Inevitable Evil? A Qualitative Comparative Analysis of Genocide Occurrence and Non- Occurrence. Paul Steinheuer University of Kent, Canterbury

Transcription

1 Inevitable Evil? A Qualitative Comparative Analysis of Genocide Occurrence and Non- Occurrence Paul Steinheuer University of Kent, Canterbury Timothy Williams Free University, Berlin Centre for Conflict Studies, Philipps-University, Marburg Paper presented at the International Studies Association Annual Convention, San Francisco, 3-6 April, 2013

2 Abstract: The relevant genocide literature identifies a host of causes for why and when genocide occurs, with war, political upheavals such as revolution, autocracy, less interdependent economies and exclusionary ideologies among the suggested factors. However, much of this evidence relies strongly on individual cases of genocide, neglecting the important role of comparison between cases and the relevance of counterfactuals. Harff s (2003) statistical study began to bridge this gap; however, as a complex phenomenon there are multiple reasons why different cases of genocide occur, which are not captured by this statistical approach. By approaching this topic with Qualitative Comparative Analysis (QCA), this paper aims to apply a different and more suitable methodology to this question than previous studies. This method allows for the occurrence or non-occurrence of genocide to be traced back not to the same variables in all cases, but to different combinations of multiple factors, each coming to a fore in different cases. By using QCA and including counterfactuals in the analysis, this paper is able to identify both necessary and sufficient conditions for genocide occurrence. The model identifies several combinations of independent variables which in interaction with each other have the same effect across several cases, either repeatedly causing genocide or preventing it.

3 Genocide is not an inevitable consequence of certain social conditions within a society. There may be extreme pluralism in a society, with highly antagonistic, polarizing ideologies, division expressed in religion, segregation, employment, social networks, and political party affiliation, a long history of reciprocal violence, and periods of highly escalated conflict. Yet the struggle may stop short of genocide. - Leo Kuper, 1981 (page 56f.) 1. Introduction This quote by Leo Kuper is an apt summary of most scholarship in the study of genocide, highlighting certain factors which facilitate genocide occurrence but shying back from being able to actually label any of these conditions necessary or sufficient. This paper aims to go beyond the previous literature and identify constellations of determinants which together can cause genocide. Given the heinous nature of the crime and the huge impact genocidal killing has had in the 20 th century it is of the utmost importance that social scientists further the endeavour to identify the central determinants of this phenomenon in order to understand why and when genocides occur. Thus the research question underlying this paper asks What combinations of factors play together to determine genocide occurrence? In 2003 Barbara Harff published her seminal paper No Lessons Learned from the Holocaust? Assessing Risks of Genocide and Political Mass Murder since 1955, a statistical analysis of determinants of genocide occurrence which received wide acclaim. While Harff s study identified several key factors which raised the risk of genocide and politicide, genocide is an inherently complex phenomenon and no one or more causes can be deemed responsible for its outbreak in all cases. Thus this paper postulates that a better methodology for studying this subject matter would be a qualitative comparative analysis (QCA) which uses Boolean algebra to identify multiple pathways to the same outcome with different combinations of variables acting in different cases. With this method, this paper will provide different profiles of factor combinations which provoke genocide outbreak, identifying necessary and sufficient conditions. In order to ensure the comparability of this QCA study with Harff s previous work, we attempt to use as many of Harff s variables as possible. This paper will first clarify some definitorial issues around the concept genocide and politicide, before presenting a theoretical framework of different determinants emphasised in previous research and explaining the merits of the chosen QCA approach. Subsequently, our case choice will be explained in detail, as well as the variable operationalisations, particularly detailing which variable of Harff s this study kept, and which were replaced. Finally, the results of the QCA will be presented in a truth table

4 and expanded in an empirical analysis, before bringing together these results and identifying how this study has developed the work of Harff and what remains to be researched. 2. What is Genocide? A Contentious Definition of an Elusive Concept While genocides have been occurring as long as mankind has been living in societies, the term genocide was not coined until after World War II. Polish jurist and émigré to the United States Raphael Lemkin combined the Greek word genos (nation, people, race) with the Latin caedere (to kill), in an attempt to codify in international law the horrors of the Holocaust, in order that they not be repeated. In 1948, the United Nations General Assembly (1948) after much discussion and dilution passed the Genocide Convention, under which genocide is constituted by any of the following acts committed with intent to destroy, in whole or in part, a national, ethnical, racial or religious group, as such: (a) Killing members of the group; (b) Causing serious bodily or mental harm to the members of the group; (c) Deliberately inflicting on the group conditions of life calculated to bring about its physical destruction in whole or in part; (d) Imposing measures intended to prevent births within the group; (e) Forcibly transferring children of the group to another group. This definition, however, is widely criticised by academics, as it proves too exclusive in its target identification, ignoring victimisation of political, economic and social groups, such as around 300,000 mentally ill and thousands of homosexuals killed during the Holocaust or the 15-20m Soviet citizens killed as class enemies (Chalk 1989). Furthermore, the UN definition is too inclusive regarding constitutive means (mental harm, birth prevention, transfer of children), and many posit that a degree of physical harm should be included to warrant the term genocide. Also, the Convention neglects the possibility of non-state actors engaging in such actions; finally, while the stipulation of intentionality is important to the definition, it provides great difficulty empirically. While this final point has been debated hotly in the research community, it has also been pointed out that it is not plausible that a group of some considerable size is victimized by man-made means without anyone meaning to do it! (Jonassohn 1992, 21). Nonetheless, when trying to identify cases of genocide and distinguish them from cases of non-genocide, it can be difficult to pinpoint concrete evidence of intent. Due to these obvious shortcomings, academics widely opt for using alternative definitions, although there are nearly as many definitions as there are authors. One helpful definition is provided by Chalk and Jonassohn (1990, 26): a form of one-sided mass killing in which a state or other authority intends to destroy a group, as that group and membership in it are defined by the perpetrator. The advantages of this phrasing are the subjective definition of the victim, one-sidedness of violence, the

