Is Education Neglected in Natural Resources-Rich Countries? An Intergenerational Approach in Africa
  • 1 0000000404811396https://isni.org/isni/0000000404811396International Monetary Fund

The literature on the effects of natural resources on education is mixed and inconclusive. In this paper, we adopt an innovative approach by exploring the effects of mineral discoveries and productions on intergenerational educational mobility (IM), linking parents to the children education levels for more than 14 million individuals across 28 African countries and 2,890 districts. We find that mineral discoveries and productions positively affect educational IM for primary education in Africa for individuals exposed to the mineral sites and living in districts with discoveries. Specifically, the probability of upward primary IM increases by 2.7 percentage points (pp.) following mineral discoveries and 6.7 pp. following mineral productions. Downward primary IM decreases by 1.2 pp. following both mineral discoveries and productions. These positive effects are increasing for individuals born later after discoveries and productions, for males, and individuals living in the urban area. However, no significant effects are found for secondary and tertiary educational IM. Finally, we explore the income and returns to education channels through which mineral discoveries and productions affect educational IM.

Abstract

The literature on the effects of natural resources on education is mixed and inconclusive. In this paper, we adopt an innovative approach by exploring the effects of mineral discoveries and productions on intergenerational educational mobility (IM), linking parents to the children education levels for more than 14 million individuals across 28 African countries and 2,890 districts. We find that mineral discoveries and productions positively affect educational IM for primary education in Africa for individuals exposed to the mineral sites and living in districts with discoveries. Specifically, the probability of upward primary IM increases by 2.7 percentage points (pp.) following mineral discoveries and 6.7 pp. following mineral productions. Downward primary IM decreases by 1.2 pp. following both mineral discoveries and productions. These positive effects are increasing for individuals born later after discoveries and productions, for males, and individuals living in the urban area. However, no significant effects are found for secondary and tertiary educational IM. Finally, we explore the income and returns to education channels through which mineral discoveries and productions affect educational IM.

I. Introduction

Pre-COVID-19, Africa emerged out of decades of stagnant and unstable economic growth since mid-1990s, with significant progress made on education and human capital. According to Young (2012), Sub-Saharan living standards have, for the past two decades, been growing about 3.4 to 3.7 percent per annum, reflecting the African “growth miracle”. This “growth miracle” has been accompanied by significant improvements in education. Indeed, in Sub-Saharan Africa (SSA), gross enrollment in primary education almost doubled from 54 percent in 1970 to 99 percent in recent years. For secondary and tertiary education, it has been more than three and six times higher in recent years compared to 1970, from 13 to 43 percent and 1.4 to 9.4 percent, respectively (World Bank, 2020). These improvements have been observed for both rural and urban areas, as well as females and males. As a result, intergenerational mobility (IM) in education––which measures the levels of children education relative to their parents––has also significantly increased across African countries, above the level in Latin America. But it remains lower in the region compared to Western, Asian, and Eastern Europe countries (Hertz et al., 2008; Azomahou and Yitbarek, 2020; Henn and Robinson, 2021). As shown by Henn and Robinson (2021), actual and perceived social and educational intergenerational mobility constitute one of the three Africa’s latent assets that will drive its economic prosperity and bright future.

Taking stock of this progress in education, we analyze the potential effects of mineral discoveries and productions on educational IM across 28 African countries and 2,890 districts. Few studies have investigated the determinants of educational intergenerational mobility in Africa. They found that individuals and local characteristics (e.g., gender, race, urbanization, etc.), access to markets, quality of education, globalization, and investments in physical and human capital affect educational mobility (Alesina et al., 2021; Azomahou and Yitbarek, 2020; Baah and Eshun, 2020; Nimubona and Vencatachellum, 2007). To the best of our knowledge, only Alesina et al. (2021) discussed the effects of natural resources on educational IM. They show a weak association between oil, gas, and diamond discoveries and educational IM. They explain this weak association by opposing mechanisms as natural resources might be a curse, and they may also represent a wealth spurring human capital accumulation and structural transformation. Still, their analysis of the effects of natural resources on educational IM is very limited and discussed in a short paragraph as this is not their focus.

In this paper, we fill this gap in the existing and inconclusive literature. Mineral discoveries are generally known as being a curse or sources of difficulties and fragility for most African countries as illustrated by the resource curse literature. The growth in the mining sector does not necessary shift an economy towards better industry processing, services, i.e., structural transformation, education, health, job creation, and inclusive growth and development. In contrast, they might have some positive effects, especially on educational IM. Mining activities may create new opportunities that will increase households’ income (Becker and Tomes, 1979 Weber-Fahr et al., 2002; Loayza et al., 2013), enabling them to invest more in children education. They may also favor a structural transformation in the districts with natural resources (Cavalcanti et al. 2019), and therefore an increase of the returns to education, i.e., income or weath induce by education (Torche, 2014; Bütikofer et al., 2018). Finally, they may also support the provision of insfrastructures (in education in particular) financed by the revenues from the resources (Witter and Jakobsen, 2017). This analysis is particularly relevant as the African continent is home to an abundance of mineral resources.1 It 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.2 According to Minex Consulting database (2019),3 969 normal to super-giant mineral discoveries occurred in Africa between 1950 and 2019, and 396 of them (40 percent) since 2000. Moreover, the exploitation of mineral resources makes a significant contribution to the development of African economy. According to the IMF, the mining sector accounted for 8.8 percent of GDP and 51.2 percent of total exports in Sub-Saharan countries over the period 2009–19.

There is a vast and inconclusive literature on the effects of natural resources. At the macro level, most papers found that natural resources have been a curse than a blessing, as well illustrated by the resource curse literature (e.g., Corden and Neary, 1982; Sachs and Warner, 1995, 2001; Kretzmann and Nooruddin, 2005).4 Others studies found positive effects on foreign direct investments in non-resource sectors (Toews and Vezina, 2017), or ambiguous effects on macroeconomic activity and financial conditions (Arezki et al., 2017; Seri, 2021). At the local level, a positive effect of natural resources has emerged in recent analyses focusing on African countries or other developing countries. (e.g., Fisher et al., 2009; Cust and Mensah, 2020; Bhattacharyya and Mamo, 2021). Specifically, the literature of the effects of natural resources on education is also inconclusive. Some papers found that natural resources exert a decrease in education level in developing countries (Leamer et al., 1999; Gylfason, 2001 and Ahlerup et al., 2020), while others revealed that natural resources are positively associated with human capital accumulation, notably through the increase of public spending in education (Kim and Lin, 2017; Pegg, 2010; Stijns, 2006). Other studies highlighted that the effects depend on the quantity or the quality of education, the levels of education or the characteristics of individuals (e.g., Farzanegan and Thum, 2020; Gradstein and Ishak, 2020).

Our paper complements these studies by exploring the effects of mineral discoveries and productions on educational IM. We use a large dataset of more than 14 million individuals from 28 African countries, 2,890 districts. To do so, we rely on two main sources: i) the Integrated Public Use Microdata Series (IPUMS), and ii) the Minex Consulting datasets. We first start by providing a panorama of the trends, dynamics, and disparities of educational IM across countries, regions of the continent, and the characteristics of the individuals, using the conditional absolute measure of IM as in Alesina et al. (2021). Second, we empirically study the effects of mineral discoveries and productions on primary and secondary/tertiary educational IM by employing a generalized difference-in-differences method in a quasi-natural experiment. Our quasi-natural experiment relies on the plausible exogeneity of mineral discoveries that revert specific characteristics, specifically the unpredicted time of discoveries, the unpredicted geographical location, and the lag between the natural resources discoveries and beginning of production (Horn, 2011; Khan et al., 2016; Arezki et al., 2017; Cavalcanti et al., 2017). Third, we explore the channels through which mineral resources discoveries and productions may affect educational IM such as job creation, and the returns to education.5

We find that mineral discoveries and productions positively affect educational IM for primary education in Africa. Indeed, the probability of upward primary educational IM, i.e. the probability for a child born from uneducated parents or parents with less than primary education attainment to achieve at least primary education, increases by 2.7 pp. following mineral discoveries and 6.7 pp. following mineral productions. The probability of downward primary educational IM, i.e. the probability for a child born from parents with at least primary education attainment to be uneducated or have less than primary education attainment, decreases by 1.2 pp. following both mineral discoveries and productions. About the size of the effects, it is relatively small when compared to the increase of educational IM across the different cohorts, signaling that, other factors have also played a significant role in improving IM in Africa.

To put our findings into perspective and extrapolate them to Africa, we show that the number of individuals born up to 15 years after mineral discoveries and who have completed at least primary education while their parents have not, increases by 662 thousand in Africa over the period 1950-2000. This figure stands at 581 thousand for individuals born up to 15 years after mineral productions. Similarly, the number of individuals born up to 15 years after mineral discoveries and who have not completed at least primary education while their parents have completed it, decreases by 371 thousand. This figure stands at 124 thousand for individuals born up to 15 years after mineral productions. These figures would have been even higher to millions of individuals if we would have considered all the individuals born after the discoveries and productions, and not only those born up to 15 years after the event. However, our results show that the effects of mineral resource discoveries and productions on the probability of upward and downward secondary and tertiary educational IM are not statistically significant.

