Why Do Some Oil Exporters Experience Civil War But Others Do Not? A Qualitative Comparative Analysis of Net Oil-Exporting Countries

Transcription

1 THIS IS A DRAFT! PLEASE DO NOT DISTRIBUTE OR CITE WITHOUT AUTORS PERMISSION Why Do Some Oil Exporters Experience Civil War But Others Do Not? A Qualitative Comparative Analysis of Net Oil-Exporting Countries Paper prepared for presentation at the SGIR 7th Pan-European Conference on IR, Stockholm, 9-11 September 2010 Matthias Basedau & Thomas Richter 1 German Institute of Global and Area Studies (GIGA) According to quantitative studies, oil seems to be the only resource that is robustly linked to civil war onset. However, recent debates on the nexus of oil and civil war have neglected two major considerations: there are a number of historically peaceful oil-rentier states, and there have been few efforts to explain why some oil-exporting countries have experienced civil war and others have not. Methodologically, the debate has been dominated by research using either quantitative methods or case studies, with little genuine medium-n comparison. This paper aims to fill this gap by studying the conditions of civil war onset among net oil exporters using (crisp-set) Qualitative Comparative Analysis (csqca), a technique best used to study samples between 10 and 50 cases. QCA is also more suited to identifying necessary and sufficient conditions. At the same time, this technique looks at country-specific configurations but outperforms case studies in yielding generalizable results. Considering a sample of 44 net oil exporters between 1970 and 2008, we test pertinent factors such as the abundance (per capita) and dependence of oil, settlements of excluded ethnic groups in oil reserve areas and outside protection. The results point to a combination of necessary and sufficient conditions that has been largely ignored until now: although low rents per capita and a lack of outside protection are necessary conditions for civil war onset among oilexporting countries, a combination of high dependence and low abundance sufficiently explains the onset of civil war. 1 German Institute of Global and Area Studies, Neuer Jungfernstieg 21, Hamburg, Germany; Matthias Basedau is at the Institute of African Affairs, basedau@giga-hamburg.de; Thomas Richter is at the Institute of Middle East Studies, richter@giga-hamburg.de. 1

2 1. Introduction 2 According to a large body of quantitative literature, oil seems to be robustly linked to the onset of civil war 3 (Ross 2004a; Hegre und Sambanis 2006; Dixon 2009). However, recent debates on the nexus of oil and civil war have neglected two major considerations: there are a number of historically peaceful oil-rentier states, and there have been few efforts to explain why some oil-exporting countries have experienced civil war and others have not. Methodologically, the debate has been dominated by research using either quantitative methods or case studies, with little genuine medium-n comparison. This paper aims to fill this gap, researching why some oil-exporting countries experience civil war while others do not using a relatively new methodological approach: (crisp-set) Qualitative Comparative Analysis (csqca), a technique best used to study samples between 10 and 50 cases. QCA is also more suited to identifying necessary and sufficient conditions than most existing quantitative techniques. At the same time, this technique looks at countryspecific configurations but outperforms case studies in yielding generalizable results. The paper proceeds as follows. We first review the literature on the resource-conflict link, pointing to specific deficiencies. We then outline our general empirical strategy, starting with a short introduction to csqca. Subsequently, we present our dependent variable (civil war) and our independent variables (oil abundance, oil dependence, overlap of ethnic exclusion and oil reserves, and outside protection), which are then tested using a sample of 44 net oil exporters between 1970 and After outlining our empirical strategy, we perform the analysis and discuss the results. Finally, we draw conclusions and highlight opportunities for future research. 2. Literature Review 2.1 Theoretical Approaches Collier and Hoeffler s (2004; initially 1998) influential work on greed and grievance has inspired many studies on the link between resources and conflict. Collier and Hoeffler (2004) argue that an abundance of primary commodities increases the likelihood of civil war onset by 2 We are grateful to Georg Strüver for excellent research assistance. 3 If not indicated otherwise, civil war denotes an armed conflict (UCDP/PRIO-definition) that has produced at least 1,000 battle-related deaths in a given year (see Gleditsch, Wallensteen, und Eriksson 2002). 2