5 stipulation of intentionality, and option for non-state and state actors, most elements not found in their entirety in other definitions. However, some authors would not go as far as this and allow victims to be any group a perpetrator chooses. Thornberry (1991, 68) posits that political groups are formed optionally, not as inevitable groups and thus should be excluded from the genocide category. So as to include these cases in her study anyway, Barbara Harff (2003, 58) includes the crime of politicides, defining genocides and politicides as the promotion, execution, and/or implied consent of sustained policies by governing elites or their agents or, in the case of civil war, either of the contending authorities that are intended to destroy, in whole or part, a communal, political, or politicized ethnic group. To ensure the comparability of this study with Harff s original study, we will remain with this definition; however, we will subsume both categories of genocide and politicide under a more inclusive term of genocide, defined to include both the restrictive genocide definition and the extended social, economic and political categories suggested in politicide. Thus, the term genocide in this paper refers to the same subject matter as Harff s genocide and politicide. 3. Why are Genocides and Politicides Expected to Occur? Central Determinants and Hypotheses In this section we discuss the main determinants of genocide, as identified in previous research, particularly the study by Harff (2003), and subsequently deduce preliminary research hypotheses. As will be discussed in section 4, it would be unwise to include too many variables into the QCA; thus, this section will also explain which of the theoretically important variables will be included in the analysis and which will be factored out. Political Upheaval The most commonly cited determinant of genocide occurrence is political upheaval (see Chalk/Jonassohn 1990; Krain 1997; Melson 1992; Rummel 1994; Weitz 2003) and also finds a prominent position in Harff s study. Harff (2003, 62) defines political upheaval as an abrupt change in the political community caused by the formation of a state or regime through violent conflict, redrawing of state boundaries, or defeat in international war. In this context, Melson (1992) and Krain (1997) emphasise the importance of revolutions, but it is also a category in which decolonisation would equally be well placed. The logic behind both these is that political upheavals provide a political opportunity structure which is conducive to starting genocide and in the throes of political upheaval, reigning elites may feel particularly threatened by certain groups, they believe could try to exploit the upheaval to topple them. With the constraints of a previous system removed, the legitimacy of the political community and its identity (including which groups constitute it) can be

6 called into question (Melson 1992, 21). Political upheaval provides the opportunity to re-define the demos, the circle of people included as constituents of the state, in a deinstitutionalised setting (Mann 2005). The empirical founding of this is immense and Harff (2003) finds significant impact of political upheaval on the risk of genocide occurrence, while Krain (1997) suggests that extraconstitutional changes (a similar concept) have a significant impact on this likelihood. These empirical findings suggest that political upheaval should be seen as a necessary and sufficient condition. Hypothesis 1: Political upheaval is a necessary and sufficient condition to all genocide occurrences. War A further condition often associated with genocide occurrence is the presence of war (Melson 1992; Rummel 1994; Weitz 2003), and for Krain (1997) a country s involvement in civil war is the single most significant determinant of genocide occurrence in his statistical study. Nonetheless, war should be seen more as a facilitative condition than a determinant, as, logically, it is never strictly the cause of genocide occurrence: the participation in a war is not a motivation to attempt the destruction of a part of the population, it merely allows such a destruction to happen, e.g. as other policy options are closed off or the state becomes increasingly autonomous from other internal social forces (Melson 1992, 19); furthermore, war provides the opportunity to scapegoat certain unwanted groups and portray them as internal enemies of the state. Due to its purely facilitative nature, however, war will not feature as a variable in the analysis, as in Harff, but is used to aid in the choice of counterfactual cases (see section 5). Pluralistic Societies and Ethnic and Religious Cleavages Defined as states with re-existing internal cleavages and real opposition (Fein 1990, 39), pluralistic states have often come into existence during the process of colonisation and decolonisation when multiethnic states emerged as borders were determined arbitrarily by colonial masters (Kuper 1981). In her paper, Harff (2003) suggests several measures of ethnic and religious cleavages, two of which will be chosen to be tested here: First, continual political and economic discrimination of a minority can lead to this group s political mobilisation, and in turn can lead to repression by the state; in some circumstances, this repression could escalate into genocide. Only if a spiralling of discrimination is seen will a genocide occur, although there can be different reasons for such spiralling to occur and discrimination to intensify. Secondly, if elites are disproportionately from one section of society, underrepresented groups may challenge this, prompting the elites to frame security and ideology in communal terms, e.g. advocating racial denigration or exclusion of the other groups. These ethnic factors, as argued here, could participate in causing genocide where they occur (sufficient), but they

