Resources and Governance in Sierra Leone’s Civil War

Abstract We empirically investigate the role of natural resources, and governance in explaining variation in the intensity of conflict during the 1991–2002 civil war in Sierra Leone. As a proxy for governance quality we exploit exogenous variation in political competition at the level of the chieftaincy. As a proxy for resources we use data on the location of pre-war mining sites. Our main result is that neither governance nor resources robustly explains the onset or duration of violence during the civil war in Sierra Leone.


Introduction
Over two-thirds of African countries experienced an episode of civil conflict in the past decades and the search for determinants of the onset, duration and intensity of conflict remains an important topic of debate. One dominant strand in the literature focusses on the economic motives for groups to enter into conflict. Participants in armed conflicts are motivated by material gains or a desire to improve their economic situation, such as the grabbing of natural resource rents. In the literature on the resource curse, this has been referred to as the 'greed perspective'. Other reasons for engaging in conflict have to do with identity, rather than income. This includes concerns about injustice, lack of political rights, social marginalisation, and ethnic or religious divisions. The relative importance of these competing explanations remains ill understood and controversial, and presumably varies from one location to the next. This paper seeks to explain how natural resources and governance quality affect conflict intensity in the civil war that ravaged Sierra Leone between 1991 and 2002. Bad governance in this context implied the exclusion of certain social groups in the development process. Hence we argue that governance quality is correlated with grievances (but we do not deny that alternative interpretations might exist). We analyse spatial and temporal patterns in the conflict data, and link them to exogenous variation in the quality of governance at the chiefdom level (based on the intensity of competition for the chieftaincy) and georeferenced locations of pre-war (diamond) mines. Sierra Leone is a poster child of the resourcebased perspective, and its so-called 'blood diamonds' feature prominently in many essays on African conflict. For instance, Collier, Hoeffler, and Rohner (2009, p. 13) note: 'The most celebrated cases are the diamond-financed rebellions in Sierra Leone and Angola'. However, (other) academics have emphasised and implicated the many weaknesses in Sierra Leone's institutional domain. Authors like Richards (2005, p. 588) point out that 'institutional failure, and not criminal 'greed', should be regarded as the motor [of intensity against the backdrop of an intense and prolonged war. We cannot exclude the possibility that diamonds or bad leadership (at the macro level), invited or shaped the war across all chiefdoms. This paper is organised as follows. In Section 2 we briefly summarise the literature on the determinants of conflict, focusing on analyses that include resources and institutions. Section 3 presents the context, introduces our data, and outlines our identification strategy. This section contains evidence from colonial times to support the identification strategy. Section 4 presents the empirical results, showing that neither resources nor governance affect the intensity of local violence. The conclusions ensue.

