## 1. Relationship Between Female Labor Force Participation and Size of Manufacturing Sector

## Source details

**Canonical URL:** [1. Relationship Between Female Labor Force Participation and Size of Manufacturing Sector](https://www.imf.org/-/media/files/publications/wp/2019/wpiea2019091.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2019/wpiea2019091.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2019/wpiea2019091.pdf.json)

---

### I. Introduction — research focus and headline findings
- Research question: gender implications of digitalization, artificial intelligence, and machine-learning for job displacement using a task-based approach and individual-level PIAAC data for 30 advanced and emerging economies.
- Methodology:
  - Individual-level RTI (routine task intensity) constructed from PIAAC task composition, skills, and demographics.
  - Individual-level probabilities of automation derived by mapping occupation-level estimates (Frey and Osborne, 2017) to worker characteristics using an Expectation Maximization (EM) approach similar to Arntz, Gregory, and Zierahn (2017).
- Major empirical findings:
  - Women, on average, perform more routine or codifiable tasks than men across sectors and occupations.
  - Women perform fewer analytical/abstract tasks that are complementary to technological change.
  - Self-selection into occupations explains most gender differences in task composition, but cross-country heterogeneity is substantial.
  - Within the sample, 54 million male and female workers in 30 countries (28 OECD member countries, Cyprus, and Singapore) are at a high risk of being displaced by technology (i.e., facing higher than 70 percent likelihood of being automated) within the next two decades.
  - Sample-wide shares at high risk (probability >= 0.7): Female 11 percent; Male 9 percent.
  - Estimated number of female jobs at high risk within next two decades: 26 million female jobs in the 30-country sample.
  - Vulnerable groups: less well-educated and older female workers (aged 40 and older), and those in clerical, service, and sales positions.

### II. Measuring routineness (RTI index) — construction and components
- RTI definition (individual i):
  - RTI_i = Routine_i − Abstract_i − Manual_i
  - RTI index ranges from zero to one, with higher values indicating more routine activities.
- Components:
  - Abstract: analytical and interpersonal tasks (e.g., writing reports, solving complex problems, negotiating).
  - Routine: lack of job flexibility and little learning on the job (repetitiveness, codifiability).
  - Manual: two manual task types from PIAAC (hand and finger dexterity; long physical work) classified into routine and non-routine manual per prior literature.
- Estimation:
  - Principal component analysis (PCA) used to derive indices for subcomponents; components standardized to lie between zero and one; final RTI standardized to lie between zero and one.

### III. Gender RTI gap — magnitudes and cross-country patterns
- Sample average RTI components (Table 2):
  - RTI Level: Female 0.52; Male 0.46.
  - Abstract: Female 0.48; Male 0.56.
  - Manual: Female 0.40; Male 0.47.
  - Routine: Female 0.51; Male 0.47.
  - All t-test p-values reported as 0.00.
- Cross-country heterogeneity and correlates:
  - Female RTI level highest in Eastern and Southern European countries, lowest in Scandinavian and Central European countries.
  - Example: RTI for female workers in Lithuania is 36 percent higher than in Norway.
  - Gender RTI gap highest in Japan, the Slovak Republic, Singapore, and Estonia.
  - Gender RTI gaps are smaller in countries with higher female labor force participation (FLFP).
  - Gender RTI gaps are positively associated with the size of manufacturing sector (as a percentage of GDP).
- Selected country RTI and FLFP entries (Table A2):
  - Austria — FLFP 70.56; RTI Level (Female) 0.51; RTI Gap 1.16; T-test 0.00
  - Belgium — 61.11; 0.50; 1.12; 0.00
  - Canada — 74.18; 0.53; 1.14; 0.00
  - Japan — 63.64; 0.52; 1.21; 0.00
  - Sweden — 77.60; 0.48; 1.07; 0.00
  - Turkey — 32.19; 0.53; 1.05; 0.00
  - United Kingdom — 70.00; 0.51; 1.10; 0.08
  - United States — 66.36; 0.50; 1.14; 0.00

### IV. Decomposition of the gender RTI gap
- Econometric controls: age, education, partner/children, immigrant status, ability scores (numeracy, literacy, problem-solving), experience, on-the-job training, part-time, occupation fixed effects, sector fixed effects, country fixed effects.
- Decomposition results (Table 3):
  - Unconditional Male-Female RTI Gap: 0.07.
  - Conditional Male-Female RTI Gap (with controls): 0.05.
  - Gap explained by included covariates: 0.01 (18.82 percent of the unconditional gap).
  - Percent of gap explained by components:
    - Age, numeracy, literacy, training: 6.07 percent.
    - Education: -0.87 percent.
    - Sector: 0.92 percent.
    - Occupation: 12.70 percent (p<0.01).
- Interpretation:
  - Occupational choice (types of tasks performed within occupations) is the largest documented contributor to the gender RTI gap.