7 do not seem to present mechanisms which must necessarily be present in all cases, as it is conceivable that genocide could occur without strong ethnic factors. In her statistical analysis, Harff found empirical support only for the latter of the two posited connections. In this paper, the primary analysis will include the last measure of ethnic cleavages focussing on disproportionate ethnic representation in a country s elite as this is what Harff found to be most significant. Nonetheless, the other variable will also be included as a comparison. Hypothesis 2: Official state discrimination, or biased elite representation are sufficient, although not necessary for genocide to occur. It is expected that they will have similar impact in the analysis. Autocratic Regime Rummel s (1994) mantra has adapted Lord Acton s iconic phrase to read Power kills; absolute power kills absolutely. In countries with limited and accountable power, democides (as he terms mass murder by governments of their own citizens) are less likely to happen because of cross pressures and the associated political culture. This argument can be focused on a more narrow definition of democracies, in which no one group can become the driving force and there is a democratic culture which involves debate, demonstration, and protests as well as negotiation, compromise, and tolerance (Rummel 1994, 23). Inversely, autocratic regimes have a free hand at dealing with disagreeable groups with no other actors constraining their actions or contesting one-sided genocidal action, suggesting that autocracy is a necessary condition for genocide occurrence. While Harff finds autocracies to be three-and-a-half times more likely to commit genocide than democracies, Krain finds no statistical relation. In an effort to test this discrepancy, this variable will be included in the study. However, one can expect that it only comes to a fore as a sufficient condition in cases in which there is an exclusionary ideology perpetuated by the elites who then have free hand to act on their ideology, or in cases of political upheaval, as the autocratic elites must fear losing power and thus attempt to consolidate their position by drastically changing the playing field, as was evidenced by Hutu radicals in the 1994 Rwanda genocide. This is not necessarily the case for ethnic cleavages, however, as the autocratic regime does not necessarily need to see this diversity as negative; rather political upheaval threatens, and exclusionary ideologies provide a motivation. Hypothesis 3: The presence of an autocratic regime is necessary for genocide occurrence and could be sufficient in combination with political upheaval or an exclusionary ideology. Ideology Chalk and Jonassohn (1990) posit that modern genocide is implemented as an ideology, selecting victims according to who they are, and not by what they have or where they are as it was in

8 premodern genocides. Harff also emphasises the importance of exclusionary ideologies and finds that they significantly impact the probability of genocide occurrence. She defines an exclusionary ideology as a belief system that identifies some overriding purpose or principle that justifies efforts to restrict, persecute, or eliminate certain categories of people (Harff 2003, 63). Freeman (1991) posited that while neither a strong state, nor a pluralistic society were strictly necessary for genocide occurrence a genocidal ideology is. We disagree that such an ideology should be necessary for genocide occurrence, as there are several examples of cases in which it is a rational decision due to internal politics, rather than an ideological impulse that provokes genocide; however, an elite that subscribes to an exclusionary ideology and has the capacity to follow through, needs no other factors for genocide to occur. For instance, Mann (2005) demonstrates that the most important genocidal ideology in modernity has been the ideology of the unified nation-state, from which the victim group is excluded; this becomes possible because in the process of building a new nation, a new identity can be created which redefines the demos (citizens of the state) to include only one ethnos (ethnic sub-group in the country), thus expelling the victim group from the security of citizenship and opening up the possibility of their destruction. Given the evidence in a breadth of scholarly literature this variable will also be included in the analysis. Hypothesis 4: An exclusionary ideology is sufficient for genocide to occur. Economic and Political Autarky A variable not commonly mentioned in the genocide literature, but that nonetheless played an important part in Harff s analysis is the degree of international involvement and interdependence a country has. The more interdependent that a country is with others, the less likely they are to choose genocide as a course of action because the repercussions of international reactions to this move would hit them harder. On the other hand, this means that relatively isolated states leaders will work on the assumption that they can act without having to fear too strong economic repercussions. For instance, countries with relatively independent economies will have less to fear from economic boycotts. This, however, is a facilitative condition that will not cause genocide independently as there is then no intrinsic motivation for committing genocide. In Harff s analysis, political interdependence has no significant impact, while economic interdependence does. In order not to incorporate too many variables, this QCA will thus include only economic interdependence. Hypothesis 5: Economic autarky is facilitative for genocide occurrence in some conditions, although it will have no stand-alone impact.

