## wpiea2022042-print-pdf

## Source details

**Canonical URL:** [wpiea2022042-print-pdf](https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022042-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2022/english/wpiea2022042-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2022/english/wpiea2022042-print-pdf.pdf.json)

---

### Data and occupational categorization (Section 3)
- Source: UK Labour Force Survey (LFS), quarterly frequency.
- Analysis period for baseline: 2001:Q2-2019:Q4 (consistent SOC2000 occupational classification).
- Sample: employed workers aged 21 to 65; main focus on females with no qualifications above secondary education (A-levels or similar).
- Sample size (employment regressions): 922,952 (unemployed and out of labor force excluded).
- Wage sample sizes:
  - manual: 64,525
  - routine: 129,853
  - abstract: 54,652
- Occupational categorization:
  - Three types — abstract, manual, routine — via crosswalk between SOC2000 and ISCO08, merging Cortes et al. (2020) routine groups.
  - Routine group formed by grouping routine-cognitive and routine-manual occupations into a single routine group.
  - Manual and abstract correspond to non-routine-manual and non-routine-cognitive categories, respectively.
- Wage variable: gross hourly pay pre-computed in LFS, values deflated to 2010 prices with the UKCPI in the decomposition exercise.

### Empirical strategy and counterfactuals (Section 5–7)
- Models:
  - Employment linear probability model: Employed in Occ.ict = α0 + α1 ageict + α2 age2ict + α3 age3ict + γt + δc + εict.
  - Wage equation: Log(w)ict = same specification.
- Age modelled via cubic polynomial; cohorts in 5-year windows with Cohort 1940 (born 1940–1944) as reference.
- Counterfactual decompositions of average wages W_A_t:
  - Only cohort effects: W_A,COHORT_t (equation 4).
  - Only year effects: W_A,YEAR_t (equation 5).
- Lifetime earnings computation:
  - Y_c = Tc Σ_{t=tc0}^{Tc} β^{t} Σ_o p_co_t × w_co_t with β = 0.975 (discount rate of 2.5% per year).
  - Assumption: use 2001 year effects for prior years and 2019 year effect for subsequent years (assume change permanent).
  - Counterfactual lifetime exercises: set cohort effects equal to Cohort 1940; set year effects equal to their 2001 value.

### Stylized occupational shifts by cohorts (Section 4 summary)
- Life-cycle patterns:
  - Abstract jobs: concave/hump-shaped over age.
  - Routine jobs: mild U-shape with lowest point just before age 40; share in 0.5-0.6 interval across most ages absent aggregate/cohort shifts.
  - Manual jobs: almost flat around 0.3 over the life-cycle.
- Cohort heterogeneity:
  - For workers aged 40+ in 2019, share in abstract occupations rose over time; unchanged for earlier ages.
  - Younger cohorts shifted more toward manual jobs compared to earlier cohorts.
  - Interpretation: reallocation away from routine induced “upgrading” for older generations but “downgrading” for younger entrants.

### Estimated period and cohort effects on employment propensities and wages (Section 6 summary)
- Period (year) effects (2001–2019):
  - Routine jobs fell by around 13 percentage points between 2001 and 2019.
  - Abstract jobs rose by 10 percentage points in the same interval.
  - Wages:
    - routine and manual jobs show positive trend so that in 2019 wages are more than 15% higher than in 2001.
    - abstract jobs show a 9% increase in 2019 relative to 2001.
  - Business cycle: positive wage growth until 2007 and steep decline during the Great Recession across occupations.
- Cohort effects:
  - Non-college females born since the late 1970s experienced a pronounced fall in the permanent component of real wages across all occupation types.
  - Example: those born in 1990-1994 receive wages that are 14% lower than workers born in 1965-1969 within all occupations.
- Wages by occupation and age:
  - Abstract jobs pay more than routine at every age; manual jobs pay the least.
  - Abstract jobs show steeper wage progression over the career; manual jobs show virtually no progression.

