The Geography of Conﬂict Diamonds:
The Case of Sierra Leone
)and Andrew Crooks2
1Biocomplexity Institute of Virginia Tech, Arlington, VA 22203, USA
2Computational Social Science Program,
George Mason University, Fairfax, VA 22030, USA
Abstract. In the early 1990s, Sierra Leone entered into nearly 10years
of civil war. The ease of accessibility to the country’s diamonds is said
to have provided the funding needed to sustain the insurgency over the
years. According to Le Billon, the spatial dispersion of a resource is a
major deﬁning feature of a war. Using geographic information systems
to create a realistic landscape and theory to ground agent behavior, an
agent-based model is developed to explore Le Billon’s claim. Diﬀerent
scenarios are explored as the diamond mines are made secure and the
mining areas are moved from rural areas to the capital. It is found that
unexpected consequences can come from minimally increasing security
when the mining sites are in rural regions, potentially displacing conﬂict
rather than removing it. On the other hand, minimal security may be
suﬃcient to prevent conﬂict when resources are found in the city.
Keywords: Agent-based modeling ·Geographic information systems ·
Civil war ·Conﬂict
In the early 1990s, Sierra Leone, a small country on the western coast of Africa,
entered into nearly 10years of civil war. Sparked by an abusive government and
fueled by an illicit diamond market, the decade-long war killed an estimated
70,000 and displaced another 2.6 million people . It is said that the primary
driver of the war was the country’s most abundant and valued resource, dia-
monds . While the resource has resulted in growth in other countries such as
Botswana, Sierra Leone has experienced some of the highest levels of poverty
in the world. Unlike the diamond mines of Botswana, however, the alluvial dia-
mond mines of Sierra Leone cover widespread areas in remote parts of the country
where mining areas cannot be easily fenced and security is minimal .
Le Billon  argued that the spatial dispersion of a resource is a major deﬁn-
ing feature of a war, impacting the type of conﬂict that may emerge. An agent-
based model (ABM) was developed of the resource-driven conﬂict to explore
Springer International Publishing Switzerland 2016
K.S. Xu et al. (Eds.): SBP-BRiMS 2016, LNCS 9708, pp. 335–345, 2016.
DOI: 10.1007/978-3-319-39931-7 32
336 B. Pires and A. Crooks
Le Billon’s  theory. Some of the earliest ABMs of rebellion include Axelrod’s 
model of new political actors and Epstein’s  civil violence model. More recent
ABM’s have explored in-group dynamics and ethnic salience (e.g., [9–11]). While
the ABM presented here shares similarities with prior ABMs that have explored
income, resources, and identity as drivers of conﬂict, it also introduces some key
diﬀerences. Utilizing geographic information systems (GIS) and socioeconomic
data of the country, a landscape and population that better represent the actual
setting being modeled is created while the behavior of agents draws from theory.
Diﬀerent scenarios are run as the diamond mines are made more secure and the
mining areas are moved to the capital. It is found that unexpected consequences
can come from minimally increasing security over the diamond mines in rural
regions. For instance, while minimal increases in government control stopped
rebel activity in the south, it displaced the conﬂict to a district in the north,
which had not seen violence in prior runs of the model.
Theorists have pointed to opportunity, along with motivation and group identity,
as indicators of war (e.g., [5,6]). Opportunity can come in the form of ﬁnancing,
the availability of recruits, and the ability to garner these resources with relative
ease, which can be due to factors such as geography, economics, and availabil-
ity. Others have focused on the ﬁnancing of war through “lootable” resources
(e.g., ). Le Billon , while agreeing that lootable resources are a factor in con-
ﬂict, argues further that the spatial dispersion of a resource is a major deﬁning
feature of a war, impacting the type and duration of rebellion.
