TOWARDS A BETTER UNDERSTANDING OF GENOCIDE

Our conceptual model draws on the work above in two ways. First, we believe that an improved understanding of genocide will draw on insights from both the qualitative and quantitative literature and not just the literature on genocide per se, but from the larger field and much broader work on domestic conflict and contentious politics. Here, one of the more powerful theoretical approaches seeks to understand individual instances of killing, torture, rape and beating that take place during the genocide (violations) as a result of individual calculations of costs and benefits by authorities as well as citizens, where varying levels of mobilization and structural conditions affect these costs and benefits. Second, drawing on the insights from the qualitative literature on genocide, initially studying a single case across space and time will allow for a better operationalization of structural conditions and mobilization processes – the former drawn from the literature on genocide and the latter drawn from the literature on conflict and contentious politics. For our study, we investigate the Rwandan genocide of 1994.

Why Study Rwanda?

The Rwandan genocide of 1994 is important not just because of what took place in 1994, but also because of what has taken place repeatedly throughout Rwandan history (e.g., the mass killings of 1959, 1962 and 1973), what is taking place today and what might take place in the future. Awareness of the case is quite extensive. Even casual observers of international politics are aware that Hutu extremists perpetrated acts of violence targeting all Tutsis and moderate Hutus. The genocide in 1994 took place over one hundred days (from April to July) when between 500,000 and 1,000,000 people were murdered with untold numbers suffering from other offenses (e.g., rape, torture, intimidation).

The rate of killing exceeds that of the Holocaust and it makes this the "most efficient mass killing since the atomic bombings of Hiroshima and Nagasaki" (Gourevitch 1998). Despite familiarity, however, it is not clear why certain parts of Rwanda experienced many violations, of different types, while others experienced hardly any at all. Related to this, mass participation (which involved not just the military, police and the Interawahamwe [local militias], but also farmers, students, common household members and priests), presents something of a puzzle. For us, the most interesting puzzle is how, when so many individuals could engage in rape, kidnapping, beating, torture, and killing, frequently against people that they knew – including close friends and family members, did others resist the threats made against them and instead of joining in the organized and spontaneous violence, placed themselves at risk by either refusing to participate or actively aiding those most threatened. A wide variety of answers have emerged – none of which are backed by any detailed, systematic analysis. Third, there is great deal of variation within the Rwandan political economy with different regions being affiliated with diverse political organizations, historical experiences with conflict and natural endowments thus serving as an ideal environment within which to assess variation.

A fourth reason for studying the case is that in many ways the genocide is still politically, economically and culturally relevant to and present within contemporary Rwandan (and indirectly European and American) society. The Rwandan government, with the assistance of the United States as well as other countries, is currently engaged in an effort of truth and reconciliation within “Gacaca” courts – which literally mean “justice in the grass.” These courts, which rely exclusively upon eyewitness testimony, are local efforts in every political cell of the country. Understanding the particulars of what took place in Rwanda during the time of the genocide would be useful for better understanding the Gacaca process. It is quite likely that having experienced large numbers of violations, certain locales will not have many witnesses, thereby allowing many individuals to be freed and potentially enraging others in different parts of the country. It is also likely the case that locales without extensive violations will have many witnesses but great hesitancy with coming forward in the highly visible and public court system. Here, again, perpetrators could be set free threatening the perceived legitimacy of the court as well as the centralized state political system.

The final reason for exploring the case is that many of the factors that we speculate prompted the genocide in the first place still exist. Based on interviews during the summer of 2002 of numerous political elites, rural mass publics and prisoners, the common thread across all sectors of society is fear of future large-scale political violence. There is still animosity between the different groups involved in the conflict (Rwanda’s majority Hutu population and the Ugandan Tutsi minority currently in power as well as the exiled Hutu Power [Pawa] organization that engineered the genocide – with members currently within Congo and Canada, among other countries). Rwanda is still incredibly poor and the highly regimented structure of government that existed before the genocide is still operative. Additionally, within the near future 40,000 of the 120,000 individuals currently in prison for genocidal activities will be returned to the general population. Given the common finding within the genocide literature that genocides often recur in the same countries this point becomes particularly important.

