## Is Education Neglected in Natural Resources-Rich Countries? An Intergenerational Approach in Africa

## Source details

**Canonical URL:** [Is Education Neglected in Natural Resources-Rich Countries? An Intergenerational Approach in Africa](https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022160-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2022/english/wpiea2022160-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2022/english/wpiea2022160-print-pdf.pdf.json)

---

### Overview and context
- Sample and scope:
  - Sample: more than 14 million individuals from 28 African countries and 2,890 districts.
  - Main data sources: Integrated Public Use Microdata Series (IPUMS) and Minex Consulting datasets.
  - Mineral discoveries: 969 normal to super-giant mineral discoveries in Africa between 1950 and 2019; 396 of them (40 percent) since 2000.
  - Study coverage: 406 mineral discoveries in 28 African countries over 1950–2000 (71 percent of the 573 discoveries between 1950 and 2000).
  - After merging: 331 districts out of 2,890 identified where mineral sites were discovered or entered production.
- Regional and education context:
  - Sub-Saharan Africa living standards growth: about 3.4 to 3.7 percent per annum.
  - Gross enrollment changes since 1970:
    - Primary: from 54 percent in 1970 to 99 percent in recent years.
    - Secondary: from 13 percent in 1970 to 43 percent in recent years.
    - Tertiary: from 1.4 percent in 1970 to 9.4 percent in recent years.
  - Mineral endowment: Africa hosts 30 percent of the world’s mineral reserves, 40 percent of the world’s gold and up to 90 percent of some minerals like chromium and platinum.
  - Mining contribution (Sub-Saharan countries, 2009–19): mining sector accounted for 8.8 percent of GDP and 51.2 percent of total exports.

### Research question and empirical approach
- Objective:
  - Analyze effects of mineral discoveries and productions on educational intergenerational mobility (IM) across 28 African countries and 2,890 districts.
- IM measures (absolute, conditional):
  - Upward primary IM: probability that a child from parents with less than primary education attains at least primary education.
  - Downward primary IM: probability that a child from parents with at least primary education attains less than primary education.
  - Upward secondary/tertiary IM: probability that a child from parents with at most primary education attains at least secondary education.
  - Downward secondary/tertiary IM: probability that a child from parents with at least secondary education attains primary or less.
- Identification and estimation:
  - Quasi-natural experiment using generalized difference-in-differences (GDID).
  - Quasi-exogeneity arguments: unpredicted timing of discoveries, unpredicted geographic location, lag between discovery and production.
  - Dynamic checks: leads-and-lags model following Angrist and Pischke (2009) and Maurer (2019) to test parallel trends and dynamic effects.
  - Baseline treatment window: 30 years around event (individuals born 15 years before and after discovery/production); robustness windows: 40, 50, 60 years.
  - Controls and fixed effects: district FE, cohort FE (decade baseline or year-of-birth in robustness), census-year FE, individual and household controls.

### Main empirical findings — primary education
- Average treatment effects:
  - Probability of upward primary educational IM increases by 0.027*** (2.7 pp.) following mineral discoveries.
  - Probability of upward primary educational IM increases by 0.067*** (6.7 pp.) following mineral productions.
  - Probability of downward primary educational IM decreases by -0.012*** (-1.2 pp.) following both mineral discoveries and productions.
- Dynamic effects (selected):
  - Born 0–5 years after discovery/production: upward 0.032*** (3.2 pp.) for discovery and 0.075*** (7.5 pp.) for production; downward -0.014*** and -0.013***, respectively.
  - Born 6–10 years after: upward 0.042*** and 0.076***; downward -0.015*** and -0.020***.
  - Born 11–15 years after: upward 0.062*** and 0.153***; downward -0.029*** and -0.028***.
- Extrapolated population impacts (Africa, 1950–2000, individuals born up to 15 years after event):
  - Individuals who completed at least primary education while their parents had not:
    - Increase by 662 thousand after mineral discoveries.
    - Increase by 581 thousand after mineral productions.
  - Individuals who did not complete at least primary education while their parents had completed it:
    - Decrease by 371 thousand after mineral discoveries.
    - Decrease by 124 thousand after mineral productions.
  - Note: figures consider only individuals born up to 15 years after discovery/production.

### Main empirical findings — secondary and tertiary education
- Overall:
  - Effects of mineral discoveries and productions on upward and downward secondary/tertiary educational IM are generally not statistically significant across specifications.
  - One specification shows slight significance at the 10 percent level, but there is no consistent evidence of an effect overall.
- Heterogeneity:
  - By region and discovery size, secondary/tertiary effects are heterogeneous:
    - Giant and super-giant discoveries show statistically significant positive effects on upward secondary/tertiary IM in some specifications.
    - Regional differences: mining reduces upward secondary/tertiary IM in Eastern and Northern Africa (negative significant) and increases upward secondary/tertiary IM in Southern Africa and Western & Central Africa (positive significant).

### Heterogeneity and subgroup results
- By region (Eastern, Northern, Southern, Western & Central):
  - Primary upward IM: mining positive and strongly significant across regions.
  - Primary downward IM: mining reduces downward IM in Eastern Africa and Northern Africa (negative and significant), not consistently elsewhere.
- By size of discoveries (Minex categories):
  - Moderate and major discoveries: positive and significant increases in upward primary IM and reductions in downward primary IM.
  - Giant and super-giant: mixed; giant/super-giant can show positive effects at secondary/tertiary levels.
- By gender:
  - Positive primary IM effects are higher for males than females (especially for production-related effects).
  - Example: probability of upward primary IM after production: males ≈ 8.4 percent vs females ≈ 4.9 percent.
- By urban-rural residency:
  - Positive effects on primary IM larger in urban areas than rural areas.
  - Secondary/tertiary effects significant in urban areas but not in rural areas.