According to Le Billon , conﬂict characteristics are aﬀected by two geo-
graphic factors: (1) the location of resources as they relate to the country’s center
(proximate versus distant) and (2) the concentration of resources (point versus
diﬀuse). Distant resources (i.e., in remote areas) are easier for rebel forces to
capture and control, while proximate resources are easier to secure and are less
likely to be captured (e.g., coﬀee). Diﬀuse resources are widespread over large
geographic areas, making the resource more diﬃcult to secure (e.g., alluvial dia-
monds). Point resources, however, are concentrated in small geographic areas
and typically require mechanized extraction (e.g., kimberlite diamonds) making
them easier to secure and less likely to be exploited . Assuming an environ-
ment that is ripe for conﬂict, the geographic features of a resource can inﬂuence
the type of conﬂict. This relationship is illustrated in Table 1.
3 Model Development
An ABM was developed in MASON  to explore the role of geography in the
resource-driven war. GIS data was utilized to create the modeling landscape,
while socioeconomic data provided initial agent attributes. Due to the localized
nature of social processes, including civil violence, ABM combined with GIS is
ideal for modeling the unique environment of Sierra Leone and the long-lasting
The Geography of Conﬂict Diamonds: The Case of Sierra Leone 337
Table 1. The relationship between the resource dispersion and conﬂict type .
Concentration/Relation to center Diﬀuse Widely spread with
in small areas
Distant Warlordism Secession
Located in remote territories
Proximate Rioting / mass rebellion State control or coup
Close to center of power
conﬂict it endured. For brevity, a high-level overview of the model is presented
here. The detailed model description using the Overview, Design Concepts, and
Details (ODD) protocol , the source code, and data to run the model can
be downloaded from https://www.openabm.org/model/4955/. The model’s ini-
tialization process is discussed in Sect. 3.1; the agents’ behavior is discussed in
Sect. 3.2; and the model’s outputs are reviewed in Sect. 3.3.
3.1 Model Initialization
The modeling world encompasses the country of Sierra Leone, an area of approx-
imately 71,740 km2. Each run of the simulation begins by reading in the spatial
dataset and building the environment using data from the Global Administra-
tive Areas database , the Oak Ridge National Laboratory , the Peace
Research Institute Oslo , and OpenStreetMap . The agent population is
created using data from the Republic of Sierra Leone 1985 and 2004 Population
and Household Census  and the Oak Ridge National Laboratory , while
socioeconomic data, which provided information on age distribution, income
levels, and employment statistics, came from the Republic of Sierra Leone 2004
Population and Housing Census [22,23] and Statistics Sierra Leone’s Annual
Statistical Digest . Due to the computational constraints of modeling the
complete population of Sierra Leone (approximately 4.9 million), the population
within each parcel is reclassiﬁed to equal one percent of the total population.
Model runs performed at varying populations yielded similar qualitative results.
Note that households are not explicitly modeled here. The idea of a household is
used only to ensure that agents can be assigned an income even if unemployed.
Table 2summarizes the input parameters used in the model.
The model proceeds in one-month increments. While the decision to join the
rebellion may occur in a short time period (hours or even minutes), there is
a lag of weeks or even months between the time someone makes that decision
(or is forced to make that decision) and the time they are actually ready for
combat . In addition, the war lasted years. From a modeling perspective, we
are interested in capturing the dynamics of the conﬂict over the years, not days
or hours. We also need to consider the balance between spatial and temporal
computational resources. The modeling world is the entire country of Sierra
Leone and simulates the dynamics of a war as it spreads a country.
338 B. Pires and A. Crooks
Table 2. Input parameters and variables.