Moreover, we also explore the dynamic effects of mineral discoveries and productions on educational IM, i.e. whether the effects vary with the distance between individuals’ birth years and the year of mineral discoveries and productions, and test the assumption of parallel trend of the GDID model by estimating a leads and lags model following Angrist and Pischke (2009) and Maurer (2019). Our results show that mining activities positively affect primary educational IM for all age groups, but the effects are non-significant for secondary/tertiary educational IM, confirming our baseline findings. Interestingly, we find that the positive effects of mineral activities on educational IM are increasing for individuals born later after the discovery and beginning of mining production, while being non-significant or low for individuals born before the discovery or production, also confirming that the parallel trend assumption is verified. Our baseline results are robust to several robustness checks. We also analyze the sensitivity of our findings across African regions, size of mineral discoveries, gender, and urban-rural living area. Overall, our findings show that the effects of mining activities are different by African regions and size of the discoveries. Also, the positive effects are higher for males than females (only for primary education), and individuals living in urban than rural areas (both for primary and secondary/tertiary education).

Furthermore, we discuss two transmission channels through which the positive effects of mineral discoveries and productions on educational primary IM operate, including the income effect proxied by parents working in the mining sector, and the returns to education. First, our results show that the mining sector creates new job and income opportunities for parents, allowing them to invest more in their child’s education attainment (Becker and Tomes, 1979). Second, we uncover that the economic dynamism and creation of new jobs following the discoveries of mineral resources lead to an increase in the demand for skilled workers and thereby boosting the returns to education (Torche, 2014).

Our paper unveils the potential impact of mineral discoveries and productions on educational IM in Africa. It shows that, on average, mineral discoveries and productions have led to an improvements of primary education and intergenerational mobility at local levels. It adds to the existing knowledge of the effects of natural resources in general, and on education in particular, and identifies some potential channels through which they impact educational IM. Our paper has many policy implications. We discussed them in the concluding section. In short, this paper calls for a better management of the resources by the government and companies. Adequate policies should be put in place to extract the benefits of the resources, including policies aiming at creating jobs and facilitating entreprise development, and ensuring equitable access to the benefits of the resources by all people independently of gender and location.

The rest of the paper is organized as follows. Section II reviews the literature and places this study among the existing papers. Section III discusses the data, explains the construction of the educational IM, and presents some stylized facts on educational IM in Africa by decade, gender, at district and country level. Section IV provides stylized facts on both educational IM and mineral discoveries and productions. Section V describes the methodology. Section VI presents our main findings. Section VII explains the transmission channels. Sections VIII and IX discuss the robustness checks and the sensitivity of our findings, respectively. Section X concludes and discusses potential policy implications.

II. Review of literature

Our paper closely relates to three strands of the literature, notably the general effects of natural resources, the relationship between natural resources and education, and the literature on the determinants of intergenerational mobility.

A. General effects of natural resources

At the macroeconomic level, most papers in this literature show that natural resources have been a curse than a blessing, as well illustrated by the resource curse literature. They found that natural resources are generally associated with the deterioration of economic and institutional conditions, the occurrence of conflicts, an appreciation of real exchange rate, which induces a loss of competitiveness and de-industrialization of the economy, as well as with weak fiscal policy stance and unsustainable debt accumulation (e.g., Corden and Neary, 1982; Sachs and Warner, 1995, 2001; Kretzmann and Nooruddin, 2005; Collier and Hoeffler, 2005; Ross, 2004, 2006; Van Der Ploeg, 2011; Keen, 2012; Van Der Ploeg and Poelhekke, 2017). However, the macroeconomic effects of natural resource discoveries of oil, gas, and minerals on economic activity seem to be mixed. While some papers find negative impacts of giant discoveries on fiscal policy, debt, conflict, poverty, and inequality (Kretzmann and Nooruddin, 2005; Harding et al., 2020; Lei and Michaels, 2014; Tsui, 2011; Smith and Wills, 2018), others show positive effects on foreign direct investments in non-resource sectors (Toews and Vezina, 2017), or ambiguous effects on macroeconomic activity and financial conditions (Arezki et al., 2017; Seri, 2021).

At the local level, a positive effect of natural resources has emerged in more recent analysis focusing on African countries or other developing countries. They show that natural resources are associated with a reduction of inequality, poverty, and an increase of living standards, income, and welfare. In fact, Goderis and Malone (2011) find that resource exploitation booms reduce income inequality in resource-rich countries, while Fisher et al. (2009) show an evidence of the reduction of poverty in the mineworkers’ population in Tanzanian artisanal mines of gold and diamond. Zabsonré et al. (2018) reveal for Burkina Faso that gold exploitation led to better living standards, an increase in per capita household expenditures, and a reduction of poverty in the mining areas. Marlet (2020), using mining exploitation in Ghana, finds that mining activities tend to increase migration flows up to 200 km from the treated district by reducing migration costs through the construction of roads and infrastructures. Moreover, they also induce an increase of income and improvement of welfare by 1.3 percent. In contrast, some papers find that mining activities can create some environmental issues by increasing pollution and metal toxicity (see, e.g., Hausermann et al., 2018; von der Goltz and Barnwal, 2019).

The literature also supports the benefits and positive role of natural resources discoveries on local economic development, governance and conflicts, provisions of public goods and welfare. Cavalcanti et al. (2019) find evidence of a positive impact of oil and gas discoveries on local development and urbanization in Brazil. Cust and Mensah (2020) reveal that oil, gas, and mineral discoveries positively impact the citizen’s expectations, which is materialized by a decrease in outward migration and an increase in fertility in the short term. Bhattacharyya and Mamo (2021) show that oil and mineral discoveries reduce the likelihood of conflict in 48 African countries, which is mainly driven by an improvement of economic development and efficient political distribution patronage in districts with discoveries.

B. Natural resources and education

The literature on the effects of natural resources on education is also inconclusive, but it identifies channels through which natural resources may affect education. While some papers find that natural resources favor a decrease in education levels in developing countries, others rather point out to a positive effect. On the negative effects, Leamer et al. (1999) find that the abundance of natural resources entails a delay of industrialization, and it lowers education levels in Latin American resource-rich countries as workers do not need high skills to work in the natural resources sector. Gylfason (2001), Cokx and Francken (2016), and Ahlerup et al. (2020) show that natural resources crowd out investments in education in resource-rich countries, decrease public education expenditures relative to GDP, and reduce educational attainment, respectively. On the positive effects, some papers find that natural resources abundance is positively associated with human capital accumulation, notably through the increase of public spending in education (see, e.g., Stijns, 2006; Kim and Lin, 2017, and Pegg, 2010). A possible channel is that the mining activity may create new opportunities that will increase households’ income (Becker and Tomes, 1978, Weber-Fahr et al., 2002; Loayza et al., 2013), enabling them to invest more in children education. It may also favor a structural transformation in the district with natural resources (Cavalcanti et al. 2019), and therefore an increase of the returns to education, i.e., income or wealth induce by education (Torche, 2014; Bütikofer et al., 2018). Finally, it may also support the provision of insfrastructures (in education in particular) financed by the revenues from natural resources (Witter and Jakobsen, 2017). We test these channels for the effects of mineral discoveries and productions on educational IM.

Other studies find that the effects of natural resources on education may diverge depending on whether the focus is the quantity or the quality of education, the levels of education, and characteristics of individuals (age, gender). In fact, Farzanegan and Thum (2020) show a positive effect of oil rents on the quantity of education measured by government spending in primary and secondary education, particularly in countries with sound quality of institutions. In contrast, they find a negative effect of oil rents on the quality of education, defined as “an increase in cognitive skills obtained from an additional year of schooling”. This negative effect is driven by the low demand and supply for high-quality education. On the demand side, the phenomenon of resource curse in those countries, by leading to an increase in the size of the non-tradable sector, requires less skilled workers with lower level of human capital. On the supply side, the lower incentive to attract local qualified teachers and the lack of long-term opportunities for foreign or migrant teachers reduce the quality of education. Moreover, Gradstein and Ishak (2020), using IPUMS data on 18 African countries, find that oil price booms occuring in early childhood (ages 0-4) enhance educational attainment and other derived outcomes, but reduce them when occuring in the adolescence (ages 10-14), especially for girls.

C. Intergenerational mobility in education and its determinants

Very few papers exclusively focus on educational IM in Africa. Focusing on South Africa, Nimubona and Vencatachellum (2007) show that educational IM is higher for white than black people. They find that access to credit market and quality of schools are the main determinants of lower educational IM for black people. Baah and Eshun (2020) reveal that economic and educational IM in Ghana is one of the lowest in the world. In addition, they find that globalization enhances IM, thereby recommending policies aiming at expanding globalization. Moreover, they find that FDI and expansionary fiscal policy improve IM while unemployment has an exactly opposite effect on it. Other papers have conducted cross-country analysis based on several African countries. Alesina et al. (2021) employ measures of absolute mobility to estimate educational intergenerational mobility since independence using census data on 27 African countries. After maping IM cross-country and within country variation, they find that colonial investments in the transportation network and missionary activities were associated with higher upward mobility. Intergenerational mobility was also higher in regions close to the coast and national capitals as well as in rugged areas without malaria. Upward mobility is higher and downward mobility is lower in regions that were more developed at independence, with higher urbanization and employment in services and manufacturing. Finally, they also reveal that early exposure of children to regions with higher (lower) upward IM significantly improve (decrease) the likelihood of completing primary schooling. In addition, Azomahou and Yitbarek (2020) analyze the educational IM across 9 Sub-Saharan African countries over 50 years, using two measures of intergenerational educational persistence. They reveal that educational intergenerational persistence has reduced among the birth cohorts in all countries, particularly after the 1960s due to huge investments in human capital following independence and drastic changes in the educational systems. Even in the light of declining educational intergenerational persistence in the region, countries such as Ghana, Guinea, Nigeria, and Uganda experienced higher intergenerational mobility while Comoros and Madagascar had the lowest. Also, intergenerational persistence in education was found to be stronger from mothers to their children, and daughters’ education is more correlated with their parents’ education than that of sons.

