An Economic Profile of Peasant Perpetrators of
Genocide
Micro-level Evidence from Rwanda
Philip Verwimp*
Economics Department, Catholic University of Leuven, Naamsestraat 69, 3000 Leuven,
Belgium and Research Fellow, Genocide Studies Program, Yale University, USA
philip.verwimp@econ.kuleuven.ac.be
November 2003
Abstract of the paper
This paper presents the results of a research project in which we have traced 350 Rwandan household who were
part of a rural household survey before the Rwandan genocide (1994). Economic, demographic and agricultural
data from an extensive 1989-1992 survey can be linked with the condition of the household at the time of the
Genocide Transition Survey (2000). This allows us to study the fate of the household members during the
genocide. Our results show that age, sexe, the sex of the head of the household, the size of rented land, personal
off-farm income, gross household income and farm-level anti-erosion investment significantly determine the
probability of a household member to become a perpetrator of genocide. Tbese results are interpreted in the
political economy of Rwanda.
J.E.L. : C31, J43, O55
Keywords : peasants, survey research, genocide, Rwanda
* This research was funded by the Fund for Scientific Research (Flanders, Belgium) which enabled field work in
Rwanda. We owe many thanks to Dan Clay for the support during the initial stage of the project and the
permission to use the original data. We are indebted to Lode Berlage, Stefan Dercon, Alison Desforges, and Ben
Kiernan for critical comments and support throughout the research. I also thank the participants of the 7th Spring
Meeting of Young Economists in Paris and participants in seminars in Leuven and in The Hague (Institute of
Social Studies) for critical comments and suggestion. All responsibility remains with the author.
1. INTRODUCTION
In his book on the social origins of dictatorship and democracy, Barrington Moore (1967,
1993) writes that the Nazis best succeeded in appealing to peasants whose land holdings were
relatively small and unprofitable in the particular area in which they were located. Moore cites
several publications that demonstrate that the Nazis won 70 to 80% and sometimes 100% of
the vote in areas with small farms and poor soils, heavily dependent on sensitive markets for
young cattle and hogs. Parts of an area known as Geest and parts of Hanover show association
between small farms and voting Nazi. In Nuremberg too, Nazis got very high percentages in
areas of relatively low land values, middle-sized farms and generally marginal agriculture
dependent on the urban market.1 According to Moore, a specific type of agrarian relation
between peasants, landlords and the state is favourable to the development of fascism. He
writes that peasant revolutions have occurred under regimes where the political and landed
elite has not succeed in transforming an agrarian economy based on traditional and feudal
relations into a modern economy based on commercial agriculture. The elites of what Moore
calls “labour-repressive” regimes have preferred to maintain peasant society while squeezing
more dues, taxes and services out of the peasantry (p. 434-435). A repressive-repressive regime
is not just a intensive-intensive economy (as opposed to a capital-intensive economy) but also
a system where non-market (i.e., political) mechanisms make sure that there is an adequate
labour force for working the soil, which keeps the peasants on their farms. The attraction of
Nazism to small farmers then was the result of the advance of capitalism (commercial
agriculture), with its problems of prices and mortgages that seemed to be controlled by hostile
city middlemen and bankers. Nazi propaganda presented the romantic image of an idealised
peasant. As Moore writes,
“The Nazis were fond of stressing the point that, for the peasant, land is more than a means
with which to earn a living; it has all the sentimental overtones of Heimat. Physiocratic and
liberal notions found themselves jumbled together in these doctrines of the radical right”
“A firm stock of small and middle peasants,” said Hitler in Mein Kampf, “has still been at all
times the best protection against social evils as we have them now” And further,
“Industry and commerce retreat from their unhealthy leading positions and fit into the general
framework of a national economy based on need and equality. Both are then no longer the basis
for feeding the nation, but only a help in this”.
1
Moore, B., Social Origins of Dictatorship and Democracy, Beacon Press, 1993, p. 449.
2
Interestingly, the former Rwandan president, Juvénal Habyarimana shared these views on
peasantry. He too considered the peasants as the basis of society; he too did not like the
liberties of urban life and he too considered food self-sufficiency to be Rwanda’s prime
2
objective. We recall that 95% of the Rwandan population resided in the rural areas.
Umuganda, in addition, was a prime example of a policy to control and mobilise peasant
labour. As we have repeated several times, Habyarimana professed to solve a problem that
was inherently unsolvable: modernising Rwandan society without changing its social structure.
However, there is almost no data available to test Moore’s hypothesis for Rwanda. Did
peasants with small and unproductive landholdings participate more or less in the genocide
compared to wealthy peasants or landlords? Scholars writing on the Rwandan genocide
consider both these questions very important and at the same time largely unanswered. Peter
Uvin for example writes that with the exception of the studies by André and Platteau (1995)
and Longman (1995), we do not possess the micro-data necessary to test different hypotheses.
3
Claudine Vidal puts it as follows:
“In reality, according to my knowledge, systematic research on the adhesion of peasants to the
4
genocide and on the voluntary or forced enrolment of peasant-killers has not yet been undertaken”.
Granted that voting for the Nazis is different from participating in genocide, the hypothesis
put forward is worth researching. Moore has argued that fascism appealed to small peasants
because of its anti-capitalist rhetoric in which the Jews were presented as a commercial elite of
city bankers and traders. Exactly the same propaganda was used to describe the Tutsi in
Rwanda. In this paper we therefore research the economic profile of the peasants who were
attracted by the genocidal rhetoric of the Habyarimana regime.
