high risk vs 12 percent of the female workforce.
  - France, the UK, and the US: proportion at high risk is roughly similar between men and women.
  - Finland and Poland: the female workforce is significantly less exposed to automation than male counterparts.
- Global projection:
  - Applying sector-wise probabilities to ILO employment estimates for 2017 yields an estimate of 180 million female jobs at high risk of being automated, comprising more than 14 percent of the global female workforce.
  - Much of global automation-driven displacement projected to be in agriculture and retail trade (more than 50 million female jobs potentially displaced in each), followed by accommodation and food services.

### Opportunities and challenges
- Opportunities:
  - Job growth likely in traditionally female-dominated sectors such as health, education, and social services—jobs requiring cognitive and interpersonal skills and thus less prone to automation.
  - Digital transformation may confer greater flexibility in work, potentially benefiting women.
- Challenges:
  - Women are underrepresented in STEM sectors anticipating job growth, such as engineering and ICT.
  - Underrepresentation of women in professional and managerial positions across sectors places them at higher risk of displacement by technology.
  - New jobs created by technology and higher output could offset job losses, but outcomes are uncertain.
  - Gig economy and platform-mediated flexible work arrangements are not captured in the micro-data and could alter outcomes for women.

### Policy recommendations (crucial role for policy)
- Endowing women with requisite skills:
  - Early investment in women in STEM fields and peer mentoring to break down gender stereotypes and increase retention.
  - Fiscal instruments for those already in the workforce (e.g., tax deductions for training, portable individual learning accounts) to remove barriers to lifelong learning.
- Closing gender gaps in leadership positions:
  - Family-friendly policies to boost retention and career progression.
  - Consider setting recruitment and retention targets for organizations, promotion quotas, and mentorship and training programs to promote female talent into managerial positions.
- Bridging the digital divide:
  - Public investment in capital infrastructure and ensuring equal access to finance and connectivity to close gender gaps in access to and use of new technologies.
- Easing transitions for workers:
  - Ensure gender equality in support for displaced workers through active labor market policies.
  - Link training and benefits to individuals rather than jobs to improve reemployment prospects.
  - Adapt social protection systems to new forms of work, including portability of benefits and consideration of noncontributory schemes where appropriate.
- Other targeted instruments:
  - Publicly subsidized vouchers, portable individual learning accounts, preferential loans, tax deductions, and payroll taxes dedicated to subsidizing training are cited as mechanisms to encourage lifelong learning and retraining, with consideration to target retraining incentives to sectors where women face large skill gaps.

### Caveats and limitations
- Estimates consider only technical feasibility given the current state of technology; depth and speed of future technological advancement and diffusion are unknown.
- Jobs at risk of automation is not the same as actual job loss; potential offsetting by new work opportunities is possible.
- Analysis does not capture the "gig" economy due to data constraints; flexible work arrangements intermediated by digital platforms could alter outcomes for women.
- Extrapolations beyond the micro-data sample assume immediate transferability of technology across economies; technological diffusion is heterogeneous and results should be interpreted with caution.

### Methodological notes (RTI, ICT indices, probability of automation)
- RTI index construction:
  - RTI index components: abstract, manual, and routine, constructed using PIAAC questionnaire items and principal component analysis.
  - RTI formula: RRTRI_i = ROUTINE_i − ABSTRACT_i − MANUAL_i, with the final index standardized between zero and unity.
- Probability of automation estimation:
  - Start from Frey and Osbourne (2017) occupation-level probabilities (0 to 100 percent) and map to individuals using an Expectation-Maximization (EM) algorithm and PIAAC individual characteristics (age, education, literacy and numeracy skills, task characteristics).
  - EM estimation carried out on the US sample; estimated parameters then applied to the full sample to predict individual-level probabilities.
- Key empirical relationships (preserved coefficients and signs as reported):
  - RTI Index correlates positively with probability of automation (0.05***).
  - Routine Index: 0.02***; Abstract Index: -0.05***; Manual Index: -0.01*** (Table A3, N 42945).
  - Selected predictors of higher predicted probability of automation include: Female 0.04***; Fingers and hand use 1.04***; Selling 2.96***; Exchanging information 0.74***.
  - Selected predictors of lower predicted probability of automation include: Education high (ISCED 5A, 6) -0.46***; Solving complex problems -1.39***; Using programming language -4.12***; Presenting -4.96***; Reading books -4.66***.
  - Model sample size for selected coefficient table: N 2065.
- Data coverage and limitations:
  - PIAAC surveys administered in two rounds between 2011 and 2016; sample includes 30 countries and adults aged 16 to 65.
  - Missing regressors in certain country samples restrict some model variants (e.g., Austria, Ireland, Singapore lack payment scheme information; Cyprus, France, Italy, Spain lack problem solving skills; Canada lacks payment scheme and experience-required variables).