Resources, Governance, and Conflict
A large and rapidly growing literature in economics and political sciences studies the causes and consequences of civil war (refer to Blattman & Miguel, 2010, for a survey). A recent overview focusing on the multifaceted role of natural resources as a determinant of conflict is provided by Nillesen and Bulte (2014). It is impossible to do justice to this literature on these pages, but we will try to summarise some key lessons, setting the stage and motivating our own analysis.
For several years, the leading explanation for conflict were the so-called 'greed' and 'grievances' hypotheses. 1 The work of Collier and Hoeffler (1998, 2004Collier, Hoeffler, & Söderbom, 2004;Collier, Hoeffler, & Rohner, 2009) has been extremely influential in advancing the former perspective. Among other things, they document an inverted U-shaped relationship between natural resource exports and the incidence of conflict. This is explained by the interaction between various effects. On the one hand, resource rents constitute a 'prize' that rebels might want to grab, and facilitate or finance on-going rebellions. But resource rents also enable incumbent governments to suppress the opposition (see also Humphreys, 2005;Papaioannou & van Zanden 2015, Ross, 2004. The opportunity costs of rebelling also feature prominently in such an economic framework, linking the incidence of violence to public goods provision (and allocative decisions by, as well as capacity of, the statesee Basedau & Lay, 2009). The empirical evidence supporting the resource perspective is mixed, and the effects of the presence or exports of commodities like oil and diamonds are more subtle and conditional than envisaged in early studies (for example, Lujala, Gleditsch, & Gilmore, 2005;Ross, 2004;but also Elbadawi & Sambanis, 2002;Fearon & Laitin, 2003). Indeed, several recent studies suggest that the impact of resources on conflict is conditional on, for example, income (Østby, Nordås, & Rød, 2009) and the physical location of the resource (Lujala, 2010).
While it is easy to use a cross-section model and correlate various measures of resource richness to either the onset, incidence or duration of conflict, it is notoriously difficult to jump to causal inference. In particular, potential problems with omitted variables remain. 2 In an effort to attenuate such concerns, analysts have estimated fixed-effects panel models, often leveraging identification from exogenous variation in the prices of key primary commodities. While also producing mixed evidence, these models tend to (further) erode support for the resource curse hypothesis. For example, Brückner and Ciccone (2010) find that the outbreak of violence is likely to follow a downturn in commodity prices. Similarly, Bazzi and Blattman (2013) find little evidence that price spikes initiate conflict. In contrast, they argue that higher commodity prices are associated with an increased likelihood of the cessation of violence. Such a finding runs counter to the perception of rebels seeking to grab prizes, but instead suggest that resource rents may increase state capacity (enabling the provision of public goods) or increase the opportunity costs of conflict (through enhanced employment in the primary sector). Other analyses seek to identify causal effects by focusing on (exogenous) resource discoveries. For example, Cotet and Tsui (2013) study the discovery of oil fields and find they do not trigger conflict. 3 The ambiguity of this literature is rather at odds with insights from case studies, or studies focusing on specific countries such as Colombia (Angrist & Kugler, 2008;Dube & Vargas, 2013), Sierra Leone (Bellows & Miguel, 2009;Humphreys & Weinstein, 2008), or Sudan (Olsson & Fors, 2004). These studies, together with others that seek to better understand the perspective of prospective rebels (for example, Weinstein, 2005), provide support for the idea that certain resources can play a role in initiating or sustaining conflict. 4 This micro evidence is corroborated by robust results of a recent disaggregated study of the dynamics of conflict across the African continent (Arezki et al., 2015;Berman et al., 2014). 5 Starting from the premise that conflicts have a spatial dimension, and that country-year variation in conflict status may be too coarse to capture key features, both Berman et al. (2014) and Arezki et al. (2015) adopt a grid-based approach to investigate if mineral mines invite conflict. The studies arrive at opposing conclusions based on the time frame under study. Where Berman et al. (2014) find that minerals invite conflict (and that such conflicts may later spread to other parts of the country), Arezki et al. (2015) extend the time frame and find the evidence disappears.
To sum up, the literature on the resource-conflict nexus provides mixed signals about the impact of natural resources on violence. The leading alternative explanation is related to governance, typically associated with relative deprivation, social exclusion or marginalisation of specific social groups. In his seminal work, Gurr (1970) argues how relative deprivationthe tension between a person's actual state and her beliefs about what should be achievabledetermines the potential for collective violence. Ample anecdotal and case study evidence suggests a clear link between relative deprivation and conflict. For example, considering the case of Sierra Leone, the writings of Keen (2005), Richards (2005) and Peters (2006) clearly sketch how the disconnect between an urban elite and rural hinterland, combined with exploitative agrarian and patronage institutions, has been conducive to widespread support for societal transformationeven through violence (see below).
But capturing such ideas in an econometric framework has been far from straightforward. Early efforts have tried to capture social and institutional variables through aggregate inequality measures (such as Gini coefficients), but largely failed to produce significant associations (for example, Collier & Hoefler, 2004;Fearon & Laitin, 2003). Other work has focused on so-called horizontal inequality (based on inequality coinciding with identity-based cleavages, see Østby et al., 2009;Stewart, 2000), or on ethnic diversity and conflict (for example, Esteban, Mayoral, & Ray, 2012;Horowitz, 1985;Montalvo & Reynal-Querol, 2005). Østby et al. (2009) adopt a disaggregated approach to studying (horizontal) inequality and conflict. The latter study finds that both inter-and intra-regional inequalities increase the risk of violence, suggesting that the quality of local governance is a key factor explaining conflictbad governance tends to translate into poor economic performance (say, through inadequate provision of public goods) and does little to ameliorate local income differentials. This is consistent with the interpretation of Fearon and Laitin (2003) that state capabilities are at the heart of many crises of violence. It also naturally links the literature on grievances and conflict to the literature on the quality of governance as determined by precolonial factors (for example, Michalopoulos & Papaioannou, 2013), experiences during the colonial era (Acemoglu, Johnson, & Robinson, 2001;Mamdani, 1996) or postcolonial reconstruction efforts (see Casey, Glennerster, & Miguel, 2012;King & Samii, 2014).