### V. Estimating probability of automation — mapping occupation-level estimates to individuals
- Base occupational automatability: Frey and Osborne (2017) occupational probabilities (702 occupations, 0 to 100 percent scale).
- Mapping approach:
  - EM algorithm and a weighted Generalized Linear Model (GLM) relate occupation-level probabilities to individual-level PIAAC characteristics (age, education, gender, literacy and numeracy skills, and task variables).
  - Reweighting per iteration using a normal density-based weight function (Ibrahim, 1990) until convergence.
- Robustness:
  - Alternative occupational-level index: Brynjolfsson, Mitchell, and Rock (2018) “suitability for machine learning” (SML) index mapped via same EM approach.
  - SML differences: SML values can exceed one and are subjective rubric-based; higher SML indicates higher capability of ML to substitute tasks.

### VI. Gender differences in probability of automation — levels, high-risk groups, and correlates
- Cross-country averages (Table 4):
  - Probability of Automation (average): Female 0.40; Male 0.38; Gender ratio = 1.06 (*** p<0.01).
  - Share of workers at high risk (probability >= 0.7): Female 10.65; Male 8.84; Ratio = 1.20 (*** p<0.01).
  - SML (average): Female 3.47; Male 3.45; Ratio = 1.01 (*** p<0.01).
  - Share of workers at high SML risk (SML >= 3.5): Female 27.04; Male 6.70; Ratio = 4.04 (*** p<0.01).
- RTI–automatisability correlations (Table 5):
  - RTI Index coefficient 0.05** (positive association with Probability of Automation).
  - Routine Index coefficient 0.02*** (positive).
  - Abstract Index coefficient -0.05*** (negative).
  - Manual Index coefficient -0.01*** (negative).
  - Observations: 42,945.
- Country heterogeneity (Table A6 highlights):
  - Women face higher probabilities of automation than men in Japan, Singapore, and South Korea.
  - Sweden and Finland show men with greater likelihoods.
  - France, the UK, and the US show roughly similar proportions of men and women at high risk.
  - Turkey exhibits a reversed gap (Female probability 0.33; Male 0.41; Ratio 0.81***), potentially due to low female labor force participation and selection effects.
  - Selected country probability and high-risk entries:
    - Austria — Prob Female 0.48; Male 0.43; Ratio 1.11***; Share high risk Female 15.84; Male 10.09; Ratio 1.57***
    - Canada — 0.41; 0.39; 1.04**; 11.75; 9.79; 1.20**
    - Japan — 0.44; 0.34; 1.30***; 14.28; 4.20; 3.40***
    - Singapore — 0.36; 0.30; 1.18***; 8.59; 5.06; 1.70***
    - South Korea — 0.39; 0.33; 1.19***; 10.34; 4.29; 2.41***
    - United States — 0.39; 0.39; 1.01; 9.46; 9.20; 1.03
- Individual-level correlates (Table A5 selected coefficients):
  - Female — -0.08*** (coefficient in regression of probability of automation).
  - Age (60-65) — 0.13***.
  - Education Medium (ISCED 3, 4, 5B) — -0.26***.
  - Education High (ISCED 5A, 6) — -0.64***.
  - Using programming language — -4.16***.
  - Reading professional publications — -4.85***.
  - Selling — 3.22***.
  - Fingers and hand use — 0.93***.
  - Sector (Private=0; Public/Non-profit=1) — -0.18***.
  - Firm Size (11-1000) — 0.07***; Firm Size (>1000) — -0.03***.
  - Income decile 10 — -0.44***.
  - Not challenged enough at work — -0.16***.
  - Need more training — 0.08***.
  - Computer use at work — -0.02***.
  - Constant — 1.33***.
  - Country Fixed Effects — Yes.
  - N — 65,876.
- Interpretations:
  - Higher risk associated with selling, adherence to manuals, and heavy fingers-and-hands tasks.
  - Lower risk associated with training others, solving complex problems, writing articles, programming, and reading professional publications.
  - Probability of automation decreases with higher education, income level, and firm size; risk < 1 percent for workers with a bachelor’s degree or higher (statement).
  - Women with lower secondary education or less: nearly 50 percent at high risk; less than 40 percent of men with same education at high risk (statement).
  - Workers in SMEs have higher risk than workers in large enterprises; women more likely than men to face risk in SMEs but less likely in large firms relative to male counterparts.

### VII. Sectoral and occupational patterns in automation risk
- High-risk sectors (Table A7 and text):
  - Accommodation and food services; wholesale & retail trade; transportation identified as high-risk.
  - Retail trade employs 14 percent of both female and male workforce in the sample.
  - Accommodation and food services employs 7 percent of female workforce versus 5 percent of male workforce.
- Sectoral gender patterns:
  - Women overrepresented in education, health, and social services—relatively low average risk sectors—but within-sector gender gaps may persist (e.g., in education women can face higher likelihoods than men).
  - In retail trade, female workers are significantly less likely than male workers to perform abstract tasks (fewer women in managerial positions).
- Occupational risk patterns (Table A8 highlights):
  - Highest-exposure occupations on average: elementary occupations, clerks, and service workers.
  - Selected occupation probability and high-risk entries:
    - Clerks — Prob Female 0.54; Male 0.46; Ratio 1.16***; Share high risk Female 16.15; Male 10.24; Ratio 1.58***
    - Crafts & trade — 0.55; 0.52; 1.06*; 16.60; 13.95; 1.19
    - Elementary — 0.64; 0.64; 1.00; 38.35; 38.65; 0.99
    - Professional — 0.19; 0.20; 0.97*; 0.46; 0.22; 2.10*
    - Service, shop, & market — 0.55; 0.51; 1.08***; 25.25; 16.32; 1.55***
    - Plant/machine operators — 0.62; 0.60; 1.03; 30.60; 28.59; 1.07
  - Legislators, managers, and professionals are relatively insulated (less than 1 percent at high risk), but within professionals women can be twice as likely as men to be among the highest-risk subgroup.