*Source: EXECUTIVE SUMMARY, sdn1807*

### EXECUTIVE SUMMARY  ____________________________________________________________________________ 4

### EXECUTIVE SUMMARY

### New landscape of work
- New technologies—digitalization, artificial intelligence, and machine learning—are changing the way work gets done at an unprecedented rate.
- Key questions addressed: gender implications of changing work; vulnerability of women’s jobs to displacement by technology; policies to ensure technological change narrows rather than widens gender gaps.

### Main findings on routineness and gender gaps
- Data and methodology:
  - Individual-level task-composition data from the Program for the International Assessment of Adult Competencies (PIAAC) survey for adults 16 to 65 years of age, covering 28 OECD member countries, Cyprus, and Singapore.
  - Constructed a Routine Task Intensity (RTI) index at the individual level (index ranges from zero to one) following Autor and Dorn (2013) as modified by De La Rica and Gortazar (2016).
- Stylized facts:
  - Women, on average, perform more routine or codifiable tasks than men across all sectors and occupations—tasks more prone to automation.
  - The RTI index, on average, is 13 percent higher for female workers across the sample of 30 countries.
  - Women perform fewer tasks requiring analytical and interpersonal skills and more repetitive, low-flexibility tasks.
  - Cross-country heterogeneity is significant: RTI index level for female workers ranges from 0.45 to 0.61 across countries, with exposure highest in eastern and southern European countries and lowest in Scandinavian and central European countries.
  - The selection of women into specific sectors and occupations explains most of the gender differences in RTI.
  - The gender gap in RTI and ICT use is lower in sectors and countries where female labor force participation (FLFP) is higher.

### Quantified risk of automation (current technology, next two decades)
- Sample of 30 countries (28 OECD member countries, Cyprus, and Singapore):
  - 10 percent of the male and female workforce (54 million workers) are at high risk (facing higher than 70 percent likelihood of being automated) of being displaced within the next two decades.
  - A larger proportion of the female workforce is at high risk than the male workforce: 11 percent of the female workforce versus 9 percent of the male workforce.
  - 26 million female jobs in these 30 countries are potentially at stake.
  - Disproportionately exposed groups: less well-educated female workers; older female workers (aged 40 and older); female workers in low-skill clerical, service, and sales positions.
  - Extrapolation: about 180 million female jobs are at high risk of being displaced by automation globally.

### Opportunities and challenges
- Opportunities:
  - Job growth likely in traditionally female-dominated sectors such as health, education, and social services—jobs requiring cognitive and interpersonal skills and thus less prone to automation.
  - Digital transformation may confer greater flexibility in work, potentially benefiting women.
- Challenges:
  - Women are underrepresented in STEM sectors anticipating jobs growth, such as engineering and ICT.
  - Underrepresentation of women in professional and managerial positions across sectors places them at higher risk of displacement by technology.
  - New jobs created by technology and higher output could offset job losses, but outcomes are uncertain.

### Policy recommendations (crucial role for policy)
- Endowing women with requisite skills:
  - Early investment in women in STEM fields and peer mentoring to break down gender stereotypes and increase retention.
  - Fiscal instruments for those already in the workforce (e.g., tax deductions for training, portable individual learning accounts) to remove barriers to lifelong learning.
- Closing gender gaps in leadership positions:
  - Family-friendly policies to boost retention and career progression.
  - Consider setting recruitment and retention targets for organizations, promotion quotas, and mentorship and training programs to promote female talent into managerial positions.
- Bridging the digital divide:
  - Public investment in capital infrastructure and ensuring equal access to finance and connectivity to close gender gaps in access to and use of new technologies.
- Easing transitions for workers:
  - Ensure gender equality in support for displaced workers through active labor market policies.
  - Link training and benefits to individuals rather than jobs to improve reemployment prospects.
  - Adapt social protection systems to new forms of work.

