Leave none to claim the land

Abstract

More than 200 years after its first publication, the Malthusian thesis is still much debated, albeit in a modified form. Rather than predicting a global catastrophe, most neo-Malthusians stress the local character of the relationship between population pressure, natural resource scarcity, and conflict as well as its dependency on the socio-political and economic context. This softened version of Malthus’s thesis has received little empirical support in cross-country studies. In contrast, a number of subnational analyses have provided some evidence for local conditional Malthusian catastrophes, although ‘catastrophe’ is a big word since these studies have largely focused on low-intensity violence. This article adds to the small body of subnational studies, but focuses on a high-intensity conflict – the Rwandan genocide. In particular, it provides a meso-level analysis of the relation between population pressure and the intensity of violence measured by the death toll among the Tutsi across 1,294 small administrative units. The results indicate that the death toll was significantly higher in localities with both high population density and little opportunity for young men to acquire land. This finding can be interpreted as support for the neo-Malthusian thesis. On the other hand, it is possible that another mechanism operated – in densely populated areas, it may have been relatively easy for the elite to mobilize the population, because of dependency relations through the land and labor market. Alternatively, in densely populated areas, there may have been more lootable assets, and the violence may have been opportunistic rather than driven by need or by fear.

Keywords

genocide land scarcity neo-Malthusianism population Rwanda

Introduction

The power of population is so superior to the power of the earth to produce subsistence for man that premature death must in some shape or other visit the human race. The vices of mankind are active and able ministers of depopulation. They are the precursors in the great army of destruction, and often finish the dreadful work… (Malthus, 1798: 13)

The idea of Thomas Malthus that population is checked by famine, disease, or war in order to keep the per capita food production above the subsistence level was hugely influential and controversial even at the time of its first publication, receiving praise from John Stuart Mill and fierce criticism from David Ricardo, Karl Marx, and Friedrich Engels. By the end of the 19th century, when the Industrial Revolution and the transformation of agriculture that it had brought about allowed rapid population growth to go hand-in-hand with increases in food production per capita, the facts seemed to prove Malthus wrong and, for a while, he was perceived by many as a failed prophet of doom. However, for at least two reasons, Malthus’s Essay on the Principle of Population continues to fuel scholarly debate today.

First, the industrial revolution was largely based on cheap, readily available energy contained in coal and later in other fossil fuels. As these non-renewable resources are shrinking, the question emerges as to how the current food-production system can continue to satisfy growing demand. Moreover, the burning of fossil fuels has raised environmental concerns, with climate scientists warning of global warming and its adverse effects on food production. Joint with other forms of environmental degradation and the population explosion in the Third World, these observations have led to renewed claims about the end of prosperity by a number of neo-Malthusians, who predict that rapid population growth in combination with environmental degradation will lead to widespread hunger and armed conflict (e.g. Ehrlich, 1968; Kaplan, 1994). Much as in the case of Malthus, this doomsday scenario is heavily criticized, most prominently by the ‘resource optimists’ or ‘cornucopians’, who argue that resource scarcity is largely exaggerated and that, if it occurs, scarcity can be dealt with by market pricing and innovation (Simon, 1981; Lomborg, 2001).

Second, although the industrial revolution unleashed the greatest increase in food production ever seen, enabling the population to increase sevenfold since Malthus’s day, hunger, famine, and malnutrition are with us still. This observation has fed the idea that, rather than a global Malthusian catastrophe, the world may experience local catastrophes if local market and non-market institutions fail to enable households to meet their subsistence requirements and cannot resolve the tensions that may stem from this failure. The best-known proponent of this idea is Thomas Homer-Dixon, who qualifies the impact of scarcity on conflict by stressing its conditionality on the form of resource scarcity and the context in which it occurs: the developing versus the developed world, the level of ethnic or religious fractionalization in society, and the absence of institutions that provide a peaceful alternative for resolving the scarcity problem (Homer-Dixon, 1999).

Many efforts have been made to test this more moderate neo-Malthusian view empirically. Most of these efforts have relied on cross-country analysis and explained the onset or incidence of armed conflict in terms of country-level measures of population density, population growth, deforestation, soil degradation, and water scarcity (e.g. Collier & Hoeffler, 1998; Urdal, 2005; Theisen, 2008). Overall, these studies have provided very little evidence in support of a causal relationship between a decline in resource quantity or quality and violent conflict, even when including low-intensity conflicts in the sample (using a threshold of 25 battle-related deaths per year instead of the more conventional 1,000). The latter is done to account for the neo-Malthusian view that resource scarcity leads to low- rather than high-intensity internal conflicts (Homer-Dixon, 1999). A recent study by Esteban, Morelli & Rohner (2010) looks at the other extreme by focusing on mass killings using a minimum threshold of 50,000 fatalities. Analyzing country and ethnic-group-level panel data (1960–2007), the authors do find that, ceteris paribus, high population density is a significant predictor of mass killings.

The scarce empirical evidence for the neo-Malthusian hypothesis in large-N studies may be partly due to the inability of such studies to account fully for the local nature and conditionality of the tension between resource supply and demand. For example, national figures may hide local scarcities, and unobserved country characteristics can bias the results – for example, good institutions may foster both peace and population growth. Hence, scholars have started to focus on local-level analyses. Pioneering work by Thomas Homer-Dixon has taken the case-study approach in studying Pakistan, Mexico, Gaza, Rwanda, and South Africa. Overall, the conclusion on the basis of the case studies is that ‘environmental scarcity causes violent conflict’ (Homer-Dixon, 1994: 39). However, the case-study approach has been strongly criticized for sample selection bias and its inability to determine causal relations (Gleditsch, 1998; Gleditsch & Urdal, 2002).

This criticism has triggered more rigorous quantitative analyses. First, using geo-referenced time-series data, Raleigh & Urdal (2007) examine the relationship between resource scarcity and conflict at the level of geographical squares of 100 km×100 km and find that population growth and density are associated with increased risk of conflict, although the effects of land degradation and water scarcity are inconclusive. Second, Urdal (2008), looking at the variation of low-intensity conflict across 27 Indian states in the period from 1956 to 2002, finds a positive link between resource scarcity and conflict. Third, analyzing data for 25 Indonesian provinces for the period from 1990 to 2003, Østby et al. (2011) find that the risk of low-intensity conflict is higher in provinces where both population growth and inequality are high, while each of these factors is not significant in isolation. Finally, Bundervoet (2009) finds that communal land pressure significantly increased the probability of being killed in the 1993 wave of violence in Burundi, a country that is similar to Rwanda in a number of relevant characteristics, not least rapid population growth and extreme land scarcity.

