## Chapter 3 — April 2018 World Economic Outlook (WP/19/191)

## Source details

**Canonical URL:** [Chapter 3 — April 2018 World Economic Outlook (WP/19/191)](https://www.imf.org/-/media/files/publications/wp/2019/wpiea2019191-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2019/wpiea2019191-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2019/wpiea2019191-print-pdf.pdf.json)

---

### I. Research question and empirical approach
- Research question:
  - Did the decline in manufacturing employment drive increases in income inequality in advanced economies between the 1980s and the 2000s?
- Empirical approach:
  - Decomposition exercise using GE(0) (mean log deviation) to separate within- and between-sector inequality.
  - Shift-share analysis and simulations (thought-experiments) with Luxembourg Income Study (LIS) microdata.
  - Focus countries (seven advanced economies with detailed employment data): Austria (1987, 2007), Denmark (1987, 2007), Finland (1987, 2007), France (1989, 2005), Germany (1989, 2007), United Kingdom (1986, 2007), United States (1986, 2007).
- Key measurement choices:
  - Inequality measured by Gini coefficient and generalized entropy GE(0).
  - Sector assignment by household head’s sector (ISIC 3.1): (1) agriculture and fishing; (2) services; (3) manufacturing; (4) other industry (mining, construction, and electricity).

### II. Data, definitions, and controls
- Data source and periods:
  - Luxembourg Income Study (LIS) micro-level harmonized surveys; comparison late 1980s versus 2000s; years after 2008 excluded.
- Income concept and adjustments:
  - Disposable household income = labor income + capital income + social security transfers − income taxes − social security contributions.
  - Equivalization by square root of household members; top-/bottom-coded; person-level adjusted weights (number of household members × household sampling weights).
- Skill classification (ISCO):
  - High skill: ISCO 1 and 2 (managers and professionals).
  - Middle skill: ISCO 3–8, 10.
  - Low skill: ISCO 9 (laborers/elementary).
- Demographic controls used in manufacturing wage-premia regressions:
  - Region, gender, age, education (less than high school; high school; college and above); for the United States only: race (White, African-American, Asian).

### III. Stylized facts and shift-share decomposition findings
- Cross-country patterns:
  - Many countries experienced a rise in disposable income inequality between the late 1980s and the 2000s.
  - United States: most unequal in both periods; about average in sample in terms of change in inequality.
- Shift-share decomposition (GE(0) intertemporal decomposition into four terms per eq.3):
  1. Intertemporal changes in pure within-sector inequality.
  2. Effect of changes in sectoral employment shares on the “within” component.
  3. Effect of changes in sectoral employment shares on the “between” component.
  4. Changes in relative average sectoral income levels.
- Main empirical messages from decomposition (Figures 5 and 6, described in text):
  - Change in inequality mostly driven by changes in pure within-sector inequality (largest component).
  - Changes in sector size (primarily decline in manufacturing) generally contribute toward increasing inequality in almost all countries, but the contribution is small relative to within-sector changes.
  - Effects of sector size on between-sector inequality and changes in average sector income levels typically offset each other by opposing signs and similar magnitudes.
  - Interpretation: pure within-sector inequality likely arises from factors separate from structural transformation (e.g., taxation, social safety nets, unemployment benefits).

### IV. Manufacturing wage premia and evidence on job displacement
- Manufacturing wage premia estimation:
  - Personal-level regressions: ln(wage_i) regressed on sector indicators controlling for skill, education, gender, age, region, race; coefficient β_M for manufacturing relative to services is manufacturing premium.
  - Household-level regressions: dependent variable ln(disposable household income), household assigned sector by household head.
- Empirical magnitudes (summary):
  - United States personal-level manufacturing wage premium declined from around 14 percent to 7 percent since the 1980s.
  - Germany’s personal-level manufacturing premium generally between 8 and 14 percent.
  - Personal-level premium about 6 to 8 percentage points higher than household-level estimates for Germany and the United States.
  - Household-level: in the 1980s U.S. manufacturing household premium notably large—twice the size of the next largest premium (Denmark); only U.S. and Denmark premia statistically different from zero in the 1980s. By the 2000s U.S. premium declined substantially. Other countries’ household-level premia about 4–8 percent.
- Displacement and wage loss evidence (U.S., CPS 1994–2008; summarized in text):
  - Around 55 percent of displaced manufacturing workers who found a new job moved to the service sector.
  - Of those who switched to services and were employed after manufacturing job loss, over 60 percent experienced a decline in wages relative to their manufacturing wage.
  - Among those who worked in manufacturing for more than 10 years before losing their job, around 80 percent experienced a wage decline.
  - Median wage decline among displaced manufacturing workers was around 35 percent; around 45 percent for those with more than 10 years in manufacturing.