### Caveats and limitations
- Estimates consider only technical feasibility given the current state of technology; depth and speed of future technological advancement and diffusion are unknown.
- Jobs at risk of automation is not the same as actual job loss; potential offsetting by new work opportunities is possible.
- Analysis does not capture the "gig" economy due to data constraints; flexible work arrangements intermediated by digital platforms could alter outcomes for women.

*Source: EXECUTIVE SUMMARY, sdn1807*

### 12.      Gender RTI and ICT gaps and female labor force participation. The gender RTI gap―the

### 12.      Gender RTI and ICT gaps and female labor force participation.

### Relationship between gender RTI/ICT gaps and female labor force participation (FLFP)
- The gender RTI gap—the ratio of the female RTI level to the male RTI level—is correlated with FLFP: gender RTI gaps are smaller in countries with higher FLFP.
- Turkey is an outlier, exhibiting both the lowest FLFP and the lowest gender RTI gap.
- Gender gaps in ICT use (ICT use index measures level and frequency of computer use at work) are inversely related to FLFP, suggesting that in countries with low FLFP women entering the labor force are more likely to work in jobs that are more intensive in ICT use.
- Note: The gender ICT gap is defined as the ratio of the male ICT use index to the female ICT use index. Gender differences in ICT use are not statistically significant in Greece, Italy, Lithuania, and Turkey.

### Decomposition of the gender RTI gap: drivers and intensive margin
- Decomposition results show:
  - Nearly 13 percent of the unconditional gender RTI gap is explained by occupational choice.
  - The intensive margin (type of tasks women perform and their job positions within occupations) is the main driver of gender disparities in routineness in the workplace.
- Methodology note: decomposition based on individual-level regression; bars indicate share of unconditional RTI gap explained by given sets of variables. Statistical significance levels indicated: *** p<0.01; ** p<0.05; * p<0.1.

### Gender RTI gaps across occupations and sectors
- Gender RTI gaps are persistent across all occupations and increase with the average routineness level of the occupation.
- Clerks and elementary occupations are among the most routine occupations and make up a large share of women’s overall employment, indicating higher potential exposure of women to automation due to occupational choices.
- Sectoral patterns:
  - Retail trade and services employ a significant proportion of the female labor force and have a large gender RTI gap.
  - Sectors that employ large proportions of the female labor force (e.g., health and education) have lower gender RTI gaps, suggesting greater participation by women on the extensive margin in a sector lowers relative exposure to automation.
- Note: occupation categories based on ISCO-08; sector categories based on ISIC rev. 4. Gender differences in RTI are statistically significant at 1 percent level for all occupations.

### Job routineness and gender wage gaps
- Findings:
  - Sectoral differences and other individual and job characteristics (education, experience, presence of children) explain a large share of existing gender wage gaps.
  - Job routineness also matters: RTI differences account for nearly 5 percent of the average unconditional gender wage gap, beyond other factors.
  - Example: In the US, a 5 percent wage gap driven by RTI differences would translate into $26,000 in forfeited lifetime income.
- Methodology note: wage decomposition estimated at the individual level; statistical significance levels reported.

### Quantifying the risk of automation for women: estimates and methods
- Approach:
  - Start with occupational-level estimates from Frey and Osbourne (2017) and project them onto individual worker characteristics (age, education, gender, literacy and numeracy skills, task characteristics) using PIAAC data and an expectation-maximization (EM) algorithm.
  - This yields individual-level probabilities of automation that incorporate demographics, skills, and job tasks rather than assuming identical tasks within occupations.
- Relationship: probability of automation has a strong, statistically significant, positive relationship with RTI and a statistically significant negative relationship with abstract and manual components of the RTI index.

### Higher probability of automation for women: magnitudes
- Average probability of automation:
  - Women: 40 percent.
  - Men: 38 percent.
  - Difference: 2 percent higher for women (statistically significant).
- High-risk definition: probability of automation >= 0.7 (>= 70 percent).
- High-risk shares and counts:
  - 10 percent of all male and female jobs in the sample are at high risk (about 54 million workers in the sample).
  - 11 percent of female workers (or 26 million jobs) are at high risk.
  - 9 percent of male workers (or 28 million jobs) are at high risk.
- Differences in probability of automation and share of workers with high automatability across gender are statistically significant at 1 percent level.

