Natural resource abundance and human capital accumulation

World Development Vol. 34, No. 6, pp. 1060–1083, 2006 Ó 2006 Elsevier Ltd. All rights reserved 0305-750X/$ - see front matter www.elsevier.com/locate/worlddev

doi:10.1016/j.worlddev.2005.11.005

Natural Resource Abundance and Human Capital Accumulation JEAN-PHILIPPE STIJNS * Northeastern University, Boston, MA, USA Summary. — This paper studies the link between resource abundance and human capital accumulation. It reviews the commonly used indicators of resource abundance and human capital accumulation. The case for a form of resource curse in human capital accumulation is not robust to reasonable changes in these indicators. In fact, subsoil wealth and resource rents per capita are shown to be significantly correlated with improved indicators of human capital accumulation. If mineral wealth is what authors have in mind when they refer to natural resource abundance, then they should choose indicators that measure this concept as accurately as possible. Ó 2006 Elsevier Ltd. All rights reserved. Key words — human capital, education, natural resources, resource curse, economic development

1. INTRODUCTION It is widely assumed in the literature that natural resources tend to slow down economic growth in countries that possess or discover them. Sachs and Warner (1995, 1999) have most notably made this claim. It deserves careful scrutiny if only because of its potential implications for development policy. In particular, the Extractive Industries Review (EIR) commissioned by the World Bank (2003) argues that international financial institutions should cease lending for hydrocarbon projects by 2008 and should limit lending for other mining activity to those countries with effective governments. In a press release, the World Bank (2004) has politely dismissed the basic thrust of the EIR’s recommendations and announced that ‘‘management [. . .] would continue investments in oil, gas, and mining production, as these will continue to be an essential part of the development of many poor nations.’’ However, subsequent developments seem to indicate that the questions raised by the EIR are likely to continue haunting the World Bank particularly its Mining Department. The purpose of this paper is to shed some new light on this debate. Because resource abundance is likely to have a large variety of possibly conflicting effects on different sectors

and functions of the economy, the paper restricts its attention to the link between resource abundance and human capital accumulation. The question at stake here is: Do natural resource-abundant countries tend to accumulate more or less human capital than resource-poor countries?

* We are grateful to several anonymous referees for useful suggestions and to George Akerlof, Pranab Bardhan, Brad DeLong, Chad Jones, Steve Pische, Ted Miguel, David Romer, Gavin Wright, and the participants of seminars at the University of California at Berkeley and Northeastern University for valuable discussion and suggestions. We have benefited from presenting related research at the 2001 WEAI Conference, the Second World Congress of Environmental and Resource Economists, the 2002 All-UC Economic History Conference, and the 2003 AEA meetings. We also thank Tamara Springsteen and Jennifer Jefferson for helpful editing, as well as Brian Baker, Oscar Brookings, Megan Bushor, and Alan Dyer for kindly proofreading a draft of this paper. We are solely responsible for the opinions expressed here and for any remaining errors. The beginning of this research was undertaken while the author was an SSRC Fellow in Applied Economics. Financial support from the Social Science Research Council is gratefully acknowledged. Final revision accepted: November 9, 2005.

1060

NATURAL RESOURCE ABUNDANCE AND HUMAN CAPITAL ACCUMULATION

That human capital accumulation accompanies mineral activities should be positive news from the perspective of economic development. Human capital accumulation is a crucial issue for economic development in all countries. Barro (1997, 2001) argues that education permanently increases the efficiency of the labor force by fostering democracy. He also argues that human capital facilitates the absorption of superior technologies from leading countries. This technology-absorption effect is supposed to be especially important at the secondary and higher education levels. Similarly, Aghion, Caroli, and Garcia-Penalosa (1999) assert that education creates better conditions for good governance by improving health and enhancing equality. Development economists, most notably Sen (1999), stress the importance of education, and in particular the importance of educating women in developing countries. The marginal social returns of education for growth are considered sizeable at the human capital levels characteristic of developing economies. Additionally, given the high degree of income inequality prevailing in these countries, education is often considered a better indicator of the median level of development than gross domestic product per capita. Let us briefly review the limited literature dealing with the nexus between resource abundance and human capital accumulation. Gylfason (2001) shows that public expenditure on education relative to national income, expected years of schooling for girls, and gross secondary enrollment are all inversely related to the share of natural capital in national wealth across countries. He concludes that natural capital appears to crowd out human capital, thereby slowing the pace of economic development. Gylfason asserts, ‘‘nations that are confident that their natural resources are their most important asset may inadvertently—and perhaps even deliberately!—neglect the development of their [other] resources, by devoting inadequate attention and expenditure to education.’’ He goes on to add, ‘‘their natural wealth may blind them to the need for educating their children.’’ Birdsall, Pinckney, and Sabot (2001) start by observing that most governments around the world extol the benefits of education while claiming that their investment in education is limited because of a lack of money. Indeed, the EIR (World Bank, 2003) reports that governments ‘‘believe that [extractive] industries

1061

contribute to [. . .] poverty alleviation [and that] revenues from extractive industries can be used for [. . .] education [. . .]’’ (Vol. 3, Annex 5, p. 102). As Birdsall et al. (2001) note, if limits on human capital investment primarily result from binding government constraints, resource abundance should induce additional investment, all else equal. Yet, these authors argue that the data tell another story: resource-abundant countries, on average, invest less in education than other countries. Just how surprising we can find the paradoxical result reported by Birdsall and her coauthors is debatable. On one hand, Wade (1992) argues that, in Latin America for example, governments controlled by the owners of natural resources have no incentive to invest in basic skills. The idea is that in resource-abundant countries, with plentiful foreign exchange, there is no incentive for the political elite to invest in basic skills so as to export the manufactures needed to pay for imports. Rather, the resource-owning elite have a tendency in these circumstances to use the country’s resources to invest in highly skilled labor, particularly in the form of college-level education for their children. On the other hand, it is surprising that while mineral states tend to lavishly spend their revenues on numerous development projects and programs (see, e.g., Ascher, 1999), education would be the only exception. It is even more surprising to read that in regard to education, the same mineral states actually spend less than other states. In fact, in an under appreciated paper about resource abundance and economic growth, Davis (1995) finds human capital accumulation indicators to be higher in mineral countries than non-mineral countries. The results in this paper support his conclusions. This paper explains why Gylfason (2001) and Birdsall et al. (2001) have reached different conclusions. It improves upon Davis (1995) by using richer human capital data and better resource abundance measures. This paper is organized as follows. Section 2 presents the data used, paying close attention to the different indicators of resource abundance and human capital accumulation used in the literature in order to better understand why they lead us to strikingly different conclusions. Section 3 reports and comments on linear correlation coefficients between these various resource abundance and human capital accumulation indicators. These correlation coefficients are bootstrapped to generate confidence intervals

1062

WORLD DEVELOPMENT

and deal with non-normality and sample representativity issues. Section 4 discusses some extensions to the results in Section 3 results and suggests some candidates for country case studies based on the variables surveyed in this paper. Finally, Section 5 presents the conclusions of this paper and offers some recommendations for future research. 2. DATA One of the main purposes of this paper is to compare how strongly associated various measures of resource abundance are with different measures of human capital accumulation. To this end, we have collected data about the most commonly used variables to measure resource abundance and human capital accumulation. This section describes these variables and clarifies the differences among the indicators used in the literature. Section 2(a) deals with resource abundance indicators, while Section 2(b) discusses human capital accumulation indicators. (a) Resource abundance indicators (i) Share of natural capital in national wealth Gylfason (2001) introduces the share of natural capital in national wealth. He defines it as ‘‘natural capital’’ over the sum of the same natural capital, the perpetual inventory value of ‘‘produced assets’’—referred to in this paper simply as ‘‘physical capital’’—and ‘‘human resources’’ measured as a residual, that is, as the present value of GDP net of the other national wealth components. In other terms, this ratio, which we call the natural capital:total capital ratio, is defined as Share of natural capital in national wealth 

Natural Capital Natural Capital þ Produced Assets þ Human Resources

The original variables come from the World Bank (1997) and are estimated for 1994. Natural capital is the sum of ‘‘subsoil wealth,’’ and what this paper refers to as ‘‘green capital,’’ that is, timber, non-timber benefits of forests, cropland, pastureland, and the opportunity cost of protected areas. In turn, subsoil wealth is the present value of a constant stream of economic profits on ‘‘resource rents’’ on various fuels and minerals, that is, gross profit on extraction less depreciation of capital and normal return on capital.

