How Do Natural Resources Influence Civil War?

A Medium-N Analysis

January 2, 2003

Michael Ross
Department of Political Science, UCLA
[email protected]

1
Abstract

Recent studies have found that natural resources and civil war are highly correlated. What

accounts for these correlations?

The causal mechanisms behind the correlations are not well-explained by previous

studies, although some have offered plausible hypotheses. Identifying the causal mechanisms

could help resolve contradictions among earlier studies; address the problems of endogeneity and

spuriousness; and help find appropriate policy responses. Discovering the causal mechanisms,

however, it not a straightforward task: data on key dimensions of conflict are scarce, of poor

quality, and difficult to quantify.

This paper uses a novel “medium-N” method to analyze these causal mechanisms. It

begins by explaining the research design. It then describes eleven hypotheses about how

resources may influence a conflict; specifies the observable implications of each; and reports

which of these implications can be observed in a sample of 13 civil wars. It finds that resource

wealth and civil war are not linked by a single mechanism, but a variety of mechanisms, which

may help account for the inconsistencies of earlier studies; that two of the most widely-cited

causal mechanisms do not appear to be valid; and that resources appear to play different roles in

separatist and non-separatist conflicts.

2
Introduction

Recent studies have demonstrated that natural resources and civil war are highly correlated.

According to Collier and Hoeffler [1998, 2001], states that rely heavily on the export of primary

commodities face a higher risk of civil war than resource-poor states. Fearon and Laitin [2002]

and de Soysa [2002], each using unique data sets, find that oil-exporting states are more likely to

suffer from civil wars. Fearon [2002] also shows that the presence of certain types of resources

(gemstones and narcotics) tends to make wars last longer; similarly, Doyle and Sambanis [2001]

demonstrate that civil wars are harder to end when they occur in countries that depend on primary

commodity exports. Buhaug and Gates [2002] show that the presence of mineral resources in a

conflict zone tends to increase a conflict’s geographical scope.

There is little agreement among these and other scholars on why natural resources have

these effects: most have little to say about causal mechanisms – the processes that link

“resources” to “conflict.” Journalists often claim that resources have “fueled” a given conflict but

are vague about how this occurred.

Identifying the mechanisms that link resources to civil war would make these theories

more complete and persuasive: statistical correlations can only take us so far. It would also

address three problems in the natural resources-civil wars literature. First, it could help resolve

nagging concerns about endogeneity and spuriousness. The natural resource-civil war

correlation, for example, might be the opposite of what it appears: civil wars might produce

resource dependence by forcing a country’s manufacturing sector to flee while leaving its

resource sector – which is location-specific and cannot depart – the major force in the economy

by default. Even though most scholars employ lagged independent variables in their regressions,

this does not rule out reverse causality: since civil wars are not recognized as “beginning” until

they have generated at least a thousand combat-related deaths, the wars might be preceded by

years of low-level violence that drives off manufacturing firms, producing a higher level of

resource dependence before the civil war officially commences.

3
 The natural resource-civil war correlations could also be spurious: both civil war and

resource dependence might be independently caused by some unmeasured third variable, such as

the weak rule of law. A state where the rule of law is weak might be unable to attract investment

in its manufacturing sector, and hence would depend more heavily on resource exports; it might

also face a heightened risk of civil war through a different process. The result could be a

statistically-significant correlation between resource dependence and civil war, even though

neither factor would cause the other.

Second, identifying the causal mechanisms could help settle disagreements among the

statistical studies over which resources matter, and what dimensions of conflict they tend to

influence. Collier and Hoeffler [2001], for example, find that primary commodities of all types –

including oil, minerals, and agricultural goods – are linked to the onset of war. Both Fearon and

Laitin [2002] and De Soysa [2002] dispute this claim, and suggest that only oil matters. Collier,

Hoeffler, and Söderbom [2001] suggest that primary commodities have no influence on the

duration of conflict, a claim that is apparently contradicted by Doyle and Sambanis [2001] and

Stedman [2001]. Fearon [2002], meanwhile, suggests that contraband commodities, like

diamonds and drugs, make wars last longer. A closer look at case studies may help resolve some

of these contradictions.

Finally, different mechanisms suggest different policy interventions. For example, if

mining causes conflict because it produces grievances over environmental degradation and access

to jobs, the solution might be greater community involvement by mining firms. But if conflicts

occur because mining provides extortion opportunities for rebel groups and warlords, the solution

might be stricter mine site security and less community involvement. The UN Security Council,

the World Bank, and the G8 have all been engaged in policy responses to the resource-civil war

issue, making this concern highly salient.

Identifying the correct causal mechanisms, however, it not a straightforward task. Some

of the purported causal mechanisms have been carefully specified, but most have not. Once

4
specified, it is not clear how these mechanisms can be tested. If we had sufficiently high-quality

data for a large number of civil wars, we could test for causal mechanisms by placing intervening

variables on the right-hand side of regressions on war onset or war duration. Unfortunately, data

on the requisite dimensions of conflict are scarce and typically of poor quality; moreover, some of

the key intervening variables are difficult to quantify.

Another approach might be to conduct case studies. Case studies can tell us a great deal

about causal mechanisms by what is commonly called “process-tracing” – making explicit or

implicit inferences about the salience of the intervening variables that link cause and effect. Yet

when the theory we wish to examine is probabilistic (as most political science theories are), it

becomes virtually impossible to generalize from one or two case studies to a larger population,

since the cases we examine have only a certain probability of exhibiting the hypothesized

phenomena [King, Keohane, and Verba 1994, 209-211]. When our units of analysis are

substantially heterogeneous – as civil wars appear to be – generalizing from one or two cases is

more hazardous still.

One solution might be to commission a large number of case studies from country experts

to achieve some measure of generality. Yet, as anyone who has read an edited volume of case

studies knows, it is hard to aggregate the results of different case studies made by different

authors, since they typically their define their variables and construct counterfactuals in different

ways. It is not simply that “the whole is less than the sum of the parts”; rather, the parts often

cannot be summed.

How then can we better understand the mechanisms behind these natural resources-civil

war correlations? This paper suggests and illustrates one possible method, which I call a

“medium-N” analysis. While it lacks some of the desirable properties of both carefully-executed

statistical tests and high-quality case studies, it can nonetheless advance our understanding of

poorly-understood causal relationships – particularly when data are scarce and difficult to

quantify.

5
 This paper begins with an explanation of the medium-N method, and then discusses the

selection of cases. Section Three describes eleven causal mechanisms that might account for the

resource-civil war correlation, and suggests how they might be confirmed or disconfirmed in case

studies. Section Four presents the results from an analysis of 13 cases of recent civil wars. The

final section concludes.

The paper reaches six main conclusions: that there is good case-study evidence that

certain types of natural resources – oil, gemstones, and drugs – can influence the onset and

duration of civil wars; that most civil wars in the sample were influenced by natural resources

through several mechanisms simultaneously, which may help account for some of the analytical

muddle of some earlier studies; that resources do not necessarily make conflicts longer or more

severe – they can also shorten conflicts and promote cooperation among opposing sides; that two

of the most commonly-cited causal mechanisms do not appear to be valid; and that resources

appear to play a different role in separatist conflicts than in non-separatist conflicts. It also

describes an unanticipated causal mechanism that may link resource wealth and civil war.

1. The Medium-N Method

The medium-N approach is a method of aggregating data from a sample of case studies about

causal relationships, and drawing inferences from this data. To make the cases comparable, it

employs similar types of counterfactuals – and hence makes similar kinds of causal inferences –

within each of the cases. It retains some of the valuable properties of case studies, including

close attention to the validity of concepts, and to causal linkages; a capacity to account for

variables that are difficult to measure; and a sensitivity to case-specific factors. It also borrows

some of the important properties of statistical analysis, including a concern with generality; an

explicit and transparent method, which facilitates replication; and the use of clear and

consistently-applied assumptions about causal relationships. Although I use the method here to

study causal mechanisms, it can be applied to the study of other types of causal relationships.

6
 The analysis comes in two variants: a hypothesis-testing version and a hypothesis-

development version. The hypothesis-testing method has five steps. First, the researcher begins

with hypotheses about causal relationships that are both general and falsifiable. They must then

state the observable implications of each hypothesis – what he or she would expect to observe at

the case-study level if the hypothesis were true, in a manner that is general enough to cover a

wide range of cases but specific enough to minimize spurious inferences. The third step entails

the selection of cases, and the fourth step, the investigation of each case to determine the presence

or absence of each of the observable implications. Note that the researcher focuses on causal

relationships within each case, and does not make cross-case comparisons.1 In the final step the

researcher aggregates the results and draws inferences from the sample about the larger

population of cases.

The hypothesis-testing method itself has two variants, depending on how the cases are

selected. If the cases are randomly selected and sufficient in number, the researcher’s inferences

about causal relationships in the sample should also be valid for the larger population.2 A

researcher may also select cases based on a “most likely” research design, examining the set of

cases in which the hypothesized causal relationship is most likely to be found. The traditional

“most likely” research design, described by Eckstein [1975], employs a single case study to

establish the plausibility of a hypothesis, or more commonly, to refute it. As King, Keohane, and

Verba [1994] note, this method is flawed in several ways, most importantly because most causal

hypotheses in the social sciences are probabilistic, and hence cannot be falsified with a single

case. The medium-N method helps alleviate this problem by allowing a researcher to evaluate a

1
The absence of comparisons between cases distinguishes this method from George’s [1979]
“structured, focused” case study method. It also circumvents the problem described by Collier
and Mahoney [1996]: scholars who examine a small number of randomly-selected cases can be
hobbled by the difficulty of making comparisons across heterogeneous units.
2
In small-N research, the random selection of cases is typically inappropriate; I am assuming that
a medium-N analysis opens the door to random selection, although this issue is beyond the scope
of this paper. There is an extensive literature on case selection for qualitative analysis. Some
important recent contributions including Geddes [1990], King, Keohane, and Verba [1994], and
Collier and Mahoney [1996].

7
relatively large number of “crucial cases” simultaneously. While selecting only “most likely”

cases – as I do here – does not enable the researcher to confirm the validity of a causal

relationship in some larger population, it should enable him or her to disconfirm one, if the

observable implications are not found.