### Age, education, firm size, and automation risk
- Age:
  - Younger people are at the highest risk: 48 percent of women in the 16–19 age group fall into the high risk of automation category.
  - Women aged 20 to 29 have moderately lower odds of displacement than men in the same bracket (differences not statistically significant).
  - Older cohorts (older than 40) of working women are at significantly higher risk for automation than men in the same age cohorts.
- Education:
  - Risk of automation is decreasing in education, numeracy and literacy skills.
  - Risk is less than 1 percent among workers who have a bachelor’s degree or higher.
  - Nearly 50 percent of women with lower secondary education or less are at high risk, relative to less than 40 percent of men with the same education level.
- Firm size:
  - Workers in small and medium enterprises (SMEs) face a much higher risk of being automated compared to workers in large enterprises (greater than 1000 workers).
  - Female workers are significantly more likely than male workers to face risk of automation in SMEs, but significantly less likely to face risk of automation in large firms relative to male counterparts.

### Cross-country heterogeneity
- Proportion at high risk is roughly similar between men and women in France, the UK, and the US (large service-dominated economies).
- Countries with small or statistically insignificant gender differences in exposure: Belgium, Denmark, Germany, Slovakia, Sweden.
- Examples of notable heterogeneity:
  - Japan: 4 percent of the male workforce is at high risk vs 12 percent of the female workforce.
  - In some countries (Austria, Cyprus, Israel, Korea) women are disproportionately exposed to high risk of automation.
  - In Finland and Poland the female workforce is significantly less exposed to automation than male counterparts.

*Italic: Source: "12.      Gender RTI and ICT gaps and female labor force participation." (SDN1807 chapter content).*

### 21.      Sectoral variation in likelihood of automation. Overall, our analysis suggests the

### 21.      Sectoral variation in likelihood of automation. Overall, our analysis suggests the

### Sectoral exposure and gender distribution
- Accommodation and food services, retail trade, and transportation sectors have the largest exposure to risk of automation (Figure 10, left panel).
- Retail trade and accommodation and food services employ roughly similar proportions of the male and female workforce.
- Men are disproportionately represented in the transport sector and therefore more exposed to risk of automation owing to their sectoral choice.
- Women are vastly overrepresented in education, health, and social services, with 34 percent of the female workforce working in these sectors relative to only 11 percent of the male workforce.
- Within-sector gender gaps in risk of automation vary; despite the large share of female workers in education, women face a higher likelihood of being automated than men in that sector.
- Footnote statistics preserved:
  - 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.

### Occupation and job-characteristic drivers of gender gaps
- Differences in occupation choice and job characteristics within occupations drive gender gaps in automation threat.
- Elementary occupations are most exposed: nearly 40 percent of workers are at high risk of being displaced (Figure 11).
- Roughly similar proportions of male and female workforce are employed in elementary occupations, with no statistically significant difference in numbers at risk.
- Women are overrepresented among service workers, and female service workers face a higher risk of automation than male counterparts.
- Legislators, managers, and professionals: less than 1 percent of the workforce in these occupational categories are at high risk of automation.
- Among professionals, women are twice as likely as men to be at high risk for automation.
- In retail trade (overall very high risk), female workers are significantly less likely than male workers to perform abstract tasks, reflecting fewer women in managerial positions.
- Footnote statistic preserved:
  - About 20 percent of the overall female workforce falls under this occupation, relative to 13 percent of the male workforce.

### Recent occupational shifts and implications
- Between 1994 and 2016 (sample of OECD countries), most job growth has been on the high-skill end; women have benefited more than men.
- Occupational shifts observed: away from clerical occupations towards service and retail workers, technicians, and professionals—a trend more pronounced for women.
- Women are increasingly opting into occupations relatively more insulated from automation risk; increases in managerial and professional roles are notable.
- Educational trends cited:
  - In 1994, slightly more than 60 percent of male and female graduates in the United States were enrolled in university.
  - By 2014, the figure for women had jumped to 71 percent, while that for men was broadly unchanged.

### Global projection of jobs at high risk
- Method: apply sector-wise probability of being at high risk of automation to ILO employment estimates by sector (most recent estimates for 2017).
- Estimate: 180 million female jobs are at high risk of being automated, comprising more than 14 percent of the global female workforce.
- At the global level, much of this automation will be driven by agriculture and retail trade, with more than 50 million female jobs being potentially displaced in each, followed by accommodation and food services.
- Caveat: extrapolation assumes immediate transferability of technology across economies; technological diffusion is heterogeneous—exercise caution in interpreting results beyond micro-data sample.