Conflict in Sierra Leone
Sierra Leone suffered from a civil war between 1991 and 2002. Over half of the population was displaced, an estimated 50,000 Sierra Leoneans were killed, and thousands were victims of amputations, rapes, and assaults (Smith, Gambette, & Longley, 2004). Explanations for the civil war in Sierra Leone have mainly (and perhaps too simplistically) centred around resource wealth and local grievances. Some authors point to the prominent role of extraction and smuggling of (blood) diamonds in starting or sustaining the conflict. Keen (2005, p. 212) documents how armed groups participated in diamond smuggling during the conflict, and argues that the control of diamond-rich areas was an important objective for warring groups as 'battles were largely restricted to the areas with the richest diamond deposits'. The role of diamonds in shaping the dynamics of the war also featured prominently in the case against the former president of Liberia, Charles Taylor, at the Special Court for Sierra Leone (SCSL), who allegedly aided the RUF rebel group.

Resources, governance and conflict 281
Other scholars argue that the insurgency was principally motivated by bad governance. The dismal state of governance at the national level in Sierra Leone is extensively discussed by Reno (1995). But governance issues are also manifest in the rural areas, governed by an intricate system of patronclient relationships, spearheaded by paramount chiefs. Individuals are dependent on these highly exclusionary traditional institutions if they want to access property or gain political rights. Enforced community labour and the lack of opportunities created by this system resulted in a large class of excluded, low-status individuals (mostly young men, descending from slaves) that felt disenfranchised and who believed they had little stake in economic development (for example, Fanthorpe, 2001;Richards, 1996Richards, , 2005Sawyer, 2008). Matters are worsened by abuse of the local judicial system to advance the interests of the privileged class (Mokuwa et al., 2011). Moreover, in the decades before the war, some chiefs enriched themselves through illicit diamond deals, while doing little to provide public services such as health care and education (Bratton, Van de Walle, & Lange, 1997;Reno, 1995;Richards, 1996). Considering this evidence, Sierra Leone seems to fit the conventional wisdom that African chiefs may be unaccountable despots (Mamdani, 1996), with their position of authority fortified by colonial systems of indirect rule allowing them to avoid accountability to their local constituencies (Boone, 2003). Such (de facto) chiefly powers have persisted over time through systems of clientelism (Acemoglu & Robinson, 2008). Richards (1996) emphasises that the initial motivations of the main rebel group (the RUF) were idealistic and guided by a strong sense of political grievances related to the perceived failings of the corrupt institutional structure. RUF propaganda complained about exploitation, and railed against 'the raping of the countryside to feed the greed and caprice of the Freetown elite and their masters abroad' (Richards, 1996, p. 27). RUF propaganda also emphasised the almost feudal relationships in the class-based agrarian society that characterises the hinterland, as is evident from their slogan 'no more master, no more slave!'. Indeed, grievances in rural Sierra Leone are more likely to be associated with governance and class-based production relations than with ethnic tensions between the countries major ethnic groups (the Mende and Temne). For example, Glennester, Miguel, and Rothenberg (2013) document that ethnic issues are not important for the provision of public goods.