### Decomposition of wage gap changes (2001–2019) and explanatory power (Section 6–7)
- Observed LFS mean wages (GBP, deflated to 2010 prices):
  - W 21–34: 8.15 (2001) → 8.20 (2019) = 0.7% Δ
  - W 35–49: 8.30 (2001) → 9.81 (2019) = 18.2% Δ
  - W 50–64: 7.79 (2001) → 9.60 (2019) = 23.2% Δ
  - Young vs prime gap: 1.45 (2001) → 1.75 (2019)
- APC Model predictions:
  - W 21–34: 7.87 → 8.53 = 8.3% Δ
  - W 35–49: 8.26 → 9.65 = 16.8% Δ
  - W 50–64: 7.72 → 9.42 = 22.0% Δ
  - Predicted young-prime gap change explained by regressions: 0.73 (of 1.45 observed) = 50% explained.
  - Predicted young-old gap change explained: 1.04 (table indicates 60% explained).
- Decomposition within APC model:
  - Cohort effects explain the majority of the model-explained widening:
    - 77% (0.56 out of 0.73) of the young-prime-age gap explained by the model is due to cohort effects.
    - 85% (0.89 out of 1.04) of the young-old gap explained by the model is due to cohort effects.
  - Within cohort contributions: most driven by cohort effects on wages within occupations; remainder (about one quarter to one fifth) driven by shifts in employment shares.
- Counterfactual series:
  - Year effects (cyclical) similar across age groups on average.
  - Cohort effects drive diverging trends: young non-college females experience a fall in average wages; prime-age stable; older mild rise.
  - Post-2008 cyclical downturn affected younger workers more severely (not fully captured by model interactions).

### Lifetime earnings and lifecycle projections (Section 7)
- Cohort effects on present value of lifetime earnings (relative to Cohort 1940):
  - Cohorts up to 1980: cohort effects contributed positively to lifetime earnings.
  - Since Cohort 1965: successive cohorts show a lower cohort effect compared to preceding cohort; cohort effect turns negative for cohorts born since 1985.
- Period (year) effects:
  - Monotonically increasing between 2001 and 2019; assuming 2019 effect persists implies aggregate growth benefits future years of cohorts.
- Decomposed drivers of lifetime effects:
  - Decline in within-occupation wages is the primary driver of reduced lifetime earnings for post-1975 cohorts.
  - Employment propensity (shares across occupations) accounts for roughly one third of the positive cohort contribution until Cohort 1975; after 1975, within-occupation wage declines dominate the fall in lifetime earnings.
  - For period effects, effects on wages are the main driver; shifts toward abstract and manual jobs account for about one tenth of increased lifetime earnings prospects.

### Robustness and sensitivity checks
- Alternative occupation categorization, sample changes (including/excluding foreign-born, excluding self-employed), and additional controls (marital status, number of children, region, part-time, self-employment, contract type) — results hold with minor quantitative changes.
- Expanded time sample back to 1986 (using crosswalks across SOC classifications) adds Cohorts 1930 and 1935; hump-shaped differences accentuated when comparing 1970s cohorts to 1930 cohort.
- Quantile regressions at the 10th, 25th, 50th, 75th, 90th percentiles show period and cohort effects broadly hold across wage distribution.
- Alternative estimation methods (maximum entropy, intrinsic estimator, proxying period effects with IT equipment relative prices and labor productivity) reported in Online Appendix — robustness maintained.
  - Proxying period effects with IT price and labor productivity yields similar baseline results but can amplify linear cohort trends (example: fall in probability of routine jobs from Cohort 1940 to Cohort 1990 of almost 30 percentage points in one specification).

### Possible drivers of negative cohort effects for recent non-college female cohorts (Section 8)
- Increased university attendance and potential sorting:
  - Rising female university attendance could induce negative selection into non-college group, lowering average wages of non-college group; empirical evidence mixed.
- Routine-replacing technical change and industry reallocation:
  - Decline in routine jobs uneven across ages and industries.
  - Prime-age and older non-college females saw decline in routine jobs concentrated in low-wage sectors (trade, manufacturing).
  - Younger workers experienced routine job decline across a broader set of industries, with reallocation concentrated into relatively low-wage industries such as health and elderly care (partly driven by population ageing).
- Family formation, tax and welfare reforms, and work incentives:
  - 1990s–2000s reforms including Working Families Tax Credit (WFTC) and Income Support changes increased in-work benefits.
  - WFTC: subsidy for working mothers involved a 25 percent rise (in constant wage levels) in maximum award for single mothers of one child and a drop in the withdrawal rate from 70 to 55 percent.
  - Changes shifted childbearing later and increased part-time work, especially among lower-educated women; part-time work tends to entail lower pay, less training, and limited transitions to full-time permanent roles.
- Rise in non-standard work:
  - Increasing self-employment and flexible/temporary contracts (zero-hours): more than a third of zero-hours contract workers were under age 24 in 2017; almost half of zero-hours contracts in low-skilled occupations.
- Declining bargaining power:
  - Young, low-skilled women (single mothers) may face lower bargaining power, self-select into flexible local firms, reducing wages and reinforcing low-wage traps.