9 4. QCA an Intermediate Method for Multi-Causal Phenomena The methodology to be used in this paper is Qualitative Comparative Analysis (QCA), popularised in the social sciences by Charles Ragin (1989; 2008). This method categorises cases by combinations of factors leading to the occurrence or non-occurrence of the outcome, and reduces these combinations to the lowest common denominator using Boolean algebra. By executing this reduction one can show which variables are necessary and which are sufficient for causing the outcome. The dependent variable can be caused not by the same variable in all cases, but by different combinations of multiple factors, each coming to a fore in different cases. QCA is thus an intermediate method between quantitative and qualitative methodologies, as it relies on the qualitative judgement of the researcher when categorising the cases, but uses a mathematical technique to categorise and reduce complexity. Its key advantages vis-à-vis statistical analyses are its ability to cope with multiple pathways to an outcome, each consisting of different combinations of variables. While statistical analyses pit different variables against each other to establish the variable with the highest significant impact, QCA follows a conjunctural logic which allows for variables to act differently in different cases depending on their interaction with other factors. Another key strength, the importance of the qualitative input of the research when coding the variables, is also its main weakness. As a set-theoretical method, it can only cope with dichotomous variables, thus losing much information about the countries being studied and not very well allowing for ambiguous cases to be coded appropriately, as they must fit the box. However, there are methodological developments towards fuzzy set QCA (fsqca) which allows for a more continuous categorisation. However, fsqca cannot work with a dichotomous outcome such as genocide occurrence, so that this paper will remain with classic QCA, pointing out contentious categorisations. Thus with this method, this paper hopes to go beyond Harff s statistical analysis and be able to pinpoint several key combinations that can lead to the occurrence of genocide. To carry out a QCA, first all values of variables, both independent and dependent, must be coded dichotomously for all cases is this condition present or absent in each case? Next, a so-called truth table is created in which each logical combination of independent variables constitutes one row. For instance, if a study has three variables A, B and C leading to the outcome E, there will be eight rows of possible combinations in the truth table. In this example capital letters signal the presence of a variable, while the symbol ~ before the letter denotes the variable s absence in the left-hand column of Figure 1. The asterisk * is to be read as a logical and, meaning that all these variables must happen together for the effect to occur. The truth table then assigns to these combinations whether the outcome is present (E) or not (~E); in most rows this coding is clear, but in some rows there are cases with diverging outcomes meaning that the researcher must decide whether to include the rows

10 as present, absent or ambiguous. Ambiguous cases (or Don t care in QCA terms and represented by a dash -) mean that this particular combination of variables is permitted to be both E and ~E during the Boolean reduction. Also rows can be excluded from the analysis. Logical combination Number of cases with outcome present Number of cases with outcome absent Classification for QCA reduction 1 A * B C 2 E 2 A * B * ~C 1 5 ~E 3 A * ~B * C 4 E 4 ~A * B * C - 5 A * ~B * ~C 3 E 6 ~A * B * ~C 1 ~E 7 ~A * ~B * C 3 ~E 8 ~A * ~B * ~C 2 E Figure 1: Demonstration QCA truth table By reducing the cases, one can find two combinations of factors that demonstrate sufficiency for the outcome to occur: by combining A * B * C and A * ~B * C, one can say that B is irrelevant for the outcome, but that the presence of A and C together are sufficient. Likewise A * ~B * ~C and ~A * ~B * ~C can be reduced to ~B * ~C. Altogether then the Outcome E can be described as being caused by the following solution formula, in which + should be read as a logical or, meaning that either combination is sufficient for the occurrence of E: A * C + ~B * ~C Interestingly, C is an important part of both outcomes, although in one combination it is present and in the other absent. A statistical regression analysis would not be able to pick up on this and would most likely discount the influence of C on the outcome, even though it just acts differently in different combinations. Logically, A * ~B * C and A * ~B * ~C can be reduced to A * ~B also; however, this brings no extra classification of cases not already covered so is normally excluded from the results. 5. Model Specification 5.1 Case Selection Key to any empirical study is selecting the right cases, and in order to avoid any bias in this selection process (for instance by choosing cases which are a good fit with a researcher s preconceived ideas of such phenomena and the relevant hypotheses) we have chosen to work from the same list as Harff (2003). She has compiled a list of 42 cases (Harff 2009a) classified as genocide according to her definition as explained above and building on the Political Instability Task Force s (PITF) State Failure Problem Set which in its updated version today includes all countries in the world from the years

