Natural Resources, Conflict, and Conflict Resolution: Uncovering The Mechanisms

JOURNAL OF
10.1177/0022002705277545
Humphreys / NATURAL
CONFLICT
RESOURCES,
RESOLUTION
CONFLICT, AND CONFLICT RESOLUTION
Natural Resources, Conflict,

and Conflict Resolution
UNCOVERING THE MECHANISMS
MACARTAN HUMPHREYS
Department of Political Science
Columbia University
The interpretation of the resource-conflict link that has become most publicized—the rebel greed
hypothesis—depends on just one of many plausible mechanisms that could underlie a relationship between
resource dependence and violence. The author catalogues a large range of rival possible mechanisms, high-
lights a set of techniques that may be used to identify these mechanisms, and begins to employ these tech-
niques to distinguish between rival accounts of the resource-conflict linkages. The author uses finer natural
resource data than has been used in the past, gathering and presenting new data on oil and diamonds produc-
tion and on oil stocks. The author finds evidence that (1) conflict onset is more responsive to the impacts of
past natural resource production than to the potential for future production, supporting a weak states mecha-
nism rather than a rebel greed mechanism; (2) the impact of natural resources on conflict cannot easily be
attributed entirely to the weak states mechanism, and in particular, the impact of natural resources is inde-
pendent of state strength; (3) the link between primary commodities and conflict is driven in part by agricul-
tural dependence rather than by natural resources more narrowly defined, a finding consistent with a “sparse
networks” mechanism; (4) natural resources are associated with shorter wars, and natural resource wars are
more likely to end with military victory for one side than other wars. This is consistent with evidence that
external actors have incentives to work to bring wars to a close when natural resource supplies are threatened.
The author finds no evidence that resources are associated with particular difficulties in negotiating ends to
conflicts, contrary to arguments that loot-seeking rebels aim to prolong wars.
Keywords: civil war; resource-conflict link; resource dependence
1. INTRODUCTION
In early April 1975, President Tombalbaye of Chad appealed on national radio for
popular vigilance, warning that members of the army were plotting a coup against him.
He explained that if anyone wanted to know why a coup was being plotted, the answer
lay with the oil in the Doba fields in the south. This appeal turned out to be
Tombalbaye’s last public address. On April 13, he was killed during Chad’s first suc-
AUTHOR’S NOTE: My thanks to Nicholas Sambanis, Iain Lustik, Robert Bates, Doron Hadass, Bryan
Graham, David Hecht, Michael Ross, Jim Fearon, Habaye ag Mohamed, and Jeremy Weinstein for generous
help with ideas and with data and to Brendan McSherry for terrific research assistance.
JOURNAL OF CONFLICT RESOLUTION, Vol. 49 No. 4, August 2005 508-537
DOI: 10.1177/0022002705277545
© 2005 Sage Publications
508
Humphreys / NATURAL RESOURCES, CONFLICT, AND CONFLICT RESOLUTION 509
cessful coup d’etat.1 Unfortunately for Tombalbaye, he had had increasingly strained
relations with France, his chief military backer, ever since he allowed U.S. corpora-
tions to prospect for oil in the ex-colony. The U.S. corporation was successful where
French prospectors had failed. And France took umbrage.2 Ever since, oil has had a
striking prominence in the intrigues of Chadian politics. The government of Hissène
Habré is reported to have received U.S. support in exchange for his support of U.S. oil
corporations while politicians and NGOs argue that the present president, Idriss Déby,
previously Habré’s right hand man, was offered military support from France’s Elf if
he would overthrow the Habré régime and give France a stake in the southern oil
fields.3 The oil in the south has also been seen to be fuelling southern ambitions for a
federalist state, if not for outright separation. The leader of the present southern oppo-
sition, Ngarlejey Yorongar, has been incarcerated for accusing the previous southern
leader, Wadal Kamougué, of allowing himself to be co-opted by the government in
exchange for money from Elf.4 In short, in the eyes of Chad’s political leaders, control
of oil revenues has been central to Chadian politics for almost thirty years; it has made
and broken political leaders, has incited violence, and has shaped political agendas.
But what is perhaps most striking about Chadian oil is that, up to 2003, not a single
drop had been pumped.
The role of oil in Chad’s politics illustrates some of the complexity of the linkages
between natural resources and conflict. Contrary to popularized images of resource
conflicts in Africa, oil did not lead to rival warlords establishing resource-funded local
monopolies in Chad—the stories that link resources to conflict here center on gaining
tight control of the state rather than on the creation of chaotic environments. And while
natural resources may produce conflict by leading to shadowy states, the weakness of
Chad’s state structure cannot be attributed to its dependence on oil revenues to buoy it;
until recently none has been earned. Chad’s experience comes closer to what Michael
Ross (2002) terms a “booty futures” story—in which resources matter because reve-
nues can be raised in advance to gain control of them. In such a context, policy pre-
scriptions emanating from conflict research—that rebel financing needs to be cut off
through the infiltration of quasi-criminal trade routes and robust opposition to rogue
states—fall wide of the mark.
The problem is that unless we understand the mechanisms linking resources to con-
flict, the advice of conflict scholars will be of limited use to the policy community. And
unless we test these mechanisms, we will be unsure of the generality of processes we
observe in individual cases, such as those seen in Chad. In this article, I explore the
diversity of mechanisms that may link natural resources to conflict and develop and
employ strategies to identify which mechanisms are likely to be in operation when.
1. This explanation is provided by Mohammet Sally, a leader of FROLINAT, the dominant Chadian
rebel group at the time (interviews, N’djamena, April 2003). The leader of the coup, Wadal Abdelkadar
Kamougué, denies any connection between the coup and Doba oil (interviews, N’djamena, April 2003).
2. On French frustration see Nolutshungu (1996). Jacques Foccart (1999) describes how by 1969 the
involvement of U.S. corporations in Chad was seen as an insult to the prestige of the French army.
3. This claim, emanating apparently from the leader of the opposition, Yorongar, has been argued by
Agir ici -Survie (1999) and Verschave (2000). And according to Verschave, Déby also had at least tacit mili-
tary support from the French troops stationed in N’djamena.
4. Agir ici -Survie (1999) and interviews with Yorongar (N’djamena, April 2003).
510 JOURNAL OF CONFLICT RESOLUTION
The article is organized as follows. Drawing largely on the experiences of Sahelian and
West Africa states as well as work by scholars who have studied a wider set of cases
and identified relevant mechanisms (notably, Le Billon 2001; Snyder 2002; Ross
2002, 2003, 2004a), I catalogue a series of mechanisms that may link natural resources
to conflict onset and conflict duration. With the aim of engaging with the econometric
literature, the task then is to find ways econometrically of differentiating between the
effects of these rival mechanisms. I discuss four strategies for doing so, indicating how
the different strategies may be used for the problems at hand. In the final section I turn
to the data, building on the model developed by Fearon and Laitin (2003); I begin the
task of identifying the work of rival mechanisms on conflict onset and duration and
testing a core set of the relations discussed in previous sections.
2. NATURAL RESOURCES AND CIVIL WARS:

AN EMBARRASSMENT OF MECHANISMS
Highly influential research by Paul Collier and Anke Hoeffler at the World Bank5
suggests that countries whose wealth is largely dependent on the exportation of pri-
mary commodities—a category that includes both agricultural produce and natural
resources—are highly prone to civil violence. In explaining the correlation between
primary commodities and conflict, Collier and Hoeffler argue that conflict may be
explained either by greed or by grievances, such as feelings of ethnic or political
marginalization. They conclude (in large part based on the correlation between pri-
mary commodities and conflict) that to understand the causes of contemporary civil
wars we should forget about political and cultural arguments and focus instead on the
greed of rebels and especially on their trade in natural resources. 6
The problem is that the correlation between commodities and conflict does not
imply either that rebels are greedy or that they finance their campaigns through the
trade in natural resources. The correlation could arise, for example, if conflict, or even
expectations of a conflict, causes other economic activities, such as tourism and manu-
facturing, to cease, leaving only extractive industries to function.7 But even if the rela-
tionship is not so spurious, there are at least six rival families of mechanisms that could
explain the relationship between natural resources and war onset and duration; rebel
greed is just one of them.8 Here are the six.
5. See especially Collier and Hoeffler (2000b).

6. Different versions of this research vary in the extent to which evidence for “greed” simply means
evidence for the economic “opportunity’to fight. Collier and Hoeffler argue, “We test a ‘greed’theory focus-
ing on the ability to finance rebellion, against a ‘grievance’theory focusing on ethnic and religious divisions,
political repression and inequality. We find that greed considerably outperforms grievance.” Collier and
Hoeffler (2000b) suggest furthermore that the relevant “opportunity” is in fact the opportunity for predation,
arguing, “Our model suggests that what is actually happening is that opportunities for primary commodity
predation cause conflict.” For a recent statement of this research position, see Collier (2003).
7. See Bannon and Collier (2003) and Collier’s (1999) study of the impact of conflict on Uganda’s
economic structure.
8. For an excellent discussion of a number of these mechanisms, as well as case study evidence sup-
porting them, see Ross (2004a).
The greedy rebels mechanism, with three variants. The first, emphasized by Collier
and Hoeffler, is that domestic groups may engage in quasi-criminal activity to benefit
from resources independent from the state. The second, argued by Fearon and Laitin
(2003), is that natural resources increase the “prize” value of capturing the state. The
first variant should lead to the local expulsion of the state, as in Colombia; the second,
to bids to gain state control, as in Chad or, as argued by Engelbert and Ron (2004), the
Republic of Congo. Plausibly the first variant may lead to the second, or vice versa, as
in Sierra Leone, where control of the diamond areas sufficiently weakened the state as
to make state capture appear easy. In a third variant, if natural resources are concen-
trated in a particular region of a country, this may ground beliefs among dissatisfied
groups that a seceding state could be viable or even prosperous.9 As with the feasibility
mechanism discussed below, the greedy rebels mechanism does not require that rebels
control resources directly; it may be sufficient to extract rents from those who do, as
has been done with oil extractors in Colombia, Cabinda, and Nigeria.
The greedy outsiders mechanism. Rather than resulting from the greed of rebels, as
emphasized by recent literatures, the existence of natural resources may be an incen-
tive for third parties—states and corporations—to engage in or indeed foster civil con-
flicts. Hence, for example, the escalation of the civil war in the Democratic Republic of
Congo has resulted in part from the involvement of neighboring states seeking raw
materials (Dashwood 2000; Meldrum 2000; Willum 2001). The secessionist bid in
Katanga in Congo was supported if not instigated by the Belgian firm Union Minière
du Haut Katanga. And evidence suggests that the French oil corporation Elf took
actions that led to an escalation of the conflict in the Republic of Congo (Verschave
2000).
The grievance mechanism. Natural resource dependence could in fact be associated