### Conclusions and policy implications (Section 8)
- Main findings:
  - Marked worsening of labor compensation for young non-college women in the UK over 2001–2019.
  - Job polarization (period effects) contributed but played a limited role relative to cohort-specific declines in within-occupation wages for recent cohorts.
  - Cohorts born since 1980 experienced a steep rise in manual employment and a fall in routine occupations, and cohort wage effects indicate a slowdown across all occupation types.
  - Within-occupation wage declines for recent cohorts explain most of the aggregate wage gap between young and older non-college workers since 2001.
- Potential long-term consequences:
  - Lower lifetime earnings, lower retirement savings, and longer life expectancies for current young non-college cohorts.
  - Limited wage and occupational progression could exacerbate gender inequality and reduce aggregate productivity.
- Research and policy directions:
  - Further structural modeling to quantify channels: negative selection into non-college group, role of labor market training, flexible employment contracts, family policies.
  - Policy focus implied: addressing the within-occupation wage deterioration for young non-college women, considering training, job quality, bargaining power, and family-support policies to improve long-run earnings trajectories.

*Source: wpiea2022042-print-pdf - "Are Low-Skill Women Being Left Behind? Labor Market Evidence from the UK", Working Paper No. WP/22/42 — Sections 3–8 and Appendices.*

### Section 3 describes the data and the occupational categorization we apply. Section 4 presents

### Section 3–7 overview

### Data and occupational categorization (Section 3)
- Section 3 describes the data and the occupational categorization we apply.

### Descriptive evidence on occupational composition (Section 4)
- Section 4 presents descriptive evidence on changes in the occupational composition of employment of non-college females and discusses differences across cohorts.

### Empirical strategy (Section 5)
- Section 5 outlines the empirical strategy.

### Estimation of age, period, and cohort effects (Section 6)
- Section 6 presents the estimation of age, period, and cohort effects in employment propensities and wages.

### Counterfactual exercises and relevance of cohort effects (Section 7)
- In Section 7, we carry out a set of counterfactual exercises to quantify the relevance of the cohort effects for earnings inequality across ages and for workers’ expected lifetime income.

*Source: wpiea2022042-print-pdf - Section 3 describes the data and the occupational categorization we apply. Section 4 presents*

### Section 8 discusses potential channels driving the lower economic prospects for younger cohorts

### wpiea2022042-print-pdf - Section 8 discusses potential channels driving the lower economic prospects for younger cohorts

### Relation to the literature
- Focus: non-college female workers in the UK and asymmetric occupational changes across cohorts.
- Links to prior findings:
  - Job polarization and occupational composition shifts (Cortes et al., 2020; Bussolo et al., 2018; Autor and Dorn, 2009).
  - Improvement in female labor market prospects largely tied to abstract-intensive service jobs (Cortes et al., 2021).
  - Non-college females show lowest wages with little life-cycle growth in the UK (Costa Dias et al., 2021).
- Contribution: quantify the role of occupational shifts in widening the wage gap between younger and older non-college female workers over the past 20 years using a life-cycle (cohort) perspective.

### Data and sample
- Source: UK Labour Force Survey (LFS), quarterly frequency.
- Analysis period for baseline: 2001:Q2-2019:Q4 (consistent SOC2000 occupational classification).
- Sample: employed workers aged 21 to 65; main focus on females with no qualifications above secondary education (A-levels or similar).
- Sample size (employment regressions): 922,952 (unemployed and out of labor force excluded).
- Wage sample sizes: manual 64,525; routine 129,853; abstract 54,652.
- Occupational categorization: three types — abstract, manual, routine — via crosswalk between SOC2000 and ISCO08, merging Cortes et al. (2020) routine groups.
- Wage variable: gross hourly pay pre-computed in LFS, values deflated to 2010 prices with the UKCPI in the decomposition exercise.