### V. Thought-experiment simulations: methodology and baseline results
- Thought-experiment baseline assumptions:
  - Instant structural transformation: all manufacturing job losses between the 1980s and the 2000s realized immediately and displaced workers move to services.
  - All displaced workers are middle-skill manufacturing households reassigned to service sector and set income at the 25th percentile of low-skilled service disposable household incomes in the 1980s.
  - Number of jobs lost = difference in total households employed in manufacturing between 2000s and 1980s (sector by household head).
  - Middle-skill manufacturing households reallocated randomly until reallocation count matches observed job losses; household weights used and scaled to total household population.
  - Move to the 25th percentile interpreted as an unfavorable upper bound for inequality impact.
- Simulation mechanics:
  - For each country compute hypothetical GE(0) under scenario, keeping other households fixed.
  - Confidence intervals from 100 random seeds for household selection.
- Baseline simulation findings (Figure 9, described in text):
  - Except for the United States, the negative effect on inequality attributable solely to manufacturing job loss is very small—negligible compared to actual change in inequality between the 1980s and the 2000s.
  - United States: simulated effect explains around a quarter of the overall increase in inequality from the 1980s to the 2000s.
  - Occasional negative simulated effects (e.g., Denmark, Finland): in very equal societies, adding mass of “poorer” people can—depending on GE(0) weights—make the society overall more equal.

### VI. Sensitivity experiments for the United States
- Scenario 1 — Eliminate manufacturing household wage premium:
  - Reduce initial incomes of all U.S. manufacturing households by the estimated 1980s manufacturing household premium (set manufacturing premium = zero); repeat baseline reallocation.
  - Result: Eliminating the manufacturing household wage premium makes almost no difference compared to baseline simulation.
  - Interpretation: U.S. manufacturing household premium large relative to other countries but small relative to the income loss from moving from median middle-skilled manufacturing wage to the 25th percentile low-skilled service wage.
- Scenario 2 — Compress U.S. initial income distribution toward median:
  - Target average generalized entropy of Finland, Austria, Denmark, and Germany (target GE(0) ≈ 0.1).
  - Iterative compression solved with α = 0.7 producing target generalized entropy ≈ 0.1; compression preserves median and income ordering while moving incomes closer to median.
  - Result: With compressed initial distribution, increase in inequality due to manufacturing job loss approximately halved relative to baseline; manufacturing employment decline would explain about ten percent of the actual increase in U.S. inequality.
  - Interpretation: Initial inequality level is crucial—lower initial inequality substantially reduces the contribution of manufacturing decline to rising inequality.

### VII. Strengths, weaknesses, and broader implications of simulations
- Potential underestimation channels:
  - Assumption that displaced workers immediately find service employment understates costs from unemployment spells; long unemployment spells can be costly and permanent.
  - If increased competition in services depresses low-skill service wages, income losses could be larger.
- Potential overestimation channels:
  - Not all displaced middle-skill manufacturing workers move to low-pay service jobs; evidence suggests about one third of U.S. displaced workers moved to high-skilled jobs.
  - Many reallocated workers may not end at the low end of the low-skill distribution.
- Net implications drawn from thought experiments:
  - Except for the United States, unlikely that manufacturing employment decline was associated with significant increases in inequality in the studied advanced economies.
  - United States is the only country in the sample where manufacturing decline may have been an important contributor to rising inequality; high initial income inequality in the U.S. in the 1980s likely exacerbated this effect.
  - High manufacturing wage premium in the U.S. appears to have played a minor role relative to initial inequality.
  - Emerging markets that are still expanding manufacturing but have high inequality may face larger future inequality challenges when reducing manufacturing employment in favor of services.

### VIII. Main empirical conclusions and policy-relevant takeaways
- Sample and scope:
  - Seven advanced economies (Austria, Denmark, France, Finland, Germany, United Kingdom, United States), late 1980s to 2000s, using LIS household survey data; no years after 2008 considered.
- Core conclusions:
  - Declining manufacturing employment generally contributes to rising inequality, but the contribution is minor relative to other forces affecting inequality.
  - United States exception: baseline simulation indicates about a quarter of the rise in inequality between late 1980s and 2000s may be attributable to manufacturing job losses.
  - Eliminating the manufacturing wage premium in the U.S. would not have significantly reduced that contribution.
  - Compressing U.S. 1980s income distribution to match the average of Finland, Austria, Denmark, and Germany (α = 0.7 to target GE(0) ≈ 0.1) would substantially lessen the impact of manufacturing decline—reducing the contribution to about ten percent of actual increase in inequality.
- Broader considerations beyond income metrics:
  - Income measures understate non-income losses from manufacturing decline: loss of good pensions, health insurance, and job stability in the U.S. exacerbated socio-economic consequences beyond distributional metrics.