The present article adds to this small body of subnational empirical studies by analyzing the relationship between population pressure and mass killings in the Rwandan genocide at the level of 1,294 small administrative units. It is similar to Bundervoet (2009) in its focus on mass killings in one single year rather than low-intensity conflict over a number of years. In addition, instead of explaining the onset or incidence of violence as in Raleigh & Urdal (2007), Urdal (2008), and Østby et al. (2011), it explains the variation in the intensity of violence, measured by the proportion of Tutsi killed across administrative units. The following section reviews the literature on the role of population pressure and land scarcity in the Rwandan genocide. Next, the research design is presented, including a description of the empirical framework and the data used. The final two sections present the results and conclude.

Land scarcity and genocide in Rwanda

Civil war broke out in Rwanda at the end of 1990 when the Rwandan Patriotic Front (RPF), a rebel army consisting mostly of Tutsi exiles, started launching attacks from Uganda. These initial attacks were followed by a two-year period of intermittent hostilities and negotiations between the government and the RPF, which eventually led to a ceasefire in July 1992 and a power-sharing agreement. However, on 6 April 1994, the plane carrying President Juvénal Habyarimana of Rwanda was shot down, whereupon Rwanda descended into chaos. Within hours, the Forces Armées Rwandaises (FAR), the Interahamwe militia,¹ administrators, and ordinary people started to kill Tutsi and moderate Hutu. Simultaneously with the onset of genocide, the civil war between the FAR and the RPF recommenced. In the areas where the RPF liberated the population from the genocidal regime, the RPF allegedly engaged in reprisal killings of Hutu. Late in June 1994, the RPF took power and the massive killings came to an end, but, until the late 1990s, insurgency and counterinsurgency operations by the FAR and the RPF, respectively, continued in the Northwest along the Congolese border.

Although the Rwandan genocide was concentrated in a relatively short period of time, April–June 1994, all the regions were affected and the death toll among Tutsi was staggeringly high, estimated at around 800,000 or approximately 75% of Rwanda’s Tutsi population (Prunier, 1995; Verpoorten, 2005). According to Eck & Hultman (2007), the Rwandan genocide is the largest example of one-sided violence in the post-Cold War era.² Many have tried to understand this massive violence from a political, social, anthropological, cultural or economic point of view (e.g. André & Platteau, 1998; Des Forges, 1999; Mamdani, 2001; Newbury, 1998; Olson, 1995; Prunier, 1995; Verwimp, 2005). None of these studies fail to mention that, in the years preceding the genocide, the Rwandan rural population was fighting an uphill battle against land scarcity and soil degradation.

In particular, on the eve of the genocide, Rwanda was Africa’s most densely populated non-island nation and, due to both high fertility rates and a young population, the annual population growth rate remained high at around 3%. Moreover, despite high population pressure, the Rwandan population had remained overwhelmingly rural and dependent on agriculture, with over 90% of the people relying on small-scale farming. For a number of decades, Rwandan peasants had responded to population pressure by expanding the area under cultivation and by intensifying agricultural production by means of evolving to a system of near-continuous cropping and mixed cropping, intense crop maintenance, and fertilization with manure to increase productivity. Until the mid-1980s, these responses enabled the food production to increase at the same rate as did the population (Olson, 1995).

However, by the end of the 1980s, almost all of the marginal land had been taken up for cultivation and many of the intensification techniques had reached their limits. For example, because of the lack of pasture land, it became increasingly difficult to keep livestock for manure (Clay, 1996). At the same time, no significant progress was made towards modern agricultural intensification, leaving the use of improved seeds and fertilizer in Rwanda well below that of the sub-Saharan average. Consequently, not only the average farm size but also farm productivity was drastically decreasing (Clay, 1996). Furthermore, the rural poor and landless had few opportunities to earn income outside agriculture, partly because the few jobs in the non-farm sector were largely monopolized by the rural elite through patronage relationships (André & Platteau, 1998). At the same time, falling coffee prices in combination with a liberalization policy as part of a Structural Adjustment Program added to the economic hardship.

The strong correlation between the timing of the genocide and the failure to maintain per capita food production in the face of population pressure has led scholars to hypothesize that land scarcity was an important ingredient in explaining the mass killings (e.g. Newbury, 1998; Mamdani, 2001; Prunier, 1995). This hypothesis is backed up by a large amount of anecdotal evidence. For example, the 595-page Human Rights Watch report Leave None to Tell the Story mentions the word ‘land’ 54 times, that is, on average almost once every 10 pages (Des Forges, 1999). The opportunity to acquire land emerges as an important incentive for the killers:

Authorities offered tangible incentives to participants. They delivered food, drink, and other intoxicants, parts of military uniforms and small payments in cash to hungry, jobless young men. They encouraged cultivators to pillage farm animals, crops, and such building materials as doors, windows and roofs. Even more important in this land-hungry society, they promised cultivators the fields left vacant by Tutsi victims. (Des Forges, 1999: 10–11)

Land also emerges as a very particular asset, not only because – unlike other lootable property – it is immobile and cannot be easily divided by the killers but also because of strong de facto inheritance rights and the absence of de jure private property rights, meaning that all the land reverted to the community and could be redistributed only after all its Tutsi ‘owners’ were killed or chased away:

As early as mid-April in some places, burgomasters ordered their subordinates to prepare inventories of the property of Tutsi who had been killed or driven away. One reason for the lists of people killed, initiated also at this time, was to identify which households were completely eliminated, meaning that their property was available for redistribution, and which had some survivors, meaning the land would be available only after further killing. (Des Forges, 1999: 299)

At first sight, the anecdotal evidence suggests that land scarcity played a significant role in accounting for the mass killings in Rwanda. However, there is an important competing hypothesis that is based on the drastic change in the political climate in Rwanda in the years preceding the genocide. By 1992, the war with the RPF had considerably weakened Habyarimana’s authority and, owing to pressure from both inside and outside Rwanda, he was forced to enter a coalition with domestic opposition parties and negotiate a peace settlement with the RPF. The Peace Agreement, signed in Arusha in August 1993, stripped many powers from the office of the president, transferring them to the transitional government in which opposition parties as well as the RPF received a large number of cabinet posts.

Percival & Homer-Dixon (1995) and Olson (1995), among others, argue that, with the prospect of having to share power, a part of the elite decided to take matters into their own hands and carefully planned both the genocide and politicide, that is, the killing of Tutsi, who were perceived as RPF sympathizers, as well as moderate Hutu and Hutu of opposition parties. Hence, although, in their account of events leading up to the genocide, Percival & Homer-Dixon (1995: 1) acknowledge that ‘environmental scarcity and population growth are critical issues in Rwanda’ and that ‘before the recent violence, they clearly threatened the welfare of the general population’, they argue that ‘many factors were operating in this conflict, and environmental and population pressure had at most a limited, aggravating role’. Instead, they attribute a central role to the insecurity of the regime and the elite generated by the civil war with the RPF and the Arusha peace accords.

