Fiche du document numéro 25270

Num
25270
Date
Thursday December 15, 2005
Amj
Auteur
Fichier
Taille
254222
Pages
20
Titre
Machetes and Firearms: The Organization of Massacres in Rwanda
Lieu cité
Mot-clé
Cote
Journal of Peace Research 2006 43: 5
Source
Type
Article de revue
Langue
EN
Citation
See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/240704714

Machetes and Firearms: The Organization of Massacres in
Rwanda
Article in Journal of Peace Research · January 2006
DOI: 10.1177/0022343306059576

CITATIONS

READS

48

240

1 author:
Philip Verwimp
Université Libre de Bruxelles
83 PUBLICATIONS 1,685 CITATIONS
SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Conflict and Coffee Production View project

All content following this page was uploaded by Philip Verwimp on 02 September 2014.
The user has requested enhancement of the downloaded file.

Journal of
Peace Research
http://jpr.sagepub.com/

Machetes and Firearms: The Organization of Massacres in Rwanda
Philip Verwimp
Journal of Peace Research 2006 43: 5
DOI: 10.1177/0022343306059576
The online version of this article can be found at:
http://jpr.sagepub.com/content/43/1/5

Published by:
http://www.sagepublications.com

On behalf of:
Peace Research Institute Oslo

Journal of Peace Research Replication Data

Additional services and information for Journal of Peace Research can be found at:
Email Alerts: http://jpr.sagepub.com/cgi/alerts
Subscriptions: http://jpr.sagepub.com/subscriptions
Reprints: http://www.sagepub.com/journalsReprints.nav
Permissions: http://www.sagepub.com/journalsPermissions.nav
Citations: http://jpr.sagepub.com/content/43/1/5.refs.html

>> Version of Record - Dec 15, 2005
What is This?

Downloaded from jpr.sagepub.com by guest on October 11, 2013

© 2006 Journal of Peace Research,
vol. 43, no. 1, 2006, pp. 5–22
Sage Publications (London, Thousand Oaks, CA
and New Delhi) http://jpr.sagepub.com
DOI 10.1177/0022343306059576

Machetes and Firearms: The Organization of
Massacres in Rwanda*
PHILIP VERWIMP
Institute of Social Studies, The Hague, and Households in Conflict Network
This article is a quantitative study of the use of machetes and firearms during the 1994 genocide in
Rwanda, Kibuye Prefecture. The machete is an agricultural tool owned by most Rwandan households
and is believed to have been the prime instrument of killing during the genocide. The article addresses
the question to what extent individual characteristics of victims (gender, age, occupation) and aspects of
the Rwandan genocide (location of atrocities, point in time during the genocide) determined the perpetrators’ use of modern rather than traditional weapons to kill individual victims. An original database
developed by the organization of the survivors of the genocide (IBUKA) is used. The data were collected
from 1996 to 1999 and contain information on the deaths of 59,050 victims. Logistical regression analysis is performed to explain the use of either a traditional weapon or a firearm to kill the victims. The
analysis shows that the probability of being killed with a firearm depended on the location where the
victim was killed (more particularly, on whether or not the victim was killed in a large-scale massacre);
on the commune of residence and the age of the victim; on the number of days after 6 April the victim
was killed; and on interaction effects between the latter two variables and the gender of the victim. The
importance of individual characteristics, location of atrocities and timing for the use of different kinds
of weapons adds to our understanding of the organized nature of the Rwandan genocide.

Introduction
In 1994, the popular press portrayed the
Rwandan genocide as a tribal war between
ethnic groups. Since then, scholarly research
has rejected this view (Des Forges, 1999;
Uvin, 1998; Prunier, 1995). Hatred between
conflicting parties is very often one of the
consequences of a conflict but rarely the
* The first version of this article was written when the
author was a research scholar from the Fund for Scientific
Research, Flanders, Belgium. The author owes a debt of
gratitude to the Fund for the grants that enabled two
research stays in Rwanda in 1999 and in 2000. The author
expresses his gratefulness to IBUKA for the permission to
use their data file (see http://www.hicn.org/data_resources.
html).The author is indebted to Nils Petter Gleditsch,
Glenn Martin and several anonymous referees for detailed
comments and many remarks and suggestions that have
improved the article. All remaining errors are the author’s
responsibility. Correspondence: p.verwimp@hicn.org.

main cause of such a conflict. Hatred or
grievance is not a necessary (and even less a
sufficient) condition for genocide or less
lethal ethnic conflicts to occur. Krain (1997)
found that ethnic fractionalization is uncorrelated with the onset of genocide or political mass murder. The ethnic label put on
many conflicts often masks the real reasons
behind those conflicts. Econometric analysis
of conflict data for many countries (Collier
et al., 2003; Easterly, 2001; de Soysa, 2002)
has demonstrated that income, export,
governance, institutions, population density
and the presence of mineral wealth significantly influence the probability of civil war
or genocide, and that the inclusion of these
variables weakens or eliminates the effect of
an ethnic variable.

5

6

jour nal of P E A C E R E S E A RC H

Comparatively little research, and very
little quantitative research, has been carried
out on the use of traditional weapons and
firearms. Notwithstanding the importance of
small arms during conflicts, systematic
studies of their use during intensive episodes
of violence are nearly absent (Small Arms
Survey, 2002: 155). Cross-national econometric studies rarely introduce variables capturing the presence of (small) arms among
the population or variables measuring the
quantity or quality of the firearms used by
the army or the rebel groups involved in the
conflict. Such studies often stress the curse of
mineral wealth, but they neglect the fact that
one does not kill a human being with a
gallon of oil or a sparkling diamond.
Without weapons, a conflict between
groups, countries or factions among the
population would not reach the toll on
human lives that we have seen in the recent
bloody conflicts in Sierra Leone, the Democratic Republic of Congo or Sudan, to name
just a few.
This article presents micro-level evidence
of the use of firearms and traditional
weapons during the genocide in Rwanda.
My intention is to contribute to the understanding of the importance of the use of
weapons in Rwanda and, more generally, the
importance of small arms and light weapons
in other kinds of conflicts that target or
include civilians, both as victims and
assailants. The article offers a quantitative
study of the use of weapons during the
genocide in Kibuye Prefecture based on
large-scale data collection in Rwanda. The
article addresses the question to what extent
individual characteristics of victims (gender,
age, occupation) and aspects of the Rwandan
genocide (location of atrocities, point in time
during the genocide) determined the perpetrators’ use of modern rather than traditional
weapons to kill individual victims. If individual characteristics, location of atrocities
and timing have an impact on the use of

volume 43 / number 1 / january 2006

different kinds of weapons, this will add to
our understanding of the organized nature of
the Rwandan genocide.