with grievances rather than greed.10 There are at least four variants of this mechanism.
First, countries with middling levels of dependence on natural resources may be expe-
riencing transitory inequality as part of the development process.11 Second, economies
that are dependent on natural resources may be more vulnerable to terms of trade
shocks. These could cause instability and dissatisfaction within groups that suffer
from the shocks (in this case, the problem is not with dependence on natural resources
per se but that natural-resource-dependent economies are likely to be dependent on a
small number of commodities for their export earnings).12 Third, the process of extrac-
tion may produce grievances, for example, through forced migration. Ross, for
9. Such arguments have been made for the cases of Biafra in Nigeria, Katanga in Congo, Cabinda in
Angola, Casamance in Senegal, Bougainville in Papua New Guinea, and southern Sudan. See also Ross
(2003).
10. The introduction of the terms greed and grievance is unfortunate, not least because the distinction
between them appears to be a moral rather than a positive one.
11. This may follow for example from the Kuznets curve hypothesis that predicts transitory inequality
resulting simply from the fact that different parts of an economy may develop at different rates.
12. Vulnerability will also be more likely if the risks associated with the commodities are highly corre-
lated—either in terms of price fluctuations or in terms of production conditions, such as the weather. There is
no reason to expect that an economy exporting a diversified portfolio of natural resources will be particularly
susceptible to income shocks.
example, describes externalities of the extraction process itself in Aceh and Papua
New Guinea such as environmental damage and loss of land rights. Or finally, natural
resource wealth may be seen as more unjustly distributed than other wealth—as has
been claimed in Sierra Leone and Nigeria. In Niger, the insurgent groups stressed not
just that the north received little investment from the political center in the south but
also that the south relied economically on revenues gained from the uranium wealth of
the north, with no visible returns to the north. Such are the fears presently in Chad:
with no expectation that any of the oil revenues will accrue to their region, local leaders
in the Doba area have petitioned to the oil corporations for direct compensation, in the
form of scooters for each village leader.13
The feasibility mechanism. Natural resources could provide a way to finance rebel-
lions that have been started for other reasons, thereby increasing the prospects of suc-
cess.14 This can occur either through control of production during conflict, or, in prin-
ciple, through the sales of booty futures. Insofar as natural resource dependence
matters through feasibility effects, it is a “permissive cause” rather than a “root cause”
of conflict. Some scholars argue that because motivations for conflict are ubiquitous
only permissive causes of this form matter, nevertheless insofar as there is variation in
motivation, the feasibility explanation implies that there is a need to take account of
root causes when responding to conflicts. In principle, there should also be observable
differences in the conduct of wars, and of negotiations, between those that are con-
ducted to access resources and those that are financed by resources but conducted to
achieve other goals (although of course motivations may change over time).
The weak states mechanism. State structures may be weaker in natural-resource-

dependent economies. There are two prominent variants of the argument, both of
which focus on the strength of state-society linkages. One variant focuses on the soci-
ety side of weak society-state relations and holds that when citizens are untaxed by
governments, they have less power over them: they may have less information about
government activity, weaker incentives to monitor government behavior, and fewer
instruments at their disposal to withdraw support from governments; accordingly,
resource-dependent states may have little compulsion to respond to the demands of
their citizens or create structures that engage their citizens. The implications of these
arguments are ambiguous insofar as they may result in greater insulation but less resil-
ience of the state. The second channel focuses on the state side of state-society rela-
tions, arguing that governments that rely on natural resources rather than taxation have
weakened incentives to create strong bureaucratic institutions.15 The argument is a
13. Formally the Chad deal provides for 5 percent of revenues to be spent in the region of oil extraction
in the south. Few in the south believe that the government, dominated by northerners, will implement this
deal (interviews with village chiefs, Doba, April 2003).
14. As noted above, this interpretation of resources providing “opportunity” (as opposed to motivation)
is indeed a “softer” interpretation of the result that is sometimes suggested by Collier and Hoeffler.
15. Although, as argued by Snyder and Bhavnani (2005 [this issue]), the degree to which resources lead
to a weakened capacity for tax raising is partly endogenous to policy choices.
long-standing one for oil-dependent states and has been recently stressed by Fearon
and Laitin (2003), who argue that oil states are more likely to have weak structures
because they have less need for intrusive bureaucracies to raise revenue.16 The result
may be a state such as Mobutu’s Zaire that is divorced from the domestic economy.
The sparse networks mechanism. Plausibly, the importance of natural resources

may lie in their impacts on the daily economic activities of the citizens of an economy
and how these in turn affect attitudes of citizens or relations between citizens. One way
in which this more indirect route could function is as follows. Natural-resource-
dependent economies may have weak manufacturing sectors—an effect exacerbated
by “Dutch Disease” dynamics17—and correspondingly low levels of internal trade.
Insofar as internal trade is associated with greater levels of social cohesion and interre-
gional interdependence, the weakness of the manufacturing sector and the fragmenta-
tion of an economy into independent enclaves of production may raise conflict risks.
The argument that dense trade networks reduce conflict risks is already well estab-
lished in the study of international conflicts. Liberal theorists argue that where trade is
mutually beneficial, to fight with a trading partner would be to commit “commercial
suicide.”18 Related arguments claim that, through exchange, trading partners develop
greater understanding for each others’ cultures. Political philosophers, meanwhile,
suggest that trade reduces the risk of conflict because trade alters cultures: that there is
something about trade that makes people less violent.19 Empirical research demon-
strates that once proximity is taken into account, states that trade with each other are
less likely to fight each other.20 However, good cross-national measures of the density
of internal trade or the sparseness of internal networks do not exist, and so this
hypothesis, tested at the international level, is untested at the within-country level.
There are then many possible mechanisms underlying the relationship between
natural resources and conflict that need not imply that civil war is typically a result of
greed. Different mechanisms require different sorts of policy responses. But knowing
what response is most appropriate requires thinking more carefully about ways to dis-
tinguish the workings of these different mechanisms.
16. See also Moore (2001) on the role of “unearned state income” on political development and Sørli
(2002) on the relationship between oil and “rentier” states.
17. “Dutch Disease” describes the effect of a rise in the price of nontradables relative to tradables that
adversely impacts on nonboom exporting sectors. The effect may also lower growth if manufactured exports
are more growth-enhancing than nontradables. Since growth is negatively associated with conflict, it could
be that natural resources effect conflict via their impact on growth.
18. Angell (1933).
19. The classic statement is that “wherever there is commerce, manners are gentle” (Montesquieu
1749, quoted in Hirschman 1982). Recent experimental evidence also suggests that market relations are
associated with greater “fair-mindedness.” See Ensminger (2001).
20. See Oneal and Russett (1999), Russett (2002), and Doyle (1997) for explanations of the source of
the dispute between liberals and realists and evidence that, when variables such as geographic proximity are
controlled for, trade reduces conflict.
3. NATURAL RESOURCES AND DURATION
There are also multiple avenues through which resources may affect duration. I
point to seven families of such mechanisms.
The feasibility mechanism. Plausibly natural resource financing creates longer

wars by enabling rebel groups to keep fighting. The more general statement is Napo-
leon’s: an army marches on its belly. In terms of military feasibility, it might not matter
much whether what is in the belly comes from trade in natural resources, local abun-
dance of food supplies, extortion, subscriptions, overseas aid, or production by the
troops. In any case, the marginal effect of local natural resource endowments may be to
increase duration.
The duration of the Angola conflict is in part explained by the fact that both sides
had access to natural resource financing to support their combatants. Across Sahelian
states, too, there is evidence for the idea. The wars in Senegal and Sudan have been
long in duration, and rebel groups have been able to benefit directly from the
resources—both commodities and nontraded foodstuffs—around them. Conflicts in
Niger and Mali have been more difficult to sustain. The regions occupied by insurgents
in these countries have not been wealthy enough to support a protracted struggle. 21
The military balances mechanism. The financing available to protagonists is of use

not just for keeping combatants alive and fighting but for protecting assets and inflict-
ing damages on the opposing side: these are effects that influence the ease of victory
and may also influence the ease of negotiated settlements. Insofar as the link between
natural resources and conflict operates through this mechanism, we do not expect a
monotonic relationship between the assets available to any one side and duration. The
expectation from the quantitative study of international wars is that wars with financial
asymmetries are likely to last less long: balance makes for longer wars. In this case, the
marginal impact of quantities of natural resources on duration depends on whether
they tend to produce more symmetric or less symmetric forces. One implication of this
logic is that changes in international policy and actions on conflict financing (such as
the regulation of the diamond trade) can increase conflict duration if they lead to
greater symmetry between combatant sides.
The theory and evidence from negotiated settlements to end civil wars is less clear,
although researchers (e.g., Zartman 2000) suggest that balance facilitates negotiated
settlement. The opposite argument can however also be made: that asymmetries
reduce the uncertainty around the outcome of conflict, leading to a lower likelihood of
bargaining failures.
Michael Ross (2002) suggests that “booty futures” financing—financing to secure
assets that can be gained after a war ends—can be associated with longer wars. It is not
21. A Sahelian counterexample to the logic is provided by Chad. The FROLINAT rebellion, lasting
from 1966 to around 1980, was protracted but took place almost exclusively in a resource poor part of the
country, in the B.E.T. region of the north. The reason that FROLINAT could keep going when comparable
rebellions in Mali and Niger could not is that FROLINAT had access to prolonged foreign support, in this
case from Libya.
clear why this should be so, however: insofar as the mechanism requires the ability to
make credible commitments, we may expect it to work to the advantage of one side
only—the government side (that is, the side that already has rights that it can sell and
that can, in principle, be enforced with the help of the international system) and
thereby lead to asymmetries and possibly shorter wars. In any case, whichever side is
financed by booty futures, if benefits accrue only after a conflict ends, we would
expect financiers to act only when they expect wars to be short, or to act to ensure that
wars are short.
The fragmented organizational structures mechanism. A new and fruitful line of