Parameter Range Default value Reference
Initial number of agents 1–4.9 million 49,000 [15,19]
Percentage of population in the initial
0–1 0.005 
Age Grouped in age ranges 0–6, 7–17 18–64 
Income level 1–3 1–3 
Empl oyment st atus 1–4 1–4 
Vision 0–370 25 Authors estimation
Likelihood to mine if food poor 0–1 0.01 [5,6]
Likelihood to mine if total poor 0–1 0.05 [5,6]
Rebel threshold if adult and not a
0–1 0.1 [5,6]
Rebel threshold if adult miner 0–1 0.01 [5,6]
Rebel threshold if minor 0–1 0.01 [5,6]
Distance to diamond mines 0–1 0–1 [14,16]
Remoteness 0–1 0–1 [17,25]
Government control over mines 0–1 0 
Maximum parcel risk 0–1 0–1 Authors estimation
3.2 Agent Behavior
The PECS (Physical conditions, Emotional state, Cognitive capabilities, and
Social status) framework is a cognitive architecture that provides a ﬂexible
framework to model human behavior . Using PECS to implement agent
behavior, Fig. 1provides details on the speciﬁc motives (i.e., needs) and the
set of potential actions available to the agent. The Intensity Analyzer is respon-
sible for determining the action-guiding motive from the set of possible motives.
Two sub-models discussed here – the Needs Model and the Opportunity Model
– are incorporated into the Intensity Analyzer to determine agent behavior.1
The Needs Model. As illustrated in Fig. 1, agents can have three motives:
(1) the need for basic necessities such as food and shelter, (2) the need for
security of employment, housing, and ﬁnancials, and (3) the need to main-
tain the home. These motives represent the two most fundamental levels from
Maslow’s  hierarchy of needs.2Agents meet these needs through a house-
hold income, whether from employment in the formal market, employment in
the illicit diamond market, or employment of a “household” member. While
1A third sub-model, the Identity Model, activates the identity of the agent based on
the outcome of the Needs and Opportunity Models. A detailed description of this
sub-model is provided in the ODD, which can be downloaded from https://www.
2While the Needs Model is responsible for determining the agents’ motive, and as
such, could be called the “Motives Model”, it was instead named after the humanistic
needs theory for which it draws from to highlight the application of theory.
The Geography of Conﬂict Diamonds: The Case of Sierra Leone 339
Set of possible motives
(1) Need for basic necessities,
such as food, water, and
(2) Need for security of
employment, shelter, and
Set of possible actions
(2) Get employment as
(3) Remain employed with
(3) Need to maintain
household (4) Stay home
Fig. 1. Motives and actions via the Intensity Analyzer (adapted from ).
the Needs Model determines the action-guiding motive, the Opportunity Model
helps determine the ﬁnal goal and, subsequently, the ﬁnal action the agent will
The Opportunity Model. Drawing from opportunity-based theories, which
have stressed such factors as the accessibility to resources, the geographic concen-
tration of rebels, and economic factors, agents in the model require opportunity
to join the illicit mining industry or to rebel. The ﬁrst factor of opportunity is
the accessibility to resources. This is driven by three criteria: the presence of
diamond mines, the remoteness of the area, and the level of government con-
trol (or security) surrounding the resource . The second factor is economic in
nature. We use a simple likelihood to mine variable to model this, where the
lower income brackets are the most vulnerable to joining the conﬂict. The ﬁnal
factor is the concentration of rebels within an agent’s “vision”. The more geo-
graphically concentrated the rebels, the easier it is to overcome challenges of
collective action and to mobilize. In the model, if the ﬁrst two factors of oppor-
tunity are met, then there exists the opportunity to mine in the illicit market.
If the third factor is also met, then there exists the opportunity to rebel. In the
case of those agents forced to rebel, however, economic factors are not consid-
ered, as these cases were largely children abducted and violently coerced to join
the conﬂict .
The Action Sequence. An agent can perform one of three activities at each
time step: mine, rebel, or do nothing. If an agent stays home or works in the
formal market, the agent will do nothing (agents “going to work” is not explicitly
modeled). If the agent joins the illicit diamond market, that agent will leave its
current employer and will join the diamond mining industry. If the agent’s income
level was zero, it is increased to one. An agent who becomes a rebel, on the other
hand, does not work for any employer, as the agent is either being forced to rebel
or is seeking to take control of a mining area for purposes beyond that of the
average independent miner.