The hypothesis-testing version of the method works best when a research program has

already produced hypotheses suitable for empirical analysis. When a research program is less

mature, the second, hypothesis-generating version, may be more appropriate. The hypothesis-

generating approach provides a framework for inductive reasoning about causal relationships. It

has four steps. It begins with the selection of cases; once again, they may be either randomly

selected or chosen on a “most likely” basis. Next, the researcher examines the relevant

phenomena in each of the cases, and in the third step, he or she develops general, falsifiable

hypotheses to explain the observed relationships. The fourth step once again entails listing the

observable implications of each of the hypotheses, to make them more concrete and to facilitate

future, out-of-sample testing. These hypotheses themselves – and their accompanying

illustrations in the case studies – are this method’s final products.

The key step in this second variant is case selection. A common problem with small-N

studies is the biased selection of cases, which are (or are suspected to be) chosen to validate or

falsify a given hypothesis [Bennett 1997]. A medium-N study with randomly-selected cases, or

“most likely” cases (selected through an explicit process), provides a relatively unbiased starting

point for developing concepts and testable hypotheses. Executed properly, the hypothesis-

generating method can help scholars specify potential causal mechanisms in general, falsifiable

terms, determine their observable implications, illustrate their plausibility, and ultimately, render

them suitable for future out-of-sample testing, or further exploration through formal models.

The two types of medium-N studies, hypothesis-generating and hypothesis-testing, can

also be combined. Care should be taken in doing so. There is a norm in political science – albeit

one that is often honored in the breach – against combining inductive and deductive methods. A

8
key reason for the norm is to guard against scholarship that uses a single data set (or case study)

to generate a hypothesis, and to subsequently “test” it. But it is possible – and indeed can be

fruitful – to use the same data set to first test pre-existing hypotheses, and then to produce

alternative hypotheses for future out-of-sample testing. Indeed, only after determining whether

existing hypotheses can fully account for the observed phenomena is it appropriate to generate

new hypotheses. This is what I do below: having tested eleven pre-existing hypotheses with my

thirteen cases, I find that they do not fully account for the natural resources-civil war correlations

I observe. I then develop four additional hypotheses for future out-of-sample testing.

The medium-N approach is designed, in part, to bring quantitative and qualitative

research methods closer together. Scholars who favor qualitative methods typically focus on the

validity of concepts within one or several cases, and pay less attention to generality; scholars who

use quantitative methods tend to focus on the generality of their concepts across many cases but

are less concerned about their validity within cases. The medium-N method can force

researchers to be concerned simultaneously with validity and generality; this may help the two

traditions develop more congruent norms about concept validity and hypothesis testing.3

The medium-N method has important limitations. It sacrifices some of the close attention

to detail found in comprehensive case studies, and leaves little room for more extensive

counterfactuals within each of the cases. It cannot identify partial correlations across a large

number of observations the way regressions can. Moreover, some casual relationships may have

no implications that can be readily observed in case studies, at least without extensive primary

research. In my study below, two of the eleven hypotheses I examine cannot be confirmed or

falsified in the case studies I conduct. Still, at key junctures in the development of a research

program, the medium-N method may make an important contribution.

3
The method might be seen as one way to operationalize what Charles Ragin [1994] calls the
process of “double-fitting,” that is, developing concepts with appropriate degrees of both validity
and generality. On recent approaches to narrowing the gap between large and small-N studies,
see Coppedge [2002]; Lieberman [2002]; and Bennett [1997].

9
2. Case Selection

The 13 cases in the sample were selected from the Collier-Hoeffler list of 39 civil wars that began

or continued between 1990 and 2000. The cases were chosen on a “most likely” basis: the

sample includes all civil wars that occurred between 1990 and 2000 in which scholars, non-

governmental organizations, or United Nations agencies suggested that natural resource wealth,

or natural resource dependence, influenced the war’s onset, duration, or casualty rate. I employ a

broad definition of “natural resources,” including oil, gas, gemstones, nonfuel minerals, timber,

and agricultural commodities, including illicit drugs. While there are additional countries where

natural resources may have influenced low-level conflicts, the sample is limited to cases that meet

the common definition of a civil war: a conflict between a government and an organized rebel

movement that produces at least one thousand battle-related deaths.4

The thirteen cases vary by conflict type, and include three separatist wars (Sudan,

Indonesia, and Burma) and ten non-separatist wars (Afghanistan, Angola, Cambodia, Colombia,

Republic of Congo, Liberia, Peru, Sierra Leone, and two successive wars in the Democratic

Republic of Congo [DRC]).

If I were trying to determine whether resource wealth is correlated with civil war, this

would be the wrong set of cases to look at, since in these cases such a link is likely due to the

selection method. But this is not my concern: the resource dependence-civil war correlation has

already been established by the large-N studies discussed above. What I wish to research are the

causal processes that link the variables together.

4
Low-level conflicts that may be linked to natural resources include the Bougainville rebellion in
Papua New Guinea; the Cabinda conflict in Angola; the West Papua rebellion in Indonesia; the
conflict in Senegal’s Casamance region; and the independence movement in Western Sahara. For
a more extensive discussion of these and other cases, see Le Billon (2001).

10
 The “most likely” method of case selection enables me to carry out three types of

analysis. First, if the causal mechanisms that others have proposed can be illustrated in these

thirteen cases, I may deem them “plausible.” Second and more powerfully, if a purported causal

mechanism is not observed in this set of cases, I can infer that it is unlikely to be valid more

generally. Similarly, by observing whether a mechanism is absent in all of the separatist or non-

separatist conflicts in the sample, I can make inferences about the mechanism’s validity in each

sub-category of conflicts. Finally, I can use the cases to develop new hypotheses about causal

mechanisms; to determine whether these mechanisms are valid for a larger set of countries would

require out-of-sample tests, which I do not carry out.

The most-likely research design does not permit me to make cross-national inferences

within my sample (except for comparisons between two subcategories of cases, the separatist and

non-separatist conflicts); nor can I make valid inferences for the larger population of states about

the frequency, or relative weight, of the causal mechanisms I observe.

Since the sample only includes cases in which resource wealth is likely to have an effect

on the onset or duration of a civil war, I am unlikely to find – and indeed, do not find – evidence

that the resource-civil war correlation is spurious, or that civil wars cause resource dependence

instead of the reverse. But by determining whether the resource wealth-civil war link is internally

valid in a substantial number of cases, I can ease (or heighten) suspicions that the correlation is

spurious or reversed.

3. Eleven Hypothesized Mechanisms

There are at least eleven ways that a country’s natural resource wealth might influence the

initiation or duration of a civil war [Figure 1].5 The first five mechanisms listed below describe

5
I use the terms “resource wealth” and “resource dependence” interchangably here. Most of the
large-N studies measure the correlation between civil war and resource dependence, defined as

11
ways that resource wealth could lead to the onset of conflict; the next four suggest ways that

resource wealth could influence the duration of a conflict; and the final two describe how

resource wealth might influence the intensity of a conflict, i.e., the casualty rate.6

I include the hypotheses on conflict intensity because it is possible that the resource

wealth-civil war correlation is produced solely (or partly) by an intensity effect. To become

classified as a civil war, a conflict must pass a certain threshold, producing at least one thousand

combat-related deaths over some period of time. The presence of resource wealth might turn

low-intensity conflicts into high-intensity conflicts without influencing the total number of

conflicts; this could produce a statistical correlation between resource dependence and the

incidence of civil war by increasing the number of conflicts that cross the critical threshold.

Similarly, if resource wealth increased the number of years in which the conflict crossed the

thousand-death threshold without influencing the conflict’s beginning and end dates, it could

produce a spurious correlation between resource wealth and duration. Hence it is valuable to

explore whether resource wealth has an influence on the intensity of civil wars.

The eleven hypotheses below were taken from other scholars’ accounts of resource-based

civil wars. Often they discussed causal mechanisms briefly or indirectly; I have tried to turn their

implicit hypotheses into explicit ones. Some of the mechanisms are linked to each other. By

treating them as discrete mechanisms, I can specify them more clearly and test them more

directly. None of the mechanisms are mutually exclusive, and most scholars cited below discuss

multiple causal mechanisms.

Onset of Civil War

the ratio of natural resource exports (including oil, gas, minerals, and agricultural commodities) to
GDP. Most scholars treat this as an indicator of the relative abundance of natural resource wealth
in the economy. But since this indicator is sensitive to changes in the size of the non-resource
sector, and the size of GDP, it is a less-than-ideal measure. In the case studies I examine the
effects of commercially-exploited resource wealth on conflict, which eliminates this problem.
6
In Section Four, I inductively develop and illustrate four additional hypotheses, based on my
analysis of the thirteen cases.

12
The presence of resource wealth might cause the onset of civil wars in five ways; the first four

can be tested with case studies, while the final one can be explored statistically.

Perhaps the most influential hypothesis on resources and conflict comes from the work of

Collier and Hoeffler [2001]; I refer to it as the “looting” mechanism. Collier and Hoeffler

suggest that explanations for civil wars fall in two categories: those that focus on the motives or

“grievances” of rebel organizations, and those that focus on their funding. The most significant

funding opportunities for insurgents, they suggest, tend to come from exportable natural

resources: if rebels can extract and sell resources, or extort money from those who do, then they

are more likely to launch a civil war.

The Collier-Hoeffler argument comes from their observation that natural resources offer

rebel groups unusual funding opportunities, because resources typically produce rents and are

location-specific. If rebels try to loot or extort money from manufacturing firms, the firms will

relocate to a safe area or be forced out of business; but if they extort money from resource firms,

the firms cannot relocate, and can often make payments to rebels and still turn a profit. States

whose economies are more heavily based on resource exports should therefore also face a higher

risk of civil wars.

In their empirical tests, Collier and Hoeffler [2001, 16] find that the effect of a country’s

primary commodity exports on its conflict risk is “both highly significant and considerable”; they

state, “we have interpreted (this correlation) as being due to the opportunities such commodities

provide for extortion, making rebellion feasible and even attractive.”7 They reject the possibility

that primary commodities lead to conflict through a grievance mechanism.

The “looting” mechanism might be stated as:

7
Collier and Hoeffler estimate that the correlation between resource dependence and civil war is
curvilinear, suggesting that the risk of civil war declines when resource dependence reaches
exceptionally high levels, at which point “the increased tax revenue eventually augments the
capacity of the government to defend itself sufficiently to offset the enhanced finances of the
rebels.” Other scholars, such as Hegre [2002], estimate the correlation to be linear.

13
 H1: Primary commodities increase the probability of civil war by enabling nascent rebel

groups to raise money either by extracting and selling the commodities directly, or by

extorting money from others who do.