More generally, the trends and drivers of intergenerational mobility or persistence in education have been studied in the literature (see, e.g., Corak, 2013; Chetty et al., 2014; Chetty and Hendren, 2018; Howell, 2019; Engzell and Tropf, 2019). Overall, the intergenerational mobility in education has increased over time, but some heterogeneities and disparities across regions remain. Hertz et al. (2008) analyze trends in the intergenerational persistence of education over 50 years in 42 countries, including 19 developing countries and 3 SSA countries.6 They find that the educational IM has improved in almost all regions of the world.7 The western developed countries have higher educational IM than in any region of the world, especially for the Nordic countries. They are followed by the Eastern bloc countries and Asian countries. However, educational IM is lower in Latin American countries and African countries. Interestingly, Henn and Robinson (2021), using the 2015 World Bank Intergenerational Database, find that educational IM is higher in Africa than in South Asia, MENA, and Latin American countries, with countries like Botswana, Kenya, Mauritania, and Cape Verde displaying approximately the same educational IM as high-income countries.8

III. Data sources and IM index

A. Data sources

Our data mainly come from two sources. We use data on education and individual characteristics from Integrated Public Use Microdata Series (IPUMS)9 and data on mineral discoveries and production from Minex Consulting Dataset (2019).

1) IPUMS data

The Integrated Public Use Microdata Series (IPUMS) database covers 82 national censuses surveys from 28 African countries: Benin, Botswana, Burkina Faso, Cameroon, Egypt, Ethiopia, Ghana, Guinea, Kenya, Lesotho, Liberia, Malawi, Mali, Mauritius, Morocco, Mozambique, Nigeria, Rwanda, Senegal, Sierra Leone, South Africa, South Sudan, Sudan, Tanzania, Togo, Uganda, Zambia, and Zimbabwe.10 It contains information on more than 130 million of individuals, including on their demographic characteristics, occupation, household members, the relationship between household members, and place of residence. Regarding education, IPUMS reports data on the total years of schooling and whether the individuals completed primary, secondary, tertiary education levels. For this study, we follow Alesina et al. (2021) and use the educational attainment for both parents and childrens instead of the years of schooling given their higher coverage. Our sample is based on the availability of information on district level and residency, education and individuals characteristics (gender, age) as well as whether individuals co-reside with their biological/step parents or immediate older generation. We have harmonized the boundaries of districts following Alesina et al. (2021) to deal with admininistrative boundaries changes.11 We focus on the individuals aged between 16 and 50, and born between 1950 and 2000.12 The final sample covers more than 14 million individuals across 2,890 districts. Table 20 in Appendix A describes the data for each census and country.

2) Mineral discoveries data

Our data on mineral discoveries and productions in Africa come from Minex Consulting Dataset (2019). This dataset provides geolocalized information on discoveries, their size (moderate, major, giant, super-giant), the status of the mine (closed, feasibility study, operating, underdeveloped), and the type of minerals.13 After merging this dataset with the IPUMS data, we identify 331 districts out of the total of 2890 in which mineral sites were discovered or entered in production. Figure 11 in Appendix C displays the evolution of the number of discoveries over time in all African countries. 969 mineral discoveries occurred in Africa between 1950 and 2019, of which 573 (60 percent) between 1950 and 2000. This study covers 406 mineral discoveries in 28 African countries over the period 1950–2000 (i.e., 71 percent of the 573 mineral discoveries). We therefore cover a large share of mineral discoveries in Africa.

Figure 1 maps the location of these discoveries across Africa by the size of mineral discoveries. We observe that mineral discoveries have been concentrated in Southern Africa (48.3 percent) and Western and Central Africa (34.2 percent). A relatively few discoveries occurred in Eastern Africa (11.9 percent) and Northern Africa (5.7 percent). Looking at the sizes of mineral discoveries, they have been mostly moderate (45.3 percent), followed by major (29.8 percent), giant (21.7 percent) and super-giant giant (3.2 percent). Moderate and major discoveries were mainly found in Western and Central Africa while giant and super-giant were located in Southern Africa. The minerals discovered were mostly gold (34 percent), bulk metals (18.4 percent), precious minerals (15.8 percent), and base metals (15 percent). Table 22 and Table 23 in Appendix C present some statistics on the mineral discoveries by regions, size, and types.

Figure 1:
Figure 1:

Location of mineral discoveries, by size, for all African countries, 1950–2019

Citation: IMF Working Papers 2022, 160; 10.5089/9798400218071.001.A001

Source: Authors’ calculations based on Minex Consulting datasets (2019)

B. Construction of educational IM

In this paper, we use absolute educational IM measures as in Alesina et al. (2021).14 We define both an upward and downward educational IM for the primary and secondary/tertiary education levels. First, upward primary educational IM is defined as the probability for a child born from uneducated parents or parents with less than primary education attainment to achieve at least primary education. Downward primary educational IM is defined as the probability for a child born from parents with at least primary education attainment to be uneducated or have less than primary education attainment. Second, upward secondary/tertiary educational IM is defined as the probability for a child born from parents with at most primary education background to achieve at least secondary education. Downward secondary/tertiary educational IM is defined as the probability for a child born from parents with at least secondary education background to achieve primary education or be uneducated.15 To identify the old generation benchmark for each child, we use the average of education attainment for their biological/step parents, rounded to the nearest integer. In the robustness section, we use the minimum or maximum of the education levels of the biological/step parents as benchmark. We also consider the immediate older generation and broaden the definition of parental authority to include uncles/aunts (in law), parents-in-law, grand-parents, and grand-uncles/aunts in the reference group, to take into account fostered, abandoned, or orphan children.

Practically, first, for each individual (parents and children), we compute two educational attainment variables Pjith and STjith measuring the primary and secondary/tertiary educational attainment, respectively. Pjith takes that value of one if the individual j born in district i and year h, and surveyed in year t has completed at least the primary education, and zero otherwise. Similarly, STjith takes that value of one if the individual j born in district i and year h, and surveyed in year t has completed at least the secondary education. Second, for each child j, we compute two averaged measures of parents’ educational attainment, PPjith and PSTjith, as the average of Pjith and STjith rounded to the nearest integer, respectively, for the biological/step parents if both cohabit with the child, or for only the father/step-father or mother/step-mother if the child lives with only one of its parents. Third, we compare the educational attainment for each child j cohabiting with at least one parent to the average educational attainment of its parents and obtain our absolute measures of educational as follows:16

  • i. Upward primary IM: IMUPjith={1ifPjith=1andPPjith=00ifPjith=0andPPjith=0(1)

  • ii. Downward primary IM: IMDPjith={1ifPjith=0andPPjith=10ifPjith=1andPPjith=1(2)

  • iii. Upward secondary/tertiary IM: IMUSTjith={1ifSTjith=1andPSTjith=00ifSTjith=0andPSTjith=0(3)

  • iv. Downward secondary/tertiary IM: IMDSTjith={1ifSTjith=0andPSTjith=10ifSTjith=1andPSTjith=1(4)

C. Cohabitation-selection issues

Our analysis might be subject to cohabitation-selection issues, as our sample includes both youth children and adults, and individuals’ educational attainment is also determined by their cohabitation with their parents or not.17 For adults, the cohabitation-selection issues are more severe as the intensity of self-selection is increasing with age. Moreover, it is more accentuated for adults’ women, especially for those who got married at younger age, as some African countries are patrilocal (Heckert et al., 2021). Since our focus is also on secondary education, our sample must include adults at age where secondary education is mostly completed despite the likelihood of co-residing with parents at this age being low (see Figure 2).

Figure 2:
Figure 2:

Cohabitations rates

Citation: IMF Working Papers 2022, 160; 10.5089/9798400218071.001.A001

Source: Authors’ calculations based on IPUMS dataset

Interestingly, we show in Figure 2 that the cohabitation selection is independent from the discoveries of natural resources. Both districts with and without discoveries exhibit the same patterns of cohabitation by age and gender, therefore the cohabitation selection issues might have limited impact on our analysis.18 If the cohabitation rate was higher in districts with discoveries than districts without discoveries, it would have caused our estimates of the effects of discoveries on upward (downward) IM to be upward (downward) bias. Indeed, as we expect discoveries to have a positive effect on educational IM and given that individuals living with their parents are likely to have higher educational attainments, then, higher cohabitation rates in districts with discoveries than without discoveries would have resulted in higher positive estimates of the effects of discoveries on educational IM.