Dynamics of the Sierra Leonean War
The civil war in Sierra Leone lasted between 1991 and 2002, and eventually engulfed all 149 chiefdoms of the country. However, there is considerable variation in conflict intensity across time and space. Figure 1 shows the number of conflict events such as deaths and injuries over time (see Supplementary material). 6 Conflict dynamics across space are mapped in Figure 2. Violence peaked on several occasions. There was much violence in the eastern part of the country, in 1991, when RUF rebels entered Sierra Leone from Liberia. The violence later spread north and west towards Kenema, Bo and the Freetown peninsula. Subsequent peaks in violence followed in 1994-1995, and again in 1997. In January 2002, the war was declared over, and an internationally-brokered peace agreement was signed. In what follows, we exploit the variation across space and time to examine how resources and governance relate to conflict in Sierra Leone.

Data
Conflict. Our main dependent variable is conflict intensity, derived from two sources. Panel A of Table 1 summarises our data. We use data from a nationally representative household level survey conducted by the Institutional Reform and Capacity Building Project (IRCBP) in 2007. IRCBP was a project funded by the Wold Bank to assist the government of Sierra Leone in the decentralisation process. The dataset contains data on 6345 randomly selected households from within 635 randomly selected villages across Sierra Leone's 149 chiefdoms. 7 Respondents were asked about a range of war experiences, including death of family members, maiming, fleeing, being a refugee and the destruction of household assets. We use this information to construct an index at the chiefdom level, Resources, governance and conflict 283 indicating the average number of events experienced by households during the war. On average, households experienced 2.4 of these events. Importantly, while this dataset provides detailed information on the exposure of households to conflict, it does not contain a temporal dimension, so it is not useful to distinguish between different stages of the conflict.
Time-variant data is available from the 2004 No Peace Without Justice (NPWJ) conflict mapping project (Smith et al., 2004). The project aimed to help identify human rights violations and later helped establish the Special Court for Sierra Leone. As part of the process NPWJ chronologically and geographically mapped all conflict events for Sierra Leone during the war. Data were collected from key persons throughout the country, and supplemented with open source materials (see Smith et al., 2004, for further details). The NPWJ report contains data on 1997 conflict events. We create an annual conflict event variable counting conflict events per chiefdom. Specifically, for each year we sum observations that involve the killing, raping, maiming or abduction of people, and the burning of houses. Averaging conflict events, there were on average eight conflict events per chiefdom, per year. The total range of this variable is from 0 to 40 events. Correlation between the IRCBP and NPWJ data is modest at 0.2 (p = 0.02).

Governance.
To proxy the quality of bad governance, we use a measure of power of the paramount chief, created by Acemoglu, Reed, and Robinson (2014b). 8 In Sierra Leone, chiefs must come from so called 'ruling families' (or ruling houses). The number of such families is small and displayed in Figure 3: the average number per chiefdom is four, and across the Chiefdoms the number ranges from one to 12. Only selected members from these elite families were officially recognised by British colonial authorities in the nineteenth century as legitimate leaders of the chieftaincy. This institutional arrangement is clearly undemocratic, but was nevertheless perpetuated after independence. Acemoglu et al. (2014b) argue that the number of ruling families is a useful proxy for the intensity of political competition, as it determines the number of potential challengers for the chieftaincy. Political competition is a key factor influencing the quality of governance. The main hypothesis is that as competition for political power intensifies, the spoils of governing will have to be shared more widely in order to garner sufficient support, so that policies tend to be more inclusive. The number of ruling families per chiefdom is summarised in Figure 3. This hypothesis is borne out by the data. Acemoglu et al. (2014b), after demonstrating that the number of ruling families is a source of exogenous variation in local political power, proceed to show a reverse relation between political power and the provision of public goods (or specific development outcomes). Following Acemoglu et al. (2014b), we interpret the number of ruling families as a proxy for the quality of governance. We examine whether it explains variation in the intensity of conflict, assuming that the number of ruling families is related to local grievances (through the degree of 'inclusiveness' of policy making). If, as Richards (1996) argues, rebels were motivated by abusive leaders, we expect more conflict in places with a smaller number of ruling families.
Resources. Following the 'blood diamond' narrative, we take the number of diamond mines as our proxy for greed-based explanations for conflict. The data comes from the Armed Conflict Location and Event Data dataset from the Peace Research Institute Oslo (PRIO), and contains all pre-war registered diamond mining sites. Figure 4 provides mining sites, and demonstrates these were mainly clustered in the eastern provinces. However, there are also mines in the northern areas.