Theoretical Approaches and Hypotheses to Test Them:

In the literature on mass political violence, there are numerous theoretical themes or approaches. Below we develop sets of hypotheses drawn from two that are particularly relevant to the Rwanda context in order to better understand the relative importance of each approach (theory). The first theoretical approach is the structuralist view. From this perspective, individual acts and their national level aggregations result from an interaction of stochastic shocks (such as the Tutsi invasion of 1993) and sets of underlying “structural” factors such as regime structure/change, poverty, physical proximity, ethnic hatred and collective memories of previous events which set the context within which group action occurs (e.g., Piven and Cloward 1979; Skocpol 1979; McAdam 1982; Muller 1985; Kitschelt 1986; Tilly 1995; Davenport 1999). The second approach views events like the Rwandan genocide resulting from particularly intense political competition and concurrent mobilization along with incentives for individual survival. Here, working from a rational-political framework, political party mobilization factors, and elites’ ability to alter individual’s cost-benefit calculations directly linked to individual’s perception of their survival odds more directly affect the dynamics of the violent actions we observe (e.g., Goldstein 1978; Tilly 1978; Della Porta 1995; Lichbach 1995; Francisco 1996; Moore 1998). The distinctions between the two approaches are important. If we are able to find differing degrees of support for the hypotheses of one approach compared to the other, we may be able to identity better intervention strategies for both preventing future violence as well as reconciliation efforts for previous events. Next, we detail these theoretical approaches and our hypotheses related to them.

Structuralist Factors: The Contentious Politics Model. Within existing literature, the decision calculus of political authorities, who are contemplating the benefits and costs of genocidal activity, is influenced by diverse structural characteristics – specifically domestic crisis and state penetration. Following this general argument, but attempting to analyze relevant explanatory variables in a different way, we assess the sub-national variation that exists within several structural factors. For example, by most accounts, the ongoing civil war (1990-1994) in Rwanda would serve as the major contextual influencing violations. Many Hutus within Rwanda by the then Hutu government’s propaganda about the coming confrontation with the Rwandan Patriotic Front (RPF) and were informed that they would be killed by this organization if and when they took power – continuing a pattern of Tutsi domination and violence that had extended over generations (e.g., African Rights 1995; Gourevitch 1998; Human Rights Watch 1999; Mamdani 2001). As information about the civil war and movements of RPF into Rwanda came to individuals within the country via radio and newspaper but also by word of mouth, it is anticipated that,

Hypothesis 1a: The frequency and scope of violations increased as the RPF commanded more territory within Rwanda.

Moreover, in line with the disaggregation argument, we further expect that it would be easier to mobilize individuals when the RPF threat was more credible and immediate. Here,

Hypothesis 1b: The frequency and scope of violations were decreased and resistance increased in areas further away from RPF commanded territory within Rwanda.

According to the quantitative literature but also the anecdotal literature on genocide, it is extremely important that the state have control over the society in question in order to influence relevant activity. With this control, authorities could provide leadership, resources, personnel from government locations, communication, reinforcement for local perpetrators, and intimidation of those reluctant to participate; in short, control provides for the conveyance and enforcement of “selective incentives” to ensure individual participation in mass activity. With these concerns, it appears that pre-1994 Rwanda was built for genocide. As noted by Mamdani (2001, 144),

“(i)nspired by the ten-cell system in Tanzania (and South Korea [Kimonyo 2000, 10]), (beginning in 1978 the country was divided) into ten prefectures, and these into 143 communes of about 30,000 persons. Each commune comprised eight collines (hills), with each hill divided into secteurs, and each sector into two cellules (cells). Each cell was to comprise fifty families.”

As he continues,

“Neither the prefect nor the burgomaster, in charge of the prefecture and commune, respectively, was elected; both were administratively appointed from above. The prefect was like the colonial chief: he decided how many acres of coffee should be cultivated in each commune, and how many people should be put to work on roads and for how long. He alone was responsible for maintaining public order and tranquility and he had the authority to suspend the execution of ministerial orders when circumstances so required (Mamdani 2001, 144)”.

In this context, it appears that all one would have to do is direct all functionaries to the task of mass killing and it would be uniformly applied across the relevant territorial jurisdiction. But, this is all in the abstract. Actual state penetration and capacity likely reveals significant variation across the whole country. This is especially important within the Rwandan context because here the urban and rural environments were very different from one another, the capitol city (Kigali) was quite different from every other part of the country, and terrain within the countryside was extremely difficult to traverse (in that there are thousands of hills and not many paved roads). Because of this, we expect that:

Hypothesis 2a: Violations were decreased and resistance was increased in frequency and scope when Kigali was distant;

Hypothesis 2b: Violations were decreased and resistance was increased in frequency and scope when proximity to a road was high;

Hypothesis 2c: The scope of violations was decreased and resistance was increased when terrain was complex (mountainous);

Hypothesis 2d: Violations were decreased and resistance was increased in frequency and scope when communal/public spaces such as churches, schools, outdoor markets, sports stadiums were unavailable; and,

Hypothesis 2e: Violations and resistance were both decreased in frequency and scope if the locale existed next to another country.