One has to be careful, however, to consider Nazism as a rural movement. It is not because the
Nazis idealised German peasants that they can be considered a rural movement. Several
historians argue that the Nazis had a catchall ideology, a total concept of a nation in which
each group (farmers, labourers, bureaucrats, women, children, capitalists, teachers, soldiers,
2
3
4
Verwimp, Ph., Development Ideology, the Peasantry and Genocide, Rwanda represented in Habyarimana’s
speeches, Journal of Genocide Research, November 2000.
Uvin, P. Aiding violence, The development enterprise in Rwanda, 1998, p.200-201 and p.219.
Vidal, C., Question sur le rôle des paysans durant le génocide des Rwandais tutsi, Cahiers d’Eudes
africaines, 150-152, 1998, p.332. Author’s translation from the French text.
3
5
etc.) had to contribute to the power of the Third Reich. Renton writes that the Nazis
recruited cadres in the cities more then in the countryside. Contrary to rhetoric too, the Nazis
as well as the Habyarimana regime, placed agriculture, in terms of allocation of government
budget, consistently behind commerce and industry. Hitler personally sponsored and helped
to plan the Volkswagen or “people’s car”. The Nazis built monuments to the “unknown
engineer” (Renton, p. 145). As far as Rwanda is concerned, we will see this dual approach: the
regime idealised the peasantry but, as Claudine Vidal observes, the leaders at the national as
well as at the local level lived urban lives and used all sorts of means to distinguish themselves
6
from peasant life. We thus need the necessary data to investigate which class of people was
especially attracted to a genocidal solution in Rwanda. Poor peasants? Day labourers? Rich
farmers? Urban elites?
This paper offers an empirical analysis of peasant participation in the Rwandan genocide. The
author collected data in Rwanda that allows, among other things, to test Moore’s hypothesis
for Rwanda. After a short review of two empirical papers on peasant participation in Rwanda,
we describe the methodology of the fieldwork. Part 3 then presents descriptive statistics. Part
4 tests Moore’s hypothesis. Parts 5 and 6 analyse the importance of the land and labour
market and present regression results. Part 7 interprets the empirical results in the political
economy of Rwanda.
André is one of the few persons to research the link between bad economic conditions and
the genocide. She spent 14 months on a hill in Gisenyi Prefecture in 1988 (five months) and
1993 (nine months) enabling her to gather detailed information on the rural livelihoods of
peasants. In her fieldwork, in Gisenyi Prefecture, she focussed on land transactions, land
disputes and the effects of land scarcity in general. The interval (five years) between her first
and her second stay allowed her to follow up changes over time. The results are astonishing:
the incidence of quasi-landlessness is increasing rapidly; land holdings have become extremely
fragmented; average size of land holding per capita decreases steadily; tensions over land
5
6
Renton, D., The Agrarian Roots of Fascism: German Exceptionalism Revisited, The Journal of Peasant
Studies, Vol 28, No 4, July 2001, p. 142-145; Griffin, R., Fascism: a reader , Oxford University Press,
1995
Vidal, C, Sociologie des Passions, Editions Karthala, Paris, 1991, p. 30-31
4
within the household are rising; young people postpone marriage because they cannot find
land; the (illegal) land market is very active. 7
After the genocide, André tried to collect data on the fate of the household members in her
data set. For that purpose, she travelled to the Kivu and interviewed people in refugee camps.
She found (1998, p. 40) that people, who fell victim to the 1994 events, were not a random
selection of her sample. From information on 32 victims, André found that 10 of them had
comparatively large land properties that 11 of them were land-poor and malnourished and that
10 others were considered either troublemakers or youngsters engaged in militias. André and
Platteau (her co-author) conclude that the 1994 events provided a unique opportunity to settle
scores, and they consider these people victims of the war (1998, p. 39). The problem with this
research, however, is that it provides a profile of victims, not of killers. André and Platteau
consider members of youth militia as victims of the war (1998, p. 41). The question, however,
is whether these people really are victims. Some of them may have been killed in the act of
committing murders themselves. It is certain that score-settling occurred during the genocide,
but André and Platteau do not investigate participation in genocide (they do not even use the
word). They analyse the characteristics of people killed (no matter how or where) and from
this they derive that land disputes must have been the reason behind their death. One would
have wished that the authors not only looked for the characteristics of those they consider
victims, but also for the characteristics of perpetrators.8 They show that they have tried to
look at killer profiles when they write that (without supplying data)
“In our study area, it is noticeable that the most violent people tend to be young and to come from
poor, yet not the most extremely poor family backgrounds. Bleak prospects for the future and a sense
of meaninglessness in life, rather than struggle-for-survival under the harshest circumstances, seem to
lead young people into violence whether through enlistment in militias or otherwise.”
They also note that one Tutsi woman was the first to be killed. She was an earlier victim of a
failed murder attempt by an anti-Tutsi young radical in January 1993. The authors write that
(p. 40-41)
“it is probably a simplification to view her assassination as a purely racial act. As a matter of fact, she
was hated for many reasons, particularly because she came from the south of the country and was
therefore considered to be a stranger, and because she inherited a relatively large land property upon
7
8
André, C and Platteau, J-Ph., Land Relations under unbearable stress: Rwanda caught in the Malthusian
Trap, Journal of Economic Behaviour and Organisation, vol 34, 1998. First published as a Cahiers de la
Faculté des Sciences Economiques et Sociales, no 164, January 1996.
The authors indicate that they cold not obtain this information for their own security (p. 41)
5
the death of her husband of whom she was the fourth wife (an anomaly in a society where women do
not inherit from their husband). She was involved in many land disputes, which were clearly not of her
own making.”