Resources, governance and conflict 285
Controls. To improve the precision of our estimates and to control for factors correlated with both conflict and resources or governance, we also introduce a vector of control variables in some models. We mostly draw from the IRCBP data and use variables commonly used in the conflict literature (see Collier & Hoeffler, 1998). Unfortunately, like the IRCBP conflict data, we lack panel data on these variables. As a measure of ethnic fractionalisation, we use a Herfindahl index (one minus the sum of squared fractions of each of the 18 ethnic groups, or the probability that two randomly drawn individuals are from different ethnic groups). Religious fractionalisation is created in the same manner for all 15 religions. As a proxy for per capita income we use an asset index. Respondents were asked to indicate which assets they possessed, from a list of 10 assets that included mobile phones, generator, television, bicycle, and so forth. As a proxy for education we use a household level dummy indicating whether the household head had any education. To control for ease of movement within a chiefdom we use road density (km road per square km area) from the Geographic Information System (GIS) data. Finally, we control for chiefdom surface areas, as incidence and number of conflicts within a chiefdom may be correlated with its size. 9

Identification
Our ambition is to explain the spatial variation in the intensity of conflict throughout the war. However, we start with a simple cross-section model based on aggregate data, using both crosssection data from the IRCBP set as well as aggregate conflict data from the NPWJ data: where C i refers to our measure of conflict events for chiefdom i throughout the 1991-2002 war, with i = 1, . . .,149, Chiefs i and Mines i are time-invariant binary variables capturing, respectively, whether the chief in chiefdom i is 'strong' and whether the chiefdom contains known diamond sites before the war started. We define 'strong chiefs' as chiefs ruling chieftaincies in which the number of ruling families is smaller than the average value (that is, chiefdoms with less than four ruling families) 10 ;ε it is an error term. In some models, we control for a range of variables plausibly correlated with violence, X i and include district fixed effects, α D (D = 1,. . .,12) to control for common factors at the district level, and zoom in on intra-district variation in resources and governance. Next, we explore determinants of conflict during different stages of the conflict. We estimate the following panel model: where C it refers to our measure of conflict events for chiefdom i in year t, with t = 1991,. . .,2001. 11 To examine whether the impact of resources and governance varies over the course of the war, we now interact our chief and mine variables with a vector of year dummies, T t . Again, we estimate Equation (2) with and without our set of controls, and district fixed effects, α D . Figure 2 illustrates how the conflict started in the Gola Forest region, in the east of Sierra Leone, and subsequently spread to other chiefdoms. Augmenting the panel model above, we also control for spatial autocorrelation by including conflict events in neighbouring chiefdoms. Specifically, we estimate models containing a spatial lag, P j2NðiÞ C j;tÀ1 , capturing the sum of all (lagged) conflict events in those chiefdoms j bordering chiefdom i (see also Van der Windt & Humphreys, 2015;Zhukov & Stewart, 2012). The spillover term allows us to test whether conflict diffuses over space, and attenuates concerns about spurious correlations brought about by geographical factors shaping both clusters of governance quality or resource availability, as well as the intensity of violence. In addition, to capture the persistence of conflict, we also add a measure of lagged conflict in chiefdom i: C it-1 . In sum, we estimate the following model: Finally, we create a new dependent variable, C o it , indicating each time a conflict starts (zero else) in chiefdom i, and estimate a conflict onset model. Since conflict is duration-dependent, we now add a count variable indicating the number of years a conflict event lasted: C d it . We also add its squared term (see Beck, Katz, & Tucker, 1998). 12