If this is correct, then in case studies we should observe rebel organizations raising money, prior

to the start of the civil war, through the extraction and sale of natural resources, or from the

extortion of resource firms.8

The second possible mechanism – which has been widely cited by policy analysts and

journalists – is a “grievance” mechanism. It suggests that resource extraction creates grievances

among the local population, due to land expropriation, environmental hazards, insufficient job

opportunities, and the social disruptions caused by labor migration; these grievances, in turn, lead

to civil war.9 Klare [2001, 208], for example, suggests that “resource wars” are caused in part by

logging or mining firms that are “ravaging the environment” and “driving off the people who

have long inhabited the area or depriving them of any benefits from the appropriation of their

traditional lands.” Gedicks [2001] and Switzer [2001] offer similar arguments; so do many

journalists [e.g., Onishi 2002].10 These arguments suggest

H2: Resource wealth increases the probability of civil war by causing grievances over

insufficiently compensated land expropriation, environmental degradation, inadequate

job opportunities, and labor migration.

8
The looting mechanism suggests a second observable implication: if looting resource firms is
easier, or more sustainable, than looting non-resource firms, we should observe rebel groups
gaining a greater fraction of their financing from the resource sector (relative to its size in the
economy) than from other economic sectors. This would be hard to test unless rebel
organizations agree to have their finances audited.
9
Some might argue that resource wealth should promote economic growth, and hence reduce
poverty and grievances. In fact, resource exploitation appears to reduce economic growth rates
[Sachs and Warner 1995; Leite and Weideman 1998], and to increase poverty rates, child
malnutrition, and infant mortality [Ross 2001].
10
These theories might be seen as part of a larger literature that argues that grievances, often
proxied by poverty or inequality, tend to influence the danger of civil war. See, for example,
Muller and Weede [1990]; Auvinen [1997]; Dudley and Miller [1998].

14
 If resource exploitation leads to civil war through a grievance mechanism, we should

observe the rebels criticizing resource firms, or the resource sector, in their propaganda; and we

should see them make resource firms a target of their violence, apart from looting or extortion

attempts. Of course, neither of these indicators would prove that insurgents are truly motivated

by resource-related grievances. But they would imply that the rebels believe that resource issues

are salient concerns in the population they wish to mobilize, and that raising these issues will help

them build support.

A third possibility is that resource wealth, if it is located on a country’s periphery, or in

an area populated by an ethnic minority, will give local residents a financial incentive to establish

a separate state, thus raising the risk of a civil war. Le Billon [2001] discusses this mechanism;

Collier and Hoeffler [2002] offer it as well. It implies

H3: Resource wealth increases the probability of civil war by giving residents in

resource-rich areas an incentive to form a separate state.

If this mechanism is valid, we should observe in case studies that a) the conflict is a

separatist war; b) the conflict began after the separatist region was identified as having

exploitable resource wealth; and c) that the rebel group discusses the unfair distribution of

resource wealth in its propaganda. We would not necessarily observe rebels attacking resource

firms in this instance, since they should in principle support resource extraction and may not wish

to alienate companies working in the sector. To distinguish the second and third mechanisms

from each other – since both entail local grievances around resource extraction – I look for

evidence of the second mechanism only in non-separatist conflicts, and the third mechanism only

in separatist conflicts.

A fourth mechanism – closely related to the first and second – might be called a

“warlord” mechanism. It generally applies to settings where the state is weak, and where

resources are highly valuable and can be extracted by small groups of workers with little training

or capital (such as timber and alluvial diamonds). If the state is too weak to enforce property

15
rights, individual producers may instead pay local strongmen (e.g., warlords) to protect

themselves. In this type of setting, civil war would be the result of competition among warlords

for control of these payments. Several scholars have used formal models to explore the properties

of these warlord conflicts.11 Although different models suggest slightly different links between

the presence of warlords and the outbreak of conflict, a general form of this argument might be

phrased as:

H4: Resource wealth increases the probability of civil war by generating conflict among

non-state actors who seek to control access to exploitation rights.

If this hypothesis is correct, we should observe in case studies the rise of warlord-type

organizations in the extractive region before the war breaks out; and we should see these groups

directly contribute to the conflict by aiding the rebel movement, or impeding the government’s

efforts to suppress it.12

Fearon and Laitin [2002], among others, have suggested a fifth mechanism: that resource

wealth – in particular, oil – causes “state weakness,” which in turn increases the probability of

civil war. The claim that oil wealth influences the character of the state has a long heritage

among Middle East scholars: they commonly suggest that oil wealth relieves governments of the

need to levy taxes, which in turn produces a state that is less responsive to its citizens. 13 Karl

[1997] developed this argument further, suggesting that oil wealth also tends to impede the ability

of states to resolve social conflicts. Fearon and Laitin adopt this argument, and further suggest

that the absence of a “socially intrusive and elaborate bureaucratic system to raise revenues” will

make states more susceptible to civil war. This claim could be stated as

11
Important contributions include Hirschleifer [1989]; Grossman [1991]; Skaperdas [1992,
2002]; and Azam [2002].
12
This mechanism partly overlaps with the looting mechanism. If a warlord-type group itself
became the ‘rebel organization,’ I would treat it as a case of the looting mechanism. If the
presence of warlords contributes to the outbreak of civil war, even though the rebel group did not
arise from a warlord’s organization, I would treat it as a case of the warlord mechanism.
13
See, for example, Mahdavy [1970]; Beblawi and Luciani [1987]; Crystal [1990]; Chaudhry
[1997].

16
 H5: Oil wealth increases the probability of civil war by weakening the state’s

bureaucracy.

It is difficult to know what the observable implications of this hypothesis are at the case study

level. One possibility is that oil will foster greater bureaucratic corruption, which will drain funds

from the military; this would leave a state more vulnerable to civil war. This is not easy to

observe at the case-study level, although it can be examined statistically.

Table 2 reports the results of a cross-national OLS regression for 130 countries that

explores this mechanism. The dependent variable is military spending as a fraction of

government spending in 1997, the most recent year for which comprehensive data were

available.14 After controlling for the natural log of GDP per capita, column one shows that oil

dependence is strongly and positively related to military spending as a fraction of government

spending.15 Columns two, three, and four exchange oil for other commodities – non-fuel

minerals, non-food agricultural goods, and food crops – and shows the results are mixed. If oil

dependence leads to civil war by weakening a state’s military forces, it is not evident in these

regressions.

The mechanisms that may link oil to bureaucratic weakness – and more problematically,

bureaucratic weakness to subsequent conflict – could be diffuse and subtle. Advocates of this

mechanism must further specify its logic before it can be tested with case studies.

Duration

Natural resource wealth may influence the duration of civil war, independent of its effects on the

incidence of civil war. There are four mechanisms that could either lengthen or shorten a

conflict, depending on how they occur; three of them have implications that can readily be

observed in case studies.

14
Data are taken from World Bank [2001] and UNCTAD [1995].
15
All of the explanatory variables are measured for 1995.

17
 The first mechanism, once again, is looting. Many observers – including scholars,

NGOs, and analysts from international organizations – have suggested that resource wealth can

lengthen a conflict if it enables the rebels to fund themselves, and hence continue fighting instead

of being crushed or forced to the negotiating table.16 Many journalistic accounts of recent wars in

the mineral-rich states of Central and West Africa – including Liberia, Sierra Leone, the

Democratic Republic of Congo, and Angola – allude to this mechanism when they claim that

resources are “fueling” a conflict.

The mechanism entails two key assumptions: that the rebels are the weaker side; and that

strengthening the weaker side tends to lengthen conflicts. In fact, there is evidence from

interstate conflicts to support the latter claim: Bennett and Stam [1996] find that international

conflicts tend to last longer when the two sides have more equal resources.

If we assume that when the weaker side in a civil war gains additional resources, the

conflict will be lengthened, we must also assume that when the stronger side gains additional

resources, the conflict will be shorted, by bringing about a quicker victory or settlement.17 This

implies

H6: Resource wealth tends to increase (decrease) the duration of civil wars when it

provides funding to the weaker (stronger) side.

If this mechanism has occurred, there should be evidence that one side or the other has raised

money from the resource sector – through looting, extortion, or other means – after the war

began. If both sides raised funds from the resource sector simultaneously, I infer that the net

16
See, for example, UN Panel of Experts [2000, 2001]; Sherman [2000]; and the reports of
Global Witness, a London-based NGO, at www.globalwitness.org.
17
This raises several important problems for the coding of case studies. First, a judgement must
be made about the relatively military strength of the two sides before the resource is exploited, to
avoid the problem of endogeneity. Second, it is necessary to restrict the analysis to contested
resources. Virtually all governments derive at least a fraction of their revenue from the sale of
natural resources; but I only treat these as relevant if they are located in the contested terrain. For
civil wars that are national in scope, I treat all resources as contested.

18
effect has been to lengthen the conflict, based on the conjecture that combat is likely to continue

as long as the weaker party does not run out of money. 18

Some scholars have suggested a second duration-related mechanism: that resource wealth

discourages peace settlements, if wartime looting is sufficiently profitable. Sherman [2000: 699],

for example, suggests that

“Rebel groups in Angola, Sierra Leone, Democratic Republic of Congo (DRC) and

elsewhere enrich themselves through the sale or exchange of diamonds…economic

interests not only shape the conflict, but, if the economic advantage of fighting outweighs

that of peace, perpetuate it as well.

Once again, this mechanism has a seldom-noticed corollary: if one or both parties believe

that peacetime profits would be greater than wartime profits, it could help induce them to reach a

settlement. In hypothesis form, this “incentive” mechanism and its corollary may be stated as

H7: Resource wealth tends to increase (decrease) the duration of civil wars by offering

combatants a financial incentive to oppose (support) a peace settlement.

This is a slippery mechanism to observe in case studies. It should not be sufficient to

observe that war is profitable for some combatants: this is virtually inevitable when combat takes

place on resource-rich territory, and it hardly proves that the parties are deliberately lengthening

the conflict. What we must determine is whether high level officers, who have the ability to

negotiate (or block) a treaty, believe they would profit more if the war continues than if it comes

to a negotiated end. If this is occurring we should observe a) evidence that resource looting is

generating personal profits for high-level officers; b) evidence that they would not be

compensated in some comparable way by a proposed peace treaty; and c) evidence that they

chose not to sign or adhere to an unprofitable peace accord. Conversely, if resource wealth is

18
This is less likely to be true in separatist conflicts than in non-separatist conflicts. As Fearon
[2002] points out, separatist and non-separatist conflicts appear to have substantially different
characteristics: separatist conflicts tend to last longer, and often continue even when the separatist
movement is at an overwhelming financial disadvantage.

19
facilitating a peace accord, we should observe officers who support a peace agreement

subsequently profit from – or attempt to profit from – the resource industry.