We further investigate the potential bias of the cohabitation selection. We do so in Table 1 by computing the differences in educational attainment for both individuals co-residing with biological/step parents (selected individuals) and those that do not, both in districts with and without discoveries, separately. We test the significance of the differences through a Khi-2 test. We show that the unconditional likelihood of not completing primary education is higher for individuals not co-residing with their biological/step parents than those who do.

Table 1:

Differences in educational attainment between individuals living with biological/step parents or not.

article image

This difference is more accentuated in districts without discoveries than districts with discoveries. Similarly, the likelihood of completing primary and secondary education is higher for those living with their biological/step parents than those who do not, also more pronounced in districts without discoveries. These differences are significant as indicated by the Khi-2 tests. As a result, if we were to consider individuals not co-residing with parents in our study, the level of educational attainment would have been on average lower in districts without discoveries than districts with discoveries. Thus, if the cohabitation selection creates any biases, our estimates of the effects of discoveries on upward (downward) IM would be downward (upward) bias. Therefore, as we expect discoveries to have a positive effect on IM (i.e., increase upward IM and decrease downward IM), these effects should be considered as a lower bound.

D. Stylized facts on educational IM in Africa

In this section, we briefly describe the trends of educational IM in Africa as well as their disparities across countries, gender, and residency. To do so, we calculate conditional educational IM, netting country/districts, cohort and census effects. Specifically, we regress educational IM indices on country or district fixed effects αi, cohort fixed effects γt, and census-year fixed effects λt. The model is as follows:

IMjit = αi + γt + δt + εjit      (5)

Country or district fixed effects αi reflect the conditional likelihood of each type of educational IM at the country or district levels, netting the cohort and census effects. Cohort fixed effects γt reflect the conditional likelihood of each type of educational IM, netting the country/district and census effects. We do so to better compare the educational IM across individuals, cohorts, districts, and over time, especially by purging the differences between countries/districts, cohorts, and census-year specific effects. In addition, we estimate conditional educational IM by country and gender, country and individuals’ residency (urban and rural), by cohort and gender, cohort and residency, and cohort and discovery dummy by introducing country and gender fixed effects, country and residency fixed effects, cohort and gender fixed effects, cohort and residency fixed effects, and cohort and discovery fixed effects, respectively.19

1) Trends of IM by decade

Overall, we observe that primary and secondary/tertiary educational IM have significantly improved in Africa over time, independently of gender and residency (urban/rural areas). The average trends of primary and secondary/tertiary educational IM are displayed in Figure 3 for both upward and downward mobility, respectively, and for five cohorts between 1950 and 2000. We show that upward primary IM has steadily increased across cohorts, from 35.1 percent for the 1950s cohort to 57.7 percent for the 1990s cohort. Similarly, downward primary IM has steadily decreased, but at a slower pace, from 29.8 to 23 percent between the 1950s and 1990s cohorts, respectively. Moreover, secondary and tertiary educational IM have experienced similar trends. Upward secondary and tertiary educational IM has steadily increased from 10.8 to 32.9 percent, while downward secondary and tertiary educational IM has steadily decreased from 45.1 to 35.8 percent between the 1950s and 1990s cohorts, respectively. In contrast to primary education level, downward IM has always been elevated than upward IM for secondary and tertiary levels, but the gap has closed over time. Finally, upward (downward) educational IM has been higher (lower) at primary level than secondary and tertiary level.

Figure 3:
Figure 3:

Educational IM by birth cohorts

Citation: IMF Working Papers 2022, 160; 10.5089/9798400218071.001.A001

Source: Authors’ calculations based on IPUMS dataset

We next look at these trends by gender in Figure 4. In general, both males and females have had an increase in the probability of upward educational IM and a decline of the probability of downward educational IM for both primary and secondary/tertiary levels. We also observe that the gender gap in favor of males has narrowed over time, for instance, with the probability of upward (downward) secondary/tertiary educational IM for females being higher (lower) than that of males in the last decades.

Figure 4:
Figure 4:

Educational IM by birth cohorts and gender

Citation: IMF Working Papers 2022, 160; 10.5089/9798400218071.001.A001

Source: Authors’ calculations based on IPUMS dataset

Finally, we explore the trends by residency as presented in Figure 5. We find that the general trends are also confirmed for both individuals living in urban and rural areas. More specifically, we show that educational IM has always been higher in urban areas than in rural areas. Although the residency gap has diminished over time, it has remained significant across cohorts.

Figure 5:
Figure 5:

Educational IM by birth cohorts and residency

Citation: IMF Working Papers 2022, 160; 10.5089/9798400218071.001.A001

Source: Authors’ calculations based on IPUMS dataset

2) Educational IM at country and district levels

a. Country-level educational IM

We display in Figure 6, the country-level educational IM. It shows that educational IM is uneven across countries, and that upward and downward IM are negatively correlated, i.e., countries with the highest upward IM tend to have the lowest downward IM, and inversely. These findings hold for both primary and secondary/tertiary education levels. Upward primary IM ranges between 13 percent in South Sudan and 98 percent in Mauritius, and downward primary IM between close to zero for Egypt, Mauritius and Botswana and 58 percent in South Sudan. Upward secondary/tertiary IM ranges about 14 percent in Sierra Leone and Sudan and more than 70 percent in Egypt and Nigeria, while downward secondary/tertiary IM is between 1 percent in Egypt and 71 percent in Togo.

Figure 6:
Figure 6:

Ranking: Country-level educational IM

Citation: IMF Working Papers 2022, 160; 10.5089/9798400218071.001.A001

Source: Authors’ calculations based on IPUMS dataset
b. Country-level educational IM by gender

We also present the country-level gender gap in educational IM in Figure 12 in Appendix D. First, we show that gender gap, in favor of males for the primary level, is more pronounced for individuals living in countries with lowest values of upward IM, and in countries with highest values of Downward IM. More specifically, upward primary IM is higher for males than females in Togo (14.9 percent), Liberia (11.6 percent), Sierra Leone, Zambia, Uganda (around 8–9 percent).20 It is rather higher for females than males in Lesotho (25.9 percent), Botswana (15.5 percent), Nigeria (5 percent), and South Africa (4.2 percent). Similarly, downward primary IM is higher for females than males in Togo (9.8 percent), Liberia (7 percent), Guinea (6.2 percent), and Sierra Leone, Benin, and South Sudan (4–6 percent). It is rather higher for males than females in Lesotho (16.1 percent), and Botswana, Burkina Faso (5.4 percent). Second, in contrast, gender gap in favor of males, for the secondary/tertiary level, is less related to whether individuals are living in countries with highest values of upward or downward educational IM. For instance, upward secondary/tertiary IM is higher for males than females in Rwanda, Egypt, Liberia (8–9 percent), Uganda, Malawi and Zambia (6–6.5 percent). It is rather higher for females and males in South Africa, Morocco (4.5–5 percent), Sudan, Lesotho (3.9 percent), Mauritius (3 percent), Nigeria (2.5 percent), and Burkina Faso (2 percent). Moreover, downward secondary/tertiary IM is higher for females than males in Botswana (14.2 percent), Malawi (11.3 percent), Togo, Sierra Leone (around 9 percent), Ghana, and Benin (around 8 percent). It is rather higher for males than females in Morocco (7.2 percent), Mauritius (6.8 percent), South Africa (5.8 percent), Zimbabwe, Burkina Faso (around 4.5 percent), Sudan, Lesotho and Mozambique (3 percent).

c. Country-level educational IM by residency

We also report the country level residency gap in educational IM in Figure 13 in Appendix D. First, we show that upward (downward) IM are higher (lower) for individuals living in urban than rural areas, for both primary and secondary/tertiary levels, and for all countries. Second, individuals living in urban areas tend to do far better than those in rural areas in countries with the lowest values of upward IM and the highest values of downward IM. Indeed, the countries with the highest values of upward primary IM for individuals living in urban than rural areas are Ethiopia (57.6 percent), Sudan (44.8 percent), Burkina Faso (42.3 percent), and Guinea (40.2 percent). The countries with the lowest residency gap for upward primary IM (always in favor of individuals living in urban areas) are Mauritius (5 percent), South Africa (7.6 percent), Nigeria (10.5 percent), and South Sudan (12.1 percent). Similarly, the countries with the highest values of downward primary IM for individuals living in urban than rural areas are Ethiopia (50.8 percent), Burkina Faso (43.3 percent), and Sierra Leone (34.4 percent). The countries with the lowest residency gap for downward primary IM are Mauritius (0.9 percent), Nigeria, South Africa (close to 3 percent), and Egypt (4.7 percent). Third, we find that the residency gap for upward secondary/tertiary IM is higher in Ethiopia (32.9 percent), Malawi (21.8 percent), Morocco (21.6 percent), and Guinea (21.2 percent), while it is lower in South Sudan, Sierra Leone, and Botswana (5–6 percent). We also show that the residency gap for downward secondary/tertiary IM is higher in Burkina Faso (51.9 percent), Ethiopia (40.9 percent), and Morocco (34.9 percent), while it is lower in South Sudan (-2.8 percent), Nigeria (7.7 percent), and Egypt (9.1 percent).