Given the political structure of the country, it makes sense that if an official within the particular geographic locale in question (such as a prefect or burgomaster), adopted a position against the Hutu Power faction, then they would be able to reduce the likelihood that relevant information would be passed down and mobilized within the 10-cell structure below them. As a result,

Hypothesis 2f: Violations were likely decreased and resistance increased when an official was either affiliated with Tutsis and/or Moderate Hutus and increased when these individuals were eliminated/withdrawn.

The ability of the Hutus to spread an irrational culture of fear could also be hindered by the presence of outsiders. Although much attention has been given to the fact that the UN and individual countries withdrew from Rwanda, it is clear that such efforts had an impact when they were in effect. Therefore,

Hypothesis 2g: Violations were decreased when “Safe-Zones” existed and increased when they were eliminated/withdrawn.

The final hypothesis within this category concerns regional variation. A fair amount of attention has been given to the regional variation of political and agricultural culture that exists within Rwanda, which is directly relevant to the likelihood of violations. Although a small country, approximately the size of Vermont, with a population in 1994 of about 8 million, there were great differences across the regions. For example, there was a history of strong Hutu political power within the North (e.g., Ruhengeri) – the place from which the leadership during the Second republic emerged (governing the country from 1973 to 1994). Within this area, the focal point for numerous Tutsi military incursions, there was also a long history of anti-Tutsi persecution. In contrast, the center of the country (e.g., Gitarama, Kigali) was generally associated with a more conciliatory Hutu position. These were the individuals largely associated with the First republic (the post-revolutionary/independent regime), which ruled the country from 1962 to 1972. From this, we thus expect that

Hypothesis 3: Prior violations and resistance in a locale increased later violations and resistance within the same place.

Political Party Competition and Mobilization: The Rationalist Political Model

Significant discussion has been given to the role played by a political organization known as “Hutu Power” in the genocide (a political organization that challenged the existing president’s political party, which advocated not democratizing the political system, opening the borders to the Tutsi living outside of the country as well as negotiating with the RPF [Mamdani 2001]). Individuals in Hutu Power had an incentive to participate in violations for they were the ones who would lose the most politically and economically from an RPF victory (many had benefited from the expropriation of land from Tutsis earlier and their removal from positions in schools and throughout the workplace). These individuals were also the ones that had been increasing in popularity within the second republic (support for the existing party had been slowly deteriorating leading up to April 1994 [Kimonyo 2000]), and they were the ones whose ideology/identity was most closely associated with a pro-Hutu/anti-Tutsi position (where the former would receive preferential treatment in all matters and the latter would be essentially banned and removed from Rwandan society). Even more attention has been given to the various state militia organizations that were present at the time – commonly referred to as the Interawahamwe (social clubs associated with Hutu Power that were latter militarized and which served as the local enforcement mechanism for the organization [e.g., Human Rights Watch 1999]). Individuals in the Interawahamwe had an incentive to participate in violations for they were the ones that had the most to gain from the direct confiscation of land as well as the ones most likely to follow political organizations given their high degree of social vulnerability (they were poor and largely isolated from state-directed political activity).

Working alongside of and within the 10-cell structure identified above, these two political organizations engaged in violations and mobilized other individuals within the society not directly affiliated with either group to participate within violations thereby serving the role as political entrepreneurs. The process itself has been described numerous times (e.g., Gourevitch 1998; Human Rights Watch 1999; Berkeley 2001; Kuperman 2001; Mamdani 2001): members of Hutu Power and/or the militias would show up within a cell, bring all non-moderate Hutus together in one location, inform them who was to be killed, eliminate any resistance within this crowd and then, as a group, eliminate those targeted. Afterward the killing, the original political group would move on to the next location. Importantly, the presence of Hutu Power and their militias were not spread evenly throughout Rwanda (Ruhengeri and Gisenyi – two northern prefectures had a strong presence of Hutu Power when others did not [Kimonyo 2000]). Due to the geographic dispersion of the party organization and mobilization dynamics, we propose:

