To probe and test the robustness of these results and consider potential mechanisms, I perform three additional analyses. First, to analyze if the definition of genocide
influenced the findings, I examine a smaller subset of cases based on a more restricted definition. Then, as some may argue that not every country considered was truly at risk of genocide, I restrict the risk set to countries that experienced a civil war, the strongest predictor of genocide. Lastly, as some may suggest that an exclusionary ideology is part of the process of genocide, I treat it as a dependant variable and analyze how the factors considered thus far influence its presence.
To start, it has been argued that the guidelines the PITF used to create their dataset were too broad (e.g., Mayersen 2010). While there will never be complete agreement regarding which cases constitute genocide, I address these concerns by restricting the 30 cases included above to a smaller sample of 15 cases, relying upon Fein’s (1993a) coding as explained above. Due to the small number of events, I again
start with simpler analyses of each individual predictor and time. All effects shown in Table 2.2 remain similar except the effect of trade openness, which is no longer significantly associated with lower odds of genocide. Multivariate analysis with rare events discrete time logit models (not shown)43 showed some evidence of instability in the predictors due to the small number of events, though results generally affirmed those reported in Table 2.3. Specifically, exclusionary ideologies, civil wars, coups, and autocratization remain associated with higher odds of genocide in these models, while trade openness and salient elite ethnicities are no longer significantly associated, though the direction and magnitude of the effects remain similar. International war is
significantly associated with the odds of genocide in these models.
As to the potential argument that some countries, such as Great Britain or Canada, were not truly at risk of genocide during the time period examined in this study, I restrict the sample to the 51 countries that experienced a civil war between 1955 and 2005 and set the time of risk to the first year a civil war took place. In other words, countries are not considered at risk of genocide unless a civil war had taken place. This results in a slightly smaller number of genocides (28), illustrating that civil war is not a necessary risk factor of genocide.
Key findings remain in these models (not shown), as aspects of intergroup
dynamics, national politics and economy, and international dimensions influence the odds of genocide. Ethnolinguistic diversity is not significantly associated with higher odds of genocide, though salient elite ethnicity and exclusionary ideologies are. Turning to the state, the key difference is that larger populations have higher odds of genocide, which
43 Variables that were not significant are not included (such as youth bulge or democracy)
suggests that the effect of population is mediated by the effect of civil war in Table 2.3. Beyond this, the situation in which the regime operates significantly impacts the odds of genocide, with coups and revolutions each showing strong, positive effects. In terms of international dimensions, international war is significantly associated with higher odds of genocide, though the effects of trade openness and membership in INGOs are not
significant, again suggesting that the influence of the international system may become less efficacious when a country plunges into civil war. Thus far, both tests of robustness confirm most findings but call the effect of trade openness into question. In light of this, and as this variable is only significant at the .1 level in the final model, future studies should analyze the mechanisms through which trade may impact the onset of genocide.
Lastly, as the exclusion of a group is a key element of genocide, some may argue that an exclusionary ideology is part of the process of genocide and could be treated as a dependant variable. Indeed, both Mayersen (2010) and Hagan and Rymond-Richmond (2009) begin their process-oriented models of genocide with the establishment of an out- group and an ideology that excludes that out-group. Further, as many genocides were planned prior to the year in which they took place, predicting an exclusionary ideology may be an important aspect of understanding genocide.
Due to these considerations as well as potential issues of endogeneity, I examine how theories reviewed in this paper explain whether an exclusionary ideology existed in any of the 153 countries in the dataset over the time period analyzed. As outlined in Table 2.1 and further detailed in Appendix B, exclusionary ideologies are defined as “belief systems that are articulated by governing elite that identify some kind of
of people who are defined as antithetical to that purpose or principle.” More than one- third of the countries examined had an exclusionary ideology at some point during the study, and many of the countries that experienced genocide had an exclusionary ideology present at some point prior to the violence. Independent variables can again be found in Table 2.1 and are analyzed under the assumption that theories that pertain to the
occurrence of genocide also pertain to the presence of an exclusionary ideology. To test their influence, I utilize logistic regression with clustered standard errors, as an
exclusionary ideology is not a singular event and could manifest multiple times over multiple years.44
Results are presented in Table 2.4, and it is immediately clear that many of the variables associated with the occurrence of genocide are also associated with
exclusionary ideologies. In fact, more of the variables tested are significantly associated with the presence of an exclusionary ideology (Table 2.4) than with the occurrence of genocide (Table 2.3), suggesting that many theories about the onset of genocide may more accurately explain the presence of an exclusionary ideology.