### Caveats, technological progress, and the gig economy
- Estimates assume current levels of technology and prevailing bottlenecks in computer-controlled equipment; thus analysis presents a lower bound for potential impact.
- Further computing improvements could automate more tasks than current predictions.
- Estimates are based on technological feasibility, not economic feasibility—automation may be possible but prohibitively costly relative to prevailing labor costs.
- Reference to alternative findings:
  - Hawksworth, Berriman, and Goel (2018) find female workers could be more affected by automation over the next decade, but male jobs could be at higher risk in the longer term.
  - Results are robust to alternate occupation-level automation measures by Brynjolfsson, Mitchell, and Rock (2018).
- Gig economy:
  - Gig work can provide additional income avenues on an on-demand or short-term basis.
  - Platforms facilitating delivery may reallocate employment across tasks and sectors, with ambiguous implications for women.
  - Analysis of the gig economy is crucial for comprehensive assessment of future-of-work opportunities and challenges.

### Sector deep-dives: ICT and Health & Social Services
- ICT sector:
  - Share of women at high-risk of displacement in ICT is significantly greater than share of men (Figure 13, top panel).
  - Men more often occupy managerial and professional positions in ICT: 87 percent for men versus 72 percent for women—positions likely to entail abstract tasks.
  - Nearly four times as many women employed as clerks or service workers in ICT: 26 percent for women versus 7 percent for men.
  - Women make up less than 20 percent of tertiary school graduates in computer science in OECD countries (OECD, 2018).
  - Implication: persistent gender differences in field of study may limit women's benefit from new STEM job opportunities.
- Health and social services:
  - Lower likelihood of automation is driven by smaller gender gaps in managerial and professional positions and smaller gaps in abstract and manual tasks (Figure 13, bottom panel).
  - Manual index gap is reversed: women perform more non-routine manual work (e.g., nursing).
  - With aging populations in advanced economies, demand for health and social care services will increase.
  - Coping with aging will require more human workers and greater use of AI, robotics, and advanced technologies to complement and boost productivity.
  - Workers in these sectors will need digital and softer skills.

### Policy conclusions and options
- New technologies are likely to be disruptive for women’s labor market outcomes; some jobs will be displaced or fundamentally changed, but many new jobs will be created.
- Best policy approaches vary across countries depending on development level, existing FLFP, and speed of technological impact.
- General policy focus: reduce workplace barriers for women and foster gender parity in education, training, access to technology, and support for displaced workers.
- Bringing more women into the workforce:
  - Policies and infrastructure that help reconcile work and family life (e.g., leave policies that allow women to retain positions at work; availability of high quality, affordable child care; flexible work arrangements) are particularly effective.
  - Female labor supply is more responsive than men's to specific tax policies; replacing family taxation with individual taxation and providing tax relief for low-income families can increase women's employment rates.
- Empowering women in the workplace:
  - Invest in human capital—skills provide the most important safeguard against displacement and enable benefit from new opportunities.
  - Demand is rising for advanced cognitive skills (critical thinking, problem solving, learning agility) and socio-emotional skills (creativity, curiosity, adaptability); demand for narrow job-specific skills is waning.
  - Urgent need to adapt and reform education systems and workforce training to reduce skill-mismatches and remove barriers to lifelong learning.
  - Employers should play a more active role in education and training, including divulging information on future skills demand and providing better learning opportunities.

*Italicized source attribution preserved from original content unit.*

### 32.      Building skills early. Raising the quality and quantity of the human capital of new labor

### 32.      Building skills early. Raising the quality and quantity of the human capital of new labor market entrants will be important

### Early investment in education and STEM
- Ensuring universal and high quality basic and secondary education is critical to help individuals adapt to new technologies (Autor, 2013).
- Balance needed between:
  - endowing women with the necessary tools in specialized STEM fields, where they are at a relative disadvantage, and
  - providing a general curriculum that emphasizes critical thinking, decision-making, problem-solving skills, and empathy (Deming, 2015).
- Early investment measures highlighted:
  - targeted early investment in women in STEM fields with a digital and quantitative focus and peer mentoring to help break down gender stereotypes and increase retention (Dasgupta and Stout, 2014; Dasgupta and Dennehy, 2017; OECD, 2018).
- Key empirical finding:
  - the risk of automation is less than 1 percent among female workers who have a bachelor degree or higher, compared with 50 percent among women with lower secondary degrees—highlighting the importance of higher education.
- Tertiary attainment:
  - Across most advanced economies, more women than men are now tertiary graduates; ensuring skill acquisition and long-term skill matching in tertiary education remains important.