11 However, some of these genocides have occurred with very low absolute fatality numbers and in order for this study to show the central dynamics leading to genocide occurrence in general, it will restrict itself to the worst instances of genocide. This is particularly important to be able to differentiate the cases of genocide clearly from the non-genocides included in the study; in cases of less extreme genocide, there would be a bit of overlap with bloody civil wars in which civilians are targeted while these are conceptually different, there is sometimes empirically a problem in differentiation between them, particularly with low numbers of death. Hence, this extreme case approach will highlight the variables which lead to genocide and non-genocide in the most direct way possible. This paper thus takes all cases of genocide from Harff s list which score four or more (out of five) on the variable DEATHMAG, which signifies a magnitude of at least 64,000 annual deaths, for at least one year during the genocide. In order to rule out interaction effects over time, each country is only permitted to be chosen as a case once, with the higher fatality instance being selected. Ultimately thus there are 11 cases of genocide included, ranging from the 1959 suppression of counterrevolutionary elements of society in China to the 1994 genocide in Rwanda. 2 Non-genocides conceptually should be cases in which genocide did not occur but in which it is plausible that they could have. Such counterfactual cases are relatively hard to identify reliably, and the best attempt of this in this paper is to take countries in which according to the State Failure Problem Set an armed conflict (ethnic or revolutionary war) is being fought and then take the conflicts with the highest number of fatalities, including both combatants and non-combatants. The logic behind this approach is that most genocides occur in the context of war (one consensus point within genocide studies), and that this provides a conducive environment for genocide to come about; this is a context in which it is plausible that genocide could occur, and yet it does not. Again, to rule out interaction, non-genocides are not selected in a country in which a genocide has occurred (also in countries with lower magnitude genocides not included in the study), and only one conflict is included per country. In order to ensure that the same number of cases of genocide and non- 1 Operational guidelines applied by Harff (2003, 58f.) in compiling this list are: 1. Complicity of the state or contending authorities; 2. Evidence of intent; 3. Victims as members of an identifiable group; 4. Policies and practices that cause prolonged mass suffering ; 5. Threat posed to survival of the group by these actions. 2 The case list is Cambodia ( ), Indonesia ( ), Nigeria ( ), Pakistan (1971), Rwanda (1994), Sudan ( ), Afghanistan ( ), Bosnia ( ), Burundi ( ), China (1959), Iraq ( ).

12 genocide are included the threshold is set at more than 5,000 fatalities in the worst year of a conflict, resulting in 11 cases too Variables operationalisations In most key variables operationalisation I follow in Harff s footsteps, taking the data from her model dataset (Harff 2009b), which is based on the State Failure Problem Set. The QCA model for this paper features the occurrence of genocide as the dependent variable (coded 1 for occurrence and 0 for non-occurrence) and five key variables as independent variables. The values of these variables are selected from the first year of the genocide or non-genocidal conflict (or as close as possible when data points are missing), in order to avoid reverse causality by which the genocide or non-genocide itself would impact these variables which are supposed to be determining its outbreak. Here we will only briefly detail the operationalisation of variables taken from Harff (2003), please see her study for further details. Political upheaval (P) Harff (2009c) operationalises political upheaval (P) using the the sum of the maximum magnitude of events in the prior 15 years, including revolutionary wars, ethnic wars, and regime crises from the State Failure Problem Set. In the cases included in this study, political upheaval values range from 0 to 33, although two clusters are apparent: value 4 and below, and 9 and above; thus we code anything as P = 0 if the value is 4 or below, anything of 9 or higher as P = 1, making 7 cases of political upheaval and 15 with this variable absent. Exclusionary ideology (I) Harff (2009c; see also Harff 2003, 63) codes exclusionary ideology herself, as belief systems that are articulated by governing elite, and that identify some kind of overriding purpose or principle that is used to restrict, persecute, or eliminate categories of people who are defined as antithetical to that purpose or principle. Hereby, 1 signals the presence of such exclusionary ideologies among the elites (6 cases), and 0 their absence (16 cases). Autocratic regime (A) This indicator of whether a regime is autocratic (1) or non-autocratic (0) is taken from the Polity IV dataset variable POLITY, which subtracts a country s autocracy score (scale 0-10) from its democracy 3 The cases are Colombia ( ), Lebanon ( ), Liberia ( ), Mexico (2007 and ongoing), Nicaragua ( ), Romania (1989), Russia ( ), Tajikistan ( ), Yemen ( ), Zimbabwe ( ), Chad ( ), Congo-Brazzaville ( ), Croatia ( ).

13 score (scale 0-10); if POLITY is between -10 and 0, A is coded as 1 (19 cases), otherwise it is coded 0 (3 cases). These data points are again taken from Harff s dataset. 4 Discrimination resulting from ethnic and religious cleavages (D) Regarding ethnic and religious cleavages and discrimination, Harff (2003) tested several different variables, although primarily of interest here is only the one which was significant in Harff s study: whether elites disproportionately hail from one ethnic group. Nonetheless, another measure regarding state-level discrimination has also been included in the QCA as an alternative measurement. For the main variable of discrimination due to elite ethnicity we differentiate between elite ethnicity not being salient (0) (5 cases) and ethnicity being salient whereby the political leadership is representative of a majority or minority communal group or coalition (1) (17 cases). In Harff s analysis, she divides the latter category into two; however, in order to dichotimise this variable appropriately, it is theoretically more plausible to distinguish between whether ethnicity is salient than between majority and minority representation, then including majority representation with the non-salient category. A second variable which was created for the State Failure Problem Set from Minority at Risk data is official state discrimination. To create a dichotomous variable here, we added the values for active political discrimination and active economic discrimination of a group together, and then coded any variable below the average as 0 (14 cases) and above the average as 1 (8 cases). In fact, there were no values close to the average, making this dichotomisation clearly justifiable. Economic autarky (E) Data were missing for most relevant countries in Harff s dataset, so a new variable was created using World Bank data. Import and export data were added together (as percentages of the country s gross domestic product in the relevant year) to indicate the trade openness of a country. Case were coded as economically autarkic (1) if their trade percentage was below the average of all cases in our dataset (12 cases), and as economically interdependent (0) if these trade indicators were above average (10 cases). Again, no cases are close to the average, making none of the codings controversial. 4 For the precise coding procedure please see Marshall et al. (2011, 15ff.).