I will comment in detail on this argument further on, but first I turn to two studies that provide micro-empirical analyses of the relationship between land scarcity and genocide in Rwanda. André & Platteau (1998) studied the impact of land scarcity on the land-tenure system and intracommunity tensions and disputes in Rwanda on the eve of the genocide. They rely on an in-depth study of a small community in Northwest Rwanda, where, both in 1988 and in 1993, detailed information was collected for 87 out of the 124 households living in the community. Among other things, the authors found evidence of a drastic decrease in per capita landholdings, increasing inequality in land endowments, a strikingly high number of land conflicts, and a lack of access to off-farm alternatives to help quasi-landless households make ends meet. In addition, it is demonstrated that young men belonging to the lower landownership classes were postponing marriage for lack of land. For instance, between 1988 and 1993, the share of women and men aged 20 to 25 still living with their parents increased from 39% to 67% and from 71% to 100%, respectively. Based on the quantitative information as well as their day-to-day observations of life in the community, the authors argue that due to extreme land scarcity, intracommunity tensions – land disputes in particular – were rising and becoming increasingly difficult to settle.

After the genocide, these authors collected information about the whereabouts and experiences of the individuals in their sample. Of particular importance for the purpose of the present study is that the findings indicate that a disproportionately large number of the victims of the 1994 violence belonged to households with relatively large landholdings or to households that were involved in pre-1994 land conflicts. With respect to the profile of the perpetrators, the authors report that ‘the most violent people tend to be young and to come from poor, yet not the most extremely poor family backgrounds’. However, the community under study included only one Tutsi. André & Platteau (1998) considered this ethnic homogeneity to be an advantage for their analysis of the link between land scarcity and violence, since in the context of ethnic homogeneity they could focus on studying killings of Hutu by Hutu, allowing them to disregard ethnic hatred as a motivation for the killings.

Verwimp (2005) studied the profile of perpetrators in a sample of 402 Hutu adult males across three Rwandan provinces. He found that those men living in a household that rented a lot of land for cultivation relative to its own, as well as the relatively well-off with a high percentage of income earned off-farm, were disproportionately represented among the genocide perpetrators. However, neither the size of landholdings owned nor the soil quality significantly predicted the likelihood of being a genocide perpetrator, a result that is somewhat in line with the findings of André & Platteau (1998) that it was not the poorest who participated most in the violence. Interpreting his findings, Verwimp (2005) hypothesized that the relatively high participation of the local elite can be explained by their having had the most to defend, that is, their privileged economic and political position, certainly in view of the changing macro-political situation in the early 1990s. On the other hand, the quasi-landless peasants who depended on the land-rental and labor market had something to gain but also had to defend the little they had, which left little choice but to obey the local elite whom they depended on for land rental and wage work.

Among the studies discussed above, Percival & Homer-Dixon (1995) most explicitly downscaled the causal role of land in the Rwandan genocide. The authors made three arguments in support of their view: (1) there is no evidence for large popular participation in the genocide; (2) violence broke out first in and around Kigali City and in the northern region before spreading to the southern region, where the population density was highest; and (3) the planners of the genocide belonged to the economic and political elite who arguably suffered the least from the country’s scarce natural resource base.

Each of these arguments needs to be qualified. First, since the study of Percival & Homer-Dixon (1995), there has been mounting evidence of widespread popular participation in the genocide. For example, based on detailed fieldwork in five administrative communes and in-depth interviews with prisoners, Strauss (2004) estimated that there were 175,000 to 210,000 perpetrators, and, in 2000, the government held 109,499 detainees on genocide charges, while the number of accused persons not detained was 49,066 (Office of the Prosecutor, 2002). In 2005, the number of genocide suspects emerging from the information round of the transitional justice system (Gacaca) was as high as 510,000 suspects (Government of Rwanda, 2005).³

Second, one cannot draw conclusions based on a comparison between the northern and southern regions without controlling for factors other than population density. In particular, several political and historical factors can account for the commencement of the killings in and around Kigali City and the Northwest rather than the most densely populated South. Most importantly, the South was the site of the pre-colonial Tutsi kingdom that had ruled since the 15th century, whereas the North had remained dominated by Hutu kingdoms until the end of the 19th century (Newbury, 1998). Hence, the share of Tutsi in the southern provinces was relatively large, and the ethnic groups were interconnected through family and friendship relations, two factors that may explain why, despite higher population densities, resistance against mass killings was stronger in the southern provinces.

Third, even though there is by now widespread agreement that elite security in the face of the Arusha accords is likely to be the most important factor explaining the onset of violence (given ample evidence that the genocide was planned and orchestrated by the Hutu elite), elite insecurity can hardly account for the intensity of the violence, since the militia and FAR army could not openly have killed 800,000 people in barely three months without the active or passive support of a large part of the population. Furthermore, it can be argued that the civil war with the RPF and the subsequent peace agreement was a proximate rather than a root cause for elite insecurity. The RPF consisted mostly of Tutsi exiles, who fled previous waves of violence between 1959 and 1990,⁴ and several scholars stress that these previous violent encounters can be explained by the resentment of Hutu towards the concentration of wealth in the hands of Tutsi. For example, Tutsi kept complete control over land use and access to land and reserved large land areas as pasture. When they were driven away, the vast areas that had been pasture were converted to crops, an extensification process that contributed significantly to rising food production (Olson, 1994; Newbury, 1998).⁵

In sum, since 1994, there has been mounting evidence of widespread popular participation in the genocide in both the North and the South, and historical facts suggest that the origin of the RPF army is rooted in an interethnic power struggle over land and other resources. This suggests that, whereas the proximate cause for the onset of the genocide was elite insecurity, the latter may itself have been rooted in a struggle over scarce natural resources, and, when it comes to explaining the intensity of the violence, the root causes were likely to be ethnic polarization (manipulated and transformed into ethnic hatred by the extremist elite), the struggle over control of scarce resources, or both. It seems that the latter cannot be discarded as an important factor in view of the findings of the study of André & Platteau (1998) in an ethnically homogenous community. On the other hand, both in André & Platteau (1998) and Verwimp (2005), the profile of perpetrators does not coincide with that of the poorest or most land-deprived households. This suggests that the role of land in explaining the civilian participation in mass killings is complex and that the underlying mechanisms still remain largely uncovered.

Research design

From this review of the anecdotal evidence and previous studies, I infer that, both at the macro and the micro level, neither land scarcity nor elite insecurity was a sufficient factor in isolation. At the macro level, elite insecurity put the bandwagon in motion, channeling existing grievances towards Tutsi hatred and unleashing the underlying tension over access to land to add to the scale and intensity of the genocide. At the micro level, a large part of the local elite feared losing their privileged status and therefore actively supported the killings and could easily persuade those depending on them for access to land and wage work to join in.

If this assessment holds true, we should find that killings were more severe in areas where the underlying tension over land was highest and where many depended on the local elite for access to land. This cannot be directly tested because of the lack of detailed data on land conflicts, land tenure, and land inequality, but it is plausible to assume that these features were more prevalent in areas with high population density, high population growth, or both, as well as the areas where landlessness was prevalent (among the young generation). The testable hypotheses therefore are:

H1: The genocide intensity was greater in localities with higher population densities.