### VIII. Intergenerational trends and occupational shifts (1994–2016)
- Aggregate and cohort patterns:
  - Since the 1990s, shifts away from clerical and elementary occupations toward service sector and professional jobs have been more pronounced for women than men.
  - Shift-share analysis indicates most job growth in OECD countries has been on the high-skill end, and women have benefited more than men.
  - Gender automation gaps between men and women are smaller for younger cohorts, even among high-risk groups.
  - Cross-sectional limitations: dataset does not track workers over time; observed generational patterns could reflect lifecycle effects.
- Selected occupational generational differences (Table 6 excerpts):
  - Clerks — Difference in Automatability (Older Generation) 1.88***; Female Share of Occupation 21.48. Younger Generation 1.11; 19.44.
  - Professional — Older Generation 4.51*; 26.39. Younger Generation 0.92; 28.20.
  - Service, shop, & market — Older Generation 2.01***; 15.36. Younger Generation 1.58***; 21.16.
  - Legislators, senior officials, & managers — Older Generation 1.62; 8.77. Younger Generation 0.65; 5.05.
  - Plant/machine operators — Older Generation 0.96; 1.75. Younger Generation 1.25; 1.23.
- Sectoral generational differences (Table 7 excerpts):
  - Accommodation & food — Older Generation 1.76*; Female Share 26.95. Younger Generation 1.64**; 33.14.
  - Financial services — Older Generation 7.73**; 9.20. Younger Generation 1.85; 4.61.
  - Education — Older Generation 3.93**; 2.86. Younger Generation 1.31; 0.61.
  - Manufacturing — Older Generation 1.82***; 16.89. Younger Generation 1.13; 13.21.
  - Wholesale & retail — Older Generation 1.63***; 26.92. Younger Generation 1.19; 25.07.
- Interpretation:
  - Across-the-board decline in women’s relative exposure to automation for younger cohorts; gender differences significant for older generations but substantially lower and often insignificant for younger cohorts.
  - Exception: Accommodation and food services remains highly exposed with large female shares.
  - Younger people (ages 20 to 39) are at high risk given current technologies, potentially driven by selection of less-educated workers into the labor force and junior-level positions being more automatable.
  - Education attainment increases for women (example: US university enrollment among female graduates rose to 71 percent by 2014 versus slightly more than 60 percent in 1994).

### IX. Mechanisms, caveats, and broader labor-market considerations
- Automation effects:
  - Can reduce employment by replacing workers and by reducing new-job creation in some sectors or create new high-skilled jobs—this can particularly affect young entrants.
  - Estimates assume current levels of technology and prevailing bottlenecks in use of computer-controlled equipment; results present a lower bound for potential impact of automation.
  - Estimates are based on technological feasibility rather than economic feasibility: tasks could be automated given current technology, but costs may be prohibitive relative to prevailing labor costs.
  - Job losses could be offset by new work opportunities and higher output potential from falling costs and prices.
- Data and methodological caveats:
  - Analysis does not capture the emerging “gig” economy.
  - The dataset does not track workers over time; observed generational patterns could reflect lifecycle effects such as a “child bearing penalty”.

### X. Policy implications and suggested priorities
- Priorities:
  - Policies to smooth transitions of younger women workers into less automatable roles and to facilitate human capital accumulation at early ages.
  - Ensure adequate safety nets and support for older, displaced workers to manage lifecycle risks and automation exposure.
- Further research priorities:
  - Examine drivers of gender RTI gaps across countries.
  - Evaluate differences in probability of automation across sectors and countries.
  - Deeper assessment of the gig economy’s opportunities and challenges for women.

### XI. Data and definitions (Appendix 1) — survey basis and sample
- Survey: OECD Programme for the International Assessment of Adult Competencies (PIAAC) administered in two rounds between 2011 and 2016.
- Coverage: 30 countries; adults ages 16 to 65; collects demographic, work, and cognitive-skill measures (numeracy, literacy, problem-solving).
- Task-frequency variables used to estimate occupational automatability via an EM algorithm.
- Sample size examples (Table A1):
  - Canada: PIAAC sample 19,403; IALS sample 4,683.
  - United Kingdom: PIAAC sample 5,911; IALS sample 4,582.
  - United States: PIAAC sample 3,560; IALS sample 2,697.
  - Austria: PIAAC sample 3,737; IALS sample n.a.
  - Belgium: PIAAC sample 3,386; IALS sample 1,155.