research argues that that the extent to which the benefits from conflict require joint
production may affect rebel organizational structures. Le Billon (2001) focuses on the
centralization of the production process—with concentrated resources, such as oil,
requiring more organizational cohesion and allowing for more hierarchical organiza-
tional structures than diffuse resources such as cattle. As argued by Ross (2003), goods
also vary in how they can be marketed, and this, too, can affect group cohesion.22
Weinstein (2005 [this issue]) focuses on benefits that can be targeted to individuals in
the short run and those that accrue only in the long term and are conditional on the
success of the rebellion.
These organizational effects may matter for war duration. Researchers have sug-
gested, for example, that we should expect a positive relationship between the cohe-
siveness of a rebel organization23 or its degree of hierarchy24 and the duration of a con-
flict. The experiences of Senegal and Mali, however, both suggest that if anything the
lack of cohesiveness leads to longer conflicts. While cohesiveness may improve the
fighting capacity of a group and thereby delay any military victory over the group by
the government, in a context where military victory is unlikely, cohesiveness may
instead lead to an improved ability to reach a negotiated settlement. Conversely, the
lack of cohesion can prevent effective negotiation by preventing the formulation of a
coherent ideology or set of demands as well as by resulting in an inability on the part of
rebel groups to convince the state that they can deliver what they offer.25
22. Some goods, particularly bulky goods destined for export to a well-structured international market,
will have more centralized marketing channels than less bulky goods sold in less structured markets. These
qualities can be described as variations in a resource’s “obstructability.”
23. Collier, Hoeffler, and Söderbom (2001), for example, argue that “[many] rebel organizations face
severe problems of maintaining cohesion: hence the much shorter duration of such wars.” As suggested by
Nicholas Sambanis in comments, a useful distinction may be drawn between the ability of leaders of a given
group to enforce orders and the existence of multiple factions, each with its own leadership structures. Lack
of the former type of cohesion may make military victory for the government more likely, while lack of the
latter may prevent negotiated resolution.
24. More hierarchical structures may lead to longer wars because the leadership is less likely to suffer
personally from the costs of the conflict and is more likely to gain a large share of benefits. However, if a set-
tlement can be negotiated that benefits the leadership, more hierarchical organizations may be better able to
guarantee the adherence of the organization to the terms of the settlement.
25. While the literature on bargaining (e.g., Schelling 1960) suggests that fragmentation, by producing
limited mandates, may strengthen a negotiator, this logic only holds when the limits are within the bargaining
set. If placed outside the bargaining set, the bargainer may be seen as being unable to deliver any deal.
The possibility of pork mechanism. A related argument focuses on peace negotia-

tions. The argument suggests that having rents to divide among faction leaders greases
the wheels of a negotiation process; if rents can be provided to multiple sides, the argu-
ment goes, their adherence will be more reliable. Englebert and Ron (2004) argue, for
example, that the availability of oil monies facilitated adherence to peace agreements
in the Republic of Congo. Related arguments have been made to encourage the timely
allocation of aid monies following peace negotiations; similar arguments have been
made to explain coalition formation in pork politics in the United States (consider, for
example, Lohmann and O’Halloran’s [1994] model of the “universal log-roll”). The
argument should be most applicable in situations where allocations of pork are condi-
tional upon a return to peace—for example, if an oil industry is incapable of
functioning during times of conflict but can restart afterwards.
This logic of these arguments suggests that if resource exploitation depends on
peace, then the presence of natural resource endowments should make negotiation
more likely to succeed. However, cooperative game theorists might expect pork poli-
tics incentives to work in just the opposite direction. In the context of distributive poli-
tics, these theorists predict that if any coalition reaches an agreement, new coalitions
will typically be able to form that will have an incentive to overturn that agreement.26
This suggests that credible commitments should be especially difficult to achieve in
negotiations: the presence of natural resources that yield transferable rents should
make negotiation more difficult by providing subgroups with incentives to renegotiate
ex-post.
The domestic conflict premium mechanism. Groups that benefit during conflict
may prefer to fight than to win and therefore act as spoilers to peace processes. If, as
argued by Keen (1998) and Collier (2000c), natural resource endowments are associ-
ated more with greed-inspired rebellions, then the fighters in these conflicts may not
have an interest in the success of negotiations. Weinstein (2005), though not directed
to the question of conflict duration, nonetheless adds a new element to such domestic
conflict premium arguments. The nature of the resources available to a group, he
argues, can structure the characteristics of a group’s membership. Groups with natural
resource wealth may be more likely to attract “consumers” that benefit from the
rewards that take place during conflict and less capable of attracting “investors” who
may be driven by benefits that are realized only after successful collective action.
A caveat to these arguments is necessary. To provide the link between the benefits
of war and a conflict’s duration, we need to know not just that individuals benefit in
wartime but that they believe that they benefit more than they would in times of peace
(see, for example, Collier 2000a). The real puzzle is, What prevents parties from
agreeing on a peaceful arrangement that leaves everyone better off? Typical answers to
this question focus on the ability of agents to make credible commitments to each other
to honor agreements made in war time. Another possibility is a feasibility constraint
on negotiators: individuals may do well out of war because they are engaged in illegal
26. More formally, in noncollegial distributive games, the set of equilibrium points—the core—is
generically empty.
activities that they would not be able to undertake during peacetime; if so, compensat-
ing protagonists after a conflict may not be financially feasible for a state that rejects
future trade in these commodities.
This, it seems, is what leads to the special relationship between illegal economic
activities, the illegal drug trade in particular, and protracted conflicts. Over the course
of the Casamance conflict, for example, rebels have increasingly become reliant on
natural resources—notably cannabis, cashew nuts, and timber. Control of these indus-
tries would likely shift in the absence of the conflict and appears to have rendered pro-
ductive negotiations difficult. More generally, groups that get financing from contra-
band fight in longer wars (Fearon 2004). In these instances, unless states are willing to
turn a blind eye to trade in illegal goods by protagonists of conflict, settlements will be
difficult to achieve. In the absence of alternative forms of compensation, peace in such
conflicts then may require victory rather than negotiation.
The international conflict premium mechanism. Insofar as third parties can bring
pressure to bear on the resolution of conflicts, their incentives can help determine the
duration of conflicts. Neighboring states can provide sanctuary for rebel bases within
their borders, and they can provide logistical support to one side or another. They can
also facilitate mediation. Which of these they decide to do may depend again on the
optimal benefits they can expect to achieve during wartime relative to those they could
gain in a negotiated settlement, a feature that can often depend on the resource
endowments of their neighbors.
The experience of Sahelian states is again rich in this regard. The greatest interna-
tional influence on the duration of the Mali conflict was probably that exerted by Alge-
ria. The Mali conflict took place in a region bordering southern Algeria—one home to
Berber populations living in similar conditions to those of Tuareg groups in Mali.
Algeria had security reasons not to want to see the conflict drag on; but it also failed to
gain economically from the conflict. It used its control over supply routes and over
Tuareg exiles and refugees in Algeria to place pressure on the rebels. And as a major
supplier of oil as well as military and economic aid, it placed pressure on Bamako.
Countries neighboring Senegal did not have the same fears of a spread of the conflict,
and neither Guinea-Bissau nor Gambia had strong motivations to bring the conflict to
an end. Indeed, both countries have benefited from the war economy associated with
the conflict—Guinea-Bissau, through the routing of cashew exports through the zone
and by acting as a market to areas more isolated from Senegalese markets; and Gam-
bia, through its involvement with the routing of cannabis and timber exports through
the country (Evans 2002). As a result, not only have the rebels not come under pressure
from these sources, but they have benefited financially and militarily from their
relations with them.
The sparse networks mechanism. An analogy of the sparse networks mechanism

for conflict onset may help explain conflict duration. Just as dense linkages within
economies may prevent a conflict from breaking out, so may they help resolve them.
Such an argument has been made for the Sierra Leone war by Paul Richards (1996),
who suggests that aid enclaves may have hindered conflict resolution by limiting the
extent of economic exchanges.27 An illustration of the logic is also provided by the

conflict in Mali: of two rebel groups, ARLA and FPLA, for only one of them, the
FPLA, did the members have dense economic exchanges with the Songhoi commu-
nity in the north. The suspension of these exchanges during the conflict was expensive
for both the Tuareg fighters in the FPLA and the members of the Songhoi civil defense
militias; the desire to reestablish them led to a successful series of negotiations
between these groups while the ARLA continued fighting until ultimately they were
defeated.
4. MECHANISMS AND ESTIMATION
With so many possible mechanisms, is it possible to differentiate between them?

And is it possible to do so in a way that can engage with the econometric literature?
The challenge seems particularly daunting when multiple mechanisms may well be in
operation at the same time. One view is that econometric work can only produce corre-
lations and that qualitative work is required to identify the mechanisms that underlie
these correlations. In this section, I argue that this view is based on a somewhat false
distinction between qualitative and quantitative work; I argue that it is indeed possible
to use econometric methods to help differentiate between multiple mechanisms, and I
suggest four relatively standard techniques that can be employed to do so.
To describe the techniques and their domains of application, I will assume that the
outcome of interest, y, is binary: hence, in any particular case i, the event did or did not
occur. Since law-like generalizations are typically not attainable in social science, we
can treat y as a random variable distributed according to some density function, such as
the Bernoulli function, conditional upon some set of explanatory variables. The aim
then is to estimate the properties of the function in a way that can distinguish between
rival mechanisms whose effects may easily be conflated.
I proceed by discussing two different ways—referred to as systems with “Type B”
and “Type A” mechanisms by Elster (1998)—in which rival mechanisms may relate to
each other. For each of these, I identify two simple “fixes”—in all cases techniques
that already exist—that can be used to help distinguish between mechanisms when
multiple mechanisms exist.
SYSTEMS WITH TYPE B MECHANISMS
Following Elster (1998), a system with Type B mechanisms is one in which multi-
ple mechanisms may work simultaneously, possibly with opposite effects.28 In such
situations, we may incorrectly infer from a simple correlation that an independent
variable has no effect on a process, even though it has multiple effects, possibly work-
27. Richards (1996) argues that this mechanism was plausibly at work in the Sierra Leone case.
28. Elster (1998) focuses on cases in which effects work in opposite directions. When effects work in
the same direction, there will not be ambiguity about the direction of the net impact of the independent vari-
able, but distinguishing between mechanisms may nonetheless be important if responses to different mecha-
nisms differ.
ing in different directions. The challenge here is to identify the opposing effects.
Should multiple mechanisms work in the same direction, the challenge is to assess the
different contribution of the different mechanisms.
The simplest econometric fixes for such indeterminacy problems arising from Type
B mechanisms use more fine-grained data. Two approaches stand out.
Disaggregating the Explanatory Variable