Fearon [1999] has suggested a third duration-related mechanism: that wartime resource

looting could create discipline problems within one or both of the opposing armies, in a manner

that impedes a peace accord. Both rebel and government armies in developing states commonly

have difficulty maintaining the chain of command during wartime. Fearon argues that the

presence of lootable resources could make principal-agent problems worse by giving soldiers an

incentive to accumulate personal wealth instead of obeying their commanding officers. These

discipline problems could make it more difficult for negotiators to forge a binding, enforceable

settlement.

It is conceivable, once again, that resource wealth could have the opposite effect: soldiers

on one side or the other might be so intent on getting rich that they refuse to fight, hastening their

army’s military defeat or forcing them to sign a peace agreement. These arguments imply

H8: Resource wealth tends to increase (decrease) the duration of civil wars by creating

discipline problems among combatants that make a peace settlement more (less) difficult.

If this “discipline” mechanism has lengthened a conflict, we should observe the following: a) that

at least one army in the conflict suffers from major discipline problems; b) that disobedient

officers are personally benefiting from resource looting; and c) that these discipline problems

have made it harder for that party to sign or adhere to a peace settlement.

If the discipline mechanism has shortened a conflict, we should again see a) that at least

one army suffers from major discipline problems; b) that disobedient soldiers are personally

benefiting from resource looting; and c) that these problems helped cause their army’s defeat, or

forced it to sign a peace agreement that it might otherwise have opposed.

This “discipline” mechanism is similar to the previous “incentive” mechanism: both

focus on the ways that lootable resource wealth can create personal wealth for soldiers that make

20
them more (or less) inclined to settle. While the “incentive” mechanism focuses on commanding

officers, the “discipline” mechanism focuses on rank-and-file soldiers.

Finally, Fearon [2002] has suggested a fifth way that resource abundance might lengthen

separatist conflicts. He specifies a model in which the likelihood that rebels will settle a conflict

through an agreement for regional autonomy, depends, in part, on whether they believe the

government is likely to adhere to it. The model suggests that if the region has resource wealth the

government is more likely to renege on any such agreement, in order to gain access to future

resource revenues; even if the government does not plan to renege, the rebels are more likely to

expect them to renege, and hence should be more hesitant to sign a peace accord.19 The net result

is that separatist conflicts over resource-rich regions should be unusually difficult to settle, due to

the commitment problems that are aggravated by resource wealth. It may be phrased as

H9: Resource wealth tends to increase the duration of separatist civil wars by making it

less likely that the government will adhere to a peace accord that gives the region fiscal

autonomy.

Unfortunately, this final mechanism is hard to verify in case studies unless we know a great deal

about the perceptions and motivations of rebel leaders. For this reason, I do not attempt to test it

in these case studies.

Intensity

Resource wealth might also influence the intensity of civil conflicts, producing more (or fewer)

conflict-related deaths over time. Two mechanisms might bring this about.

The most obvious mechanism is resource-related combat, in which opposing armies do

battle over resource-rich territory. Many observers of Africa’s recent civil wars have suggested

that combatants are “fighting for control” of a resource, implying

19
The Fearon model includes the further suggestion that resource wealth discourages a peace
settlement when it provides rebels with a source of wealth during combat. Since this is already
covered by the “looting” and “incentive” hypotheses (H6 and H7) I do not include it here.

21
H10: Resource wealth tends to increase the casualty rate during a civil war by causing

combatants to fight for resource-rich territory that would otherwise have little value.

I look for evidence in each case of resource-related battles, over sites that had little or no intrinsic

strategic value.

The second mechanism might reduce the intensity of civil wars. Keen [1998] describes a

number of wars in which battlefield opponents lay down their arms and cooperate to extract

resources; this suggests

H11: Resource wealth tends to decrease the casualty rate during a civil war by causing

combatants to cooperate in resource exploitation.

If this type of cooperative plunder occurs, there should be reports of substantial wartime trade

and cooperation in resource exploitation between the two sides.20 From this we might infer that

the presence of resource wealth has reduced the casualty rate.

4. Results from Case Studies

The causal mechanisms observed in the thirteen cases are summarized in Tables 3 through 6.21

Overall there is good evidence to support both the claim that resource wealth makes conflict more

likely (in 6 of 13 cases), and the claim that resource wealth tends to make conflicts last longer (in

8 of 13 cases). Within the sample, the influence of resource wealth on conflict intensity varies

greatly – appearing to increase the casualty rate in two cases, having a mixed effect in eight, and

no effect in three.

20
I am deliberately omitting a third possible mechanism: that resource looting enables one or both
combatants to arm themselves with more lethal equipment and hence kill each other at a faster
rate. It is not obvious that greater military spending produces more lethal combat; moreover, I am
already assuming that resource revenues influence the duration of conflict and do not wish to
double-count.
21
I conducted one of the case studies – Indonesia – using primary sources and field work in June-
July 2000. I based the other twelve case studies on secondary sources, including academic
studies, interviews with country experts, United Nations reports, journalistic accounts, and reports
from non-governmental organizations. When data are missing or ambiguous, I note this either in
the text or in the appendix of case studies.

22
 The most striking finding may be that there is no evidence in the sample of the looting

mechanism, and little if any evidence of the grievance mechanism. There were several notable

differences between separatist and non-separatist conflicts. Resources also contributed to the

onset, duration, and intensity of conflict in four ways that were not predicted by the eleven

hypotheses.

Incidence of Conflict: Evidence

Resource wealth contributed to the outbreak of conflict in six of the thirteen cases: in two cases

(Indonesia and Sudan), resource wealth appeared to create an incentive for a separatist rebellion;

in two cases (Sierra Leone and the DRC 1996), the presence of resource-based warlords appeared

to contribute to the outbreak of conflict; and in four cases (Congo Republic, DRC 1998, Sierra

Leone and Sudan) resources seemed to contribute to the outbreak of conflict in ways that were

not predicted by the hypotheses [Table 4]. To account for these latter four cases, I develop two

new hypotheses.

There were no cases of the looting mechanism (H1) that Collier and Hoeffler suggest: in

these thirteen cases, nascent rebel groups never gained funding before the war broke out from the

extraction or sale of natural resources, or from the extortion of others who extract, transport, or

market resources. If interpreted strictly, the Collier-Hoeffler looting mechanism gains no

support from these cases.22 While there is abundant evidence that rebel groups engage in looting

after a war begins (discussed below), in this sample no rebel group funded its start-up costs from

the resource sector. Perhaps those who are able to loot during peacetime are too well-employed

to attack the government.

22
The closest that any case came to validating the looting hypothesis was Sudan. The Sudan
People’s Liberation Army funded itself in the early 1980’s by conducting cattle raids and by
kidnapping and ransoming westerners. They also extorted small sums from western firms that
wished to protect equipment they had left in Sudan to build an oil pipeline [Anderson 1999,
O’Ballance 2000]. Oil production itself, however, only began in 1999.

23
 There was also no evidence that the grievance mechanism (H2) has led to civil war,

although the case of Sierra Leone is ambiguous. In general, however, no non-separatist civil wars

were associated with complaints about land expropriation, environmental degradation,

insufficient employment opportunities, or pressures caused by labor migration to resource-rich

areas. This does not suggest that these grievances are illusory: they are real and ubiquitous. But

except in separatist conflicts, they never appeared to contribute to the outbreak of a civil war.

There may be an exception to this pattern: the case of Sierra Leone, where the evidence is

ambiguous. The war in Sierra Leone began in March 1991 when the Revolutionary United Front

(RUF) first crossed the border from Liberia. The following January RUF conducted operations in

diamond-rich southeastern Sierra Leone; beginning in September 1992, RUF and government

troops fought for control of the diamond-rich areas.

The case of Sierra Leone exhibits one of the indicators of a grievance-based conflict:

RUF propaganda complained about resource exploitation, railing against “the raping of the

countryside to feed the greed and caprice of the Freetown elite and their masters abroad” [cited in

Richards 1996].23 Whether or not it met the second criteria – attacking resource firms – is

uncertain: although RUF conducted operations in Sierra Leone’s diamond-producing operations

and drove out many Lebanese diamond traders, these operations may have simply been part of

RUF’s diamond-looting tactics. Hence it is unclear whether resource grievances helped initiate

the war in Sierra Leone.24 Even if Sierra Leone is coded as an example of a “grievance”

mechanism, it is notable that the grievances exploited by the rebel group concerned the

distribution of resource wealth, not land expropriation, labor migration, environmental damage or

lack of job opportunities.

23
Specialists disagree over whether the RUF leadership actually believed these charges [Richards
1996], or simply used this rhetoric for recruitment purposes [Abdullah 1998]. For the purposes of
this analysis, however, this dispute is irrelevent.
24
On RUF’s 1992 activities in the diamond-rich areas, see Richards [1996]; Reno [1998];
Abdullah and Muana [1998].

24
 The sample includes three separatist civil wars; in two of them (Indonesia and Sudan)

there is evidence of the “separatist” mechanism (H3). The Indonesian civil war occurred in the

northwest province of Aceh. The rebel group – widely known as GAM (Gerakan Aceh Merdeka,

Aceh Freedom Movement) – began in 1976, shortly before a large natural gas facility began

operations. GAM’s 1976 “Declaration of Independence” denounced the Indonesian government

for stealing Aceh’s resource revenues, but it did not criticize the natural gas facility itself, or

Mobil (now ExxonMobil), which operates the facility.25 One of its first acts was to attack the

plant [Robinson 1998].

The war in the Sudan began in 1983 when Sudanese President Numeiry took a series of

measures that upset the delicate balance between the predominantly Muslim north and the heavily

Christian and Animist south; among these measures was his decision to place newly discovered

oil in the country’s south under the jurisdiction of the north, and to build an oil refinery in the

north instead of the south. The Sudan People’s Liberation Army (SPLA) subsequently

complained that the north was stealing the resources of the south, including oil; demanded that

work cease on a pipeline to take oil from the south to the refinery in the north; and in February

1984, attacked an oil exploration base, killing three foreign workers and bringing the project to a

halt [O’Ballance 2000, Anderson 1999]. To date, the conflict has killed an estimated two million

people.

The warlord mechanism (H4) was apparent in two cases (Sierra Leone and the 1996-97

war in the Democratic Republic of Congo); in each case, the rise of local warlords indirectly

contributed to the initiation of conflict. Before the outbreak of civil war, Sierra Leone’s diamond

fields were controlled by a network of armed gangs, private armies and paramilitary forces that

sold protection services to miners and traders [Hirsch 2001, Richards 1996]. The presence of

these organizations weakened the government’s authority in the diamond areas; when the

25
Indeed, the founder of GAM, Hasan di Tiro, was a businessman who failed in his effort to win
a bid for a work contract at the natural gas facility.