*Source: Chapter 3 of the April 2018 World Economic Outlook (IMF Working Paper WP/19/191).*

### Chapter 3 of the April 2018 World Economic Outlook.

### Chapter 3 — April 2018 World Economic Outlook (WP/19/191)

### I. Introduction — question and approach
- Research question: Did the decline in manufacturing employment drive increases in income inequality in advanced economies between the 1980s and the 2000s?
- Empirical approach:
  - Decomposition exercise to identify how the decline in manufacturing employment might have affected within- and between-sector inequality.
  - Simulations (thought-experiments) with Luxembourg Income Study (LIS) microdata to quantify how much of the change in inequality could be attributed to the decline in manufacturing employment.
  - Focus on seven advanced economies with detailed employment data: Austria (1987, 2007), Denmark (1987, 2007), Finland (1987, 2007), France (1989, 2005), Germany (1989, 2007), the United Kingdom (1986, 2007), and the United States (1986, 2007).
- Key measurement choices:
  - Inequality measured using the Gini coefficient and generalized entropy — specifically GE(0) (mean log deviation).
  - Shift-share analysis and decomposition using GE(0), which is decomposable.
  - Sector assignment based on household head’s sector (ISIC 3.1): (1) agriculture and fishing; (2) services; (3) manufacturing; (4) other industry (mining, construction, and electricity).

### II. Data and definitions
- Data source: Luxembourg Income Study (LIS) micro-level harmonized surveys (household- and personal-level).
- Time comparison: late 1980s versus 2000s using the specific country years listed above; years after 2008 excluded to avoid Global Financial Crisis effects.
- Income concept: disposable household income = labor income + capital income + social security transfers − income taxes − social security contributions.
- Equivalization: household income equivalized by the square root of the number of household members; top-/bottom-coded for outliers; inequality measures weighted by person-level adjusted weights (number of household members × household sampling weights).
- Skill classification (ISCO):
  - High skill: managers and professionals (ISCO 1 and 2).
  - Middle skill: other skilled workers (ISCO 3–8, 10).
  - Low skill: laborers/elementary (ISCO 9).
- Demographic controls used in manufacturing wage-premia analysis: region, gender, age, education (less than high school; high school; college and above), and for the United States only: race (White, African-American, Asian).

### III. Stylized facts and shift-share findings
- Cross-country patterns:
  - Many countries experienced a rise in disposable income inequality between the late 1980s and the 2000s.
  - The United States stands out as the most unequal country in both the 1980s and the 2000s, despite being about average in the sample in terms of the change in inequality.
- Methodological note: GE(0) chosen because it is decomposable into within- and between-sector components (mathematical derivations in Appendix A).
- Shift-share decomposition: change in inequality over 20 years decomposed into components including pure within-sector changes in inequality, changes in sector size (structural transformation), and changes in sectoral income levels.
- Observed implication: shift-share analysis can shed light on the relative importance of structural transformation versus economy-wide trends, but cannot identify all factors that produced more equitable distributions in some countries.

### IV. Manufacturing wage premia and U.S. context
- Hypothesized mechanisms linking manufacturing decline to rising inequality:
  - Loss of “manufacturing wage premium”: workers switching from manufacturing to services may incur large income losses if manufacturing wages were substantially higher.
  - Interaction with initial inequality: the same absolute or relative wage loss produces larger effects on measured inequality in countries with higher initial wage dispersion.
- Evidence excerpted in text:
  - In the United States (LIS, 1980s): a median middle-skill manufacturing worker earned 107% of an overall median worker; a 25th percentile low-skill service sector worker earned 50% of an overall median worker.
  - In Finland (LIS, 1980s): a median middle-skill manufacturing worker earned 103% of an overall median worker; a 25th percentile low-skill service sector worker earned 80% of an overall median worker.
- U.S. Current Population Survey (1994–2008) evidence summarized in text:
  - After manufacturing job loss, around 55% of displaced manufacturing workers who found a new job moved to the service sector.
  - Of those who switched to the service sector and were employed after manufacturing job loss, over 60% experienced a decline in wages relative to their manufacturing wage.
  - Among those who worked in manufacturing for more than 10 years before losing their job, around 80% experienced a wage decline.
  - Median wage decline among displaced manufacturing workers was around 35%; around 45% for those with more than 10 years in manufacturing.