Hypothesis 4a: Violations were decreased and resistance increased when Hutu Power and militia presence was low; and,

Political diffusion was extremely important. Unlike many areas within the social sciences that discuss such issues, in the Rwandan context the mechanisms are fairly clear. During the genocide it has been noted that “runners” and drums were used to communicate between different cells about what had already taken place as well as what was to take place. Much attention has been given to the use of radio (e.g., Schabas 2000; Power 2002), but more precise information about who was to be abducted and/or killed was provided by runners and disseminated lists provided by political authorities. Consequently, it is expected that

Hypothesis 4b: Violations were decreased and resistance increased when Hutu Power and militia contiguity in the area was low.

Of course, Hutu Power and militias were more than just present within an area; they instigated and engaged in violations as well. This is important for some attention has been given to the fact that many violations employed militias and everyday Rwandans in the targeting individuals in their homes. These activities (which were generally smaller in magnitude) involved robbery, rape, arson and murder (Kuperman 2001. 15). Other violations (which tended to be larger) involved agents of the Hutu government including the regular army, reservists, the Presidential Guard and national police, each of which had differing proximity to the inner circle of the Hutu Power organization. These differences in perpetrators should influence the pace and scope of contentious activity: specifically,

Hypothesis 4c: The frequency of violations should be increased when the perpetrators included militias and everyday Rwandans but the magnitude of violations should be increased when the perpetrators included the army, reservists, Presidential Guard and national police.

Hypothesis 4d: The frequency and magnitude of violations should be increased and resistance decreased when police stations or military bases were nearby.

Personal Survival: The Individual Rationalist Approach

Following the arguments of Moore (1966), Wolf (1969), Scott (1976), Paige (1997), we also wish to control for what might be thought of as the “moral economy of the genocide participant,” which considers the overarching economic context that Rwandans found themselves in. In crude terms, it is suggested that within an essentially rural, peasant and extremely poor society (which clearly describes Rwanda [Human Rights Watch 2001]), most peasants are most interested in their survival. In this context, if local leaders generally provided for basic needs, then it is expected that obedience to various social tasks (regardless of what they involve) would occur. If basic needs were not met, however, then disobedience can be expected (Scott 1976). Within Rwanda, obedience takes on a very specific meaning for as many authors have identified, political leaders framed genocide participation as a form of communal labor obligation (Mamdani 2001, 146-147). Historically the state called on individuals to clear forests, build roads and to farm, and in the situation of genocide to participate in violations. Drawing upon the moral economy theory it is expected that:

Hypothesis 5: Violations were decreased and resistance increased in frequency and scope when there were communal economic benefits to be distributed.

The next few hypotheses concern the potential targets of violations and how their size, behavior and characteristics influenced relevant behavior. It is expected that Rwandans in 1994 were particularly attuned to the fact that a larger number of Tutsi and moderate Hutu represented a greater “threat” to those following the Hutu Power position; this was frequently addressed within newspapers, radio broadcasts, within town meetings and via megaphone. It was also clear that larger numbers also represented a more difficult situation for perpetrators to overcome (Kuperman 2001, 15). From this, one could anticipate that:

Hypothesis 6a: Violations were increased when the number of Tutsi and Moderate Hutu was small.
Accepting the collective action literature on contentious mobilization facilitated by small groups, however, we would also expect that:

Hypothesis 6b: Resistance was increased when the number of Tutsi and Moderate Hutu was small
Alternatively, incorporating a temporal element, it may be the case that

Hypothesis 6c: A large number of potential victims increased the number of actual victims in the long term but in the short term it decreased them as it facilitated resistance and served as a deterrent to violations.

In many agriculturally based societies, individuals frequently have plots of land that are some distance from one another, individuals take their products to market which are often miles away, and when crops fail individuals have to move to more arable land. In Rwanda not only do these dynamics exist but there is also a history filled with internal and interstate conflicts that have led to numerous population movements within the country. As individuals tried to calculate whether or not to trust someone in their neighborhood, it should be expected that newcomers would have fewer qualms about killing individuals (if targeted) and that neighbors would have fewer qualms about killing newcomers (again, if targeted). As a result,

Hypothesis 6d: Violations were decreased when the number of relatively newcomers into a location was low.