### Transmission channels — tested mechanisms
- Income channel (parents working in mining):
  - Interaction results:
    - Having one parent working in mining raises likelihood of upward primary IM by 0.022** (2.2 pp.) following discoveries and by 0.063*** (6.3 pp.) following productions.
    - Interaction reduces likelihood of downward primary IM (example reported -0.011**).
  - Implies parental employment in mining is a measurable pathway.
- Returns-to-education channel (Wealth index and LIDO score):
  - Mincer-type GDID results:
    - Primary completed → LIDO: 0.021***; → Wealth index: 0.041***.
    - Secondary/tertiary completed → LIDO: 0.048***; → Wealth index: 0.097***.
  - Interaction with mining raises returns: Primary completed X Yes Mining yields positive and significant coefficients for both LIDO and Wealth index.
  - Second-step linking returns to IM:
    - LIDO score 0-1 coefficient for upward primary IM: 0.592***.
    - Wealth index 0-1 coefficient for upward primary IM: 0.433***.
    - LIDO and Wealth index negatively associated with downward primary IM: -0.352*** and -0.252***, respectively.
- Sectoral reallocation:
  - Exposure to discoveries/productions decreases likelihood to work in agriculture by 0.011 (−1.1 pp.) for exposed individuals.
  - Increases likelihood to work in manufacturing by around 0.8–1.4 pp. and in services by around 0.8–1.4 pp., depending on specification.

### Robustness and sensitivity
- Robustness checks that preserve main findings:
  - Alternative control groups (only countries/districts with discoveries): primary IM results remain significant; secondary/tertiary remain insignificant.
  - Alternative fixed-effects: birth-year FE, common time trend, district × cohort FE — main findings unchanged.
  - Alternative time windows: 40-year, 50-year, 60-year windows — primary IM effects remain significant.
  - Alternative treatment exposure: include individuals born 10 and 15 years before events — baseline findings confirmed.
  - Alternative IM definitions (parental benchmarks, min/max parental education): mining increases upward and reduces downward primary IM; secondary/tertiary insignificant.
  - Using all mineral discoveries and productions (not just first events): primary IM effects persist.
  - Adding conflicts as control (district-level conflict >25 deaths): conflict associated with lower upward primary IM (-0.009***); inclusion does not alter mining effects on primary IM.

### Stylized facts and trends
- Cohort trends (1950s to 1990s):
  - Upward primary IM increased from 35.1 percent (1950s cohort) to 57.7 percent (1990s cohort).
  - Downward primary IM decreased from 29.8 percent to 23 percent between the 1950s and 1990s cohorts.
  - Upward secondary/tertiary IM increased from 10.8 percent to 32.9 percent.
  - Downward secondary/tertiary IM decreased from 45.1 percent to 35.8 percent.
- Country- and district-level variation:
  - Upward primary IM ranges from 13 percent in South Sudan to 98 percent in Mauritius.
  - District-level averages (Total districts: 2,888):
    - Upward primary IM mean: 0.562, cv: 0.520.
    - Downward primary IM mean: 0.259, cv: 0.861.
  - Districts with discoveries generally show higher upward primary IM by around 4 pp. on average and lower downward primary IM by around 4 pp. on average (period-wide means).

### Interpretation and relative magnitude
- Effect size interpretation:
  - Positive impacts on primary educational IM are statistically significant but relatively small compared to cohort-driven increases in IM, implying other factors have contributed substantially to IM improvements in Africa.
- Scope and limits:
  - Results for secondary/tertiary IM are non-significant overall though heterogeneous by region, size, gender, and residency.
  - Population extrapolations limited to cohorts born up to 15 years after events.

### Policy implications
- Resource management and distribution:
  - Better management of mineral resources by governments and companies to ensure extraction benefits are captured for broader development goals.
  - Consider channeling mining revenues into a fund and redistributing among districts to reduce regional disparities and broaden benefits beyond district-level effects.
- Labor and enterprise policies:
  - Promote job creation and facilitate enterprise development linked to mining activities (labor market flexibility, business linkages between mining companies and local SMEs).
- Equity and access:
  - Targeted policies to reduce inequality of opportunities across gender and urban-rural divides following mineral discoveries; ensure equitable access to mining benefits independent of gender and location.
- Education and human capital:
  - Leverage mining-driven economic opportunities to bolster primary education attainment and intergenerational mobility, while addressing gaps in secondary and tertiary pathways.

*IMF Working Paper No. WP/2022/160 — Is Education Neglected in Natural Resources-Rich Countries? An Intergenerational Approach in Africa (authors’ calculations based on IPUMS dataset and Minex Consulting dataset (2019)).*

### References .............................................................................................................