14 Coding problems The following variables had no data available in Harff s dataset (or the alternative sources which were used) for these specific cases, thus necessitating a manual coding after a qualitative assessment. None of the five codings is particularly controversial. Colombia exclusionist ideology: The period known as La Violencia saw a violent civil conflict between the Conservative and Liberal parties; the two sides were in a bitter struggle vying for power, however this was not founded on a belief system that justified the restriction or elimination of the other side. Each side simply wanted to take power. Thus this variable was coded 0. Colombia discrimination due to elite ethnicity: In La Violencia the ethnicity of the elite was not particularly salient, as the main issue was their social status. While ethnicity did run as a more or less parallel cleavage (white landowners and indigenous peasants), the conflict itself was not framed in ethnic but in social and economic terms, thus dictating a coding of 0 for this variable. Iraq economic autarky: This case is coded as economically interdependent (not autarkic = 0) as the country was vulnerable to outside economic pressure due to its high oil exports. While in 1988 the effects of the Iran-Iraq war ( ) had reduced previously high oil exports and miscellaneous imports to a lower level, this recovered to a certain degree in Lebanon economic autarky: The Lebanese economy was extremely vulnerable to external influence and thus coded as not autarkic (0). As one of the most open economies in the region it had very high exports and imports. North Yemen economic autarky: The North Yemeni economy has been classified as autarkic (1) as the country s economy was characterised by low trade relations with other countries prior to 1962, with the economy based on feudal relations and an underdeveloped infrastructure, rendering the economy relatively isolated from the rest of the world. 6. Why do Genocides Occur? Empirical Results from the QCA Case data To approach this data with QCA, first each variable must be specified for each of the cases. Figure 2 lists all cases included in this study and their respective values for the relevant variables. 5 Purely by looking at this unsorted table it is plain to see that autocratic regimes (A) are present in all genocide cases, making them necessary for the outcome of genocide occurrence. However, some cases of non- 5 The alternative measurement of ethnic discrimination is not featured here, but results will be reported as necessary during the analysis.

15 Figure 2: Overview of cases with their respective values on all major included variables. genocide also occur in autocracies thus an autocracy is a necessary condition for genocide occurrence, but not sufficient. In a next step, one can analyse what conditions are sufficient to cause genocide occurrence. This analysis is the cornerstone of QCA and requires the re-categorisation of this information into a truth table sorted by combinations of variables. Truth table Genocide Occurrence (G) Political Upheaval (P) Exclusionist Ideology (I) Economic Autarky (E) Elite Ethnicity Discrimination (D) Autocratic Regime (A) Cambodia Indonesia Nigeria Pakistan Rwanda Sudan Afghanistan Bosnia and Herzegovina Burundi China Iraq Colombia Lebanon Liberia Nicaragua Russia Tajikistan Yemen Zimbabwe Chad Congo Brazzaville Croatia Total The truth table (Figure 3) has 32 rows to accommodate all possible logical combinations of the dichotomous variables. Of these theoretical combinations 13 have actual empirical cases, while 19 have no empirical observations; however, this does not mean that these combinations do not actually exist, but merely that it is not part of this sample. As there is no way of knowing whether this combination would in reality actually be a genocide or not, combinations of variables with no empirical cases are coded as don t cares ( - ) meaning that they can take on the property of genocide or non-genocide during the reduction processes. Of the 13 combinations with empirical