This statement suggests that ethnicity, region of origin and land disputes interfere with each
another. A Tutsi widow, from the south, living in an area where she is the only Tutsi, is clearly
in a very weak position to defend her land rights. It is not a surprise then that she was the first
victim of the genocide in that area.
Longman (1995) compared the relationship between the local elite and the peasants in two
Rwandan communes, Kirinda and Biguhu. He found that, before the genocide, the local elite in
Kirinda acted in an authoritarian self-serving way vis-à-vis the population. In Biguhu, on the
other hand, relations between the local elite and the population were based on co-operation
and understanding. It turned out that the elite and the population in both communes acted
differently during the genocide. In Kirinda, the elite organised a mob to kill the local Tutsi in
an attempt to re-establish their authority whereas in Biguhu participation was minimal and
clearly initiated from outside.
9
2. The Tracing Methodology of the Genocide Transition Survey (GTS)
In order to research the fate of the members of Rwandan households in transition from civil
war and genocide to a situation of relative peace, we needed data at the household level. This
data had to be unbiased, meaning that whatever information on households one could find
(e.g., income, household composition, location, farm size, etc.), that information should be
collected independently of the behaviour of household members during the genocide. This
condition is not satisfied, for example, when one considers a sample of perpetrators or a
sample of prisoners. The author therefore decided to trace the rural households interviewed
by Dan Clay from Michigan State University and the Department of Agricultural Statistics
(DSA) before the genocide. Clay had interviewed 1248 rural households from 1989 to 1992 in
all Rwandan prefectures. He collected detailed data in the demographic, economic and
agricultural situation of farm households. This data set is a unique source to study the
livelihood of Rwandan peasants before the genocide. The research strategy would provide the
9
Longman, T., Genocide and Socio-Political Change : Massacres in two Rwandan Villages, Issue: a Journal
of Opinion, 1995
6
researcher with data on the fate of rural households during and after the genocide, data based
on a pre-genocide sample of rural households.
The intent of D. Clay and the DSA was to first conduct an agricultural survey. That is why the
agricultural data are particularly detailed. They include the crops grown, the number of parcels,
the size of each parcel for each crop grown, the degree of intercropping, the use of fertiliser,
the slope of the fields, the length of the anti-erosion ditches per field, the soil quality of each
field and so on. In a personal conversation Dan Clay explained that he could not ask the
ethnic affiliation of the interviewed farm households, because the government did not want
this. In the political climate of the 1989-1992 period, ethnicity was indeed a very sensitive
subject.
10
The 1989-1992 survey also has data on off-farm activities such as the number of days each
member of the household worked outside the family farm, the income earned from this
activity, the kind of off-farm activity. However, households who did not own or cultivate land,
mostly young wage labourers not living with their parents, were excluded from the sample.
Full-time off-farm workers living with their parents were thus included in the sample as part
of the farm household. We stress that households with very small landholdings were included
in the sample. Nevertheless, the choice of Clay and the DSA not to include landless
households reduces the representative significance of the 1989-1992 sample.
D. Clay had all household survey data computerised, but could not provide a list of the
location and the names of the surveyed households. Since the former regime and its allies had
to evacuate Kigali in a hurry and did not have the time to destroy their archives (they only
destroyed or stole the computers), we believed to have a (small) chance of finding the old
surveys. Digging in the archives in the Ministries of Agriculture, Economic Planning and their
respective statistics department did not result in a list of the households. However, under a
layer of dust in one of the archives, we finally found the original questionnaires with the
location and the names of the heads of households mentioned on the first page. On the
whole, we managed to find the references of 73% of the originally surveyed households in that
10
It still is in Rwanda today, maybe even more as official parlance does not use the ethnic categories
anymore. For our research strategy this meant that we had to approach this subject with much caution.
7
archive. For seven of the ten prefectures, the references were almost complete and for three
they were almost completely missing.
For reasons of budgetary limitations, however, we could not trace all 1248 households. One
must realise that survey research in general and in Rwanda in particular is very expensive and
time consuming. Furthermore, a genocide transition survey is not free from security concerns.
The genocide took place in 1994, but it is very much present in Rwandan society today. In the
summer of 2000 we decided to trace households in three prefectures, Gitarama, Kibuye and
Gikongoro. In total 352 households were surveyed by Professor Clay. With 160 households
(ten clusters) in the first and 96 households (six clusters) both in the second and third
prefecture. These prefectures were chosen for a variety of reasons. Firstly, we had the
information needed to find the households in these prefectures. Secondly, at that time, these
areas were safer to work in than Ruhengeri or Gisenyi. Thirdly, Imidugudu policy
(villagesation) was implemented to a lesser degree in these prefectures compared to Kibungo
and rural Kigali leaving more hope to find the households in the same location as before the
war. Fourthly, the prefectures had a mix of a very complete genocide (Gikongoro and Kibuye)
and a less complete genocide (Gitarama). Fifthly, the prefectures had a sizeable Tutsi
population before the war. (This does only distinguish them from the northern prefectures).
Sixthly, the prefectures encompass both very poor and not so poor communes.
We designed the questionnaires for the Genocide Transition Survey and decided to proceed in
two phases. A team of research assistants, selected at the National University of Rwanda in
Butare (one for each commune, the equivalent of one for each cluster of 16 households)
would try to find the households in the indicated sectors. In the first stage, the research
assistants would not take a detailed questionnaire, but would only take down a limited amount
of information. We first wanted to know whether or not we were able to trace the households
in their original dwellings. The information collected in this first stage was the following:
-
can we locate at least one member of the household surveyed in 1989-1992?
are the head of household and his wife alive or dead?
what is the ethnic group of the head of household and of his wife?
how many members did the household have in 2000?