### Empirical strategy and counterfactuals
- Modelled employment propensity and log wages within each occupation as functions of age, year, and 5-year cohort dummies:
  - Employment linear probability model: Employed in Occ.ict = α0 + α1 ageict + α2 age2ict + α3 age3ict + γt + δc + εict.
  - Wage equation: Log(w)ict = same specification.
- Age modelled via cubic polynomial; cohorts in 5-year windows with Cohort 1940 (born 1940–1944) as reference.
- Counterfactual series to decompose changes in average wages W_A_t:
  - Only cohort effects: W_A,COHORT_t (equation 4).
  - Only year effects: W_A,YEAR_t (equation 5).
- Lifetime earnings computation:
  - Y_c = Tc Σ_{t=tc0}^{Tc} β^{t} Σ_o x p_co_t x w_co_t with β = 0.975 (discount rate of 2.5% per year).
  - Assumption: use 2001 year effects for prior years and 2019 year effect for subsequent years (assume change permanent).
  - Counterfactual lifetime exercises: set cohort effects equal to Cohort 1940; set year effects equal to their 2001 value.

### Occupational shifts by cohorts (stylized facts)
- Life-cycle patterns:
  - Abstract jobs: concave/hump-shaped over age.
  - Routine jobs: mild U-shape with lowest point just before age 40; share in 0.5-0.6 interval across most ages absent aggregate/cohort shifts.
  - Manual jobs: almost flat around 0.3 over the life-cycle.
- Cohort heterogeneity:
  - For workers aged 40+ in 2019, share in abstract occupations rose over time; unchanged for earlier ages.
  - Younger cohorts shifted more toward manual jobs compared to earlier cohorts.
  - Interpretation: reallocation away from routine induced “upgrading” for older generations but “downgrading” for younger entrants.

### Employment propensities and wages — estimated effects (Figure 4 summary)
- Period (year) effects (2001–2019):
  - Routine jobs fell by around 13 percentage points between 2001 and 2019.
  - Abstract jobs rose by 10 percentage points in the same interval.
  - Wages: routine and manual jobs show positive trend so that in 2019 wages are more than 15% higher than in 2001; abstract jobs show a 9% increase in 2019 relative to 2001.
  - Business cycle: positive wage growth until 2007 and steep decline during the Great Recession across occupations.
- Cohort effects:
  - Non-college females born since the late 1970s experienced a pronounced fall in the permanent component of real wages across all occupation types.
  - Example: those born in 1990-1994 receive wages that are 14% lower than workers born in 1965-1969 within all occupations.
- Wages by occupation and age:
  - Abstract jobs pay more than routine at every age; manual jobs pay the least.
  - Abstract jobs show steeper wage progression over the career; manual jobs show virtually no progression.

### Decomposition of wage gap changes (2001–2019) and explanatory power
- Observed average wages (LFS) and APC model predictions (values in GBP deflated to 2010 prices):
  - LFS mean wages:
    - W 21–34: 8.15 (2001) → 8.20 (2019) = 0.7% Δ
    - W 35–49: 8.30 (2001) → 9.81 (2019) = 18.2% Δ
    - W 50–64: 7.79 (2001) → 9.60 (2019) = 23.2% Δ
    - Young vs prime gap: 1.45 (2001) → 1.75 (2019)
  - APC Model predictions:
    - W 21–34: 7.87 → 8.53 = 8.3% Δ
    - W 35–49: 8.26 → 9.65 = 16.8% Δ
    - W 50–64: 7.72 → 9.42 = 22.0% Δ
    - Predicted young-prime gap change explained by regressions: 0.73 (of 1.45 observed) = 50% explained.
    - Predicted young-old gap change explained: 1.04 (of 1.22 observed? table shows 60% explained).
- Decomposition within the APC model:
  - Cohort effects explain the majority of the model-explained widening:
    - 77% (0.56 out of 0.73) of the young-prime-age gap explained by the model is due to cohort effects.
    - 85% (0.89 out of 1.04) of the young-old gap explained by the model is due to cohort effects.
  - Within cohort contributions: most driven by cohort effects on wages within occupations; remainder (about one quarter to one fifth) driven by shifts in employment shares.
- Counterfactual series (Figure 5):
  - Year effects (cyclical) similar across age groups on average.
  - Cohort effects drive diverging trends: young non-college females experience a fall in average wages; prime-age stable; older mild rise.
  - Post-2008 cyclical downturn affected younger workers more severely (not captured by model interactions).