Agents that are miners or rebels will move on the modeling landscape. These
agents need to be near the diamond mines, but at the same time it is assumed
340 B. Pires and A. Crooks
that they will want to move to a location that will minimize its potential level
of risk as much as possible. Utilizing cost surfaces developed to create the initial
landscape, an agent will move to a parcel within its vision that is closer to
the diamond mines but more remote than its current location. The agent will
continue to move until it cannot ﬁnd any parcel within its vision that would be
better (i.e., closer to the mines and more remote) than its current location.
3.3 Model Output
The model exports a set of comparative statistics, including the number of agents
by a set of labor attributes and income levels. Statistics are collected at the
district-level by time step so that changes in the conﬂict’s dynamics can be
assessed across time and geographic location. The spatial dynamics of the conﬂict
as it evolves across time are observed through the interface during model runs.
4 Simulation Results
This section describes the model results. First, sensitivity testing was performed
to ensure the model was working as intended and to establish qualitative agree-
ment of model results to empirical data of the conﬂict. To determine initial
default parameter values, the model was calibrated by adjusting parameter set-
tings and selecting values based on observed visual results that most closely
replicated the actual conﬂict from a qualitative perspective. Figure 2shows
average intensity levels of rebel activity. Because the model does not simulate
events, intensity here is a function of that proportion of the total population
Fig. 2. A visual comparison of model results to actual events. a: Average model results
using default parameter values. b: Actual event intensity .
Next, two experiments were performed to explore Le Billon’s  theory.
As discussed in Sect. 2, Le Billon  examined four types of conﬂicts and the
environmental factors required for each to emerge. To explore this theory, two
experiments were performed: (1) an experiment where resources are distant
The Geography of Conﬂict Diamonds: The Case of Sierra Leone 341
and government control (i.e., security) is varied, and (2) an experiment where
resources are moved closer to the country’s center and government control is var-
ied (all other parameter values are set to the default values shown in Table 2).
The Impact of the Spatial Dispersion of a Resource on Conﬂict Type
When Resources are Distant. To explore the potential impact on a con-
ﬂict between having distant, diﬀuse resources, which is associated with conﬂicts
of warlordism, and distant, point resources, which is associated with secession
attempts, government control is varied in increments of 0.05 and the diamond
mines, whose relation to the “center” of the country is already distant, are main-
tained at their current locations. Government control of zero represents the min-
imum securities typically found with diﬀuse resources while government control
of one represents the increased security over point resources.
Figure 3illustrates the spatial dynamics of rebel intensity as government
control is increased. Results shown are the average rebel intensity during year
10 of the conﬂict. Figure 3a–b show that at lower levels of government con-
trol, the resulting violence was widespread with some regions experiencing very
high levels of rebel activity. In this case, the resulting spatial dynamics of the
violence was similar to the actual areas where conﬂict was the most intense.
Because of the geographic similarities between the real-world case of warlordism
in Sierra Leone and model results, the model output supports the theory that dis-
tant, diﬀuse resources are associated with conﬂicts of warlordism. As expected,
Fig. 3c–d show that with increasing government control, the intensity of rebels
and the geographic spread of the violence decreased. There are a few unex-
pected results, however. For instance, while minimal increases in government
control were enough at times to stop rebel activity in the south, the conﬂict
looked to be displaced to a district in the north. As government control was
increased systematically to simulate a resource situation going from diﬀuse to
point, rebellion occurred in smaller, more contained areas, often on the bound-
aries of the country. Given these spatial dynamics and the unique geographical
location and size of the conﬂict, a situation of secession may be feasible.
The Impact of the Spatial Dispersion of a Resource on Conﬂict Type
When Resources are Proximate. Freetown is the country’s capital, most
populated city, and main ﬁnancial center . Freetown can thus be considered
the “center” of Sierra Leone. In this second experiment, the diamond mines are
moved to Freetown and results are observed as government control is varied from
zero to one at increments of 0.05. Figure 4shows the spatial dynamics of rebel
intensity as government control is increased. Results shown are the average rebel
intensity during year 10 of the conﬂict.