8
-
to what category (broadly speaking) does the head of household belong? Is s/he a
genocide survivor, is s/he in prison, is s/he abroad?
what was the age of the head of household in 2000?
Table 1: Survey Sites of the Genocide Transition Survey
Gitarama
commune
Nyamabuye
Ntongwe
Mugina
Tambwe
Musambira
Runda
Taba
Nyakabanda
Masango
Murama
Gikongoro
commune
Musange
Rwamiko
Nyamagabe
Nyamagabe
Muko
Kinyamakara
Kibuye
commune
Mabanza
Kivumu
Rutsiro
Gitesi
Gisovu
Bwakira
Table 2: Descriptive data by prefecture (n=340)*
Number of households in GTS
with Hutu head of household
with Tutsi head of household
with Twa head of household
with ethnicity unknown
Gitarama
155
133
17
5
0
Gikongoro
96
90
5
0
1
Kibuye
89
81
5
0
3
* during the GTS no information at all could be found for 12 households
The research assistants were told to approach the households in a prudent manner, taking
time to explain that the research was only done for scientific purposes and that we neither
belonged to the Rwandan judiciary, nor to the International Tribunal or to the government. In
fact, if research assistants thought that households were not approachable, they could gather
the information for this first stage of the project by talking to neighbours or to government
authorities. Since tracing of households requires extensive knowledge of the area of residence,
9
we decided to work with research assistants from the communes itself. However, since this
kind of work is very sensitive, we decided not to take research assistants from the sector
where the households were located, but from a different sector within the commune. This
would at least give a minimum guarantee of security, would save transportation cost and
would make local knowledge of the area of residence available to the researcher. Relying on
earlier tracing done in Indonesia by Duncan Thomas (et al.) (1998), we decided to choose
research assistants with a good knowledge of mathematics11. Since the survey is a quantitative
research project, research assistants with a good grasp of mathematics would not only be
helpful in the tracing stage of the project (as they were in Indonesia), but also in the actual
completion of the questionnaires.
The following characteristics determined our choice of research assistants:
-
Having a good grasp of mathematics; in practice this came down to having a university
degree
Having resided in the commune were the interviews take place before and during the
genocide
Being at least 21 years old
Previous experience in survey research, especially interviewing
Being able and willing to undertake survey research on the genocide. This came down to
my question whether the candidate-enumerator was on good terms with both the Tutsi
and the Hutu population of his/her commune of origin
For the first stage of the project, a one day training session was organised. Research assistants
were told not to be satisfied with one source of information, but to cross-check their
information with members of both ethnic groups. In the end, we employed 22 research
assistants, one for each cluster of 16 households. Eighteen research assistants had a university
degree or were about to finish their university studies. Four research assistants had taken
teacher training during their high school studies and were currently teaching in primary
schools. Eight research assistants were female, 14 were male. Research assistants came from
both ethnic groups and all of them originated from the communes where the survey was to be
held.
11
Duncan, T., Frankenberg, E. and Smith, J., Lost but not Forgotten, Attrition in the Indonesian Family Life
Survey, Paper presented at the Conference on Data Quality in Longitudinal Surveys, Institute for Social
Research, University of Michigan, October 1998.
10
The advantage of a tracing exercise is that the researcher has information on the households
that is not available to the research assistants. The 1989-1992 database provided a very good
12
instrument to check whether the research assistants found the intended households. After
the first stage, the information supplied by the research assistants was checked with the
database and the researcher was able to criticise unfinished work. Research assistants were also
given an opportunity to correct information collected upon their return to the household for
the second stage of the project.
The 352 households in the prefectures of Gitarama, Kibuye and Gikongoro interviewed by
Clay and the DSA taken together have over 1,900 household members. In the first stage of
the project, we managed to find information on 340 of these households and on 1,800
persons included in the 1989-1992 survey. This success may be attributed to the preparation
of the tracing exercise and to the constant presence in the field of the researcher, but it was
also due to the organisation of Rwandan society. Indeed, communal authorities have a detailed
and well-organised record of all the inhabitants in each sector and cell. They carefully register
every birth, death or migration in or out of the commune. The success of the tracing exercise
was also a result of Rwandan culture: people in the hills know each other and know of each
other’s whereabouts. Even when a person is not present on the hill, neighbours have
information on where they can be found. When we say that we found information on
approximately 1,800 of the 1,900 people in the original sample, this does not mean that they
were all present in their original dwellings, as will be documented in the Tables.
3. Criminological data and its categorisation
In one of the most sensitive and difficult questions in the survey, research assistants were
asked to find out what the most appropriate “type” was of each member of the household
under investigation. Knowing very well that few persons would fit into one type, research
assistants had to register the most appropriate characterisation for each person choosing from
the following types: (1) victim/survivor; (2) perpetrator; (3) thief; (4) innocent; (5) protector;
(6) type unknown. When desired or appropriate, research assistants could fill in two types for
12
Apart from the data on the households, the researcher also had another way of controlling the tracking done
by the research assistants. During the 1989-1992 agricultural survey, the households were given a solid
basket as a kind of reward for their co-operation with the survey. The research assistants were told to ask
whether the household received this item and if they still had it.