_Italicized source attribution: IMF staff (content drawn from the provided chapter/section of wpiea2019091)._

### 1. Relationship Between Female Labor Force Participation and Size of Manufacturing Sector

### 1. Relationship Between Female Labor Force Participation and Size of Manufacturing Sector 
and the Gender RTI Gap

### Major Sections
- 1. Relationship Between Female Labor Force Participation and Size of Manufacturing Sector 
    and the Gender RTI Gap _______________________________________________________________________ 10
- 2. Gender Differences in RTI by Occupation______________________________________________________ 12
- 3. Gender Gap in High Risk of Automation Across Sectors _______________________________________ 18
- 4. Gender Gap in High Risk of Automation Across Occupations _________________________________ 19
- 5. Changes in Occupational Shares by Gender (1994-2016) ______________________________________ 20
- 6. Gender Gaps in High Risk of Automation by Generation and Educational Level _______________ 21

### Tables
- 1. Questionnaire Items Used to Construct RTI ____________________________________________________ 8
- 2. Gender Differences in RTI Components ________________________________________________________ 9
- 3. RTI Decomposition: Drivers of RTI Gap ________________________________________________________ 11
- 4. Gender Differences in Probability of Automation and SML ____________________________________ 15
- 5. Correlation of RTI Components and Probability of Automation _______________________________ 16
- 6. Gender Gaps in High Risk of Automation by Generation and Occupation _____________________ 22
- 7. Gender Gaps in High Risk of Automation by Generation and Sector __________________________ 23

### Appendices
- 1. Data and Definitions of RTI and ICT Use Indices _______________________________________________ 26
- 2. Additional Tables ______________________________________________________________________________ 27

*Source: wpiea2019091 - 1. Relationship Between Female Labor Force Participation and Size of Manufacturing Sector (IMF PDF).*

### REFERENCES ____________________________________________________________________________________ 35

### REFERENCES (Content unit: wpiea2019091)

### I. Introduction — key findings and research focus
- Research question: gender implications of digitalization, artificial intelligence, and machine-learning for job displacement using a task-based approach and individual-level PIAAC data for 30 advanced and emerging economies.
- Methodological highlights:
  - Individual-level RTI (routine task intensity) constructed from PIAAC task composition, skills, and demographics.
  - Individual-level probabilities of automation derived by mapping occupation-level estimates (Frey and Osborne, 2017) to worker characteristics using an Expectation Maximization (EM) approach similar to Arntz, Gregory, and Zierahn (2017).
- Major empirical findings:
  - Women, on average, perform more routine or codifiable tasks than men across sectors and occupations.
  - Women perform fewer analytical/abstract tasks (e.g., information-processing skills) that are complementary to technological change.
  - Self-selection into occupations explains most gender differences in task composition, but cross-country heterogeneity is substantial.
  - Within the sample, 54 million male and female workers in 30 countries (28 OECD member countries, Cyprus, and Singapore) are at a high risk of being displaced by technology (i.e., facing higher than 70 percent likelihood of being automated) within the next two decades.
  - Across the sample, 11 percent of the female workforce versus 9 percent of the male workforce face high risk of automation (sample-wide comparison).
  - Less well-educated and older female workers (aged 40 and older), and those in clerical, service, and sales positions are disproportionately exposed.

### II. Measuring routineness (RTI index) — construction and components
- RTI index definition (individual i):
  - RTI_i = Routine_i − Abstract_i − Manual_i
  - RTI index ranges from zero to one, with higher values indicating more routine activities.
- Component descriptions:
  - Abstract: analytical and interpersonal tasks (e.g., writing reports, solving complex problems, negotiating).
  - Routine: lack of job flexibility and little learning on the job (repetitiveness, codifiability).
  - Manual: two manual task types from PIAAC (hand and finger dexterity; long physical work) classified into routine and non-routine manual per prior literature.
- Estimation procedure:
  - Principal component analysis (PCA) used to derive indices for subcomponents; components standardized to lie between zero and one; final RTI standardized to lie between zero and one.

### III. Gender RTI gap — magnitudes and cross-country patterns
- Sample average RTI components (Table 2):
  - RTI Level: Female 0.52; Male 0.46.
  - Abstract: Female 0.48; Male 0.56.
  - Manual: Female 0.40; Male 0.47.
  - Routine: Female 0.51; Male 0.47.
  - All t-test p-values reported as 0.00 (statistically significant).
- Cross-country heterogeneity:
  - Female RTI level highest in Eastern and Southern European countries, lowest in Scandinavian and Central European countries.
  - Example comparisons given: RTI for female workers in Lithuania is 36 percent higher than in Norway.
  - Gender RTI gap highest in Japan, the Slovak Republic, Singapore, and Estonia (sample-wide statement).
- Correlates:
  - Gender RTI gaps are smaller in countries with higher female labor force participation (FLFP).
  - Gender RTI gaps are positively associated with the size of manufacturing sector (as a percentage of GDP).