Finally, we will consider the interactive influence of the strategic choices that individual and perpetrators made in their mutual attempts to avoid one another and to find one another respectively. In some areas (such as Bisesero [African Rights 1997]) individuals cooperated and engaged in overt activities and organized to defend themselves; in other areas (reminiscent of Schindler in the German case) individuals simply hid others or assisted them with escaping. Many of these efforts were successful for a time, but these activities tended to draw more resistors and survivors as well as greater numbers of (and better armed) perpetrators. As a result, we expect this would correlate with larger-scale violations. Accordingly, we expect that:

Hypothesis 6f: in the short term, prior resistance should decrease the frequency of violations, but in the long term they should increase their magnitude.

Additionally, some attention should be given to the fact that there was only a finite number of targets/victims that existed within each geographic locale. As a result,

Hypothesis 6g: Violations that involved large numbers of killings should reduce the subsequent number of violations that took place within the same locale.

Variable Operationalization

The data required to investigate the hypotheses identified above continue an effort already underway (www.genodynamics.com). In 2001, we began compiling information about violations at the district and prefecture level, from two reports: Leave None to Tell the Tale by Human Rights Watch (1999) – the international human rights organization, and Death, Despair and Defiance (1995) by African Rights – a human rights organization based in Rwanda. To date, we have coded information about what was done (the specific violation), who engaged in this activity (specifically what organizational affiliation they maintained), against what targets/individuals (Hutu, Tutsi, Twa as well as the political orientation and profession of the victim), at what location (cell, district, secteur and prefecture) and at what time (day). After several trips to Rwandan over the course of the next two years, we have also obtained an internal government report documenting the same activity from the Rwandan government (the Ministry of Youth, Culture and Sport). After translating the government documents from French into English and coding all three sources, we collected the data for our dependent variables. During this time, we also collected information relevant to several independent variables as well. With this grant application, we seek support for the reformatting of compiled information in order to facilitate the creation of numerous variables as well as support for the collection and entry of new information. Generally in order of the hypotheses identified, each will be discussed below. Following this, we will discuss the methodological technique that will be used and then offer a brief conclusion.

Violations and Resistance. To measure the frequency and magnitude of violations as well as identify perpetrator: everyday citizen, militia, police, army, Presidential Guard and reservist – and victim characteristics: Moderate Hutu, Tutsi and Twa (a third but numerically small ethnic group), we utilize the database that was already constructed (see Appendix 2 for codebook). To measure the frequency and type of resistance we use the same database, which we have already complied, which includes the date, location, action and occasionally the number of victims and profession of the perpetrator. In an effort to gauge the validity of the data on violations and resistance already compiled, we will compare data from three additional sources that were directed toward the comprehensive listing of events within a single location. These data come from three sources: 1) a Rwandan survivor’s association called Ibuka (which collected extraordinarily detailed information on the Kibuye prefecture), 2) African Rights (which compiled a listing of victims/activities in Gishyita and Gisovu communes - also in Kibuye), and 3) the international human rights organization Physicians for Human Rights (which compiled a listing of victims/activities in Kibuye).

Domestic Crisis (Civil War). For measures regarding RPF civil war activity, we rely upon the information provided within Kuperman (2001, 43), which presents a concise account of RPF troop movements between April and July. This prefecture-level information was derived from interviews, UN documents and archival research. Given the prefecture-level information, we develop a measure for the mean distance from RPF activity by district and prefecture, discussed below in the section entitled “Distance” and within Appendix 3. These measures are added to the database after they have been converted into district and prefecture-day format. We will also attempt to obtain more refined information, through additional interviews exploiting our existing contacts within the current government and through archival research at NUR.

Distance (State Penetration). Measures for proximity to Kigali (the number of miles from this city to all other points within the country), the mean distance from roads and the sheer amount of level terrain – by district and prefecture, will be obtained from the Center for Geographic Information Systems (GIS) at the National University of Rwanda in Butare (NUR). In 1992, the GIS center engaged in an effort to geocode various economic and demographic factors by longitude and latitude as well as place name. To accomplish this, satellite images were taken of the entire country and all information was digitized and placed into Arcview. In order to calculate the three measures identified above, we will build what is referred to as a “layer,” discussed briefly in Appendix 3. These measures are added to the database after they have been converted into district and prefecture-day format. To identify the number of public spaces by district and prefecture (i.e., locations of churches, schools, government buildings and markets), the data house at the GIS-NUR will again be relied upon supplemented with GIS location data collection on our part to account for currently missing data (the NUR data cover approximately 90% of the country). In order to use this information, we employ the same process of extraction utilized above.