Small Arms in Rwanda
In January 1994 (three months before the
genocide), Human Rights Watch presented
evidence that the government was buying
weapons that would be paid for partly in
cash and partly with the future harvest of the
Mulindi tea plantation (Human Rights
Watch, 1994: 14–18). The report states that
the regime was distributing weapons among
the population, thereby using the Rwandan
administrative organization as part of a socalled civilian self-defence programme. For
instance, in August 1991, Colonel Nsabimana, chief of staff of the Rwandan army,
proposed to provide a gun for every administrative unit of ten households: ‘at least one
person per Nyumba Kumi should be armed’
(unit of ten households; Human Rights
Watch, 1994: 27). Human Rights Watch has
documented how, in 1992–93, burgomasters
(the head of the communal authority)
ordered quantities of arms and ammunition
that far exceeded the needs of their local
police forces (Des Forges, 1999: 97–99).
They ordered guns, Kalashnikovs, machine
guns, grenades and large quantities of
ammunition. The report also documents the
purchase of arms by the rebels, the Rwandan
Patriotic Front (RPF). Goose & Smith
(1994: 86–96) describe and criticize the
arms sales to Rwanda in detail.
According to a private source, three boxes
of machetes were shipped from Belgium to
Rwanda with the machetes hidden between
sheets.1 From the corporate sector, there is
plenty of evidence that wealthy businessmen
as well as the directors of state-owned companies used the resources at their disposal to
1

Personal communication with the author, Kigali, August
2000.

Philip Ver wimp

import, transport and distribute firearms as
well as traditional weapons (Guichaoua,
2002).
A number of other sources inform about
the presence of arms in urban or rural areas.
A man identified only as Jean-Pierre documented the existence of dozens of hidden
stocks of firearms in Kigali in January 1994
for the head of the Belgian Peacekeeping
Force (Marchal, 2001: 133–140). Des
Forges (1999: 106–108, 140) used Colonel
Bagosora’s diary to provide more details
about the civilian self-defence programme
and the distribution of weapons. Bagosora
lists how many recruits should be trained,
what kind of weapons they should receive
and how recruits should be trained to use
them. Realizing that the supply of firearms
was limited, Colonel Nsabimana proposed
that the civilian population should be
instructed in the use of machetes, spears,
swords, bows and arrows. Nsabimana also
distributed a memorandum, dated 21 September 1992, in which he defined as enemies
the Tutsi and their accomplices, more particularly all members of the RPF, the Tutsi
inside the country and Hutu opponents of
the regime (Des Forges, 1999: 62–63,
99–100). The military authorities at the
national as well as the local level prepared
themselves to fight an enemy not at the war
front, but dispersed among the population.
Firearms were handed out to communal

MASSACRES

IN

R WA N D A

councillors, soldiers, army reservists and
trained militia (Des Forges, 2003: 23). This
policy was clearly meant to militarize the
rural areas and to draw the rural population
into the conflict, however far they lived from
the frontline. Ordinary farmers were told to
participate in the self-defence programme
with their own farming tools. Table I shows
that 83% of rural households owned one or
more machetes at the time of the National
Agricultural Household Survey (1984).
Almost all surveyed households owned a hoe
or a hack. In a radio address, four days after
the February 1993 attack by the RPF, President Habyarimana advocated a self-defence
force armed with traditional weapons, an
idea he repeated in a speech to army commanders on 13 March when he called for the
population to ‘organise to defend itself ’ (Des
Forges, 2003: 23–24).
From our knowledge of the distribution
of arms to the civilian population (mostly
machetes) and to the local police forces
(mostly firearms), we can derive a general
hypothesis on the objectives of the elite: The
objective of the regime was to kill as many Tutsi
as possible under the constraint that firearms
and bullets were in short supply. We thus
expect that firearms, grenades and bullets
were not used at random but in a targeted
and efficient way. We expect a high number
of victims killed by firearms and grenades in
places where Tutsi sought refuge or were told

Table I. Share of Rural Households Owning a Hoe (Ifuni), a Hack (Isuka), a Machete or an Axe, 1984 (%)
Number owned

Hoe

Hack

Machete

Axe

0
1
2
3
4
5+

8.7
16.5
30.1
18.8
9.5
10.0

1.0
17.0
47.1
21.3
8.5
5.0

17.3
64.1
16.3
1.9
0.2
0.2

41.9
52.9
4.9
0.3
0.0
0.0

N = 2,081 rural households.
Source: National Agricultural Household Survey (1984).

7

8

jour nal of P E A C E R E S E A RC H

to go, such as schools, churches and sports
stadiums. In these places, firearms and
grenades could be used in a very efficient way
to kill large numbers of Tutsi. Similarly, we
expect to see a high percentage of killings
with machetes and other traditional weapons
in individual attacks outside large-scale massacres. The latter took place most often ten
days to two weeks into the genocide. If we
observe such patterns – the topic of the
empirical tests in this article – then we may
have added some insight into the organized
nature of the genocide.
From the literature on armed conflicts,
one expects that age and gender matter in the
odds of dying by firearms. Young men are
most often not only the main perpetrators of
murder but also the main victims. Is this also
the case in the Rwandan genocide? Perpetrators of genocide often first target specific
individuals who may be able to resist genocidal policy, within the ranks of the perpetrating group itself or among the victimized
group. We will be able to test this statistically
by exploiting one of the few observable
differences in rural Rwanda, having or not
having a job outside agriculture. We expect
the former group to suffer more from genocidal killing because holders of off-farm jobs
in rural communities have high social status
(De Lame, 1996).
From this, I derive five hypotheses:
H1: Younger adults were more likely to be
killed by modern weapons than
children and the elderly.
H2: Men were more likely than women to
be killed by modern weapons.
H3: Persons employed outside the agricultural sector were more likely to be
killed by modern weapons.
H4: The risk of being killed by a modern
weapon was greatest some weeks into
the genocide.

volume 43 / number 1 / january 2006

H5: The risk of being killed by a modern
weapon was greatest in a large-scale
massacre.
To see the non-trivial nature of these
hypotheses, it may be useful to discuss the
counterfactual: If the genocide had not been
organized centrally, would one observe the
same pattern of the use of firearms? This
would mean that killers who happened to
own a firearm spontaneously killed Tutsi
along these same patterns. In order to see
that the counterfactual is not a plausible
alternative hypothesis, one has to understand
that ordinary Rwandans do not own
firearms. Neither do they spontaneously
wield weapons against their neighbours. In
order to overcome the collective action
problem, the local and national organizers of
the genocide gave money and promised jobs
and parcels of land to potential participants
in the killing operations. Participants were
also allowed and encouraged to rape Tutsi
women and to take them as concubines, a
kind of reward for ‘patriotic’ behaviour
(Verwimp, 2003).
Recall: the scale of the genocide; the speed
of the genocide; the gathering and guarding
of Tutsi in larger compounds such as schools,
churches and stadiums;2 the training of
Interahamwe and their presence in many
localities; the use of radio propaganda from
the capital Kigali and the broadcasting of
incitements to kill that went completely
unpunished.3 This does not mean there were
no local initiatives, but it offers strong indications of central command and central
organization. As it happens, local initiatives
could take place only with central approval,
either implicit or explicit.
2

For an analysis of the speed of the genocide, the death
toll and the determinants of survival, see Verwimp (2004).
3 The 1991 population census showed that 58% of urban
households and 27% of rural households in Rwanda
owned a radio. In 1978, radios were found in only 18% of
the households.