3 providing the opportunity for armed rebel activity and the related motive of greed rather than by spurring conflict-promoting grievances, such as the political and economic deprivation experienced by, for instance, ethnic or religious groups. These ideas have been further developed, extended and modified in the literature. Generally, natural resources can promote violence through three major causal mechanisms (Humphreys 2005; Ross 2004b; see also Le Billon 2008). 1) Motivation to take up arms may result from resource-related grievances, such as ecological damage or the withholding of resource revenues; costs and benefits relating to resources are the driving forces of conflict. 2) Resources also provide the opportunity for conflict by making rebellion or warfare in general financially (or militarily) feasible, particularly through the lootability of resources. 3) Resources may make indirect mechanisms work, directly providing neither motive or opportunity but exerting a detrimental influence on other areas such as state institutions (the weak state ) and socio-economic development ( Dutch disease ), which in turn makes civil war more likely. All of these mechanisms involve numerous contextual conditions, particularly resourcespecific conditions (Ross 2008; Basedau 2005). In fact, resources are a fairly imprecise concept. They have several characteristics that may influence their impact on peace and war. Among these characteristics are the resource type (oil, diamonds or other) and the mode of extraction, which affects the feasibility of rebellion. As Le Billon (2001; see also Auty 2001) notes, the exploitation of so-called distant and diffuse resources such as alluvial diamonds, timber or drugs can hardly be controlled by the central government. Hence, rebels can loot them more easily than deep-shaft gems or off-shore oil, which would require sophisticated technical know-how. 4 Moreover, a country s resource dependence and abundance are not identical (Ross 2006, 266; De Soysa 2002, 8-9). Dependence means that rents from resources are the most important source of income relative to other value-adding activities, whereas abundance or wealth refers 4 Such point resources are more likely to trigger power struggles over the control of the central state or, if concentrated in certain, presumably peripheral regions (referred to as point and distant regions), secessionist uprisings (Le Billon 2001, 31). 3

4 to the absolute resource rents available in per capita terms (or relative to global reserves or production). It can be easily illustrated that the values of these two variables may differ substantially. 5 Dependence and abundance have different implications for the likelihood of civil war. Resource dependence may be especially violence-enhancing at higher levels. Countries are vulnerable to price shocks, which in turn create economic crisis and (thus) make conflict more likely (Basedau und Lay 2009). Moreover, resources may foster the onset of internal conflict only in economies with a monolithic structure that offers limited alternative income sources. In contrast, higher resource abundance per capita will probably foster peace when governments employ particular distributive policies. According to rentier state theory (Mahdavy 1970; Luciani 1987; Smith 2004; Snyder und Bhavnani 2005), which has been largely disregarded in the debate over the resource-conflictlink, governments can use resource revenues for costly policies such as large-scale redistributional schemes and security apparatuses, buying off potential dissidents or effectively suppressing rebellion. In contrast, in motivating mechanisms of grievance, as mentioned above, material distribution and repression work in the opposite direction, creating a profound challenge to the resource-conflict link suggested by the conventional wisdom of the existing literature. Whether or not high rents will be available depends not only on output but also on input as part of resource governance. Careful management of the resource sector may include effective taxation structures (Snyder und Bhavnani 2005) and private rather than state ownership (Luong und Weinthal 2006). Moreover, rent income often disappears into the pockets of the elite or multinational companies. Important resource-specific conditions also include the external structure of demand. Powerful importing countries may be ready to intervene militarily, either directly or by supporting warring factions. All of these conditions have been declared to constitute a key research agenda (Collier und Hoeffler 2005, 627). 5 Nigeria and Saudi Arabia, for instance, were almost equally dependent on oil exports in 2002; these exports accounted for 38.9% and 38.5% of their GDPs, respectively. However, had governments decided to pay out the proceeds from oil exports to their citizens, Nigerians would have been given a mere US$ 140, whereas Saudi Arabian citizens would have earned US$

5 The likelihood of conflict may also depend on non-resource-specific characteristics such as the general level of development, relations between identity groups, geographical and demographical factors and the general quality of state institutions as well as elite behavior. All of these conditions may affect the likelihood of violence independent of resources. Hegre and Sambanis (2006) have empirically identified non-resource-related conditions such as population size, income level and growth and rough terrain as robust correlates of major civil war. However, more importantly, resource- and non-resource specific conditions are likely to interact (Ross 2008; Basedau 2005). In a resource-dependent economy, wealth is mainly derived from resource income. The relationships between identity groups will be affected when resources are discovered in a region that differs from the rest of the country or when the region in question does not get a fair share of the revenues but instead has to suffer from ecological or socio-economic stress connected to resource production. 2.2 Results from Quantitative Studies A multitude of quantitative studies has tried to demonstrate that natural resources increase the risk of civil war onset. The results vary. Ross (Ross 2004a) has analyzed 14 quantitative studies of the resource-conflict link and finds that primary commodities as a whole cannot be robustly linked to civil war onset or duration. According to his conclusions, the type of resource definitely matters. Whereas no study under review by Ross can establish a relationship between agricultural commodities and violence, lootable resources such as narcotics, timber and (alluvial) diamonds seem to influence the duration of a conflict, although they do not make the onset of civil war more likely (Lujala, Gleditsch, und Gilmore 2005). 6 Studies differ in their conclusions regarding whether diamonds cause or sustain civil war (Le Billon 2008, 352). Only oil-exporting countries seem to be particularly prone to civil war onset (Ross 2006). Further studies question the notoriety of oil in this respect: Smith (2004) finds a positive effect of oil dependence on regime stability and peace in developing countries. According to Hegre and Sambanis (2006, 531), only oil exports (and not production or other resources) are marginally robustly linked to minor armed conflict. Others (Fearon und Laitin 2003; Fearon 2005) have concentrated their criticism on the opportunity or feasibility mechanism and propose that the oil-violence nexus works through a weak state mechanism (see Ross 2006, ) or can be attributed to the effects of a sparse networks mechanism (Humphreys 2005). 6 They find that secondary diamonds are positively linked to the onset of ethnic civil wars only, whereas primary diamonds lower the risk of civil war onset and duration. 5