### V. Main empirical findings
- Overall contribution of manufacturing decline to inequality change:
  - Across the seven studied countries, the decline in manufacturing employment does not explain a large share of the change in inequality. Factors other than manufacturing are likely more important in explaining changes in overall inequality.
- United States exception:
  - The United States is the country with the highest contribution of manufacturing decline toward rising inequality in the sample.
  - Simulation result: for the United States, about a quarter of the rise in inequality could be attributed to the loss of manufacturing jobs.
- Role of exacerbating factors:
  - This paper examines two factors that might exacerbate the manufacturing–inequality link: the manufacturing wage premium and the initial level of income inequality.
  - Finding: high initial inequality in the United States may have made manufacturing job loss particularly costly, more so than an initial manufacturing wage premium.

### VI. Analytical contributions and scope
- Novelty: first paper (to the authors’ knowledge) to explore how manufacturing wage premia and initial inequality may exacerbate the relationship between manufacturing employment decline and increasing inequality.
- Scope limits:
  - Analysis restricted to seven advanced economies with harmonized LIS surveys in specified years.
  - No years after 2008 considered.
  - Sectoral decomposition uses household-head sector assignment and ISIC 3.1 sector grouping.

*Source: Chapter 3 of the April 2018 World Economic Outlook (IMF Working Paper WP/19/191).*

### Appendix A) for each of the seven advanced economies we study is presented in Figure 3.

### Appendix A) for each of the seven advanced economies we study is presented in Figure 3.

### Within- and between-sector inequality (levels)
- Decomposition covers four sectors: agriculture, manufacturing, services, and other industry (mining, construction, and electricity), plus a “missing” category for unavailable sector information.
- Aggregate country-wide inequality equals the sum of all within- and between-sector components for the entire population.
- Main empirical observation:
  - Most variation in inequality is due to inequality within sectors (red bars), rather than between sectors (blue bars).
  - Between-sector inequality would be large only if some sectors had uniformly high wages and others uniformly low wages.
- Caveat:
  - Figure 3 weights each sectoral component by sector size; therefore it is difficult from Figure 3 alone to infer whether inequality is higher within manufacturing or services because sectoral components sum to total economy-wide inequality.
- Note: The between-sector inequality (blue bar) is the sum of the between components for each sector, including the “missing” category.

### Manufacturing vs. services (unweighted comparison)
- Figure 4 compares manufacturing and services inequality without weighting by sector size.
- Empirical patterns:
  - Inequality in manufacturing and services are strongly correlated across countries: if inequality is high (low) in services it is also high (low) in manufacturing.
  - Observations lie below the 45° line, indicating inequality is somewhat higher in the service sector in all countries.
  - Greater wage dispersion in services is expected because services are typically larger and more diverse than manufacturing.
- Additional data:
  - Appendix Table 3 provides points in the wage distributions for different sectors and skill levels.

### Change in inequality over time: shift-share decomposition
- Method:
  - Changes in generalized entropy (GE(0)) between the 1980s and the 2000s are decomposed (see eq.3 in Appendix A) into four interpretable terms:
    1. Intertemporal changes in pure within-sector inequality.
    2. Effect of changes in sectoral employment shares on the “within” component.
    3. Effect of changes in sectoral employment shares on the “between” component.
    4. Changes in relative average sectoral income levels (Mookherjee and Shorrocks, 1982).
- Key findings (Figure 5 and Figure 6):
  - The change in inequality is mostly due to changes in pure within-sector inequality (the blue bars are the largest).
  - Overall versus within-sector change in inequality aligns closely along the 45° line.
  - Changes in sector size (mostly decline in manufacturing employment in this sample) tend to contribute towards increasing inequality in almost all countries (green and red bars are almost always positive in Figure 5).
  - The effect of changes in sector size on within-sector inequality is generally small (red bars are small).
  - Effects of sector size on between-sector inequality (green bars) and changes in average sector income levels (yellow bars) typically offset each other by being of opposite sign and similar magnitude; thus direct and indirect between-sector changes are not important drivers of overall inequality change over this period.
- Interpretation:
  - Pure within-sector inequality likely arises from factors separate from structural transformation, such as changes in taxation, social safety nets, unemployment benefits, etc.