### IV. Decomposition of the gender RTI gap
- Econometric specification includes female indicator, individual controls (age, education, partner/children, immigrant status), ability scores (numeracy, literacy, problem-solving), job controls (experience, on-the-job training, part-time), occupation fixed effects, sector fixed effects, country fixed effects.
- Decomposition results (Table 3):
  - Unconditional Male-Female RTI Gap: 0.07.
  - Conditional Male-Female RTI Gap (with controls): 0.05.
  - Gap explained: 0.01 (18.82 percent of the unconditional gap explained by included covariates).
  - Percent of gap explained by components:
    - Age, numeracy, literacy, training: 6.07 percent.
    - Education: -0.87 percent.
    - Sector: 0.92 percent.
    - Occupation: 12.70 percent (statistically significant at p<0.01).
- Interpretation:
  - Occupational choice (the intensive margin—types of tasks performed within occupations) is the largest documented contributor to the gender RTI gap.

### V. Estimating probability of automation — mapping occupation-level estimates to individuals
- Base occupational automatability: Frey and Osborne (2017) occupational probabilities (702 occupations, 0 to 100 percent scale).
- Limitations of pure occupation-level assignment noted: assumes homogeneous tasks within occupations and identical technology access across workers.
- Mapping approach:
  - Use EM algorithm and a weighted Generalized Linear Model (GLM) to relate occupation-level probabilities to individual-level PIAAC characteristics (age, education, gender, literacy and numeracy skills, and task variables).
  - Reweighting per iteration using a normal density-based weight function (Ibrahim, 1990) until convergence.
- Robustness check:
  - Alternative occupational-level index: Brynjolfsson, Mitchell, and Rock (2018) “suitability for machine learning” (SML) index mapped to individuals via the same EM approach.
  - SML differences: SML values can exceed one and are subjective rubric-based; higher SML indicates higher capability of ML to substitute tasks.

### VI. Gender differences in probability of automation — levels, high-risk groups, and correlates
- Cross-country averages (Table 4):
  - Probability of Automation (average): Female 0.40; Male 0.38; Gender ratio = 1.06 (*** p<0.01).
  - Share of workers at high risk (probability >= 0.7): Female 10.65; Male 8.84; Ratio = 1.20 (*** p<0.01).
  - SML (average): Female 3.47; Male 3.45; Ratio = 1.01 (*** p<0.01).
  - Share of workers at high SML risk (SML >= 3.5): Female 27.04; Male 6.70; Ratio = 4.04 (*** p<0.01).
- RTI and automatisability correlations (Table 5):
  - RTI Index positively associated with Probability of Automation: RTI Index coefficient 0.05**.
  - Routine Index coefficient 0.02*** (positive).
  - Abstract Index coefficient -0.05*** (negative).
  - Manual Index coefficient -0.01*** (negative).
  - Observations: 42,945.
- Country heterogeneity:
  - Women face higher probabilities of automation than men in Japan, Singapore, and South Korea.
  - The result reverses in Sweden and Finland, where men face greater likelihoods.
  - France, the UK, and the US show roughly similar proportions of men and women at high risk.
  - Outlier noted: Turkey exhibits a reversed gap potentially due to low female labor force participation and selection effects.
- Individual-level correlates:
  - Higher risk associated with tasks: selling, adhering to manuals, using fingers and hands extensively.
  - Lower risk associated with tasks: training others, solving complex problems, writing articles.
  - Probability of automation decreases in education, income level, and firm size.
  - Risk of automation < 1 percent for workers with a bachelor’s degree or higher (statement).
  - Women with lower secondary education or less: nearly 50 percent at high risk of automation; less than 40 percent of men with the same education level at high risk (statement).
  - Workers in SMEs have higher risk of automation compared to workers in large enterprises (>1000 workers). Women are more likely than men to face risk in SMEs but less likely in large firms relative to male counterparts.

### VII. Sectoral and occupational patterns in automation risk
- Sectors with high risk of automation (Figures 3 and A7 references):
  - Accommodation and food services; retail trade; transportation highlighted as high-risk sectors.
  - Retail trade employs 14 percent of both female and male workforce in the sample.
  - Accommodation and food services employs 7 percent of female workforce versus 5 percent of male workforce.
- Sectoral gender patterns:
  - Women overrepresented in education, health, and social services—sectors at relatively low average risk of automation—but within-sector gender gaps may persist (e.g., in education women can face higher likelihoods than men).
  - Retail trade: overall high automation risk, and female workers are significantly less likely than male workers to perform abstract tasks (fewer women in managerial positions).
- Occupational risk patterns (Figure 4 and Table A8 references):
  - Highest-exposure occupations on average: elementary occupations, clerks, and service workers.
  - Women are overrepresented among service workers and face higher risk than male counterparts.
  - Legislators, managers, and professionals are relatively insulated (less than 1 percent at high risk).
  - Among professionals, although overall threat is low and women on average at lower risk than men, women are twice as likely as male counterparts to be among the highest-risk subgroup within professionals.

### VIII. Intergenerational trends and shifts in occupational composition
- Positive trends:
  - Since the 1990s, shifts away from clerical and elementary occupations toward service sector and professional jobs have been more pronounced for women than men in the sample.
  - Shift-share analysis (1994–2016) indicates most job growth in OECD countries has been on the high-skill end, and women have benefited more than men.
  - Figure 5 (referenced): shift away from clerical occupations toward service and retail workers, technicians, and professionals—a trend more pronounced for women.
- Cohort patterns:
  - Aggregate gender gaps primarily driven by older cohorts of women.
  - Gender automation gaps between men and women are smaller for younger cohorts, even among high-risk groups (e.g., less-well educated, clerical and sales positions).
  - Caution: cross-sectional data limits inference on cohort dynamics, but findings are consistent with other literature showing decreased labor market barriers to female employment in male-dominated occupations over time.