Philip Ver wimp

A Brief Political History
The formation of the Rwandan state was the
result of the century-long expansion of the
central territory (ancient Rwanda) in which
adjoining territories were put under the
control of the King of Rwanda. This process
took place in the 18th and 19th centuries,
particularly under the reign of King Rwabugiri. The king’s army consisted of warriors
(Intori) equipped with long spears. The
central state was characterized by a high
degree of organization in which the king and
his advisers decided on all important
matters. This inner circle of power was
always made up of a small group of Tutsi,
originating from two clans. The large
majority of Tutsi and Hutu had no access to
power or privilege. The extent to which the
two groups differed in their main economic
activity is contested in the literature, but
most scholars agree that before colonization
people involved in cattle breeding were considered Tutsi, whereas cultivators were predominantly considered Hutu. A significant
part of the land was reserved for pastures
(Ibikingi).
The advent of colonialism (first by
Germany, then by Belgium) brought farreaching change to the country. The
colonizers observed the socio-political composition of the elite and the peasantry and
concluded that the Tutsi were a different
race. Attracted by their high stature, facial
characteristics and leading position in
society, the colonizer (church and state)
concluded that the Tutsi originated from
northern Africa and that they were related to
the Caucasian race, thereby genetically predestined to rule. The Hutu, on the other
hand, were considered Bantu people, a black
race, predestined to be ruled.
From 1959 to 1962, a Hutu-led revolution took political power out of the hands of
the ruling Tutsi elite. Not only the elite, but

MASSACRES

IN

R WA N D A

also thousands of Tutsi civilians were driven
out of their homes and had to take refuge in
neighbouring countries. Grégoire Kayibanda, a Hutu educated in missionary
schools, became president and set up the
First Republic. Following the revolution, the
percentage of Tutsi in the Rwandan population declined sharply. Said to be 17.5% of
the population in 1952, Tutsi were counted
as 8.4% in 1991 (Des Forges, 1999: 40).
Habyarimana, defence minister in the
Kayibanda government, took control in a
coup d’état in 1973 that removed President
Kayibanda from power. The main reason for
this coup supposedly was that the Kayibanda
regime favoured Hutu from Gitarama and
other prefectures in the south.4 While the
landed interest of the northern elite
(Abakonde) was preserved by the Hutu revolution, they were not given access to lucrative business opportunities and political
power by the Kayibanda regime.
From 1974 to 1976, Habyarimana consolidated his political power. He outlawed
political parties and created his own Revolutionary Movement for Development
(MRND). According to Prunier (1995), the
MRND was a truly totalitarian party: every
Rwandan had to be a member of the MRND,
and all burgomasters and préfets were chosen
from among party cadres. Habyarimana
institutionalized Umuganda, compulsory
communal labour, and had peasants participate in village animation sessions to honour
him. He starved or executed 56 businessmen
and politicians closely related to the Kayibanda regime. All citizens were under tight
administrative control. Every five years, the
president was re-elected with 99% of the
vote.
In October 1990, a group of about 7,000
Tutsi rebels, former refugees and their sons
4

In a 1980 interview, Habyarimana mentions the ethnic
problem as one of the reasons for his coup d’état (cited by
Sato, 1980: 238; see also Verwimp, 2003).

9

10

jour nal of P E A C E R E S E A RC H

attacked Rwanda from Uganda. They had
brought equipment and arms from the
Ugandan military. The following years were
marked by a low-intensity civil war and
ongoing peace negotiations. In 1993, a peace
agreement was reached in Arusha whereby
political power would be divided between
the rebels and the government. Economic
decline, political manipulation of ethnic animosities and civil war all contributed to the
disintegration of Rwandan society in the
1990–93 period. Christophe Mfizi, close
supporter of the president, broke with the
MRND in 1992, after discovering statesponsored massacres of Tutsi in several
locations in northern Rwanda. He claimed
that a group called the ‘zero network’ had
penetrated the highest levels of government
(Mfizi, 1992).
When, on 6 April 1994, President Habyarimana came back from a meeting in
Arusha, his plane was shot down by two
missiles over Kigali airport. Hutu extremists
in his regime used the death of the president
to turn Rwanda into hell on earth: in only
100 days, more than half a million (500,000
to 800,000) Tutsi and Hutu opponents of
the regime were killed. A genocide, prepared
by the network around Habyarimana, was
executed with the participation of thousands
of ordinary people.

Descriptive Statistics
Method Used by IBUKA
The organization of the survivors of the
genocide, named IBUKA, has undertaken a
large research project with the main objective
of finding all the names of the victims of the
genocide in Kibuye Prefecture. They proceeded alongside the administrative organization of Rwandan society. Kibuye Prefecture
is divided into nine communes. Each
commune, with on average 50,000 inhabitants, is subdivided into several sectors.
These sectors, in turn, consist of several cells.

volume 43 / number 1 / january 2006

Commune by commune, sector by sector
and cell by cell, IBUKA collaborators went
into all families of surviving Tutsi and Hutu
to find the names of the murdered Tutsi. The
project was financed by the Dutch Embassy
in Rwanda and employed about two
hundred enumerators. The enumerators
came from or were familiar with the
commune where they were doing the interviews. The majority of the enumerators had
high-school training behind them. It was not
easy for IBUKA to find experienced enumerators, since almost all educated Tutsi were
killed. This lack of experience, together with
IBUKA’s decision to use only survivors as
enumerators, had a negative effect on the
quality of the data-collection process in some
communes. A supervisor for each commune
monitored the work of about 20 enumerators, at least one enumerator per sector. The
enumerators and supervisors did not receive
any statistical or interview training. The
result is a dictionary with the names of
almost 60,000 victims of genocide in the
prefecture, published in December 1999.5
Whenever possible, the project also registered the age and the profession of the
victim, the place where the person was killed
and the weapon used. The present article
uses the IBUKA data file to analyse the statistics of genocide in Kibuye Prefecture. The
data file was created not for statistical
purposes, but for the documentation of
genocide victims in Kibuye. This remains its
main value. Given the death toll among
Tutsi, the very difficult living conditions
after the genocide, and the lack of training
and adequate research facilities, the result of
the project is all the more remarkable.
Using the data for statistical purposes is
problematic, since the author did not take
5

The assignment of one enumerator to one sector prevented the duplication of records by different enumerators.
Furthermore, to facilitate registration and monitoring and
to avoid duplication, a number was given to each surveyed
household. However, I have not checked the data specifically for duplication. See IBUKA (1999).