(ii) Natural capital:physical capital ratio We introduce the natural capital:physical capital ratio with the aim of removing a major source of distortion implicit to the share of natural capital in national wealth. The presence of human capital in the denominator of Gylfason’s ratio is troublesome if the intended use of this ratio is to correlate it with human capital accumulation itself. Indeed, if Country A is more successful at investing a share of its natural wealth in education than Country B, then ceteris paribus Gylfason’s ratio will identify Country B as more resource abundant than Country A. This renders suspicious claims made using this ratio that resource-abundant countries invest less in education. (iii) Subsoil wealth:physical capital ratio We introduce this second ratio, subsoil wealth:physical capital, with the purpose of distinguishing between the effects of subsoil wealth (minerals and fuels) and green capital. Besides the caveats mentioned below about primary export intensity, it is hard to find an argument that would imply that the human capital accumulation effects of the opportunity cost of protected areas should be similar to those of primary products, let alone minerals and fuels. In any case, it is doubtful that protected areas are what authors have in mind when they make the claim that resource abundance is detrimental to human capital accumulation. (iv) Green capital:physical capital ratio The idea behind this third ratio is to see what role resources other than subsoil wealth play in driving the correlation between the share of natural capital in national wealth (or natural capital:total capital ratio) and human capital accumulation measures. This definition implies that what we call ‘‘green capital’’ is equal to natural capital minus subsoil wealth. (v) Subsoil wealth per capita This indicator suggests itself naturally when we measure human capital accumulation by average years of education, which is an indicator of human capital accumulation per capita. Further, scaling subsoil wealth by population rather than physical or total capital removes an important bias intrinsic to other scaling factors. Indeed, if a country is successfully industrialized based, at least in part, on its mineral wealth, then its physical capital stock—and thus total capital—are likely to reflect this success. The problem is that the subsoil

NATURAL RESOURCE ABUNDANCE AND HUMAN CAPITAL ACCUMULATION

wealth:physical capital and the subsoil wealth: total capital ratios are going to attribute these ‘‘success stories’’ to resource-poor countries by construction, thereby heavily stacking the cards in favor of finding that these ‘‘resourcepoor’’ cases of faster economic development have corresponded to faster accumulation of human capital. Scaling subsoil wealth by population avoids this problem. Dividing subsoil wealth, as defined above, by country population yields subsoil wealth per capita. (vi) Arable land per capita Birdsall et al. (2001) use Auty’s (1995, 1997, 2001) classification of countries. Auty’s complete classification is based on arable land per capita, country size, and the nature of exports. Yet, as Birdsall et al. use it, a country is categorized as resource-poor simply if cropland per head was smaller than 0.3 hectares in 1970. It is this categorization that Birdsall et al. (2001, Tables 4.1 and 4.2, pp. 58 and 63–64, respectively) rely on solely to conclude that resource-rich countries accumulate less human capital than resource-poor countries. However, it is questionable whether arable land per capita can be used as a reliable indicator of resource abundance when many notoriously resourcerich countries, such as Saudi Arabia, have little arable land to speak of. Additionally, the 0.3 hectares in 1970 sounds somewhat arbitrary, and we would like to analyze the impact of switching from this binary classification to a continuous variable. Therefore, we use arable land per capita instead. We calculate this variable using arable land and population data from the World Bank’s World Development Index (WDI) database. (vii) Primary export intensity Sachs and Warner (1995) proposed primary export intensity, defined as primary export, thus encompassing all agricultural, mineral, and energy exports, divided by gross national product. Sachs and Warner measured this variable in 1970, and they applied it to economic growth rather than human capital accumulation. However, because of the popularity of this indicator in the literature, we also consider it. This paper reproduces Sachs and Warner’s calculations for every year in the sample, using the World Trade Data 1 sets described in Feenstra, Lipsey, and Bowen (1997). Sachs and Warner use SITC (Standard International Trade Classification) categories 0–2, 4, and 68, corresponding to non-fuel primary activities. 2 They also

1063

include SITC category 3 covering fuels. For the denominator, they use Gross National Product. We take this variable from the World Bank’s WDI database. Stijns (2005) criticizes the use of primary export intensity precisely for being a measure of natural resource intensity of exports, thus trade specialization in the primary sector, rather than the intended measure of resource abundance, which is a direct measure of factor endowment. Further, it is doubtful that the agricultural and mining sectors play similar roles in economic development. Justifications for making an empirical distinction between these two sectors are numerous. These justifications include the use of very different technologies and the fact that the rents typically generated in each sector differ substantially in magnitude. However, Sachs and Warner’s measure lumps these two very different sectors together. (viii) Agricultural export intensity There are two reasons behind the use of agricultural export intensity is this paper. The first is that it should indicate whether a measure of trade specialization in agriculture differs in its effects on human capital accumulation from an endowment measure such as arable land per capita. The second reason for using this indicator is to check for the consequences of aggregating the agricultural sector with the mineral and fuel sectors when using Sachs and Warner’s primary export intensity variable. This ratio is obtained by combining data from the World Bank’s WDI about the share of agricultural raw materials, as well as about food exports in total merchandise exports, with data on total merchandise exports and total output from the World Trade data sets. 3 (ix) Mineral export intensity While mineral export intensity has the same numerator as the share of minerals in exports, the denominator is gross domestic product rather than total exports. This paper introduces this indicator to check whether the choice of scaling factor for mineral exports, total exports, or gross domestic product has an influence on results. The second purpose of this ratio is to illuminate the implications of using Sachs and Warner’s primary export intensity as an indicator of resource abundance, and, thus, of including primary activities other than fuel and mineral extraction. GDP data for the denominator comes from the WDI database.

1064

WORLD DEVELOPMENT

(x) Share of minerals in exports Davis (1995) uses mineral exports as a share of total exports. He collected data for 1991. We have computed this variable for all years in the dataset, again using the World Trade Data. Like Davis, we use SITC categories 3, 68, and 971, covering all mineral and energy exports. This paper uses ‘‘share of minerals in exports’’ and ‘‘mineral export share’’ interchangeably throughout. This ratio is aimed at measuring the extent of a country’s trade specialization in mineral exports. An important justification for using this ratio is that trade specialization in minerals may affect human capital accumulation differently than mineral abundance. Indeed, the (standard) Hecksher–Ohlin theorem only implies that the exports of a country be intensive in their abundant factors. In the case of minerals, this theorem does not necessarily imply that a mineral-abundant country (as compared to the rest of the world) exports a higher share of minerals, just that its exports be more mineral-intensive. Therefore, a country could be richly endowed in minerals and not concentrate its exports in the minerals sector. In other words, a country could export manufactured products that embed its mineral production and yet export few minerals per se. Finally, there is the issue of how relevant a country’s stock of human capital is to trade specialization in the minerals sector. de Ferranti et al. (2002, Chap. 2) present results that imply that countries with a comparative advantage in tropical agriculture, raw materials, petroleum, and labor-intensive manufactures are typically characterized by lower-than-average education and knowledge scores. These results make us expect a negative correlation between trade specialization in minerals (which should result from a comparative advantage therein) and human capital. Unfortunately, data limitations brought de Ferranti and his coauthors to exclude deposits and minerals as determinants of comparative advantage.

specific average costs of extraction from the world price for the resource in question, all expressed in current US dollars. For minerals, the levels of total resource rents are calculated as

(xi) Resource-rent intensity We introduce resource-rent intensity to capture the contribution of mineral and fuel rents to the aggregate income of resource-abundant countries. This ratio uses the World Bank’s (1997) estimates for resource rents, further explained in Hamilton and Clemens (1999). The basic approach used by World Bank to calculating resource rents for non-renewable resources is to subtract country- or region-

(b) Human capital accumulation indicators

Rent  World price  mining costs  milling and beneficiation costs  smelting costs  transport to port  ‘‘normal’’ return to capital. For crude oil, unit rents are calculated as the world price less lifting costs. Currently, natural gas does not have a single world price yet. Hamilton and Clemens estimate its world price by averaging free-on-board prices from several points of export worldwide, following which they calculate unit rents as they do for oil. In addition to timber, coal, oil, and natural gas, the minerals they cover include zinc, iron ore, phosphate rock, bauxite, copper, tin, lead, nickel, gold, and silver. Data problems led them to exclude diamonds from their estimates. Dividing this variable by current price GDP data gives us resource-rent intensity. (xii) Resource rent per capita Introducing resource rents per capita captures the contribution of rents in the mineral industry to income per capita. This contribution in absolute terms to income per capita measures the mineral revenues available (on average) for each inhabitant to accumulate human capital. This indicator suggests itself naturally when we measure human capital accumulation using indicators that are averages over the country’s (relevant) population, such as average years of education, literacy rates, life expectancy at birth, and, to a lesser extent, secondary enrollment rates. Dividing resource rents by a country’s population yields resource rents per capita. Multiplying this number by the GDP deflator, also taken from the World Bank’s WDI, yields real rents per capita.