*Italicized source attribution: IMF staff (content drawn from the provided chapter/section: REFERENCES and related sections of wpiea2019091).*

### Appendix 1 contains further details on the International Adult Literacy Survey used for this analysis.

### wpiea2019091 - Appendix 1 contains further details on the International Adult Literacy Survey used for this analysis.

### Summary findings on automation risk and gender
- Using individual-level task data, women on average perform more routine tasks than men across all sectors and occupations; these tasks are most prone to automation.
- Estimated number of female jobs at high risk of automation within next two decades: 26 million female jobs in 30 countries (28 OECD member countries, Cyprus, and Singapore).
- High automatability is defined as having probability of automation >= 0.7.
- Female workers face a higher risk of automation compared to male workers across all occupations, with significant heterogeneity across sectors and countries.
- Positive trends: increased educational attainment for women and occupational shifts into roles relatively more insulated from automation (notably movement away from clerical and service positions toward professional roles), but underrepresentation in managerial positions remains a vulnerability.

### Generational and occupational differences
- Younger people (ages 20 to 39) are at highest risk of losing jobs to automation given current technologies; this is potentially driven by selection of less-educated workers into the labor force and junior-level positions being more automatable.
- Older cohorts of working women (ages 40 to 65) are at significantly higher risk for automation than men in the same age cohorts.
- Inter-generational shift toward higher educational levels, providing relative insulation from automation risk for younger cohorts.
- Education attainment change example: in the US, in 1994 slightly more than 60 percent of male and female graduates were enrolled in university; by 2014 the figure for women had jumped to 71 percent, while that for men was broadly unchanged.

- Table 6 (selected occupational entries — Difference in Automatability (Female vs Male); Female Share of Occupation (percent)):
  - Clerks: Older Generation 1.88***; 21.48. Younger Generation 1.11; 19.44.
  - Professional: Older Generation 4.51*; 26.39. Younger Generation 0.92; 28.20.
  - Service, shop, & market: Older Generation 2.01***; 15.36. Younger Generation 1.58***; 21.16.
  - Legislators, senior officials, & managers: Older Generation 1.62; 8.77. Younger Generation 0.65; 5.05.
  - Plant/machine operators: Older Generation 0.96; 1.75. Younger Generation 1.25; 1.23.

- Interpretation from Table 6:
  - Shift away from clerical occupations and toward professional occupations for younger cohorts of women.
  - Gender automatability gap narrows or becomes statistically insignificant for younger generations in several occupations and even reverses in some (e.g., professionals, legislators and senior managers).
  - Rising share of women in service and retail occupations remains concerning given high automation risk in those occupations.

### Sectoral differences and inter-generational trends
- Table 7 (selected sector entries — Difference in Automatability (Female vs Male); Female Share of Sector (percent)):
  - Accommodation & food: Older Generation 1.76*; 26.95. Younger Generation 1.64**; 33.14.
  - Financial services: Older Generation 7.73**; 9.20. Younger Generation 1.85; 4.61.
  - Education: Older Generation 3.93**; 2.86. Younger Generation 1.31; 0.61.
  - Manufacturing: Older Generation 1.82***; 16.89. Younger Generation 1.13; 13.21.
  - Wholesale & retail: Older Generation 1.63***; 26.92. Younger Generation 1.19; 25.07.

- Key sectoral patterns:
  - Across-the-board decline in women’s relative exposure to automation for younger cohorts.
  - Gender differences in automatability are statistically significant in almost all sectors for older generations but substantially lower and often insignificant for younger cohorts (reductions up to four-fold in cases such as education and financial services).
  - Exception: Accommodation and food services remains highly exposed to automation with large female shares.

### Mechanisms, caveats, and broader labor-market considerations
- Automation can reduce employment by replacing workers and by reducing new-job creation in some sectors or creating new high-skilled jobs—this can particularly affect young entrants.
- Estimates assume current levels of technology and prevailing bottlenecks in use of computer-controlled equipment; results present a lower bound for potential impact of automation.
- Estimates are based on technological feasibility rather than economic feasibility: tasks could be automated given current technology, but costs of automation may be prohibitive relative to prevailing labor costs.
- Job losses could be offset by new work opportunities created by technology and higher output potential from falling costs and prices.
- Analysis does not capture the emerging “gig” economy (flexible, independent work arrangements), which could make it easier for women to combine paid work with family responsibilities and potentially improve labor market outcomes.
- The dataset does not track workers over time; observed generational patterns could reflect widening gender gaps over women’s lifecycle (for example, a “child bearing penalty” documented in other literature).

### Policy implications and suggested priorities
- Policies to smooth transitions of younger women workers into less automatable roles and to facilitate human capital accumulation at early ages are important.
- Ensure adequate safety nets and support for older, displaced workers to manage lifecycle risks and automation exposure.
- Further research priorities include: examining drivers of gender RTI gaps across countries; evaluating differences in probability of automation across sectors and countries; and deeper assessment of the gig economy’s opportunities and challenges for women.