### References (Excerpt) — Is Education Neglected in Natural Resources-Rich Countries? An Intergenerational Approach in Africa

### Overview and context
- Study scope and data:
  - Sample: more than 14 million individuals from 28 African countries and 2,890 districts.
  - Main data sources: Integrated Public Use Microdata Series (IPUMS) and Minex Consulting datasets.
  - Timeframe and events: mineral discoveries between 1950 and 2019 (Minex Consulting database reports 969 normal to super-giant mineral discoveries in Africa between 1950 and 2019, and 396 of them (40 percent) since 2000).
- Regional context and education trends:
  - Sub-Saharan Africa living standards growth: about 3.4 to 3.7 percent per annum (Young, 2012).
  - Gross enrollment changes since 1970:
    - Primary: from 54 percent in 1970 to 99 percent in recent years.
    - Secondary: from 13 percent in 1970 to 43 percent in recent years.
    - Tertiary: from 1.4 percent in 1970 to 9.4 percent in recent years.
  - Mineral endowment: Africa hosts 30 percent of the world’s mineral reserves, 40 percent of the world’s gold and up to 90 percent of some minerals like chromium and platinum.
  - Mining contribution: in Sub-Saharan countries over 2009–19, the mining sector accounted for 8.8 percent of GDP and 51.2 percent of total exports.

### Research question and empirical approach
- Objective:
  - Analyze effects of mineral discoveries and productions on educational intergenerational mobility (IM) across 28 African countries and 2,890 districts.
- IM measure:
  - Conditional absolute measure of educational IM as in Alesina et al. (2021).
- Identification strategy:
  - Quasi-natural experiment using generalized difference-in-differences (GDID).
  - Quasi-exogeneity arguments: unpredicted timing of discoveries, unpredicted geographic location, and lag between discovery and production.
  - Dynamic checks: leads and lags model following Angrist and Pischke (2009) and Maurer (2019) to test parallel trends and dynamic effects.

### Main empirical findings
- Effects on primary educational IM:
  - Probability of upward primary educational IM increases by 2.7 pp. following mineral discoveries.
  - Probability of upward primary educational IM increases by 6.7 pp. following mineral productions.
  - Probability of downward primary educational IM decreases by 1.2 pp. following both mineral discoveries and productions.
- Effects on secondary and tertiary educational IM:
  - Effects of mineral discoveries and productions on the probability of upward and downward secondary and tertiary educational IM are not statistically significant.
- Extrapolated population-level impacts (Africa, 1950–2000, individuals born up to 15 years after event):
  - Individuals who completed at least primary education while their parents had not:
    - Increase by 662 thousand after mineral discoveries.
    - Increase by 581 thousand after mineral productions.
  - Individuals who did not complete at least primary education while their parents had completed it:
    - Decrease by 371 thousand after mineral discoveries.
    - Decrease by 124 thousand after mineral productions.
  - Note: figures consider only individuals born up to 15 years after discovery/production; larger totals would result if all cohorts after events were included.
- Heterogeneity and sensitivity:
  - Effects vary by African region and size of mineral discoveries.
  - Positive effects are higher for males than females (only for primary education).
  - Positive effects are larger for individuals in urban than rural areas (for both primary and secondary/tertiary education).
  - Dynamic results: positive effects on primary IM increase for individuals born later after discovery and start of production; effects non-significant or low for individuals born before discovery/production, supporting parallel trend assumption.
- Robustness:
  - Baseline results robust to several robustness checks reported in the paper.

### Proposed channels
- Income effect:
  - Mining sector creates new job and income opportunities for parents, enabling greater investment in children’s education (Becker and Tomes, 1979).
  - Empirical evidence: parents working in the mining sector proxy.
- Returns-to-education effect:
  - Economic dynamism and new jobs following discoveries increase demand for skilled workers, thereby boosting returns to education (Torche, 2014).
- Additional channels tested (summary note):
  - Reallocation of individuals across broad sectors (agriculture, manufacturing, services): population exposed to discoveries/productions more likely to work in manufacturing and services.
  - Provision of infrastructure (proxy: access to electricity and clean water) tested but not presented due to proxy limitations; results available upon request.

### Interpretation and relative magnitude
- Effect size interpretation:
  - Positive impacts on primary educational IM are statistically significant but relatively small compared to cohort-driven increases in IM, implying other factors have contributed substantially to IM improvements in Africa.
- Scope and limits:
  - Results for secondary/tertiary IM are non-significant.
  - Population extrapolations limited to cohorts born up to 15 years after events.

### Policy implications (as discussed in the paper)
- Resource management and distribution:
  - Call for better management of mineral resources by governments and companies.
  - Policies to ensure extraction benefits are captured for broader development goals.
- Labor and enterprise policies:
  - Promote job creation and facilitate enterprise development linked to mining activities.
- Equity and access:
  - Ensure equitable access to mining benefits independent of gender and location.
- Education and human capital:
  - Leverage mining-driven economic opportunities to bolster primary education attainment and intergenerational mobility, while addressing gaps in secondary and tertiary pathways.

*IMF Working Paper: Is Education Neglected in Natural Resources-Rich Countries? An Intergenerational Approach in Africa*

### 1.3 percent. In contrast, some papers find that mining activities can create some environmental issues by

### IMF WORKING PAPERS Is Education Neglected in Natural Resources-Rich Countries? An Intergenerational Approach in Africa

### Natural resources and education
- Literature is inconclusive on whether natural resources improve or worsen education; studies document both negative and positive channels.
- Negative channels documented:
  - Resource abundance can delay industrialization and lower education levels (Leamer et al., 1999).
  - Natural resources can crowd out investments in education, decrease public education expenditures relative to GDP, and reduce educational attainment (Gylfason, 2001; Cokx and Francken, 2016; Ahlerup et al., 2020).
- Positive channels documented:
  - Natural resources can increase public spending in education and human capital accumulation (Stijns, 2006; Kim and Lin, 2017; Pegg, 2010).
  - Mining activity may raise household income, enabling greater investment in children’s education (Becker and Tomes, 1978; Weber-Fahr et al., 2002; Loayza et al., 2013).
  - Resource-driven structural transformation can increase returns to education (Torche, 2014; Bütikofer et al., 2018).
  - Revenues from resources may finance education infrastructure (Witter and Jakobsen, 2017).
- Heterogeneity in effects:
  - Effects differ by quantity vs. quality of education, education level, age, and gender (Farzanegan and Thum, 2020; Gradstein and Ishak, 2020).
  - Oil rents can increase government spending in primary and secondary education but reduce quality (Farzanegan and Thum, 2020).
  - Oil price booms in early childhood (ages 0-4) can enhance attainment, but booms in adolescence (ages 10-14) can reduce attainment, especially for girls (Gradstein and Ishak, 2020).