6 Recent studies have found evidence that some of the rentier mechanisms are present within oil states in particular. Fjelde (2009) finds that the interaction of high levels of corruption and appropriable resources (oil wealth) reduces the conflict proneness of a country by offsetting the destabilizing effect of resource abundance. According to Basedau and Lay (2009), oil dependence increases the risk of civil war onset, creating a U-shaped relationship, whereas high levels of abundance (as indicated by an inverted U-shape relationship) are apparently used to engage in large-scale distribution and the establishment of a huge and effective security apparatus. 7 Brunnschweiler and Bulte (2009) have also tested both resource dependence and abundance and find that resource abundance reduces the likelihood of civil war onset, whereas dependence seems to be a result rather than a cause of civil war. 2.3 Results from Comparative Studies (Small & Medium N) Methodologically, research on the onset of civil war is dominated by quantitative studies employing mostly logistic regressions. Although there are numerous case studies that provide country-specific evidence on the resource-conflict link, genuinely comparative perspectives are rare. In particular, medium-n studies are very much missing from this field. There are some edited volumes and monographs that compare several pertinent cases (Basedau und Mehler 2005; Collier und Sambanis 2005; Oliveira 2007). A comparison of three diamond producers (Sierra Leone, Ghana and Guinea) by Snyder and Bhavnani (2005) shows that modes of extraction, the tax base from resource production and spending patterns as already suggested in the theoretical section above help to influence the complex relationship between natural resources and civil war. One of the few systematic small-to-medium-n studies in this debate is Ross (2004b). Ross explicitly and exclusively tests a number of causal mechanisms potentially leading to civil war onset using 13 cases in which resources and conflict are most likely interrelated. He finds evidence of neither the opportunity ( looting ) nor motive ( grievances ) mechanisms for creating conflict, but he does find evidence of separatism in two cases (Sudan, Indonesia) and of two other mechanisms in three cases (Congo Republic: future booty ; DRC: foreign 7 The U-shaped relationship suggests that very low and very high levels of dependence are linked to a greater likelihood of civil war, whereas medium levels of dependence make civil war onset less likely. They find an inverse U-shaped relationship between abundance and the likelihood of civil war: states with medium levels of abundance are the most war-prone, whereas very low and high levels of abundance are related to a lesser likelihood of civil war onset. 6

7 intervention; Sierra Leone: foreign intervention and future booty). Empirical support for the mechanisms at work increases with conflict duration, particularly for looting (in all but two cases), although some of the mechanisms have shortened the wars. However, the sample cannot explain differences in violence given the rather constant dependent variable, as frankly admitted by the author himself (Ross 2004b, 49). A study by Basedau and Lay (2009) uses both logistic regression and a medium-n analysis that concentrates on 27 highly dependent net oil exporters to isolate the possible effects of abundance and material rewards that may explain the absence or presence of civil war onset. According to this analysis, abundant income from resources is apparently used to engage in large-scale distribution and the establishment of a huge and effective security apparatus. In addition, oil-abundant states often enjoy protection from abroad, suggesting that international influence may not only spur conflict but also contribute to internal peace. A paper by Wegenast and Basedau (2009) tests a medium-n sample of 17 highly ethnically fragmented, low-income, and resource-rich countries (a strategy used to keep important contextual conditions constant) and attempts to look more closely at individual cases. The findings lend support to the assumption that the combination of ethnic exclusion and resource availability (here, oil and diamonds) renders civil war onset very likely. Basedau and Wegenast s (2009) medium-n study of Africa s 15 main oil and diamond producers confirms these results. According to these findings, the existence of high dependence and minorities at risk in producing regions strongly increase the likelihood of civil war. Lower income from resources per capita and substantial production of lootable resources in peripheral regions are apparently necessary conditions for civil war. 2.4 Preliminary Conclusion Three observations from previous research on the nexus of material resources and civil war onset seem important for our paper: First, oil (and oil exports) is the only resource that is more or less robustly connected to civil war. Secondly, relatively little effort has been invested in explaining the variation within the group of oil-exporting countries. The relatively few studies on this subject that exist suggest that certain resource-specific characteristics (e.g., abundance and dependence) and the relation of resources to the ethnic diversity of a country, for instance, might be potential factors. Finally, research strategies other than quantitative approaches or single case studies are extremely rare. In particular, medium-n analyses are, 7