The first, and most obvious, approach is to identify finer data that can distinguish
between rival stories. When the relationship being studied is as general as that between
natural resources and conflict, econometricians dispose of a degree of discretion in
selecting what measures to use. In practice, the choice of measure may be driven by
mundane concerns of data availability. Sometimes it is also driven by the more worry-
ing concern of which measure produces the “best” results, where the metric for best is
too often the size of the t-statistic. With a focus on mechanisms, however, the selection
can be determined by the ability of different measures to identify divergent observable
implications of rival mechanisms. If multiple mechanisms are simultaneously in oper-
ation, and each has an independent effect on the outcome, then multiple measures may
be able to capture the effects of these rival mechanisms. Let me illustrate the point by
focusing on rival mechanisms used to explain the relation between national oil sales
and conflict.
The weak states mechanism described above is based on the idea that states that
depend on natural resources for their revenues rather than on taxation have weak state
structures and hence are less capable of mustering support against a threat. A rival
mechanism—the greedy rebels mechanism—focuses instead on the idea that control
of the state becomes more valuable to men with guns when there are oil revenues in the
bargain. Similarly, the feasibility and the greedy outsiders mechanisms suggest that an
important determinant of conflict is the financing that can be raised on the strength of
future production (“booty futures” financing). Fearon and Laitin (2003) recognize
both the weak states mechanism and the rebel greed mechanism, but they cannot dis-
tinguish between these two on the basis of a single correlation.29 Nonetheless, insofar
as they affect different aspects of governance, these different mechanisms have very
different implications for policy.
Qualitative differences in the mechanisms, however, can form the basis for quanti-
tative tests to distinguish between them. The first mechanism works through the sup-
ply side of tax revenues—where the revenues come from—the second set of mecha-
nisms work through the demand side—who benefits from the revenues. The first
mechanism is also more backwards looking, the latter set more forward looking. In
principle, the two sets of mechanisms can be distinguished: the first suggests a correla-
tion between conflict and the past size of revenues from oil relative to tax revenue (or
better, some direct measure of bureaucratic depth); the latter suggests a correlation
29. The measure used by Fearon and Laitin (2003)—an indicator of whether oil exports constitute at
least a third or all exports—is not tailored to individuate the effect of the two mechanisms described by the
authors.
between conflict and future discretionary government income. If measures of these

characteristics of economies are not too highly correlated with each other, multivariate
analysis should be able to distinguish between their different effects. In particular,
should a measure of petroleum reserves have an effect on conflict risks, or conflict
duration, after controlling for actual production, then this would provide strong evi-
dence for the greedy rebel and booty futures ideas: in particular, the booty futures
argument would be able to explain a correlation that neither the “weak state”
mechanism nor the grievance mechanism is incapable of explaining.
Disaggregating the Dependent Variable

Consider a situation in which it can be observed, after the fact, which paths have
dominated. In this case, the possibly inconsistent linkages between the explanatory
variables and the dependent variable of interest can be identified by generating a more
fine-grained dependent variable that records the paths that have dominated and then
employing standard techniques for studying multicategory dependent variables.
Depending on how the processes relate to each other, methods such as multinomial
logit, conditional logit, or multivariate30 probit techniques may be appropriate.31 A
related approach can be used even if it cannot, in all cases, be observed after the fact
which paths have dominated. In particular, Gordon and Smith (2004) present an ele-
gant model in which there are multiple mechanisms that cannot be observed but for
which, in some cases at least, it is clear whether one or other path led to an outcome.
Consider now an illustration of a case in which only one pathway is pivotal and, ex
post, that pathway is observable. I noted above that there is ambiguity in the expected
impact of natural resources (x) on conflict termination (y) via its influences on the
cohesiveness of rebel groups. Cohesion may have military implications for rebels,
allowing them to fight for longer and reducing the probability of a military defeat,
leading, plausibly, to longer wars. But it may also allow rebels to negotiate more effec-
tively, increasing the probability of attaining a negotiated settlement. Disaggregating
these effects may be achieved by using a more fine-grained variable y* that records
whether a conflict persists, whether it ends through a military victory, or whether it
ends through negotiation. If we use data with sufficient variation in these outcomes,
we can then use the techniques noted above to work out what the impact of resources
are on the likelihood of military victory and, separately, on the likelihood of negotiated
settlements. In effect, by estimating its impact on a more finely dependent variable, we
identify rival channels linking an independent variable to a more coarsely defined
dependent variable.
30. A multivariate probit approach may be appropriate if it the operation of multiple mechanisms can
ex post be recognized and, hence, the dependent variable can be disaggregated into a series of distinct suc-
cess measures along different paths.
31. An assumption implied by using the standard multinomial logit in this context is that the relative
probability of a negotiated settlement or a continuation of the war is independent of the probability of a mili-
tary victory. This is consistent with a view that there are separate diplomatic and military channels working
more or less independently, but it is more problematic if increases in the chances of a military victory for one
side are likely to reduce (or increase) the efficacy of negotiation. The plausibility of the model can be tested
using either Hausman or Small-Hsiao tests.
SYSTEMS WITH TYPE A MECHANISMS
Again following Elster (1998), assume that there are two possible processes that
link some independent variable x to the outcome variable y, but that for any observa-
tion, only one of these two mechanisms applies In this context, Elster refers to a “Type
A” mechanism problem as one where “the indeterminacy concerns which (if any) of
several causal chains will be triggered.” We can consider two cases, one in which the
process that determines which mechanism will operate is known, and the second in
which it is stochastic or unknown.
Type A Mechanisms Where the Sorting Procedure Is Known

If the process for determining which path is selected is known, then the problem
posed by Type A mechanisms is not difficult. For example, assume that the impact of x
on y depends on some third variable w, then it is sufficient to specify this relationship in
the likelihood function and proceed as usual. For linear specifications this is most sim-
ply done by introducing an interactive term, w ´ x, in the list of regressors. More gener-
ally, appropriate functional forms can be specified directly depending on the way w is
believed to matter.
As an example, consider the following simplified version of the argument that the
impact of natural resources depends on institutions (Snyder 2002; Smith 2004). Let x
denote natural resource endowments and let w denote an indicator variable for strong
states. In a linear framework, one formalization of the argument that natural resources
induce chaos if and only if state structures are weak is given by the hypothesis that the
coefficient on x is positive and is equal in magnitude but opposite in sign to the coeffi-
cient on w ´ x, and hence, that if w = 1, the marginal effect of x is uniformly zero, but if
w = 0, then the marginal effect is positive. Should natural resources reduce the risk of
conflict if state structures are strong but increase it if they are weak, then we expect the
coefficient on x to be positive and smaller in magnitude but opposite in sign to the coef-
ficient on w ´ x.
Type A Mechanisms Where the Sorting Procedure Is Not Known

If, however, the selection mechanism is unknown, then it may still be possible to
write the outcome as a result of two or more rival functional forms where the actual
selection of the functional form is governed by a stochastic process.32 For this type of
problem, the method of “switching regressions” may be used to determine the individ-
ual characteristics of each of the rival processes plus the properties of the “switching
equation.” The switching equation—which determines which process is relevant for
32. For example, we could have
ì y i ~ f ( X )with probability q ( X )
í
î y i ~ g( X )with probability1 - q ( X )
In this case, the expected value of y may be written in a single equation as
y = q(X)f(X) + [1 – q(X)]g(X).
any given case—can itself be a function of a series of explanatory variables. This

allows us to determine, first, the conditions under which going down one path is more
likely than going down the other, and second, the properties of each of the two paths.
We have then multiple and fairly standard ways in which we can test the more
“thick” explanations that arise from qualitative work without losing the generality that
quantitative methods can help deliver. In the next section, I explore whether these tech-
niques can be effective in helping to understand the relation between resources and
conflict.
5. RESULTS: MECHANISM IDENTIFICATION
In this section, I begin to use the techniques described above to identify or exclude
individual mechanisms that may underlie the basic relation between resource endow-
ments and conflict.
DATA
The family of possible measures that could be used to capture the abstract notion of
natural resource abundance is large. The most common measure that has been used is
the value of primary commodity exports as a share of GDP or of total exports. Alterna-
tives include stocks of natural resources, or measures of sales or stocks of different
types of commodities; and many different normalizations of each type of measure can
also be used.
For some of the mechanisms discussed in the literature, there is a concern that mea-
sures used to identify these mechanisms do not measure what we think they measure.
For example, in motivating their research and interpreting their result, Collier and
Hoeffler focus on resources such as diamonds and drugs. Yet these commodities are
unlikely to be captured by the measure they employ—the share of primary commodity
exports in GDP. Illegal commodities are certainly excluded and diamond flows are
also likely not to figure in official data, at least when states are weak. The Collier and
Hoeffler measure and Fearon and Laitin’s (2003) oil measure also include reexports—
primary commodities that are shipped through the country but not necessarily pro-
duced within the country. Hence, in Collier and Hoeffler’s data, Singapore appears as
one of the most natural-resource-dependent economies, while Sudan and Burma fea-
ture as countries with among the lowest levels of dependence on natural resources.
Such reexports bear no relation to the stories provided by Collier-Hoeffler and Fearon
and Laitin.
Even ignoring these issues, there is a problem that data that has been used is not suf-
ficiently fine to distinguish between those mechanisms for which effects operate
through the impacts of resources on the desires and calculations of a small number of
political actors, as suggested by the greedy rebels mechanism or the feasibility mecha-
nism, and mechanisms that function through deeper economic and social structures
such as suggested by the sparse networks mechanism.
As a first cut at responding to these shortcomings, I collected new data on the arche-
typical lootable resource: diamonds. The diamonds data is taken from the Mining
Annual Review (various years), the Metals and Minerals Annual Review (various
years), and the Diamond Registry (based on U.S. Geological Survey data). The advan-
tage of this measure is, first, that it is more fine grain than export data presently
employed, but second, that the sources do not rely only on export data but also on
information gathered from actors in the industry and information provided by mining
corporations, in particular the sources attempt to provide estimates of total diamond
production, including diamonds that are exported clandestinely. The measures of dia-
monds differ from those used by Lujala, Gleditsch, and Gilmore (2005 [this issue])
insofar as they do not contain either information regarding where in the country the
diamonds are mined or whether they derive from alluvial or kimberlite sources; they
have an advantage, however, in that they record quantities mined and not simply the
existence or number of mining sites.
I then collected a measure of the level of production and proven reserves of oil—a
less lootable resource but one whose benefits are still subject to capture by small num-
ber of actors. The measure of oil production records the average amount of oil
extracted per day in a given year, measured in millions of barrels per day. The oil data is
derived from measures reported in the BP Statistical Review of World Energy/BP Sta-
tistical Review of the World Oil Industry (various years), PennWell Corporation’s Oil
& Gas Journal, the U.S. Department of Energy, the OPEC Bulletin, and Petroleum
Economist. In cases where there are differences in a single source between contempo-
raneously published figures and later figures, I use the latest figures under the assump-
tion that these correct for past reporting errors and may better reflect private informa-
tion at the time. In cases where there are differences between sources, I use a simple
average of the estimates of the multiple sources.33
One important difference between this measure and other measures used in the lit-
erature, such as those used by Fearon and Laitin (2003) and Ross (2003), is that it does
not include oil reexports and so allows us to distinguish between extraction, which
involves large rents, and the more industrial oil processing sector. Unlike the Fearon
and Laitin variable, the measure used here is continuous.
The measure of estimated reserves is recorded in billions of barrels and is some-
what vaguely defined as “the volume of oil remaining in the ground that geological and
engineering information indicate with reasonable certainty to be recoverable from
known reservoirs under existing economic and operating conditions.” The oil reserves
data is derived from the same sources as listed above. The correlation between the pro-
duction and reserves measures is .65 while these two have a correlation with measure
33. After interpolation, if no source reports known reserves, then it assumed that that known reserves
are 0. Problematic cases with known production but no reported reserves include Indonesia (1973-1980),
Nigeria (1971-1980), Russia (1993-1997). Dropping these cases results in a loss of about 5 percent of the
observations in Table 1 but leaves the core results on the difference between the effects of production and the
effects of reserves unaltered; dropping these cases in Table 2 results in loss in observations of more than 20
percent and does alter the results, leaving only reserves significant in model III and both reserves and produc-
tion significant in model VI. The results in Table 4 become stronger when these cases are dropped.
Fearon and Laitin (2003) oil-exporting states dummy variable of .36 and .46,
respectively.
A primary advantage of these oil measures is that they allow us to distinguish
between oil produced in the past and oil still in the ground and, hence, of potential
value in the future. Measured in quantities, conversion factors and published prices
allow for conversion of these measures into values.34 Below, however, I simply use
measures of total production, plus production per capita.
Finally, as a rough measure of economic structures, I use the recorded share of agri-
cultural value added in national income, drawn from the World Bank’s (2003) World
Development Indicators.
WEAK STATES, SPARSE NETWORKS, AND BOOTY FUTURES
These measures allow us to test a number of the mechanisms relating to conflict