A Historical Prelude to Grievances and Chiefly Power
Based upon data from archival research in the National Archives in London, 13 we show evidence from colonial times to support the interpretation that the number of chief families is related to the quality of governance. From the archival data we construct a measure of grievances at the chiefdom level between 1920 and 1940, capturing the frequency with which local riots against the chief were sufficiently serious to draw the attention of the Britishoccasionally inviting a (military or administrative) response. There were on average two such events over the 20-year period per chiefdom. When regressing this grievance variable on the number of ruling families, we find a strong, statistically significant, negative relationship. Specifically, the coefficient of the ruling family variable equals −0.36 (p-value = 0.01). There is also ample anecdotal evidence in the archives to link powerful paramount chiefs to the abuse of power. One (British) district commissioner stated 'The Kpaka chiefdom [only one ruling family] of the Pujehun district, has for many years been misgoverned' and that '. . . chief Momo Rogers has proved himself to be a most unsatisfactory and unjust ruler almost from the first years of his tenure' (CO267/636). The charges against this chief were numerous but centred around the fact that the chief had been enriching himself by levying forced labour, extracting illegal fines and forcing contributions from his people. The acting governor reported that 'the chief had made himself so unpopular among the people of the chiefdom that there has developed an atmosphere of considerable strain and tension . . . and that severe disturbance of the peace is considerable' (CO260/55). In several cases, the misrule of chiefs was so severe that colonial officials intervened in local affairs to restore order by deposing the chief, despite the fact that they had strict orders not to do so (CO 270/49); for instance in 'the Imperri chiefdom [two ruling families] has for some years shown active discontent against its paramount chief . . . until the government found it necessary to intervene and steps for the deposition of the chief were taken'. Table 2 reports results for the cross-chiefdom analysis (coefficients are standardised). In columns (1)-(3) we use data from the IRCBP data set, and in columns (4)-(6) we use aggregated conflict events as reported in the NPWJ dataset. Columns (1) and (2) provide early support for the greed as well as the bad governance perspective, as both the presence of diamond mines and the strong chief dummy are correlated with variation in local conflict intensity. Consider column (1). Chiefdoms with strong chiefs are associated with a 0.32 standard deviation change in victimisation, and chiefdoms with mines have a 0.67 standard deviation increase in victimisation. These are sizable effects (a Wald test reveals that the two coefficients are not significantly different from each other: p-value equals 0.26). However, the results suggest, resources and governance do not robustly explain variation within Resources, governance and conflict 287 districts. When we include district fixed effects (column 3), the coefficients shrink, and the coefficient of the governance proxy even switches sign. Similar patterns emerge when we use the aggregate NWPJ conflict variable. Across columns (4)- (6), the governance is not significant (and, indeed, of the 'wrong sign'), but the diamond variable is. Overall, Table 2 provides some support for the role resources played in the conflict. However, it is well-known that aggregate data may be too coarse to detect meaningful effects when there is heterogeneity in the underlying data. Specifically, governance or resources may matter during specific stages of the warinviting conflict, or prolonging itand such effects may be masked in a cross-section analysis that lumps all conflict events together. To probe this important issue we now turn to our panel data and report our main results in Table 3.