8 with a few exceptions, virtually absent from the field of conflict studies. To the best of our knowledge, no study has ever tried to investigate the oil-conflict link via formal configurational methods such as csqca using a medium-n sample. 3. A Short Introduction to Crisp Set QCA (csqca) Qualitative Comparative Analysis (QCA) or Configurational Comparative Analysis, as it has recently been (re)named (Rihoux und Ragin 2009, bd. 51), is a relatively new formal data analysis technique that is concerned with the systematic matching and contrasting of cases to establish common causal relationships by eliminating all other possibilities (Berg- Schlosser u. a. 2009, 2). Introduced by the sociologist Charles Ragin (Ragin 1987) during the late 1980s and developed further during the last two decades (Ragin 1994; Ragin 2000), QCA has become an interesting complement to prevailing statistical approaches within macrocomparative research (Grofman und Schneider 2009). However, as already mentioned, to the best of our knowledge, QCA has rarely been used as a data analysis technique within the field of peace and conflict studies. 8 Although it can be argued that QCA attempts to combine the advantages of qualitative (caseoriented) and quantitative (variable-oriented) techniques, it is better suited to comparing an intermediate number of cases (medium-n) than are most quantitative, statistically oriented methods. From a qualitative perspective, QCA is useful because it still allows a closer look a different kinds of cases; it does not level down case characteristics to an average degree within a given sample as statistical techniques usually do. From a quantitative perspective, QCA permits to analyze the influence of more than one or two variables on a single outcome or event, without getting lost into case specifics. As a data analysis technique, QCA systematically combines the method of agreement and the method of difference 9 the two famous principles of J.S. Mill s logic of comparison in order to identify necessary 10 and sufficient 11 conditions (independent variables) or 8 The only application we know of is (Clément 2004), and she deals more with state failure than with civil war. 9 The method of agreement eliminates all similarities but one. The method of difference establishes the absence of a common cause if all other circumstances are identical (Berg-Schlosser u. a. 2009). 10 A condition is defined as necessary if it must be present for a certain outcome to occur (Ragin 1987, 99). However, necessary conditions might be present even the outcome does not occur. 11 A condition is defined as sufficient if by itself it can produce a certain outcome (Ragin 1987, 99). Sufficient conditions are not present if the outcome does not occur. 8

9 combinations of conditions in relation to a certain outcome (the dependent variable). 12 QCA s two basic questions are therefore as follows: which conditions (or which combination of conditions) are necessary for a certain outcome to occur, and which conditions (or which combination of conditions) are sufficient for a certain outcome to occur? As a result, QCA has the ability to unravel causally complex structures like equifinality, multifinality, and asymmetric causality (Grofman und Schneider 2009, 662). Whereas standard statistical techniques give prominence to the net effect of single variables, QCA is used to detect different and sometimes overlapping conjunctions of configurations or conditions that may all led to the same outcome (Ragin 2008, ). Boolean algebra 13 is used to analyze truth tables in which cases are pictured according to the binary values of their conditions and respective outcomes. 14 An uppercase letter represents the 1 value for a given binary condition or an outcome, whereas the lowercase letter represents the 0 value for that binary condition or outcome. A hyphen - represents the don t care value for a given binary condition or outcome, meaning that it can be either present or absent. 15 In crisp set QCA, there are two main Boolean operators used. The logical AND is symbolized by the * multiplication sign and the logical OR is represented by the + addition symbol. To establish a connection between the conditions and the outcome, arrows are used. An arrow to the right signifies a sufficient relationship between conditions and outcome, whereas an arrow to the left symbolizes a necessary relationship. Moreover (and this is really the heart of the matter of using QCA as a data analysis technique), Boolean minimization is used to reduce empirical complexity. In the words of Charles Ragin, If two Boolean expressions differ in only on causal condition yet produce the same outcome, then the causal condition that distinguishes the two expressions can be considered irrelevant and can be removed to create a simpler, combined expression (Ragin 1987, 93). 12 Please note that QCA applicants usually refer to independent variables as conditions and to dependent variables as outcomes. We will exclusively use this terminology from now on. 13 Boolean algebra was developed in the mid-nineteenth century by the mathematician and logician George Boole and is based on logical rather than numeric characters. For a detailed introduction to this technique as used in QCA, see Ragin 1987, ; Rihoux und De Meure Because we use only crisp set QCA (csqca) in this paper, other versions (like, for instance, multi-value QCA (mvqca) and fuzzy-set QCA (fsqca)) are not be examined here. For mvqca, see Cronqvist 2005; Cronqvist und Berg-Schlosser 2009 and for fsqca, see especially Ragin It is important to note that the hyphen is not an intermediate value between 1 and 0. 9

10 Consider the example of two states that have high degrees of primary commodities per GDP (sxp) and experienced negative growth during a given time period (gy1) but significantly differ in their population size (pop) one with more than 20 million inhabitants and one with just only 1 million. Also assume that in both countries, a civil war started to take place (onset). Table 1 shows the corresponding truth table. Table 1: A simple binominal Truth Table used for csqca Country Sxp Gy1 Pop Onset A B Expressed in Boolean language, this truth table can be written as: SP*gy1*POP + SP*gy1*pop ONSET 16 Following the rule described above, Boolean minimization allows the replacement of these two terms with a single, simpler expression because the population size turns out to be irrelevant to the onset of civil war. SP*gy1 ONSET As a result of Boolean minimization, more parsimonious expressions are created that show combinations or configurations of conditions that together form a sufficient condition for the occurrence of a certain outcome (Rihoux und De Meure 2009, 34-39). As part of these configurations, the initial single condition operates as a so-called INUS factor (an insufficient but necessary part of a condition that is itself unnecessary but sufficient for the outcome) (George und Bennett 2005; Mahoney 2008). These are causally relevant factors that are almost always overlooked when standard statistical techniques are applied (Grofman und Schneider 2009, fn. 4). 16 Read this expression as indicating that the presence of many primary commodity exports relative to GDP, combined with negative GDP growth and a large population, OR the presence of many primary commodity exports relative to GDP, combined with negative GDP growth and a small population, are sufficient for civil war onset. 10