16 Combination Political Upheaval (P) Exclusionist Ideology (I) Economic Autarky (E) Elite Ethnicity Discrimination (D) Autocratic Regime (A) Cases (of which G) Consistency Outcome G 1 ~P ~I ~E D A (1) ? 2 ~P ~I E D A (2) 0.5? 3 P I E D A (3) ~P ~I ~E D ~A (0) ~P ~I ~E ~D (0) ~A 6 ~P ~I ~E ~D A (0) ~P ~I E ~D A (0) ~P I E ~D A (1) ~P I E D A (1) P ~I ~E D A (1) P ~I E ~D A (0) P ~I E D A (1) P I ~E D A (1) ~P ~I E ~D ~A (0) - 15 ~P ~I E D ~A (0) - 16 ~P I ~E ~D ~A (0) - 17 ~P I ~E ~D A (0) - 18 ~P I ~E D ~A (0) - 19 ~P I ~E D A (0) - 20 ~P I E ~D ~A (0) - 21 ~P I E D ~A (0) - 22 P ~I ~E ~D ~A (0) - 23 P ~I ~E ~D A (0) - 24 P ~I ~E D ~A (0) - 25 P ~I E ~D ~A (0) - 26 P ~I E D ~A (0) - 27 P I ~E ~D ~A (0) - 28 P I ~E ~D A (0) - 29 P I ~E D ~A (0) - 30 P I E ~D ~A (0) - 31 P I E ~D A (0) - 32 P I E D ~A (0) - Figure 3: Truth table listing all logical combinations of variables, their empirical frequency and the coding as used in the analysis. The dash - signifies don t care (meaning that these combinations can be used as either genocide occurrence or non-occurrence in the reduction algebra), while? indicates empirically ambiguous combinations, of which both genocides and non-genocides are to be found in the dataset (in the analysis these? will be turned to - or 0, as specified). examples, 11 are clear regarding the outcome and can thus be coded as such. The remaining two combinations encompass 4 cases each (combinations 1 and 2), and are marked by question marks (?), as they are not clearly consistent and their coding must be discussed in detail. Possible solutions would be to code them both as 0, so as not to include any non-genocides in the genocide category, or vice versa as 1 to ensure that all genocides are covered, even if some non-genocides get mixed up

17 in the categorisation, too. Alternatively, however, the two fields could be excluded completely from the analysis or marked as don t cares, swinging both ways depending on what is needed in the reduction. For combination 1 it makes sense to code this as 0 as there are three non-genocides in this category and only one genocide the one genocide case in this context is Bosnia, which is thus now coded as non-genocide. 6 Combination 2 on the other hand has both two genocides (Pakistan and Burundi) and two non-genocides (Tajikistan and Chad) and this was run twice with this combination once excluded and once coded as don t care. While there were slight differences in the outcome, these were relatively small. We here report the results for the combination 2 being excluded completely from the analysis. Nonetheless, this does mean that the genocide cases of Bosnia, Pakistan and Burundi will not be able to be explained in this model. Results and Interpretation Key determinants of genocide occurrence As mentioned above, one very central insight from this QCA procedure is that the presence of autocracy is a necessary condition for the occurrence of genocide. The truth table, however, lets one also find out further central conditions which are sufficient for genocide occurrence, that is that the presence of this combination will always lead to the outcome, although not all instances of the outcome will be caused by this combination. When the data from Figure 3 is reduced according to the principles of Boolean algebra (see section 4), one receives the following formula: Solution 1: P * D * A + I * E * A Substantively this means that genocide occurs in autocratic countries in which either there is political upheaval combined with an elite whose ethnicity is salient, or there is an exclusionary ideology combined with an autarkic economy. Which cases are covered by these combinations are listed in Figure 4, along with their consistency and coverage scores. Consistency scores indicate how good a fit a particular combination is, that is how many non-genocides are also described by this term. Coverage scores show how many of the genocide cases can be explained by this solution raw scores detail how many cases are covered by this combination, while unique coverage means the cases which are explained only by this combination. To a certain degree, there is a trade-off between the consistency and coverage scores, as a solution with a higher consistency (thus explaining only the phenomenon itself and less other cases too) will possibly not be able to explain the same amount of breadth as a very inclusive solution which explains all cases but also includes some non-genocides 6 As it is controversial coding Bosnia as a non-genocide, the analysis was also run with this combination coded as don t care ; the results were nearly identical, the only difference being that in the more parsimonious solution A *D was a combination, and was not interchangeable with P * D and E * D. This is a small difference given that the cases covered are identical. Also, on alternative coding for this case please see footnote 7 below.

18 Cases covered Raw coverage Unique coverage Consistency P * D * A I * E * A Cambodia, Indonesia, Nigeria, Rwanda, Sudan, Iraq Cambodia, Indonesia, Sudan, Afghanistan, China Solution coverage: Solution consistency: 1.0 Figure 4:Solution 1 case coverage and consistency by combinations too. The consistency scores of 1 in this case, however, indicate that there are no non-genocides occurring with these combinations of variables, thus making these combinations truly sufficient for causing genocide occurrence. The score for overall coverage is explained by the exclusion of Pakistan and Burundi, and with the re-coding of Bosnia as 0. Thus these three cases are not explained by this conjunctural solution. Both the combinations are substantively plausible. First, an autocratic system with elites whose ethnicity is particularly salient which comes into a time of political upheaval is susceptible to genocide, as these elites will want to try and cling to their position of power in the upheaval by mobilising along ethnic lines. This mobilisation can then easily lead to a demonisation of the ethnic group which the elites fear could take over power and the subsequent call for their elimination. A prime example of this combination is the 1994 genocide in Rwanda in which the autocratic Hutu majority government (in which their Hutu ethnicity was certainly salient) feared that it would lose power through the monumental changes happening in the country, most specifically through the power-sharing agreement which was being signed in the Arusha Accords. This agreement would have given the Tutsi ethnic group significant powers in government, thus causing Hutu radicals to mobilise along ethnic lines against this agreement and against the Tutsis as a whole. This attempt to retain power during a political upheaval underlines the dynamics of this combination very well. 7 While the P * D * A emphasises the motivational structure that can lead some elites to pursue genocide as a strategy, I * E * A also incorporates the importance in some cases of an opportunity structure which is conducive to genocide. This second combination again in an autocratic regime 7 A very similar situation was present in Bosnia in 1992 in the context of the dissolution of Yugoslavia. The autocratic regime had elites whose ethnicity was certainly salient, however, according to Harff s dataset Bosnia was not in a context of political upheaval in Even if the situation in Bosnia does not match the State Failure Problem Set definition of political upheaval, it can well be argued that Bosnia was undergoing a severe and unsettling transition which, as in Rwanda, caused elites to mobilise along ethnic lines to sustain their power. In order to stay consistent with Harff s analysis, this value for political upheaval was not changed, but an alternative coding would lead to a fit on this case to, and thus also higher coverage scores.