11
the same person, starting with the most appropriate. This happened when, for example, a
person had both killed Tutsi (perpetrator) as well as hidden Tutsi (protector) in their own
house or when a person both protected Tutsi in his or her house and looted property (thief) in
someone else’s house. In this way, we tried to capture the ambiguity and complexity of the
people’s behaviour. When the situation was too dangerous to ask the household directly about
its members’ involvement in the genocide, indirect methods of investigation were used.
Research assistants could then ask the local authorities and neighbours questions on the
whereabouts of household members. We could not even envisage that the life of an
enumerator would come under threat. Table 3 presents the information found by the skilled
and determined team of research assistants. From a total of 1,838 household members of
which we were able to register the ethnicity, we found 1,657 Hutu, 155 Tutsi and 26 Twa.
Among the 461 Hutu adult males (age > 12), we found 59 perpetrators, 44 thieves, 334
innocent and 22 protectors.
We thus have 59 male adult perpetrators, 6 female adult perpetrators and 5 children, totalling
70 perpetrators. In order to get a profile of the adult male perpetrators, we compare their
characteristics (registered before the genocide) with those of all Hutu males in the sample. The
latter namely constituted the group from which potential perpetrators were drawn and as such
form the baseline for comparison in the Table 4.
In our sample 12.8% of all adult Hutu males who were alive in March 1994 participated in the
genocide. The average age of the adult perpetrators was 33 years (Table 4 does not include
child perpetrators). Among the educated (having completed primary school or more), the
perpetrators represent 21.1% of all educated male Hutu (15 out of 71). Perpetrators are thus
over-represented among the educated. This is also the case for adult male Hutu with a parttime or full-time off-farm activity, where perpetrators represent 25% of this group.
12
Table 3: Type of person during genocide (most appropriate), n = 1838
Hutu
Tutsi
Twa
Total
#
1,657
%
100
#
155
%
100
#
26
%
100
Age > 12
Age < = 12
1,056
601
63.7
36.3
99
56
63.8
36.2
16
10
61.5
38.5
1,056
8
65
94
733
44
105
100
0.8
6.6
9.5
68.0
4.5
10.6
99
99
0
0
0
0
0
100
100
16
0
1
2
10
0
3
100
25.0
601
0
5
15
562
1
18
100
0
0.5
2.5
93.7
0.2
3.1
55
55
0
0
0
0
1
100
100
10
100
5
50.0
5
50.0
ADULTS
Age > 12
Victim/survivor
Perpetrator
Thief
Innocent
Protector
Unknown type
8.3
16.7
50.0
CHILDREN
Age <= 12
Victim/survivor
Perpetrator
Thief
Innocent
Protector
Unknown type
HUTU ADULTS WITH KNOWN TYPE, n = 947
Male
#
%
Total
461
100
Victim/survivor
0
0
Perpetrator
59
13.2
Thief
44
10.4
Innocent
334
70.6
Protector
22
5.2
Female
#
486
8
6
50
399
22
%
100
1.8
1.4
11.0
81.1
4.8
In absolute numbers, most Hutu perpetrators were either married or bachelors. However,
among the divorced, the widowed and those who lived out of wedlock, the perpetrators were
particularly well-represented (24.2%), compared to the overall percentage of 12.8%. Adult
male perpetrators are more strongly represented among heads of households than among sons
13
living with parents. However, in half of the households with at least one son as a perpetrator,
the head of the household is a woman, making the (oldest) son the acting male head of the
household. In the other households with at least one son-perpetrator and with the head of the
Table 4: Profile of adult male Hutu, n = 461
(1) Personal characteristics
# All Hutu
adult males
Number
Perpetrators
%
#
461
59
12.8
Average
Minimum
Maximum
Education
34.5
13
93
33
16
68
Never attended school
Incomplete primary
Completed primary
Post primary (CERAI)
or secondary
Primary or higher
No information
156
208
57
21
22
10
13.4
10.6
17.5
14
71
26
5
15
1
35.7
21.1
Main occupation
Cultivator
Of which without off-farm activity
with off-farm activity
333
220
113
51
22
29
15.3
10.0
25.6
Full-time off-farm occupation
of which
businessman
administrator
artisan
manual worker
All Off-farm income earners
14
3
3
2
6
127
3
1
2
0
0
32
21.4
33.3
66.6
0
0
25.2
3
0
0
1
1
6.1
0
0
20.0
Age
Table 4: continued
Pupil/Student
Domestic worker
Other
Unemployed
No information
63
13
3
7
28
14
Marital Status
Married
Bachelor
Divorced
Wedlock
Widower
All broken relations
No information
181
222
4
25
4
33
24
23
27
1
7
0
8
12.7
12.1
25.0
28.0
0
24.2
Family position
Head of the household
Son living with parents
of which with female head
No information
218
236
57
7
32
27
13
14.6
11.4
22.8
household (the father of that son) still living and present in the household, it was only the son
who participated, not the father. This means that in most cases in our sample, one male
member of the household participated, namely either the head of the household or a son of
the household. As if households decided to supply the labour of one person per
13
household to the genocidal effort. Off all households with at least one perpetrator, 81%
(43 out of 53 households) counted exactly one perpetrator. This means that 19% of these
households had more than one perpetrator. Even then the father-son combination is seldom
observed, the households often have several brothers or man and wife participating together
in the genocide.