H2: The genocide intensity was greater in localities with higher population growth.

H3: The genocide intensity was greater in localities where a large proportion of young men were single (for lack of sufficient land to marry).

In 1994, localities with high population density were not necessarily those with high population growth. In fact, it is documented that population pressure in the most densely populated rural areas led to demographic responses of the affected population, including lower fertility rates and migration towards less densely populated areas (Olson, 1994). In addition, communities with similar degrees of population density and population growth may still differ with respect to landlessness among young men because of distributional issues. Hence, apart from including the explanatory variables of interest in isolation, they will be included jointly as well as in interaction.

Besides the decimation of Tutsi, other forms of violence took place in Rwanda. However, since lack of documentation on these forms of violence constrains their measurement, the empirical analysis in this article focuses on the killings of Tutsi, which may lead to an underestimation of the impact of population pressure on violence (provided that the genocide of Tutsi is positively correlated with other forms of violence, e.g. revenge killings, killings of moderate Hutu, etc.).

Empirical framework

Although small in geographical size, Rwanda has a very diverse ecosystem with many different micro-climates. In addition, in many areas, one finds hills with steep slopes where population density and intensified cultivation have led to severe erosion. Hence, one of the main challenges in the empirical framework is to control for subnational differences in climate, soil type, and soil degradation. In the main specification, I estimate the empirical relation between population pressure and mass killings at the level of the smallest codified administrative unit, that is, for 1,294 rural administrative sectors, which have a mean size of 13.5 km² and counted on average 4,824 inhabitants in 1991.⁶ In order to reduce the influence from unobserved heterogeneity in soil and climate, I add fixed effects for each of the 137 Rwandan rural communes included, which is one administrative unit above the sector, with a mean size of 135.5 km² and on average 48,681 inhabitants in 1991.

The estimated model can be written as follows:

g e n o c i d e_t o l l_{1994, i c} = α_{0} + α_{1} p r e s s u r e_{1991, i c} + c o n t r o l s_{i c} Π^{'} + η_{c} + ε_{i c}

with $g e n o c i d e_t o l l_{1994, i c}$ the death toll among Tutsi in sector i: 1…1294, located in commune c: 1…137; and $p r e s s u r e_{1991, i c}$ pre-genocide population density in sector i, and $c o n t r o l s_{i c}$ including the 1991 sector-level population size as well as the distance of a sector to the nearest provincial capital city and main road. The latter two variables are included to capture the variation in access to non-farm work, which is likely to be higher close to roads and urban centers, but it may also account for the genocidal campaign having been orchestrated by the central and local administration. Finally, $η_{c}$ are commune fixed effects and $ε_{i c}$ are idiosyncratic errors.

In the second specification, I replace the commune fixed effects by province fixed effects, $η_{p}$ , in order to run the test with $p r e s s u r e_{1991, i c}$ measured as commune-level population growth and marriage delay, two measures for which sector-level data are not available. The second model can be specified as follows:

g e n o c i d e_t o l l_{1994, i c} = β_{0} + β_{1} p r e s s u r e_{1991, i c} + c o n t r o l s'_{i c} Ω' + η_{p} + ε'_{i c}

with $p r e s s u r e_{1991, i c}$ now equaling the 1991 sector-level population density, the 1978–91 commune-level population growth, the 1991 commune-level proportion of young single men (aged 20–35), or a combination of these variables. To account for intercommune variation, $c o n t r o l s_{i c}^{'}$ now also includes the proportion of Tutsi in a commune as well as all or a subset of the following variables: (1) the 1991 commune-level infant mortality to control for baseline wealth, (2) the 1991 commune-level proportion of young men that have little education to further reduce heterogeneity and because a large pool of poorly educated young men has been associated with armed conflict in the literature (e.g. Collier & Hoeffler, 1998), and (3) the percentage of interethnic marriage in the population to account for the interconnectedness between Hutu and Tutsi.

The proportion of Tutsi in a commune and the percentage of interethnic marriage in the population also account for possible measurement error in the genocide death toll, which is explained below and in detail in the online appendix. However, these control variables are not crucial as the results remain similar when excluding them.

The dependent variable

The outcome variable $g e n o c i d e_t o l l_{1994, i c}$ proxies the death toll among Tutsi in a sector. It is calculated based on information about the pre-genocide Tutsi population, the surviving Tutsi population, and the natural death rate. Using this information, the sector-level estimate of the number of Tutsi killed in proportion to the 1994 Tutsi population can be specified as follows:

g e n o c i d e_t o l l_{1994, i c} = (1 - \frac{(g e n o c i d e_s u r v i v o u r s_{i 2005}) {(1 - d_{n})}^{- 11}}{T u t s i_p o p u l a t i o n_{i 1994}}) - d_{n}

with $g e n o c i d e_s u r v i v o r s_{i 2005}$ being the sector-level number of genocide survivors alive in 2005 taken from the Gacaca – the transitional justice system in Rwanda (Government of Rwanda, 2005); $d_{n}$ a national level estimate of the natural death rate between 1994 and 2005 based on the 2000 DHS (Timaeus & Jasseh, 2004); and $T u t s i_p o p u l a t i o n_{i 1994}$ an estimate of the pre-genocide sector-level Tutsi population derived from the 1991 population census (Government of Rwanda, 1991; Minnesota Population Center, 2010).⁷

The variable $g e n o c i d e_t o l l_{1994, i c}$ is subject to several sources of measurement error. In the online appendix, I give a detailed discussion of the possible causes and consequences of measurement error. In summary, the sources of error include the following: (i) $d_{n}$ is a national-level estimate of the natural death rate and may not be appropriate if there is large subnational variation in post-genocide death rates; (ii) $g e n o c i d e_s u r v i v o r s_{i 2005}$ may include Hutu widows who were married to Tutsi; and (iii) the accuracy of $T u t s i_p o p u l a t i o n_{i 1994}$ hinges on the reliability of the 1991 population census and the extent of unobserved within-commune variation in the population growth and the share of Tutsi in the population.

These sources of errors probably account for 96 anomalous cases (out of the 1,390) for which $g e n o c i d e_t o l l_{1994, i c}$ is negative. I removed these anomalous cases from the sample in the main results of this article, and I performed a series of robustness checks when including them (see below). A number of the control variables included in $c o n t r o l_{i c}^{'}$ attenuate the impact of possible measurement error. For example, the 1991 commune-level proportion of interethnic marriages is likely to be positively related to the number of Hutu counted as survivors; the commune-level infant mortality rate accounts for part of the subnational variation in natural death rates; and the sector-level distances to the nearest road and town may capture part of the within-commune-level variation in population growth.