### Data and definitions (Appendix 1)
- Survey basis: Organisation for Economic Cooperation and Development (OECD) Programme for the International Assessment of Adult Competencies (PIAAC) administered in two rounds between 2011 and 2016.
- Sample coverage: 30 countries included in the analysis; survey covers adults ages 16 to 65 and collects demographic, work, and cognitive-skill measures (numeracy, literacy, problem-solving) used as proxies for workers’ ability.
- Task-frequency variables: variables describing frequency at which respondents perform specific tasks at work are used to estimate occupational automatability via an expectation-maximization (EM) algorithm relating individual and job task characteristics to occupational-level risk of automation.
- Sample size examples from Table A1:
  - Canada: PIAAC sample 19,403; IALS sample 4,683.
  - United Kingdom: PIAAC sample 5,911; IALS sample 4,582.
  - United States: PIAAC sample 3,560; IALS sample 2,697.
  - Austria: PIAAC sample 3,737; IALS sample n.a.
  - Belgium: PIAAC sample 3,386; IALS sample 1,155.

*Source: https://www.imf.org/-/media/files/publications/wp/2019/wpiea2019091.pdf*

### APPENDIX 2. Additional Tables

### APPENDIX 2. Additional Tables

### Gender differences in Routine Task Intensity (RTI) by country
- Table A2 reports: Female Labor Force Participation Rate (percent); RTI Level (Female); RTI Gap (female vs. male) defined as RTI gap = Female RTI/Male RTI; and T-test (Male-Female) P Val for each country.
- Selected country entries (as reported):
  - Austria — 70.56; 0.51; 1.16; 0.00
  - Belgium — 61.11; 0.50; 1.12; 0.00
  - Canada — 74.18; 0.53; 1.14; 0.00
  - Japan — 63.64; 0.52; 1.21; 0.00
  - Sweden — 77.60; 0.48; 1.07; 0.00
  - Turkey — 32.19; 0.53; 1.05; 0.00
  - United Kingdom — 70.00; 0.51; 1.10; 0.08
  - United States — 66.36; 0.50; 1.14; 0.00
- Sources: International Labour Organization; PIAAC survey; and IMF staff calculations.
- Note: The t-test evaluates whether there is a significant difference between the mean values for males and females.

### Gender differences in RTI by occupation
- Table A3 reports: Share of Female Workers (percent); RTI Level (Female); RTI Gap (female vs. male); T-test (Male-Female) P-Value by occupation (ISCO 2-digit groups).
- Occupation entries (as reported):
  - Professional — 30; 0.50; 1.07; 0.00
  - Technicians & assoc. prof. — 21; 0.53; 1.17; 0.00
  - Legislators, senior officials, & managers — 8; 0.45; 1.14; 0.00
  - Clerks — 18; 0.6; 1.16; 0.00
  - Plant/machine operators — 1; 0.59; 1.12; 0.00
  - Elementary — 2; 0.55; 1.08; 0.00
- Note: The share of female workers represents the proportion of the total female labor force employed in each occupational classification. Occupational classification corresponds with the respondent’s 2-digit ISCO code.

### Occupational classifications (ISCO mapping)
- Table A3.1 lists ISCO 2-digit occupations and corresponding ISCO 4-digit components. Examples:
  - Professional: Science and engineering professionals; Health professionals; Teaching professionals; Business and administration professionals; Information and communications technology professionals; Legal, social and cultural professionals.
  - Clerks: General and keyboard clerks; Customer services clerks; Numerical and material recording clerks; Other clerical support workers.
  - Elementary: Cleaners and helpers; Agricultural, forestry and fishery laborers; Laborers in mining, construction, manufacturing and transport; Food preparation assistants; Street and related sales and service workers; Refuse workers and other elementary workers.

### Gender differences in RTI by sector
- Table A4 reports: Share of Female Workers (percent); RTI Level (Female); RTI Gap (female vs. male); T-test (Male-Female) P-Value by sector (1-digit ISIC rev 4).
- Sector entries (as reported):
  - Health — 21; 0.48; 1.05; 0.00
  - Wholesale & retail — 16; 0.51; 1.19; 0.00
  - Education — 17; 0.51; 1.06; 0.00
  - Manufacturing — 8; 0.54; 1.17; 0.00
  - Construction — 2; 0.57; 1.38; 0.00
  - Financial services — 6; 0.55; 1.19; 0.00
- Note: The share of female workers represents the proportion of the total female labor force employed in each sector.