Where the characteristics of the households are concerned (see Table 5), households with at
least one perpetrator have 10% more members, 8% in adult equivalents, compared to the
average Hutu household. They own, on average equal sizes of land, but in terms of adult
equivalents, they have less land. This indicates a relative abundance of labour on farms with at
least one perpetrator. This explains why they rent more land than average. Households with at
least one perpetrator have higher incomes than others. This is especially true because they
have a higher income from off-farm activities and beer sales. On average, households with at
least one perpetrator have a gross income, which is 25% higher than the income of the
average Hutu household, and is 15% higher when reckoned in adult equivalents. This
difference also applies to beer sales and the difference is more then 100% higher for income
13
This result is an indication that some households may have regarded participating in the genocide as a
state-directed obligation. Peasant households under the Habyarimana regime had to contribute labour to
the weekly Umuganda (a collective form of labour) and this was usually performed by one person per
household. The 1993 FIDH human rights report on Rwanda as well as Desforges (1999) also describe
how local authorities summoned people to participate in the killing, calling it a ‘special Umuganda’.
15
from off-farm activities. Total monetary income (the sum of off-farm work, beer and crop
sales) represents 50% of the gross income in households with at least one perpetrator and only
40% in the
Table 5: Profile of Households with Hutu head of the household (n=282)
(2) Household characteristics
# All Hutu
households
Number
household size 1991
adult equivalents (ae)
gross income
(1) Auto-consumption
(2) Crop sales
(3) Beer sales
(4) off-farm income
(5a) livestock cons
(5b) livestock sales
(6) transfers received
gross income per ae
monetary income (2)+(3)+(4)+(5b)
food consumption
size of owned land (in ares)
size of owned land (ae)
% of cultivated land rented
distance to paved road (in km)
wages paid to employees (in RWF)
tot. hh labour in ae days
on-farm hh labour
labour hired in (in days per year)
labour hired out
282
5.1
4.7
42,230
21,473
3,760
4,838
6,080
1,768
2,906
1,386
9,804
17,.595
29,276
Households with
at least one perpetrator
#
% difference
54
5.6
5.14
52,872
22,410
4,141
6,054
13,604
2,826
2,681
1,155
11272 +15
26,80
32,758
+9
+9
+25
+4
+10
+25
+123
+59
-8
-17
+50
+11
93.9
21.9
10
32
93.8
18.7
14
30
0
-14
+40
-6
1.622
916
710
26
65
2.625
997
859
42
118
+61
+8
+20
+61
+81
(1)
(2)
(3)
(4)
Includes crop consumption from own production.
Includes sales of all crops (food, domestic cash crops and export crops).
Includes the sales of artisan brewed beer (banana and sorghum been).
Includes income from skilled and unskilled off-farm work and from business activities other than
beer sales.
(5) Includes livestock and livestock products consumed from own production and sold.
(6) Includes all kinds of gifts of food, beer and livestock received.
16
average Hutu household. Households with at least one perpetrator eat 60% more meat, milk
and eggs as shown by data on auto-consumption of livestock. These households consume in
general 12% more food than the average Hutu household. The income from off-farm
activities is also reflected in the number of days worked off the farm, which is almost double
the number for the average Hutu household. They also hire in more labour and thus pay a
higher wage bill. This means that, on average, households with at least one perpetrator employ
more people, compared to households with no perpetrators.
Since we are dealing with averages, we should consider in more detail the composition of the
households with a least one perpetrator. It could be that these averages result from a group of
high-income earners and landed households on the one hand and poor landless households on
the other hand. In Table 6, we therefore compare households without and with at least one
perpetrator, while distinguishing three groups (tertiles) on the basis of land owned, land rented
and income.
Table 6: Distribution of Hutu households over land and income tertiles, n = 279
Landholding tertiles in ae*,**
Lowest
Middle
Highest
< 10.3 are
10.3
> 23.2 are
#
%
#
%
#
%
No perpetrator
At least one perpetrator
75
17
81.5
18.5
71
20
78.0
22.0
81
15
84.4
15.6
*The tertiles for owned land were calculated with inclusion of the Tutsi households who do not figure
in the presented data.
**When we take land holdings per household instead of per adult equivalents, we do not observe a
distinction between perpetrator households and non-perpetrator households.
% Landrenting “tertiles”*
Lowest
0
#
%
No perpetrator
At least one perpetrator
133
20
87
13
Middle
< 10%
#
%
Highest
> 10 %
#
%
44
12
78.5
21.5
52
20
72.2
27.8
* tertiles is put between brackets because they are not of equal size.
17
Gross Income tertiles per ae*
Lowest
< 5338
#
%
No perpetrator
At least one perpetrator
79
17
82.3
17.7
Middle
5338
#
%
82
15
84.5
15.5
Highest
> 10.900
#
%
75
22
77.3
22.7
* The income tertiles were calculated with inclusion of the Tutsi households who do not figure in the
presented data.
Off-farm income “tertiles” per ae*
Lowest
=0
#
%
No perpetrator
At least one perpetrator
95
12
88.8
11.2
Middle
< 911
#
%
Highest
> 911
#
%
79
13
85.9
14.1
63
29
68.5
31.5
* From the whole sample (including Tutsi), 37% of the households have no off-farm income.
This group composes the first (lowest) group. The limits for the middle and highest groups were
therefore set at 31.5% of the households of the entire sample each.
Off-farm income as percentage of gross income, tertiles*
Lowest
Middle
=0
< 17%
#
%
#
%
Highest
>17 %
#
%
No perpetrator
At least one perpetrator
57
26
91
12
88
11.5
81
15
84.4
15.6
67.9
31.0
* From the whole sample (including Tutsi), 37% of the households have no off-farm income. This
group makes up the first (lowest) group. The limits for the middle and highest groups were
therefore set at 31.5% of the households of the entire sample each.