onset identified in section 2.
Consider first the sparse networks mechanism. I argued that the effects of natural
resource dependence may derive primarily from its impacts on the structure of econo-
mies rather than through the particular rent-seeking incentives that natural resources
provide for government or rebel elites. If indeed what matters is the weakness of eco-
nomic integration within a country, then we should expect to see conflict risks arising
in nonindustrial economies (after conditioning on national income) whether or not
these countries possess natural resource wealth, narrowly defined. If so, the problem
of primary commodity dependency may be more general than the problem of natural
resource dependency. For the purpose of this study, I use the share of agriculture in
national income as a measure of economic structure.
Tables 1 and 2 report on tests of these ideas. The models reported use Fearon and
Laitin’s (2003) data and model of civil war onset as a baseline but employ a rare events
logit specification (King and Zeng 1999) and add alternative measures of economic
activity (brief descriptions of all other variables are provided in Table 5). I begin in
Table 1 by looking at the African sample—that is, at those cases that motivated much
of the discussion above. I then turn to the global sample in Table 2.
The results in equation V of Table 1 indicate that there are indeed effects associated
with structural features of economies, as measured by the degree of agricultural
dependence of economies—an effect likely to be picked up by studies that use primary
commodities as a proxy for natural resources. The magnitude of the coefficients are
such that a rise in agricultural dependence from an African mean of 33 percent (about
the level of Cote d’Ivoire in 1990) to one standard deviation above the mean to 50 per-
cent (about the level of Sierra Leone in 1990) increases the chances of conflict onset by
about 1.5 percentage points, all else being equal (all other variables set at their means,
95 percent confidence intervals: 0.3 to 3.3 percentage points).
The results in equation V indicate furthermore that these effects are independent of
effects related to oil and diamonds. This lends prima facie support to the notion that the
34. The data set attached to this article includes an annual index of oil prices along with code to repli-
cate the results of this article using the value of oil production rather than quantities.
TABLE 1
Differential Effects? Africa Sample
(Dependent Variable: FL Measure of Civil War Onset)
Independent Variable I II III IV V VI
Oil production (per capita) 13.137 14.635 14.229

(1.49) (1.88)* (2.00)**
Oil reserves (per capita) 2.074 1.38 1.502
(0.54) (0.63) (0.79)
Diamond production (per capita) 0.592 0.539
(3.53)*** (3.97)***
Agricultural value added
as share of GDP 0.038 0.04
(2.28)** (2.37)**
Observations 1,482 1,503 1,482 1,482 1,283 1,283
NOTE: All the variables reported above were entered with a one-period lag. All equations were estimated us-
ing rare events logit and post-1960 data. Controls in all equations include Fearon and Laitin’s (FL; 2003)
measures of Lag of War, Lag of GDP, Log of Population, Log of Mountainousness, Non-Contiguity, Instabil-
ity, Democracy, Ethnic Fractionalization, Religious Fractionalization. Absolute value of robust z-statistics
shown in parentheses.
*Significant at 10%. **Significant at 5%. ***Significant at 1%.
problem of primary commodity dependence is not simply one of the availability of

lootable or obstructable resources but may also relate to structural feature of econo-
mies and how these structure social relations. These findings, identified in the African
sample, are in evidence in the global sample in Table 2; more weakly in equation V and
more strongly in equation VI.
Nonetheless, the estimated coefficients on the oil and diamond measures in equa-
tion V of Tables 1 and 2 also suggest that while economic structures matter, the prob-
lem is not simply one of “modernization” or the lack of it, but an outcome of the politi-
cal incentives resulting from particular types of natural resource commodities: oil and
diamonds. We now try to distinguish between these incentives.
I noted above that the “weak states” and the “grievance” mechanisms linking
resources to conflict could be distinguished from some “greedy rebels” mechanisms
insofar as the former would indicate a relationship between past production and con-
flict and the latter suggests a relationship between future production potential and con-
flict. One way to study this relationship econometrically is to look for different
impacts of historical oil production and the potential for future oil production, as mea-
sured by reserves. Turn then to the estimated coefficients on the measures of oil pro-
duction and of proven reserves in Table 1. We see here evidence of a positive relation-
ship between conflict onset and past oil production, at least once we control for oil
reserves. The size of the coefficient is such that an increase in production from 0 to 0.1
barrels per person per day (roughly the level produced in the Republic of Congo in
1997) is associated with a rise in the chances of a conflict onset by about 5 percentage
points (all other variables held at their means; 95 percent confidence intervals are 0.15
TABLE 2
Differential Effects? Global Sample
(Dependent Variable: FL Measure of Civil War Onset)
Independent Variable I II III IV V VI
Oil production (per capita) 2.38 4.61 2.954

(3.93)*** (4.43)*** (6.24)***
Oil reserves (per capita) 0.353 –0.254 0.182
(5.31)*** (1.05) (2.59)**
Diamond production (per capita) 0.549 0.516
(3.67)*** (5.09)***
Agricultural value added as
share of GDP 0.021 0.025
(1.87)* (2.17)**
Observations 5,208 5,240 5,208 5,208 3,721 3,721
NOTE: All the variables reported above were entered with a one-period lag. All equations were estimated us-
ing rare events logit and post-1960 data. Controls in all equations include Fearon and Laitin’s (FL; 2003)
measures of: Lag of War, Lag of GDP, Log of Population, Log of Mountainousness, Non-Contiguity, Insta-
bility, Democracy, Ethnic Fractionalization, Religious Fractionalization. Absolute value of robust z-statis-
tics shown in parentheses.
and 17.5 percentage points).35 Importantly, however, we find no similar evidence of a

correlation between reserves and conflict, once we control for production. These pat-
terns are also found at the global level. Both production and reserves enter signifi-
cantly in equations I and II of Table 2. The marginal effect of oil is, however, much
weaker in the global sample compared to the African sample, a similar rise from 0 to
0.1 barrels per person in the global sample is only associated with a 0.4 percentage
point rise in the probability of conflict onset. As with the African sample, the coeffi-
cient on reserves loses significance once we account for production in equation III of
Table 2. In equation V, estimated using a more restricted sample, the coefficient on
reserves enters significantly, but its effect, controlling for production, is substantively
much weaker than the effect estimated in equation II. This evidence is then supportive
of the weak state structures and the resource grievances hypotheses (although it cannot
distinguish between them) but fails, particularly for the African subsample, to provide
support for the greedy rebels mechanism, greedy outsiders mechanism, or booty
futures variant of the feasibility mechanism.
35. These effects, though strong in this model, are, as in previous work on natural resources, highly sen-
sitive to the sample and the model used. In some of the equations, significance on the coefficients of interest
is lost when a standard logit specification is used rather than a rare events logit specification, and in some
cases, relationships are no longer significant when total resources rather than per capita resources are used.
The data files that accompany this paper also provide results for nonrare events logit and logit fixed effects
specifications. In the nonrare events version, which is subject to bias, much weaker results are found for the
per capita measures, although many of the results for aggregate quantities are not affected. No effects are
identified for either oil or diamonds in fixed-effects specification, indicating that most of the effects are due
to cross-national rather than intertemporal variation. The results are robust to the taking the log of the oil
measures.
THE POLITICAL ECONOMY OF EXTRACTION HYPOTHESIS
We have some evidence then that natural resources relate to conflict through their
impacts on state capacity. In section 4, however, I discussed a somewhat more complex
relationship between resources and state strength, indicating that the degree to which
natural resources have adverse effects depends on prior levels of state strength. In that
discussion, I suggested the use of a simple interaction specification to test the idea that
whether lootable resources lead to conflict will depend on the degree of state strength.
To test this hypothesis, I again use the new measures of oil. And for a more lootable
commodity, I focus again on diamond production. Measuring state strength—or spe-
cifically the strength of institutions that can distribute wealth efficiently—poses
numerous problems. I use three proxies, all of which are imperfect. The first is Fearon
and Laitin’s (2003) measure of political instability—whether a state has undergone a
large change in its political institutions over the past three years. Such changes may,
but need not, indicate weakness of state structures. The second measure is a combina-
tion of Fearon and Laitin’s instability measure and their “anocracy” measure: it takes
the value of 1 if a state is a robust democracy or a robust dictatorship and a 0 otherwise.
The third measure I employ is Evans and Rauch’s (1999) measure, designed to record
the “Weberianness” of state structures. A major drawback of this measure, however, is
that it has no time series component and covers relatively few countries.36 All three
measures are likely to be endogenous to conflict; if conflicts or expected conflicts
weaken state capacity, especially in resource-dependent states, then this will bias us
towards finding a result different from zero.
In Table 3, I report the results of the tests of interactive effects between the three nat-
ural resource measures and the three state strength measures. In general, I find that
while (with the exception of the Weberianness measure), the state strength measures
and the natural resources measures typically enter significantly and with the expected
sign, the interaction term typically fails to enter significantly. The exception is for the
case of oil production; here I find weak evidence that oil production has especially
adverse effects in weak states. The coefficients in equations I and II indicate that past
oil production has an adverse effect on all polities but that this effect is augmented for
weak states. The coefficients in equation III indicate that for this measure of state
strength, past oil production is associated with higher risks of conflicts in weak
states but may in fact be associated with lower risks in strong states. This effect cap-
tures the Snyder-Smith thesis quite well. It is only however observed for the case of
oil production—and, notably, is not found for the case of diamonds production—and
in all cases the estimated coefficients in the interaction terms attain significance only at
the 90 percent level.37
36. As this measure is time-invariant, I employed it in both a cross-section and as a fixed effect in a
panel.
37. These results are in general weaker when total measures rather than per capita measures are used.
The exception is that in this case, one of the interactive terms on diamonds enters significantly, but this again
appears to be a fragile result.
528
TABLE 3
The Political Economy of Extraction:
Global Sample (Dependent Variable: FL Measure of War Onset)
I II III IV V VI VII VIII IXa