### Manufacturing wage premia
- Estimation approach:
  - Personal-level regressions: ln(wage_i) regressed on sector indicators controlling for skill-level, education, gender, age, region, race, etc.; the coefficient β_M for manufacturing (measured relative to services) is the manufacturing premium.
  - Household-level regressions: analogous regressions where i indexes households; a “manufacturing household” has a household head employed in manufacturing. Household-level premia are expected to be lower than personal-level premia.
- Empirical magnitudes:
  - United States personal-level manufacturing wage premium declined from around 14 percent to 7 percent since the 1980s.
  - Germany’s personal-level manufacturing premium generally hovered between 8 and 14 percent.
  - Personal-level premium is about 6 to 8 percentage points higher than household-level estimates for Germany and the United States.
  - At household level (Figure 8):
    - In the 1980s the United States manufacturing household premium is notably large—twice the size of the next largest premium (Denmark); only the U.S. and Denmark premia were statistically different from zero in the 1980s.
    - By the 2000s the U.S. manufacturing premium declined substantially.
    - Other countries (United Kingdom, Finland, France, Denmark) develop household-level manufacturing premia of about 4–8 percent.
- Implication:
  - A sizable manufacturing premium can make manufacturing workers reluctant to switch sectors because service-sector jobs (even well-matched ones) would on average offer lower salaries.

### How much could declining manufacturing employment increase inequality? — Thought-experiment framework
- Data constraints:
  - LIS allows cross-country inequality analysis but not longitudinal tracking of individuals; the study uses thought-experiments with 1980s and 2000s cross-sections.
- Baseline scenario assumptions:
  - Instant structural transformation: all manufacturing job losses between the 1980s and the 2000s are realized immediately and displaced workers move to services.
  - All displaced workers are middle-skill manufacturing households who are reassigned to the service sector and set their disposable household income at the 25th percentile of the distribution of low-skilled service sector disposable household incomes in the 1980s.
  - The number of jobs lost is defined as the difference between the total number of households employed in manufacturing in the 2000s and the 1980s (sector assigned by household head).
  - Middle-skill manufacturing households to be reallocated are picked randomly until the reallocation count matches observed job losses; household weights are used and scaled to total household population size.
  - The assumed move to the 25th percentile is intentionally unfavorable for inequality and is interpreted as an upper bound of manufacturing decline’s effect on inequality.
- Simulation methodology:
  - For each country the hypothetical inequality under the scenario is computed using generalized entropy (GE(0)), keeping other households fixed.
  - Confidence intervals on simulations computed from 100 random seeds determining which households are reallocated.
- Baseline simulation results (Figure 9):
  - Two main messages:
    - Except for the United States, the negative effect on inequality attributable solely to manufacturing job loss is very small—negligible compared to the actual change in inequality between the 1980s and the 2000s.
    - In the United States the simulated effect explains around a quarter of the overall increase in inequality from the 1980s to the 2000s.
  - Explanation for occasional negative simulated blue bars (e.g., Denmark, Finland):
    - If a country is very equal, wage differences between low- and middle-skilled workers may be small; increasing the mass of “poorer” people can—depending on distributional weights of GE(0)—make the society overall more equal.

### Sensitivity experiments for the United States
- Scenario 1 — Eliminate manufacturing wage premium:
  - Reduce initial incomes of all U.S. manufacturing households by the estimated 1980s manufacturing household premium (i.e., assume manufacturing premium = zero) and repeat baseline reallocation.
  - Result (Figure 10): Eliminating the manufacturing household wage premium makes almost no difference compared to the baseline simulation.
  - Interpretation: Although U.S. manufacturing household premium in the 1980s was large relative to other countries, its magnitude is small relative to the income loss from moving from the median middle-skilled manufacturing wage to the 25th percentile low-skilled service wage (see Appendix C).
- Scenario 2 — Compress U.S. initial income distribution:
  - Compress 1980s U.S. household incomes toward the median to match average generalized entropy of Finland, Austria, Denmark, and Germany (target GE(0) ≈ 0.1).
  - Compression method:
    - For each U.S. household compute deviation from median income; set new income y_new = median * α + (deviation)*α? (text defines iteratively solving for α); solution found: α = 0.7 produces target generalized entropy ≈ 0.1.
    - Compression preserves median income and ordering of household incomes; all incomes move closer to median.
  - Result (Figure 11): With compressed initial distribution, the increase in inequality due to manufacturing job loss is approximately halved relative to the baseline; manufacturing employment decline would explain only about ten percent of the actual increase in inequality in the United States.
  - Interpretation: Initial inequality level is crucial—lower initial inequality substantially reduces the contribution of manufacturing decline to rising inequality.