Environments with proximate, diﬀuse resources are associated with conﬂicts
of mass rebellion or riots near the center of power. When government control
is low, this experiment seeks to simulate this environment, as shown in Fig. 4a.
While rebel activity emerged in the model, it was largely contained to the capi-
tal and its surrounding areas. Although the resources were placed in Freetown,
342 B. Pires and A. Crooks
Fig. 3. Average model results in year 10 when resources are distant. a: Government
control is 0.0. b: Government control is 0.2. c: Government control is 0.4. d: Government
control is 0.6.
Fig. 4. Average model results in year 10 when resources are proximate. a: Govern-
ment control is 0.0. b: Government control is 0.25. c: Government control is 0.35.
d: Government control is 0.45.
The Geography of Conﬂict Diamonds: The Case of Sierra Leone 343
which is located in the district of Western Area Urban, its neighboring district
(Western Area Rural) actually experienced higher levels of rebel activity (see
Fig. 4a–b). Given the geographic location of rebel activity and the spread of the
violence in the model at low levels of government control, these results provide
support to the idea that diﬀuse, proximate resources are associated with rebel-
lion. From Fig. 4, we ﬁnd that only minimal increases in government control are
required to rapidly drop the intensity of rebel activity, supporting the idea that
proximate resources are easier for the government to control. As government
control was maximized, an environment with proximate, point resources is mod-
eled, as shown in Fig. 4c–d. These types of resources are associated with conﬂicts
of state control or coups. A coup would occur in the country’s center of political
power, however, at relatively low government control levels (0.25 and above), no
rebel activity ensues in the capital. Thus, we cannot support or reject the notion
that proximate, point resources are associated with coups.
Since diamonds were discovered in Sierra Leone, the government has been unable
to control the activity and provide residents with the beneﬁts of having the
resource . Through the interplay of ABM and GIS, the model presented
explores Le Billon’s  theory and the impact that the unique environmental
and socioeconomic attributes of a region and its population can have on the
onset of conﬂict. The resulting intensity and spatial characteristics of conﬂict in
the model provided support to Le Billon’s  theory that the spatial dispersion
of a resource can lead to warlordism, secession, and mass rebellion. However, the
model did not implement the necessary detail to support Le Billon’s  claim
that proximate, point resources lead to a coup. Furthermore, in future work,
agent movement could be empirically calibrated to the displacement levels of the
population. Nevertheless, by applying simple behavior we were able to explore
theory and test “what if” scenarios. When an environment is ripe for conﬂict, this
type of model could potentially provide insight into the locations most prone to
conﬂict and the characteristics of a conﬂict. Diﬀerent conﬂict types may require
unique intervention strategies , an important consideration for policy.
1. UN Development Programme: Case Study Sierra Leone: Evaluation of UNDP
Assistance to Conﬂict-Aﬀected Countries. UNDP, Evaluation Oﬃce, New York,
2. Leoa, I.: Youth marginalisation and the burdens of war in Sierra Leone. In: Freedom
From Fear, pp. 26–28 (2010)
3. Goreux, L.: Conﬂict Diamonds, Africa Region Working Paper Series, No. 13. The
World Bank, Washington, DC (2001)
4. Le Billon, P.: The political ecology of war: natural resources and armed conﬂicts.
Polit. Geogr. 20, 561–584 (2001)
344 B. Pires and A. Crooks
5. Fearon, J.D., Laitin, D.D.: Ethnicity, insurgency, and civil war. Am. Polit. Sci.
Rev. 97, 75–90 (2003)
6. Lujala, P., Gleditsch, N., Gilmore, E.: A diamond curse? civil war and a lootable
resource. J. Conﬂ. Resolut. 49, 538–562 (2005)
7. Axelrod, R.: A Model of the Emergence of New Political Actors. Working Papers
93–11-068, Santa Fe Institute, Santa Fe, NM (1993)
8. Epstein, J.M.: Modeling civil violence: an agent-based computational approach.
PNAS 99, 7243–7250 (2002)
9. Bhavnani, R., Miodownik, D.: Ethnic polarization, ethnic salience, and civil war.
J. Conﬂ. Resolut. 53, 30–49 (2009)
10. Miodownik, D., Bhavnani, R.: Ethnic minority rule and civil war onset how iden-
tity salience, ﬁscal policy, and natural resource proﬁles moderate outcomes. Conﬂ.