(i) Average years of education Average years of education is the most direct and preferred indicator of accumulation of human capital at the country level. Gylfason (2001) uses this data for females. This data for females and for both genders combined comes from Barro and Lee (2001). Barro and Lee (1993) extend their previous (1993) esti-

NATURAL RESOURCE ABUNDANCE AND HUMAN CAPITAL ACCUMULATION

mates of educational attainment up to 1995 and provide projections for 2000. Reporting results for females is important because, first, we would expect the female educational attainment variable to capture the median level of human capital accumulation. Second, female education is considered important on its own if only with respect to labor market participation and demographic transition in developing countries. (ii) Net secondary enrollment rate The net secondary enrollment rate has been very common in the empirical economic growth literature, starting as early as Mankiw, Romer, and Weil (1992). Gylfason (2001) and Birdsall et al. (2001) also use this indicator. Davis (1995) uses primary school enrollment instead. The net secondary enrollment rate is defined as the number of children of official secondary school age enrolled in secondary school divided by the size of the population in the same age bracket. 4 These data come from the UNESCO Institute for Statistics via the WDI database. It is possible to argue in favor of the use of a lagged school enrollment indicator to measure current human capital on the grounds that today’s useful human capital in terms of labor productivity is the result of schooling some years ago. However, this paper does not consider a lag in secondary enrollment rate because it is conceptually difficult to imagine that today’s stock of mineral reserves influenced yesterday’s investment in human capital more than today’s investment in human capital. Moreover, the more we extend the lag in the secondary enrollment rate, the more we run the risk of capturing reverse causality running from yesterday’s human capital stock to today’s mineral reserves. Indeed, David and Wright (1997) argue that during the second half of the 19th century and the first half of the 20th century, the US mineral base reflected in large part the superior geological knowledge of American engineers. (iii) Adult literacy rate Davis (1995) and Birdsall et al. (2001) use this indicator. The adult literacy rate is particularly interesting in the context of developing countries and when the distribution of human capital is a concern. Indeed, literacy rates tell us more about the median skill levels than other average indicators. This paper’s adult literacy rate data comes from the UNESCO Institute for Statistics via the WDI database.

1065

(iv) Life expectancy at birth Davis (1995) uses life expectancy at birth for both genders. Health is considered to be an important component of workers’ human capital because workers’ efficiency often depends critically on their health conditions, particularly in developing countries. Life expectancy at birth should also be more informative than average indicators regarding the median level of human capital accumulation. This paper’s life expectancy data are World Bank staff estimates based on various sources and they are collected from the WDI database. (v) Public expenditure on education as a percent of aggregate expenditure Gylfason (2001) uses public expenditure on education as a percent of aggregate expenditure. This ratio aims to measure the amount of resources used as input in the public production of human capital. However, there are several caveats to be made. First, public expenditure on education as a percent of aggregate expenditure does not take into account the needs of the population even if it does provide a measure of the political ‘‘effort’’ at providing public education with respect to the size of the economy. Second, by definition, public expenditure as a share of GDP does not cover private expenditure on education. It could be that resource rents trickle down on at least the upper and middle classes and allow them to afford more private schooling, particularly in developing countries. Or conversely, public expenditure on education funded by resource rents may crowd out private expenditure, so that any positive association between resource abundance and public expenditure on education may be misleading. However important these considerations, a rigorous empirical investigation of these hypotheses falls outside the more modest scope of this paper. Fortunately, the other human capital variables, namely average years of education, net secondary enrollment rate, adult literacy, and life expectancy at birth, do not suffer from this public vs. private complication. We collect data on public expenditure on education as a percent of aggregate expenditure from the UNESCO Institute for Statistics via the WDI database. 3. MAIN CORRELATIONS This section analyzes simple linear (Pearson) correlations between the natural resource

1066

WORLD DEVELOPMENT

abundance and human capital accumulation indicators discussed in the previous section. Tables 1 and 2 report these correlation coefficients. Additionally, these tables provide biascorrected and bootstrapped 95% confidence intervals for the reported correlation coefficients. Finally, the tables report whether correlation coefficients test as significantly different from zero at the usual 10–5–1% level of confidence based on bootstrapped 90–95–99% confidence intervals. Table 1 reports correlation coefficients among the complete sample of observed countries. On the grounds that the effect of resource abundance in developing countries may operate on human capital accumulation differently than in the complete sample of countries, Table 2 reports correlation coefficients within the subsample of observed developing countries. Section 3(b) discusses Table 1, whereas Section 3(c) discusses Table 2. Resource abundance indicators can be classified into two categories: measures available across countries only for a single year, and measures available across countries and over time. The upper part of Tables 1 and 2 reports correlations involving cross-sectional indicators of resource abundance, whereas the lower part of Tables 1 and 2 reports correlations involving panel indicators of resource abundance. Before discussing these results, let us first explain the paper’s use of bootstrapping. (a) Bootstrapping confidence intervals for correlation coefficients Many key observations made in this section and the following concern the statistical significance of correlation coefficients between resource abundance and human capital indicators. There are two major issues regarding such observations. First, it is unclear just how representative is the sample of the larger set of countries that exist or could exist. Second, hypothesis testing usually relies on some assumption regarding the population that may or may not be true. In the case of correlation coefficients, it is typically assumed that the variables are following a bi-normal distribution in the population. It turns out that an omnibus test rejects the null hypothesis of bi-normality for the vast majority of the pairs of variables considered in this paper. 5 Fortunately, the statistical method known as bootstrapping (Efron, 1979) is known to do a surprisingly effective job of addressing both concerns.

Due to limited space, it is not possible to explain bootstrapping here. Efron and Tibshirani (1993) offer a detailed introduction to bootstrapping, and Lunneborg (1985) explains how the application of bootstrapping to correlation coefficients is done. Tables 1–4 report the bias-corrected 95% bootstrapped confidence intervals of the correlation coefficients. These tables also report for each correlation coefficient a test of statistical significance at the 10%, 5%, or 1% level. It is straightforward to test this hypothesis based on the corresponding 90–95–99% confidence interval. The correlation coefficient is statistically significant when the whole corresponding interval and thus both of its bounds fall on the same side of zero. (b) Correlations among all countries (i) Cross-sectional indicators of resource abundance The World Bank estimates for the components of country wealth are only available for 1994 and, therefore, this is also the case for ratios such as the natural capital:total capital ratio (Gylfason’s share of natural capital in natural wealth), the natural capital:physical capital ratio, the subsoil wealth:physical capital ratio, and the green capital:physical capital ratio, as well as for subsoil wealth per capita. The upper part of Table 1 provides Pearson correlation coefficients between these measures and human capital accumulation indicators among all countries. The net secondary enrollment rates, adult literacy rates, life expectancy at birth, and public expenditure on education as a share of GDP are all averaged over the period 1991– 95. The reason for this averaging is to keep as many countries as possible for the calculation of correlation coefficients despite the fact that there are many countries for which these measures are not reported every year. The exception is total years of education for both genders and for females. These data are available at 5-year intervals, and this table uses values for 1995. Gylfason’s ratio, the natural capital:total capital ratio in 1994, is significantly (a 6 .01) associated with lower values for all human capital accumulation indicators. The same is true of the natural:physical capital ratio. However, correlations are weaker between human capital indicators and the natural:physical capital ratio than with Gylfason’s ratio. This consistent weakening indicates that the presence of human capital in the denominator of Gylfason’s ratio is biasing results in the direction of his

Total years of education

Secondary enrollment rate (net)

Adult literacy rate

Life expectancy at birth

Public expenditure on education (% of GDP)

0.69* (0.78, 0.58) [70]

0.71* (0.83, 0.55) [44]

0.60* (0.75, 0.39) [61]

0.78* (0.85, 0.67) [81]

0.58* (0.66, 0.48) [72]

1994

0.55* (0.63, 0.44) [70]

0.54* (0.63, 0.43) [70]

0.58* (0.69, 0.41) [44]

0.54* (0.73, 0.2) [61]

0.56* (0.67, 0.4) [81]

0.42* (0.5, 0.33) [72]

1994

0.04 (0.23, 0.14) [60]

0.02 (0.22, 0.16) [60]

0.23* (0.37, 0.07) [37]

0.03 (0.24, 0.37) [50]

0.00 (0.27, 0.15) [67]

0.23* (0.35, 0.09) [60]

1994

0.55* (0.63, 0.44) [60]

0.54* (0.62, 0.41) [60]

0.69* (0.82, 0.49) [37]

0.51* (0.75, 0.1) [50]

0.53* (0.69, 0.32) [67]

0.41* (0.57, 0.32) [60]

1994

0.45* (0.27, 0.57) [60]

0.45* (0.27, 0.57) [60]

0.06 (0.3, 0.36) [37]

0.04 (0.07, 0.39) [50]

0.16* (0.07, 0.34) [67]