### Data sources and construction of educational IM
- Data sources:
  - Integrated Public Use Microdata Series (IPUMS): covers 82 national census surveys from 28 African countries, containing information on more than 130 million individuals. Sample focus: individuals aged between 16 and 50, born between 1950 and 2000. Final sample: more than 14 million individuals across 2,890 districts.
  - Minex Consulting Dataset (2019): geolocated mineral discoveries with size (moderate, major, giant, super-giant), status (closed, feasibility study, operating, underdeveloped), and mineral type.
- Mineral discovery and production coverage:
  - 969 mineral discoveries in Africa between 1950 and 2019.
  - 573 (60 percent) of these between 1950 and 2000.
  - This study covers 406 mineral discoveries in 28 African countries over 1950–2000 (i.e., 71 percent of the 573 mineral discoveries).
  - After merging, 331 districts out of 2,890 are identified where mineral sites were discovered or entered production.
- Spatial and categorical distributions of discoveries (1950–2019):
  - Regional shares: Southern Africa 48.3 percent; Western and Central Africa 34.2 percent; Eastern Africa 11.9 percent; Northern Africa 5.7 percent.
  - Size shares: moderate 45.3 percent; major 29.8 percent; giant 21.7 percent; super-giant 3.2 percent.
  - Mineral types: gold 34 percent; bulk metals 18.4 percent; precious minerals 15.8 percent; base metals 15 percent.
- Construction of absolute educational intergenerational mobility (IM) measures:
  - Upward primary IM: probability that a child from parents with less than primary education attains at least primary education.
  - Downward primary IM: probability that a child from parents with at least primary education attains less than primary education.
  - Upward secondary/tertiary IM: probability that a child from parents with at most primary education attains at least secondary education.
  - Downward secondary/tertiary IM: probability that a child from parents with at least secondary education attains primary or less.
  - Parents’ benchmark is the average of biological/step parents’ attainment rounded to the nearest integer; alternative benchmarks (min/max, immediate older generation, extended relatives) used in robustness checks.
  - Binary indicator variables P and ST and averaged parental measures 푃푃 and 푃푆푇 are used to define IM indicators (see equations (1)–(4) in the text).

### Cohabitation-selection issues and diagnostics
- Cohabitation-selection concern:
  - Sample includes both youth and adults; cohabitation with parents affects measured attainment and cohabitation patterns vary with age and gender.
  - Adults have more intense self-selection; patrilocal marriage patterns can accentuate selection for women.
- Empirical checks:
  - Cohabitation rates by age and gender are similar in districts with and without discoveries, suggesting limited bias from differential cohabitation patterns.
  - Table 1 analysis: the unconditional likelihood of not completing primary education is higher for individuals not co-residing with biological/step parents than for those co-residing. Differences are significant with Khi-2 tests p-value 0.000.
  - Implication: if cohabitation selection biases exist, estimates of the effects of discoveries on upward (downward) IM would be downward (upward) biased, so estimated positive effects of discoveries should be considered lower bounds.

### Stylized facts and empirical findings on educational IM
- Trends by cohort (1950s to 1990s):
  - Upward primary IM increased from 35.1 percent (1950s cohort) to 57.7 percent (1990s cohort).
  - Downward primary IM decreased from 29.8 percent to 23 percent between the 1950s and 1990s cohorts.
  - Upward secondary/tertiary IM increased from 10.8 percent to 32.9 percent.
  - Downward secondary/tertiary IM decreased from 45.1 percent to 35.8 percent.
  - Upward (downward) IM is higher (lower) at primary level than at secondary/tertiary level.
- Gender and residency patterns:
  - Both males and females show increased upward IM and decreased downward IM over time; gender gaps have narrowed and in recent cohorts females often outperform males for secondary/tertiary upward IM.
  - Educational IM has consistently been higher in urban areas than rural areas; residency gaps have diminished but remain significant.
- Country-level ranges and disparities:
  - Upward primary IM ranges from 13 percent in South Sudan to 98 percent in Mauritius.
  - Downward primary IM ranges from close to zero (Egypt, Mauritius, Botswana) to 58 percent in South Sudan.
  - Upward secondary/tertiary IM ~14 percent in Sierra Leone and Sudan to more than 70 percent in Egypt and Nigeria.
  - Downward secondary/tertiary IM ranges between 1 percent in Egypt and 71 percent in Togo.
  - Significant within-country and across-district heterogeneity documented; upward primary IM more unequal in South Sudan, Sudan, Ethiopia, Burkina Faso; less unequal in Mauritius, South Africa, Zimbabwe.
- District-level and discovery-related stylized facts:
  - Mapping across 2,890 districts shows large within-country variation; some district-level IM estimates can be unstable with small sample sizes.
  - Country case: South Africa
    - Total discoveries in sample: 108.
    - In the 60 districts with mineral sites, upward IM was 92 percent (primary) and 46 percent (secondary/tertiary), higher by 2–3 pp. than in 156 districts without discoveries.
    - Downward IM was 2 percent (primary) and 25 percent (secondary/tertiary) in districts with discoveries, lower by 1–2 pp. than in districts without discoveries.
  - Mean differences across districts with vs. without discoveries (period-wide averages):
    - Upward (downward) IM for primary education is on average higher (lower) in districts with discoveries than in districts without discoveries by around 4 pp.
    - For secondary education the opposite holds: upward (downward) IM is on average lower (higher) in districts with discoveries than in districts without discoveries by around 4–6 pp.
  - Dynamics by cohort and discovery status:
    - For older cohorts (1950s–1960s), upward IM was lower in districts with discoveries; for more recent cohorts (1980s–1990s), upward IM in districts with discoveries increased and closed or overtook the gap with districts without discoveries.
    - Downward IM showed the reverse pattern: higher in discovery districts for old cohorts, gap narrowed over time.
    - Conclusion: IM has been more dynamic in districts with discoveries, suggesting mineral discoveries and productions contributed to changing the geography of opportunities across regions.
  - Gender dynamics in discovery vs. non-discovery districts:
    - Improvements in IM occurred for both sexes; females often did better than males in districts with discoveries for recent cohorts (gender gap in upward primary IM closed earlier in districts with discoveries).
- Summary interpretation from stylized facts:
  - Mineral discoveries are associated with faster improvements in intergenerational educational mobility in affected districts over time, with important heterogeneity by education level, cohort, gender, and residency.