d. Mapping of district-level educational IM across Africa

Finally, we map educational IM across 2,890 districts in Africa in Figure 7.21 Table 24 and Table 25 in Appendix E also report the summary statistics of district-level educational IM by country for primary and secondary/tertiary levels, respectively. They show both large within country and cross-districts variations. Overall, we find that within country disparities are larger in countries with lower educational mobility, with some exceptions. First, we show that upward primary IM is more unequal in South Sudan, Sudan, Ethiopia, and Burkina Faso (countries with lower average upward primary IM) and less unequal in Mauritius, South Africa, and Zimbabwe (most countries with higher average upward primary IM).22 Regarding the probability of downward primary IM, it is more unequal in Botswana, Mauritius, and Egypt, and less unequal in South Sudan, Rwanda, Liberia, and Lesotho. Moreover, upward primary IM varies less across regions than downward primary IM (coefficient of variation is 1.6 times higher for the latter than the former). Second, we find that upward secondary/tertiary IM varies more across districts in Ethiopia, Sudan, Malawi, and Cameroon (many countries with lower average upward secondary/tertiary IM). It varies less across districts in Botswana, Mali, Lesotho, and Senegal (most countries with higher or milder average upward secondary/tertiary IM). Moreover, downward secondary/tertiary IM varies more across districts in Botswana, Egypt, Ethiopia, and Malawi (paradoxically, most countries with lower average downward secondary/tertiary IM). It varies less across districts in Senegal, Ghana, Morocco, and Tanzania (paradoxically, most countries with higher or milder average downward secondary/tertiary IM).

Figure 7:
Figure 7:

District-level educational IM

Citation: IMF Working Papers 2022, 160; 10.5089/9798400218071.001.A001

Source: Authors’ calculations based on IPUMS dataset

IV. Stylized facts on educational IM and mineral discoveries and production

As a forestate of the effects of mineral resource discoveries and productions on intergenerational educational IM, we present in this section some stylized facts on both conditional IM in education and mineral discoveries and productions. South Africa, the country with the highest number of discoveries in our sample (108 total discoveries), illustrates well the case of countries where educational IM is higher in districts with discoveries/productions (see Figure 8). In the 60 districts where mineral sites were found, upward IM was 92 and 46 percent for primary and secondary/tertiary levels, respectively, higher by 2–3 pp. than in the 156 districts without any discoveries. Similarly, downward IM was 2 and 25 percent in districts with discoveries for primary and secondary/tertiary levels, respectively, lower by 1–2 pp. than in districts without discoveries.

Figure 8:
Figure 8:

District-level educational IM in South Africa

Citation: IMF Working Papers 2022, 160; 10.5089/9798400218071.001.A001

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

We also show the mean differences of district-level conditional IM between districts with and without discoveries over the period of study (see Table 26 and Table 27).23 We find that upward (downward) IM for primary education is, on average, higher (lower) in districts with discoveries than in districts without discoveries by around 4 pp. The opposite result holds for secondary education: upward (downward) IM is on average lower (higher) in districts with discoveries than districts without discoveries by around 4–6 pp. Second, we also present the summary statistics of IM by districts with and without discoveries for each country in Table 26 and Table 27. Upward IM is on average higher in districts with discoveries than districts without discoveries in 12 and 7 countries out of 26 countries (with both districts with and without discoveries) for primary and secondary levels, respectively. Downward IM is, on average, lower in districts with discoveries than districts without discoveries in 11 and 9 countries out of 26 for primary and secondary levels, respectively (see also Figure 16).

However, these stylized facts could hide differences in the dynamics of IM in districts with and without discoveries, and particularly progress that has happened in districts with discoveries. Figure 9 shows that while upward IM for primary and secondary/tertiary education was lower in districts with discoveries among the old cohorts (1950s and 1960s), it has significantly increased and closed the gap in these districts for more recent cohorts (1980s and 1990s) to stand above the one in districts without discoveries. Similarly, downward IM for primary and secondary/tertiary education was higher in districts with discoveries than without discoveries for old cohorts, and the gap has narrowed over time and for more recent cohorts. Therefore, IM has been significantly more dynamic in districts with discoveries. As a result, mineral discoveries and productions seem to have contributed to change the geography of the land of opportunities across African regions.

Figure 9:
Figure 9:

Dynamics of country-level IM by district with and without discoveries and cohort

Citation: IMF Working Papers 2022, 160; 10.5089/9798400218071.001.A001

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

In addition, we find that improvements in IM have occurred for both females and males, with females doing better than males in districts with discoveries (see

Figure 14 in Appendix D). The gender gap of upward primary IM in favor of males has closed early in districts with discoveries (for 1970s cohort) than districts without discoveries (for 1990s cohort). The gender gap of secondary and tertiary IM in favor of males quickly turns to be in favor of females in districts with discoveries (for cohorts 1960s–1990s), which happened 20 years later in districts without discoveries (for cohorts 1980s– 1990s). Likewise, downward IM for primary and secondary/tertiary was higher for females than males in districts without discoveries contrary to districts with discoveries. Still, the gap has closed or widened for recent cohorts in districts without and with discoveries, respectively, with females performing better than males. We also find that IM improvements have occurred both in urban and rural areas in all districts (see Figure 15 in Appendix D). However, the gap between urban and rural areas has remained significant despite greater improvements in rural areas. There are, however, no significant differences in the dynamics of IM for rural and urban districts with and without discoveries.

Finally, we look at the dynamics of IM for individuals born around the first discovery and beginning of production. They are presented in Figure 10. These dynamics show that the timing of mineral discoveries and productions may constitute structural breaks in IM dynamics for individuals born before and after the discovery or production. Indeed, we show that the likelihood of upward IM for primary and secondary education has significantly increased after mineral discoveries and productions for individuals born after the discovery or production. Similarly, the likelihood of downward IM for primary and secondary education has decreased after the mineral discoveries and productions for individuals born after the discovery or production, while it has sometimes increased for individuals born in years running up to discovery or production. While we cannot plot the dynamics of the counterfactuals, we show that IM has on average significantly accelerated after mineral

Figure 10:
Figure 10:

Dynamics of Unconditional IM around the first discovery and start of production

Citation: IMF Working Papers 2022, 160; 10.5089/9798400218071.001.A001

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

V. Empirical methodology

To estimate the potential effects of mineral discoveries and productions on educational IM, we adopt an experimental approach and exploit the exogeneity of natural resource discoveries. First, it is plausible that the timing of mineral resource discoveries is exogenous due to the uncertainty related to the timing of the discovery and exploration success. While the technology used for exploration has improved over time, it is still highly improbable to predict the timing and success likelihood of finding a mineral field in a particular region (Khan et al., 2016; Arezki et al., 2017; Cavalcanti and others, 2019; Seri, 2021). Moreover, the exact location of mineral resources discoveries is purely exogenous as it depends on random geographical factors of the area. Therefore, while some regions may be endowed with mineral resources, it is improbable to find any resources in others. Second, mineral discoveries provide a significant source of revenues and represent a major economic shock that can affect the trajectory of the development in countries and districts they are found. They can also change the habit of individuals and their expectations about their own and children's future. Third, as shown by Horn (2011) and Arezki et al. (2017), there is a significant lag between the discoveries of natural resources and the beginning of their production, around five to six years. This allows us to study the effects of both mineral discoveries and productions on educational IM, separately. These features stand at the heart of our identification strategy and allow capturing a causal effect of mineral discoveries and productions on educational IM in African countries. Throughout the paper, we analyze both the effects of the first discoveries and productions on upward and downward mobility, and later in the robustness section, of multiple and successive discoveries and productions.

We employ a generalized difference-in-differences (GDID) strategy in a quasi-natural experiment and estimate treatment effects by comparing the changes in educational IM between a treatment group (people with exposure to the mineral discoveries and productions) and a control group, across pre-discovery/production and post-discovery/production. By doing so, our goal is to identify how educational IM has evolved following discovery/production for a group of people with exposition compared to a group of people born in the same district and around the discovery or production but not exposed to it, while controlling for the dynamics of educational IM in other districts without any discovery/production.

To capture the effects of the discovery/production, first, we focus in our baseline on a period spanning 30 years around it, i.e., we consider in the regressions, individuals born 15 years before and after the discovery or production. In the robustness, we expand this window and consider larger window periods of 40, 50, 60 years around the discovery or production. We define different expositions to the discovery or production as well as various control groups. First, focusing on a window of 30 years, we consider all the individuals born the year of the discovery or production to up to 15 years after it to be in the treatment group. In an alternative specification, we assume that individuals born 5 years before the discovery or production will still be exposed to it as they start their education around the date of the discovery or production; therefore, the treatment group comprises individuals born five years before the discovery or production to up to 15 years after. Second, in the first specification, we consider as control groups (i) individuals born in districts with a discovery or production and between 15 years before to one year before the discovery or production, and (ii) individuals born in districts without any discovery or production over the period of study. Similarly, the second specification includes in the control groups, (i) individuals born in districts with a discovery or production and born between 15 years to 5 years before the discovery or production, and (ii) individuals born in districts without any discovery or production.