### Discussion of simulation strengths and weaknesses
- Potential underestimates of the true negative effect:
  - Assumption that displaced workers immediately find service employment (versus unemployment) understates possible negative effects from unemployment spells; long unemployment spells can be costly and permanent (Walker, 2013).
  - If increased competition in services depresses low-skill service wages, income losses for displaced workers could be larger (Autor, 2015).
- Potential overestimates:
  - Not all displaced middle-skill manufacturing workers necessarily move to low-pay service jobs; evidence suggests about one third of U.S. workers who lost manufacturing jobs moved to high-skilled jobs (Alichi et al., 2013).
  - Many remaining workers who move to service-sector jobs may not end up at the low end of the low-skill distribution.
- Net implications from the thought experiments:
  - Except for the United States, it seems unlikely that manufacturing employment decline was associated with significant increases in inequality in advanced economies.
  - In the sample, the United States is the only country where manufacturing decline may have been an important contributor to rising inequality; high initial income inequality in the U.S. in the 1980s likely exacerbated this.
  - High manufacturing wage premium in the U.S. appears to have played a minor role relative to initial inequality.
  - Emerging markets that are still expanding manufacturing but have high inequality may face larger future inequality challenges when reducing manufacturing employment in favor of services.

### Conclusion (synthesis of findings)
- Sample: seven advanced economies (Austria, Denmark, France, Finland, Germany, United Kingdom, United States), late 1980s to 2000s, using Luxembourg Income Study household survey data.
- Main conclusions:
  - Declining manufacturing employment generally contributes to rising inequality, but the contribution is minor relative to other forces affecting inequality.
  - Exception: United States, where baseline simulation indicates about a quarter of the rise in inequality between late 1980s and 2000s may be attributable to manufacturing job losses.
  - Eliminating the manufacturing wage premium in the U.S. would not have significantly reduced that contribution.
  - Compressing the U.S. 1980s income distribution to match the average of Finland, Austria, Denmark, and Germany (α = 0.7 to target GE(0) ≈ 0.1) would substantially lessen the impact of manufacturing decline—reducing the contribution to about ten percent of actual increase in inequality.
- Broader considerations:
  - Income measures understate non-income losses from manufacturing decline: loss of good pensions, health insurance, and job stability in the U.S. exacerbated socio-economic consequences beyond distributional metrics.

*Source: Authors’ calculations based on the Luxembourg Income Study database, as presented in the specified content unit.*

### REFERENCES

### REFERENCES (wpiea2019191-print-pdf - REFERENCES)

### Key literature cited
- Alichi, A., Mariscal, R., & Muhaj, D. (2017). “Hollowing Out: The Channels of Income Polarization in the United States.” International Monetary Fund Working Paper WP/17/244.
- Austin, B., Glaeser, E., & Summers, L. H. (2018). “Saving the heartland: Place-based policies in 21st century America.” In Brookings Papers on Economic Activity Conference Drafts.
- Autor, D. (2015). “Why are there Still So Many Jobs? The History and Future of Workplace Automation.” Journal of Economic Perspectives 29(3): 3-30.
- Autor, D. & Dorn, D. (2013). “The growth of low-skill service jobs and the polarization of the US labor market.” American Economic Review 103(5): 1553-97.
- Autor, D., Dorn, D., & Hanson, G. H. (2018). “When work disappears: Manufacturing decline and the falling marriage-market value of young men.” IZA Discussion Papers 11465.
- Bárány, Z. L., & Siegel, C. (2018). “Job Polarization and Structural Change.” American Economic Journal: Macroeconomics 10 (1): 57–89.
- Case, A., & Deaton, A. (2017). “Mortality and morbidity in the 21st century.” Brookings Papers on Economic Activity 397.
- Charles, K. K., Hurst, E., & Notowidigdo, M. J. (2016). “Housing Booms, Manufacturing Decline, and Labor Market Outcomes.” The Economic Journal 129(617): 209-248.
- Goos, M., & Manning, A. (2007). “Lousy and lovely jobs: The rising polarization of work in Britain.” The Review of Economics and Statistics 89(1): 118-133.
- Gould, E. (2018). “Explaining the Unexplained: Residual Wage Inequality, Manufacturing Decline, and Low‐Skilled Immigration.” The Economic Journal.
- Helper, S., Krueger, T., & Wial, H. (2012). “Why does manufacturing matters?” Metropolitan Policy Program at Brookings 1775.
- International Monetary Fund (2018) “Manufacturing Jobs: Implications for Productivity and inequality” Chapter 3 of the April 2018 World Economic Outlook, Washington D.C.
- Lawrence, R. Z. (2017). “Recent Manufacturing Employment Growth: The Exception That Proves the Rule.” National Bureau of Economic Research Working Paper w24151.
- Luxembourg Income Study (LIS). “Methodological Notes.” https://www.lisdatacenter.org/data-access/key-figures/methods/ (Accessed January 2019).
- Luxembourg Income Study (LIS) Database. (Multiple countries, multiple years; November 2017 – March 2018). http://www.lisdatacenter.org. Luxembourg: LIS.
- Mookherjee, D., & Shorrocks, A.F. (1982). “A Decomposition Analysis of the Trend in UK Income Inequality.” The Economic Journal 92(368): 886-902.
- Shorrocks, A.F. (1980). “The Class of Additively Decomposable Inequality Measures.” Econometrica 48(3): 613–625.
- Walker, W. R. (2013). “The transitional costs of sectoral reallocation: Evidence from the clean air act and the workforce.” The Quarterly Journal of Economics 128(4): 1787-1835.