11 There exists a shared consensus among experts on csqca that a good configurational analysis will present at least three technical components (Wagemann und Schneider 2007; Grofman und Schneider 2009): the truth table, the solution formulas and measures of fit like coverage and consistency. The truth table presents the data to be analyzed, showing the conditions (the independent variables) as columns, the cases as rows and the value of the outcome (the dependent variable). However, the cases are not always confined to a single row but are rather grouped together if they are associated with a single combination of conditions with the same outcome. Using Boolean language, the solution formula shows the causally relevant conditions linked to the outcome, as already explained above. Coverage and consistency are basic measures of fit (Ragin 2006). The consistency of a sufficient condition measures the proportion of instances of this condition coded as either 1 or 0 in csqca that overlap with the given value of the outcome also coded as either 1 or 0 in relation to the total number of cases with the same value for this condition. The higher the consistency level of a condition, the closer it is to being a consistently sufficient condition of the outcome. Furthermore, coverage will exist only when conditions are sufficiently consistent to be regarded as sufficiently explaining the outcome (Grofman und Schneider 2009, 665). 17 The coverage of a sufficient condition indicates the proportion of the outcomes that are explained by this condition. It is then the number of cases in which Y is present simultaneous to (as the consistently sufficient condition or configuration of conditions) relative to all cases with Y. 4. Empirical Analysis 4.1 Sample Selection Because we are mainly interested in explaining civil war onset and peace among petroleumproducing countries, our base sample consists of all countries that were net oil exporters for at least one year between 1970 and 2008, the period for which reliable data is available (the base sample). 18 To exclude pure oil traders like Singapore, for instance, we also excluded all those 17 Please note that in csqca, as we have introduced it in this paper, the level of consistency of a single condition or a configuration of conditions must always be or 100 per cent. 18 The data are taken from UN comtrade (SITC Rev. 1,33) online: (access ). 11

12 countries with no proven reserves of mineral fuels using the PRIO Petrodata (subsample1). 19 Following Ross (2006), we further reduced our sample size, only keeping those countries with non-trivial amounts of oil production: that is, countries with more than $100 per capita in rents per year from exporting oil and gas in at least one year between 1970 and This two-step sample selection procedure leaves us with 44 petroleum-producing countries (subsample2) Outcome (Dependent Variable): Civil War The dependent variable or the outcome is the onset of civil war, which is operationalized using UCDP/PRIO definitions and data (Gleditsch, Wallensteen, und Eriksson 2002). According to the definition by UCDP/PRIO, an armed conflict is a contested incompatibility that concerns government and/or territory where the use of armed force between two parties, of which at least one is the government of a state, results in at least 25 battle-related deaths. We define a civil war as any armed conflict that has caused at least 1,000 battle-related deaths in a given year. We positively coded civil war onset for a country when at least one such conflict began or if an ongoing conflict or a new episode of a conflict caused at least 1,000 battle-related deaths during at least one year. All other constellations, including those featuring armed conflicts with fewer than 1,000 victims, were coded as not experiencing civil war onset (cf. Appendix 2 - Codebook at the end of this paper for more details). On this basis, there are 30 cases in the sample that have not experienced civil war onset and 14 cases in which at least one civil war has broken out between 1970 and Conditions (Independent Variables) Because we aim to provide the most parsimonious explanations possible, we restrict our independent variables to four pertinent conditions 22, two of which are novel in civil war research. (see for more coding details and data sources, Appendix 2 at the end of this paper). 19 Data are from PRIO/CSCW: Petroleum Dataset v. 1.2, online: (access ). 20 Data are from the World Bank Adjusted Net Saving Data Center: Oil and Gas Rents, online: (access ). 21 A list of all three samples can be found in Appendix 1 at the end of this paper. 22 Limiting our analysis to four conditions inevitably raises the question of why we use these and not others. As already mentioned, QCA is best used with limited number of factors explaining a certain outcome. Having selected these four conditions does not mean that we are completely ruling out others that might be important. In particular, non-resource-specific conditions such as state strength might be included in future research. However, some potentially influential conditions were excluded based on theoretical considerations. For instance, because 12