Given the nature of our data, multiple discoveries or productions and multiple treatments and control groups within and across districts, our GDID model allows for (i) specific IM across districts by introducing district fixed-effects αi, (ii) the common change in IM to vary across cohorts (decade in baseline, and each year of birth in the robustness) and years of census/survey by introducing cohort fixed-effects or year-birth fixed effects γt and census-year fixed effects δt, respectively, and (iii) different timing of the discovery or production for different treated groups. This allows to filter out all rigid characteristics specific to districts, cohorts or years of birth, and census year found to be critical in explaining education IM by Alesina et al. (2021). The model is estimated using a linear probability specification and obtained as follows:

IMjith = αi + γt + δt + βdh + Xjitθ + εjit      (6)

Where IMjith is our measure of upward (and downward) mobility for primary or secondary/tertiary levels that takes the value one if the child j in district i has on average a higher (lower) education than its biological/step-parents knowing that its biological/step-parents have not completed (have completed) primary or secondary/tertiary education, respectively. dh is a dummy that takes the value of one if the individual is in the treated group (e.g., born between the year of the discovery or production to up to 15 years later) and zero if the individual is in the control group (e.g., either born between 15 years to one year before the discovery or production, or born in districts without any discovery or production). β is the coefficient of interest. It captures the treatment effect of mineral discovery/production on upward and downward educational IM by comparing educational IM in the treated and control groups. Xit is a set of control variables including the gender of individuals and of their household head, the occupation of household head, the dummies of cohabitation with biological/step-parents (i.e, with only biological/step-father, only biological/step-mother, or both biological/step-father and mother), the size of the household, and urban/rural residency. εjit is the idiosyncratic term.

Our model requires the parallel trends assumption to hold, i.e., in the absence of the discovery or production, the change of educational IM would have been the same in both the treated and control groups. This assumption is violated when there are unobserved factors that are correlated with both the exposition to the discovery or production and the timing of the discovery or production. As discussed above, we have good reasons to believe that the timing of mineral discoveries is exogenous. Regarding the exposition to the discovery or production, since we focus on a relatively short period around the discovery or production in our baseline and include either cohort fixed effects or year-birth fixed effects, we limit the risks that other shocks or interventions polluted our findings. However, since we cannot test for the parallel trends’ assumption in our GDID, we apply the following strategy to test the robustness of our findings and implicitly verified whether this assumption holds. First, we analyze the dynamic effects and conduct a standard leads-and-lags test following the literature (see, e.g. Angrist & Pischke, 2009, Maurer, 2019). This allows testing whether the effects of discoveries occurred after the discovery or production and tend to intensify thereafter. Second, we cross validate our findings by using different control groups while dropping from the control group all individuals in (i) countries without mineral discoveries (Mauritius and South Sudan) and (ii) in districts without mineral discoveries. This reduces the heterogeneity and differences in characteristics between our treated and control groups.

VI. Results

In this section, we present our main findings, starting with the baseline results for the primary and secondary/tertiary levels of education, before discussing the dynamic effects of mineral discoveries and productions on educational IM.

A. Baseline results

1) Educational primary IM

Table 2 reports the baseline results for the various sample compositions and definitions of the variables. The dependent variable is the occurrence of upward (columns 1–4) or downward (columns 5–8) educational IM. As explained in the previous section, we consider a time window of 15 years before and after the mineral discoveries or productions. We start with the results in Panel (A) where control variables are not included in the estimates.

Table 2:

Baseline results, primary education

article image
Note: This table presents the baseline results for the probability of upward (columns 1–4) and downward (columns 5–8) educational IM as a function of mining discoveries and production without (Panel A) and with (Panel B) a number of other control variables. Standard errors are in parentheses. Disc-B/A means that the treatment group is after the discovery and the control group is before the discovery; Disc-5 is when individuals born 5 years before the discovery are included in the treatment group; Prod-B/A means that the treatment group is after the beginning of production and the control group is before production; Prod-5 is when individuals born 5 years before mining production are included in the treatment group. *Indicates significance at 10% level, **significance at 5% level, and ***significance at 1% level.

In columns (1) and (5), the treatment group includes individuals born after the mineral discoveries or productions. The estimates then provide the differences in the likelihood of upward or downward primary educational IM for individuals in districts with mineral discoveries or productions born after (treatment group) and before (control group) the discovery within the exposure window period of 15 years. We do so by controlling for the likelihood of educational IM in districts without mineral discoveries/productions. The results show that the probability of experiencing an upward educational IM is 2.7 pp. higher for an individual born after a discovery of a mining site compared to an individual born before the discovery (column 1). In other words, individuals who are born from parents who are uneducated have better chances of achieving at least primary education if exposed and living to a district with a mining discovery. Inversely, the likelihood of experiencing a downward educational IM for individuals born after the discovery of a mining site is 1.2 pp. lower than those before the mining discovery (column 5). That said, individuals who are born from educated parents are less likely to do less than their parents if exposed and living to a district with a mining discovery. About the size of the effects, it is relatively small when compared to the increase of IM across the different cohorts, signaling that, other factors have also played a significant role in improving IM in Africa.

To put our findings into perspective and extrapolate them to Africa, we show that the number of individuals born up to 15 years after mineral discoveries and who have completed at least primary education while their parents have not, increases by 662 thousand in Africa over the period 1950-2000. This figure stands at 581 thousand for individuals born up to 15 years after mineral productions. Similarly, the number of individuals born up to 15 years after mineral discoveries and who have not completed at least primary education while their parents have completed it, decreases by 371 thousand. This figure stands at 124 thousand for individuals born up to 15 years after mineral productions. These figures would have been even higher to millions of individuals if we would have considered all the individuals born after the discoveries and productions, and not only those born up to 15 years after the event.

In columns (2) and (6), we expand the treatment group to include individuals who are born 5 years before the discovery. In fact, these individuals may have not started their education at the time of the discovery as they do not meet the minimum years of schooling in most African countries. The coefficient of interest in column (2) remains broadly the same as in column (1), suggesting that the change in the time exposition to the mining discovery does not after the likelihood of upward primary educational IM. However, the coefficient is significantly lower in column (6) than in column (5), implying that individuals born after the discovery of the mining sites would have a lower likelihood of downward mobility than those born just before (five years before) the discovery.

In columns (3) and (7), we use an alternative definition of mining activity where the binary variable is now equal to one for districts where mining production has started and zero otherwise. The estimates compare the likelihood of upward or downward educational IM between individuals born after and before the beginning of the mining production. We find that the probability of experiencing an upward educational IM is 6.7 pp. higher for an individual born after the mining production compared to those born before the beginning of the mining production. The coefficient associated with mining is higher in column (3) than in column (1), implying that mining productions tend to have, on average, higher positive effects on the likelihood of upward primary educational IM than the discoveries. These higher effects could be explained by the increase in investments required to start production, which may create more jobs and income opportunities for district residents than the proceeds associated with the exploration and discovery of mining sites. Indeed, the start of the production reveals a new information that would affect the parents own and child’s life expectation. We also observe that our coefficient of interest remains broadly unchanged in column (7) compared to column (5), suggesting that individuals born after the discovery and the beginning of the mining production have the same likelihood of experiencing downward primary educational IM. We also include the individuals born 5 years before the beginning of production in the treatment group in columns (4) and (8). We find that the coefficient associated with mining production is lower in column (4) than in column (3), while remaining the same in columns (7) and (8).

In Panel (B), we control for several covariates that can affect the likelihood of upward and downward educational IM as presented in the previous section (see Table 28 in Appendix F for the coefficients associated with each control variable). We find that the coefficients associated with mining are highly significant at 1 percent level in all columns, and their magnitude are broadly equal to those found in Panel (A). Thus, our findings remain unchanged when we control for individual characteristics.

2) Educational secondary and tertiary IM

We now turn to secondary/tertiary education level. The results are reported in Table 3. We find that the coefficient associated with mineral discoveries and productions is not statistically significant in all columns, except the slightly significance in column (5) at 10 percent level.

Table 3:

Baseline results, secondary and tertiary education

article image
Note: This table presents the baseline results for the probability of upward (columns 1–4) and downward (columns 5–8) educational IM as a function of mining discoveries and production without (Panel A) and with (Panel B) a number of other control variables. Standard errors are in parentheses. Disc-B/A means that the treatment group is after the discovery and the control group is before the discovery; Disc-5 is when individuals born 5 years before the discovery are included in the treatment group; Prod-B/A means that the treatment group is after the beginning of production and the control group is before production; Prod-5 is when individuals born 5 years before mining production are included in the treatment group. *Indicates significance at 10% level, **significance at 5% level, and ***significance at 1% level.

The result suggests that the likelihood for individuals born after the mineral discoveries and productions to experience an upward or downward secondary/tertiary educational IM is not statistically different from that of individuals born before the start of mining activities discovery or production. The insignificance of the effects of mining activities on secondary and tertiary educational IM could be due to the presence of mixed effects, both positive and negative, that are offsetting each other. For instance, as illustrated by Gradstein and Ishak (2020), a positive oil price shock is found to have a positive effect on educational attainment for childs (ages 0–4), and a negative effect for adolescents (ages 10–14). Indeed, when youth children or adults reach secondary and tertiary education levels, they face a significant trade-off between education and employment. They are now able to do some domestic tasks or work outside of the household, and therefore be forced or decide to drop out of school. This situation could be particularly prone when there is a mining site within the district. For instance, Ahlerup et al. (2020) found that individuals who had gold mines within their district when they were adolescent have significantly lower educational attainment. This is explained by some myopic educational decisions when employment in gold mining is an alternative. However, some other papers reveal a positive effect of subsoil wealth on average years of primary, secondary, and tertiary education level (Stjins, 2001). In addition, the same reasons driving the benefits of mineral discoveries and productions on the primary education may still be playing a role, therefore rebalancing the negative effects at higher levels of education.