0.39** (0.06, 0.66) [60]

All

Females

1994

0.69* (0.78, 0.55) [70]

Share of natural capital in national wealth

Natural capital:physical capital ratio

Subsoil wealth:physical capital ratio

Green capital:physical capital ratio

Subsoil wealth per capita

(continued next page)

NATURAL RESOURCE ABUNDANCE AND HUMAN CAPITAL ACCUMULATION

Table 1. Correlations across all countries

1067

1068

Table 1—continued Total years of education

Secondary enrollment rate (net)

Adult literacy rate

Life expectancy at birth

Public expenditure on education (% of GDP)

Females

1975–2000

0.17* (0.06, 0.25) [584]

0.17* (0.05, 0.25) [583]

0.11** (0.03, 0.18) [380]

0.08 (0.16, 0.02) [603]

0.02 (0.09, 0.05) [878]

0.09** (0, 0.17) [542]

1975–2000

0.10*** (0.21, 0) [536]

0.09** (0.21, 0.01) [536]

0.08*** (0.25, 0.03) [314]

0.02 (0.08, 0.09) [471]

0.09* (0.16, 0.03) [668]

0.07** (0.12, 0.01) [504]

1975–2000

0.12* (0.2, 0.04) [423]

0.10** (0.18, 0.01) [423]

0.10** (0.22, 0.03) [271]

0.01 (0.1, 0.08) [400]

0.12* (0.19, 0.04) [553]

0.14* (0.2, 0.07) [400]

1975–2000

0.22* (0.3, 0.15) [506]

0.23* (0.31, 0.1 6) [505]

0.27* (0.37, 0.16) [324]

0.13* (0.21, 0.04) [467]

0.20* (0.27, 0.12) [671]

0.12* (0.19, 0.04) [459]

1975–2000

0.07 (0.16, 0.02) [555]

0.06 (0.15, 0.02) [555]

0.03 (0.18, 0.08) [319]

0.02 (0.07, 0.09) [494]

0.02 (0.09, 0.02) [703]

0.03 (0.08, 0.03) [523]

1975–2000

0.18* (0.23, 0.14) [486]

0.20* (0.24, 0.15) [486]

0.27* (0.34, 0.2) [257]

0.05 (0.13, 0.03) [407]

0.17* (0.22, 0.11) [569]

0.16 (0.06, 0.36) [547]

1975–2000

0.08** (0, 0.15) [475]

0.08** (0, 0.16) [475]

0.02 (0.15, 0.11) [244]

0.16* (0.07, 0.23) [395]

0.06 (0.02, 0.13) [551]

0.03 (0.02, 0.1) [547]

Arable land per capita

Primary export intensity

Agricultural export intensity

Mineral export share

Mineral export intensity

Resource rent intensity

Resource rents per capita

Notes: 95% bias-corrected bootstrapped confidence internal between brackets; # of observations between square brackets. * Significant at the 1% level. ** Significant at the 5% level. *** Significant at the 10% level.

WORLD DEVELOPMENT

All

Total years of education All

Females

Secondary enrollment rate (net)

Adult literacy rate

Life expectancy at birth

Public expenditure on education (% of GDP)

0.56* (0.73, 0.34) [57]

0.67* (0.79, 0.52) [60]

0.45* (0.62, 0.28) [52]

0.50** 0.51* (0.67, 0.08) (0.71, 0.15) [25] [57]

0.45* (0.6, 0.26) [60]

0.39* (0.52, 0.25) [52]

0.08 (0.21, 0.46) [46]

0.19 (0.14, 0.38) [47]

0.11 (0.1, 0.38) [41]

0.48*** (0.72, 0) [46]

0.40* (0.6, 0.12) [47]

0.41* (0.61, 0.23) [41]

0.06 (0.03, 0.43) [46]

0.20** (0.06, 0.26) [47]

0.36** (0.12, 0.57) [41]

Share of natural capital in national wealth

0.63* 0.62* 0.59* (0.77, 0.42) (0.75, 0.42) (0.79, 0.17) 1994 [49] [49] [25]

Natural capital:physical capital ratio

0.53* (0.63, 0.29) 1994 [49]

Subsoil wealth:physical capital ratio 1994 Green capital:physical capital ratio

0.39** (0.07, 0.64) [40]

0.50* (0.6, 0.29) [49] 0.39** (0.06, 0.63) [40]

0.10 (0.2, 0.46) [19]

0.51* 0.63* 0.55* (0.67, 0.39) (0.62, 0.33) (0.86, 0.21) 1994 [40] [40] [19]

Subsoil wealth per capita 1994

0.53** (0.33, 0.68) [40]

0.56** (0.38, 0.7) [40]

0.05 (0.33, 0.55) [19]

(continued next page)

NATURAL RESOURCE ABUNDANCE AND HUMAN CAPITAL ACCUMULATION

Table 2. Correlations among developing countries

1069

1070

Table 2—continued Total years of education

Adult literacy rate

Life expectancy at birth

Public expenditure on education (% of GDP)

0.13** (0.23, 0.01) [433]

0.01 (0.14, 0.14) [264]

0.05 (0.14, 0.05) [555]

0.17* (0.25, 0.09) [705]

0.11* (0.17, 0.05) [425]

1975–2000

0.02 (0.13, 0.13) [386]

0.03 (0.1, 0.14) [386]

0.08 (0.19, 0.07) [204]

0.07 (0.03, 0.13) [428]

0.01 (0.09, 0.05) [509]

0.10** (0.02, 0.2) [387]

1975–2000

0.03 (0.13, 0.06) [301]

0.00 (0.1, 0.09) [301]

0.12*** (0.25, 0.01) [172]

0.05 (0.05, 0.14) [360]

0.03 (0.11, 0.05) [418]

0.04 (0.05, 0.14) [301]

1975–2000

0.03 (0.08, 0.14) [406]

0.04 (0.07, 0.13) [406]

0.01 (0.13, 0.15) [211]

0.05 (0.05, 0.11) [452]

0.03 (0.03, 0.09) [548]

0.12* (0.05, 0.19) [407]

1975–2000

0.09*** (0.18, 0) [366]

0.11** (0.2, 0.01) [365]

0.14** (0.26, 0.01) [215]

0.09*** (0.17, 0) [421]

0.12* (0.22, 0.03) [511]

0.10** (0.02, 0.23) [346]

1975–2000

0.05 (0.12, 0.03) [354]

0.08** (0.14, 0.01) [354]

0.13*** (0.23, 0) [161]

0.01 (0.08, 0.1) [377]

0.04 (0.11, 0.05) [431]

0.56* (0.22, 0.73) [424]

0.15* (0.05, 0.24)

0.16* (0.05, 0.25)

0.10 (0.06, 0.27)

0.21* (0.13, 0.29)

0.15** (0.05, 0.23)

0.15* (0.08, 0.24)

Females

1975–2000

0.15* (0.26, 0.04) [434]

Arable land per capita

Primary export intensity

Agricultural export intensity

Mineral export intensity

Mineral export share

Resource rent intensity

Resource rents per capita 1975–2000

Notes: 95% bias-corrected bootstrapped confidence internal between brackets; # of observations between square brackets. * Significant at the 1% level. ** Significant at the 5% level. *** Significant at the 10% level.

WORLD DEVELOPMENT

Secondary enrollment rate (net)

All

Total years of education

Secondary enrollment rate (net)

Adult literacy rate

Life expectancy at birth

Public expenditure on education (% of GDP)

All

Females

1994

0.07 (0.24, 0.13) [77]

0.05 (0.24, 0.16) [77]

0.06 (0.26, 0.11) [50]

0.00 (0.26, 0.25) [67]

0.09 (0.24, 0.12) [90]

0.07 (0.15, 0.07) [79]

1994

0.07 (0.26, 0.3) [39]

0.05 (0.31, 0.29) [39]

0.06 (0.37, 0.2) [28]

0.03 (0.34, 0.23) [28]

0.08 (0.33, 0.17) [43]

0.11 (0.36, 0.12) [38]

1994

0.07 (0.26, 0.22) [43]

0.03 (0.21, 0.22) [43]

0.31 (0.47, 0.2) [28]

0.00 (0.2, 0.33) [33]

0.03 (0.12, 0.16) [47]

0.05 (0.3, 0.31) [43]

1994

0.10 (0.16, 0.36) [40]

0.14 (0.13, 0.38) [40]

0.19 (0.5, 0.17) [29]

0.12 (0.19, 0.37) [28]

0.03 (0.18, 0.2) [43]

0.02 (0.29, 0.33) [40]

1994

0.22 (0.41, 0.04) [76]

0.20 (0.4, 0.02) [76]

0.24 (0.5, 0.12) [47]

0.28** (0.49, 0.05) [65]

0.38* (0.55, 0.18) [87]