### Encouraging lifelong learning (policy instruments and challenges)
- Current adult education systems often reinforce gender and economic disparities: more reskilling/upskilling by more educated, high-income workers with digital literacy and internet access.
- Women, typically more exposed to automation risk, often:
  - work in occupations with lower on-the-job learning opportunities, and
  - are the least likely to participate in training.
- Fiscal and incentive instruments to promote lifelong learning:
  - publicly subsidized vouchers to ease access to formal education,
  - portable individual learning accounts (example: France),
  - preferential loans,
  - tax deductions (example: Netherlands),
  - tax benefits and incentives for private-sector human capital investment (e.g., payroll taxes dedicated to subsidizing training opportunities, public grants for subsidizing training).
- Recommendation: consider targeting retraining tax incentives to particular sectors where women continue to face large skill gaps relative to men.

### Fostering gender parity in management positions
- Progress and gaps:
  - Women are, on average, more educated than men in most advanced economies and participate more fully in professional and technical occupations than two decades ago.
  - Women’s chances to rise to positions of leadership are only 28 percent of those of men (WEF, 2017b).
- Economic impact:
  - Greater involvement of women in senior management has been found to strengthen economic performance (Cuberes, Newiak, and Teignier, 2017; Sahay, Čihák, and others, 2018).
  - Payoffs from gender diversity are significantly higher in high-tech and knowledge-intensive sectors.
- Policy options to retain and promote female talent:
  - family-friendly policies (discussed elsewhere in the source) to boost retention and career progression,
  - setting recruitment and retention targets for organizations,
  - mentorship and training programs to promote female talent into managerial positions,
  - mandatory gender quotas on hiring and promotion of women in public and private sectors could be considered.
- Example:
  - In Norway, following quota legislation for corporate boards in 2003, women representation on all company boards increased to about 40 percent (OECD, 2016).
- Financial sector statistic:
  - IMF research finds 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 and a higher share of women on bank boards is associated with greater financial resilience.

### Bridging the digital gender divide (access and infrastructure)
- Digital access and gender gaps (ITU, 2017):
  - 60 percent of the global population, most of them women in emerging and developing economies, still have no access to the internet;
  - 250 million fewer women are online than men;
  - 200 million fewer women than men own a mobile phone across developing countries.
- Consequences:
  - Lack of digital infrastructure is particularly detrimental for women’s labor market outcomes.
  - Digital technologies and flexible work can boost women’s employment rates and foster more gender-balanced career paths, but only if access is extended.
- Policy and investment recommendations:
  - Ensure equal access to finance to overcome affordability barriers for women,
  - Investment (public, private, or public-private partnerships) to support technological adoption and close digital gender gaps,
  - National connectivity policies should apply a gender lens (example: Finland defined access to the internet at broadband speeds as a legal right and pursued universal access).

### Easing transitions for workers (adjustment policies and social protection)
- Smoothing adjustment amid technological change:
  - Older female workers are at disproportionately higher risk of displacement than male counterparts; ensuring gender equality in support for displaced workers is essential.
  - Short-term active labor market policies can provide direct support:
    - job search assistance,
    - retraining,
    - income or geographical mobility support,
    - employment incentives such as hiring and wage subsidies (IMF, 2018).
  - Potential bias risk: policies may be inadvertently biased against women if schemes do not support sectors where women dominate (OECD, 2017).
  - Recommendation: link training and benefits to individuals rather than jobs to improve reemployment prospects for both men and women.
  - The tax/benefit system can play an important redistributive role when needed.
- Adapting social protection to nonstandard work:
  - Traditional social insurance eligibility conditional on contribution records disadvantages nonstandard workers (and many women).
  - Tax and benefit systems need revamping to close coverage gaps and allow portability of benefits to prevent loss of entitlements when workers move between jobs.
  - Several countries are testing basic income guarantees and other noncontributory schemes (examples in the source: expanding social pensions and earned income tax credit) to address income decline and uncertainty from automation—measures that could be particularly beneficial for older women at high displacement risk.