13 Dependence: As discussed in section 2.1, greater dependence on oil might make civil war more likely. High levels of dependence indicate that oil is the main source of wealth in a country, which may have a number of harmful consequences that may indirectly increase the risk of civil war or, using the classification of section 2.1, may serve as a proxy for the indirect mechanism(s): such dependence makes countries vulnerable to price shocks in the world oil market. Dependence on an export commodity such as oil fosters a rent-seeking mentality and, resultantly, weaker institutions and governance. Finally, in an economy dependent on oil, it is also more likely that economic conflict will centre on this very commodity. Dependence on oil is certainly the most conventional measure of resources in the resource-conflict literature and has been employed in many, if not most, quantitative studies (cf. for an overview Ross 2004a). We measure dependence on oil (dependence5) using the ratio of available fuel exports to GDP. To minimize endogeneity problems, we use the mean between 1970 and the year before the possible civil war onset according to the above definition. If no civil war begins, the mean for the whole period of investigation is calculated. Because csqca requires binary-coded conditions, we had to select a cut-off point. Looking at the actual distribution of the values in the sample, we see that 0.15 (indicating that 15% of the GDP consists of fuel exports) emerges as a sort of natural threshold range because no country shows this or similar values (for the distribution of dependence5, see Appendix 3 at the end of this paper). We used this value to create a new binary variable dependence5_015. All values below this point have been coded as 0 and all values equal to and above as 1. Abundance: As also previously discussed, dependence and abundance must not be confused (see section 2.1). According to our understanding, oil abundance indicates what oil rents are available per capita. The per capita measure is particularly useful because it allows us to proxy the capacity of governments to employ peace-buying rentier mechanisms, thus testing possible motivation mechanisms and providing a proxy for state capacity. Large-scale distributional policies and an effective security apparatus are extremely expensive, and income must be considered in relation to population size. We measure abundance in oil and gas (rentpc3) as the average of oil and gas rents per capita between 1970 and a possible civil war onset (or, if no war occurs, until 2008). Data on estimated income from these rents come oil is a less lootable resource than diamonds, we do not include an indicator of lootability or any proxy for the opportunity mechanism in a narrow sense. 13

14 from the World Bank Data on adjusted net savings. 23 For abundance, we were able to identify a pertinent threshold. Variable pre-testing identified a cut-off-point of $440 per capita from rents above which no civil war began (see the Appendix 4 at the end of this paper for the distribution of rentpc3 in relation to ucdpintens3). Similar as with dependence, we used this value to create a new binary variable rentpc3_440. All values below this point have been coded as 0 and all values equal to and above as 1. Overlap: Conflicts like those in Nigeria (Biafra, Niger Delta), Indonesia (Aceh) or (southern) Sudan illustrate that the interaction of ethnic and/or regional heterogeneity and oil reserves might be a particularly dangerous combination, though this condition has rarely been tested. The combination of resources and a shared group identity facilitates mobilization and provides a significant motive for uprising (motive mechanisms). This risk might increase if groups are already politically or otherwise aggrieved. Furthermore, resources may contribute to ethnic insurgency by supplying the necessary financial means for armed rebellions (opportunity mechanism). We therefore created a third variable that captures whether oil reserves are located in areas where ethnic groups settle that are excluded from the central political framework in a given country. To measure this overlap, we used PRIO data on the location of oil reserves (Lujala, Gleditsch, und Gilmore 2005) and matched them to georeferenced data from the Ethnic Power Relations Data (see Wucherpfennig u. a. 2010). We positively coded for overlap of ethnic exclusion and resource endowment (overlap2) when overlap between the settlement of an excluded group and oil location could be identified within the period of investigation or, in the case of civil war onset, before the outbreak of civil war. If the area in which the excluded group was settled covered the entire state territory, we did not code this as an overlap. All other constellations were also rated as 0. Because this procedure required subjective visual assessment, we performed inter-subjective coding via a group of three different independently functioning coders. Outside protection: External variables have been rarely used in studies of the resourceconflict link. Because oil is probably the most strategic resource in the world today, it attracts interest that may also include a military dimension. External powers may be ready to intervene militarily or to support warring factions within a given oil country to secure a 23 Data can be found at (access, ). Please note that rents (rents3) include both oil and gas rents, not just oil. 14

15 supply of black gold. However, this is not to say that oil-exporting countries only face threats from abroad. On the contrary, many oil-producing countries have established close relations with internationally powerful countries, which are most often also large oil importers. The most well known example is certainly the special relationship between the US and Saudi Arabia, in which the latter provides oil to the former and receives a security guarantee in return. We thus created a fourth variable to measure outside protection (outsideprotection). We consider substantial outside protection to exist if the net oil exporter in question hosts a permanent military base of a permanent UNSC member country (US, British, Chinese, French, or Russian military bases) which we coded as We defined a military base as a (relatively permanent) facility with regular combat forces (not just an airfield or depots). Military bases were only counted as outsideprotection when there was a military presence before the onset of civil war and not as in Afghanistan or Iraq as a result of an external military invasion. 4.4 Model and Expectations We use the following model (1) to test for the necessary and sufficient conditions of civil war onset and peace using crisp set QCA. (1) ucdpintens3 = f (rentpc3_440, dependence5_015, overlap2, outsideprotection) The four variables dependence, abundance, overlap and outside protection form a specific risk profile. According to our theoretical assumptions, the greatest risk of civil war should be found in countries that are A) highly dependent on oil, B) experience low oil abundance, C) feature a geographical overlap between oil reserves and excluded ethnic groups and D) do not enjoy outside protection. In contrast, we should expect that countries that are a) not dependent on oil, b) enjoy a great abundance of rents, c) do not have a geographical overlap between excluded ethnic groups and oil reserves and d) possess the protection by a powerful outsider will not experience civil war. 24 In the case of the UNSC members themselves, outside protection was only positively coded when there were military bases from other UNSC members. For instance, China and Russia, both without any foreign military bases do not have outside protection according to our understanding. 15