### Methodology: Generalized Entropy and shift-share analysis
- Inequality measure: generalized entropy index (mean log deviation, GE(0)).
- GE(0) definition (eq. 1): GE(0) = (1/n) ∑ ln( y_i / y ), where i indexes households, n is total number of households, y_i is income of household i, and y is mean income.
- Decomposition (eq. 2): GE(0) = ∑ s_k GE(0)_k + ∑ s_k ln(1/π_k), where s_k = e_k / E is employment share of sector k, and π_k = y_k / y is relative mean income of sector k. (Sector of employment of household head used for sectoral inequality.)
- Intertemporal decomposition (eq. 3) yields four terms interpreted as:
  - (1) intertemporal changes in pure within-sector inequality;
  - (2) effect of changes in sectoral employment shares on the “within” component;
  - (3) effect of changes in sectoral employment shares on the “between” component;
  - (4) changes in the relative average sectoral income levels (Mookherjee and Shorrocks 1982).
- Generalized entropy general formula (footnote 17): GE(α) = (1/n) ∑ [(y_i / y)^{α} - 1] / [α(α-1)] when α ∉ {0,1}; when α=0 GE given by eq.1.

### Regression results: manufacturing wage premia (appendix B)
- Table 1 (personal level; dependent variable = ln(gross hourly wage)):
  - Sector: Manufacturing coefficients: 0.122 ***, 0.129 ***, 0.144 ***, 0.104 *** (standard errors: (0.018), (0.018), (0.006), (0.007))
  - Sector: Other Industry coefficients: 0.076 ***, 0.063 **, 0.161 ***, 0.122 *** (standard errors: (0.026), (0.026), (0.011), (0.009))
  - Sector: Agriculture coefficients: -0.492 **, -0.218 ***, -0.405 ***, -0.280 *** (standard errors: (0.213), (0.081), (0.031), (0.032))
  - Skill: High coefficients: 0.201 ***, 0.520 ***, 0.373 ***, 0.514 *** (standard errors: (0.048), (0.047), (0.013), (0.012))
  - Skill: Medium coefficients: 0.012, 0.274 ***, 0.145 ***, 0.205 *** (standard errors: (0.033), (0.043), (0.011), (0.011))
  - Education: High coefficients: 0.431 ***, 0.464 ***, 0.488 ***, 0.566 *** (standard errors: (0.042), (0.036), (0.011), (0.011))
  - Age coefficients: 0.096 ***, 0.095 ***, 0.067 ***, 0.058 *** (standard errors: (0.006), (0.005), (0.002), (0.001))
  - Age^2 coefficients: -0.001 ***, -0.001 ***, -0.001 ***, -0.001 *** (standard errors: (0.000) across)
  - Sex: Male coefficients: 0.138 ***, 0.140 ***, 0.318 ***, 0.236 *** (standard errors: (0.018), (0.017), (0.006), (0.005))
  - Region Fixed Effects: Yes, Yes, Yes, Yes
  - Race Fixed Effects: No, No, Yes, Yes
  - Number of Observations: 73,648 4,127 6,865 51,584 (presented in that order)
  - R2 (as presented): 0.310.490.450.32
  - Note: observations with negative gross hourly wage excluded; outliers top-coded at ten times the median; sample restricted to household members employed full time; personal sampling weights used; robust standard errors in parenthesis.
- Table 2 (household level; dependent variable = ln(disposable household income), equivalized):
  - Sector: Manufacturing sample of coefficients (selected as presented): 0.017, 0.018, 0.033 **, 0.049 ***, 0.006, 0.092 ***, 0.018, 0.058 ***, 0.005, 0.028, 0.014, 0.042 ***, 0.079 ***, 0.045 *** (standard errors vary; first few: (0.012), (0.029), (0.014), (0.005), (0.010), (0.014), (0.020), (0.016), (0.020), (0.019), (0.031), (0.014), (0.008), (0.010))
  - Sector: Other Industry sample coefficients include: -0.006, -0.076, 0.024, 0.044 ***, 0.002, 0.052 ***, -0.105 ***, -0.008, -0.041, 0.018, -0.016, 0.050 **, 0.017, 0.018 (standard errors presented in table).
  - Sector: Agriculture sample coefficients include: -0.020, -0.055, -0.299 ***, -0.380 ***, -0.103 ***, 0.048 **, -0.232 ***, -0.137 ***, -0.067, -0.033, -0.251 ***, -0.328 ***, -0.422 ***, -0.062 ** (standard errors presented).
  - Skill: High coefficients (selected): 0.295 ***, 0.276 ***, 0.008, 0.237 ***, 0.093 ***, 0.331 ***, 0.271 ***, 0.420 ***, 0.261 ***, 0.467 ***, 0.355 ***, 0.386 ***, 0.297 ***, 0.408 *** (standard errors vary).
  - Skill: Medium coefficients (selected): 0.166 ***, 0.149 ***, -0.003, 0.104 ***, 0.007, 0.147 ***, 0.137 ***, 0.145 ***, 0.101 ***, 0.245 ***, 0.137 ***, 0.176 ***, 0.122 ***, 0.171 ***.
  - Education: High coefficients (selected): 0.210 ***, 0.212 ***, 0.132 ***, 0.103 ***, 0.223 ***, 0.223 ***, 0.315 ***, 0.314 ***, 0.352 ***, 0.279 ***, 0.000 ***, 0.421 ***, 0.523 ***, 0.619 ***.
  - Education: Medium coefficients (selected): 0.172 ***, 0.124 **, 0.034 ***, 0.050 ***, 0.054 ***, 0.090 ***, 0.088 ***, 0.151 ***, 0.196 ***, 0.093 ***, 0.000 ***, 0.159 ***, 0.285 ***, 0.344 ***.
  - Age, Age^2, Sex: Male coefficients presented with standard errors.
  - Country coverage and observations (examples):
    - Austria 1987 2007; Denmark 1987 2007; France 1989 2005; Germany 1989 2007; Finland 1987; United Kingdom 1986 2007; United States 1986 2007.
    - Number of observations: 50,585; 5,262; 2,802; 73,648; 4,127; 6,865; 51,584; 43,223; 8,161; 7,230; 5,296; 6,321; 2,954; 5,889; 4,599; 14,059; 38,791 (as presented in table rows).
  - R2 examples shown: 0.12 0.24 0.15 0.17 0.20 0.25 0.12 0.25 0.18 0.24 0.05 0.17 0.25 0.32
  - Note: regressions estimated using household-level files; households equivalized by square root of household members; households assigned sector using household head; low skill, low education, and service variables excluded to avoid multicollinearity; weighted by household sampling weights times household members; robust standard errors in parenthesis.