25
government tried to re-establish control in early 1991, it displaced thousands of illicit miners,

some of whom were recruited by RUF, the rebel organization. The government’s efforts also

drove local strongmen into alliances with RUF [Reno 1998].

In the DRC, the rise in the 1980s and early 1990s of small-scale, artisanal mining led to a

proliferation of warlords who weakened the government’s military capacity. Some warlords were

local strongmen, who sold protection to small-scale miners [MacGaffey 1991]. Others were DRC

military officers who, along with their units, had stopped receiving salaries from the cash-

strapped government; to survive, they sold their services to dealers in diamonds, gold, coffee,

timber, cobalt, and arms. One general sold protection services to alluvial diamond miners in

Kivu; another general used his unit to ship cobalt from the Shaba province to Zambia [Reno

1998]. These activities weakened the DRC military, and apparently encouraged the rebel group

(organized and led by the Rwandan and Ugandan armies) to launch its assault in October 1996.

In four the of thirteen cases, natural resource wealth helped trigger conflicts in two ways

that were not predicted by the hypotheses.

The first way was by encouraging interventions from neighboring powers: in Sierra

Leone, and the DRC 1998 war, foreign forces decided to support nascent rebel groups against

incumbent governments, in part, to gain access to natural resource wealth. In Sierra Leone,

Liberia’s Charles Taylor helped organize and support the 1991 RUF invasion in order to gain

access to Sierra Leone’s diamond fields [Abdullah 1998; UN Panel of Experts 2000].

Similarly, the Ugandan and Rwandan governments decided to organize, and fight

alongside, rebels in the DRC partly because of the DRC’s resource wealth. The UN Panel of

Experts [2001] found that Uganda’s decision to enter the war was influenced, in part, by at least

three figures who were eager to profit from the occupation of resource-rich parts of the DRC. In

Rwanda, the government’s decision to back an incursion was influenced by the belief – which

was subsequently proved correct – that resource looting would help offset the cost to the

government of the invasion, which might have otherwise been prohibitive. Once inside the DRC,

26
the Rwandan army established well-disciplined procedures for extracting Congolese resources

and using them to fund the military effort [UN Panel of Experts 2001, 2002].

Based on these two cases, a new hypothesis might stated as:

H12: Resource wealth increases the probability of civil war by increasing the probability

of foreign intervention to support a rebel movement.

The second and more surprising mechanism entailed the sale, by rebel groups, of what

might be called “booty futures” – the right to exploit mineral resources that the seller has not yet

captured. In three cases (Congo Republic, Sierra Leone, and possibly Liberia), rebel groups that

had no resources to sell, but had a chance of securing them in combat, were able to sell future

mineral rights to foreign firms or neighboring governments. The rebels then used the proceeds

from the sale of booty futures to pay soldiers and buy arms, and thus gain the capacity to capture

the promised resource.

The special danger of selling booty futures comes from its self-fulfilling properties. If

the rebel group was unable to sell the future right to exploit the resource, it might not have the

funds it needs to capture the resource itself. Selling the future right to the resource makes its

seizure possible. Without the futures market, the rebel offensive – and perhaps the conflict itself

– would be less likely.26

In the Congo Republic, a former president, Denis Sassou-Nguesso, received $150 million

from the French oil company, Elf-Aquitaine, to help him defeat the incumbent president, Pascal

Lissouba, either by force or through a national election; the payment was clearly meant to ensure

26
In principle, rebels could sell futures for any type of war spoils, not just mineral resources. In
practice, minerals appear to be the only commodities for which future exploitation rights have
been sold. This may be because because minerals have two unusual qualities. First, they are
almost always owned by the state; hence, rebel armies that hope to displace the state – and the
firms that deal with these armies – believe they can legitimately buy and sell mineral rights.
Second, minerals can be exploited under conditions of instability and chaos, since they are often
found in remote areas, are geographically concentrated, and are processed abroad. By contrast,
agricultural commodities tend to be privately owned, and are harder to exploit during wartime
since they are less remote, geographically extensive, and often rely on the presence of a large
labor force to harvest and process locally.

27
Elf’s access to Congolese oil in a future Sassou government.27 The election never took place.

Instead, Sassou and Lissouba fought a four-month war that destroyed much of Brazzaville and

cost 10,000 lives, eventually leaving Sassou in charge [Galloy and Gruénai 1997].

Something similar occurred in Sierra Leone, when RUF launched its 1991 invasion.

RUF received material support from a variety of sources; they included the Liberian leader,

Charles Taylor, and a Sierra Leone businessman who had recently been forced out of the diamond

industry [UN Panel of Experts 2000; Reno 1998]. There is circumstantial evidence that the RUF

leadership traded this financial support for future diamond rights – in effect, using informal

mining futures to purchase their assistance.28

The notion of “booty futures,” in hypothesis form, might be stated as

H13: Resource wealth increases the probability of civil war by enabling rebel groups to

sell future exploitation rights to minerals they hope to capture.

To summarize, there is good evidence in this sample that civil wars have been triggered

by both the separatist mechanism and the warlord mechanism; however, there is no evidence of a

looting mechanism, and little or no evidence of a grievance mechanism. In addition, resource

wealth has apparently led to conflict through two unanticipated mechanisms: by encouraging

foreign powers to support insurgencies, and by enabling rebel groups to sell future mineral rights

to resources not yet under their control

27
Elf had lost its oil contract under the government of Lissouba, Sassou’s rival.
28
There may have been a similar sale of booty futures in Liberia, although the evidence is
unclear. The Liberian civil war began on Christmas Eve 1989, when Taylor led 100 troops from
the NPFL into Liberia from neighboring Côte D’Ivoire. Taylor was plainly determined to depose
Liberia’s President, Samuel Doe, and claim the presidency for himself; he was supported by the
leaders of Côte D’Ivoire and Burkina Faso, and had earlier received training from the Libyan
government.
According to Ellis [1999], just before the invasion Taylor met with businessmen who
hoped to gain access to Liberia’s iron ore and timber; Taylor reportedly received “sympathetic
attention” from them. Ellis is uncertain, however, whether these businessmen actually helped
finance Taylor’s efforts.

28
Duration of Conflict: Evidence

Resource wealth appears to have influenced the duration of ten of the thirteen conflicts: it

lengthened eight, shortened two, had a mixed effect in two and no effect in one [Table 5]. Once

again, an unanticipated mechanism – the sale of booty futures – influenced the duration of several

conflicts.

While looting played no role in the initiation of these thirteen conflicts, it played a role in

the duration of ten conflicts (H6). In other words, in these thirteen cases, rebel groups only

started to loot resources after the conflicts began. In nine of the ten cases, the looted commodity

was a type of resource that can be easily extracted, or cultivated, by small groups of unskilled

workers – mostly gemstones (five cases), drugs (two cases), or timber (two cases).

In two cases, however, rebels have used extortion and kidnapping to raise money from a

more difficult-to-loot commodity. In Colombia and Sudan, insurgents have been able to raise

money by blowing up oil pipelines and ransoming kidnapped oil workers. In both cases, the

rebels have capitalized on the precarious geography of the their country’s oil industry by

sabotaging pipelines that stretch for hundreds of miles, crossing territory where they have a

strong presence. In Colombia, two independent rebel movements bombed the pipelines 98 times

in 2000. Together they have used these attacks to extort an estimated $140 million annually; this

windfall has enabled one group, the National Liberation Army (ELN), to grow from fewer than

40 members to at least 3,000 [Dunning and Wirpsa 2001].

There was evidence in two cases (DRC 1998 and Liberia) that resource wealth

lengthened a conflict through an incentive mechanism (H7), giving combatants an economic

incentive to avoid signing, or adhering to, a peace agreement. In two other cases (Congo

Republic and Burma), however, the incentive mechanism had the opposite effect, giving

combatants a financial inducement to settle.

In Liberia, parties to the conflict signed the 1993 Cotonou accord under heavy

international pressure. But almost immediately the signatories created nominally-independent

29
surrogate groups that – because they were not signatories – could carry on with profitable

wartime looting. This practice contributed to the accord’s collapse [Alao et al. 1999, Ellis 1999].

It also implies that combatants subverted the treaty so they could continue their looting, thus

lengthening the conflict.

In the war that has plagued the DRC since 1998 – which has both the qualities of a civil

war and an international war – the profitability of resource looting for foreign governments, rebel

militias, and individual officers has substantially reduced their incentive to end the conflict .

Even though a peace accord was signed in Lusaka in July 1999, it was not implemented until

2002, in part because it forced foreign combatants to withdraw from the DRC, which hampered

their ability to siphon off the country’s remarkable resource wealth [ICG 2000, UN Panel of

Experts 2001, 2002].29

Conversely, the 1997 civil war in the Congo Republic may have been shortened by the

combatants’ agreement to share the oil revenues.30 Similarly, the Burmese government reached

settlements with the Shan State Army (in 1989) and the Kachin Independence Army (in 1994)

after agreeing to jointly exploit the opium, timber, and precious stones in rebel-held territory.

Although rebel groups already controlled these resources, the agreements made it easier for them

to attract new investment in, process, and export their goods [Lintner 1999].

There was little evidence of an incentive mechanism in two cases where others suspect it

exists: Angola and Sierra Leone [Sherman 2000]. In both cases, rebel leaders generated

enormous sums from resource looting; this may have led some observers to falsely infer that

resource wealth caused the rebels to prefer war to peace. But in each case, peace negotiators

anticipated this problem and drafted accords that would enable rebel leaders to continue getting

rich – or get even richer – in peacetime. Both the 1999 Lomé accord in Sierra Leone, and the

29
The foreign armies withdrew in 2002 only after making arrangements – by establishing joint
ventures, and by using local militias that acted as their surrogates – to continue profiting from the
DRC’s mineral wealth [UN Panel of Experts 2002].
30
Pierre Englebert, personal communication, October 2001.

30
1994 Lusaka Protocols in Angola, offered to place the rebel leader (Foday Sankoh in Sierra

Leone and Jonas Savimbi in Angola) in charge of the country’s natural resources under a unity

government. Peace would also allow the minerals sector in each country to expand by enabling

abandoned mines to reopen and new ones to develop, presenting the rebels with new

opportunities for enrichment [Le Billon 1999; Human Rights Watch 1999]. In these cases, these

peace accords failed, but for reasons other than the lure of wartime looting.