*Source: IMF Working Paper — Is Education Neglected in Natural Resources-Rich Countries? An Intergenerational Approach in Africa.*

### Appendix D). However, the gap between urban and rural areas has remained significant despite greater

### IMF WORKING PAPERS Is Education Neglected in Natural Resources-Rich Countries? An Intergenerational Approach in Africa

### Empirical methodology
- Identification strategy: exploit exogeneity of timing and location of mineral discoveries and a GDID (generalized difference-in-differences) model comparing treated (exposed) and control groups across pre-discovery/production and post-discovery/production periods.
- Treatment windows and definitions:
  - Baseline window: 30 years around event (individuals born 15 years before and after the discovery/production).
  - Robustness windows: 40 years, 50 years, 60 years around the discovery/production.
  - Alternative treatment definitions:
    - Treatment = individuals born the year of discovery/production up to 15 years after.
    - Alternative: include individuals born 5 years before discovery/production up to 15 years after.
    - Further robustness: include individuals born 10 and 15 years before discovery/production (using 60-year window).
- Model controls and fixed effects:
  - District fixed-effects 훼i, cohort (decade baseline or year-of-birth in robustness) fixed-effects 훾t, census-year fixed-effects 훿t.
  - Control variables Xjit: gender of individual and household head, occupation of household head, dummies for cohabitation with biological/step-parents, household size, urban/rural residency.
- Estimation: linear probability specification; key parameter 훽d_h captures treatment effect on upward and downward IM.
- Parallel trends: tested via leads-and-lags (dynamic effects) and multiple alternative control group specifications (dropping countries/districts without discoveries).

### Baseline results — primary education
- Main effects (Panel A, no controls; Panel B, with controls):
  - Probability of upward primary educational IM for individuals born after a discovery: 0.027*** (2.7 pp.) (column 1).
  - Probability of downward primary educational IM for individuals born after a discovery: -0.012*** (-1.2 pp.) (column 5).
  - Including individuals born 5 years before discovery: upward 0.026*** (2.6 pp.) (column 2); downward -0.005*** (-0.5 pp.) (column 6).
  - Effect of beginning of mining production (Prod-B/A): upward 0.067*** (6.7 pp.) (column 3); Prod-5 upward 0.056*** (5.6 pp.) (column 4). Downward effects remain negative and similar in magnitude (around -0.011***).
- Extrapolated population-level impacts (Africa, period 1950-2000, for individuals born up to 15 years after events):
  - Number of individuals who completed at least primary education while their parents have not increases by 662 thousand for discoveries and 581 thousand for productions.
  - Number of individuals who did not complete at least primary education while their parents have completed it decreases by 371 thousand for discoveries and 124 thousand for productions.
- Robustness to covariates: coefficients remain highly significant at 1 percent and magnitudes broadly unchanged when controls are added.

### Baseline results — secondary and tertiary education
- Table 3 findings:
  - Coefficients associated with mineral discoveries and productions are generally not statistically significant across specifications, with a slight significance in one specification (column 5 at 10 percent).
  - Interpretation: no consistent evidence that discoveries/productions affect upward or downward secondary/tertiary educational IM.

### Dynamic effects of discoveries and productions
- Leads-and-lags specification: compared cohorts born 0–5, 6–10, 11–15 years after events and 5–10, 10–6 years before events, with reference group born 11–15 years before events.
- Primary education dynamic results (selected coefficients):
  - Born 0–5 years after Disc/Prod: upward 0.032*** (3.2 pp.) for discovery and 0.075*** (7.5 pp.) for production; downward -0.014*** and -0.013***, respectively.
  - Born 6–10 years after Disc/Prod: upward 0.042*** and 0.076***; downward -0.015*** and -0.020***.
  - Born 11–15 years after Disc/Prod: upward 0.062*** and 0.153***; downward -0.029*** and -0.028***.
  - Example comparison: probability of upward primary IM is 7.6 pp. higher for individuals born 6–10 years after the beginning of mining production versus only 1.1 pp. higher for those born 10–6 years before production (column 2 example).
- Secondary/tertiary education: coefficients across dynamic windows are not statistically significant.