The different sections in Table 6 are very similar, households with a high amount of off-farm
income per adult equivalent (> 911) are most of the time also households with a high
percentage of their income from off-farm activities. The Pearson correlation between these
two Tables is 0.915 with a significance of 0.000. In none of the above four Tables does it
make much difference to present data on land, gross income and off-farm income per capita or
per household instead off per adult equivalents. The outlook of all Tables remains the same:
households with at least one perpetrator are almost equally represented over the landholding
18
tertiles, over-represented in the landrenting tertiles, somewhat over-represented in the high
gross income tertile and especially over-represented in the high off-farm income tertile.
4. LOOKING AT MOORE’S HYPOTHESIS
We drew Tables 7 and 8 to investigate Barrington Moore’s hypothesis that especially farmers
with small and unproductive landholdings were attracted to Nazism, in the case of Rwanda.
Table 7: Land size and marginal value products
Cultivated land tertiles (in ae)*
smallest
middle
highest
Marginal value product of land
Marginal value product of labour
18,820
9,996
5,862
26
35
49
* includes both owned and rented land;
Source: Byiringiro and Reardon (1996)
Byiringiro and Reardon (1996) computed marginal value products of land and labour from the
estimates of a production function with the DSA data.14 They show that there is a strong
inverse relationship between farm size and land productivity, and the opposite for labour
productivity. Their results suggest that the marginal value product of land on smaller farms is
well above the rental price of land, implying factor use inefficiency and constraints on land
access. By contrast, the marginal value product of labour on smaller farms was well below the
market wage (100 RWF), implying bottling up (surplus) of labour on smaller farms and
constraints on access to the labour market and perhaps barriers to entry into small businesses.
These results thus suggest that small farmers experience constraints on the land as well as on
the labour market.
The differences between small, middle-size and large farms presented in Table 7 are common
in a developing world context and are not particular to Rwanda. In Table 6 we have already
seen that perpetrators are not particularly over-represented among owners of small farms, but
rather among farmers renting in a lot of the land they cultivate. In order to test Moore’s
hypothesis, we now look at the following question: are perpetrators over-represented among
farmers experiencing low land and land productivity inside each of the three tertiles (small,
14
Byiringiro, F. and Reardon, T., Farm productivity in Rwanda: effects of farm size, erosion, and soil
conservation investments, Agricultural Economics, 15, 1996, p. 132.
19
middle, large)? We test this hypothesis by looking at the productivity of each farm compared
to the local (i.e., cluster level) average productivity (as Moore originally did). We also carry out
this analysis for soil quality and for rented land tertiles. The results are shown in Table 8.
Table 8: Chi-Squared Tests for differences between
perpetrator and non-perpetrator households
Owned land tertiles
>average
Rented land tertiles
>average
Marginal land productivity
0.362
0.749
Marginal labour productivity
0.445
0.040 **
0.282
0.109 *
Soil quality
0.528
0.948
0.00 ***
0.568
0.150 +
0.247
***significant at the 1% level, ** significant at the 5% level, *significant at the 11% level,
+
significant at the 15% level
The weakest, statistically non-significant results are found for Moore’s original hypothesis, to
wit on land productivity. There is no over-representation of perpetrators among farmers with
farms suffering from low land productivity. At least at this level, Moore’s hypothesis does not
seem to apply to perpetrators of the Rwandan genocide in the areas of our field work. As for
the productivity of labour, a hypothesis not directly researched by Moore but that seems
important to look at in a poor economy with lack of off-farm jobs, the results are somewhat
more statistically significant. Perpetrators seem to be over-represented among farmers having
higher than average levels of labour productivity, both when we consider owned land tertiles
as well as rented land tertiles. The statistically strongest result is delivered for soil quality. The
latter is measured by the C-value, a measure of protective land use. Lower C-values indicate
better practices. Perpetrators are not over-represented among farmers with higher than
average C-values when we consider owned land tertiles As far as rented land is concerned,
perpetrators are over-represented among farmers with good soil quality.15 These results do not
seem to confirm Moore’s hypothesis either.
15
In order limit the number of Tables presented, we do not present all cross-tabulations for small, middlesized and large farms on the one hand and the number of perpetrators in each of the productivity groups
(> or < than average) on the other hand. Perpetrators were especially over-represented among farmers
with good soil quality (low C-value) who rent in a lot of land. The data does not allow to distinguish
between the soil quality of the owned and the rented land, but one may assume that farmers will not rent
in land of poor soil quality (i.e., high C-values).
20
Thus, we find that perpetrators in general do not own smaller farms compared to nonperpetrators, do not experience lower land productivity and do not have poor soil quality. In
terms of labour productivity, households with perpetrators do better than the local average.
We do find that perpetrators, more than non-perpetrators, are active on the land rental
market. In the next section we will therefore further investigate the land and labour markets.
5. MORE EVIDENCE FROM THE RURAL LABOUR MARKET
In Table 9, households are grouped according to the kind of off-farm income at the household
level. When none of the members of a farm household has an off-farm income, the household
is put in the first group. When the household earns an income from agricultural off-farm work
and not from non-agricultural off-farm work or when the income earned in the latter is
smaller then the one earned in the former, the household is part of the second group. The
third group then consists of the households whose members earn an income in nonagricultural activities, be it as skilled professionals or unskilled workers. It turns out that the
sample, at the household level, has roughly one third of the households in each of the three
categories. From Table 9 it is clear that the households with at least one perpetrator are more
represented in the third group, which represents households with one or more members
earning an income in non-agricultural off-farm activities, than in the other groups. Thirty one
percent (31.3%) of the households of this group had one or more of their members among
the perpetrators. The sample mean is 18.8%. This average means that one in six to one in five
of the Hutu households in the sample have at least one perpetrator among their members.