Philip Ver wimp

part in the data-collection process, and the
quality of the data differs substantially
between communes and sectors. I have interviewed a number of the data collectors and
their supervisors and used statistical methods
to correct for missing data. From my interviews, I conclude that most of the respondents were Hutu, who have no incentive to
inflate the victim figures. I also had the
impression that Hutu who did not participate in the genocide felt a need to come
forward with accurate information to clear
themselves of guilt. This is important if one
wants to avoid identifying a whole ethnic
group with the perpetrator image. Some of
the respondents identified more with the survivors of the genocide than with the perpetrators. This is especially the case for Hutu
widows of Tutsi husbands. In the empirical
section of the article, several efforts were
made to test for robustness of the results.
The organizers of the data-collection
process also intended to register the cause of
death of each person. The different
categories used in the registration books
were: being Tutsi; being a Tutsi-friend;
having a Tutsi appearance; political opposition; and having a Tutsi mother. However,
this part of the data collection failed, in the
sense that IBUKA has registered only the
Tutsi victims of genocide. They either failed
to register, or registered only sporadically,
those persons who were killed for other
reasons than being Tutsi, such as Hutus
married to Tutsis. A presentation and an
analysis of the data for the mortality rate of
victims in different communes can be found
in Verwimp (2004). Another deficiency of
the data collection is the absence of rape as a
weapon used in the genocide. When a
woman died of AIDS after 1994 and the
disease was contracted because of rape in
1994, the woman is not registered as a
casualty in the data. We were not able to
investigate the resulting underestimation of
deaths from such factors.

MASSACRES

IN

R WA N D A

General Figures of Genocide in Kibuye
Based on the 1991 census and the figures
found by IBUKA, Table II shows that 12.4%
of the population of Kibuye Prefecture was
killed in the genocide, that is, approximately
83% of the Tutsi population. Table III
provides information on the genocide in
each of the communes of the prefecture.
Strong variation exists between the numbers
and the percentage killed in the communes.
Since I did not have access to the 1991 population figures according to ethnic affiliation
by commune, the exact number of Tutsi who
survived the genocide in each of the
communes could not be determined. Apart
from this, the IBUKA data file provides a lot
of other information that would normally
not be found in census data, such as the dates
and locations of the massacres and the
weapons used. It is, thus, a unique source of
information for the study of the genocide in
Kibuye Prefecture. Table IV is a good
example of one of the data problems. The
weapon that was used to kill the victim is
‘known’ in 92% of the cases. The date of
death, however, is known for 43% of the
victims. Later, we discuss the treatment of
missing values.6 Of all the tables in this
article, Table IV best documents the brutality of the genocide. Most victims were
hacked to death with traditional weapons
such as machetes or clubs. We do notice,
however, the importance of firearms, being
guns, rifles and grenades.7 From the 25,719
victims whose date of death is registered in
the data file, 20.2% were killed by a firearm.
In the subsequent analysis, ‘firearm’ will be
used as the umbrella term that includes guns,
rifles and grenades.
6

Missing values for the date of death do not necessarily
question data reliability in general. Four to five years after
the genocide, exact dates are more easily forgotten than the
place of death and the weapon involved.
7 Traditional arms are defined as all other instruments used
to kill people that are not guns, rifles or grenades. The definition of firearms corresponds with the internationally
used term ‘small arms and light weapons’. All other tools
used in the Rwandan genocide do not fit this term.

11

12

jour nal of P E A C E R E S E A RC H
Table II.

volume 43 / number 1 / january 2006

Genocide in Kibuye Prefecture
Number a

%

Population registered as Hutu
Population registered as Tutsi
Population registered as Twa
Foreign, other or undetermined
Murdered Tutsi found by IBUKA
Total population of the prefecture in 1991

399,470
71,225
1,490
1,735
59,050
473,920

84.3
15.0
0.3
0.3
12.4
100

Tutsi population registered as murdered
Tutsi population not registered as murdered

Number
59,050
12,175

% of Tutsib
82.9
17.1

a I have no access to exact figures of the population size in March 1994. Total population in the prefecture probably
reached 511,000 (≅ 473,920*(1.03)2.58) by March 1994 (two years and seven months after the 1991 census). The
prefecture represented 7% of Rwanda’s population (7.1 million).
b Accounting for population growth, the figures become 78% registered as murdered and 22% not registered.

Table III. Victims of Genocide by Commune of Residence a
Commune
Bwakira
Gishyita
Gisovu
Gitesi
Kivumu
Mabanza
Mwendo
Rutsiro
Rwamatamu
Totald

Number of inhabitants,
1991a

Number of victims
in data fileb

% of population
killed c

53,555
43,090
39,365
61,341
55,361
63,460
43,632
56,768
54,494

4,674
11,273
3,003
11,118
3,934
8,782
4,472
941
10,853

8.7
26.1
7.6
18.1
7.1
13.8
10.2
1.6
20.0

471,066

59,050

12.5

a

I did not have access to 1991 census data on ethnic affiliation per commune.
does not necessarily mean that these victims were killed in the commune of residence.
c To be decreased by 0.5% when population growth between 1991 and 1994 is taken into account.
d As stated before, the IBUKA project registered only Tutsi killed and not Hutu.
b This

Analysis
Figure 1 presents the share of victims in each
age group killed by a firearm. A clear pattern
emerges: few children and elderly people
were killed by firearms, while (relatively)
many young adults were. Of all victims in
their early 20s, 20–25% were killed by
firearms, whereas for victims in their late 50s,
the figure was 10%.

The Organization of Massacres
For about 40,000 victims (two-thirds of the
sample), I was able to find out whether or
not the victim died in a large-scale massacre.
A massacre is defined as an event where at
least 100 people were killed in one specific
location in less than three days. From the
data in the IBUKA file, 43.6% of the victims
whose location of death is registered were

Philip Ver wimp
Table IV.