### Transmission channels
- Two channels explored: income channel (parents working in mining) and returns-to-education channel (wealth index and LIDO score).

A. Income channel — parents working in mining
- Interaction of treatment with dummy “one parent works in mining”:
  - Having one parent working in mining raises the likelihood of upward primary IM by 0.022** (2.2 pp.) following discoveries and by 0.063*** (6.3 pp.) following productions.
  - Having one parent in mining diminishes likelihood of downward primary IM by 0.011 following discovery (column results reported as -0.011** in interaction).
- For secondary/tertiary IM: no significant effects for mining or the interaction term.

B. Returns to education — Wealth index and LIDO score
- Construction:
  - Wealth index: PCA on household variables following DHS conventions.
  - LIDO score: lasso-adjusted industry, demographic, and occupational score mapped from U.S. 1950 labor market to African individuals (controls for demographics).
- Mincer-type GDID results (Table 6):
  - Primary completed → LIDO: 0.021***; → Wealth index: 0.041***.
  - Secondary/tertiary completed → LIDO: 0.048***; → Wealth index: 0.097***.
  - Interaction with mining raises returns: Primary completed X Yes Mining yields positive and significant coefficients for both LIDO and Wealth index (e.g., 0.003*** to 0.045*** across specs).
- Second-step linking returns to IM (Table 7):
  - LIDO score strongly associated with upward primary IM: LIDO 0-1 coefficient 0.592*** (columns 1–2).
  - Wealth index 0-1 coefficient 0.433*** for upward primary IM.
  - Negative associations of LIDO and Wealth with downward primary IM: -0.352*** and -0.252***, respectively.
- Sectoral reallocation (Table 8):
  - Exposure to discoveries/productions decreases likelihood to work in agriculture by 0.011 (−1.1 pp.) for exposed individuals.
  - Increases likelihood to work in manufacturing by around 0.8–1.4 pp. and in services by around 0.8–1.4 pp. (columns 3–6 show increases of 0.008***, 0.012***, 0.003***, 0.014*** depending on spec).

### Robustness checks (selected)
- Alternative control groups (only countries/districts with discoveries): primary IM results remain significant; secondary/tertiary remain insignificant.
- Alternative fixed-effects structures: replacing cohort FE with birth-year FE; adding common time trend; district × cohort FE — main findings unchanged.
- Alternative time windows: 40-year, 50-year, 60-year windows — primary IM effects remain significant; secondary/tertiary remain insignificant.
- Alternative treatment exposure: include individuals born 10 and 15 years before discovery/production (using 60-year window) — baseline findings confirmed.
- Alternative IM definitions:
  - Broaden parental authority to include other immediate older relatives; use minimum and maximum parent education instead of average — findings unchanged: mining increases upward and reduces downward primary IM; secondary/tertiary insignificant.
- Use of all mineral discoveries and productions (not just first events): primary IM effects persist; secondary/tertiary effects remain generally insignificant.
- Adding conflicts as control variable (GED, dummy for district-level conflict >25 deaths):
  - Conflict associated with lower upward primary IM: -0.009***.
  - Inclusion of conflicts does not alter positive (negative) effect of mining on upward (downward) primary IM; secondary/tertiary effects remain not significant.

### Sensitivity analyses
A. By African region (Eastern, Northern, Southern, Western & Central)
- Primary education upward IM: mining positive and strongly significant across regions.
- Primary education downward IM: mining reduces downward IM in Eastern Africa and Northern Africa (negative and significant), not consistently elsewhere.
- Secondary/tertiary IM: heterogeneous:
  - Mining reduces upward secondary/tertiary IM in Eastern and Northern Africa (negative significant).
  - Mining increases upward secondary/tertiary IM in Southern Africa and Western and Central Africa (positive significant).
  - Downward secondary/tertiary IM coefficients generally not significant except some positive associations in Eastern and Northern Africa.

B. By size of mineral discoveries (Minex categories: moderate, major, giant and super-giant merged)
- Primary education:
  - Moderate and major discoveries: positive and significant increases in upward IM and reductions in downward IM (columns 1–4).
  - Giant and super-giant: coefficients for primary downward IM not always significant; moderate/major effects larger in absolute terms than giant/super-giant in some specs.
- Secondary/tertiary education:
  - Giant and super-giant discoveries: statistically significant positive effects on upward secondary/tertiary IM (columns 5–6).
  - Moderate and major less consistently significant at higher education levels.

C. By gender
- Primary education:
  - Mining effects on upward primary IM similar in some specs, but mining production benefits appear larger for males.
  - Example: probability of upward primary IM after production: males ≈ 8.4 percent vs females ≈ 4.9 percent (gender gap in production-related benefits).
  - Downward primary IM: males less likely to experience downward mobility than females in mining areas (coefficients higher in absolute terms for males).
- Secondary/tertiary education: coefficients mostly not statistically significant or inconsistently estimated for both genders.

D. By urban-rural residency
- Primary education:
  - Coefficients associated with mining generally higher in absolute terms in urban areas than in rural areas — larger positive effect on upward IM in urban areas.
- Secondary/tertiary education:
  - Effects significant in urban areas but not in rural areas.
- Implication: preconditions for positive effects of mining on educational IM (jobs, infrastructure, services) are more likely present in urban areas.