Measure of Natural Resources Oil Production (Per Capita) Oil Reserves (Per Capita) Diamonds Production (Per Capita)
b b b
Measure of Strength/Weakness Instability Strong Weberian Instability Strong Weberian Instability Strong Weberian
Natural resources 2.415 8.537 –89.911 0.336 0.956 1.603 1.11 0.937 3,204.00
(3.94)*** (3.36)*** (1.70)* (5.67)*** (1.25) (0.16) (4.27)*** (3.19)*** (.)
Strength/weakness 0.595 –0.466 –0.056 0.646 –0.51 –0.051 0.618 –0.517 –0.047
(2.39)** (1.67)* (0.72) (2.57)** (1.80)* (0.66) (2.49)** (1.85)* (0.55)
Interaction term 10.978 –5.313 18.572 0.974 –0.588 1.368 –0.004 0.543 –800.649
(1.82)* (1.93)* (1.93)* (0.43) (0.76) (1.24) (0.01) (1.27) (.)
Observations 5,170 5,167 1,339 5,170 5,167 1,339 5,170 5,167 1,339
NOTE: All the variables reported above were entered with a one-period lag. All equations were estimated using rare events logit and post-1960 data. Controls in all equations
include Fearon and Laitin’s (2003) measures of Lag of War, Lag of GDP, Log of Population, Log of Mountainousness, Non-Contiguity, Instability, Democracy, Ethnic
Fractionalization, Religious Fractionalization. Absolute value of z-statistics reported in parentheses. Note that approximately 25 percent of observations are lost due to miss-
ing data when reserves are added.
a. Computational problems encountered in estimating equation IX.
b. “Strong” is given by (1 – Instability) ´ (1 – Anocracy).
EVIDENCE ON CONFLICT DURATION MECHANISMS
To help distinguish between some of the rival mechanisms that may link natural
resources and conflict duration, I construct a simple duration model like that described
in section 4.
The duration model uses data from the Fearon and Laitin data set to construct a
binary variable indicating whether, conditional upon the existence of a conflict, the
conflict ends in a given year. Data based on Barbara Walter’s conflict termination data
set and qualitative sources was then used to split this dichotomous measure into a
three-way variable that takes the value 0 if a conflict is ongoing in a given year, 1 if a
conflict is resolved through military means, and 2 if a conflict is resolved through a
negotiated settlement. Further information on these variables can be found in Walter
(1997, 2002) and Fearon and Laitin (2003).38 The method employed to relate different
measures of natural resources to this outcome variable is multinomial logit.39
As explanatory variables, I introduce a series of controls alongside the measures of
diamond production and oil production and reserves. The effects of each of the
resource measures on duration are a priori ambiguous. Introducing the diamond mea-
sure is intended to help distinguish between the fragmented organizational structures
mechanism and the domestic conflict premium mechanism, which can have opposite
implications for duration depending on whether a victory or a negotiations channel is
pursued. Introducing the oil measure, meanwhile, is intended to help distinguish
between the military capability mechanism and the possibility of pork mechanism, in
particular to help determine whether desires to capture oil revenues are more likely to
encourage compromise at the negotiation table or one-sided support on the battlefield.
Hence, the model allows us to parse some of the duration mechanisms discussed
above. Insofar as it focuses only on “internal” factors, the model is incomplete and a
caveat is in order: since external factors are likely to be correlated with the natural
resource measures employed in this model, their exclusion may induce omitted vari-
able bias. There are also potential endogeneity problems—the data suggest, for exam-
ple, that oil production typically rises over the course of a conflict—and reserves
fall—whereas diamonds production falls during the course of a conflict.
With these caveats in mind, we turn to the results of the model in Table 4. The model
suggests that natural resources are associated with shorter conflicts—this is supported
both for the oil measures and for diamond production. This finding is counterintuitive
because there are a number of well-known long conflicts in countries rich in natural
resource—notably Sudan, Columbia, and Angola. It also runs contrary to many
38. In cases where these sources disagreed, Fearon and Laitin (2003) codings were generally employed
after consultation with other sources. In some instances, the differences in dates implied different types of
terminations from those recorded by Walter (1997). For example, where Walter records a successful settle-
ment in Lebanon in 1976, I record the war as ending with a military victory in 1990; whereas Walter records
the war in Tajikistan as ending in 1994 without a settlement, I record it as ending in 1997 with a settlement. In
cases where conflict in the Fearon and Laitin data set were not in Walter, secondary sources were used to
determine the form of termination, using Walter’s coding rules.
39. See Beck, Katz, and Tucker (1998) on the relation between logit models and duration models.
530
TABLE 4
Conflict Duration Model with Victory and Negotiation Channels
Dependent Variable I II III IV V VI VII VIII IX

“Pooled” Channel 1 Channel 2 “Pooled” Channel 1 Channel 2 “Pooled” Channel 1 Channel 2
Independent Variable Termination Victory Negotiation Termination Victory Negotiation Termination Victory Negotiation
Diamond production 0.1 0.086 0.141