The discipline mechanism (H8) appeared to influence the duration of two conflicts, albeit

in opposite ways. In Liberia it lengthened the civil war by impeding a settlement, while in

Cambodia it shortened the war by crippling the weaker side with defections.

The Liberian civil war lasted from December 1989 to August 1996. Between June 1990

and August 1996 the combatants signed fourteen peace accords, thirteen of which failed. One

important reason for these failures was competition within the parties over the control of resource

wealth – competition that made it impossible for some of them to get their own troops to comply

with the terms of the agreements [Accord 1996; Ellis 1999]. By delaying the implementation of

a peace accord, it lengthened the war.

Conversely, in Cambodia resource wealth created agency problems within the rebel

group (the Khmer Rouge) that hastened their breakdown and – at least after 1996 – shortened the

civil war. Until 1995, income from the sale of timber and gemstones had helped fund the Khmer

Rouge, and hence lengthened the civil war. But in 1996 Ieng Sary, one of the Khmer Rouge’s top

officials, surrendered to the government along with 4,000 soldiers under his command. As part of

the surrender agreement, he was allowed to retain his troops and keep control of a gem-and-

timber-rich area near the Thai border. The Khmer Rouge never recovered from his defection, and

by 1998 the Khmer Rouge had collapsed, bringing about an end to the war [Brown and Zasloff

1998; Le Billon 2000].

31
 A close look at the thirteen cases suggests that an unanticipated mechanism – once again,

the sale of booty futures – influenced the duration of three conflicts: Sierra Leone, Angola, and

the DRC.

During the war in Sierra Leone, the government saved itself from defeat twice by selling

off the right to exploit diamond fields that it did not yet control. In March 1995, RUF had taken

control of the country’s main diamond fields and advanced to within 20 miles of the capital. To

stave off defeat, the government sold future mining rights to the Kono diamond fields – then in

rebel hands – to Branch Energy, a South African company; the government then used the

proceeds to hire a South African mercenary firm, Executive Outcomes, to beat back the RUF

offensive and recapture the mortgaged diamond fields [Hirsch 2001]. Just two years later, a

deposed president, Ahmad Tejan Kabbah, sold $10 million in diamond futures to a Thai banker;

Kabbah used the revenues to hire Sandline, a London-based mercenary firm, to take back the

capital and the diamond fields [Africa Confidential 2001]. In each case, the sale of future mineral

rights helped prolong the conflict.

The sale of booty futures also lengthened the Angolan conflict. In 1992-93, the rebel

group UNITA waged an offensive that brought more than 70 percent of the country – including

all of its diamond-rich areas, and the northern oil town of Soyo – under its control. To fund a

counteroffensive, the government sold off future exploitation rights to both oil fields (still under

the government’s control), and diamond areas (some of which were under rebel control). In one

deal, the government hired International Defence and Security (IDAS), a private military services

company, to retake the diamond fields near the DRC border; the government paid IDAS with a

share of the contested diamonds [Peleman 2000; Human Rights Watch 1999].

In the 1996-97 conflict in the DRC, the sale of booty futures most likely shortened the

conflict since it generally benefited the stronger side. In this conflict, the rebel organization (the

Alliance of Democratic Forces for the Liberation of Congo/Zaïre, led by Laurent-Desiré Kabila)

received a huge resource windfall after it became clear that it was defeating the government in

32
combat.31 In April 1997, Kabila signed an $885 million contract with American Mining Fields, a

U.S. firm intent on exploiting Congolese copper, cobalt, and zinc.32 Around the same time, the

minerals parastatal, Miniére de Bakangwa, switched its support from the government to ADFL,

offering Kabila both cash and the use of its aircraft fleet [French 1997; Reed 1998]. One month

later, Kabila entered the capital and became the new President. Since Kabila’s April 1997 sale of

mineral futures helped strengthen the hand of the winning side, I infer that it helped bring about a

swifter end to the war.

These three cases suggest

H14: Resource wealth tends to increase (decrease) the duration of civil wars by enabling

the weaker (stronger) side to earn revenues by selling future exploitation rights to

minerals they hope to capture.

In sum, there is good evidence to support all three of the hypothesized mechanisms that

can be tested; there is also evidence of an unanticipated mechanism, the sale of booty futures.

While eight of the conflicts in the sample were lengthened, one was shortened, and in four

conflicts resources had a mixed effect.33

Intensity of Conflict: Evidence

There was evidence of both hypothesized effects – resource battles and cooperative plunder – in

the thirteen cases [Table 6]. Often both mechanisms appeared in the same war. There was, once

again, evidence of an unexpected mechanism influencing the intensity of conflicts. Collectively,

31
The ADFL was led by the Rwandan army and backed by the Ugandan army, who were
principally concerned with eliminating the threat created by the exiled Rwandan government in
eastern Congo. The exiled Rwandan government was led by ethnic Hutus and was responsible
for the 1994 Rwandan genocide. Kabila was a longtime political figure who had opposed
Mobutu since the early 1960s.
32
The $885 million figure represented future investment. However, it is customary in large deals
for the company to also pay a signing bonus, which would have augmented the AFDL’s revenues.
The Kabila government later reneged on the contract.
33
I determine the net effect as follows: if a case exhibited only conflict-lengthening, or conflict-
shortening, effects, I judge the net effect as “longer” or “shorter,” respectively. If a case has both
conflict-lengthening and conflict-shortening effects, I list the net effect as “mixed.”

33
resource wealth heightened the casualty rate in two wars, had no effect in three wars, and had a

mixed impact in eight wars.34

Resource battles (H10) occurred in nine of the thirteen cases, as combatants fought for

control of areas rich in alluvial gemstones (Sierra Leone, Liberia, Cambodia), opium fields and

processing plants (Peru, Burma), oil pipelines that traveled over disputed lands (Colombia,

Sudan), mines (DRC, Liberia), and commercially-valuable forests (Cambodia, Liberia).

Yet in eight of these nine cases, combatants intermittently cooperated in exploiting the

same resources they fought over (H11). In four cases (Sierra Leone, Liberia, the DRC, and

Cambodia) there were long periods in which the major parties more or less ceased their combat

and entered a kind of commercial equilibrium. Even in extraordinarily bitter wars like the one in

Sudan, profitable alliances were often struck between groups on opposing sides – in this case, to

guard the pipeline and oilfields that the rebels have long opposed [ICG 2002].

The only war that featured resource battles but not cooperative plunder was in Peru,

between the government and the hard-line Maoist group, Sendero Luminoso. Beginning around

1983, Sendero Luminoso controlled a large coca-producing area in Peru’s Upper Huallaga

Valley; they also periodically clashed with both government forces and a rival guerrilla group

over control of the coca trade. Their failure to cooperate with the Peruvian military in coca

production likely reflected both their highly-disciplined and ideological character, and their

ability to fly coca paste directly from the Upper Huallaga Valley to Colombia without passing

through government-controlled territory or airspace.

Resource battles and cooperative plunder seem to be closely linked. In the eight cases

where both occurred, it was impossible to judge which of these two effects had the greatest

34
I evaluate the net effect as follows: if a case exhibited only conflict-enhancing, or only conflict-
reducing effects, I list the net effect as “worse” or “better,” respectively. If there are both
conflict-enhancing and conflict-reducing effects, I judge the net effect to be “mixed.”

34
impact. I hence infer that they at least partially offset each other and produced a “mixed” effect

on the intensity of combat.35

In two of the cases (Indonesia and Sudan), a third effect appeared to link resource wealth

with the intensity of combat; the mechanism might be called “pre-emptive repression.” In each

case, a government facing a small, separatist rebellion in a resource-rich area acted strategically

to protect its control of the resource, by using terror against a population. Had the region been

resource-poor, the governments may have responded less harshly to these challenges, producing

fewer casualties.

In the Indonesian case, the government imposed martial law, terrorized villages, and

carried out egregious human rights abuses in Aceh between 1990 and 1998 to crush a tiny

independence movement, which had claimed that the government was unjustly appropriating

Aceh’s resource wealth. The repression ultimately backfired, triggering a larger rebel movement

and a renewal of the conflict in 2000 [Robinson 1998].

In Sudan, the pre-emptive repression has been severe: beginning in the late 1990s, the

government attempted to create a cordon sanitaire around a 936-milepipeline that brings oil from

the rebellious south to a port in the north. Since early 1999, the government has used summary

executions, rape, ground attacks, helicopter gunships, and high-altitude bombing to force tens of

thousands of people from their homes in the oil regions [Amnesty International 2000]. The

correlation between oil exploitation and pre-emptive repression has often been transparent. In

one well-documented case, Lundin Oil (a Swedish firm) discovered a major oil reserve in April

1999 at Thar Jath; a month later, government troops displaced tens of thousands of people from

the area. When fighting nonetheless erupted ten months later around the Thar Jath site, Lundin

35
In five of these nine cases (Sudan, Sierra Leone, Liberia, the DRC 1998, and Cambodia),
resources appeared to help fracture rebel or government alliances based on ethnic, religious, or
ideological grounds, and create new alliances based on commercial grounds. It was difficult to
judge whether these fractures influenced the casualty rate; moreover, it is not possible to know if
resource wealth leads to an unusually high rate of alliance fracture without examining alliance
stability in comparable resource-poor conflicts.

35
Oil suspended operations while government troops used aerial bombing, the burning of villages

and summary executions to depopulate a large area around the oilfield. Shortly after

depopulation was completed, Lundin Oil resumed operations [Christian Aid 2001].

Pre-emptive repression only occurred in separatist conflicts, at least within this sample.

This may be because governments are more willing to take repressive measures against peripheral

minority groups than members of the majority population. Perhaps governments are more likely

to expect trouble when resource exploitation occurs in regions with separatist aspirations than

when it occurs in other regions. In either case, the Indonesian and Sudanese conflicts suggest a

final hypothesis:

H15: Resource wealth tends to increase the casualty rate during a separatist civil war by

giving the government an incentive to react to small challenges with unusually harsher

countermeasures.

5. Conclusion

The paper uses a novel, medium-N method to examine the causal links between resource wealth

and civil wars. It presents 11 hypotheses about how resources may influence a conflict; it

specifies their observable implications; and it reports which of these implications can be observed

in a sample of 13 cases.

The method facilitates three types of analysis. First, if the hypothesized causal

mechanisms can be observed in these thirteen cases, they may be considered plausible. Second

and more powerfully, if a purported causal mechanism is not found in these cases, we can infer

that it is unlikely to be valid more generally. Finally, if the hypothesized mechanisms cannot

fully account for influence that resource wealth has on conflict, the cases may be used to develop

new hypotheses for future, out-of-sample testing.