0.15 (0.3, 0.08) [77]

1994

0.45* (0.55, 0.29) [76]

0.44* (0.54, 0.32) [76]

0.58* (0.69, 0.44) [49]

0.51** (0.66, 0.21) [67]

0.49* (0.6, 0.36) [89]

0.34* (0.42, 0.27) [78]

Mineral wealth:physical capital ratio

Coal wealth:physical capital ratio

Oil wealth:physical capital ratio

Gas wealth:physical capital ratio

Timber wealth:physical capital ratio

Cropland wealth:physical capital ratio

(continued next page)

NATURAL RESOURCE ABUNDANCE AND HUMAN CAPITAL ACCUMULATION

Table 3. Disaggregated resource abundance indicators

1071

1072

Table 3—continued Total years of education

Adult literacy rate

Life expectancy at birth

Public expenditure on education (% of GDP)

0.17 (0.08, 0.34) [77]

0.08 (0.09, 0.27) [50]

0.15 (0.16, 0.27) [67]

0.14 (0.08, 0.26) [90]

0.06 (0.07, 0.23) [79]

1994

0.10 (0.25, 0.33) [22]

0.05 (0.26, 0.3) [22]

0.04 (0.51, 0.31) [16]

0.25 (0.38, 0.47) [11]

0.16 (0.2, 0.33) [22]

0.23 (0.46, 0.1) [21]

1994

0.17 (0.12, 0.46) [44]

0.19 (0.07, 0.47) [44]

0.27 (0.5, 0.28) [29]

0.00 (0.18, 0.31) [33]

0.07 (0.04, 0.22) [48]

0.24 (0.15, 0.61) [44]

1994

0.37** (0.06, 0.56) [41]

0.39* (0.15, 0.58) [41]

0.05 (0.28, 0.35) [29]

0.18 (0.13, 0.37) [29]

0.21** (0.05, 0.36) [44]

0.30 (0.09, 0.61) [41]

1994

0.32*** (0.01, 0.49) [39]

0.31*** (0.1, 0.48) [39]

0.13 (0.17, 0.32) [28]

0.09 (0.18, 0.26) [28]

0.15 (0.17, 0.29) [43]

0.12 (0.1, 0.29) [38]

1994

0.51* (0.33, 0.63) [76]

0.52* (0.34, 0.63) [76]

0.43* (0.25, 0.56) [47]

0.22** (0.03, 0.37) [65]

0.32* (0.2, 0.41) [87]

0.60* (0.37, 0.76) [77]

Females

1994

0.15 (0.11, 0.33) [77]

Mineral wealth per capita

Coal wealth per capita

Oil wealth per capita

Gas wealth per capita

Timber wealth per capita

Cropland wealth per capita

Notes: 95% bias-corrected bootstrapped confidence internal between brackets; # of observations between square brackets. * Significant at the 1% level. ** Significant at the 5% level. *** Significant at the 10% level.

WORLD DEVELOPMENT

Secondary enrollment rate (net)

All

All countries Average years of secondary education

Developing countries Average years of tertiary education

Average years of secondary education

Average years of tertiary education

All

Females

All

Females

All

Females

All

Females

1994

0.64* (0.73, 0.54) [70]

0.65* (0.73, 0.54) [70]

0.53* (0.65, 0.39) [70]

0.50* (0.61, 0.37) [70]

0.55* (0.72, 0.36) [49]

0.55* (0.7, 0.39) [49]

0.47* (0.64, 0.25) [49]

0.46* (0.6, 0.28) [49]

1994

0.47* (0.54, 0.36) [70]

0.47* (0.57, 0.37) [70]

0.39* (0.48, 0.3) [70]

0.37* (0.45, 0.26) [70]

0.42* (0.54, 0.24) [49]

0.41* (G.52, 0.27) [49]

0.33* (0.45, 0.17) [49]

0.32* (0.42, 0.19) [49]

1994

0.09 (0.28, 0.08) [60]

0.07 (0.27, 0.1) [60]

0.06 (0.25, 0.13) [60]

0.05 (0.25, 0.14) [60]

0.42* (0.1, 0.61) [40]

0.43* (0.14, 0.64) [40]

0.23 (0.14, 0.53) [40]

0.22 (0.1, 0.58) [40]

1994

0.48* (0.58, 0.38) [60]

0.48* (0.59, 0.38) [60]

0.40* (0.48, 0.29) [60]

0.38* (0.46, 0.27) [60]

0.46* (0.58, 0.31) [40]

0.45* (0.55, 0.28) [40]

0.36* (0.45, 0.21) [40]

0.34* (0.43, 0.21) [40]

1994

0.40* (0.17, 0.57) [60]

0.43* (0.2, 0.58) [60]

0.36* (0.2, 0.58) [60]

0.35* (0.19, 0.56) [60]*

0.56* (0.37, 0.73) [40]

0.61* (0.42, 0.76) [40]

0.34** (0.09, 0.55) [40]

0.33* (0.1, 0.61) [40]

% of natural capital in national wealth Natural capital:physical capital ratio Subsoil wealth:physical capital ratio Green capital:physical capital ratio Subsoil wealth per capita

(continued next page)

NATURAL RESOURCE ABUNDANCE AND HUMAN CAPITAL ACCUMULATION

Table 4. Secondary vs. tertiary education

1073

1074

Table 4—continued All countries Average years of secondary education All **

Arable land per capita

Developing countries Average years of tertiary education

Females **

All

Average years of secondary education

Females *

*

All

Average years of tertiary education

Females *

0.33 (0.4, 0.25) [434]

*

All

Females *

0.30 0.16 0.13** (0.38, 0.22) (0.24, 0.06) (0.22, 0.03) [434] [434] [434]

19752000

0.14* (0.2, 0.09) [536]

0.13* 0.14* 0.13* (0.19, 0.08) (0.19, 0.09) (0.17, 0.08) [536] [536] [536]

1975–2000

0.24* (0.3, 0.16) [423]

0.22* 0.17* 0.14* 0.20* 0.16* 0.16* 0.14* (0.29, 0.15) (0.24, 0.08) (0.22, 0.06) (0.27, 0.12) (0.23, 0.07) (0.25, 0.07) (0.22, 0.05) [423] [423] [423] [301] [301] [301] [301]

1975–2000

0.08* 0.07* 0.10* 0.09* (0.13, 0.03) (0.12, 0.02) (0.14, 0.05) (0.14, 0.05) [555] [555] [555] [555]

0.03 (0.06, 0.13) [406]

0.04 (0.06, 0.13) [406]

0.04 (0.11, 0.02) [406]

0.04 (0.1, 0.02) [406]

1975–2000

0.17* (0.23, 0.09) [506]

0.02 (0.08, 0.14) [366]

0.00 (0.1, 0.1) [366]

0.03 (0.07, 0.12) [366]

0.01 (0.1, 0.09) [366]

1975–2000

0.19* 0.20* (0.23, 0.14) (0.24, 0.16) [486] [486]

Primary export intensity Agricultural export intensity Mineral export intensity Mineral export share Resource rent intensity Resource rents per capita 1975–2000

0.05 (0.02, 0.13) [475]

0.12 (0, 0.21) [584]

0.17* (0.24, 0.1) [506]

0.07 (0.01, 0.14) [475]

0.28 (0.16, 0.38) [584]

0.25 (0.14, 0.35) [584]

0.12* 0.13* (0.19, 0.06) (0.19, 0.06) [506] [506]

0.03 (0.12, 0.04) [386]

0.02 (0.1, 0.05) [386]

0.09* 0.08* (0.15, 0.04) (0.14, 0.03) [386] [386]

0.16* (0.2, 0.11) [486]

0.16* (0.19, 0.11) [486]

0.07*** (0.13, 0) [354]

0.09** (0.15, 0.02) [354]

0.07 (0.12, 0.03) [354]

0.09** (0.14, 0.01) [354]

0.08** (0.01, 0.17) [475]

0.06 (0.01, 0.16) [475]

0.12** (0.03, 0.21) [349]

0.13* (0.04, 0.22) [349]

0.09** (0, 0.18) [349]

0.06 (0.02, 0.16) [349]

Notes: 95% bias-corrected bootstrapped confidence internal between brackets; # of observations between square brackets. * Significant at the 1% level. ** Significant at the 5% level. *** Significant at the 10% level.

WORLD DEVELOPMENT

1975–2000

0.10 (0, 0.19) [584]

NATURAL RESOURCE ABUNDANCE AND HUMAN CAPITAL ACCUMULATION

conclusions. In fact, footnote 3 in Gylfason (2001) mentions that a referee was concerned about this issue, but the author dismissed it as inconsequential because results can be reproduced by ‘‘using different measures of natural resource abundance, such as the share of the primary sector in the labor force [. . .] or the share of primary exports in total exports or GDP [. . .]’’ (bottom of p. 2). Indeed, this paper finds that there are other resource abundance indicators, generally those that adopt a loose definition of resource abundance, that are negatively associated with human capital accumulation. However, this is not a reason—to the contrary—for defining resource abundance in a way that is likely to, and here observed to, bias results toward confirming the hypothesis of a negative effect on human capital accumulation. As we move from the natural:physical capital ratio to the subsoil wealth:physical capital ratio, correlations weaken further, often becoming insignificant (and even changing sign in the case of the adult literacy rate and life expectancy at birth). This consistent weakening of correlations indicates that the presence of the non-mineral and non-fuel components of natural wealth in the numerator of Gylfason’s ratio was also biasing results in the direction of his findings. In comparison, the presence of elements of ‘‘green capital’’ in the numerator of the natural:total capital ratio appears to be biasing results much more than the inclusion of human capital in the numerator does. The strongly negative and significant (a 6 .01) correlations observed for the green capital:physical capital ratio confirm this interpretation. While the subsoil wealth:physical capital ratio is significantly (a 6 .01) associated with lower ‘‘human capital input’’ measures (secondary enrollment rate and public expenditure on education as a share of GDP), it is not significantly associated with lower (or higher) ‘‘human capital output’’ measures (total years of education, adult literacy rate, and life expectancy at birth). Finally, rescaling subsoil wealth on a per capita basis instead of as a ratio to physical capital reverses results. Subsoil wealth per capita is significantly (a 6 .01) associated with more years of education (for both genders and for females), longer life expectancy at birth, and a larger share of GDP devoted to public education. Subsoil wealth per capita is not significantly associated with the secondary enrollment and adult literacy rates. These positive correlations indicate a possible bias of ratios that scale resource abundance by variables that could be

1075

themselves positively influenced by resource abundance. In the case of physical capital, Deaton (1999, p. 11) argues that ‘‘the revenue from commodity exports provides a potential source of investment funds’’ even though the quality of these investments may be questioned, especially in Africa. In other words, if a resourceabundant country successfully invests resource rents into physical capital, the type of ratios in question here will identify this country as less resource abundant, everything else equal. If the same type of country also successfully invests in human capital, this bias will mislead the researcher toward concluding that resource abundance is asso- ciated with a failure to invest in human capital. (ii) Panel indicators of resource abundance The lower part of Table 1 reports correlations between (time-varying) human capital accumulation indicators and time-varying measures of resource abundance. The net secondary enrollment and adult literacy rates, life expectancy at birth, and public expenditure on education as a share of GDP are averaged over five periods of 5 years beginning with 1971–75 and ending with 1991–95. All resource abundance indicators are averaged in the same way. Again, the reason for this averaging is to keep as many countries as possible in the calculation of correlations, despite the fact that for many countries these measures are not available for every single year. The exception is again total years of education for all and for females only. These data are available at 5-year intervals, and therefore values for 1975–80–85–90–95 have been used. All five intervals are pooled to compute the correlation coefficients reported in Table 2. However, qualitatively similar results are obtained when these correlations are computed instead period by period. The most striking feature of the lower part of Table 1 is the contrast between per capita resource abundance measures and resource intensity measures. With the exception of mineral export intensity, all intensity measures are typically significantly associated with lower human capital indicators. In contrast, arable land per capita and resource rents per capita are both typically associated with higher human capital indicators. This is again evidence that the scaling of resource abundance by development-related measures such as physical capital stocks may mislead the researcher toward finding an artificially negative association between resource abundance and human capital accumulation.

1076

WORLD DEVELOPMENT

There are also some interesting patterns to be observed when comparing correlations across resource intensity measures. Correlations weaken when we move from primary export intensity to mineral export intensity, to the extent of making all associations insignificant. The difference made by excluding agricultural exports from primary export intensity when using mineral export intensity indicates that specialization in agricultural exports plays a significantly different role for human accumulation than does mineral export specialization. This interpretation is supported by the observation that agricultural export intensity is typically significantly associated with lower human capital indicators (a 6 .05 except for the adult literacy rate). Finally, it also appears that human capital indicators are more significantly and negatively correlated with the share of minerals in exports than with mineral export intensity. One possible interpretation is to see this as a reflection of the fact that export concentration per se would be detrimental to human capital accumulation. Another candidate interpretation is related to the fact that trade specialization in minerals is detrimental to human capital accumulation. This detrimental effect could be linked to the notoriously high volatility of commodity prices, causing increased volatility of government revenues. Obviously, further evidence is required to confirm (or reject) such interpretations. (c) Correlations among developing countries Although growth theory generally considers developing countries together with other countries and in doing so assumes that a similar data generation process applies for developing and non-developing countries, development economists tend to look at developing countries in isolation on the grounds that their economies do not necessarily function and evolve in the same way as industrialized countries. Table 2 reproduces the statistics of Table 1 while restricting the sample to developing countries. This helps to evaluate to what extent the association between resource abundance and human capital indicators is different in the subsample of developing countries than in the complete sample. (i) Cross-sectional indicators of resource abundance With a few notable exceptions, it turns out that the conclusions reached concerning devel-

oping countries in Table 2 are of a similar nature to those reached in Table 1 regarding all countries. Generally speaking, the direction and significance of associations in Table 2 are similar to what is found in Table 1. In particular, in the upper part of Table 1, the natural capital and subsoil wealth:physical capital ratio are both significantly (a 6 .01, except in one case) associated with lower human capital indicators in developing countries. Among developing countries, correlations also weaken when we move from Gylfason’s ratio to the natural capital:physical ratio, with the same interpretation as in Section 2(b). Here too, the correlations with human capital accumulation indicators decrease when we remove green capital from the denominator of the ratio. So much so here, that among developing countries there is a positive correlation between the subsoil wealth:physical capital ratio and all human capital accumulation indicators. All correlations are significant (a 6 .05) except the secondary enrollment and adult literacy rates. Importantly, the correlation between the subsoil wealth:physical capital ratio and total years of education is both positive and significant. This positive and significant correlation is found for both females only and for the whole population. This finding is remarkable, as it can be argued that the subsoil wealth:physical capital ratio and total years of education are quite precise indicators of natural resource abundance and human capital accumulation. (ii) Time-varying indicators of resource abundance At the bottom of Table 2, in contrast to what was found across all countries, arable land per capita is negatively and significantly correlated with most indicators of human capital accumulation. A possible interpretation is that land abundance plays a different role in developing countries than it does in other countries. Perhaps the difference in technologies used in the agricultural sector in the industrial vs. the developing world can somehow explain the reversal of the corresponding correlation coefficients. However, the observation that agricultural export intensity is now (except in the case of the secondary enrollment rate) insignificantly correlated with human capital indicators warrants caution about accepting this interpretation of the data. In fact, all resource abundance intensity indicators display weaker and often insignificant associations with human

NATURAL RESOURCE ABUNDANCE AND HUMAN CAPITAL ACCUMULATION

capital indicators. In spite of this tendency, the share of mineral exports typically remains significantly associated with lower human capital indicators. In contrast, resource rents per capita see their correlation coefficient become larger and more significant (a 6 .01, except in one case). Finally, the share of public expenditure on education in GDP is positively associated with all panel indicators of resource abundance, with the exception of arable land per capita. Overall, there is some evidence that resource abundance plays a more important role for human capital accumulation in developing countries than in other countries. Perhaps this is not surprising in the context of the usual tax-collection difficulties faced by governments in developing countries. Indeed, according to Hirschman (1977), there is presumably a tradeoff between production and government revenue linkages. An activity such as manufacturing, which is highly interlinked with the rest of the economy, is expected to have a strong political lever with which to avoid taxation. Conversely, enclave economies are by definition economically isolated and are often run by foreigners. Hence, mineral extraction activities represent

1077

fewer votes, have less political leverage, and are very often the object of heavy corporate income and export taxation. Additionally, as Hughes (1975, p. 813) points out, ‘‘following Ricardian logic, the supply of mineral is in- elastic in the short-run, so that with a given demand the resource rent can be siphoned off without affecting the amount of mineral that will be mined.’’ Therefore, perhaps it is not so surprising to observe that, particularly in developing countries, resource rents generate an increase in the ability of governments to spend on, among other programs, public education. 4. EXTENSIONS This section explores possible extensions of the data analysis conducted in Section 3. First, Table 3 disaggregates measures of resource abundance to see if different types of natural resources impact differently on human capital accumulation. Second, Table 4 decomposes average years of education, leading to a discussion of whether resource abundance helps tertiary education at the expense of secondary education. Third, we use Figures 1 and 2 to

Figure 1. Subsoil wealth:physical capital ratio vs. total years of education. Both variables standardized.

1078

WORLD DEVELOPMENT

Figure 2. Subsoil wealth:physical capital ratio vs. total years of education for females. Both variables standardized.

suggest some natural candidate for future casestudies. (a) Disaggregating resource abundance indicators It can be argued that just as agricultural production and mineral extraction may have different effects on human capital accumulation, possibly because they generate (taxable) rents of significantly different magnitude, there is a case for also disaggregating other measures of resource abundance. It is also important to again address in this context the issue of the effect of the choice of scaling factor for various types of natural resources, namely total or physical capital vs. population. To these ends, Table 3 presents correlation coefficients between disaggregated measures of resource abundance and human capital accumulation indicators. Specifically, the upper part of Table 3 considers the ratio of mineral, coal, oil, gas, timber, and cropland wealth to physical capital. The lower part of Table 3 considers mineral, coal, oil, gas, timber, and cropland wealth per capita. Recall that mineral, coal, oil, and gas wealth are components of subsoil wealth, whereas timber and cropland wealth are com-

ponents of green capital. This data comes from the World Bank (1997) and the corresponding variables are estimated for 1994. In the upper part of Table 3, none of the ratios to physical capital based on disaggregated measures of subsoil wealth are significantly correlated with human capital accumulation. In this sense, minerals and fuels seem to have a similarly neutral effect on human capital accumulation, as was found in the upper part of Table 1 using the subsoil wealth:natural capital ratio. The difference is that in Table 3, the correlations with the secondary enrollment rate and the share of GDP devoted to public education are also insignificant. In contrast, the timber wealth:natural capital ratio is significantly associated with lower adult literacy rates and life expectancy at birth. The cropland:natural capital ratio is significantly (a 6 .01) associated with lower values for all human capital indicators. The bottom part of Table 3 yields similar conclusions in that none of the mineral, oil, or coal wealth per capita is significantly associated with any human capital indicator, whereas in the upper part of Table 1, subsoil wealth per capita is significantly associated with all human capital indicators except for the secondary enrollment and adult literacy rates. A possible

NATURAL RESOURCE ABUNDANCE AND HUMAN CAPITAL ACCUMULATION

interpretation of this data feature is that the range of resources a country is endowed with is important in terms of spurring human capital accumulation. Specifically, while a country may manage to exploit a single resource without investing much in education (perhaps by importing specialized labor), the exploitation of a diverse resource base requires the development of a wide range of skills that are only available to countries with deep stocks of human capital. The exception is gas wealth per capita, which is significantly associated (a 6 .05) with higher years of education for both genders and for females, as well as with longer life expectancy at birth. There is, however, a strong contrast between the upper and lower parts of Table 3 in that timber and cropland wealth per capita are in fact significantly associated with a higher number of average years of education. In addition, cropland per capita is also significantly associated with the rest of the human capital indicators. The same remark that is made in Section 3 applies here: scaling by a variable related to another helpful outcome of resource abundance biases results toward finding a negative relation between resource abundance and human capital accumulation. Beyond this, the difference made by scaling a country’s population as opposed to its physical capital stock can be interpreted as simply implying that being well endowed per capita with respect to agriculture or forestry does not threaten human capital accumulation, to the contrary. Yet, the negative correlations found for the timber wealth:physical capital ratio and even more so for cropland wealth:physical capital ratio may indicate that excessive productive specialization, possibly via excessive export specialization, may be detrimental to human capital accumulation, perhaps because of the volatility of the underlying prices. Alternatively, the nature of the production technologies involved in agriculture and forestry may play a role here too. The observations made in Table 3 regarding timber wealth per capita are perhaps not so surprising, as the forestry industry is known to be quite human capital intensive in industrialized countries. The successful examples of Sweden and Finland come to mind in this context (de Ferranti, Perry, Lederman, & Maloney, 2002, Chap. 3). (b) Secondary vs. tertiary education According to Wade (1992, p. 311), ‘‘governments [in Latin America] controlled by an elite

1079

that owned the natural resources saw no particular need to extend basic education throughout the labor force, for the exploitation of the natural resources did not depend on a large supply of basically skilled people. But the elites did want to expand tertiary education sufficiently to professionalize their own children. Hence, secondary education was often expensive; tertiary education was often cheap.’’ This type of statement calls for investigating whether in our data there is a tendency for resource abundance indicators to be more systematically correlated with higher average years of tertiary education or lower average years of secondary education. In fact, there is evidence for the alternate hypothesis of greater focus on secondary than tertiary education. Table 4 correlates our resource abundance indicators with average years of secondary and tertiary education for both genders and for females only. These data come from Barro and Lee (2001). Generally, average years of secondary and tertiary education are correlated in a similar way with resource abundance indicators as with average years of (total) education in Tables 1 and 2. Notably, arable land per capita is also significantly associated with higher years of secondary and tertiary education for both genders and females across all countries, while it is associated with lower values for these variables across developing countries. Subsoil wealth per capita is significantly associated with higher years of secondary and tertiary education for both genders and females both across all countries and across developing countries. In developing countries, while the subsoil wealth:physical capital ratio is significantly associated with higher average years of secondary education for both genders and for females, it is not significantly associated with average years of tertiary education. Again in developing countries, while primary export intensity is significantly associated with lower average years of tertiary education for both genders and for females, it is not significantly associated with average years of secondary education. Agricultural export intensity is significantly associated in developing countries with lower average years of secondary and tertiary education for both genders and for females. The evidence is ambiguous with respect to resource rents per capita. Across all countries, resource rents per capita are significantly associated with higher years of education for both genders, but they are not significantly correlated with years of secondary education and

1080

WORLD DEVELOPMENT

years of tertiary education for females. In contrast, across developing countries, resource rents per capita are significantly associated with higher average years of secondary and tertiary education for both genders, but insignificantly associated with years of tertiary education for females. Similarly, across developing countries, resource-rent intensity is significantly associated with lower years of tertiary education for females but not for both genders. Therefore, there is some evidence that in resource-abundant developing countries, tertiary education is expanded more systematically to the benefits of males rather than females. Subsoil wealth per capita, however, does not support this interpretation. (c) Case study candidates Although space limitations prevent this paper from exploring case studies, it is worth noting some natural candidates and suggesting questions for future country case studies. Figures 1 and 2 scatter plot subsoil wealth per capita vs. average years of education for both genders and females only, respectively. Additionally, linear regression lines appear on the figures for reference, and 90–95–99% elliptical confidence areas 6 appear for the two variables of interest under the assumption of bi-normality for these two population variables. In other words, each ellipse gives the area in which we expect observed combinations of subsoil wealth:physical capital ratios and average years of education to fall with no more than 10%, 5%, or 1% probability of failure, respectively. What can we learn from Figure 1? With respect to the link between resource abundance and human capital stocks, it appears that Chile (CL) has achieved very high average years of education compared to developing countries on average, and also among resource-abundant developing countries. At a lower level of subsoil wealth:physical capital ratio, two other countries stand out. Malaysia (MY) appears to have outperformed similarly resource-abundant developing countries, whereas Sierra Leone (SL) has clearly underperformed compared to the same countries. The 6-year difference (more than 2 standard deviations) between the average years of education for both genders observed between these two countries is remarkable because the countries are similar with respect to their subsoil wealth:physical capital ratios (0.27 for Malaysia vs. 0.29 for Sierra Leone, less than 7% of a standard deviation apart). Figure 2

yields the same qualitative observations with respect to average years of education for females. The literature might well benefit from casestudying these countries with a view to answering questions such as: why have countries like Malaysia so clearly outperformed equally resource-abundant countries like Sierra Leone in terms of human capital accumulation. Likewise, why has Chile not proportionally benefited from its higher resource abundance compared to Malaysia in terms of human capital accumulation? In other words, is there country-level evidence that there is some kind of non-linearity in the positive relationship observed at the aggregate level between resource abundance and human capital stocks? Finally, what are the complementary factors that increase the odds that a resource-abundant country invests its resource rents into human capital? 5. CONCLUSIONS There are obviously many possible extensions to and many unresolved questions about the set of indicators covered here. Indeed, even the average number of years can still be considered a relatively raw measure of human capital accumulation when we factor in that in some cases years of education do not necessarily translate into marketable skills. Similarly, one important potential byproduct of education is social capital. Unfortunately, we can only wish that more countries have data on educational achievement scores and social capital indicators. To claim a negative and significant correlation between resource abundance and human capital accumulation using Pearson correlations, we must arbitrarily pick indicators on both sides of the correlation. As a corollary, such claims are not robust to reasonable—and actually desirable—changes in the choice of indicators. Gylfason (2001) and Birdsall et al. (2001) make an important contribution to the development literature by analyzing some of the existing data regarding the nexus between resource abundance and human accumulation. However, they conclude in favor of a negative effect running from resource abundance to human capital accumulation because of the use of questionable resource abundance indicators. Gylfason’s share of natural capital in national wealth suffers from incorporating elements such as non-timber benefits of forests and the opportunity cost of protected areas

NATURAL RESOURCE ABUNDANCE AND HUMAN CAPITAL ACCUMULATION

that have little to do with a strict definition of natural resources, that is, a definition restricted to minerals and fuels. In fact, when we look at correlations between human capital indicators and the subsoil wealth:physical capital ratio instead, we find insignificant correlation coefficients for all countries and positive and significant coefficients among developing countries. Birdsall et al. (2001) use Auty’s (2001) country classification to conclude that resource-rich countries, compared to resource-poor countries, underinvest in human capital. As discussed earlier, there are some fundamental problems with their use of Auty’s classification system and their choice of an arbitrary land per capita threshold. In fact, if we look instead at correlations between arable land per capita as a continuous—rather than categorical—variable, we find positive and significant correlations between this variable and total years of education, the net secondary enrollment rate, and the share of aggregate expenditure devoted to public education. To find negative correlation coefficients between arable land per capita and human capital accumulation indicators, we have to restrict the sample to developing countries, which is actually what Birdsall et al. (2001) do in their paper. It is not the purpose of this paper to argue that restricting attention to developing countries is inappropriate. However, arable land per capita and agricultural export intensity tell us more about the effect on education of a comparative advantage and specialization in the agricultural sector than they tell us about the effect of natural resource abundance (interpreted in its usual sense as mineral wealth) on human capital accumulation. While among all countries, agricultural export intensity is negatively and significantly associated with human capital accumulation indicators, the corresponding correlation coefficients are not significant among developing countries. If development policy is the focus, it would therefore be advisable to proceed to further investigation before concluding that moving away from agriculture

1081

might favor the accumulation of human capital. Future papers on the effect of natural resource abundance on various aspects of macroeconomic development would benefit from establishing a clear distinction between factor endowments in the agricultural vs. mining sector. In other words, if mineral wealth is what authors have in mind when they refer to natural resource abundance, then they should use resource abundance indicators that appropriately measure this type of endowment and refrain from using indicators that capture a comparative advantage in the agricultural sector. Additionally, the choice of the variable used to scale resource abundance also matters. It would be wiser to abstain from scaling factors that are endogenous to a country’s development process because they tend to bias results by systematically underestimating the resource abundance of countries that successfully accumulate factors of production. Certainly, it is possible to argue that there is much more to analyzing the effects of mineral endowments and exploitation on development than the seemingly narrow focus on human capital accumulation adopted here. There is a long list of other channels of operation running from resource abundance to economic development; to name just one example, mineral activities are likely to have a profound and lasting impact on the environment. These other channels of operation should continue to be the object of research. Nevertheless, as the Extractive Industries Review argues, ‘‘consumption of finite resources could be considered sustainable if it improves the welfare of future generations by, for example, raising other forms of capital, such as human capital (if revenues are used, say, for education) or social capital’’ (World Bank, 2003, Vol. 1, p. 4). The evidence offered in this paper suggests that, at a minimum, we should carefully consider human capital accumulation before attempting to discourage mineral production, especially in developing countries.

NOTES 1. The World Trade Database (WTDB), put together by Statistics Canada, contains bilateral trade flows for all countries from 1970 to 1992, recently updated up to

1997, classified according to the Standard International Trade Classification. This database is considered to cover 98% of all trade.

1082

WORLD DEVELOPMENT

2. We have added category 971, non-monetary gold, as Davis (1995) does, for the sake of comprehensiveness. 3. See Section (vii) above for further details about this dataset. 4. In contrast, the gross enrollment rate is the ratio of the number of students enrolled in secondary school, regardless of age, to the population of the corresponding official school age. Rates above 100 can occur when

there are a lot of overage students in secondary schools. Such rates can indicate educational system ineffectiveness. 5. The results are not reported in this paper but are available from the author upon request. 6. Refer to Tracy, Young, and Mason (1992) for further details.

REFERENCES Aghion, P., Caroli, E., & Garcia-Penalosa, C. (1999). Inequality and economic growth: The perspective of the new growth theories. Journal of Economic Literature, 37(4), 1615–1660. Ascher, W. (1999). Why governments waste natural resources: Policy failures in developing countries. Baltimore and London: The John Hopkins University Press. Auty, R. M. (1995). Patterns of development: Resources, policy and economic growth. London and New York: Edward Arnold. Auty, R. M. (1997). Natural resource endowment, the state and development strategy. Journal of International Development, 9(4), 651–663. Auty, R. M. (2001). Resource abundance and economic development: Introduction and overview. In R. M. Auty (Ed.), Resource abundance and economic growth (pp. 3–16). Oxford University Press. Barro, R. J. (1997). Determinants of economic growth: A cross-country empirical study. Cambridge, MA: MIT Press. Barro, R. J. (2001). Human capital and growth. American Economic Review, 91(2), 12–17. Barro, R. J., & Lee, J.-W. (1993). International comparisons of educational attainment. Journal of Monetary Economics, 32(3), 363–394. Barro, R. J., & Lee, J.-W. (2001). International data on educational attainment: Updates and implications. Oxford Economic Papers, 53(3), 541–563. Birdsall, N., Pinckney, T., & Sabot, R. (2001). Natural resources, human capital, and growth. In R. M. Auty (Ed.), Resource abundance and economic growth (pp. 57–75). Oxford: Oxford University Press. David, P. A., & Wright, G. (1997). Increasing returns and the genesis of American resource abundance. Industrial and Corporate Change, 6(2), 203–245. Davis, G. A. (1995). Learning to love the Dutch disease: Evidence from the mineral economies. World Development, 23(10), 1765–1779. Deaton, A. (1999). Commodity prices and growth in Africa. Journal of Economic Perspectives, 13(3), 23–40. de Ferranti, D., Perry, G., Lederman, D., & Maloney, W. (2002). From natural resources to the knowledge economy: Trade and job quality. Washington, DC: The World Bank.

Efron, B. (1979). Bootstrap methods: Another look at the jackknife. The Annals of Statistics, 7(1), 1–26. Efron, B., & Tibshirani, R. (1993). An introduction to the bootstrap. New York: Chapman & Hall. Feenstra, R. C., Lipsey, R. E., & Bowen, H. P. (1997). World trade flows, 1970–1992, with production and tariff data. NBER Working Paper Series, 5910, 1–65. Gylfason, T. (2001). Natural resources, education, and economic development. European Economic Review, 45(4–6), 847–859. Hamilton, K., & Clemens, M. (1999). Genuine savings rates in developing countries. World Bank Economic Review, 13(2), 333–356. Hirschman, A. O. (1977). A generalized linkage approach to development with special reference to staples. Economic Development and Cultural Change, 25(Suppl.), 67–98. Hughes, H. (1975). Economic rents, the distribution of gains from mineral exploitation, and mineral development policy. World Development, 3(11&12), 811–825. Lunneborg, C. E. (1985). Estimating the correlation coefficient: The bootstrap approach. Psychological Bulletin, 98, 209–215. Mankiw, N. G., Romer, D., & Weil, D. N. (1992). A contribution to the empirics of economic growth. Quarterly Journal of Economics, 107(2), 407–437. Sachs, J. D., & Warner, A. M. (1995). Natural resource abundance and economic growth. NBER Working Paper Series, 5398, 1–47. Sachs, J. D., & Warner, A. M. (1999). Natural resource intensity and economic growth. In J. Mayer, B. Chambers, & A. Farooq (Eds.), Development policies in natural resource economies (pp. 13–38). Cheltenham, UK and Northampton, MA: Edward Elgar. Sen, A. (1999). Development as freedom. Oxford: Oxford University Press. Stijns, J.-Ph. (2005). Natural resource abundance and economic growth revisited. Resources Policy, 30(2), 107–130. Tracy, N. D., Young, J. C., & Mason, R. L. (1992). Multivariate control charts for individual observations. Journal of Quality Technology, 24(2), 88–95. Wade, R. (1992). East Asia’s economic success: Conflicting perspectives, partial insights, shaky evidence. World Politics, 44(2), 270–320.

NATURAL RESOURCE ABUNDANCE AND HUMAN CAPITAL ACCUMULATION World Bank. World Development Indicators. World Bank: http://publications.worldbank.org/WDI/. World Bank (1997). Are we saving enough for the future?. In Expanding the measure of wealth: Indicators of environmentally sustainable development (pp. 7–18). Washington, DC: World Bank.

1083

World Bank (2003). Striking a better balance. Final Report of the Extractive Industries Review, December. World Bank: http://www.eireview.org/. World Bank (2004). World Bank Group Board agrees way forward on extractive industries review. Washington, DC: World Bank.