(2.31)** (1.84)* (1.38)
Oil reserves 0.022 0.02 –0.099
(1.66)* (1.63) (0.49)
Oil production 0.378 0.38 0.093
(2.16)** (2.28)** (0.19)
Year 0.013 –0.005 0.076 0.014 –0.004 0.075 0.01 –0.009 0.076
(0.87) (0.29) (2.01)** (0.88) (0.28) (1.98)** (0.64) (0.55) (2.00)**
Polity score –0.009 –0.061 0.137 0.001 –0.048 0.139 –0.004 –0.053 0.137
(0.36) (1.84)* (2.27)** (0.04) (1.47) (2.30)** (0.14) (1.62) (2.30)**
Population (logged) –0.479 –0.327 –1.106 –0.454 –0.313 –0.967 –0.525 –0.398 –1.029
(3.00)*** (2.12)** (3.47)*** (2.79)*** (2.07)** (3.11)*** (3.07)*** (2.46)** (3.23)***
Ethnic fractionalization –1.677 –1.53 –2.567 –1.566 –1.439 –2.294 –1.544 –1.457 –2.36
(2.50)** (1.94)* (2.19)** (2.38)** (1.84)* (2.01)** (2.32)** (1.86)* (2.06)**
Africa dummy 0.065 –0.042 0.74 0.225 0.134 0.824 0.267 0.218 0.843
(0.17) (0.10) (0.99) (0.59) (0.31) (1.12) (0.70) (0.50) (1.15)
Constant –23.611 10.663 –145.14 –24.857 10.129 –143.072 –17.13 19.906 –144.931
(0.78) (0.34) (1.93)* (0.80) (0.32) (1.92)* (0.55) (0.62) (1.93)*
Observations 794 794 794 797 797 797 794 794 794
NOTE: Regression also includes three cubic splines (produced using Beck, Katz, and Tucker’s [1998] methodology). For all multinomial logit models, application of the
Hausman test suggests that we cannot reject the null that the independence of irrelevant alternatives (IIA) hypothesis is indeed satisfied. The Small-Hsiao test (but not the
Hausman test) suggests that the IIA assumption is violated for the diamonds model (and that the probability of negotiation success, relative to no termination, depends on the
probability of military victory). The Small-Hsiao, like the Hausman test, does not reject the null of IIA for either of the oil models.
hypotheses and claims in the literature.40 Despite the well-known cases of long wars, a
larger range of other cases make the general relationship, controlling for other factors.
In particular, there are many cases of long wars in countries that have little or no oil or
diamonds: in the data set, these include the conflicts in Afghanistan, Sri Lanka, Ethio-
pia, Mozambique, and Somalia, which all lasted for more than fifteen years. And there
are also many cases of short wars in countries that do have these resources; these
include the Biafran war in Nigeria, the Shaba insurgencies in Zaire (1978), the
Kurdish/Shiite revolts in Iraq (1991), and conflicts in Azerbaijan (1992-1994), Geor-
gia (1992-1994), Croatia (1995), Central African Republic (1996-1997), and Yemen
(1994).
Let us turn now to the mechanisms. In equations 2 and 3 (and similarly, equations 5
and 6, and 8 and 9) of Table 4, conflict termination is disaggregated into settlements
and victories. Strikingly, we see that for both oil and diamond production,
disaggregating the paths to peace suggests that natural resource abundance is associ-
ated with easier military victories. But in neither case is there evidence that natural
resources facilitate or obstruct negotiated settlements. This provides clues as to why
natural resource conflicts may be shorter.
Consider first the organizational structures mechanism and the results on dia-
monds. If revenues from diamond sales are more likely to benefit rebels than govern-
ments, and should they lead to more fragmented rebel organizational structures, then,
as suggested above, there are two opposite implications that this may have for conflict
duration: it may make military victory over rebels easier but negotiated settlements
more difficult. The evidence in Table 4 suggests that the first channel may be salient
but provides no support to the second channel. Instances of military victories in dia-
mond-rich countries include the relatively short Zairian conflicts ending in 1965 and
1978. Although there were multiple attempts at negotiation in Sierra Leone in Angola,
those conflicts also ended with the military defeat of the rebel groups and the disinte-
gration of their organizational structures. The result provides evidence against the con-
flict premium mechanism: If combatants in diamond-rich areas are primarily con-
cerned with accumulating wealth during wartime, then we should expect little success
of negotiations. The fact that diamond production is not associated with lower chances
of success for negotiations is then inconsistent with this argument.41
The results on oil in equations IV through IX of Table 4 suggest that although oil
conflicts end quickly, this is not due to their benign influence on negotiations: we find
no evidence in support of the possibility of pork mechanism.42 In the case of oil pro-
duction, we find some evidence that oil conflicts are associated with faster military
victories; this is consistent with the discussion of the military balances mechanism
40. Ross (2004b), for example, suggests that commodities that are lootable (such as diamonds) and
commodities that are obstructable (such as oil) should both lead to longer wars. Bannon and Collier (2003)
note that conflicts are likely to make countries even more dependent on natural resources and thereby make
conflicts more difficult to resolve.
41. A stronger test of the conflict premium mechanism would, however, require the use of information
on contraband.
42. The effects are qualitatively similar when per capita data is used for both production and reserves—
the oil measures are positively related to victory and negatively related to negotiation, while the diamond
measures are positively related with both—however, the estimated relations are generally weaker.
TABLE 5
Descriptions and Summary Statistics for Key Variables Used in the Analysis
Variable Units Mean Min Max Source
War Onset Binary: Did a civil war begin this year? 0.02 0 1 Fearon and
Laitin (FL; 2003)
War Ongoing? Binary: Is there a war ongoing this year? 0.15 0 1 FL
Log of Population Quantity (logged) 9.05 5.40 14.03 FL
GDP PPP Thousand Dollars 3.96 0.13 66.74 FL
Polity Index Index –0.58 –10 10 FL
Mountainousness Log of the percentage of mountainous 2.11 0 4.56 FL
terrain
Noncontiguous Binary 0.16 0 1 FL
Ethnolinguistic Index 0.40 0.00 0.93 FL (Russian
Fragmentation Measure)
Religious Index 0.38 0 0.78 FL
Fractionalization
Instability Binary: Has the countries score on the 0.15 0 1 FL
Polity democracy scales shifted in the
past three years?
New State? Binary: Is the state no more than two 0.03 0 1 FL
years old?
Anocracy Binary: Is the state neither a full 0.20 0 1 FL
democracy nor a full dictatorship?
Africa Dummy Binary: Is state an African state? 0.28 0 1 FL
Oil Production Millions of barrels per day 0.29 0 12.28 See text
Oil Production/ Barrels per person per day 0.05 0 5.14 Constructed
Per Capita
Oil Reserves Billion barrels 3.45 0 263.5 See text
Oil Reserves/ Thousand barrels per person 0.88 0 176.97 Constructed
Per Capita
Diamonds Million metric carats 0.37 0 43.80 See text
production
Diamond Carats per person 0.07 0 13.31 Constructed
Production/
Per Capita
Agricultural Percentage: Value of Agricultural 22.30 0.14 78.02 World Bank
Production as Production in GDP (2003)
a Share of GDP
State strength Binary = (1 – Instability) ´ 0.73 0 1 Constructed
(1 – Anocracy)
Weberianess Scale: How Weberian is the state 7.10 1 13.5 Evans and Rauch
bureaucracy? (1999)
Conflict 0 = conflict continued, 1 = conflict 0.10 0 2 Constructed from
Termination ended in victory for one side, 2 = FL and Walter
“Disaggregated conflict ended through negotiations (1997, 2002)
Measure” data
above. There, we noted that the implications of natural resource wealth for the ability
of one side to defeat the other are indeterminate without finer information on military
balances and control over resources. However, in cases where benefits accrue only
after the end of the war, we expect a relationship between conflict duration and the
incentives to provide military support to one side or the other: in these cases, outsiders
invest if they expect conflicts to be short, and they invest so as to make them short.
A review of the cases suggests that international action was indeed important and
likely led to rapid termination of many of these oil conflicts, although this support was
not always and everywhere in support of governments. Military victory came rela-
tively quickly, for example, for the government of Nigeria in the Biafran war; in this
case, even though France had interests in supporting the breakaway state, massive sup-
port by the British for the sovereign state of Nigeria gave the government the resources
needed for military victory. However, governments were often on the losing side, and
even when they were successful, the role of international actors was often ambiguous.
Gulf states, for example, were divided over how to respond to the war in Yemen in
1994, with some providing support for the secessionist south and others supporting the
north.43 Strikingly, the north’s ability to benefit from its sovereign status was ham-
pered: UN Security Council Resolution 931 prevented arms from flowing to either
side. Meanwhile, during the war, one company, CanadianOxy, that continued to pump
oil, placed the government’s share of the oil funds in an escrow account, only paid out
after the end of the conflict.44
6. CONCLUSIONS
Prominent research has focused on correlations without constructing tests to iden-

tify particular mechanisms that may underlie those correlations. And much of this
work has arbitrarily favored one mechanism to the exclusion of others. This article has
suggested a series of ways to correct this; it has highlighted tests that can be con-
structed to check for the workings of rival mechanisms, and it has made progress in
using these techniques to begin to parse these different mechanisms.
The empirical regularities presented in this article give grounds for favoring some
of the many explanations over others. Nonetheless, econometric tests of the effects of
natural resources on conflicts, including those presented here, continue to suffer from
severe problems of data, model specification, and in particular a sensitivity of coeffi-
cient estimates to variations in model specification (see Fearon 2005 [this issue]). In
this article, I have found evidence to support some mechanisms and reported the lack
of support for others, but I have not constructed sufficient tests to distinguish between
all the mechanisms identified in sections 2 and 3. As mechanisms become unpacked,
increasingly more fine grain becomes required to differentiate arguments satisfacto-
rily. Chief among the measures now needed are better indicators of the composition of
government revenues, their sources and uses; reliable measures of state strength;
43. “Oil States Accused of Meddling in Yemen,” Le Monde, June 12-13, 1994.
44. “Oil Firms Return to Yemen,” Agence France Presse, July 25, 1994.
better indicators of the role of foreign interests in domestic oil production; and
measures of the relative strengths of rival forces in a conflict.
In closing, I turn to the policy implications of this work. The importance of focus-
ing on mechanisms for the study of civil wars is not simply that explanations with finer
grain are intrinsically more satisfactory to the mind. It derives also from the fact that
public policy responses require stories about who is doing what and why. Different
stories underlying a single correlation have different implications. Given the caveats
above, the following are among the implications of the results from the work presented
here.
First, countries dependent on agricultural commodities are at risk, independent of
their endowments of oil and diamonds. Sierra Leone was vulnerable not simply
because it had gems but because it had not gone through a process of industrialization
and held within it clusters of rural communities with relatively weak commercial ties
between them. In identifying at-risk countries and in engaging in conflict prevention
alongside initiatives to clean up particular commodity trades, there is a need to pursue
strategies of diversification more aggressively, directed at bringing countries outside
of dependence on primary commodities more broadly defined.
Second, this research finds stronger support for the weak state structures and griev-
ance hypotheses than for the booty futures or state capture hypotheses. This, coupled
with the somewhat weaker result that natural resources have especially adverse effects
in countries that already have weak states, suggests a redirection of policy priorities.
Policy priorities from previous research have focused on protecting assets from cap-
ture and cutting off rebel financing. While these initiatives are important, the analysis
in this article indicates that greater gains could be achieved by focusing more on better
management of the extraction process and better usage of resource revenues that are
controlled by states. This result, consistent with a now vast literature on the resource
curse (see, for example, Karl 1997), suggests a series of policy responses.
One, to limit grievances induced as part of extraction, is to better regulate the
actions of extractive industries. This could be done by requiring corporate compliance
with protocols such as the United Nations draft “Norms on the Responsibilities of
Transnational Corporations and Other Business Enterprises with Regard to Human
Rights” or by corporate participation in voluntary mechanisms such the Global Com-
pact; and it can be supported by resource rich states participating in initiatives such as
the Extractive Industries Transparency Initiative.
Another is to focus on tackling the ways in which natural resource revenues weaken
state structures or induce grievances. This can be done by better management of
intertemporal revenue paths, through the establishment or permanent funds and stabi-
lization funds and rules that place caps on annual government expenditure, as, for
example, is provided for under Sao Tome and Principe’s oil revenue management law
(2004). Another way to achieve this is to prededicate resource revenues to social
development, as done in the Chad revenue management law. Governments can also
reduce the grievances associated with natural resources by providing better public
information about how and where revenues are earned and spent and allowing for
oversight of these expenditures by civil society groups. The 2004 Sao Tome oil reve-
nue management law emphasizes these principles and guarantees a role to parliamen-
tary oppositions in overseeing the expenditure of oil monies; the Chadian law, how-
ever, has comparatively weak provisions for public access to information. A more
radical possibility is to rid the government of natural resource revenues outright—by
distributing all revenues directly to citizens—as suggested recently for Nigeria by
Sala-i-Martin and Subramanian (2003). This has the advantage that private citizens
may spend these revenues better than their governments do and that governments may
start trying to satisfy constituencies and building capacity to raise revenues, rather than
relying on rents.
Third, this research has found evidence that natural resource conflicts are more
likely to end quickly and are more likely to end with military victory for one side rather
than with a negotiated settlement. One likely reason is that in the presence of lootable
resources, rebel structures are weaker and the groups more vulnerable to defeat. Evi-
dence from survey research suggests, for example, that regular fighters in Sierra
Leone’s RUF had little loyalty to the movement and would have been ready to drop
arms well before Britain finally moved in.45 Another reason is that once conflicts begin
in areas with resources whose benefits accrue only after conflicts end, external actors
with interests in those sectors have incentives to support one side or the other with a
view to bringing the war to a rapid close. These results suggest reasons to support one-
sided military interventions in resource conflicts. But they also provide a reminder that
whatever the interests of the citizens of the countries involved, outside actors are prone
to one-sided engagement in natural resource conflicts, directly or indirectly, and
sometimes inducing regime change and producing deadly effects. Both of these fac-
tors—the opportunity for engagement in resource wars and the existence of incentives
for individual nations to engage in resource wars unilaterally—imply that the policy
debate should focus more urgently on establishing workable criteria for determining
what regimes should be supported and when external strategies of regime change
should be pursued.
REFERENCES
Agir ici-Survie. 1999. Projet pétrolier Tchad-Cameroun: Dés pipés sur le pipeline. Paris: L’Harmattan.
Angell, Norman. 1933. The great illusion, 1933. New York: Putnam.
Bannon, Ian, and Paul Collier. 2003. Natural resource and violent conflicts. Washington, DC: World Bank.
Beck, Nathaniel, Jonathan N. Katz, and Richard Tucker. 1998. Taking time seriously in binary time-series-
cross-section analysis. American Journal of Political Science 42 (4): 1260-88.
Collier, Paul. 1999. On the economic consequences of civil war. Oxford Economic Papers 51:168-83.
¾¾¾. 2000a. Doing well out of war: An economic perspective. In Greed and grievance: Economic agen-
das in civil wars, ed. Mats Berdal and David Malone. Ottawa, Canada: International Development
Research Centre.
¾¾¾. 2000b. Economic causes of civil conflict and their implications for policy. In Managing global
chaos, ed. Chester A. Crocker, et al. Washington, DC: United States Institute of Peace.
¾¾¾. 2000c. Rebellion as a quasi-criminal activity. Journal of Conflict Resolution 44 (6): 839-53.
¾¾¾. 2003. The market for civil war. Foreign Policy, May/June (136): 38-45.
45. Humphreys and Weinstein (2004).