MASSACRES

IN

R WA N D A

Killings by Type of Weapon (%)
Entire file

Dates of death known

Weapon

Number

%

Number

%

Machete
Club
Gun, rifle
Grenade
Drowned
Hoe, hack
Buried alive
Latrines
Spear
Burnt alive
Pick-axe
Stoned
Hanged
Sword
Starvation
Tractor
Other
Unknown or missing

31,117
9,779
8,706
1,058
847
444
442
437
421
401
337
131
100
79
23
12
636
4,020

52.8
16.6
14.7
1.8
1.4
0.8
0.7
0.7
0.7
0.7
0.6
0.2
0.2
0.1
0.0
0.0
1.1
6.8

13,272
4,238
4,575
609
486
328
340
150
209
226
192
84
35
50
15
7
197
659

51.6
16.5
17.8
2.4
1.9
1.3
1.3
0.6
0.8
0.9
0.7
0.3
0.1
0.2
0.1
0.0
0.8
2.6

Total

59,050

Figure 1.

100

25,719

100

Share of Tutsi Murdered with a Firearm: Realized Probability by Age

Pe r c e nt a g e m u rd e re d b y f i re a r m

40

30

20

10

0
0

10

20

30

40

50

60

70

80

Age

N = 25,719

killed in such massacres. Several thousand
entries had missing values at the location of
killing. In the further analysis, I will use a

dummy variable for death in a massacre (1 if
massacre, 0 if not) when it was known where
the victim was killed (see Appendix A).

13

14

jour nal of P E A C E R E S E A RC H

Occupation and Gender
Of the registered victims, 28,465 were
farmers, 1,949 had an occupation outside
farming, 13,955 were pupils and 8,925 were
below age 7 (see Table V). These figures
confirm that Rwanda in general and Kibuye
specifically is a very rural society. Only a
small minority of the working population
did not farm. This is basically true for Hutu
as well as for Tutsi. The former may have
been more present in public administration
and the latter more in commerce, but more
than 90% of the people of both groups were
farmers. Whereas in the past, Tutsi may have
been more involved in cattle breeding than
Hutu (see Newbury, 1988; De Lame, 1996),
in the early 1990s, there was no longer any
clear ethnic specialization in agriculture or
cattle breeding. Depending on wealth, Hutu
as well as Tutsi owned cattle and grew crops.
The IBUKA data file does not mention the
landholdings or the number of cattle of the
victims and makes a distinction only
between farming and non-farming.8 Table V
shows the occupation of the victims,
together with the weapons that were used to
kill them. In each category, for those whose
date of death is registered, a slightly higher
percentage of people killed by a firearm is
observed. I do not have an immediate explanation for this. Generally, non-farmers ran a
much higher risk than farmers of being killed
by a firearm.
More men than women were killed
during the genocide in Kibuye (30,528 men
and 28,471 women, plus 51 unknown
victims of genocide). As far as weapons are
concerned, the difference between men and
women is less than for age and occupation.
Men were only slightly more often killed by
firearms.
8

More research is needed to investigate in what respect
real or perceived ethnic labour specialization and inequalities in landholdings and cattle played a role in the
genocide. For existing sources, see Des Forges (1999), Uvin
(1998), André & Platteau (1998) and Verwimp (2003).

volume 43 / number 1 / january 2006

Regression Analysis
Specification of the Model In this section,
I test with logistic regression analysis whether
or not individual characteristics of a victim as
well as general aspects of the genocide determined the perpetrator’s use of a firearm rather
than a traditional weapon. In order to show
the robustness of the effects, several logit
regressions were performed for different
specifications of the model. In all regressions,
the dependent variable is a binary choice
variable having the value 1 for a firearm and
0 for all other arms. In Appendix C, I left out
date of death and other variables as regressors
in order to have as many cases as possible
included in the analysis. All variables, except
gender, show the same sign and the same level
of significance. The level of significance of the
gender variable changes over the different
regressions, but this is not surprising, given
that almost the same percentage of men and
women were killed by firearms. The inclusion
or exclusion of a small number of cases – as
a result of missing variables other than gender
in these cases – can then have an influence on
the significance level of the gender variable.
Interaction regressors of gender with
other variables show consistent and significant effects. Apart from the weapon used, the
age and the gender of victims, data collection
in two communes (Gitesi and Rutsiro) was
almost completely missing and unsatisfactory. In all regressions except Models 5 and
6, these two communes were dropped from
the analysis. Keeping the very few observations with complete data in these two
communes in the regression analysis would
misrepresent the effect of communal
dummies. A comparison of Model 6 (the two
communes included) with Model 7 (two
excluded) shows that this drop does not
affect the results.
Clustering on Massacre Sites Almost half
of all victims in Kibuye Prefecture were killed

Philip Ver wimp
Table V.

IN

R WA N D A

Occupations and Weapons Used
Entire file

Date of death known

Killed by
firearm

%

Number

Killed by
firearm

%

28,465
1,949
13,955
8,925
5,756

4,972
621
2,619
1,195

17.5
31.9
18.8
13.4

12,905
772
6,527
3,857

2,646
240
1,375
705

20.5
31.1
21.1
18.3

59,050

9,407

17.6

24,061

4,966

20.6

Occupation

Number

Farmers
Non-farmers
Pupils
Children < 7
Missinga
Valid total
a

MASSACRES

Occupation or weapon used or both are missing.

in large-scale massacres. A massacre was
defined as an event where at least 100 people
were killed in one specific location in less
than three days. This means that the killings
of people in these massacres and the weapons
used cannot be seen as independent cases in
our sample. Indeed, the use of weapons
depends, among other things, on the
location where the killings took place. In the
case of a grenade, for example, each use
probably killed several people. One way to
solve this is to use a robust estimation technique that clusters on the event of a
massacre. In the absence of clustering, the
standard errors are underestimated. This is
the case for Model 2. Clustering affects the
standard errors and variance–covariance
matrix of the estimators but not the estimated coefficients. Models 3 and 4 therefore
present clustered logit regressions, with the
event of a massacre as the cluster variable.
Models 3 and 4 are the regressions with the
most accurate estimation techniques. Unfortunately, I had to drop one more commune
from the analysis (Mwendo), since I was not
able to conclude from the data whether or
not a large number of victims originating
from this commune died in a large-scale
massacre. I included dummies for
communes, and Bwakira commune is used
as the baseline.