In Panel (B), we include control variables (see Table 29 in Appendix F for the coefficient associated with each control variables). The results remain unchanged. The coefficients associated with mineral discoveries and productions are still statistically not significant in all columns, except column (5). In the latter, the coefficient is positive and significant at 5 percent level, suggesting that the likelihood of downward secondary/tertiary educational IM of individuals born after mining discoveries tend to be higher than those born before the discoveries. However, this finding does not hold when we expand the treatment group by including individuals born 5 years before the discovery in column (6), meaning it is not robust.

B. Dynamic effects of mineral discoveries and productions on educational IM

In this subsection, we explore the dynamic effects of mining activities based on the time distance between the years of discoveries or productions and the birth years of individuals to test the parallel trends’ assumption. As explained previously, we conduct a leads-and-lags test following the literature (Angrist & Pischke, 2009, Maurer, 2019) to analyze whether the effects of mineral discoveries on IM tend to intensify the years after the shock. To do so, we estimate the likelihood of upward and downward educational IM for individuals born 0–5, 6–10 and 11–15 years after the discoveries or the beginning of mining production, and those born 5–10 and 10–6 years before the discoveries or the first year of mining production. The reference group is given by individuals born 11–15 years before the discoveries or the beginning of mining productions. The results are reported in Table 4 for upward IM (columns 1–2 and 5–6) and downward IM (columns 3–4 and 7–8).

Table 4:

Dynamic effects of mining activities on educational IM

article image
Note: This table presents the dynamic effects of mineral discoveries and productions on primary (columns 1–4) and secondary/tertiary (columns 5–8) educational IM. Standard errors are in parentheses. Disc stands for mining discovery. Prod stands for mining. *Indicates significance at 10% level, **significance at 5% level, and ***significance at 1% level.

The results are two folds. First, we find that mining activities affect the likelihood of primary educational IM for all age groups, while the effect is not statistically significant for secondary/tertiary educational IM. In fact, the results show that the coefficients associated with mineral discoveries and productions are positive and significant in columns 1–2, suggesting that mining activities tend to increase the likelihood of upward primary educational IM for all age groups of individuals. Inversely, the coefficients associated with mineral discoveries and productions are negative and significant in columns 3–4, meaning that mining activities are correlated with lower likelihood of downward primary educational IM for all age groups. On the other hand, the coefficients associated with mineral discoveries and productions are not statistically significant for all age groups of individuals in columns 5–8. This is in line with our findings in

that mining activities do not affect the likelihood of upward and downward secondary and tertiary educational IM. Second and more importantly, we find that the coefficients associated with mineral discoveries and productions are higher for individuals born later after the discovery and the beginning of mining production. For instance, the probability of upward primary educational IM is 7.6 pp. higher for individuals born 6–10 years after the beginning of mining production against only 1.1 pp. higher for those born 10–6 years before the beginning of mining production, as compared to those born 15–11 years before the beginning of mining of production (column 2). This higher probability could be explained by the fact that it takes time for mining activities to have an impact on the local communities, particularly in terms of infrastructure provision. It also supports that the parallel trend assumption might be verified, and therefore the effects we are capturing can be fully attributed to mineral discoveries and start of productions.

VII. Transmission channels

We explore the channels through which mining activities affect the likelihood of educational IM. We focus on two channels including the income effect proxied by parents working in the mining sector, and the returns to education.24 First, the increase of income for parents working in the mining sector, due to novel abundant opportunities, will allow them to invest more in their children education attainment (Becker and Tomes, 1979). Second, this new economic dynamism, creation of new jobs, will lead to an increasing demand for skilled workers. Thus, the higher returns or benefits to education in terms of wealth and income will motivate individuals to increase their educational attaintment relative to their parents (Torche, 2014).

To capture the income effect, we include an interactive variable between the discovery or production variable and a dummy equals to one if individuals have one of their parents working in the mining sector, and the latter itself as additional variables. For the return to education channel, we employ a two-step strategy where, in the first step, the effects of the mining discovery or production on the transmission channel variable is analyzed, and in the second step, we check the correlation between the transmission channel variable and upward/downward educational IM. This allows us to test whether the effect of mineral discoveries and productions on the probability of upward and downward educational IM transits through our transmission channel variable.

A. Income channel: parents working in mining sector

We first test whether the income effect proxied by parents working in the mining sector could be a channel through which mining activities affect the likelihood of upward/downward educational IM. One would expect that mining activities will create jobs in the local communities, therefore generating a source of income which could allow parents to invest more in their children education. We define a binary variable taking the value of one if one of the parents of the child is working in the mining sector, and zero otherwise and interact it with the treatment variable of mineral discoveries or productions. The results are displayed in Table 5, with the estimates about primary education in columns 1–4, and secondary and tertiary education in columns 5–8. We find that the coefficients associated with mineral discoveries or productions and the interactive term in columns 1–4 are statistically significant and have the expected sign. First, the mineral discoveries and productions increase (decrease) the likelihood of upward (downward) primary IM. Second and more importantly, these benefits of mining discovery or production are accentuated for individuals those one of the parents works in the mining sector. We find that having one of the parents working in the mining sector raises the likelihood of upward primary IM by 2.2 and 6.3 pp. following mineral discoveries and productions, respectively. Moreover, it diminishes the likelihood of downward primary IM by 1.1 pp. following mineral discovery. For secondary and tertiary IM, we find no significant effects of both the mineral discovery or production and its interactive term with parents working in mining. This suggests that the insignificant effects of mineral discoveries and productions hold for all individuals, independently of whether they have a parent working in the mining sector or not.

Table 5:

Income channel: parents working in the mining sector

article image
Note: This table presents the effects of mineral discoveries and productions on primary (columns 1–4) and secondary/tertiary (columns 5–8) educational IM, conditional on having of the parents working in the mining sectors. Standard errors are in parentheses. Yes Parents work in mining is a dummy taking the value of one for individuals with one of the parents working in the mining sector. Yes Mining is the treatment dummy variable for either discovery or production (depending on the columns). Disc-B/A means that the treatment group is after the discovery and the control group is before the discovery; Prod-B/A means that the treatment group is after the beginning of production and the control group is before production. *Indicates significance at 10% level, **significance at 5% level, and ***significance at 1% level.

B. Returns to education: Wealth index and LIDO score

We further test a second channel of return to education to check whether invididuals in districts with mineral discoveries and productions increase their level of education to get a higher wealth and income. Indeed, individuals exposed to the mineral discoveries and productions have a higher likelihood to work in the manufacturing and services sectors, and a lower likelihood to work in the agriculture sector. Given that the manufacturing and services sector may require a higher educational attainment and skills, individuals would have to increase their education level to work in these sectors, and thus having a better socio-economic status. To study the effect of mineral discoveries and productions on the returns to education, we first estimate a Mincer-like equation using a GDID model. The dependent variables are a proxy of wealth and income, given the lack of data on income and consumption in our database. The explanatory variables include the educational levels (primary and secondary/tertiary), the mineral discoveries and productions dummy, and their interactive term.

To do so, we construct the wealth index using a principal component analysis on several variables at the household level reflecting the economic status of household, following closely the Demographic and Health Survey wealth index (Rutstein and Staveteig, 2014).25 Second, we apply the lasso-adjusted industry, demographic, and occupational (LIDO) scores computed using actual data of labor market for the United States in 1950 by Saavedra and Twinam (2020) to our individuals in Africa. This LIDO score is an occupational income ranking score used as an alternative measure of income, socioeconomic status, and labor market outcome. It is dependent on (i) the fine categories of sectors of employment based on the industry classification (e.g., agriculture, mining and extraction, manufacturing, construction, hotels and restaurants, etc.), (ii) the occupation within employment based on the occupation classification (e.g., legislators, senior officials and managers, technicians and associate professionals, service works and market sales, elementary occupations), and (iii) individuals characteristics (e.g., age, gender). When applies to workers in Africa, the cross-individual, district and country differences at each period and over time would only come from the differences in the labor market conditions (sectors of activities, and occupation within employment) and demographics. We neutralize the effects of the demographic variables in the estimations by controlling for gender and age. We show in the Figure 17 and Figure 18 in Appendix G that the LIDO Score and Wealth index are strongly correlated with PPP GDP per capita at the country level, thereby implying that they are good proxies of income in Africa.