16 5. Empirical Analysis Using Crisp Set Qualitative Comparative Analysis (csqca) 5.1 The Truth Table Table 2 below shows the truth table based on model (1) for a dataset of 44 net oil-exporting countries with at least $100 per capita rent during at least one year between 1970 and 2008 (subsample2). There are 16 possible configurations combining four binominal conditions (2 4 = 16). Strictly speaking, only 4 of these do not represent cases from our dataset. In other words, there are 4 different theoretically possible configurations that are not represented among net oil exporters. Two of the 16 possible configurations which together represent 8 cases have led to civil war onset. As row number 1 shows, Algeria, Angola, Azerbaijan, Congo, Iran, Nigeria and Russia are part of the same configuration: they experienced the onset of at least one civil war between 1970 and Additionally, Yemen also underwent a civil war but is part of another configuration (row number 3). As highlighted in Table 2, both configurations represent high-risk profiles as previously expected in section 4.4. Furthermore, there are 9 different configurations including 26 cases in which no civil war has broken out. Additionally, there is one configuration (row number 2) that includes 10 contradictory cases (C). 25 This configuration includes those net oil exporters with low abundance (per capita rents below US$440), a low degree of dependence (below 0.15), no outside protection, and overlap in terms of the location of oil reserves and a settlement populated by an excluded political group. Whereas in Colombia, Indonesia, Iraq, Peru, Sudan, and Syria, a civil war broke out between 1970 and 2008, there was no civil war in Argentina, Bolivia, Malaysia, or Mexico during that time. Because of its contradicting outcomes, this configuration had to be excluded from the csqca analysis. We will eventually discuss the importance and relevance of this configuration in the concluding section of this paper. 25 Please note that this configuration represents less than 25% of cases. In turn, this means that our model explains more than 75% of the cases in the sample. This rate is far above the degree of variance that is usually explained via common quantitative regressions. 16

17 Table 2 Truth Table: ucdpintens3 = f (rentpc2_440, dependence5_015, overlap2, outsideprotection) Row Cases No of cases rentpc2_440 dependence5_015 overlap2 outsideprotection ucdpintens3 consistency of ucdpintens3=1 1 Algeria, Angola, Azerbaijan, Congo, Iran, Nigeria, Russia Argentina (0), Bolivia (0), Colombia (1), Indonesia (1), Iraq (1), Malaysia (0), Mexico (0), Peru (1), Sudan (1), Syria (1) C Yemen Kazakhstan Oman, Libya, Trinidad & Tobago, Turkmenistan, Venezuela Cameroon, Egypt, Papua New Guinea, Tunisia Gabon, Kuwait, Saudi Arabia Ecuador Australia, Côte d'ivoire, Denmark, Netherlands, UK, Vietnam Bahrain, Brunei, Qatar, United Arab Emirates Canada Norway Please note: We have highlighted high risk conditions in this table. For instance row no.1 has 4 highlighted cells this is a configuration with the highest risk of civil war onset. In contrast to that in row no.12 none of the cells is highlighted this is a configuration with the lowest risk of civil war onset. C indicates that a contradictory outcome exists for this configuration. The numbers in brackets after the countries in this row indicate the corresponding outcome. 17

18 5.2 Identifying Necessary Conditions of Civil War Onset and Peace 26 A condition is defined as necessary if it must be present for a certain outcome to occur (Ragin 1987, 99). Thus, tests for necessity attempt to verify whether a condition is always present if a certain outcome occurs. However, necessary conditions might also be present if the outcome does not occur. 27 Table 3 Test results for necessary conditions for UCDPINTENS3 (civil war onset) 28 Conditions tested: Consistency 29 Coverage 30 RENTPC3_ rentpc3_ DEPENDENCE5_ dependence5_ OVERLAP overlap OUTSIDEPROTECTION outsideprotection There are two necessary conditions for civil war onset, as shown by the consistency levels of in Table 3. All countries that experienced civil war onset feature rents per capita below $440 (rentpc3_440) and do not have a military base affiliated with a permanent UNSC member on their territory. Additionally, spatial overlap between the settlement of politically excluded groups and oil reserves within a given country only marginally fails the test for necessity. It is interesting to note that this result only occurs because of the inclusion of Yemen. If overlap2 for Yemen had been coded as 1, the consistency level of OVERLAP2 would have been instead of Please note that unlike in the case of the tests for sufficiency, the cases that are part of the configuration with contradicting outcomes in the truth table (row 1 in Table 2) will be included in the tests for necessity. 27 For the tests measuring necessity, we have used the software fs/qca 2.5, which can be downloaded at (Access, ). 28 As previously explained, uppercase letters represent the 1 value for a given binary condition or outcome, whereas lowercase letters represent the 0 value for that binary condition or outcome. This means that for instance RENTPC3_440 represents all cases coded with 1 that have an average rent per capita above $440, whereas rentpc3_440 represents all cases coded with 0 that have an average rent per capita below $ The consistency value of a necessary condition indicates the degree to which this condition overlaps with a particular outcome relative to all cases with the same outcome. If a given condition is present in all cases with the same outcome, the consistency value will be , indicating that this is a necessary condition. 30 Coverage of a necessary condition indicates the overlap between a condition and an outcome relative to all cases with the same condition. Coverage therefore measures the proportion of instances in which the condition arises that is necessarily explained by the outcome. 18