The following variables were included in
the analysis:
• weapon used (firearm: gun, rifle or
grenade = 1, other weapons = 0);
• age and age squared;
• gender (female = 1; male = 0);
• gender*age interaction;
• occupation dummy (off-farm work = 1,
on-farm work or pupil = 0);
• the number of days after 6 April the victim
was killed;
• number of days after 6 April squared;
• gender*number of days after 6 April interaction;
• gender*number of days after 6 April squared
interaction;
• massacre dummy (1 if the victim was
killed in a massacre, 0 if not); and
• dummies for communes.
The regression results presented in Tables VI
and VII show that the probability of the
victims in Kibuye Prefecture being killed by
a traditional weapon or a firearm depended
on the person’s age, his or her occupation,
the place of residence before the genocide,
the location where the killing took place, the
number of days after 6 April the person was
killed and interaction effects of these variables with the gender of the victim.

15

16

jour nal of P E A C E R E S E A RC H
Table VI.

volume 43 / number 1 / january 2006

Accounting for Type of Weapon Used

Variable
Individual level
Age
Age2
Gender
Gender*age
Off-farm
Days after April 6
Days after April 6 sq.
Gender*days after
Gender*days after sq.

Model 1

Model 2

.0329***
(9.86)
–.0004***
(–9.52)
–.1958*
(–1.66)
–.0072***
(–3.88)
.5454***
(5.69)
.1157***
(17.29)
–.0014***
(–12.96)
.060***
(4.64)
–.0017***
(–6.32)

.0302***
(8.31)
–.0003***
(–7.57)
.056
(0.46)
–.0063***
(–3.10)
.7065***
(6.70)
.0522***
(7.42)
–.0005***
(–5.15)
.018
(1.49)
–.0007**
(–3.02)

Commune and massacre dummies
Massacre
Gisovu
Gishyata
Kivumu
Mabanza
Mwendo
Rwamatamu
Constant
N

R2
Log likelihood

1.124***
(6.81)
1.1417***
(7.38)
1.5108***
(9.69)
3.152***
(20.94)
.5965***
(3.35)
2.398***
(16.21)
–4.851***
(–28.85)
23,650
not weighted
not clustered
0.11
–10,577.96

*** significant at the 1% level, ** significant at the 5% level, * significant at the 10% level.
z-values in parentheses.

2.191***
(44.23)
.6078***
(3.53)
.1408
(0.89)
.9590***
(6.01)
2.3064***
(14.98)

2.6472***
(17.42)
–5.060***
(–29.35)
21,536
not weighted
not clustered
0.20
–8,930.64

Philip Ver wimp
Table VII.

MASSACRES

IN

R WA N D A

Type of Weapon Used, with Clustering

Variables
Individual level
Age
Age2
Gender
Gender*age
Off-farm
Days after April 6
Days after April 6 sq.
Gender*days after
Gender*days after sq.
Commune and massacre dummies
Massacre
Gisovu
Gishyata
Kivumu
Mabanza

Model 3
.0302***
(4.40)
–.0003***
(–4.30)
.056
(0.22)
–.0063***
(–2.75)
.7065***
(5.88)
.0522*
(1.90)
–.0005
(–1.59)
.018
(0.63)
–.0007
(–1.09)

Marginal
effect
.0039
–.00005

–.0008
.1149
.0068
–.00007

2.191***
(8.34)
.6078
(0.86)
.1408
(0.19)
.9590
(1.48)
2.306**
(2.40)

.3337

2.647***
(4.09)
–5.060***
(–7.35)

.4122

.4517

Model 4
.0408***
(3.94)
–.0005***
(–3.97)
.079
(0.26)
–.0088***
(–3.06)
.8313***
(6.55)
.0564**
(2.48)
–.0006**
(–2.04)
.017
(0.50)
–.0006
(–0.90)

Marginal
effect
.0049
.00006

–.0010
.1314
.0068
–.00007

2.282***
(6.77)
.051
(0.09)
–.457
(–0.74)
.5705
(1.06)
2.307***
(3.35)

.3050

2.30***
(4.53)
–4.898***
(–9.14)

.3962

.4156

Mwendo
Rwamatamu
Constant
N

R2
Log likelihood

21,536
not weighted
clustered
0.20
–8,930.64

39,488
weighted(+)
clustered
0.28
–15,126.89

*** significant at the 1% level, ** significant at the 5% level, * significant at the 10% level.
(+) STATA 9.0 does not allow one to use non-integer weights. The weights used are therefore the nearest integer value of
the figures presented as weights in Appendix D.

17

18

jour nal of P E A C E R E S E A RC H

Weighting Procedure Because data on
victims from some communes or massacres
were more complete (i.e. data on all variables
are available for more cases) than from
others, the data have to be weighted in order
to correct for cases that had to be dropped.
The weighting procedure was as follows. We
first determined whether or not a person was
killed in a massacre. We then calculated the
total number of victims per massacre and the
number of victims with known dates of
death in these massacres. We assigned a
weight to each case representing the cases
without date known for each massacre.
Then, we calculated the number of people in
each commune who did not die in a massacre
and again assigned weights representing the
number of people without date of death in
each commune. This procedure yielded 18
different weights, one for each massacre and
for each commune. These weights are listed
in Appendix D.
Discussion In all of the models, the effects
of age and days-after variables are quadratic,
rather than linear. This means that the probability that the victim was killed with a
firearm initially increases with age and
decreases again at high age. This is also true
for the days after 6 April variable: the probability of being killed with a firearm increases
in the first weeks of April and decreases
towards the end of the genocide. The significance of the results on age and number of days
after 6 April tells us that young adults were
more likely to be killed with firearms
compared to children and old people and
that the use of firearms reached its most
intensive period several weeks into the
genocide.
Since the gender variable, by itself, is not
significant in the most accurate models,
women in general did not have a smaller
chance of being killed by a firearm. The
significance of gender reveals itself in combination with other variables. Both age and days

volume 43 / number 1 / january 2006

after 6 April effects interact with the gender
variable. In Model 1, the probability of a
woman being killed by a firearm, compared
to that of a man, decreases with age
(gender*age interaction negative and significant). For the days after 6 April variable, I
found an additional positive effect for women
only in Model 1 and Model 8 at the beginning of the genocide and a negative effect
towards the end of the genocide. The effects,
especially the magnitude of the squared
effects and the interaction variables, are small.
Once we introduce the massacre dummy, the
clustering and the weighting, these interaction effects lose their significance. This
means that, once we account for the organized nature of the genocide (as in Models 3
and 4), the gender*age effect remains, but the
gender*days after effect disappears. Thus,
older women did have a smaller risk of being
killed by a firearm, but no other gender
effects seem to emerge from the analysis. This
effect is linear. The introduction of an interaction term between gender and age squared
resulted in a non-significant effect.
The only variable capturing the socioeconomic situation of the victim is occupation. In the models, non-farmers have a
significantly higher probability of being
killed with a firearm compared to farmers
and pupils. Compared to a farmer, a person
with an off-farm job had a 13% higher probability of being killed by a firearm. This is a
strong marginal effect.
The effects of the massacre dummy and
the commune dummy variables, however, are
stronger than the individual effects, demonstrating that the location where the victim
was killed (in particular, whether or not the
victim was killed in a large-scale massacre),
as well as the commune of residence of the
victim, had a strong impact on the kind of
weapon (firearm versus other) that was used.
The probability of being killed by a firearm
was 30% higher in large-scale massacres and
40% higher for residents of Mabanza and