### Lifetime earnings and lifecycle projections (Figure 6)
- Cohort effects on present value of lifetime earnings (relative to Cohort 1940):
  - Cohorts up to 1980: cohort effects contributed positively to lifetime earnings.
  - Since Cohort 1965: successive cohorts show a lower cohort effect compared to preceding cohort; cohort effect turns negative for cohorts born since 1985.
- Period (year) effects:
  - Monotonically increasing between 2001 and 2019; assuming 2019 effect persists implies aggregate growth benefits future years of cohorts.
- Decomposed drivers of lifetime effects:
  - Decline in within-occupation wages is the primary driver of reduced lifetime earnings for post-1975 cohorts.
  - Employment propensity (shares across occupations) accounts for roughly one third of the positive cohort contribution until Cohort 1975; after 1975, within-occupation wage declines dominate the fall in lifetime earnings.
  - For period effects, effects on wages are the main driver; shifts toward abstract and manual jobs account for about one tenth of increased lifetime earnings prospects.

### Sensitivity and robustness checks
- Alternative occupation categorization, sample changes (including/excluding foreign-born, excluding self-employed), and additional controls (marital status, number of children, region, part-time, self-employment, contract type) — results hold with minor quantitative changes.
- Expanded time sample back to 1986 (using crosswalks across SOC classifications) adds Cohorts 1930 and 1935; hump-shaped differences accentuated when comparing 1970s cohorts to 1930 cohort.
- Quantile regressions at the 10th, 25th, 50th, 75th, 90th percentiles show period and cohort effects broadly hold across wage distribution.
- Alternative estimation methods (maximum entropy, intrinsic estimator, using proxies for period effects) reported in Online Appendix — robustness maintained.

### Possible drivers of negative cohort effects for recent non-college female cohorts
- Increased university attendance and potential sorting:
  - Rising female university attendance could induce negative selection into non-college group (marginal graduates higher in unobserved skill distribution), lowering average wages of non-college group; empirical evidence mixed.
- Routine-replacing technical change and industry reallocation:
  - Reallocation of low-skill workers across industries uneven by age.
  - Prime-age and older non-college females saw decline in routine jobs concentrated in low-wage sectors (trade, manufacturing).
  - Younger workers experienced routine job decline across a broader set of industries, with reallocation concentrated into relatively low-wage industries such as health and elderly care (partly driven by population ageing).
- Family formation, tax and welfare reforms, and work incentives:
  - 1990s–2000s reforms including Working Families Tax Credit (WFTC) and Income Support changes increased in-work benefits; WFTC increased maximum awards and childcare support; subsidy for working mothers involved a 25 percent rise (in constant wage levels) in maximum award for single mothers of one child and a drop in the withdrawal rate from 70 to 55 percent.
  - Changes shifted childbearing later and increased part-time work, especially among lower-educated women; part-time work tends to entail lower pay, less training, and limited transitions to full-time permanent roles.
- Rise in non-standard work:
  - Increasing self-employment and flexible/temporary contracts (zero-hours): more than a third of zero-hours contract workers were under age 24 in 2017; almost half of zero-hours contracts in low-skilled occupations.
- Declining bargaining power:
  - Young, low-skilled women (single mothers) may face lower bargaining power, self-select into flexible local firms, reducing wages and reinforcing low-wage traps.

### Conclusion and policy implications
- Main findings:
  - Marked worsening of labor compensation for young non-college women in the UK over 2001–2019.
  - Job polarization (period effects) contributed but played a limited role relative to cohort-specific declines in within-occupation wages for recent cohorts.
  - Cohorts born since 1980 experienced a steep rise in manual employment and a fall in routine occupations, and cohort wage effects indicate a slowdown across all occupation types.
  - Within-occupation wage declines for recent cohorts explain most of the aggregate wage gap between young and older non-college workers since 2001.
- Potential long-term consequences:
  - Lower lifetime earnings, lower retirement savings, and longer life expectancies for current young non-college cohorts.
  - Limited wage and occupational progression could exacerbate gender inequality and reduce aggregate productivity.
- Research and policy directions:
  - Further structural modeling to quantify channels: negative selection into non-college group, role of labor market training, flexible employment contracts, family policies.
  - Policy focus implied: addressing the within-occupation wage deterioration for young non-college women, considering training, job quality, bargaining power, and family-support policies to improve long-run earnings trajectories.