The calculated genocide death toll averages 63.1% across the 1,294 rural sectors (and 56.0% when the 96 sectors with anomalous values are included). When weighted by the share of Tutsi in the population of each of the sectors, we find a nationwide death toll of approximately 66.6%. Furthermore, when taking into account the under-reporting of Tutsi in the 1991 population census and repeating the calculation with a higher estimate for $T u t s i_p o p u l a t i o n_{i 1994}$ , I find a nationwide death toll of 75.5%, which is very close to the estimates put forward by Prunier (1995) and Verpoorten (2005).⁸

Figure 1.

Kernel density of the genocide death toll

For the purpose of testing the above hypotheses, $g e n o c i d e_t o l l_{1994, i c}$ exhibits substantial variation across sectors. This is evident from the kernel density plot in Figure 1, the province level summary statistics in Table I, and the quintile map displayed in Figure 2. The map displays high values in many sectors in the North, where the genocidal regime had its largest support base and Tutsi were a small minority. In particular, high values can be detected in the northern and eastern areas where the Bahima and Bagogwe clans lived, which were pastoralists and usually identified as Tutsi.⁹ In addition, Table I and Figure 2 indicate high values for the genocide’s death toll in sectors of Butare, Gikongoro, and Kibuye, three provinces where the share of Tutsi in the population was historically high. In Gitarama, another province with a relatively high share of Tutsi in the population, we find very few sectors with a high death toll amongst Tutsi. This is in line with other evidence on the genocide’s death toll, which indicates that resistance against the genocide was strongest in Gitarama and an estimated 50% of Tutsi survived, whereas in the other provinces with a high share of Tutsi in the population, the reported survival rate was no higher than 25%.¹⁰

Table I.

The share of Tutsi in the population and the genocide death toll

	Commune level Tutsi population (1991, %)		Sector level death toll
Province	Uncorrected	Corrected ^a	Uncorrected	Corrected ^a
Butare	0.15	0.21	0.69	0.78
Byumba	0.01	0.02	0.68	0.77
Cyangugu	0.10	0.14	0.59	0.70
Gikongoro	0.14	0.19	0.71	0.79
Gisenyi	0.02	0.03	0.54	0.67
Gitarama	0.07	0.10	0.46	0.61
Kibungo	0.06	0.09	0.57	0.69
Kibuye	0.12	0.16	0.78	0.84
Rural Kigali	0.07	0.10	0.60	0.71
Ruhengeri	0.01	0.01	0.75	0.82
Total	0.08	0.11	0.63	0.73

N = 1,294 rural sectors included in the analysis; ^athe correction adjusts for the under-reporting of Tutsi in the 1991 population census by 40%.

Figure 2.

Sector-level quintiles of the genocide death toll (highest quintile in dark grey)

The explanatory variables

Population pressure is proxied on the basis of 1991 population density and 1978–91 population growth. The former is available at the sector level from the 1991 population census; the latter was calculated from the 1978 and 1991 population censuses and is only available at the commune level. There is no information available on land inequality at the level of these small administrative units. Instead, landlessness is proxied

Table II.

Summary statistics (N = 1,294 sectors, 137 communes)

Variable name	Description	Source	Mean	St.dev.	Min	Max
Genocide death toll	Sector level estimate of % of Tutsi killed in 1994	Gacaca records and 1991 population census	0.63	0.28	0.00	0.99
Population density, 1991	Sector level population density, inhabitants/km²	1991 population census	430	188	38	1,461
Population growth, 1978–91	Average annual commune level population growth	1978 and 1991 population census	0.029	0.012	0.000	0.080
Single young men, 1991	Commune level % of men aged 25–35 that are not yet married	1991 population census	0.26	0.10	0.07	0.49
Tutsi population, 1991	Commune level % of Tutsi in the total population	1991 population census	0.076	0.084	0.000	0.575
1991 population size	Sector level total population	1991 population census	4,824	2,265	836	32,686
Distance to town	Sector level distance to nearest provincial capital or Kigali City, km	Own calculation in GeoDa	26.7	17.4	1.1	101.3
Distance to main road	Sector level distance to nearest road (type 1 road), km	Own calculation in GeoDa	8.1	5.9	1.0	40.5
Infant mortality rate, 1991	Commune level infant mortality rate per thousands of children born alive	1991 population census	55.4	27.9	12.0	206.2
Low-educated young men, 1991	Commune level % of men aged 15–25 that have no education beyond primary schooling	1991 population census	0.87	0.04	0.76	0.94
Interethnic marriage (%)	Commune level % of interethnic marriage in couples with at least one Tutsi	1991 population census	0.50	0.20	0.00	1.00
Days under RPF control	Number of days that the sector was under RPF control during April–July 1994	Davenport & Stam (2009)	39.8	41.8	0.0	109.0

by the commune-level proportion of men aged 25–35 who are not married, which is taken from the 1991 population census.

Summary statistics for the explanatory variables are presented in Table II. The average 1991 population density in the 1,294 sectors included in the analysis is approximately 430 inhabitants/km² while annual population growth was about 2.9% during 1978–91, and 26% of the men aged 25–35 were still single. Figure 3 plots the kernel-density function of these three explanatory variables of interest and indicates considerable variation across sectors.

The three explanatory variables of interest are interrelated. As a result of demographic responses to population pressure, the relation between population density and population growth is negative. In particular, from a regression of population growth on population density, controlling for province fixed effects, it is found that an increase of population density by 100 inhabitants/km² is associated with a decrease in population growth by 0.19 percentage points, corresponding to 6.6% of the average of 2.9% population growth. In contrast, the relation between population density and unmarried men is positive: an increase of population density by 100 inhabitants/km² is associated with an increase in the proportion of single young men by 2.5 percentage points, which corresponds to a 10% increase of the average of 25%.

The sector-level control variables that measure the distance to nearest town and road are calculated using numerical maps in GeoDa and average 26.7 km and 8.1 km, respectively. The commune-level control variables, including measures for the share of Tutsi in the population, infant mortality, and schooling level, are calculated on the basis of the 1991 population census data. The sample average for the proportion of Tutsi is 7.6% (or 10.6% when correcting for under-reporting of Tutsi in the 1991 census). The infant mortality rate was on average 55 per thousand live births in 1991 and only 13% of men aged 15–25 had received some secondary education.

Figure 3.

Kernel densities of explanatory variables of interest

Results

Panel A of Table III presents the estimation results for Equation 1 (Models 1–4). In Model 1, including only 1991 sector-level population density and commune fixed effects as explanatory variables, the estimated coefficient α₁ is positive and strongly significant, suggesting that an increase of 100 inhabitants/km² adds 1.2 percentage points to the genocide’s death toll. When including controls for population size and distance to town and/or road in Models 2–4, the coefficient remains significant and the estimated impact varies between 1.1 and 1.6 percentage points. Regarding the control variables of distance to town and road, it is noteworthy that the results indicate that killings were more severe in remote areas, a finding that goes somewhat against the idea that the genocide was orchestrated top-down and may be explained by the absence of non-farm work to compensate for land scarcity in these localities. In Panel B of Table III, I repeat the estimates of Panel A, but account for spatial correlation in the error terms, using a distance-based spatial weighting matrix. The results remain qualitatively the same.