Philip Ver wimp

Rwamatamu communes compared to
Bwakira (the baseline). Firearms were used
more frequently in certain locations and
events, especially in large-scale massacres,
where many Tutsi were killed at the same
time and in the same place. This was the case
in the Gatwaro Football Stadium (with many
Tutsi from Mabanza commune) on 16, 17
and 18 April; in Rwamatamu commune on
12, 13 and 28 April; and in Biramba on 13
and 14 April, as well as in Bisesero and in
several other massacres. When the commune
dummy variables for Mabanza and Rwamatamu remain statistically significant and
positive after controlling for massacres and
clustering on massacres, this means that the
Tutsi residents of these communes had a
higher probability of being killed by a
firearm even when they were not killed in
one of the large-scale massacres.

Interpretation and Lessons Learned
The effect of the dummy for Mabanza on the
probability of being killed with a firearm was
stronger than for any other commune in
Kibuye Prefecture. Many victims from
Mabanza were killed in Gatwaro Stadium,
where the perpetrators of genocide fired with
machine guns at the crowd locked up inside
the stadium.9 The high prevalence of killings
with firearms in a commune may have
resulted from the size of the local stock of
firearms, the presence of army units in a
commune or the participation of militia
armed with firearms from other communes.
A reason for the high prevalence of killings
9

There are not many sources on this event, but an observation from Lieutenant Colonel Stabenreth, a French
officer, does give some indications: ‘From his investigations, he established that the Tutsi refugees who had
sought shelter at the stadium had been attacked by soldiers
and militia who had shot until they had run out of ammunition’. The officer concludes from his investigation that
the Tutsi in the stadium were killed with firearms. My
statistical result also corresponds to the observations of eyewitness Doctor Blam, who heard the intervention of
firearms and grenades during the massacre in the stadium.

MASSACRES

IN

R WA N D A

with traditional weapons in other communes
might be the distribution of machetes to the
local population. Detailed on-site investigations would be needed to confirm or refute
this hypothesis. From the statistics, however,
it is clear that significant differences in the
use of firearms existed between the locations
of killing and between the communes.
The reason why young Tutsi and men
who were working in the modern sector of
the economy had a higher probability of
being killed with firearms, compared to
other victims, can be inferred from the constraint on the behaviour of the perpetrators:
they had to save on ammunition and, thus,
used firearms only against those people who
could mount resistance. These people, in
turn, were young to middle-aged men with
a respected status in the commune.
The strong marginal effects are not due to
the enumerators, since 15 to 20 enumerators
were doing the data collection in one
commune and the communes with bad data
collection were left out of the regression.
At a more general level, the results of the
logistical regression support the five
hypotheses that were outlined earlier.
Clearly, the genocide had an organized
nature and was not a random killing spree.
In the sample, in 17.7% of all cases where
the weapon is known, the weapon is a
firearm. In the absence of any organization
behind the genocide, we would not observe
clusters of victims gathered in schools,
churches and stadiums where many (in
several massacres, 60–80%) of the victims
were killed by firearms instead of traditional
weapons. If age, gender, occupation, the
number of days after 6 April and, above all,
the location of killing and the residence of a
victim prove significant to explain the
weapon used, then genocide was anything
but a random process.

19

20

jour nal of P E A C E R E S E A RC H

volume 43 / number 1 / january 2006

Appendix A. Massacres in Kibuye Prefecture in April 1994 and Number of
Victims Registered in Data File
Place
Birambo
Bisesero
Church Kibuye
Church Mubuga
Church Nyange
Gatwaro Stadium
Kibingo
Kizenga
Mugonero Hospital
Murangara
Ngoma
Nyamagumba

Commune

Dates

Victims

Bwakira
Gishiyta
Gitesi
Gishiyta
Kivumu
Gitesi
Rwamatamu
Rwamatamu
Gishiyta
Gishiyta
Gishiyta
Mabanza

13–14
3 months
16–18
16–17
15–16
13–18
12–13
18
16
11–12
16
13–14

583
6,549
833
419
1,638
3,477
434
1,824
541
1,025
399
672

Several thousand victims were killed in ‘Bisesero’ or ‘in the mountains’. In order to determine whether or not these
killings were part of a massacre that fits our definition, I looked at all places that belong to the mountain range of
Bisesero (e.g. Karongi, Muyira, Uwingabo) and considered the victims who were killed at these places as victims of the
massacre at Bisesero. This makes sense, because massacres were committed at Bisesero during the entire period of the
genocide. I estimated that a total of some 13,000 people were killed in Bisesero between 6 April and 30 June. For the
estimation procedure of the number of victims at Bisesero, I refer to Verwimp (2003).
The number of victims who were killed in these massacres represents the victims who resided in the communes of
Bwakira, Gisovu, Gishiyta, Kivumu, Mabanza and Rwamatamu at the time of the genocide and whose place of death
was registered. The victims in the Parish of Kibuye, for example, came largely from the commune of Gitesi. Since data
collection was not performed well in this commune, these victims, numbering several thousand, do not appear in the
table.
I considered victims as victims of a massacre (dummy = 1) when it was registered that they died in one of the 12
massacres that I identified in the data. For the entry ‘in the mountains’ in the commune of Mwendo, I could not find
out if these victims were killed in a massacre or not.

Appendix B. Description of Variables in IBUKA Data File
Variable

Valid N

Mean

Std.dev.

Age
Gender
Occupation
Days after 6 April
Massacrea
Weapon
Total N

58,297
58,258
48,646
25,106
40,626
54,534
59,050

25.21

19.04

%
Male 51.7
Off-farm 4.6

12.86

11.82
Yes 43.6
Firearm 16.4

a As defined in the text, a massacre is an event where, in the course of less than three days, at least 100 people were killed
in one specific location. We have 40,626 victims in the database where we know whether or not they were killed in such
an event.

Philip Ver wimp

MASSACRES

IN

R WA N D A

Appendix C. Regression Results, Dependent Variable is Weapon Useda
Variable
Age
Age2
Gender
Gender*age
Off-farm
Days after April 6
Days after April 6 sq.
Gender*days after April 6
Gender*days after April 6 sq.
Constant
N

R2

Model 5
.04543***
–.00061***
–.09910**
.0099***
.