19 is based on the interplay between an exclusionary ideology and economic autarky, with the exclusionary ideology suggesting the motivation for genocide, while the lack of economic ties provides a conducive environment for this plan to be executed. A country with few economic ties has several key advantages for a genocidally inclined elite: first, it has relatively little to fear from economic sanctions by other countries, as these will not impact the overall economy very strongly; second, as countries seldom intervene militarily in other countries purely out of humanitarian compassion, powerful countries are less inclined to attempt prevention of the genocide militarily as they have few economic interests in the genocidal country. An alternative, more parsimonious solution The data can, however, also be reduced to an alternative format. Solution 2 tries to reduce the data to as few variables as possible. While this provides a more parsimonious presentation of the data, it lacks some of the causal insights of the above suggestion: Solution 2: I + P * D + P * ~E This solution provides three alternatives which are each sufficient for causing genocide. First, the presence of an exclusionary ideology is enough to cause genocide; second, political upheaval in a polity in which elite ethnicity is salient; third, political upheaval in a context of economic interdependence (lack of economic autarky). The cases covered and their coverage and consistency are listed in Figure 5. However, P * D could be replaced with either A * D or E * D as these all cover the same cases equally, with the same consistency and coverage scores. Compared with Solution 1, it is striking that the variable A is missing from all three combinations. This is not necessarily surprising, however, as an autocratic regime is a necessary condition, but by no means must it be a sufficient condition. The presence of an exclusionary ideology, however, is sufficient, as only cases of genocide have such ideologies; given the essence of such ideologies, this is also substantively plausible. Also, the second combination of variables (P * D) is similar to above, following the same logic of emphasising the salience of elite ethnicity in the context of political upheaval. The final combination, however, is substantively different with genocide being caused by political upheaval combined with a country which is interdependent economically. While this could be argued along the lines that trade openness allows for an easier flow of weapons, this is certainly not the central dynamic in the two cases it purports to explain (Rwanda and Iraq) and it must be said that they are both much better explained by other combinations of variables. 8 8 While millions of machetes were imported into Rwanda prior to the genocide, this cannot be seen as the central dynamic leading to genocide. This was merely an action as part of the preparations for the genocide which was caused by other, much more substantial reasons.

20 Cases covered Raw coverage Unique coverage Consistency I Cambodia, Indonesia, Sudan, Afghanistan, China, Iraq P * D Cambodia, Indonesia, Nigeria, Rwanda, Sudan, Iraq P * ~E Rwanda, Iraq Solution coverage: Solution consistency: 1.0 Figure 5: Solution 2 case coverage and consistency by combinations Altogether, the more parsimonious Solution 2 does not appear to actually simplify our understanding of genocide occurrence, with the confusing role of P * ~E and the interchangeable status of P * D, A * D or E * D; moreover, the slightly more complex combinations in Solution 1 also pay homage to the important role of the necessary condition of the presence of an authoritarian regime. Thus, the best presentation of the results of the QCA in this paper remain P * D * A + I * E * A. Alternative operationalisation of ethnic discrimination As expounded on in the theoretical section, there are several different ways in which ethnic diversity can impact genocide occurrence. The variable used above measured whether the ethnicity of elites was salient, but the QCA was conducted also with variables measuring the occurrence of state-led political and economic discrimination. This variable resulted in the same results as described above even though in Harff the variable remained insignificant. This demonstrates that the quantitative methods used previously could not pick up on similar effects that were only demonstrated using QCA s conjunctural logic. Determinants of genocide non-occurrence It is also interesting to look into what causes can be seen as sufficient and necessary for genocide not to occur in a country. By again looking at the table in Figure 2, one can say that the absence of an exclusionary ideology (~I) is a necessary condition, as all cases lack such ideologies, but it is not sufficient as there are several genocidal cases without exclusionary ideologies. While the absence of political upheaval is nearly a necessary condition, there is one case (Colombia) in which political upheaval occurs without resulting in genocide. Thus, this cannot be a necessary condition. Figure 6