Demographic Factors. As mentioned above, in 1991, a census was taken within Rwanda, which was obtained from the U.S. Census Bureau in Washington, D.C. “Population Data Systems” have frequently been use to identify and categorize individuals within a nation-state for the purposes of control and elimination and thus we view the existence of such a source as extremely important (Selzter 2001, 481, 483). Providing a rough ethnic map prior to the beginning of the genocide, within this census is information about overall population size by district, the number of Hutu, Tutsi and Twa by prefecture, “resident population” by prefecture (specifically how many individuals within a prefecture were born there or elsewhere, as well as which prefecture individuals had recently moved from), economic benefits (e.g., employment), as well as age and gender, by district. The measure of resident population is useful but not ideal. For example, it is not clear what year individuals moved to their current prefecture. This measure does provide us with some indicator of prefecture-level resident stability, which we estimate by comparing the census data to data generated by a random sample household survey conducted by the authors in July 2002. From available information, it is clear that the majority of residents do not move around (less than 750,000 comprise the mobile population, according to the 1991 census).

Political Factors. In order to develop measures of pre-genocide Hutu Power and militia presence, the orientation of officials regarding their affiliation with Hutu power or the moderate Hutu position and their withdrawal/elimination, pre-genocide violent activity directed against Tutsi and moderate Hutu as well as the existence of police stations and military bases by district, we will rely upon the archival records at NUR, the existing literature as well as interviews with Rwandan political authorities.

Geographic Factors. Indicators of international borders and regional designation (Center, North, West, South, East) were compiled from existing maps.

Methodological Technique

Within this study, there are three aspects of violations that we investigate as dependent variables: the type of violation (rape, beating, abductions, torture, killing, moderate-Hutu violations, Tutsi violations), the frequency of violations (an event count of discrete actions), and the lethality of violations (the number of individuals killed during each violation). There are three aspects of resistance that are considered: the type of resistance, the frequency of resistance activity and the number of individuals involved. Once all data is placed into the conventional space-time format (usually the nation-year but in this project the district and prefecture day), diverse methods will be used for model estimation.

The variables concerning frequency and the number of individuals involved will be examined with an event count model, specifically designed to address cross-sectional time-series. Given the data structure, this study initially relies upon a variant of negative binomial maximum-likelihood regression (Long 1997), which was specifically created in order to deal with panel data – in particular, the “xtnbreg” command in STATA (STATA 2001, 386-394). Research by Krain (1997), drawing on the work of King (1989), has identified that previous investigations of genocide (and repression/human rights violations for that matter) have not addressed carefully the problems of left censoring and over-dispersion that may lead to substantial bias in the parameter estimates. To address these issues, Negative Binomial regressions were deemed the more appropriate strategy of estimation. Poisson models are also available for the estimation of event counts, but this technique assumes that the events in question are independent and also that the data are no characterized by over-dispersion.

The variables concerning the type of behavior will be estimated with logistic regression, whereby each type will be estimated individually as a dichotomous measure. Again, a version of this estimation procedure has been specifically created within STATA in order to deal with panel data – the “xtlogit” command. This approach is taken because within each violation there may be multiple types of activities that were utilized (the types are not mutually exclusive). The model follows below.

Violations (Frequency, Deaths, Type), Resistance (Frequency, Involved, Type) =

Structure: Proximity to Kigali (H2a), Proximity to Road (H2b), Terrain Navigability (H2c), Number of Public Spaces (H2d), Next to Border (H2e), Sympathetic Official (H2f), Withdrawn Official (H2f), Safe Zones (H2g), Region Dummies (H3), Mobilization: Amount of Rwanda Under RPF Control (H1a), Proximity to RPF Territory (H1b), Presence of Hutu Power and Size of Hutu Power organization (H4a), Contiguous Presence of Militia and Size of Militia (H4c), Perpetrator Dummies (H4d),

Personal Survival: Economic Benefits (H5), Number of Tutsi and Number of Moderate Hutu (H6a), (Number of Tutsi and Number of Moderate Hutu)*Time (H6b), (Number of Tutsi and Number of Moderate Hutu)*Time2 (H6b), Number of New Residents (H6c), Original Locale of New Residents (H6d), Resistance to Violations (H6e), Resistance to Violations2 (H6e), Resistancet-1 (H6f), Violations t-1 (H6g), Cumulative Victims (H6h), Large Scale Killing (H6i), Age (Control), Gender (Control)