Manage. Peace Sci. 28, 438–458 (2011)
11. Pint, B., Crooks, A.T., Geller, A.: Exploring the emergence of organized crime
in Rio de Janeiro: an agent-based modeling approach. In: 2010 Second Brazilian
Workshop on Social Simulation (BWSS), pp. 7–14. Sao Paulo, BR (2010)
12. Luke, S., Cioﬃ-Revilla, C., Panait, L., Sullivan, K., Balan, G.: MASON: a multi-
agent simulation environment. Simulation 81, 517–527 (2005)
13. M¨uller, B., Bohn, F., Dreßler, G., Groeneveld, J., Klassert, C., Martin, R., Schluter,
M., Schulze, J., Weisse, H., Schwarz, N.: Describing human decisions in agent-based
models ODD+D: an extension of the ODD protocol. Environ. Model. Softw. 48(1),
14. GADM, Global Administrative Areas. http://www.gadm.org/countryres. Accessed
Jan 2013 (2009)
15. Oak Ridge National Laboratory. Landscan Global Population Dataset. OakRidge
National Laboratory, Oak Ridge, TN (2007)
16. Gilmore, E., Gleditsch, N., Lujala, P., Rod, J.K.: Conﬂict diamonds: a new dataset.
Conﬂ. Manage. Peace Sci. 22, 257–272 (2005)
17. OpenStreetMap, CloudMade-Map data CCBYSA 2010 (2010). http://downloads.
cloudmade.com/africa/sierra leone#downloads breadcrumbs. Accessed Dec 2010
18. Statistics Sierra Leone, Annual Statistical Digest 2005/2006. Statistics Sierra
Leone, Freetown, Sierra Leone (2006a). http://statistics.sl/ﬁnal digest 2006.pdf.
Accessed Jan 2014
19. Statistics Sierra Leone, Final Results: 2004 Population and Housing Census. Sta-
tistics Sierra Leone, Freetown, Sierra Leone (2006b). http://www.sierra-leone.org/
Census/ssl ﬁnal results.pdf. Accessed Apr 2014
20. Schmidt, B.: The Modelling of Human Behaviour. BE: Society for Computer Sim-
ulation International, Ghent (2000)
21. Maslow, A.H.: Motivation and Personality. Harper, New York (1954)
22. Braima, S.J., Amara, P.S., Kargbo, B.B., Moserey, B.: Republic of Sierra Leone
2004 Population and Housing Census: Analytical Report on Employment and
Labour Force. Statistics Sierra Leone, Freetown, Sierra Leone (2006)
23. Thomas, A.C., MacCormack, V.M., Bangura, P.S.: Republic of Sierra Leone: 2004
Population and Housing Census: Analytical Report on Population Size and Dis-
tribution Age and Sex Structure. Statistics Sierra Leone, Freetown, Sierra Leone
24. BBC, Children of Conﬂict. BBC World Service (2014). http://bbc.in/1umgSiP.
Accessed Apr 2014
25. Commonwealth Department of Health and Aged Care: Measuring Remoteness
Accessibility/Remoteness Index of Australia (ARIA). Dept. of Health, Canberra,
The Geography of Conﬂict Diamonds: The Case of Sierra Leone 345
26. Le Billon, P.: Diamond wars? conﬂict diamonds and geographies of resource wars.
Ann. Assoc. Am. Geogr. 98, 345–372 (2008)
27. Campbell, G.: Blood diamonds: Tracing the Deadly Path of the World’s Most
Precious Stones. Westview Press, Cambridge (2004)
28. Le Billon, P.: Fuelling War: Natural Resources and Armed Conﬂict. Routledge,