Looking at societal-level factors, more diverse societies have significantly lower odds of exclusionary ideologies. This provides support for the finding that ethnolinguistic diversity is not associated with higher odds of genocide and shows that more diverse societies actually have lower odds of exclusionary ideologies, perhaps due to intergroup contact (Allport 1954). In addition, as with genocide, societies in which the ethnicity of the ruling elite is a point of contention have significantly greater odds of an exclusionary ideology.
44 Note that I am analyzing the best way to include timeperiod effects in the analysis but
Table 2.4 Logistic Regression of Exclusionary Ideology, 1955-2005 Results in Odds Ratios
Model Model Model Model
Predictors 1 2 3 4 Society Elite Ethnicity (lagged) 4.426*** 3.352*** 2.854*** 2.792*** (2.299 - 8.519) (1.647 - 6.822) (1.406 - 5.790) (1.363 - 5.719) Ethnolinguistic Diversity 0.164*** 0.109*** 0.098*** 0.095*** (0.051 - 0.530) (0.032 - 0.375) (0.028 - 0.344) (0.027 - 0.336) State Democracy 0.857*** 0.860*** 0.854*** (0.802 - 0.916) (0.810 - 0.914) (0.802 - 0.908) Population (log) 1.295* 1.210 1.092 (0.995 - 1.686) (0.851 - 1.720) (0.775 - 1.540) International Colonial Past 3.894** 3.909** (1.094 - 13.855) (1.129 - 13.531) Trade Openness 0.991** 0.989*** (0.983 - 0.999) (0.982 - 0.997) INGOs (log) 0.938 0.986 (0.692 - 1.271) (0.722 - 1.347) Upheaval Civil War 2.192* (0.997 - 4.816) Coup 0.499*** (0.332 - 0.749) International War 1.412** (1.080 - 1.845) Constant 0.304*** 0.030*** 0.078 0.149 (0.181 - 0.511) (0.002 - 0.373) (0.004 - 1.665) (0.007 - 3.032) Observations 6,310 6,310 6,310 6,310 McFadden’s R2 0.072 0.195 0.240 0.249
Robust confidence intervals in parentheses *** p<0.01, ** p<0.05, * p<0.1
Turning to the type of state, democracies have much lower odds of exclusionary ideologies. State upheaval in the form of civil war is associated with higher odds of an exclusionary ideology, though the magnitude of this effect is much smaller than the effect of civil war on genocide. Revolutions (not shown due to their high correlation with civil
wars) are also significantly associated with higher odds of an exclusionary ideology, though a coup has a negative, significant effect (consistent with bivariate analyses). This may mean that regimes with exclusionary ideologies are more difficult to topple or that exclusionary ideologies seldom occur with a coup, perhaps because new leaders seek support from the population (both possibilities should be explored in future research). International war has a strong, positive effect.
In terms of other international factors, each additional percent of time spent under imperial rule is associated with a four-fold increase in the odds of an exclusionary
ideology. This underscores the finding that while colonialism is not directly associated with the onset of genocide, its effects manifest in other ways. By contract, trade openness is significantly associated with lower odds of exclusionary ideologies, providing support for the effect of certain forms of international interconnectedness.
Again, theories that pertain to the onset of genocide are also predictive of an exclusionary ideology and, in fact, better explain the latter. An exclusionary ideology is clearly linked to genocide, as genocide involves the intent to destroy a group of people. While this factor (as measured) was not present before every genocide, ideologies of exclusion and other forms of classification are often the first steps in process-oriented models of genocide (Stanton 1996; Mayersen 2010). Thus, while an exclusionary ideology is not a necessary or a sufficient condition, it is a mechanism for genocide. Future researchers’ models should examine the factors that lead to an exclusionary ideology and explore why some ideologies lead to genocide while others do not.
Duration of Genocide
Thus far, I have examined the factors associated with the onset of genocide and tested their robustness. Before concluding, I briefly turn to one other aspect of genocidal conflict: its duration. Comparative studies of genocide have yet to consider the factors that influence the duration of violence. While many of these factors are likely found at sub-national levels and potentially linked to the actions of external actors (such as armed interventions or economic sanctions), it is also plausible that more macro-level factors studied here influence the duration of genocidal violence. In fact, studies of the duration of civil war have found that, among factors related to the type of war (Fearon 2004), low economic development and moderate ethnic heterogeneity are associated with longer civil wars (Collier, Hoeffler, and Soderbom 2004).