Table IV gives the results of several specifications of Equation 2 (Models 5–11), in which the commune fixed effects are replaced with province fixed effects and a control for the pre-genocide share of Tutsi in the population is added. In the first column (Model 5), population pressure is measured again by means of sector-level 1991 population density, and we find that the estimate for β₁ (0.015) is very similar to the estimates for α₁ in Model 1. In Models 6 and 7, 1991 population density is replaced by commune-level 1978–91 population growth and the proportion of men aged 25–35 who were still single, respectively. The estimated coefficients on these alternative measures are positive, but the former does not yield a significant effect, while the latter is only weakly significant, suggesting that a ten-percentage-point increase in the proportion of single young Hutu men increases the death toll by 2.4 percentage points.

In Models 8 and 9, population density is added as an explanatory variable jointly with one of the alternative measures of population pressure. Only the estimated impact of population density is significantly positive. In Models 10 and 11, the interaction terms between population density and population growth and the share of single young men, respectively, are included. Only the latter interaction term is significantly different from zero, while statistical significance on both of the individual variables fades and their sign even reverses. This suggests that it is not population density as such that intensified the killings, but rather the specific combination of high population pressure and a large proportion of single young men (~landless).

Robustness checks

In Table V, I present the same models as in Table IV but after adding three additional commune-level control

Table III.

Sector level population density and genocide intensity

	Dependent variable: the genocide death toll
	1	2	3	4
Panel A: OLS estimates
1991 population density^a	0.012**	0.011**	0.012**	0.016***
	(0.006)	(0.006)	(0.006)	(0.006)
1991 population size^b		0.016	0.015	0.004
		(0.025)	(0.025)	(0.025)
Distance to town^b			0.093***	0.064**
			(0.032)	(0.032)
Distance to main road^b				0.059***
				(0.020)
Commune fixed effects	Yes	Yes	Yes	Yes
Obs	1,294	1,294	1,294	1,294
R²	0.341	0.341	0.347	0.353
Panel B: spatial error regression models by maximum likelihood
1991 population density^a	0.013**	0.013**	0.013**	0.016***
	(0.005)	(0.005)	(0.005)	−0.005
1991 population size^b		0.015	0.014	0.004
		(0.025)	(0.025)	−0.022
Distance to town^b			0.092***	0.062**
			(0.03)	−0.031
Distance to main road^b				0.060***
				−0.021
Commune fixed effects	Yes	Yes	Yes	Yes
Spatial error	Yes	Yes	Yes	Yes
Obs	1,294	1,294	1,294	1,294

Robust standard error in parentheses; the spatial weights matrix used in the estimation of the spatial regression model is a 1,294 by 1,294 matrix taking values 1 for pairs of sectors that are less than 20 km apart and zero otherwise; significant at the ***1%, **5% and *10% level; ^aexpressed in 100 inhabitants/km²; ^blogged values.

variables: infant mortality, the proportion of young men without secondary schooling, and the proportion of interethnic marriages. The results remain qualitatively the same, except for Model 9, where the coefficient on the proportion of unmarried young men is now also significantly different from zero, be it only at the 10% level.

Table VI lists five robustness checks for Model 5. First, population density is replaced by its natural logarithm. Second, all 96 observations with anomalous values for the calculated death toll are included. Third, all these anomalous observations are included but censored to zero. Fourth, to account for possible measurement error in the genocide death toll stemming from the inclusion of close Hutu relatives of Tutsi victims among the genocide survivors, the share of interethnic marriages is added as a control variable. Fifth, to account for the possible impact of the RPF advancement which put an end to the genocide, I included a variable that takes on the number of days the sector was under RPF control in the 109-day period that the genocide took place.¹¹ Finally, Model 5 is estimated using the commune-level calculation of the genocide’s death toll instead of the sector-level calculation, to rule out the results being driven by within-commune measurement errors in $g e n o c i d e_t o l l_{1994, i c}$ . In all cases, the qualitative result that greater population density leads to a higher estimated death toll remains. The results for the proportion of unmarried young men are also robust for the same checks (not reported).

Table IV.

Sector level population density, commune level population growth, commune level single young men and genocide intensity

	Dependent variable: the genocide death toll
	5	6	7	8	9	10	11
1991 population density^a	0.015***			0.015***	0.014***	0.027**	−0.009
	(0.005)			(0.005)	(0.005)	(0.012)	(0.012)
1978–91 population growth		0.941		1.200		2.516*
		(0.834)		(0.832)		(1.453)
Single young men (%)			0.238*		0.195		−0.191
			(0.126)		(0.127)		(0.219)
(1991 population density*						−0.004
1978–91 population growth)						(0.004)
(1991 population density*							0.082**
Single young men (%))							(0.037)
Tutsi population (%)	0.634***	0.667***	0.549***	0.668***	0.559***	0.662***	0.587***
	(0.096)	(0.097)	(0.103)	(0.098)	(0.105)	(0.098)	(0.106)
1991 population size^b	−0.019	−0.013	−0.007	−0.026	−0.018	−0.027	−0.020
	(0.023)	(0.023)	(0.022)	(0.023)	(0.023)	(0.023)	(0.022)
Distance to town^b	0.030**	0.025*	0.036**	0.028*	0.038**	0.027*	0.033**
	(0.014)	(0.015)	(0.015)	(0.015)	(0.015)	(0.015)	(0.016)
Distance to main road^b	0.046***	0.032***	0.038***	0.044***	0.049***	0.045***	0.051***
	(0.012)	(0.012)	(0.012)	(0.012)	(0.013)	(0.012)	(0.013)
Province fixed effects	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Obs	1,294	1,294	1,294	1,294	1,294	1,294	1,294
R²	0.161	0.155	0.156	0.162	0.162	0.163	0.165

Robust standard error in parentheses; significant at the ***1%, **5% and *10% level; ^aexpressed in 100 inhabitants/km²; ^blogged values.

Conclusion

The debate over the Malthusian hypothesis has shifted from a global to a local level and centers around the question of what conditions really matter. In particular, the neo-Malthusian hypothesis states that, under certain conditions such as ethnic fractionalization and a political deadlock, resource scarcity may be conducive to violence.

In the case of Rwanda, one may argue that the perfect storm that could lead to an unfolding neo-Malthusian local disaster was present. First, since the mid-1980s, food production per capita started to decline as almost all marginal land was being taken under cultivation and as intensification of production had remained very limited. Second, the rural poor and landless had few opportunities to earn income outside of agriculture. Third, the Rwandan society had a history of violent encounters between the Hutu majority and the Tutsi minority. Fourth, in the years leading up to the genocide (1990–93), sporadic attacks from Tutsi exiles at the border with Uganda and the following power-sharing arrangement led to a tense political climate that divided the government party between radical and moderate Hutu.