### Conclusion and policy implications
- Data and scope: analysis covers more than 14 million individuals across 28 countries and 2,890 districts.
- Key empirical findings:
  - Primary and secondary/tertiary educational IM improved in Africa over time, with a stronger increase in primary IM.
  - Mineral discoveries and productions causally increase upward primary educational IM and decrease downward primary IM for exposed individuals residing in districts with discoveries/productions.
  - No consistent evidence that mineral discoveries/productions affect secondary/tertiary educational IM overall (heterogeneous by region, size, gender, urban/rural).
  - Transmission operates through parents working in mining (income channel) and increased returns to education (wealth index and LIDO score), and through sectoral reallocation away from agriculture into manufacturing and services.
- Policy implications:
  - Implement accommodative policies to support enterprise development and harness job creation from mining (e.g., labor market flexibility, business linkages between mining companies and local SMEs).
  - Targeted policies to reduce inequality of opportunities across gender and urban-rural divides following mineral discoveries.
  - Consider channeling mining revenues into a fund and redistributing among districts to reduce regional disparities and broaden benefits beyond district-level effects.
- Suggested further research:
  - Investigate how mineral discoveries and productions induce local structural transformation by analyzing intergenerational mobility in occupation (possible increase in children's employment outcomes relative to parents).

*Source: Authors’ calculations based on IPUMS dataset and Minex Consulting dataset (2019).*

### References

### wpiea2022160-print-pdf - References

### Major themes and cited literature
- Citations cover the nexus of natural resources, education, and intergenerational mobility, including theoretical and empirical work on:
  - Resource booms, Dutch disease, and macroeconomic effects (Corden and Neary 1982; Sachs and Warner 1995; Van Der Ploeg 2011; Van Der Ploeg and Poelhekke 2017).
  - Natural resources, conflict, and governance (Collier and Hoeffler 2005; Ross 2004; Ross 2006; Keen 2012; Lei and Michaels 2014).
  - Natural resource discoveries and local/macroeconomic outcomes (Arezki, Ramey, and Sheng 2017; Harding, Stefanski, & Toews 2020; Toews and Vezina 2017; Khan et al. 2016).
  - Natural resources and inequality/poverty/household outcomes (Leamer et al. 1999; Loayza, Mier y Teran, and Rigolini 2013; Goderis and Malone 2011; Smith and Wills 2018).
  - Resources and education/human capital (Gylfason 2001; Stijns 2006; Cockx and Francken 2016; Farzanegan and Thum 2020; Maurer 2019; Gradstein & Ishak 2020).
  - Intergenerational mobility and educational inheritance (Becker and Tomes 1979; Corak 2006, 2013; Chetty et al. 2014; Chetty and Hendren 2018; Azam and Bhatt 2015; Alesina et al. 2021; Azomahou and Yitbarek 2020; Neidhöfer, Serrano, and Gasparini 2018).
  - Measurement and data resources (Rutstein and Staveteig 2014 on DHS wealth index; Saavedra and Twinam 2020 on LIDO score; Minex Consulting Datasets (2019); WorldBank (2020) World Development Indicators).
- Empirical studies on specific countries, minerals, and local effects are cited (e.g., Baah and Eshun 2020; Fisher et al. 2009; Hausermann et al. 2018; Zabsonre, Agbo, and Some 2018; von der Goltz and Barnwal 2019).