### Data, indices, and empirical methods (RTI and ICT use)
- Data coverage:
  - The Programme for the International Assessment of Adult Competencies (PIAAC) surveys were administered in two rounds between 2011 and 2016. The sample includes 30 countries for which data are available.
  - The survey covers adults between the ages of 16 to 65 and collects detailed demographic and work information, plus numeracy, literacy, and problem-solving skills used as proxies for workers’ ability.
  - For a shift-share analysis, the International Adult Literacy Survey (IALS) (conducted between 1994 and 1998 for 22 countries) was linked to PIAAC trend data.
- RTI and ICT indices construction:
  - RTI index components: abstract, manual, and routine, constructed using PIAAC questionnaire items and principal component analysis.
  - Components described:
    - Abstract: analytical and interpersonal tasks (e.g., writing reports, solving complex problems, negotiating).
    - Manual: two types—routine hand and finger dexterity tasks and nonroutine physical work (physical work for long hours).
    - Routine: lack of flexibility and learning indicators (inverse measures indicate repetitiveness and rule-based tasks).
    - ICT use: use internet for work-related understanding, conduct transactions on the internet, use spreadsheet software, use a programming language, level of computer use.
  - RTI index formula (as given in the source) and standardization:
    - RRTRI_i = ROUTINE_i − ABSTRACT_i − MANUAL_i (equation (1) in the source), with the final index standardized between zero and unity.
- Empirical methods and wage decomposition:
  - Regression specifications estimate the contribution of individual and job characteristics to the observed gender RTI gap and to gender wage gaps (equations (2) and (3) in the source).
  - Controls used include age, level of education, presence of partner and children, immigrant status, numeracy and literacy test scores, experience, on-the-job training, part-time status, occupation fixed effects, sector fixed effects, and country fixed effects.
  - Standard errors are clustered at the country level.
  - The decomposition follows Gelbach (2012) to attribute changes in the unconditional gender gap to control variables.

*Source: IMF staff chapter "32. Building skills early. Raising the quality and quantity of the human capital of new labor market entrants will be important."*

### ANNEX III. Estimating Probability of Automation

### ANNEX III. Estimating Probability of Automation

### Methodology: linking occupation-level automatability to individual workers
- Follow the method of Arntz, Gregory, and Zierahn (2017) to impute occupation-based probabilities of automation to individual workers using worker and job task characteristics.
- Source probabilities: Frey and Osborne (2017) provide probabilities of automation for 702 occupations on a continuous scale between 0 and 100 percent, derived from O*NET task descriptions and a two-stage process:
  - Stage 1: hand-label a subset of occupations as fully automatable using Machine Learning researchers' judgments.
  - Stage 2: probabilistic model imputes automation probabilities to the full occupation sample using nine specific job task characteristics deemed bottlenecks for automatability.
- Mapping to individuals:
  - PIAAC contains only 2-digit ISCO codes, so one PIAAC individual may map to multiple Frey and Osborne occupational probabilities.
  - Use the Expectation Maximization (EM) algorithm to estimate an individual-level regression:
    - PPAAR(ARRRRF)_{ij} = sum_i β_i X_{iij} + ε_{ij}  (equation structure as presented)
    - i indexes individuals; j indexes duplicates when multiple occupation probabilities map to one individual; X contains individual, job, and task characteristics.
  - Estimation approach:
    - Weighted Generalized Linear Model (GLM) with equal initial weights for duplicates j of individual i.
    - Iteratively update weights following Ibrahim (1990): w_{ij} = f(y_hat_hata - y_{ij} | x_{iij}, β_i) / sum(...) where f(.) is the standard normal density (expression preserved as in source).
    - Once weights converge and best fit achieved, use estimated parameters β_i to compute predicted individual probabilities of automation based on worker and job task characteristics.
  - EM estimation is carried out for the US sample only (Frey and Osborne probabilities are based on US occupational descriptions). After convergence on the US sample, the estimated parameters are applied to construct individual probabilities for the full sample using each individual’s own characteristics.
- Estimation variations:
  - Four models estimated with minor variations in the set of regressors contingent on data availability (Table A4 contains results for the US data).
  - Data limitations affecting regressors: Austria, Ireland, and Singapore lack information on payment schemes; Cyprus, France, Italy, and Spain do not collect problem solving skills; Canada does not collect payment scheme or amount of experience required for the job.