The prospect of having to share power with the opposition parties and the RPF triggered the genocidal campaign of the Hutu extremists, who carefully planned, prepared, and organized the killings of both moderate Hutu and Tutsi. However, the elite’s genocidal plans could only succeed with a fair amount of support, active or passive, from the general population. Unfortunately, the Hutu radicals succeeded in obtaining such support by playing the ethnicity card and free-riding on the underlying tensions over the extreme scarcity of land and economic opportunities.

By how much did land intensify the genocide? Analyzing the genocide’s death toll across 1,294 administrative units, I found that an increase in pre-genocide population density by 100 inhabitants per km² leads to a rise in the genocide’s death toll by 1 to 2 percentage points. Furthermore, in areas where many young men remain single, which may arguably be an indication of landlessness, fewer Tutsi survived the genocide. The estimates indicate that a 10 percentage point increase in the share of young men that are still single increases the genocide’s death toll by approximately 2.5 percentage points. Importantly, the results of the empirical model

Table V.

The relation between sector level population density, commune level population growth, commune level single young men and genocide intensity; controlling for a large number of commune level characteristics

	Dependent variable: the genocide death toll
	5′	6′	7′	8′	9′	10′	11′
1991 population density^a	0.016***			0.016***	0.015***	0.028**	−0.010
	(0.005)			(0.005)	(0.005)	(0.012)	(0.012)
1978–1991 population growth		0.947		1.154		2.439*
		(0.837)		(0.834)		(1.449)
Single young men (%)			0.265**		0.238*		−0.195
			(0.129)		(0.130)		(0.220)
(1991 population density*						−0.004
1978–91 population growth)						(0.004)
(1991 population density*							0.095**
Single young men (%))							(0.038)
Tutsi population (%)	0.664***	0.685***	0.562***	0.691***	0.574***	0.683***	0.615***
	(0.112)	(0.111)	(0.117)	(0.113)	(0.119)	(0.113)	(0.122)
1991 population size^b	−0.022	−0.013	−0.008	−0.028	−0.022	−0.029	−0.025
	(0.023)	(0.023)	(0.023)	(0.023)	(0.023)	(0.023)	(0.022)
Distance to town^b	0.027*	0.024	0.035**	0.025*	0.035**	0.025*	0.029*
	(0.015)	(0.015)	(0.015)	(0.015)	(0.015)	(0.015)	(0.016)
Distance to main road^b	0.045***	0.031***	0.037***	0.043***	0.048***	0.044***	0.049***
	(0.013)	(0.012)	(0.012)	(0.013)	(0.013)	(0.013)	(0.013)
Infant mortality rate (%)	−0.354	−0.288	−0.312	−0.366	−0.381	−0.357	−0.404
	(0.303)	(0.304)	(0.307)	(0.302)	(0.304)	(0.303)	(0.305)
Poorly educated youg men (%)	0.237	0.071	0.184	0.199	0.304	0.202	0.417*
	(0.241)	(0.238)	(0.240)	(0.241)	(0.242)	(0.241)	(0.250)
Interethnic marriages (%)	0.001	0.018	0.002	0.003	−0.011	−0.001	−0.016
	(0.047)	(0.048)	(0.048)	(0.047)	(0.048)	(0.047)	(0.048)
Province fixed effects	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Constant	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Obs	1,294	1,294	1,294	1,294	1,294	1,294	1,294
R²	0.162	0.156	0.157	0.163	0.164	0.164	0.169

Robust standard error in parentheses; significant at the ***1%, **5% and *10% level; ^aexpressed in 100 inhabitants/km²; ^blogged values.

Table VI.

Robustness checks for Model 5

Dependent variable: the genocide death toll
	Logged population density	Anomalous values included	Anomalous values included, censored to zero	Controlling for the % of interethnic marriages	Controlling for the days sector was under RPF control	Using the commune level genocide death toll ^c
	5_rob1	5_rob2	5_rob3	5_rob4	5_rob5	5_rob6
log(1991 population density^a)	0.066***
	(0.021)
1991 population density^a		0.015**	0.016***	0.015***	0.015***	0.009*
		(0.006)	(0.005)	(0.005)	(0.005)	(0.005)
Interethnic marriages (%)				−0.003
				(0.047)
Days under RPF control					0.000
					(0.000)
Tutsi population (%)	0.628***	0.971***	0.839***	0.631***	0.637***	0.891***
	(0.096)	(0.126)	(0.110)	(0.102)	(0.096)	(0.083)
1991 population size^b	−0.018	0.052	0.003	−0.019	−0.021	0.028**
	(0.022)	(0.038)	(0.024)	(0.023)	(0.023)	(0.014)
Distance to town^b	0.029**	0.042**	0.035**	0.030**	0.030**	0.021**
	(0.014)	(0.020)	(0.016)	(0.014)	(0.014)	(0.010)
Distance to main road^b	0.047***	0.066***	0.051***	0.046***	0.046***	0.043***
	(0.012)	(0.018)	(0.013)	(0.012)	(0.012)	(0.009)
Province fixed effects	Yes	Yes	Yes	Yes	Yes	Yes
Constant	Yes	Yes	Yes	Yes	Yes	Yes
Obs	1,294	1,390	1,390	1,294	1,294	137
R²	0.160	0.103	0.140	0.161	0.161	0.398

Robust standard error in parentheses; significant at the ***1%, **5% and *10% level; ^aexpressed in 100 inhabitants/km²; ^blogged values; ^cthe data from 136 communes are weighted with frequency weights equal to number of sectors in each commune and the sector level explanatory variables are averaged at the commune level.

that includes the interaction term of population density and landlessness suggest that these factors did not operate in isolation, but it is their combination that contributed to the killings.

These findings do not indicate that extreme scarcity of land resources caused the onset of the genocidal campaign, but they clearly indicate that, once the bandwagon was set in motion, resources were a significant factor in explaining the intensity of the violence. This conclusion can be interpreted as support for the neo-Malthusian hypothesis, that is, that the violence resulted from a tension between the subsistence needs and scarce resources. On the other hand, rather than driven by need, the killers may have been driven by the opportunity to loot or by the fear to lose the little they had, two alternative motives that may have been especially strong in high-density areas with many landless men. Thus, whereas this meso-level analysis has provided empirical support for a local-level relation between violence and population density, detailed individual-level analysis, based on both quantitative and qualitative evidence, is required in order to unravel the exact mechanisms underlying this relationship.

Footnotes

Replication data

The regression analysis was executed in STATA 10.1. All information and files necessary for the replication of the results in this article can be found at .