Empirical Results
Moving from column (1) to (5), we present the outcomes of increasingly complex models. Column (1) is a parsimonious specification including only diamond mines, our governance proxy, time interaction effects, and a vector of year dummies; column (2) includes (time-invariant) chiefdom Resources, governance and conflict 289 level controls; column (3) introduces district fixed effects; column (4) introduces the spatial lag and lagged dependent variable; and in column (5), we report estimates of our conflict onset model. Our main result is that neither governance nor resources robustly explains the onset or duration of violence during the civil war in Sierra Leone. Neither variables are significant in any model as level variables, so there is no evidence of a robust effect on conflict intensity spanning the entire war. In addition, none of the interaction terms for early periods (1991 and 1992) enter significantly.
Our results also do not suggest that conflict motivated by the presence of diamonds or poor governance vary over time. The interaction terms with mines tend to be insignificant throughout. The other vector of interaction terms (strong chiefs multiplied by the year dummies) also reject the hypothesis that bad governance prolongs conflict. None of the interaction terms is significant, and the 2000 interaction terms again have the 'wrong sign'. If anything, this finding suggests a reduced likelihood of conflict starting in areas with more authoritarian chiefs. The only interaction term that consistently enters significantly across the incidence models (columns 1-4) is the product of the mining dummy and the 1998 year dummy. Only in that year do we observe that conflict was more intense in diamond chiefdoms than in non-diamond chiefdoms. We are hesitant to take this as evidence, as it need not be surprising that one of our 20 interaction terms enters significantly at the 5 per cent level.
A few additional observations are noteworthy. First, we find that conflict was less intense in ethnically fragmented chiefdoms. This supports claims in the literature that ethnic tensions were not a root cause of the conflict in Sierra Leone. In contrast, there is some mixed evidence for the hypothesis that religious fractionalisation is associated with more intense violence. We also find that violence tends to persist (column 4), the coefficient on lagged conflict events in a chiefdom is positive and significant. In addition, we find that the duration of conflict matters for the probability of conflict to start (again) (column 5), the coefficient on conflict duration is significantly and positively correlated with conflict onset.

Conclusion
The civil war in Sierra Leone has ended more than a decade ago, and the most pressing current debates concerning conflict and resources are about foreign investment in mineral extraction and farming. Nevertheless, Sierra Leone remains an important case study in the growing literature on resources, governance and civil war. As a poster child for both 'greed' and the 'grievances' hypotheses, the conflict literature stands much to learn from studying Sierra Leone's history. Resources also remain the corner stone of Sierra Leone's economic development in the future, and concerns about the quality of (local) governance are still widespread.
In this study we put two simple explanations to the test. We explore whether the dynamics of local conflict during the war was correlated with the presence of diamonds or with a measure of low-quality governance. We exploit a large nationwide survey documenting how the intensity of local conflict varied across the years during the conflict, and supplement this data with data on the location of diamond mines, and with data on exogenous variation in the (potential) abusive powers of the chieftaincy. The latter data comes from Acemoglu et al. (2014b), who leverage the unique nature of institutions in Sierra Leone, where a chief must come from one of the ruling families originally recognised by British colonial authorities.
We find no support that local measures of resources or bad governance are robustly related to the intensity of local conflict. Our panel results indicate there is no correlation between the presence of diamonds or the quality of local governance, and the onset or persistence of conflict in Sierra Leone's civil war.
However, it is important to place these results in perspective. In particular; while we find that diamonds and governance do not explain variation in conflict intensity across chiefdoms, this is not the same as arguing that governance or diamonds have nothing to do with the civil war. The extremely unequal sharing of diamond rents during the reign of the (national) Shaka Stevens government (and later the Joseph Momoh government) could have created frustration and fuelled dissatisfaction with the government across all Chiefdoms. Similarly, diamonds may have helped the RUF to fund its conflict activities in all Chiefdomsnot just the ones where mining activities were concentrated. 14 With this caveat in mind, we believe our findings present a challenge to simple theories of conflict.