To examine whether factors studied here also influence the duration of genocide, I draw upon a Weibull model, as described in the methods section. The dataset is restricted to episodes of genocide, resulting in 30 countries. The genocides lasted from 1 year (the lowest possible) to 20 years (see Appendix A for the duration the PITF assigned to each case), and all independent variables are, again, described in the data section and Table 2.1. Note that I also ran analyses with Fein’s coding, and all effects presented remain similar.
Results, shown in Table 2.5, illustrate that factors that influence the onset of genocide do not generally influence its duration. Salient elite ethnicities provide one example. Yet, exclusionary ideologies, which are key factors related to the onset of genocide, are associated with significantly shorter genocides. This may suggest that
Table 2.5 Weibull Model of the Duration of Genocide, 1955 to 2005
Model Model Model Model
Predictors 1 2 3 4 Society Exclusionary Ideology -0.751* -0.976** -0.946** -1.105** (-1.575 - 0.073) (-1.870 - -0.081) (-1.839 - -0.053) (-2.069 - -0.141) Elite Ethnicity 0.116 0.338 0.437 0.959 (-0.917 - 1.148) (-1.052 - 1.728) (-1.020 - 1.895) (-0.734 - 2.653) Ethnolinguistic Diversity -2.513*** -4.454*** -4.409*** -4.741*** (-4.076 - -0.950) (-6.266 - -2.641) (-6.224 - -2.594) (-6.718 - -2.765) State Democratization -0.104 -0.092 -0.104 (-0.592 - 0.384) (-0.574 - 0.390) (-0.614 - 0.407) GDP per capital (log) -0.058 -0.018 0.142 (-0.774 - 0.658) (-0.760 - 0.723) (-0.647 - 0.931) Population (log) 1.042*** 1.021*** 1.068*** (0.684 - 1.399) (0.656 - 1.386) (0.660 - 1.477)
International and Upheaval
Trade Openness -0.004 -0.004 (-0.017 - 0.010) (-0.017 - 0.009) Civil War -0.252 (-1.282 - 0.778) International War 0.194 (-0.161 - 0.549) Coup -0.454 (-1.464 - 0.556) Constant -2.559*** -12.112*** -11.937*** -12.510*** (-3.970 - -1.148) (-16.333 - -7.890) (-16.220 - -7.655) (-17.089 - -7.930) Observations 174 174 174 174 Events 30 30 30 30 *** p<0.01, ** p<0.05, * p<0.1
genocides where there was already in- and out-groups formed may see swifter violence. Ethnolinguistic diversity is also significantly associated with shorter episodes of
genocide, which may mean that the presence of other groups helps bring the violence to a close. However, probing this effect by examining moderate levels of ethnolingustic
diversity (.4 to .6) shows that moderate levels are actually associated with longer
genocides, paralleling findings related to the duration of civil war (Collier, Hoeffler, and Soderbom 2004).45
Turning to national politics and economy, the type of government,
democratization, and economic development are not associated with the duration of genocide. A country’s population is, however, and an increase in logged population increases the duration by a factor of 3.4. This finding may indicate that larger countries have longer genocides; it may also indicate that more potential victims increases the length of the conflict. In addition, the presence of civil war—the strongest predictor of genocide—is not significantly associated with the duration of genocidal violence. Other forms of upheaval, including coups and international war, are also insignificant.
Overall, the interpretations of the effects presented here are largely speculative, and future studies should spend more time assessing the factors that influence genocidal duration and its impacts on measures like death tolls or regional stability.46 What is significant here is the finding that factors associated with the onset of genocide do not significantly impact its duration. Factors associated with the state and its context are most important for understanding the onset of genocide, though it may be factors at societal and international levels that influence the duration, calling for alternate theories as well as
45 Moderate levels of ethnolinguistic diversity are not associated with the onset of
genocide, however.
46 Again, analyses were run with Fein’s restrictive case set of genocides, and effects
remained.
analysis at sub-national levels, which I assess throughout this dissertation and to which I return in Chapter 6.