Acknowledgments

An earlier version of this paper was presented at a LICOS/CRED seminar in Leuven and the CSAE conference in Oxford. I received helpful comments from three anonymous referees from this journal, and from Giacomo De Luca, Nils Petter Gleditsch, Romain Houssa, Pieter Serneels, and Henrik Urdal. I owe thanks to the Rwandan National Service of Gacaca Jurisdiction and the Rwandan National Census Service for making available the data used in this study.

Funding

While writing this article, I was funded by the Fund for Scientific Research – Flanders. All errors and opinions expressed remain my own.

Notes

References

André

Catherine

Platteau

Jean-Philippe

(1998) Land relations under unbearable stress: Rwanda caught in the Malthusian trap. Journal of Economic Behavior and Organization 34(1): 1–47.

Bundervoet

Tom

(2009) Livestock, land and political power: The 1993 killings in Burundi. Journal of Peace Research 46(3): 357–376.

Clay

Dan

(1996) Fighting an uphill battle: Population pressure and declining land productivity in Rwanda. International Development working paper 58. Michigan State University.

Collier

Paul

Hoeffler

Anke

(1998) On economic causes of civil war. Oxford Economic Papers 50(4): 563–573.

Davenport

Christian

Stam

Allan

(2009) Rwandan political violence in space and time. Unpublished manuscript.

Des Forges

Alison

(1999) Leave None To Tell the Story: Genocide in Rwanda. New York: Human Rights Watch.

Eck

Kristine

Hultman

Lisa

(2007) One-sided violence against civilians in war: Insights from new fatality data. Journal of Peace Research 44(2): 233–246.

Ehrlich

Paul

(1968) The Population Bomb. New York: Ballantine.

Esteban

Joan

Morelli

Massimo

Rohner

Dominic

(2010) Strategic mass killings. UFAE and IAE working papers 826.10. Unitat de Fonaments de l’Anàlisi Econòmica (UAB) and Institut d’Anàlisi Econòmica (CSIC).

10.

Gleditsch

Nils Petter

(1998) Armed conflict and the environment: A critique of the literature. Journal of Peace Research 35(3): 381–400.

11.

Gleditsch

Nils Petter

Urdal

Henrik

(2002) Ecoviolence? Links between population growth, environmental scarcity and violent conflict in Thomas Homer-Dixon’s work. Journal of International Affairs 56(1): 283–302.

12.

Government of Rwanda (1991) Recensement général de la population et de l’habitat Rwanda [General population and housing census]. Kigali: Service National de Recensement.

13.

Government of Rwanda (2005) Report on Data Collection: Annexes. Kigali: National Service of Gacaca Jurisdiction.

14.

Homer-Dixon

Thomas

(1994) Environmental scarcities and violent conflict: Evidence from cases. International Security 19(1): 5–40.

15.

Homer-Dixon

Thomas

(1999) Environment, Scarcity, and Violence. Princeton, NJ & Oxford: Princeton University Press.

16.

Kaplan

Robert

(1994) The coming anarchy. Atlantic Monthly 273(2): 44–76.

17.

Lomborg

Bjørn

(2001) The Skeptical Environmentalist: Measuring the Real State of the World. Cambridge: Cambridge University Press.

18.

Malthus

Thomas

(1798) An Essay on the Principle of Population. Oxford: Oxford University Press.

19.

Mamdani

Mahmood

(2001) When Victims Become Killers. Princeton, NJ: Princeton University Press.

20.

Minnesota Population Center (2010) Integrated Public Use Microdata Series, International: Version 6.0. Machine-readable database. Minneapolis, MI: University of Minnesota.

21.

Newbury

David

(1998) Understanding genocide. African Studies Review 41(1): 73–97.

22.

Office of the Prosecutor (2002) Abantu bafungiye mu magereza kasho na burigade [People detained in prison, solitary confinement or the police stations]. Technical report. Rwandan Ministry of Justice.

23.

Olson

Jennifer

(1994) Demographic responses to resource constraints in Rwanda. Rwanda Society-Environment Project working paper 7. Michigan State University.

24.

Olson

Jennifer

(1995) Behind the recent tragedy in Rwanda. GeoJournal 35(2): 217–222.

25.

Østby

Gudrun

Urdal

Henrik

Tadjoeddin

Mohammad

Murshed

S. Mansoob

Strand

Håvard

(2011) Population pressure, horizontal inequality and political violence: A disaggregated study of Indonesian provinces, 1990–2003. Journal of Development Studies 47(3): 377–398.

26.

Percival

Valerie

Homer-Dixon

Thomas

(1995) Environmental scarcity and violent conflict: The case of Rwanda. Project on Environment, Population and Security (EPS) (http://www.library.utoronto.ca/pcs/eps/rwanda/rwanda1.htm).

27.

Prunier

Gérard

(1995) The Rwanda Crisis: History of a Genocide. London: Hurst.

28.

Raleigh

Clionadh

Urdal

Henrik

(2007) Climate change, environmental degradation and armed conflict. Political Geography 26(6): 674–694.

29.

Simon

Julian L

(1981) The Ultimate Resource. Princeton, NJ: Princeton University Press.

30.

Strauss

Scott

(2004) How many perpetrators were there in the Rwandan genocide? An estimate. Journal of Genocide Research 6(1): 85–98.

31.

Theisen

Ole Magnus

(2008) Blood and soil? Resource scarcity and internal armed conflict revisited. Journal of Peace Research 45(6): 801–818.

32.

Timaeus

Ian

Jasseh

Momodou

(2004) Adult mortality in sub-Saharan Africa: Evidence from demographic and health surveys. Demography 41(4): 757–772.

33.

Urdal

Henrik

(2005) People vs. Malthus: Population pressure, environmental degradation, and armed conflict revisited. Journal of Peace Research 42(4): 417–434.

34.

Urdal

Henrik

(2008) Population, resources, and political violence: A subnational study of India, 1956–2002. Journal of Conflict Resolution 52(4): 590–617.

35.

Verpoorten

Marijke

(2005) The death toll of the Rwandan genocide: A detailed analysis for Gikongoro Province. Population 60(4): 331–368.

36.

Verwimp

Philip

(2003) Testing the double-genocide thesis for central and southern Rwanda. Journal of Conflict Resolution 47(4): 423–442.

37.

Verwimp

Philip

(2005) An economic profile of peasant perpetrators of genocide: Micro-level evidence from Rwanda. Journal of Development Economics 77(2): 297 –323.

38.

Verwimp

Philip

(2006) Machetes and firearms: The organization of massacres in Rwanda. Journal of Peace Research 43(1): 5–22.

39.

Watson

Catherine

(1991) Exile from Rwanda: Background to an invasion. Technical report. Washington, DC: US Committee for Refugees.

40.

Yanagizawa

David

(2010) Propaganda and conflict: Theory and evidence from the Rwandan genocide. Unpublished manuscript (http://www.hks.harvard.edu/fs/dyanagi/Research/RwandaDYD.pdf).

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB