Resource abundance and internal armed conflict: Types of natural resources and the incidence of ‘new wars’

EC O L O G IC A L E C O N O M IC S 6 7 ( 2 0 08 ) 50 3 –5 13

a v a i l a b l e a t w w w. s c i e n c e d i r e c t . c o m

w w w. e l s e v i e r. c o m / l o c a t e / e c o l e c o n

ANALYSIS

Resource abundance and internal armed conflict: Types of natural resources and the incidence of ‘new wars’ Heinz Welsch Department of Economics, University of Oldenburg, 26111 Oldenburg, Germany

AR TIC LE I N FO

ABS TR ACT

Article history:

Recent armed domestic conflicts have been described as being related to natural resource

Received 8 August 2007

abundance and as being characterized by new features not present in earlier internal

Received in revised form

conflicts (multiplicity of actors, devastation of production structures). The paper develops

3 January 2008

and tests a framework that captures both the role of natural resource abundance and the

Accepted 7 January 2008

stylized facts from the descriptive literature in a simple two-sector model in which violent

Available online 12 February 2008

appropriation of natural resources imposes a negative externality on the production sector. The model predicts that the probability of armed conflict varies directly with the size and

Keywords:

value of ‘lootable’ resource endowments and inversely with variables that increase labor

Natural resources

productivity. In contrast to mineral resources, abundance of agricultural resources reduces

Mineral resources

conflict probability, by raising labor productivity. These predictions are supported by cross-

Agricultural resources

country ordered probit estimations. In quantitative terms, the negative effect of agricultural

Conflict

resources on conflict probability is almost twice as large as the positive effect of mineral

Externalities

resources.

Economic development

© 2008 Elsevier B.V. All rights reserved.

JEL classification: Q34; O13; D74

1.

Introduction

The appropriation and exploitation of natural resources have frequently been mentioned as a cause of civil wars. While several recent pertinent studies both from economics (for instance Collier and Hoeffler, 1998, 2004; De Soysa, 2000) and political science (for instance Kaldor, 1999; Klare, 2001) suggest that natural resource abundance is an important determinant of the occurrence of internal armed conflict, Ross (2004) finds the empirical linkage between natural resources and civil war to be fragile and proposes that the resource–conflict relationship should be differentiated with

E-mail address: [email protected].

respect to both the type of natural resource and the kind of civil conflict.1 With respect to the type of natural resource, early studies (starting with Collier and Hoeffler, 1998) have linked the incidence of civil war to the extent of primary commodity exports. More recently, the stock of natural capital, both renewable and

1 It will become clear below that it is the availability of easily appropriable natural resources which is seen as important for internal conflict. It is the availability aspect which is captured by the term ‘abundance’ in this paper. Through comparative advantage, resource abundance may give rise to resource dependence, that is, dependence of an economy on natural resource based activities or natural resource exports. In this sense, resource abundance refers to a more fundamental notion than resource dependence.

0921-8009/$ – see front matter © 2008 Elsevier B.V. All rights reserved. doi:10.1016/j.ecolecon.2008.01.004

504

EC O LO GIC A L E CO N O M ICS 6 7 ( 2 00 8 ) 5 0 3 –5 13

non-renewable, has been suggested to yield a more precise and differentiated measure of abundance (De Soysa, 2000). In addition, several authors emphasize that not only the physical configuration of the resource matters, but also the political and economic environment (Fearon, 2005, Humphreys, 2005, Lujala et al., 2005).2 With respect to the type of conflict, writers with a politicalscience background emphasize that internal conflict in the post cold-war period is often characterized by additional features not present in earlier internal wars (see, e.g., Kaldor, 1999). Especially, the view is taken that (a) ‘globalization’ and new links with international markets have boosted the eruption of internal resource contests, (b) internal wars of the 1990s are often characterized by a multiplicity and fragmentation of combatants, lacking a unitary leadership and organization, and involving significant portions of the population, (c) internal wars of the 1990s often entail increasing impacts on civilians, the displacement of people, and the destruction of production structures. These apparently new features of recent civil wars have led some authors to refer to them as ‘new wars’ (Kaldor, 1999) or ‘post-modern’ conflict (Duffield, 1998). The propositions concerning the emergence and characteristics of ‘new wars’, as formulated by political scientists, are mainly based on case studies and are to a considerable extent lacking theoretical foundation. On the other hand, the economics literature on civil war usually has firm theoretical underpinnings, but largely fails to recognize several aspects of the recent evidence described by political science. This literature (dating back to Grossman, 1995; Hirshleifer, 1987) portrays internal conflict as a struggle over the tax base between the existing government and a well-defined rebel organization that pursues the objective of state capture or secession. In this struggle, natural resources often figure as an element of the tax base (e.g. Collier and Hoeffler, 1998; Olsson, 2003). By setting up a game between two well-defined parties, the economics of rebellion disregards the multiplicity and fragmentation of combatants and the lack of unitary leadership and organization described in political-science literature. In addition, the involvement of and implications for the civilian population and the production sector (‘externalities’) are hardly captured, and the alleged role of ‘globalization’ is not addressed.3 Given its focus on the government and a rival organization being engaged in armed conflict, the economics of rebellion is concerned with ‘top-down’ violence, that is, violence which is mobilized by 2 The importance of institutions, especially ‘due process’ and civil rights, is acknowledged by Collier and Hoeffler (2005), who suggest that institutional considerations produce a ‘filter’ through which civil war can be linked to resource abundance. 3 Another strand of economics literature that should be mentioned deals with the ‘curse of natural resources’, that is, the phenomenon that resource-rich countries tend to show low economic performance (e.g. Sachs and Warner, 1995; Gylfason et al., 1999; Welsch, in press). One possible reason for the ‘curse’ is rent-seeking, perhaps culminating in armed conflict (see, e.g., Gylfason and Zoega, 2002). This literature also disregards the features of the ‘new wars’ mentioned above.

political leaders and entrepreneurs and which may create largescale conflict. The present paper takes another perspective. Assuming a reduced role for integrative leadership and organization and an increased involvement of the population in their role as combatants and civil victims, the paper adopts a ‘bottom-up’ approach to internal conflict according to which violence is actively embraced by ordinary people in a contest for resource rents (see Keen, 2000 for the typology used). The aim of the paper is to capture both, the role of natural resource abundance, and the stylized facts from the descriptive literature (multiplicity of actors, devastation of production structures, role of globalization) in a unified bottom-up framework of internal conflict. To accomplish this purpose, the paper develops and estimates a simple two-sector model (resource sector and production sector). Subject to the degree of property rights enforcement, people are mobile between the two sectors and will engage in predation when resource rents rise (due to ‘globalization’, say). The resulting armed conflict imposes a negative externality on the production sector. The size of the externality varies directly with the share of the combatants in the population. Since the externality reduces the remuneration rate in the production sector (marginal labor productivity), an increase in the share of combatants may be self-energizing. The main prediction from this set-up is that the probability of the number of casualties exceeding a given threshold varies directly with the amount and value of lootable mineral resources and inversely with variables that increase labor productivity, especially agricultural resources and other sorts of production capital. The model's predictions are supported by cross-sectional econometric evidence involving 54 countries, 1989–2002. Major findings are that (a) abundance of mineral assets significantly raises the probability of internal armed conflict – defined as the probability that there are at least 25 casualties – whereas the productivity of ‘normal’ production and the quality of governance reduce the conflict probability, (b) the relevant productivity variables in this relationship are agricultural and human capital, rather than manufactured capital. Importantly, it is not natural resource abundance in general, but the abundance in nonrenewable resources which breeds conflict. Availability of agricultural resources (pastureland, cropland and forest) reduces the incidence of armed conflict by raising labor productivity in ‘normal’ production. In quantitative terms, the negative effect of agricultural resources on conflict probability is almost twice as large as the positive effect of mineral resources. An important conclusion for policy is that not only the quality of governance but also schooling reduces the propensity for armed resource conflict. Both need to have a sufficient level to make predation unattractive as an alternative to production. In relating the model and results to earlier studies, it may be noted that much of the recent literature is framed in terms of a ‘greed vs. grievance’ dichotomy. Whereas the grievance hypothesis (see, e.g., Homer-Dixon, 1995, 1999) regards internal war as originating from poverty and scarcity, the greedbased explanation (Collier, 2000; Collier and Hoeffler, 1998, 2004) emphasizes voracity as a cause of conflict. The present paper reconciles these opposing views by taking an opportunity cost perspective, focusing on the relative rather than the absolute payoffs to be gained from production and resource appropriation. In this view, income that can be gained in the

EC O L O G IC A L E C O N O M IC S 6 7 ( 2 0 08 ) 50 3 –5 13

production sector represents the ‘opportunity costs of conflict’. To trigger a move from the production to the resource sector, the remuneration in the production sector need not be particularly low (‘grievance’) nor need the remuneration in the resource sector be particularly high (‘greed’). Rather, the payoffs from the two types of activity jointly determine the allocation of people. In addition, in the present model grievance of the producers may be a result of conflict, which reinforces the incentive to take resort to predatory activities. Overall, the contribution of the paper is twofold. First, it develops a model which captures existing informal propositions concerning ‘new wars’ in a unified framework, and which offers an alternative perspective on the ‘greed vs. grievance’ controversy. Second, in the light of recent doubt concerning the resources–conflict linkage, it clarifies the role of natural resources, emphasizing the importance of a consistent structural model which informs the choice of the specific variables that are relevant. The paper is organized as follows. Section 2 sets up the theoretical framework, and Section 3 presents the empirical approach and results. Section 4 provides a discussion of findings and offers conclusions.

2.

Theoretical framework

2.1.

General approach

These basic ideas from the economics of crime are retained in the present paper. However, an important additional feature of the present approach is that the violent activity leads to negative spillovers for production. If the violent activity and the ensuing externalities exceed a certain threshold, the equilibrium may collapse, that is, there will be a selfsustaining tendency for production-based remuneration to fall short of redistribution-based remuneration.5 With respect to relevant types of resources, the framework described below is meant to apply mainly to ‘lootable’ resources. Lootable resources are those that can be easily appropriated by individuals or small groups and produce low-level conflict involving a multiplicity of relatively weak combatants (Le Billon, 2001; Ross, 2003). The general modeling strategy can be described as follows. The basic idea is that the more people are active in the appropriation and exploitation of lootable resources, the greater is the probability of armed internal conflict. The model determines the number of people in the resource sector as a function of (a) variables that determine the remuneration rates in the resource and the production sector, (b) parameters that measure the quality of governance and property rights protection, and (c) the population size. The model thus yields predictions as to how (a)–(c) affect the probability of internal conflict. These predictions are then used as the basis for the specification of a binary probit regression of the occurrence of internal conflict.

2.2. As stated in the introduction, the model to be developed here is designed especially to capture some salient features of internal armed conflict of the 1990s, in particular the multiplicity and fragmentation of combatants and the increasing involvement of the civilian population, resulting in negative spillovers for production. Given this motivation, the current approach differs from models of rebellion in the tradition of Grossman (1995) or Hirshleifer (1987), which focus on the government and a rebel organization as explicit actors and in which the population is only implicitly present insofar as the two combatting parties need to mobilize and organize their support. The current model, conversely, focuses on a representative person from the population and her/his decision to participate in violent redistributive as opposed to non-violent productive activities. The role of organizations and leaders is deliberately neglected.4 Also neglected is any effort or expenditure on securing rents (e.g. stockpiling of arms), which is an important ingredient of the conflict-as-rebellion literature (see Neary (1997) for a conceptual discussion of versions of this approach). Rather, the focus is on multiple groups or individuals with (quasi) open access to the resource sector (similar as in Torvik, 2002). The current approach shares some important features with the economics of crime in the tradition of Becker (1968), in that the individual's decision as to her/his type of activity is based on the respective remuneration rates. In this literature, the resulting level of crime represents an equilibrium in which the (expected) payoffs from the two types of activity are equalized.

505

Basic model structure

We assume that the remuneration rate to be earned in the resource sector, wR, equals the total resource rent, divided by the number of relevant persons: wR ¼

pR ; LR

where π = unit profit (unit resource rent), R = resource quantity, LR = number of persons in resource sector. The remuneration rate in the production sector, wY, is the marginal product of labor: wY ¼

∂FðLY ; :::Þ ; ∂LY

where LY is the number of persons active in the production sector, and production is described by a production function Y = F(LY,…) which is strictly increasing and strictly concave in LY and satisfies the Inada conditions. As will be described in the next subsection, this basic relationship may be extended to account for the effects of armed conflict on production. Fig. 1a illustrates the basic production function neglecting the effect of conflict. The persons in the resource sector and the production sector sum up to the exogenous population (capable of work) L: LR þ LY ¼ L: Next, we assume quasi-mobility between the two sectors. This means that whenever wR exceeds wY by some factor λ ≥ 1,

4

This indicates a shift in perspective and emphasis. It is not meant to imply that aspects of organization and leadership play no role.

5 This potential for a collapse of equilibrium is akin to similar features of some rent-seeking models (e.g. Murphy et al., 1993).

506

EC O LO GIC A L E CO N O M ICS 6 7 ( 2 00 8 ) 5 0 3 –5 13

Eq. (1) would say that the net remuneration in the two sectors is equalized in equilibrium. With respect to (b), it should be noted that wR = πR / LR represents the expected remuneration rate in the resource sector, which involves the risk of being killed.8 From this point of view, in addition to property rights, λ captures aversion towards the risk of being killed in armed conflict.

2.3.

The complete model

We now augment the basic model structure to account for those factors that affect the remuneration rate in the production sector, that is, the marginal product of labor. The marginal product of labor is positively affected by various sorts of capital, such as agricultural capital (land), manufactured capital, human capital, and social capital (especially the quality of governance). These factors will be referred to as generalized production capital and collectively denoted by A. On the other hand the marginal product of labor may be negatively affected by the repercussions from conflict. These repercussions will be captured as a negative externality, and it is assumed that this externality depends on the ratio lR = LR / L. This assumption reflects the idea that a conflict of a given level – proxied by LR – has a smaller impact in a larger country. The complete production model can thus be written in the following way:
 Y ¼ F LY ; lR; A

Fig. 1.

a move from the production sector to the resource sector will take place (and the reverse happens when wR falls below λ · w Y). The equilibrium condition which determines the allocation of people to the two sectors thus reads: kwY ¼ wR :

ð1Þ

The equilibrium is illustrated in Fig. 1b (where F′: = ∂F / ∂LY). The parameter λ reflects two phenomena: (a) the strictness of property rights protection that inhibits entry into the resource sector and/or (b) aversion towards the specific risks that may be associated with activity in the resource sector. By referring to (a), one way of justifying Eq. (1) is by interpreting wR / λ as the remuneration in the resource sector net of the cost of property rights protection.6 These costs may be higher in the resource sector than in the production sector since natural resources tend to be more prone to unlawful appropriation than are produced commodities.7 In this view, 6 Snyder (2003) argues that secure property rights are of key importance for avoiding armed conflict especially in the case of ‘lootable’ resources. Hotte (2001) takes the view that property rights protection may be particularly weak in areas remote from the ‘center’. 7 See Torvik (2002). This holds especially with respect to mineral resources, see De Soysa (2000). For a discussion with respect to diamonds, see Olsson (2003).

ð2Þ

∂F N 0; ∂LY

∂F b 0; ∂lR

∂F N0 ∂A

∂2 F b 0; ∂L2Y

∂2 F b 0; ∂LY ∂lR

∂2 F N0 ∂LY ∂A

ð2aÞ

ð2bÞ

Eq. (2a) describes the role of labor, capital and good governance (social capital) as production factors, and the negative impact of armed conflict on production. Eq. (2b) states that the marginal product of labor decreases in LY (diminishing marginal product) and lR and rises in A. We focus on the implications of the conflict-related externalities. Observing lR = 1 − LY / L, the total first and second derivatives of output with respect to LY are: F0 :¼

dF ∂F 1 ∂F ¼  dLY ∂LY L ∂lR

ð3Þ

FW :¼

d2 F ∂2 F 1 ∂2 F ¼  : dL2Y ∂L2Y L ∂lR ∂LY

ð4Þ

Since ∂2F / (∂lR∂LY) b 0 (due to Eq. (2b)), the total second derivative F″ may become positive in some part of the domain. This means that a smaller amount of people employed in production need no longer imply a larger marginal product, as these people enter the conflict-prone resource sector with its potential for damaging repercussions for production. 8 The survival probability p cancels in the numerator and denominator: E(wR) = p · (πR / pLR).

507

EC O L O G IC A L E C O N O M IC S 6 7 ( 2 0 08 ) 50 3 –5 13

production capital (A) and the quality of property rights protection (as included in λ). It is positively related to the population size (L) except possibly when small countries are concerned. Most of these predictions are rather intuitive in terms of the basic logic of the model. However, it may be worth discussing two elements of the prediction in more detail. One result to be discussed concerns population size. In the present framework, the possibility of a positive effect of the population size on conflict probability arises in a trivial way because – other things equal – larger countries have a larger absolute number of people potentially active in the resource sector.9 On the other hand, however, larger countries have a higher potential output of the production sector, whereas the resource rent is constant with respect to the population size. This asymmetry creates a negative effect of the population size on the number of people actually active in the resource sector. On balance, however, the positive effect will dominate the negative effect in large countries because, due to the diminishing marginal product of labor, potential output grows less than population, thus weakening the described negative effect.10 Another result is that conflict probability is positively related to the unit value of resources. In the introduction it was stated that ‘globalization’ is sometimes mentioned as a driver of armed conflict. As Appendix B shows, these two aspects may be related: If globalization is taken to imply that transportation costs, tariffs and other factors which drive a wedge between producer prices and consumer prices decrease, it may lead to a rise in the producer price of resources even if the final willingness-to-pay function of consumers does not change. In this sense, globalization may in fact raise the incentive for activity in the resource sector and, hence, the probability of conflict. Fig. 2.

2.5. A possible shape of the production function is illustrated in Fig. 2a, while the (total) marginal product (F′) is shown in Fig. 2b. It can be seen that (disregarding λ) multiple equilibria may arise. While equilibrium B is stable, equilibrium A is not. For sufficiently high values of πR, no equilibrium may exist at all. The same reasoning holds if we consider λF′ instead of F′. It is clear that the chance that equilibria do exist is larger at larger values of λ. Moreover, the level of LY (LR) in B-type equilibrium increases (decreases) in λ. Having completed the model description, we are now interested in what factors – in addition to λ – determine the allocation of people and the implications for conflict.

2.4.

Theoretical results

In this subsection we consider the level of LR in B-type equilibrium, since LR is hypothesized to determine the probability of conflict. The possibility that the equilibrium collapses will be addressed in the next subsection. As Appendix A shows, the model implies the following statement concerning conflict probability:

To illustrate the model and results we consider a constantelasticity production function with production elasticity of labor γ b 1:
 Y ¼ FðLY ; lR ; AÞ ¼ LgY  1  l2R  Kgg  u;

Prediction

The probability of internal armed conflict – assumed to be proportional to LR – is positively related to the value (π) and amount (R) of resources and negatively related to generalized

ð5Þ

where generalized capital, A, is split into K, which comprises agricultural, manufactured, and human capital, and ϕ, which represents the role of governance. η is the scale elasticity with respect to LY and K. The equilibrium condition λ∂F / ∂LY = πR / LR thus reads:
 pR k  g  Lg1  1  l2R  Kgg  u ¼ : Y LR

ð6Þ

By observing LY = (1 − lR) · L, it can be transformed into the following condition (see Appendix C) k  gðlR ; gÞ :¼ k  g  lR  ð1  lR Þg ð1 þ lR Þ ¼

pR : uKgg Lg

ð7Þ

Consider the left-hand side. The function g(lR, γ) takes the value zero at lR = 0 and at lR = 1 and positive values for lR in 9

2.4.1.

An example

In the economics of rebellion literature the size of the population is usually meant to capture a desire for secession (see Collier and Hoeffler, 1998). 10 More discussion of potential output is provided with respect to the example presented in the next subsection.

508

EC O LO GIC A L E CO N O M ICS 6 7 ( 2 00 8 ) 5 0 3 –5 13

between these boundaries. The function is thus hump-shaped, with an interior maximum. The same properties apply to λg(·). There may thus be two equilibria (one unstable, one stable) or none at all, depending on how large the maximum of λf(·) is, relative to the expression on the right-hand side. The right-hand side gives the ratio between the resource rent and the capacity output of the economy, that is, the output that could be produced if all people were in the production sector (LY =L). The existence of equilibrium thus depends on whether the maximum of λg(·) is larger than the resource rent divided by capacity output (henceforth referred to as relative resource rent). It should also be noted that under constant returns to scale (η = 1) the right-hand side can be written as (π /ϕ(K /L)1 − γ) · (R /L). Thus, at given levels of capital per person (K /L) it is the ratio of the resource quantity and the population size (R /L) which determines the allocation of people to the two sectors. Eq. (7) can be used for an illustrative numerical computation of the conditions under which the equilibrium may collapse. Table 1 gives the maximum of g(lR, γ) and the lR's where the maximum occurs, for an array of production elasticities of labor, γ. It can be seen that in the case of complete mobility (λ = 1) a resource rent exceeding some 30 to 40% of capacity output would lead to a collapse of equilibrium. Of course, effective property rights protection can prevent such a collapse.

Table 1 – Maximum relative resource rent for which equilibrium exists γ

0.4

0.5

0.6

0.7

0.8

0.9

Max g(lR, γ) Arg max g(lR, γ)

0.3023 0.7825

0.3283 0.7403

0.3467 0.7018

0.3605 0.6667

0.3709 0.6344

0.3788 0.6047

Note: Maximum values refer to λ = 1.

production sector. Other determinants of labor productivity are human capital and manufactured capital per capita. Whereas the endowment with subsoil assets is predicted to raise the probability of armed conflict, the various productivity variables reduce this probability. In addition, as stated in the Prediction, population size reduces/raises the probability of conflict at low/ high population levels, and will be included accordingly. A final variable which affects conflict probability is the quality of governance. This variable plays a twin role. On the one hand, to the extent that it includes the enforcement of property rights in the resource sector, it reduces the probability of conflict in a direct way. On the other hand, by enhancing labor productivity in the production sector, it reduces conflict probability indirectly.12 Via both channels, governance reduces the probability of armed conflict. A stylized representation of these considerations can be written as follows: !

3.

Evidence

probðconflictÞ ¼ f mincap; humcap; manucap; agricap; pop; gov ;

3.1.

Empirical approach and data

where mincap = subsoil assets (mineral resources etc.), humcap = human capital, manucap = manufactured capital per person, agricap = agricultural capital per person (pastureland, cropland, forest), pop = population, gov = governance (civil rights and liberties). The signs below the explanatory variables indicate how these variables are expected to affect the probability of armed internal conflict. In this formulation, conflict is a binary variable, constructed from the Uppsala Conflict Data Program. It refers to ‘conflict type 3: internal armed conflict with at least 25 casualties per year’(see UCDP, 2004)13. The variable takes the value 1 if at least one pertinent conflict has occurred in a given country within the period 1989–2002, otherwise it takes the value zero. The conflict data employed use a relatively low threshold of deaths, which permits to capture conflicts whose level is below that of large-scale ‘organized’ civil wars. This, together with the time period chosen, fits the nature and time of occurrence of ‘new

In order to test the predictions of the model, we formulate an empirical analogue to the above Prediction. We first observe that in a cross-country perspective, flows of resources can be taken to be proportional to natural resource endowments. The quantity of natural resources will thus be proxied by the stock of natural capital, but the latter will be differentiated into subsoil assets (mineral resources etc.) on the one hand and agricultural capital (pastureland, cropland, and forest) on the other.11 This differentiation is motivated by the idea that the contest for resource rents refers mainly to assets like diamonds, gold, bauxite, or oil (see Olsson, 2003; Klare, 2001; Fearon, 2005). On the other hand, agricultural capital as defined above is a generic input to agricultural production and thus belongs in the production sector as conceptualized in Section 2, not in the resource sector (see De Soysa, 2000 for a differentiation between the two sorts of natural resources). Following this idea, agricultural capital per capita will be used as one of the determinants of labor productivity in the 11 In the context of the ‘natural resource curse’ Gylfason and Zoega (2002) argue that stocks of natural capital provide a better measure of natural resource abundance across countries than the various proxies that have been used in earlier studies, mainly the share of primary (i.e., non-manufacturing) exports in total exports or in gross domestic product (GDP) and the share of the primary sector in employment of the labor force. Using the share of primary exports in econometric studies of internal conflict has been criticized for implying the risk of endogeneity bias, see Ross (2004). In addition, measures that rely on the ‘primary sector’ provide no distinction between mineral and agricultural commodities, see De Soysa (2000).

þ







=þ



12 Governance or ‘social infrastructure’ has been found in several studies to be an important determinant of cross-national differences in labor productivity, see, e.g. Hall and Jones (1999). 13 Conflict is not a self-evident term. Following the Uppsala Conflict Data Program, “An armed conflict is a contested incompatibility that concerns government and/or territory where the use of armed force between two parties, of which at least one is the government of a state, results in at least 25 battle-related deaths in one calendar year.” “Incompatibility” refers to “stated… generally incompatible positions” over Government or Territory. Internal conflict in the sense of this paper is “A conflict between a government and a non-governmental party, with no interference from other countries.” A non-governmental party in this sense is a “group of people having announced a name for their group and using armed force.” For these definitions, see www.pcr.uu.se/ database/definitions_all.htm.

−0.622 (− 2.26) −0.697 (−2.29) −0.476 (−2.11) − 0.553 (−1.98)

−0.672 (−2.83)

0.027 (0.13) 0.017 (0.09)

−1.098 (− 2.53) −0.273 (− 1.04) −1.173 (−2.64) −0.163 (−0.69) −0.448 (−2.12)

Note: Dependent variable is conflict (incidence of internal armed conflict with at least 25 casualties, 1989–2002). Method: Binary Probit. Figures in parentheses are Huber–White corrected t-statistics. The estimations also include constant terms.

(−2.48) (−0.29) (0.15) (−2.23) − 1.090 − 0.547 0.014 − 0.618 (−2.55) (−0.63) (0.57) (−2.24) − 1.147 − 1.110 0.048 − 0.690 − 1.042 (− 0.54) 0.054 (0.56) − 0.465 (− 2.03) − 0.762 (− 0.41) 0.039 (− 0.43) − 0.661 (− 2.75)

− 0.432 (− 2.03)

0.564 (2.75) − 1.474 (−1.92) 0.408 (2.27) 0.573 (2.78) −1.491 (− 1.97)

ln(mincap) ln(humcap) ln(manucap) ln(agricap) ln(pop) ln(pop)-sqrd. gov-sqrd.

0.610 − 0.926 − 0.108 − 1.171 − 0.296

(2.72) (−0.84) (−0.35) (−2.52) (−1.10)

0.258 (2.22) −1.637 (−2.38)

0.291 (2.16)

0.430 (2.38)

0.596 −0.904 −0.102 −1.161 −0.745 0.023 −0.549

(2.68) (− 0.81) (− 0.33) (− 2.47) (− 0.39) (− 0.25) (− 1.92)

0.246 (2.17) − 1.596 (− 2.32)

0.273 (2.07)

(J) (E) (D) (C) (B) (A)

Table 2 – Estimation results with mineral resources and population as separate variables

(F)

(G)

(H)

(I)

EC O L O G IC A L E C O N O M IC S 6 7 ( 2 0 08 ) 50 3 –5 13

509

wars’ discussed in the descriptive literature and modeled above. The data base comprises 54 countries, namely those for which all required data are jointly available. Of these, 18 experienced at least one internal conflict (see Appendix D).14 The other data used refer to the late 1980s and early 1990s15, depending on availability. Their definitions and sources are as follows. • mincap is the stock of subsoil assets (thousand $), taken from World Bank (1997). This measure is chosen because mineral assets (oil, bauxite, gold, diamonds) are easily captured and therefore have been implicated for playing a role in recent civil conflicts.16 • humcap is the adult literacy rate (percent), taken from the World Development Report. This choice is informed by the literature on economic growth which suggests, especially, that human capital proxies should refer to stocks (not to flows, such as school enrolment rates).17 • manucap is physical capital per person ($ per capita), taken from the database of Hall and Jones (1999).18 • agricap is the stock of pastureland, cropland and forest per person ($ per capita), taken from World Bank (1997). • pop is the population size (thousand persons) taken from World Bank (2003). • gov is a democracy index, measuring to what degree civil rights and liberties are held in respect. Respect for rights is assessed on the basis of expert judgements. It is measured on a scale ranging from 1 (low levels of civil rights and liberties) to 7 (high levels). The variable gov is taken from Freedom House (2000).19 The model has been estimated using the Maximum Likelihood Binary Probit method.

3.2.

Results

Table 2 shows the estimation results for several versions of the model introduced above. Except for governance, the

14 Among these are Spain and the U.K. Given the definitions quoted in Footnote 13, the conflicts between ETA and IRA, respectively, and the respective governments qualify as internal armed conflicts in the sense of this paper. Though these conflicts are unlikely to be resource-driven, I do not deem it methodologically appropriate to delete these countries from the sample. 15 Endogeneity should thus not be a matter of concern. 16 The large scale stealing of oil from pipelines in the Niger Delta is a pertinent example. It is commonly propounded that civil wars in Angola, Liberia, the Democratic Republic of Congo, and Sierra Leone are linked to the struggle for control of oil, diamond mines and other resources (Reno, 1998). It is acknowledged that timber may be subject to resource contests, see Klare (2001). 17 See Barro and Sala-i-Martin (1995). Adult literacy rates have been found theoretically attractive and empirically significant in the growth regressions of Barro (1991). 18 See http://emlab.berkeley.edu/users/chad/HallJones400.asc. 19 The importance of civil rights and ‘due process’ for the occurrence of conflict is emphasized by Collier and Hoeffler (2005). Using other indicators of democracy (such as Polity IV of the University of Maryland, see www.cidcm.umd.edu/inscr/polity) did not appreciably affect our results.

510

EC O LO GIC A L E CO N O M ICS 6 7 ( 2 00 8 ) 5 0 3 –5 13

explanatory variables are included in logarithmic form. The governance variable is included in squared form. This is inspired by findings in earlier literature (see De Soysa, 2000) that the conflict-inhibiting influence of governance is convex. In terms of the discussion of the preceding subsection, this phenomenon could be related to the twin role of governance (direct and indirect) for conflict probability.20 The sets of versions (A)–(E) and (F)–(J) differ from each other in terms of how the size of the population is treated. Versions (A)–(E) disregard the possibility that the effect of the population size on conflict probability may turn positive when the population is large, whereas the otherwise identical versions (F)–(J) test for this possibility by including the square of the log of population as an additional variable. Versions (A) and (F) include all of the productivity variables jointly, whereas versions (B)–(E) and (G)–(J) focus on subsets of the productivity variables. Considering (A)–(E), we first note that, whatever version from this set is considered, almost all signs are as expected: Mineral resource abundance enhances conflict probability whereas labor productivity (capital per person) and governance reduce conflict probability. Though the signs of the capital variables are as expected, human capital and manufacturing capital are insignificant in version (A), whereas agricultural capital is significant. However, when included separately, as in versions (B)–(D), all capital variables are significant at conventional levels. It may thus be conjectured that the problems with version (A) are caused by multicollinearity among the productivity variables. Nevertheless, human capital and agricultural capital are both significant when included jointly while omitting only manufactured capital (see version (E)). Manufactured capital is not significant when combined with one of the other types of production capital (results not shown). In addition to these findings concerning the productivity variables, key results from these estimations are that the impact of subsoil resource abundance is significantly positive whereas governance has a significantly negative impact on conflict probability, irrespective of which particular specification is considered.21 With respect to population size, we find a negative influence in versions (A), (D) and (E), as predicted by the theoretical model in case the population is not too large. However, the coefficient is invariantly insignificant in (A)–(E). Insignificance could arise because the size of the population affects conflict probability differently at different population size. This possibility is tested in versions (F)–(J). It can be seen that (the log of) population always has a negative sign whereas the square of this variable has a positive coefficient. This is in line with what the model predicts, but all of the population variables in versions (F)–(J) are insignificant.22 With respect to

20 Experimentation with including second and first powers jointly showed that the first power was unanimously insignificant. 21 It has been suggested in the literature that countries rich in natural resources tend to have weak governance (Fearon, 2005). This would establish an additional, indirect channel through which resource abundance affects conflict probability. 22 Including the population in other ways (levels and levelssquared instead of logs and logs-squared) also implies insignificant coefficients for the population variables.

all other variables, the results from (F)–(J) are almost identical to their counterparts (A)–(E), indicating that these results are rather robust. Considering version (J), it is remarkable that the coefficients on mincap and pop have a rather similar magnitude (though a different sign). This seems to suggest that it is the ratio of the two variables (that is, subsoil assets per person) which determines conflict probability. In the light of Eq. (6) this conforms to the case that production exhibits constant returns to scale with respect to labor and the relevant sorts of capital. The results from including mineral resources per person are reported in Table 3, which has the same structure as Table 2: While versions (A)–(E) disregard the possibility that the effect of the population size on conflict probability may turn positive when the population is large, versions (F)–(J) test for this possibility by including the square of the log of population as an additional variable. It can be seen that major results from Table 2 concerning signs and significance are preserved. Especially, manufactured capital continues to be (weakly) significant only when other productivity variables are absent. The specifications from Tables 2 and 3 thus agree in that human capital and agricultural capital are the important productivity variables; manufactured capital acts as a proxy if they are omitted. With respect to the new variable, log(mincap / pop), we find it significant throughout (or at least weakly significant, as in version (C)). It is thus a robust finding that subsoil natural capital, be it measured in absolute terms (Table 2) or in per capita terms (Table 3), is an important predictor of the probability of internal armed conflict in the period considered. As concerns the role of the population size in addition to the negative effect captured by log(mincap / pop), versions (F)–(J) have positive coefficients of the square of the log of population, and these are significant at 7.1% (version (I)) or better. In version (J) of Table 3, which seems to be the most robust of the specifications examined, the marginal significance level is 5.3%. What is predicted by the model and emerged as a tendency in Table 2 is confirmed by the estimates in Table 3: The probability of internal armed conflict – defined as the probability that there are at least 25 casualties – is related to population size in a U-shaped fashion. Turning to quantitative aspects, interpretation of the coefficient values is not straightforward. Especially, estimated coefficients from a binary model cannot simply be interpreted as the marginal effect on the dependent variable. It is known, however, that ratios of coefficients provide a measure of the relative changes in the probabilities (see, e.g., Greene, 2003). In terms of the preferred specification (J) in Table 3, this means that a one-percent increase in agricultural resources per capita reduces the probability of armed conflict almost twice as much as a one-percent increase in mineral resources per capita raises it. Similarly, a onepercent increase in adult literacy reduces conflict probability 2.6 times as much as an increase in mineral resources per capita increases it. The abundance of mineral resources is thus a factor which significantly raises conflict probability, but it is far less important than a lack in agricultural and human resources.

− 0.617 (− 2.27) − 0.687 (−2.17) −0.672 (−2.83) −0.607 (−2.11)

− 0.490 (−1.98)

− 1.028 (− 2.22) − 1.136 (−2.41) − 0.393 (−1.95)

Note: Dependent variable is conflict (incidence of internal armed conflict with at least 25 casualties, 1989–2002). Method: Binary Probit. Figures in parentheses are Huber–White corrected t-statistics. The estimations also include constant terms.

− 1.090 (−2.49) 0.015 (1.93) − 0.618 (−2.22) − 1.166 (−2.61) 0.013 (1.81) − 0.695 (−2.28) 0.014 (1.83) −0.668 (−2.80)

0.016 (2.11) −0.473 (−2.09)

−0.444 (−2.09)

0.564 (2.72) − 1.473 (−1.94) 0.425 (2.33) 0.286 (2.12) 0.254 (2.18) −1.623 (−2.35) 0.495 (2.38) − 1.297 (− 1.80)

ln(mincap / pop) ln(humcap) ln(manucap) ln(agricap) ln(pop)-sqrd. gov-sqrd.

0.522 −1.099 0.031 −1.128

0.235 (1.87) 0.220 (1.95) −1.564 (−2.25)

(2.40) (−0.98) (0.10) (−2.28)

0.391 (2.10)

0.601 −0.910 −0.104 −1.164 0.016 −0.550

(2.67) (− 0.82) (− 0.34) (− 2.48) (2.02) (− 1.94)

(H) (E) (D) (C) (B) (A)

Table 3 – Estimation results with mineral resources per capita

(F)

(G)

(I)

(J)

EC O L O G IC A L E C O N O M IC S 6 7 ( 2 0 08 ) 50 3 –5 13

4.

511

Discussion and conclusions

The model and results presented above have clarified the role of natural resource abundance for the incidence of internal conflict in the following sense: • There is not a general linkage between natural resources and internal conflict, but a differentiated pattern. While mineral resource endowments raise the risk of internal conflict, agricultural resources diminish this risk. In absolute terms, the impact of agricultural resources is almost twice as large as the impact of mineral resources. • Being based on the relatively low threshold of 25 battledeaths and the period 1989–2002, the evidence presented is consistent with the view that natural resource abundance is important especially for small and intermediate internal conflict (such as rioting and warlordism). The results thus provide some structure to the disparity of empirical findings that characterizes earlier literature (Ross, 2004). In addition to the empirical results, the theoretical framework has proven useful in the following respects: (a) It has incorporated stylized features of ‘new wars’ discussed in the descriptive literature. (b) It has leveled the dichotomy between grievance-based and greed-based explanations of conflict by taking an opportunity cost perspective. (c) It has informed the choice of the variables that are relevant for empirical assessment, and their expected role. With respect to (a), the model has captured the multiplicity of combatants and the externalities inflicted on production, emphasized in the descriptive literature (Duffield, 1998; Kaldor, 1999). Specific features of conflict-as-rebellion (Grossman, 1995; Hirshleifer, 1987) have been neglected deliberately. With respect to (b), the model has treated the payoffs to be gained from resource appropriation and from production in a symmetric way, thus avoiding to give ‘grievance’ (HomerDixon, 1995, 1999) or ‘greed’ (Collier and Hoeffler, 1998, 2004) any priority over the other. Moreover, grievance has to some extent been endogenized via the externality from conflict on production. With respect to (c), the model has provided a theoretical basis for including variables which determine labor productivity, rather than including income (which is an outcome itself). Especially, the distinction between mineral resources and agricultural resources, which has been introduced ad hoc in previous literature (De Soysa, 2000), conforms to this framework: whereas mineral resources represent lootable assets, agricultural resources are a determinant of productivity in normal production. Using ‘primary commodities’ as a proxy for resource abundance fails to capture this distinction. Moreover, the framework has helped to clarify the role of the population size for the incidence of internal conflict (other than the inclination towards secession), and the twin role played by the quality of governance.

512

EC O LO GIC A L E CO N O M ICS 6 7 ( 2 00 8 ) 5 0 3 –5 13

In terms of policy conclusions, the results obtained emphasize the vital role not only of the quality of governance (especially for productivity), but also of schooling, in preventing those kinds of conflict studied here. Both factors need to have a sufficient level to make looting unattractive as an alternative to production. Finally, to put the paper in perspective, it should be noted that the framework proposed in this paper does not apply to all kinds of resource-related conflict. Several of these, like ‘water wars’ or future climate change related conflicts, originate not from the opportunities implied by resource abundance, but – on the contrary – from resource scarcity. In addition, the pertinent resources in those cases typically are not the mineral resources addressed in this paper, but resources that are essential to satisfying basic needs. Their scarcity may trigger armed conflict in much different ways than studied here. Preventing such conflicts is one of the most forceful arguments for intra-generatively and inter-generatively fair (sustainable) resource management.

Acknowledgements I appreciate helpful comments by Udo Ebert and two anonymous referees.

Fig. 3.

larger of the two otherwise equal countries may have a smaller number of LR. The intuition behind this result is discussed in the main text. Eq. (A4) establishes the prediction in the text.

Appendix B. Role of globalization Appendix A. Comparative statics in B-type equilibrium For the subsequent analysis we introduce the following notation for the partial derivative of F with respect to LY: ∂F / ∂LY =f(LY, lR, A). The second partial derivatives satisfy Eq. (2b), that is fL:=∂f/ ∂LY b 0, fl:= ∂f /∂lR b 0, fA:=∂f/ ∂A N 0. The equilibrium condition (1) can be stated as k  f ðLY ; lR ; AÞ ¼

pR : LR

ðA1Þ

  LR pR : k  f L  LR ; ; A ¼ LR L

ðA2Þ

Totally differentiating both sides and rearranging yields    pR fl R p  f þ k dLR ¼ dp þ dR  f dk  kf A dA L LR  LR  L L2R LR þ k fl 2  fL dL: L

ðA3Þ

With respect to the left-hand side it should be noted that fl / L − fL is the negative of F″ in Eq. (4). It is thus positive whenever F″ b 0. This is likely to be the case in B-type equilibrium (see Fig. 2b). Based on this reasoning we assume that the term in square brackets on the left-hand side of Eq. (A3) is positive. We then obtain: dLR N 0; dR

pup þ t ¼ c  bR; the inverse demand function facing the supplier is p ¼ ðc  tÞ  bR:

In order to determine the comparative statics of LR we write Eq. (A1) as

dLR N 0; dp

For simplicity, we neglect costs of resource extraction. Then the supplier price of the resource equals the unit profit π. The consumer price p is obtained by adding unit transportation costs, tariffs, taxes and the like, jointly denoted by t. Assuming, for illustrative purposes, a linear inverse (final) demand function,

dLR b 0; dk

dLR b 0; dA

dLR N = b 0: dL

ðA4Þ

Note that with respect to the population size L we likely have a positive impact on LR. It may be negative only in small countries. That is, when we compare two small countries, the

Fig. 3 illustrates the market outcome in case the supplier is a monopolist (where MR = marginal revenue). It is clear that both the supplier price π⁎ and the resource quantity R⁎ rise when t decreases. Similar reasoning holds in the case of Cournot oligopoly or monopolistic competition.23 If such a decrease in t reflects a reduction in transportation costs and/or tariffs, it may be interpreted as a feature of the process of globalization. In this sense, globalization raises the resource rent to be gained and provides an incentive for entry into the resource sector.

Appendix C. Derivation of Eq. (7) Eq. (6) can be rewritten as follows:
 pR ;f k  g  ½ð1  lR ÞLg1  1  l2R  Kgg  u ¼ lR L
 pR k  g  lR  ð1  lR Þg1  1  l2R ¼ : /Kgg Lg 23 There is thus no contradiction with the multiple-actors assumption.

EC O L O G IC A L E C O N O M IC S 6 7 ( 2 0 08 ) 50 3 –5 13

Observing that the left-hand side equals k  g  lR  ð1  lR Þg1 ð1  lR Þ  ð1 þ lR Þ ¼ k  g  lR  ð1  lR Þg ð1 þ lR Þ; this establishes Eq. (7).

Appendix D. Country list Argentina, Australia, Austria, Azerbaijan, Belgium, Benin, Bolivia, Botswana, Brazil, Cameroon, Canada, Chile, Colombia⁎, Denmark, Dominican Republic, Ecuador, Egypt⁎, Finland, France, Germany, Ghana, Guatemala⁎, Honduras, India⁎, Indonesia⁎, Ireland, Italy, Jamaica, Malaysia, Mexico⁎, Morocco, Netherlands, New Zealand, Norway, Pakistan⁎, Papua New Guinea⁎, Peru⁎, Philippines⁎, Saudi Arabia, Senegal⁎, South Africa⁎, South Korea, Spain⁎, Sweden, Thailand, Togo⁎, Trinidad and Tobago⁎, Tunisia, Turkey⁎, United Kingdom⁎, United States, Venezuela⁎, Zambia, Zimbabwe. ⁎One or more internal armed conflicts with at least 25 casualties per year, 1989–2002 (see UCDP, 2004).

REFERENCES Barro, R., 1991. Economic growth in a cross-section of countries. Quarterly Journal of Economics 106, 407–444. Barro, R., Sala-i-Martin, X., 1995. Economic Growth. McGraw Hill, New York. Becker, G.S., 1968. Crime and punishment: an economic approach. Journal of Political Economy 76, 169–217. Collier, P., 2000. Doing well out of war. In: Berdal, M., Malone, D. (Eds.), Economic Agendas in Civil Wars, Boulder, CO. Collier, P., Hoeffler, A., 1998. On economic causes of civil war. Oxford Economic Papers 50, 563–573. Collier, P., Hoeffler, A., 2004. Greed and grievance in civil war. Oxford Economic Papers 56, 563–595. Collier, P., Hoeffler, A., 2005. Resource rents, governance, and conflict. Journal of Conflict Resolution 49, 625–633. De Soysa, I., 2000. The resource curse: are civil wars driven by rapacity or paucity? In: Berdal, M., Malone, D. (Eds.), Economic Agendas in Civil Wars, Boulder, CO. Duffield, M., 1998. Post-modern conflict: warlords, post-adjustment states and private protection. Civil Wars 1, 65–102. Fearon, J., 2005. Primary commodity exports and civil war. Journal of Conflict Resolution 49, 483–507. Freedom House, 2000. Freedom in the World 1999–2000. Freedom House, New York. Greene, W.H., 2003. Econometric Analysis, Fifth Edition. Pearson Education, Inc., Upper Saddle River, NJ. Grossman, H.I., 1995. Insurrections. In: Hartley, K., Sandler, T. (Eds.), Handbook of Defence Economics, vol. 1. Elsevier, Amsterdam. Gylfason, T., Zoega, G. (2002), Natural Resources and Economic Growth: The Role of Investment, Central Bank of Chile Working Paper No. 142. Gylfason, T., Herbertsson, T.T., Zoega, G., 1999. A mixed blessing: natural resources and economic growth. Macroeconomic Dynamics 3, 204–225.

513

Hall, R., Jones, C., 1999. Why do some countries produce so much more output per worker than others? Quarterly Journal of Economics 114, 83–116. Hirshleifer, J., 1987. Conflict and settlement. In: Eatwell, J., Milgate, M., Newman, P. (Eds.), New Palgrave, A Dictionary of Economics. Macmillan Press, London. Homer-Dixon, T., 1995. The ingenuity trap: can poor countries adapt to resource scarcity? Population and Development Review 21, 587–612. Homer-Dixon, T., 1999. Environment, Scarcity, and Violence. Princeton University Press, Princeton. Hotte, L., 2001. Conflicts over property rights and natural resource exploitation at the frontier. Journal of Development Economics 66, 1–21. Humphreys, M., 2005. Natural resources, conflict, and conflict resolution: uncovering the mechanisms. Journal of Conflict Resolution 49, 508–537. Kaldor, M., 1999. New and Old Wars: Organized Violence in a Global Era. Stanford University Press, Stanford. Keen, D., 2000. Incentives and disincentives for violence. In: Berdal, M., Malone, D. (Eds.), Economic Agendas in Civil Wars, Boulder, CO. Klare, M.T., 2001. Resource Wars. Metropolitan Books, New York. Le Billon, P., 2001. The Political Ecology of War: Natural Resources and Armed Conflicts. Political Geography 20, 561–584. Lujala, P., Gleditsch, N.P., Gilmore, E., 2005. A diamond curse? Civil war and a lootable resource. Journal of Conflict Resolution 49, 538–562. Murphy, K.M., Shleifer, A., Vishny, R.W., 1993. Why is rent-seeking so costly to growth? American Economic Review Papers and Proceedings 83, 409–414. Neary, H.M., 1997. A comparison of rent-seeking and economic models of conflict. Public Choice 93, 373–388. Olsson, O. (2003), Conflict Diamonds, Mimeo, Department of Economics, Göteborg University. Reno, W., 1998. Warlord Politics and African States. Lynne Riener Publishers, Boulder. Ross, M., 2003. Oil, drugs and diamonds: the varying role of natural resources in civil war. In: Ballentine, K., Sherman, J. (Eds.), The Political Economy of Armed Conflict: Beyond Greed and Grievance. Lynne Rienner Publishers, Boulder, pp. 47–70. Ross, M., 2004. What do we know about natural resources and civil war? Journal of Peace Research 41, 337–356. Sachs, J.D., Warner, A.M., 1995. Natural Resource Abundance and Economic Growth. NBER Working Paper, vol. 5398. National Bureau of Economic Research, Cambridge. Snyder, R. 2003. Does Lootable Wealth Breed Disorder? A Political Economy of Extraction Framework, Mimeo, Department of Political Science, University of Illinois. UCDP, 2004. Uppsala Conflict Data Program. www.prio.no/cwp/ ArmedConflict. Torvik, R., 2002. Natural resources, rent-seeking, and welfare. Journal of Development Economics 67, 455–470. Welsch, H. in press. Resource Dependence, Knowledge Creation, and Growth: Revisiting the Natural Resource Curse, Journal of Economic Development. World Bank, 1997. Expanding the Measure of Wealth: Indicators of Environmentally Sustainable Development. Environmentally Sustainable Development Studies and Monographs Series No., vol. 17. World Bank, Washington, D.C. World Bank, 2003. Data & Statistics. www.worldbank.org.