Collier, Paul, and Anke Hoeffler. 2000a. Aid, policy and peace. Working Paper, World Bank, Washington,
DC.
¾¾¾. 2000b. Greed and grievance in civil wars. Working Paper WPS 2000-18, World Bank, Washington,
DC.
Collier, Paul, Anke Hoeffler, and Måns Söderbom. 2001. On the duration of civil war. Working Paper, World
Bank, Washington, DC.
Dashwood, Hevina S., 2000. Zimbabwe, the political economy of transformation. Toronto, Canada: Univer-
sity of Toronto Press.
Democratic Republic of Sao Tome and Principe. 2004. Lei-Quadro Das Receitas Petrolíferas.
Doyle, Michael. 1997. Ways of war and peace: Realism, liberalism, and socialism. New York: Norton.
Elster, Jon. 1998. A plea for mechanisms. In Social mechanisms: An analytical approach to social theory, ed.
Peter Hedström and Richard Swedberg. Cambridge: Cambridge University Press.
Englebert, Pierre, and James Ron. 2004. Primary commodities and war: Congo-Brazzaville’s ambivalent
resource curse. Comparative Politics 37 (1): 61-81.
Ensminger, Jean. 2001. Market integration and fairness: Evidence from ultimatum, dictator and public
goods experiments in East Africa. Working Paper, California Institute of Technology, Pasadena, CA.
Evans, Martin. 2002. The political economy of war in the Casamance. Working Paper, School of Oriental
and African Studies, London.
Evans, Peter, and James Rauch. 1999. Bureaucracy and growth: A cross-national analysis of the effects of
“Weberian” state structures on economic growth. American Sociological Review 64 (5): 748-65.
Fearon, James D. 2004. Why do some civil wars last so much longer than others? Journal of Peace Research
41 (3): 275-301.
¾¾¾. 2005. Primary commodities exports and civil war. Journal of Conflict Resolution 49 (4): 483-507.
Fearon, James D., and David Laitin. 2003. Ethnicity, insurgency, and civil war. American Political Science
Review 97 (1): 75-91.
Foccart, Jacques. 1999. Dans les Bottes du Général. Journal de l’Élysée III, 1969-1971. Paris: Fayard.
Gordon, Sanford C., and Alastair Smith. 2004. Quantitative leverage through qualitative knowledge: Aug-
menting the statistical analysis of complex causes. Political Analysis 12 (3): 233-55.
Hirschman, Albert. 1982. Rival interpretations of market society: Civilizing, destructive, or feeble. Journal
of Economic Literature 20:1463-84.
Humphreys, Macartan, and Jeremy Weinstein. 2004. What the fighters say. Center for Globalization and
Sustainable Development Working Paper no. 20, Columbia University, New York.
Karl, Terry. 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, Oxford University
Press, Oxford, for the International Institute of Strategic Studies.
King, Gary, and Langche Zeng. 1999. Logistic regression in rare events data. Department of Government,
Harvard University. http://gking.harvard.edu.
Le Billon, Philippe. 2001. The political ecology of war: Natural resources and armed conflict. Political
Geography 20 (5): 561-84.
Lohmann, Susanne, and Sharyn O’Halloran. 1994. Divided government and US trade policy. International
Organization 48 (4): 595-632.
Lujala, Päivi, Nils Petter Gleditsch, and Elisabeth Gilmore. A diamond curse? Civil war and a lootable
resource. Journal of Conflict Resolution 49 (4): 538-62.
Meldrum, Andrew. 2000. Good-bye. New Republic On-Line, April 17.
Moore, Mick. 2001. Political underdevelopment: What causes “bad governance.” Working Paper, Institute
for Development Studies, Sussex, UK.
Nolutshungu, Sam. 1996. Limits of anarchy: Intervention and state Formation in Chad. Charlottesville: Uni-
versity Press of Virginia.
Oneal, John, and Bruce Russett. 1999. Assessing the liberal peace with alternative specifications: Trade still
reduces conflict. Journal of Peace Research 36 (4): 423-32.
Richards, Paul. 1996. Fighting for the rainforest. London: Heinemann.
Ross, Michael. 2002. Booty futures. Working Paper, University of California, Los Angeles.
¾¾¾. 2003. Oil, drugs, and diamonds: How do natural resources vary in their impact on civil war? In
Beyond greed and grievance: The political economy of armed conflict, ed. Karen Ballentine and Jake
Sherman. Boulder, CO: Lynne Rienner.
¾¾¾. 2004a. How does natural resource wealth influence civil war? Evidence from thirteen cases. Inter-
national Organization 58 (1): 35-67.
¾¾¾. 2004b. What de we know about natural resources and civil war? Journal of Peace Research 41 (3):
337-56.
Russett, Bruce. 2002. Violence and disease: Trade as suppressor of conflict when suppressors matter. In Eco-
nomic interdependence and international conflict: New perspectives on an enduring debate, ed. Edward
Mansfield and Brian Pollins. Ann Arbor: University of Michigan Press.
Sala-i-Martin, Xavier, and Arvind Subramanian. 2003. Addressing the natural resource curse: An illustra-
tion from Nigeria. National Bureau of Economic Research Working Paper 9804, Cambridge, MA.
Schelling, Thomas. 1960. The strategy of conflict. Cambridge, MA: Harvard University Press.
Smith, Ben. 2004. Oil wealth and regime survival in the developing world, 1960-1999. American Journal of
Political Science 48 (2): 232-46.
Snyder, Richard. 2002. Does lootable wealth breed civil war? Resource extraction and political order in
comparative perspective. Working Paper presented at the American Political Science Association
Meetings, Boston.
Snyder, Richard, and Ravi Bhavnani. 2005. Diamonds, blood and taxes: A revenue-centered framework for
explaining political order. Journal of Conflict Resolution 49 (4): 563-97.
Sørli, Mirjam E., 2002. Resources, regimes and rebellion. Paper presented at the 43rd Annual International
Studies Association Convention, March 22-27, New Orleans, LA.
Verschave, François-Xavier. 2000. Noir Silence. Paris: Éditions des Arènes.
Walter, Barbara F. 1997. The critical barrier to civil war settlement. International Organization 51 (3): 335-64.
¾¾¾. 2002. Committing to peace. Princeton, NJ: Princeton University Press.
Weinstein, Jeremy. 2005. Resources and the information problem in rebel recruitment. Journal of Conflict
Resolution 49 (4): 598-624.
Willum, Bjørn. 2001. Foreign aid to Rwanda: Purely beneficial or contributing to war? Dissertation, Univer-
sity of Copenhagen, Denmark.
World Bank. 2003. World development indicators 2002. CD-ROM. Washington, DC: World Bank.
Zartman, I. William. 2000. Ripeness: The hurting stalemate and beyond. In International conflict resolution
after the cold war, ed. P. Stern and D. Druckman. Washington, DC: National Academy Press.

Natural Resources, Conflict, and Conflict Resolution: Uncovering The Mechanisms

Uploaded by

Copyright:

Available Formats

Natural Resources, Conflict, and Conflict Resolution: Uncovering The Mechanisms

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Natural Resources, Conflict, and Conflict Resolution: Uncovering The Mechanisms

Uploaded by

Copyright:

Available Formats

JOURNAL OF

Natural Resources, Conflict,

Keywords: civil war; resource-conflict link; resource dependence

2. NATURAL RESOURCES AND CIVIL WARS:

5. See especially Collier and Hoeffler (2000b).

The grievance mechanism. Natural resource dependence could in fact be associated

The weak states mechanism. State structures may be weaker in natural-resource-

The sparse networks mechanism. Plausibly, the importance of natural resources

3. NATURAL RESOURCES AND DURATION

The feasibility mechanism. Plausibly natural resource financing creates longer

The military balances mechanism. The financing available to protagonists is of use

The fragmented organizational structures mechanism. A new and fruitful line of

The possibility of pork mechanism. A related argument focuses on peace negotia-

The sparse networks mechanism. An analogy of the sparse networks mechanism

extent of economic exchanges.27 An illustration of the logic is also provided by the

4. MECHANISMS AND ESTIMATION

With so many possible mechanisms, is it possible to differentiate between them?

SYSTEMS WITH TYPE B MECHANISMS

Disaggregating the Explanatory Variable

between conflict and future discretionary government income. If measures of these

Disaggregating the Dependent Variable

SYSTEMS WITH TYPE A MECHANISMS

Type A Mechanisms Where the Sorting Procedure Is Known

Type A Mechanisms Where the Sorting Procedure Is Not Known

any given case—can itself be a function of a series of explanatory variables. This

5. RESULTS: MECHANISM IDENTIFICATION

WEAK STATES, SPARSE NETWORKS, AND BOOTY FUTURES

These measures allow us to test a number of the mechanisms relating to conflict

Independent Variable I II III IV V VI

Oil production (per capita) 13.137 14.635 14.229

problem of primary commodity dependence is not simply one of the availability of

Independent Variable I II III IV V VI

Oil production (per capita) 2.38 4.61 2.954

and 17.5 percentage points).35 Importantly, however, we find no similar evidence of a

THE POLITICAL ECONOMY OF EXTRACTION HYPOTHESIS

I II III IV V VI VII VIII IXa

EVIDENCE ON CONFLICT DURATION MECHANISMS

Dependent Variable I II III IV V VI VII VIII IX

Diamond production 0.1 0.086 0.141

Variable Units Mean Min Max Source

Prominent research has focused on correlations without constructing tests to iden-

45. Humphreys and Weinstein (2004).

You might also like