### Annexes and empirical content summarized in references section
- Annex A: Sample construction from raw IPUMS data — Table 20 shows stepwise sample counts across inclusion criteria for multiple countries and census years.
  - Example aggregated totals:
    - Total All Nall: 130 727 972
    - Total All Ndistrict: 130 466 872
    - Total All Neduc: 104 306 886
    - Total All Nliverelative: 48 454 385
    - Total All Nlivebiop: 45 000 794
    - Total All Nage: 18 160 723
    - Total All Nsex: 18 160 667
    - Total All Nurban: 16 214 531
    - Total All Ncont: 14 459 893
    - Total All NIMiog: 14 263 922
    - Total All NIMbiop: 14 252 719
    - Final tally indicator: 61/82
  - Country-specific row examples preserve census-year level counts (e.g., Egypt 1986: 6 799 093 initial observations; Egypt 2006: 7 282 434; South Africa totals across years: 16 141 863 Nall; Mauritius totals across years: 352 737 Nall).
- Annex B: Summary statistics of Educational IM — Table 21 reports IM summary statistics (panels for biological or step-parents and immediate older generation; primary vs secondary/tertiary; upward and downward mobility). Examples of reported statistics:
  - IM (Mean) entries include 0.508, 0.204, 0.099, 0.438 and standard deviations 0.500, 0.403, 0.299, 0.496 in specified observation samples.
- Annex C: Stylized facts on mineral discoveries — Table 22 (IPUMS countries, 1950–2000) reports:
  - Total # of discoveries: 406
  - Distribution by African regions:
    - Eastern Africa: 48 (11.82)
    - Northern Africa: 23 (5.67)
    - Southern Africa: 196 (48.28)
    - Western and Central Africa: 139 (34.24)
  - Distribution by Size of mineral discoveries:
    - Moderate: 184 (45.32)
    - Major: 121 (29.8)
    - Giant: 88 (21.67)
    - Super-Giant: 13 (3.2)
  - Distribution by Mineral categories:
    - Gold: 141 (34.73)
    - Bulk: 75 (18.47)
    - Precious: 64 (15.76)
    - Base Metal: 61 (15.02)
    - Other: 34 (8.37)
    - Mineral Sands: 17 (4.19)
    - Uranium: 14 (3.45)
  - Table 23: Composition of minerals in each metal category (e.g., Bulk = Bauxite, Coal, Iron ore, Phosphate, Potash; Precious = Diamond, Emerald, PGE, Platinum, Ruby, Rutile, Silver; Base Metal = Copper, Lead, Nickel, Zinc; Other listing of specific minerals; Mineral Sands = Mineral sands, Zircon; Uranium = Uranium).
- Annex D: Country-level educational IM stylized facts — figures (Figure 12–15) present IM by country and gender, residency, and dynamics by districts with and without discoveries across cohorts and gender/residency. Sources noted as IPUMS and Minex Consulting dataset (2019).
- Annex E: District-level educational IM — Tables 24–27 present district-level averages and coefficients of variation by country, education level (primary; secondary and tertiary), and by districts with and without mineral discovery. Examples of reported statistics:
  - Table 24 (District-Level Primary IM by country) total rows:
    - Total districts: 2888
    - Panel (A) Upward mean: 0.562, cv: 0.520
    - Panel (B) Downward mean: 0.259, cv: 0.861
  - Table 25 (District-Level Secondary and tertiary IM) totals:
    - Total districts: 2887
    - Panel (A) Upward mean: 0.342, cv: 0.517
    - Panel (B) Downward mean: 0.457, cv: 0.658
  - Table 26 and Table 27 show comparison by discovery status (yes/no) with totals:
    - Table 26 Panel (A) Upward: Total yes 1233 districts mean 0.600 cv 0.51; Total no 1625 districts mean 0.560 cv 0.52; Total All-2828 districts mean 0.560 cv 0.52.
    - Table 26 Panel (B) Downward: Total yes 73310? (note: table shows aggregated totals as "Total yes73310.310.39192880.510.56"—the table layout is preserved in source).
    - Table 27 Panel (B) totals: Total yes 73310.310.39?; Total no 2125560.350.53?; Total All-2828870.340.52? (original tables present concatenated numeric strings reflecting counts, means, and cv by grouping; refer to table cells for exact layout).
  - Figure 16: Gaps of IM by districts with and without discoveries for each country across four IM dimensions (Upward primary; Downward primary; Upward secondary and tertiary; Downward secondary and tertiary).
- Annex F: Regression baseline results with control variables — Tables 28 and 29 report estimated effects on Upward and Downward mobility for primary and secondary/tertiary education with controls. Selected reported coefficients and model details (standard errors and significance preserved exactly):
  - Table 28 (primary education, columns labeled Disc-B/A, Disc-5, Prod-B/A, Prod-5 for Upward and Downward mobility):
    - Mining coefficients (Upward mobility): 0.028*** (0.002), 0.027*** (0.002), 0.070*** (0.003), 0.059*** (0.003)
    - Mining coefficients (Downward mobility): -0.013*** (0.002), -0.007*** (0.002), -0.012*** (0.002), -0.012*** (0.002)
    - Female effects (Upward): -0.037*** (0.000) across columns; (Downward): -0.003*** (0.000)
    - Urban effects (Upward): 0.152*** (0.001); (Downward): -0.060*** (0.000)
    - Observations reported: 8 306 024; 8 306 024; 8 537 407; 8 537 407; 4 374 423; 4 374 423; 4 478 390; 4 478 390
    - R-squared values: 0.269, 0.269, 0.270, 0.270, 0.134, 0.134, 0.133, 0.133
    - # Treated: 148 633; 192 236; 53 986; 67 663; 98 793; 123 151; 36 768; 49 337
    - District FE, Cohort FE, Census-Year FE: Yes
  - Table 29 (secondary and tertiary education):
    - Mining coefficients (Upward mobility): -0.007 (0.013), -0.000 (0.011), 0.016 (0.021), 0.007 (0.020)
    - Mining coefficients (Downward mobility): 0.037** (0.019), 0.015 (0.018), -0.010 (0.024), 0.007 (0.022)
    - Female effects (Upward): -0.021*** (0.005); (Downward): 0.002 (0.005)
    - Urban effects (Upward): 0.107*** (0.004); (Downward): -0.087*** (0.004)
    - Observations reported: 3 335 415; 3 335 415; 3 461 167; 3 461 167; 323 998; 323 998; 331 618; 331 618
    - R-squared values: 0.217, 0.217, 0.217, 0.217, 0.169, 0.169, 0.169, 0.169
    - # Treated: 67 525; 89 986; 38 002; 43 715; 6 197; 7 491; 2 813; 3 380
    - District FE, Cohort FE, Census-Year FE: Yes
  - Note at bottom: Standard errors in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01.
- Annex G: Validation of LIDO score and Wealth Index — Figures 17–18 show correlations:
  - Figure 17: Correlations between LIDO score and Wealth Index (sources: Demographic and Health Survey wealth index (Rutstein and Staveteig, 2014), LIDO score (Saavedra and Twinam, 2020)).
  - Figure 18: Correlations between LIDO score, Wealth Index, and PPP GDP per capita (with separate panels “With Morocco” and “Without Morocco” and subpanels (A) LIDO score and (B) Wealth index). Sources include IFS datasets.

### Data sources and datasets explicitly cited in references
- Minex Consulting Datasets (2019): FERDI Study Major Discoveries Since 1950.
- IPUMS (raw IPUMS data used to construct sample in Table 20 and elsewhere).
- Demographic and Health Survey wealth index (Rutstein and Staveteig, 2014).
- LIDO score (Saavedra and Twinam, 2020).
- WorldBank (2020): World Development Indicators.
- International Financial Statistics (IFS) datasets (used in validation exercises).

*Is Education Neglected in Natural Resources-Rich Countries? An Intergenerational Approach in Africa — Working Paper No. WP/2022/160*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022160-print-pdf.pdf_