### Wages by sector and skill (appendix C)
- Appendix Table 3 presents percentiles (p10, p25, p50, p75, p90) of the wage distribution expressed as percent of overall median (multiplied by 100), for:
  - economy overall, manufacturing, and services;
  - by skill level: High, Medium, Low;
  - for the 1980s and 2000s;
  - for countries: USA, France, Germany, UK, Austria, Denmark, Finland.
- Examples and interpretation provided in text:
  - Example: "in the U.S. in the 1980s a median middle-skilled worker in manufacturing earns 107% of the overall median, and the 25th percentile worker in low skilled services earns 50% of the overall median wage in the overall economy. Hence, the wage cut shown in Figure 2 for the U.S. in 1986 is 107-50=57."
- Representative percentile entries (as presented in table fragments):
  - Overall (example rows): 56 76 100 130 162  |  51 71 100 137 183
  - High skill (example rows): 113 125 178 221 288  |  63 76 133 161 221
  - Medium skill (example rows): 71 85 105 132 161  |  62 80 105 138 178
  - Low skill (example rows): 61 78 93 112 145  |  58 75 98 122 144
  - Country-specific panels include numerous percentile vectors for manufacturing and services in 1980s and 2000s (see Appendix Table 3 for full numeric grid).

### Inequality and possible wage cuts during structural transformation (appendix D)
- Definition of "possible wage cut" used in simulation and figures:
  - Assume a middle-skill manufacturing sector worker at the 50th percentile for manufacturing middle-skill income switches to service sector and, due to no prior experience, is assigned the wage at the 25th percentile of the low-skill service sector wage distribution.
  - Wages expressed as percent of the overall median income in the country.
  - Formula used: [(50th percentile of manufacturing middle-skill income) – (25th percentile of service sector low-skill income)] / (50th percentile of overall income)] * 100.
- Appendix Figure 1: plots overall level of inequality against the possible wage cut for advanced countries available in LIS (based on above calculation).
- Appendix Figure 2: same relationship including emerging market economies available in LIS (to illustrate how structural transformation from manufacturing to services might develop in emerging market economies).
- Source for these calculations: Authors’ calculations based on the Luxembourg Income Study database.

*Source: wpiea2019191-print-pdf - REFERENCES (appendix material).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2019/wpiea2019191-print-pdf.pdf_