Collectively, these three types of analysis have led to six notable findings.

36
 First, there is good evidence at the case study level that natural resource wealth is

causally linked to civil conflict. This study cannot dismiss the possibility that the natural

resource-civil war correlation is, in part, spurious, or that causality runs in the opposite direction.

Indeed, there is good evidence in at least one case (Angola) that the onset of civil war made the

economy more dependent on resource exports [Minter 1994]. But in these 13 conflicts there is

strong evidence that resource wealth has made conflict more likely to occur, and last longer and

produce more casualties when it does occur.

Second, resource wealth and civil war are not linked by a single mechanism, but a variety

of mechanisms; moreover, these mechanisms can influence a conflict’s onset, duration, and

intensity. No single mechanism appeared in more than nine of the thirteen cases. Moreover, in

12 of the 13 cases, resources had more than one type of effect on conflict. This multiplicity of

causal linkages – and the absence of a single, ubiquitous mechanism – may help account for the

analytical muddle, and contradictory findings, of earlier studies.

Third, resource wealth does not always make existing conflicts worse. While the net

effect of resource wealth on conflict in this sample was harmful, the cases suggest that resources

sometimes have contradictory and even beneficial effects over the course of a civil war.

Resource wealth appeared to bring about a quicker end to two wars. And claims that resource

wealth tends to heighten the intensity of conflict may be only partly correct. Observers often note

that combatants fight for the control of natural resources, and that these battles appear to increase

the war’s overall casualty rate. But resources also lead to battlefield cooperation that may reduce

the casualty rate. In nine of the thirteen cases examined here, combatants fought battles over

resource wealth; in eight of these cases, they also laid down their arms (at other junctures) to

cooperatively exploit these same resources.

Fourth, two of the most widely-cited causal mechanisms, the looting and grievance

mechanisms, do not appear to be valid. There was no evidence in these 13 cases that rebel groups

funded their start-up costs by looting natural resources, or extorting money from resource firms.

37
Nor was there evidence that grievances over insufficiently compensated land expropriation,

environmental degradation, inadequate job opportunities, or labor migration contributed to the

initiation of non-separatist conflicts. This does not imply that such grievances are irrelevant: they

may have contributed to the rise of low-level conflicts and separatist conflicts. But neither of

these mechanisms seem to explain the link between natural resource wealth and non-separatist

civil wars.

The fifth finding is that resources appear to play a different role in separatist conflicts

than non-separatist conflicts. Grievances over the distribution of resource wealth helped initiate

two of the three separatist wars in the sample (Sudan, Indonesia), but played no role in the ten

non-separatist wars, except for the ambiguous case of Sierra Leone. These two separatist

conflicts were also the only ones to face pre-emptive repression, which is a government’s use of

terror to suppress rebel movements that may interfere with resource exploitation. This implies

that the geographical distribution of natural resources across a nation’s territory may be

important: if resource wealth is located in a region with separatist aspirations, it may help

precipitate a war, and increase the war’s casualty rate.

Finally, the paper describes several unforeseen mechanisms that link resource wealth to

subsequent conflict. One mechanism is the sale of booty futures – future exploitation rights to

resources not under the seller’s control. The sale of booty futures is a tool of the weak against the

strong: it can help provide aspiring rebel groups with the funds they need to launch attacks on

governments; it can also give governments on the verge of defeat the ability to fund

counterattacks. In this sample it contributed to the onset of at least two major wars (Sierra Leone

and the Congo Republic) and the prolongation of three (Angola, Sierra Leone, and the DRC II).

The other mechanism is pre-emptive repression. In two of the three separatist conflicts in

the sample (Indonesia and Sudan), the government enacted exceptionally harsh countermeasures

against insurgencies, because they appeared to threaten the government’s control of resource

wealth. These two mechanisms – along with seven of the nine mechanisms whose implications

38
were tested in the sample – can help account for the frequency, and duration, of civil wars in

resource-rich states.

39
Figure 1: Causal Mechanisms That May Link Resource Wealth with Conflict
Hypotheses on the Onset of Civil War

1. Looting by potential rebels Æ start-up costs fundedÆ civil war

2. Resource extraction Æ grievances among locals Æ civil war

3. Resource extraction Æ incentive for separatism Æ civil war

4. Resource extraction Æ warlords Æ state weakened Æ civil war

5. State depends on resource revenues Æ state weakness Æ civil war*

Hypotheses on the Duration of Civil War

6. Looting by weaker (stronger) party Æ more arms Æ war prolonged (shortened)

7. War (peace) appears financially profitable Æ less (more) incentive to sign peace accord Æ
war prolonged (shortened)

8. Looting by either party Æ discipline problems Æ leaders’ peacemaking (warfighting) capacity

undermined Æ war prolonged (shortened)

9. Resource wealth in separatist region Æ commitment problem Æ war prolonged*

Hypotheses on the Intensity of Civil War

10. Two sides engage in resource battles Æ more casualties

11. Two sides engage in cooperative plunder Æ fewer casualties

Deductive Hypotheses

12. Resource wealth Æ foreign intervention Æ civil war*

13. Futures contracts for resource booty Æ start-up costs funded Æ civil war *

14. Weaker (stronger) side sells futures contracts for resource booty Æ war lengthened
(shortened)*

15. Pre-emptive repression by government to protect resources Æ more casualties*

* hypothesis is not tested in the sample

40
 Table 1: Civil Wars Linked to Resource Wealth, 1990-2000
Country Duration Resources
Afghanistan 1978-2001 Gems, opium
Angola 1975-2002 Oil, Diamonds
Burma 1949- Timber, tin, gems, opium
Cambodia 1978-97 Timber, gems
Colombia 1984- Oil, gold, coca
Congo, Rep. 1997 Oil
Congo, Dem. Rep. 1996-97 Copper, coltan, diamonds, gold, cobalt
Congo, Dem. Rep. 1998-2002 Copper, coltan, diamonds, gold, cobalt
Indonesia (Aceh) 1975- Natural gas
Liberia 1989-96 Timber, diamond, iron, palm oil, cocoa,
coffee, marijuana, rubber, gold
Peru 1980-1995 Coca
Sierra Leone 1991-2000 Diamonds
Sudan 1983- Oil
Separatist conflicts are listed in italics.

Table 2: Commodity Exports and Military Spending

Dependent variable is military spending as a fraction of government spending, 1997
1 2 3 4
Oil Exports/GDP .328*** - - -
(.0711)
Mineral Exports/GDP - .304 - -
(.159)
Nonfood Agr - - .763 -
Exports/GDP (.81)
Food Exports/GDP - - - -.621***
(.184)
Log GDP/capita -1.96* -.82 -1.28 -2.03*
(.772) (.799) (.909) (.863)
Adjusted R2 .16 .03 .02 .13
N 130 115 86 86
Tests are OLS cross-sectional regressions run with Stata 7.0.
* significant at the 0.05 level
** significant at the 0.01 level
*** significant at the 0.001 level

41
Table 3: Summary of Findings
Onset Duration Intensity
Afghanistan No Longer None
Angola No Longer Mixed
Burma No Mixed Mixed
Cambodia No Mixed Mixed
Colombia No Longer Mixed
Congo, Rep. Yes Shorter None
Congo, Dem. Rep. I Yes Shorter None
Congo, Dem. Rep. II Yes Longer Mixed
Indonesia Yes None Worse
Liberia No Longer Mixed
Peru No Longer Worse
Sierra Leone Yes Longer Mixed
Sudan Yes Longer Mixed

Table 4: Origins of Conflict

Looting Grievance Separatism Warlord
Afghanistan No No - No
Angola No No - No
Burma No - No No
Cambodia No No - No
Colombia No No - No
Congo, Rep. No No - No
Congo, Dem. Rep. I No No - Yes
Congo, Dem. Rep. II No No - No
Indonesia No - Yes No
Liberia No No - No
Peru No No - No
Sierra Leone No No? - Yes
Sudan No - Yes No

42
Table 5: Duration of Conflict
Looting Incentive Discipline Futures Net Effect
Afghanistan Yes No No No Longer
Angola Yes No No Yes Longer
Burma Yes Yes* No No Mixed
Cambodia Yes No Yes* No Mixed
Colombia Yes No No No Longer
Congo, Rep. No Yes* No No Shorter
Congo, Dem. Rep. I No No No Yes* Shorter
Congo, Dem. Rep. II Yes Yes No Yes Longer
Indonesia No No No No None
Liberia Yes Yes Yes No Longer
Peru Yes No No No Longer
Sierra Leone Yes No No Yes Longer
Sudan Yes No No No Longer
*Made the conflict shorter

Table 6: Intensity of Conflict

Battles Plunder Repress Net Effect
Afghanistan No No No None
Angola Yes Yes No Mixed
Burma Yes Yes No Mixed
Cambodia Yes Yes No Mixed
Colombia Yes Yes No Mixed
Congo, Rep. No No No None
Congo, Dem. Rep. I No No No None
Congo, Dem. Rep. II Yes Yes No Mixed
Indonesia No No Yes Worse
Liberia Yes Yes No Mixed
Peru Yes No No Worse
Sierra Leone Yes Yes No Mixed
Sudan Yes Yes Yes Mixed

43
References

Abdullah, Ibrahim (1998) "Bush path to destruction: the origin and character of the Revolutionary
United Front/Sierra Leone," Journal of Modern African Studies 36 (2), 203-235.

Abdullah Ibrahim and Patrick Muana (1998), “The Revolutionary United Front of Sierra Leone,” in
Christopher Clapham (ed.), African Guerrillas. Bloomington: University of Indiana Press, 172-194.

Accord (1996), “Special Issue on Liberia.” Downloaded from http://www.c-r.org/accord on

September 1, 2001.

Africa Confidential (2001), "Chronology of Sierra Leone: How diamonds fuelled the conflict,"
downloaded on www.Africa-Confidential.com, September 1, 2001.

Alao, Abiodun, John Mackinlay, and 'Funmi Olonisakin (1999), Peacekeepers, Politicians, and
Warlords: The Liberian Peace Process. New York: United Nations University Press.

Amnesty International (2000), "Oil in Sudan - Deteriorating Human Rights," Report AFR
54/01/00ERR, May 3.

Anderson, G. Norman (1999), Sudan in Crisis. Gainesville: University Press of Florida.

Auvinen, J. Y. (1997) "Political Conflict in Less Developed Countries, 1981-89," Journal of Peace
Research 34, 177-195.