19 Of the two necessary conditions of civil war, the absence of outside protection (outsideprotection) has slightly higher coverage ( ) than a low level of per capita rents (rentspc3_440). Only % ( ) of the cases that exhibit the latter condition have experienced civil war onset. If both of these conditions are taken together, we see that if at least one is present in a country, there is roughly a 50% chance that a civil war will break out. However, if OVERLAP2 had not marginally failed the test for necessity, countries with this characteristic would be the most likely to experience civil war of all those within our sample of 44 net oil-exporting countries. Table 4 Test results for necessary conditions for ucdpintens3 (absence of civil war) Conditions tested: Consistency Coverage RENTPC3_ rentpc3_ DEPENDENCE5_ dependence5_ OVERLAP overlap OUTSIDEPROTECTION outsideprotection Unlike in necessity tests for civil war onset, there is no necessary condition for peace (or the absence of civil war), as indicated by the presence of consistency values below in Table 4 for the 8 different values of the 4 binominal conditions tested in our sample. Taken together on a more technical level the solution formula for the necessary conditions for civil war onset reads as follows: (2) rentpc3_440 + outsideprotection UCDPINTENS Read this expression as indicating that low levels of per capita rents OR a lack of outside protection necessarily lead to civil war onset. 19

20 This formula means that an average level of per capita rents below $440 or the absence of a military base associated with a permanent UNSC member are independent of each other necessary conditions for civil war onset in net oil-exporting countries. 5.3 Identifying Sufficient and INUS Conditions of Civil War onset and Peace A condition is defined as sufficient if it can independently produce a certain outcome (Ragin 1987, 99). Thus, tests for sufficiency attempt to verify whether a condition always leads to the same outcome (although sufficient conditions, unlike necessary conditions, are not (necessarily) present if the outcome occurs). In addition, INUS conditions are insufficient but necessary parts of a condition that is itself unnecessary to but sufficient for the outcome to occur. 32 As previously pointed out (cf. section 3 of this paper), Boolean minimization is used in csqca to test for sufficient conditions to produce a certain outcome and to seek the most parsimonious solution formulas including INUS conditions. This Boolean operation, however, presupposes that each possible logical configuration is covered by at least one real-world case. Logical remainders, as logically possible configurations that are not represented through empirical cases in the dataset, constrain this quest for parsimonious solutions. 33 In the next two sections, we therefore present two results of Boolean minimization: one without using logical remainders and the second including logical remainders. We call the latter the parsimonious solution and the former the complex solution Civil war onset (UCDPINTENS3) A csqca-test for sufficiency using the outcome of civil war onset (UCDPINETNS3) excluding logical remainders yields the following complex solution: (4) rentpc3_440*dependence5_015 *outsideprotection UCDPINTENS3 solution coverage: unique coverage: solution consistency: For the following tests of sufficiency, the software TOSMANA has been used. See (Access, ) for more information. 33 See Table 2 for more details. There are 4 logical remainders in our truth table (rows 13, 14, 15, and 16). 20

21 This solution means that low rents per capita, high dependence and a lack of outside protection together sufficiently explain the onset of civil war in net oil-exporting countries in our sample. Taken separately, each factor forms an INUS condition of civil war onset. This solution is valid without any simplifying assumptions regarding logical remainders and explains (as indicated by the solution and unique coverage values) more than 57% (7 out of 12) of the cases of civil war onset in the dataset. 34 To find a parsimonious solution, logical remainders are included in the test for the sufficiency of civil war onset. Formula (5) reports the most parsimonious explanation that we could develop using our dataset: (5) rentpc3_440*dependence5_015 UCDPINTENS3 solution coverage: unique coverage: solution consistency: This solution reveals that low per capita rents and high dependence are together a sufficient explanation for civil war onset. Taken separately, the two conditions form INUS conditions. However, this parsimonious solution is only valid if one assumes that all 4 of the logical remainders in Table 2 are associated with the onset of civil war in reality No onset of civil war (ucdpintens3) After having tested for sufficient conditions of civil war onset, formula (6) summarizes the results of the csqca-test for sufficiency using no civil war onset as an outcome. In a first step similar to the tests in the previous section the complex solution is presented; later, we also show a parsimonious solution formula. 34 The Appendix 5 at the end of this paper provides a more detailed overview on this. 21