### Estimation of probability of automation — regression results
- Table A5 contains cross-country regression results of the probability of automation on individual and job characteristics. The probability of automation is estimated using an expectation-maximization (EM) algorithm relating individual characteristics and job task characteristics to occupational-level risk of automation. All task variables range from 0 to 1. Statistical significance: *** p<0.01; ** p<0.05; * p<0.1.
- Selected coefficient estimates (Variable — Probability of Automation):
  - Female — -0.08***
  - Age (20-24) — -0.08***
  - Age (25-29) — 0.05***
  - Age (30-34) — -0.07***
  - Age (60-65) — 0.13***
  - Education Medium (ISCED 3, 4, 5B) — -0.26***
  - Education High (ISCED 5A, 6) — -0.64***
  - Numeracy Skills — 0.00
  - Literacy Skills — -0.00***
  - Using programming language — -4.16***
  - Using internet for work — -1.32***
  - Reading professional publications — -4.85***
  - Reading books — -5.06***
  - Training others — -2.89***
  - Presenting — -5.36***
  - Selling — 3.22***
  - Planning activities of others — -2.30***
  - Organizing own schedule — -1.09***
  - Fingers and hand use — 0.93***
  - Reading instructions — -1.47***
  - Writing articles — -4.53***
  - Filling forms — -0.87***
  - Complex math or statistics — -1.78***
  - Cooperating with others at work — -0.00*
  - Exchanging information — 0.85***
  - Consulting — 0.34***
  - Responsibility for managing staff — 0.13***
  - Sector (Private=0; Public/Non-profit=1) — -0.18***
  - Firm Size (11-1000) — 0.07***
  - Firm Size (>1000) — -0.03***
  - Job experience requirement — -0.08***
  - Income decile 2 — 0.19***
  - Income decile 3 — 0.18***
  - Income decile 4 — 0.08***
  - Income decile 5 — 0.06***
  - Income decile 6 — -0.02*
  - Income decile 7 — -0.04***
  - Income decile 8 — -0.15***
  - Income decile 9 — -0.42***
  - Income decile 10 — -0.44***
  - Not challenged enough at work — -0.16***
  - Need more training — 0.08***
  - Computer use at work — -0.02***
  - Constant — 1.33***
  - Country Fixed Effects — Yes
  - N — 65,876
- Sources: Frey and Osborne (2017); PIAAC survey; and IMF staff estimates.
- Note: Task variables are measured in shares of time. Statistical significance levels shown next to estimates.

### Gender differences in probability of automation by country
- Table A6 reports: Probability of automation and Share of workers at high risk of automation (probability >=0.7) in percent, for Female and Male, and Gender Ratio = female measure/male measure. Statistical significance on the ratio uses t-tests; significance levels: *** p<0.01; ** p<0.05; * p<0.1.
- Selected country entries (Probability of automation — Female; Male; Ratio; Share of workers at high risk — Female; Male; Ratio):
  - Austria — 0.48; 0.43; 1.11***; 15.84; 10.09; 1.57***
  - Canada — 0.41; 0.39; 1.04**; 11.75; 9.79; 1.20**
  - Japan — 0.44; 0.34; 1.30***; 14.28; 4.20; 3.40***
  - Singapore — 0.36; 0.30; 1.18***; 8.59; 5.06; 1.70***
  - South Korea — 0.39; 0.33; 1.19***; 10.34; 4.29; 2.41***
  - Turkey — 0.33; 0.41; 0.81***; 4.89; 14.49; 0.34***
  - United States — 0.39; 0.39; 1.01; 9.46; 9.20; 1.03
- Note: High automatability is defined as probability of automation >= 0.7.

### Gender differences in probability of automation by sector
- Table A7 reports: Probability of automation and Share of workers at high risk (probability >=0.7) in percent, for Female and Male, and Gender Ratio = female measure/male measure; significance levels indicated.
- Selected sector entries (Probability — Female; Male; Ratio; Share high risk — Female; Male; Ratio):
  - Accom. & food — 0.60; 0.56; 1.07*; 34.91; 26.00; 1.34*
  - Construction — 0.49; 0.39; 1.26***; 9.93; 5.53; 1.80*
  - Financial services — 0.42; 0.31; 1.37***; 7.17; 1.89; 3.80**
  - Info. & comm. — 0.36; 0.27; 1.33***; 7.03; 2.39; 2.93**
  - Manufacturing — 0.47; 0.42; 1.14***; 15.01; 10.68; 1.40**
  - Wholesale & retail — 0.57; 0.53; 1.08***; 27.85; 20.90; 1.33***
- Note: High automatability defined as probability >= 0.7.

### Gender differences in probability of automation by occupation
- Table A8 reports: Probability of automation and Share of workers at high risk (probability >=0.7) in percent, for Female and Male, and Gender Ratio = female measure/male measure.
- Selected occupation entries (Probability — Female; Male; Ratio; Share high risk — Female; Male; Ratio):
  - Clerks — 0.54; 0.46; 1.16***; 16.15; 10.24; 1.58***
  - Crafts & trade — 0.55; 0.52; 1.06*; 16.60; 13.95; 1.19
  - Elementary — 0.64; 0.64; 1.00; 38.35; 38.65; 0.99
  - Professional — 0.19; 0.20; 0.97*; 0.46; 0.22; 2.10*
  - Service, shop, & market — 0.55; 0.51; 1.08***; 25.25; 16.32; 1.55***
  - Plant/machine operators — 0.62; 0.60; 1.03; 30.60; 28.59; 1.07
- Note: High automatability defined as probability >= 0.7.

*Sources: Frey and Osborne (2017); PIAAC survey; International Labour Organization; and IMF staff calculations.*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2019/wpiea2019091.pdf_