### Empirical relationships: RTI components and automatability (Table A3)
- Correlations of RTI Components and Probability of Automation:
  - RTI Index: 0.05*** (standard error (0.00) as shown)
  - Routine Index: 0.02*** (standard error (0.00))
  - Abstract Index: -0.05*** (standard error (0.00))
  - Manual Index: -0.01*** (standard error (0.00))
- Sample size:
  - N 42945

### Estimated predictors of individual probability of automation (Table A4)
- Model sample size:
  - N 2065
- Selected estimated coefficients (variable — coefficient with significance stars preserved exactly as presented):
  - Female: 0.04***
  - Age (20–24): -0.06***
  - Age (25–29): 0.08***
  - Age (30–34): -0.06***
  - Age (35–39): -0.01***
  - Age (40–44): 0.04***
  - Age (45–49): 0.03***
  - Age (50–54): 0.03***
  - Age (55–59): 0.02***
  - Age (60–65): 0.15***
  - Education medium (ISCED 3, 4, 5B): -0.25***
  - Education high (ISCED 5A, 6): -0.46***
  - Numeracy skills: 0.00***
  - Literacy skills: -0.00***
  - Problem-solving skills: -0.00***
  - Solving simple problems: -0.92***
  - Solving complex problems: -1.39***
  - Work physically for long: -0.74***
  - Fingers and hand use: 1.04***
  - Reading instructions: -1.66***
  - Reading professional publications: -3.92***
  - Reading books: -4.66***
  - Reading manuals: 0.12***
  - Writing articles: -4.30***
  - Filling forms: -1.14***
  - Calculating shares: -0.46***
  - Complex math or statistics: -1.21***
  - Using internet for work: -1.08***
  - Using programming language: -4.12***
  - Using communication software: -1.03***
  - Not challenged enough at work: -0.14***
  - Need more training: 0.06***
  - Computer use at work: 0.01***
  - Cooperating with others at work: -0.00***
  - Exchanging information: 0.74***
  - Training others: -2.87***
  - Presenting: -4.96***
  - Selling: 2.96***
  - Consulting: 0.39***
  - Planning own activities: 0.49***
  - Planning activities of others: -2.13***
  - Organizing own schedule: -1.23***
  - Influencing: -4.43***
  - Negotiating: 0.22***
  - Sector (private=0; public/non-profit=1): -0.16***
  - Firm size (11–1000): 0.10***
  - FirmsSize (>1000): 0.01***
  - Responsibility for managing staff: -0.12***
  - Educational Requirements of Job: -0.46***
  - Experience Required in Job: -0.04***
  - Pay Scheme: -0.07***
  - Income decile 2: 0.18***
  - Income decile 3: 0.16***
  - Income decile 4: 0.08***
  - Income decile 5: 0.09***
  - Income decile 6: 0.04***
  - Income decile 7: 0.03***
  - Income decile 8: -0.06***
  - Income decile 9: -0.32***
  - Income decile 10: -0.33***
  - _cons: 1.18***
- Interpretation from coefficients (preserve direction and magnitude as reported):
  - Positive coefficients indicate association with higher predicted probability of automation (e.g., Fingers and hand use 1.04***; Exchanging information 0.74***; Selling 2.96***).
  - Negative coefficients indicate association with lower predicted probability of automation (e.g., Reading books -4.66***; Presenting -4.96***; Using programming language -4.12***; Influencing -4.43***).
  - Education and higher skill/task measures generally associated with lower predicted probability (Education high -0.46***; Solving complex problems -1.39***; Complex math or statistics -1.21***).
  - Routine task intensity correlates positively with probability of automation (RTI Index 0.05*** in Table A3).

### Implementation notes and data caveats
- EM estimation executed on US sample only because Frey and Osborne (2017) probabilities are based on US occupational descriptions; after convergence, parameters are applied across the full sample to predict individual-level probabilities using each individual’s characteristics.
- Missing regressors in some country samples restrict the set of regressors used in some model variants:
  - Austria, Ireland, Singapore: no payment scheme information.
  - Cyprus, France, Italy, Spain: no problem solving skills.
  - Canada: no payment scheme or amount of experience required for the job.

*Source: ANNEX III. Estimating Probability of Automation, sdn1807 — International Monetary Fund*

---


_Source: https://www.imf.org/-/media/files/publications/sdn/2018/sdn1807.pdf_