*Source: wpiea2022042-print-pdf - Section 8 discusses potential channels driving the lower economic prospects for younger cohorts*

### References

### References

### Key methodological and empirical foundations
- The paper builds on literature linking technological change and task content to employment and earnings dynamics, citing: Acemoglu and Autor (2011); Autor, Levy, and Murnane (2003); Autor and Dorn (2009, 2013).
- Job polarization, routinization, and automation literature referenced includes: Goos and Manning (2007); Bussolo, Torre, and Winkler (2018); Grigoli, Koczan, and Topalova (2020); Nedelkoska and Quintini (2018).
- Age-period-cohort identification and estimation methods referenced include: Browning, Crawford, and Knoef (2013); Yang et al. (2008); Fannon and Nielsen (2019).
- Gender, cohort, and lifecycle labor supply and wage dynamics literature cited includes: Goldin (2006); Goldin et al. (2017); Blundell et al. (2016a, 2016b, 2021); Costa Dias, Joyce, and Parodi (2021); Bhalotra and Fernandez (2018); Olivetti and Petrongolo (2014, 2016).

### Appendix A — Additional tables and figures: substantive notes
- Figure A.1: Unemployment rate and labor force inactivity rate by age group for non-college females in the UK: 1997-2018.
  - Notes: left panel plots unemployment rate by age group; right panel plots Out-of-Labor-Force rate for same groups. Grey shaded area encompasses dates of the Great Recession. Sources: LFS and authors’ calculations.
- Table A.1: Reports p-values from two-sided F-tests of equality of cohort coefficients versus Cohort 1970 for:
  - Employment (Manual, Routine, Abstract) and Log Wages (Manual, Routine, Abstract).
  - Cohort columns: 1940, 1945, 1950, 1955, 1960, 1965, 1970, 1975, 1980, 1985, 1990.
  - Example entries (preserved exactly): Employment Manual row includes "0.00   0.00   0.30   0.45   0.32   0.20-0.00   0.00   0.00   0.00".
  - Note: Table reports p-values for tests that coefficient for respective cohort equals that of Cohort 1970; first three rows for employment regressions as in (2); last three rows for log wages regressions as in (3).

- Figures A.2–A.10: Robustness and alternative estimation visual results reported for:
  - Age polynomials, period effects, and cohort effects for employment shares and for wages (as in Eq. (2) and Eq. (3)).
  - Robustness exercises include: including immigrants; excluding self-employed; adding individual controls; quantile regressions at 10th, 25th, 50th, 75th, and 90th percentiles; extending sample backwards until 1986; alternative occupational categorization; maximum entropy estimation; intrinsic estimator; and proxying period effects using IT equipment relative prices and labor productivity.
  - Notes consistently: Age polynomials for wages reported as exponential of estimated polynomial for log wages; period and cohort effects accompanied by 95 percent confidence intervals or standard errors comprising the 95 percent confidence interval. Sources: LFS and authors’ calculations.

### Appendix B — Data appendix: occupational categorization and alignment
- Occupational classification period used: 2001-2019, working with double-coded SOC2000 variable after 2011 due to changes across SOC 1990, SOC 2000, and SOC 2010.
- Mapping approach used: OCC 2000 → US SOC 2000 → US SOC 2010 → ISCO 08 → UK SOC 2010 → UK SOC 2000. For US crosswalks, when multiple occupations combined into single ISCO-08 code, simple averages used.
- Ad-hoc adjustments using CASCOT-based mapping from Dickerson et al. (2012). Specific adjustments preserved exactly:
  - UK SOC 2010 9232 was originally paired with ISCO 08 9613, which in turn was originally paired with US SOC 373019, which has no information on O*NET. We thus paired it with 474051 based on CASCOT information from Dickerson et al. (2012).
  - UK SOC 2010 3561 and 3565 were originally paired with ISCO 3359 which was originally paired with US SOC 452011 (agricultural inspectors), but the original UK SOC occupations resemble more civil sector clerks. We thus repaired them with the ISCO 08 occupations corresponding to the US SOC 2010 codes that are matched with them in the CASCOT-based mapping from Dickerson et al. (2012).
- Occupational groups defined by adapting Cortes et al. (2020):
  - Routine group formed by grouping routine-cognitive and routine-manual occupations into a single routine group.
  - Manual and abstract correspond to non-routine-manual and non-routine-cognitive categories, respectively.
- When using SOC 1990 waves, mapping used: ISCO 08 → ISCO 88 → ISCO 88 (COM) → SOC 1990; mapping between SOC 1990 and KOS via modal-based matching using LFS 1991 double-coded occupations.
- Figure B.1: Plots shares of routine, abstract, and manual jobs for females without post-secondary education from 1986 to 2019. Notes: series breaks due to occupational classification changes are marked by vertical lines. Source: LFS and authors’ calculations.

- Alignment with Autor et al. (2003) RTI index:
  - RTI formula: RTI = Log(Routine Task) – Log(Abstract Task) – Log(Manual Task).
  - Mapping used: OCC 1990 → OCC 2000 → US SOC 2000 → US SOC 2010 → ISCO 08 → UK SOC 2010 → UK SOC 2000.
  - Table B.1: Summary Statistics of Routine Task Index by Occupation Type in 2001 (preserved exactly)
    - Columns: Males Mean Median 25th pct. 75th pct. | Females Mean Median 25th pct. 75th pct.
    - Routine: "0.72 0.56 -0.31 1.61   2.35 2.51 0.81 3.34"
    - Abstract: "-0.03 -0.45 -0.65 0.53   -0.12 -0.46 -1.15 0.67"
    - Manual: "0.06 0.09 -0.30 0.67   0.45 0.32 -0.27 0.96"
  - Note: Each line reports mean, median, 25th and 75th percentiles of RTI distribution across occupations in each category, separately for males and females. Source: LFS and authors’ calculations.

### Appendix B.2 — Alternative occupational categorization robustness
- Alternative classification follows Bhalotra and Fernandez (2018), merging UK SOC 2000 with task content measures computed by Autor and Dorn (2013).
- Procedure: For each occupation compute percentile in distribution of each task component (routine, manual, abstract); assign occupation to category corresponding to highest percentile.
- Table B.2: Cross-distribution of baseline and alternative categorizations for 2001 and 2018 by worker groups (preserved exactly)
  - Alternative header: 2001 2019
  - Columns: Manual Routine Abstract Manual Routine Abstract
  - Baseline rows (preserved formatting as in source):
    - Manual 83.5 10.9 5.7 82.6 10.2 7.2
    - Routine 6.0 68.7 25.3 6.4 62.6 31.0
    - Abstract 2.8 5.7 91.4 3.9 5.0 91.0
    - All 25.6 41.8 32.6 29.2 32.4 38.5
- Authors report that overall methods are consistent and majority of workers remain in same category; Figure A.7 shows baseline results broadly robust to alternative categorization.

### Appendix C — Alternative estimation approaches and results
- Discussion of identification choices:
  - Zero-sum restrictions on period effects (e.g., Deaton and Paxson (1994)) are not applicable here because year effects are expected to capture long-run structural trends.
  - Browning et al. (2013) "maximum entropy" method uses bounded dependent variables to define identifiable coefficient set and estimates expected values from the most informative distribution within that set.
  - Yang et al. (2008) "intrinsic estimator" extracts effects by regressing dependent variables on transformed variables from eigenvectors corresponding to non-zero eigenvalues of X'X.
  - Proxy approach: period effects proxied by price of IT equipment relative to price of all capital goods for employment shares, and by labor productivity for wages.
    - Prices of IT equipment and capital in the UK over 2001-2017 taken from EU KLEMS 2019 database.
    - Annual labor productivity data taken from Office of National Statistics website.
  - Restriction applied when using IT equipment proxy: estimate (2) simultaneously for three occupation types using Seemingly Unrelated Regression and add restriction that three coefficients on proxy variables sum to zero. Coefficient on labor productivity proxy left unrestricted.

- Results summary:
  - Figures A.8 (maximum entropy) and A.9 (intrinsic estimator): results very similar to baseline; single-year cohorts add noise but overall features unchanged. Linear trends largely captured by period effects; non-linearities in cohort and age effects remain.
  - Figure A.10 (proxying period effects with IT price and labor productivity):
    - Most baseline results carry through.
    - Substantial difference: cohort effects of employment propensities now show pronounced linear trends. Example: cohort effects indicate a fall in probability of routine jobs from Cohort 1940 to Cohort 1990 of almost 30 percentage points.
    - Non-linear components of cohort effects remain similar: routine job cohort effects exhibit concave shape; manual job cohort effects exhibit convex shape.
  - Note on maximum entropy: bounded dependent variable requirement technically unsuitable for wage equation; authors specify very large upper bound above all wage levels in data and report main results hold in practice.

*Source: "Are Low-Skill Women Being Left Behind? Labor Market Evidence from the UK", Working Paper No. WP/22/42 — References and Appendices (figures, tables, and data appendix).

---


_Source: https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022042-print-pdf.pdf_