–2.0260***
54,505
(entire file)
0.009

Model 6

Model 7

Model 8

.0402***
–.00056***
.0882**
–.0087***
.6788***

.0406***
–.00055***
.0756*
–.0083***
.5546***

.0285***
–.0004***
–.3296 ***
–.0074***
.4611***
.08226***
–.0012***
.0812***
–.0022***

–1.9817***
53,080
(entire file)
0.012

–1.9323***
41,409
(2 comm.
left out)
0.011

–2.3948***
23,650

0.026

*** significant at the 1% level, ** significant at the 5% level, *significant at the 10% level.
a Gitesi and Rutsiro communes are left out because of lack of data.

Appendix D. Weights Used in the Weighted Regression in Model R4a
Massacre site/commune

Count

Date known

Weight

Integer weight

Birambo
Bisesero
Church Mubuga
Church Kibuye
Church Nyange
Mugonero Hospital
Kibingo
Kizenga
Murangara
Ngoma
Nyamagumba
Gatwaro Stadium

583
6,549
419
833
1,638
541
434
1,824
1,025
399
672
3,477

265
2,488
370
730
1,466
324
421
1,145
542
190
656
1,370

2.20
2.63
1.13
1.14
1.12
1.67
1.03
1.59
1.89
2.10
1.02
2.53

2
3
1
1
1
2
1
2
2
2
1
3

Bwakira
Gishyita
Gisovu
Kivumu
Mabanza
Rwamatamu

3,871
3,440
1,435
2,198
3,549
8,434

1,046
1,227
784
1,763
1,321
7,507

3.70
2.80
1.83
1.24
2.68
1.12

4
3
2
1
3
1

a In order to determine the weights, we first looked at the number of victims in each large-scale massacre for which we
have the date of death. Weights were then computed to adjust for the total number of victims in each massacre. Then,
the remaining victims per commune of residence were computed and weights were given again to adjust for the number
of victims without known date of death. Integer weights were used because STATA 9.0 does not allow for non-integer
weights.

21

22

jour nal of P E A C E R E S E A RC H

References
André, Catherine & Jean-Philippe Platteau,
1998. ‘Land Relations Under Unbearable
Stress: Rwanda Caught in the Malthusian
Trap’, Journal of Economic Behaviour and
Organisation 34(1): 1–47.
Collier, Paul; V. L. Elliott, Håvard Hegre, Anke
Hoeffler, Marta Reynal-Querol & Nicholas
Sambanis, 2003. Breaking the Conflict Trap:
Civil War and Development Policy. Washington, DC: World Bank.
De Lame, Danielle, 1996. Une colline entre mille
ou le calme avec la tempête. Transformations et
blocages du Rwanda rural [An Ordinary Hill or
Silence Before the Storm: Transformations
and Limits in Rural Rwanda]. Tervuren:
Musée Royal de l’Afrique Central.
de Soysa, Indra, 2002. ‘Paradise Is a Bazaar?
Greed, Creed and Goverance in Civil War,
1989–1999’, Journal of Peace Research 39(4):
395–416.
Des Forges, Alison, 1999. Leave None To Tell the
Story. New York: Human Rights Watch.
Des Forges, Alison, 2003. ‘The Striking Force:
Military and Militia in the Rwandan
Genocide’, paper presented at the conference
‘Unfolding of Genocide’, Butare, November.
Easterly, William, 2001. ‘Can Institutions
Resolve Ethnic Conflicts?’, Economic Development and Cultural Change 49(4): 687–705.
Goose, Stephen D. & Frank Smyth, 1994.
‘Arming Genocide in Rwanda’, Foreign Affairs
73(5): 86–96.
Guichaoua, A., 2002. Presentation at a conference on gender and genocide, University of
Namur, Belgium, May.
Human Rights Watch, 1994. Arming Rwanda:
The Arms Trade and Human Rights Abuses in
the Rwandan War. New York, January.
IBUKA, 1999. Dictionnaire Nominative des
Victimes de Génocide [Nominal Dictionary of
Victims of Genocide]. Kigali.
Krain, Matthew, 1997. ‘State-Sponsored Mass
Murder: The Onset and Severity of Genocides
and Politicides’, Journal of Conflict Resolution
41(3): 331–360.
Marchal, Luc, 2001. Voor de poorten van de hel,
getuignis van een peace-keeper [Before the Gates
of Hell: Testimony of a Peace-Keeper].
Leuven: Van Halewijck.

View publication stats

volume 43 / number 1 / january 2006
Mfizi, Christophe, 1992. ‘Reseau zero, lettre
ouverte à Monsieur le Président du MRND’
[Zero Network: Open Letter to the President
of the MRND], Éditions Uruhimbi, Kigali,
Juillet–Août.
National Agricultural Household Survey, 1984.
Kigali: Department of Agricultural Statistics.
Newbury, Catharine, 1988. The Cohesion of
Oppression: Clientship and Ethnicity in
Rwanda, 1860–1960. New York: Columbia
University Press.
Prunier, Gerald, 1995. The Rwanda Crisis: History
of a Genocide. New York: Columbia University
Press.
Sato, Y., 1980. Interview with President J.
Habyarimana, 12 July. Discours, Messages et
Entretiens de Son Excellence le GénéralMajor Habyarimana Juvénal, Président de la
République Rwandaise [Speeches, Messages
and Discussions by his Excellency MajorGeneral Juvénal Habyarimana, President of
the Republic of Rwanda]. Kigali: Présidence
de la République.
Small Arms Survey, 2002. Geneva: Institute of
International Studies.
Uvin, Peter, 1998. Aiding Violence: The Development Enterprise in Rwanda. West Hartford,
CT: Kumarian.
Verwimp, Philip, 2003. Development and
Genocide in Rwanda: A Political Economy
Analysis of Peasants and Power Under the
Habyarimana Regime. PhD dissertation,
Economics Department, University of
Leuven, January.
Verwimp, Philip, 2004. ‘Death and Survival
During the 1994 Genocide in Rwanda’, Population Studies 58(2): 233–245.

PHILIP VERWIMP, b. 1970, PhD in Economics (Catholic University of Leuven, 2003);
Fulbright-Hays Scholar, Yale University
(2004); Poverty Economist, World Bank
(2005); Co-Director, Households in Conflict
Network (2005– ); Lecturer, Institute of Social
Studies, The Hague (2005– ). Main interests:
development, political violence, dictatorship,
genocide. Current research: micro-analysis of
violent conflict.

Haut

fgtquery v.1.9, 9 février 2024