Azam, Jean-Paul (2002), “Looting and Conflict between Ethnoregional Groups,” Journal of
Conflict Resolution 46 (1): 131-153.

Beblawi, Hazem (1987), "The Rentier State in the Arab World," in The Rentier State, eds. Hazem
Beblawi and Giacomo Luciani, New York: Croom Helm, 49-62.

Bennett, Andrew (1997), “Lost in the Translation: Big (N) Misinterpretations of Case Study
Research,” Paper presented at the 1997 Annual Meeting of the International Studies Association..

Bennett, D. Scott and Alan C. Stam III (1996) “The Duration of Interstate Wars, 1816-1985,”
American Political Science Review 90(2), 239-257.

Brown, MacAlister and Joseph J. Zasloff (1998), Cambodia Confounds the Peacemakers. Ithaca:
Cornell University Press.

Buhaug, Halvard and Scott Gates (2002) "The Geography of Civil War," Journal of Peace
Research 39 (4), 417-433.

Chaudhry, Kiren (1997), The Price of Wealth: Economies and Institutions in the Middle East.
Ithaca: Cornell University Press.

Christian Aid (2001), "The Scorched Earth." Downloaded from http://www.christian-

aid.org.uk/indepth/0103suda/sudanoil.htm on September 1, 2001.

Collier, David and James Mahoney (1996), “Insights and Pitfalls: Selection Bias in Qualitative
Research,” World Politics 49:56-91.

44
Collier, Paul and Anke Hoeffler (1998) "On economic causes of civil war," Oxford Economic
Papers 50, 563-573.

--- (2001), "Greed and Grievance in Civil War," Manuscript, October 21.

--- (2002), "The Political Economy of Secession," Manuscript, June 30.

Collier, Paul, Anke Hoeffler, and Mans Söderbom (2001), “On the Duration of Civil War.” Paper
presented at the Workshop on Civil Wars and Post-Conflict Transitions, University of California,
Irvine, May 18-20 2001.

Coppedge, Michael, “Explaining Democratic Deterioration in Venezuela Through Nested

Induction,” Paper presented at the 2002 Annual Meeting of the American Political Science
Association, Boston.

Crystal, Jill (1990), Oil and Politics in the Gulf: rulers and merchants in Kuwait and Qatar. New
York: Cambridge University Press.

De Soysa, Indra (2002) "Paradise is a Bazaar? Greed, Creed, and Governance in Civil War, 1989-
99," Journal of Peace Research 39 (4), 395-416.

Doyle, Michael and Nicholas Sambanis (2000) "International Peacebuilding: A Theoretical and
Quantitative Analysis," American Political Science Review 94 (4), 779-801.

Dudley, Ryan and Ross Miller (1998) "Group Rebellion in the 1980s," Journal of Conflict
Resolution 42, 77-96.

Dunning, Thad and Leslie Wirpsa (2002), “Andean Gulf? Oil and the Political Economy of
Conflict in Colombia and Beyond,” Manuscript May 31.

Eckstein, Harry (1975), “Case Studies and Theory in Political Science,” in Fred Greenstein and
Nelson Polsby (eds.), Handbook of Political Science (Vol 7). New York: Addison-Wesley, pp. 79-
138.
Ellis, Stephen (1999), The Mask of Anarchy: The Destruction of Liberia and the Religious
Dimension of an African Civil War. New York: New York University Press.

Fearon, James D (1999), "Weak States, Rough Terrain, and Large-scale Ethnic Violence," Paper
presented at the Annual Meeting of the American Political Science Association, Atlanta, GA,
September 1999.

--- (2002), "Why Do Some Civil Wars Last So Much Longer Than Others?," Manuscript, July 12.

Fearon, James D. and David D. Laitin (2002), "Ethnicity, Insurgency, and Civil War," Manuscript,
May 7.

French, Howard W. (1997) "The Great Gold Rush in Zaire," New York Times (18 April), A6.

Geddes, Barbara (1990), “How the Cases You Choose Affect the Answers You Get: Selection
Bias in Comparative Politics,” Political Analysis 2: 131-152.

45
Gedicks, Al (2001), Resource Rebels: Native Challenges to Oil and Mining Corporations. Boston:
Southend Press.

George, Alexander (1979), “Case Studies and Theory Development,” in Paul Lauren (ed.)
Diplomacy: New Approaches in Theory, History, and Policy. New York: Free Press, pp. 43-68.

Galloy, Martine-Renée and Marc-Éric Gruénai (1997), “Fighting for power in the Congo,” Le
Monde diplomatique, November.

Grossman, Herschel I. (1991), “A General Equilibrium Model of Insurrections,” American

Economic Review 81 (3): 912-921.

Hegre, Havard (2002), "Some Social Requisites of a Democratic Civil Peace: Democracy,
Development, and Armed Conflict," Manuscript, August 23.

Hirsch, John L. (2001), Sierra Leone: Diamonds and the Struggle for Democracy. Boulder: Lynne
Rienner.

Hirschleifer, Jack (1989), “Conflict and Rent-Seeking Success Functions,” Public Choice 63(2):
101-112.

Human Rights Watch (1999), Angola Unravels: The Rise and Fall of the Lusaka Peace Process.
New York: Human Rights Watch.

International Crisis Group (2000), "Scramble for the Congo: Anatomy of an Ugly War," ICG Africa
Report No. 26, December 20, 2000.

--- (2002), God, Oil, and Country: Changing the Logic of War in Sudan. Brussels: IGC Press.

Karl, Terry Lynn (1997), The Paradox of Plenty: Oil Booms and Petro-States. Berkeley: University
of California Press.

Keen, David (1998), "The Economic Functions of Violence in Civil Wars," Adelphi Paper 320,
International Institute of Strategic Studies, London.

Klare, Michael (2001), Resource Wars: the new landscape of global conflict. New York:
Metropolitan Books.

King, Gary, Robert O. Keohane, and Sidney Verba (1994), Designing Social Inquiry: Scientific
Inference in Qualitative Research. Princeton: Princeton University Press.

Le Billon, Philippe (1999), "A Land Cursed by its Wealth? Angola's War Economy (1975-1999),"
UNU/WIDER Series: Work in Progress No. 23.

--- (2000) "The Political Ecology of Transition in Cambodia 1989-1999: War, Peace and Forest
Exploitation," Development and Change 31, 785-805.

--- (2001), “The political ecology of war: natural resources and armed conflicts,” Political
Geography 20, pp. 561-584.

46
Leite, Carlos and Jens Weidmann (1999), "Does Mother Nature Corrupt? Natural Resources,
Corruption, and Economic Growth," IMF Working Paper WP/99/85.

Lieberman, Evan S. (2002), “Seeing Both the Forest and the Trees: Nested Analysis in Cross-
National Research” Paper presented at the 2002 Annual Meeting of the American Political Science
Association, Boston.

Lintner, Bertil (1999), Burma in Revolt: Opium and Insurgency Since 1948. Chaing Mai
(Thailand): Silkworm Books.

Mahdavy, Hussein (1970), "The Patterns and Problems of Economic Development in Rentier
States: The Case of Iran," in Studies in Economic History of the Middle East, ed. M. A. Cook,
London: Oxford University Press, 428-467.

MacGaffey, Janet (1991), The Real Economy of Zaïre. Philadelphia: University of Pennsylvania
Press.

Minter, William (1994), Apartheid's Contras: an inquiry into the roots of war in Angola and
Mozambique. London: Zed Books.

O'Ballance, Edgar (2000), Sudan, Civil War, and Terrorism 1956-1999. New York: St. Martin's.

Onishi, Norimitse (2002), “As Oil Riches Flow, Poor Village Cries Out,” New York Times,
December 22, A1.

Peleman, Johan (2000), “Mining for Serious Trouble: Jean-Raymond Roulle and his Corporate
Empire Project,” in Abdel-Fatau Musah and J. ‘Kayode Fayemi, eds., Mercenaries: An African
Security Dilemma. London: Pluto Press, pp. 155-168.

Ragin, Charles (1994), Constructing Social Research: The Unity and Diversity of Method.
Thousand Oaks: Pine Forge Press.

Reed, William Cyrus (1998), "Guerrillas in the Midst," in African Guerrillas, ed. Christopher
Clapham, Bloomington: Indiana University Press, 134-154.

Reno, William (1998), Warlord Politics and African States. Boulder: Lynne Rienner.

Richards, Paul (1996), Fighting for the Rain Forest: War, Youth, and Resources in Sierra Leone.
Oxford: James Currey.

Robinson, Geoffrey (1998) "Rawan is as Rawan Does: The Origins of Disorder in New Order
Aceh," Indonesia 66 (October), 127-156.

Ross, Michael (2001), Extractive Sectors and the Poor. Washington D.C.: Oxfam America.

Sachs, Jeffrey D. and Andrew M. Warner (1995), "Natural Resource Abundance and Economic
Growth," Development Discussion Paper No. 517a, Harvard Institute for International
Development, Cambridge.

Sherman, Jake (2000), “Profits vs. peace: the clandestine diamond economy of Angola,” Journal of
International Affairs 53:2, 699-719.

47
Skaperdas, Stergios (1992), “Cooperation, Conflict, and Power in the Absence of Property Rights,”
American Economic Review 82 (4): 720-739.

--- (2002), “Warlord Competition,” Journal of Peace Research 39 (4): 435-446.

Stedman, Stephen John (2001), "Implementing Peace Agreements in Civil Wars: Lessons and
Recommendations for Policymakers," International Peace Academy and Center for International
Security and Cooperation, May 2001, New York.

Switzer, Jason (2001), "Armed Conflict and Natural Resources: The Case of the Minerals Sector,"
International Institute for Environment and Development, Minerals, Mining and Sustainable
Development Project, No. 12, July.

UN Conference on Trade and Development (1995), UNCTAD Commodity Yearbook. Geneva:

UNCTAD.

UN Panel of Experts (2000), "Report of the Panel of Experts Appointed Pursuant to Security
Council Resolution 1306 (2000), Paragraph 19, in Relation to Sierra Leone," United Nations
Security Council S/2000/1195, December 20, 2000.

--- (2001), "Report of the Panel of Experts on the Illegal Exploitation of Natural Resources and
Other Forms of Wealth of the Democratic Republic of Congo (S/2001/357)," United Nations
Security Council, April 12, 2001.

--- (2002), “Final report of the Panel of Experts on the Illegal Exploitation of Natural Resources
and Other Forms of Wealth of the Democratic Republic of the Congo,” Security Council
S/2002/1146, October 16, 2002.

World Bank (2001), World Development Indicators, CD-ROM.

48