The results for the LIDO score and Wealth score are reported in Table 6 in columns 1–3 and 4–6, respectively. We find that an increase in the levels of education are associated with a higher LIDO score and Wealth index in all districts (with or without mineral discoveries) and all individuals (exposed to mineral discoveries/productions or not). This implies higher returns to education in Africa. More importantly, we find that the returns to education are higher in districts with mineral discoveries or productions (columns 1 and 4), and they are even greater for individuals exposed to the mineral discoveries and productions (columns 2–3 and 5–6, respectively). In a second step equation, we look how the LIDO score and Wealth index correlate with the educational IM. Table 7 shows a strong and positive (negative) association between the LIDO score and Wealth index with the upward (downward) primary educational IM. Thus, our findings suggest that in districts with mineral discoveries and productions, and specifically for individuals born after the mineral discoveries and productions, the returns to higher education has played a key role as an incentive to achieve higher educational levels and greater educational IM.

Table 6:

Returns to education channel: Mincer-type equation

article image
Note: This table presents a mincer-type equation where the effects of mineral discoveries and productions on the LIDO score (columns 1–3) and Wealth index (columns 4–6), conditional on educational attainment. Standard errors are in parentheses. Yes Mining is a dummy for district with discovery or not (columns 1 and 2) or the treatment dummy variable for either discovery (columns 2 and 5) or production (columns 3 and 6). With Disc. means with discovery. Disc-B/A means that the treatment group is after the discovery and the control group is before the discovery; Prod-B/A means that the treatment group is after the beginning of production and the control group is before production. *Indicates significance at 10% level, **significance at 5% level, and ***significance at 1% level.
Table 7:

Returns to education channel: primary educational IM and LIDO score, Wealth index

article image
Note: This table presents the effects of mineral discoveries and productions as well as of LIDO score and wealth index on primary upward (columns 1– 2) and secondary/tertiary (columns 3–4) educational IM. Standard errors are in parentheses. Mining is the treatment dummy variable for either discovery or production (depending on the columns). Disc-B/A means that the treatment group is after the discovery and the control group is before the discovery; Prod-B/A means that the treatment group is after the beginning of production and the control group is before production. *Indicates significance at 10% level, **significance at 5% level, and ***significance at 1% level.

To further reinforce this channel, we also whether mineral discoveries and productions affect the reallocation of individuals across the broad sectors, i.e., agriculture, manufacturing, and services. Specifically, the mineral discoveries and productions, by creating new opportunities, may change the economic propects at the local level. They can accelerate the demand for skilled workers by more capital-intensive and better paid activities in the manufacturing and services sectors, whereas individuals may be less prone to work in labor-intensive agricutlture sector. The results are reported in Table 8. First, we show that the likelihood to work in the agriculture sector decreases by 1.1 pp. for individuals exposed to the mineral discoveries or productions (see columns 1–2). Second, we also find that the likelihood to work in the manufacturing or services sector increases by around 0.8–1.4 pp., respectively (see columns 3–4 and 5–6). These findings suggest that mineral discoveries and productions lead to the shift from lower skilled to higher skilled jobs, and therefore boost the demand for better educated people in Africa.

Table 8:

Returns to education channel: effects of mineral discoveries and productions on sectoral employment

article image
Note: This table presents the effects of mineral discoveries and productions on the likelihood to work in the agricultural (columns 1–2), manufacturing (columns 3–4), and services (columns 5–6) sectors. Standard errors are in parentheses. Mining is the treatment dummy variable for either discovery or production (depending on the columns). Disc-B/A means that the treatment group is after the discovery and the control group is before the discovery; Prod-B/A means that the treatment group is after the beginning of production and the control group is before production. *Indicates significance at 10% level, **significance at 5% level, and ***significance at 1% level.

VIII. Robustness checks

We undertake various robustness tests to check the validity of our results to alternative samples and specifications.

A. Use of alternative control groups (only countries/districts with discoveries)

We start by using an alternative definition of the control groups. In our baseline, we compared the likelihood of upward and downward educational IM between individuals born after the discovery or the beginning of mining production, and those born before the discovery or beginning of production or in regions and countries without discoveries, regardless of their residing countries or districts. In this robustness check, we restrict the control group to individuals living in countries or districts where mining sites are discovered or with production activities. The results are reported in Table 9, with the estimates for primary educational IM being in Panel (A) and those for secondary and tertiary educational IM in Panel (B). As in the baseline, we find that the coefficients associated with mining activities are significant for primary educational IM, and insignificant for secondary and tertiary educational IM. Therefore, mining activities affect the likelihood of upward and downward primary educational IM in countries and districts with mineral resources, while the effect on secondary and tertiary educational IM is not statistically significant. Our results remain unchanged and are robust to the change of control groups and samples.

Table 9:

Robustness check: using different control groups

article image
Note: This table presents the effects of mineral discoveries and productions on upward and downward educational IM on a sample of countries (columns 1–4) and districts (columns 5–8) with mining activities. Standard errors are in parentheses. Disc-B/A means that the treatment group is after the discovery and the control group is before the discovery; Prod-B/A means that the treatment group is after the beginning of production and the control group is before production. *Indicates significance at 10% level, **significance at 5% level, and ***significance at 1% level.

B. Use of alternative structures of fixed effects and time variable

We then perform a series of consistency checks based on the structure of the fixed effects. The results are in Table 10, with Panel (A) being for primary education and Panel (B) for secondary and tertiary education. In columns 1–4, we replace the cohort fixed effects by birth year fixed effects and therefore compare individuals born within the same year instead of cohort since they may have experienced different shocks in 10 years. In columns 5–8, we include both the birth year fixed effects and the common time trend to capture the evolution of IM and rule out the possibility that individuals born before and after the discoveries or the beginning of mining production were already on differential growth trajectories in their education outcomes, i.e., a change in the educational IM indices that would have happened even in the absence of the mining activities. These factors could include particularly the family background of individuals (rich, poor and others). In columns 9–12, we control for cohort fixed effects and common time trend to filter out all persistent cohort related differences that could affect the likelihood of educational IM of individuals born before and after the discoveries of mining sites: for example, the availability of school infrastructure and change in the education system. Table 10 shows that our main findings remain unchanged even after accounting for all these different structures of fixed effects and time variable.

Table 10:

Robustness check: inclusion of fixed effects

article image
Note: This table presents the effects of mineral discoveries and productions on upward and downward educational IM, using different sets of fixed effects reported at the bottom of the table. Standard errors are in parentheses. Disc-B/A means that the treatment group is after the discovery and the control group is before the discovery; Prod-B/A means that the treatment group is after the beginning of production and the control group is before production. *Indicates significance at 10% level, **significance at 5% level, and ***significance at 1% level.

C. Use of alternative time window around the mineral discoveries or productions

Another important element of our analysis is the time window around the mineral discovery and production. In the baseline, we used a time window of 30 years (15 years before and after the discovery or the beginning of production). We test the robustness to alternative time windows, including 40 years (i.e., 20 years before and after), 50 years (25 years before and after) and 60 years (30 years before and after the discovery/production) to account for the individuals who take more time to complete their education and the potential long-lasting effect of mining activities. The results are reported in Table 11, with Panel (A) for primary education, and Panel (B) for secondary and tertiary education. We find that the coefficients associated with our variable of interest are significant for the primary education, while insignificant for the secondary and tertiary education, thus in line with the findings in the baseline estimates. This confirms that our baseline results are not driven by the choice of the time window around the discovery or the beginning of mining production. Moreover, the effects of mineral discoveries or productions tend to be long-lasting, also for cohorts born years after the discovery or production.

Table 11:

Robustness check: using alternative time window

article image
Note: This table presents the effects of mineral discoveries and productions on upward and downward educational IM, considering different time windows around discovery or production. Standard errors are in parentheses. Disc-B/A means that the treatment group is after the discovery and the control group is before the discovery; Prod-B/A means that the treatment group is after the beginning of production and the control group is before production. *Indicates significance at 10% level, **significance at 5% level, and ***significance at 1% level.

D. Use of alternative time exposition to the mineral discoveries or productions

We further explore whether our results are robust to the change in the time exposition to the mining activities. We expand the definition of the treatment group to account for individuals who may not have completed their education before the discovery or the beginning of mining production. Our baseline estimates compare the likelihood of individuals born before and after the start of mining activities. However, some individuals born before the start of mining activities may complete their education after the beginning of mining activities. In this case, those individuals will benefit from the economic and social impact from the exploitation of mining resources. We have already tried to account for these individuals by including in the treatment group the individuals born 5 years before the discovery/production in Table 2 and Table 3. In this robustness check, we include in the treatment group individuals born 10 and 15 years before the discovery or the start of mining production. To this end, we consider a longer time window of 60 years (30 years before and after the discovery/production). The results are reported in Table 12, with the primary educational IM in Panel (A), and secondary and tertiary educational IM in Panel (B). In all cases, we confirm our findings that mining activities have a statistically significant positive (negative) effect on the probability of upward (downward) primary educational IM, while the effect on the probability of secondary/tertiary educational IM is not statistically significant.

Table 12:

Robustness check: use of alternative time exposition to mining activities

article image
Note: This table presents the effects of mineral discoveries and productions on upward and downward educational IM, considering different treated group expositions. Standard errors are in parentheses. Disc-B/A means that the treatment group is after the discovery and the control group is before the discovery; Prod-B/A means that the treatment group is after the beginning of production and the control group is before production. *Indicates significance at 10% level, **significance at 5% level, and ***significance at 1% level.