The Table also shows that very few of the households without off-farm income are represented
among the households with at least one perpetrator. The Chi-Square test shows that the
difference in participation between the groups is highly significant.
The results of Table 9 are confirmed in Table 10. Table 10 shows data on 460 Hutu adult
males. Off-farm income earners are over-represented among the perpetrators. In fact, the
percentage of perpetrators in each occupational group rises with the degree of detachment
from agricultural work. Only 8.4% of the members of the group performing no off-farm work
(peasants) were perpetrators. This figure rises to 20% for the group whose members perform
agricultural work on someone else’s farm (day labourers, employees) and rises even to 28% for
the group whose members perform non-agricultural work outside the family farm.
21
Table 9: Off-farm work and participation at the household level, n = 282
Category of off-farm work
(1) No off-farm work
(2) Off-farm agricultural work
(3) Off-farm non-agricultural work
All Hutu households
#
103
96
83
At least one perpetrator
#
%
12
11.7
15
15.6
26
31.3
Total households
282
53
18.8
Chi-Square Test of equality between households with and without perpetrators
Value
Person Chi-Square
Number of Valid Cases
Degrees of
Freedom
2
12,616
282
p
0.002***
*** significant at the 1% level
The interpretation of Table 10 can be further refined when we look at the composition of the
households of perpetrators and non-perpetrators. Compared to their number in the whole
sample, heads of households and sons living in female headed households (thus acting as
heads) are particularly well-represented among perpetrators with off-farm jobs. Together, they
account for 25 of the 31 perpetrators with off-farm jobs.
Table 10: Off-farm work and participation at the individual level, n = 460
Category of off-farm work
(1) No off-farm work
(2) Off-farm agricultural work
(3) Off-farm non-agricultural work
Individual participation in genocide
All Hutu adult males
Perpetrators
#
#
%
333
28
8.4
60
12
20.0
67
19
28.4
Total
460
59
12.8
Chi-Square Test of equality between perpetrators and non-perpetrators
Value
Person Chi-Square
Number of Valid Cases
23,030
460
Degrees of
Freedom
2
p
0.002***
*** significant at the 1% level
22
Referring to the methodological part, we recall that households with no land whatsoever were not
included in the original 1989-1992 sample. Landless young people “who hang out in the
street”, often described as the core of the Interahamwe, are therefore not included in the
sample where they no longer lived with their parents. However, according to Danielle De
Lame, the situation of these landless households (not included in the sample) is similar to the
quasi-landless households (included in the sample). From our data, we can observe that 12 out
of 17 quasi-landless households with perpetrators (low landholding tertile in Table 6 rent a lot
of land (they rent more then they own). These are households that were still able to rent land
for cultivation. Other poor households, landless households, not had the chance to find a
landlord and became landless. The situation of these land-renting, quasi-landless peasants and
their household members was all but enviable, the landlord could, e. g., evict them from the
16
land at any time.
6. ECONOMETRIC SPECIFICATION AND REGRESSION RESULTS
In this section, we use a binary logit model, representing the probability of an event occurring,
in which the dummy variable y equals 1 for perpetrators of genocide and 0 for nonperpetrators. This probability of y occurring depends on a number of personal, household and
land characteristics. In addition we include dummies for clusters at the sectoral level. Formally,
17
(
)
Pr ob ( y = 1) = L ∑ k β x
k k
with x = personal, household and local characteristics
k
As determinants of the probability of participation in genocide, we use the following variables
in the regression:
Personal characteristics
Age and age squared in years;
Read and write: dummy equal to 1 for persons who can read and write and 0 for others;
Off-farm income (natural logarithm) of personally earned off-farm income in RWF;
Household characteristics
The sex of the head of the household is equal to 1 for females and 0 for males;
16
17
Danielle De Lame, Royal Museum of Central Africa, Tervuren, personal communication on the issue of
quasi-landless peasants, June 2002. Unfortunately, the data I use does not mention the kind of land
rental relationships that the farmer-households were engaged in. I therefore have no data describing
client-patron ties that determined land rentals.
The derivation and explanation of this model can be found in Liao, T.F., Interpreting probability models,
Quantitative applications in the social sciences series, Sage publications, 1994
23
The gross income (natural logarithm): the sum of income from production for own
consumption, crop sales, livestock, off-farm income and transfers;
The percent income earned from off-farm activities at the household level;
Land characteristics
Land size is measured as the area of owned cultivated land per adult equivalent;
The percentage of cultivated land rented;
Anti-erosion is the number of meters of anti-erosion ditches per are on the household farm;
Number of years cultivating
Soil quality is the C-value, a measure of protective land use (lower values indicate better
practices);
Table 11: Results of the binary logistic regressions,
dependent variable perpetrator or non- perpetrator
R1
Variables
Coefficient
R2
coefficient
R3
Coefficient
marginal
effect
Individual level
Age
Age2
Read and write
Ln off-farm y
Household level
Sex of the head
Ln gross y (ae)
% y fr. off-farm
0.2187 ***
(0.071)
-0.0028 ***
(0.0009)
-0.2972
(0.335)
0.0218
(0.0465)
1.4352 ***
(0.442)
0.5544 **
(0.229)
2.3549 ***
(0.916)
Land characteristics
Land owned (ae)
% land rented
anti-erosion eff.
years cultivating
soil quality
Constant
0.1694 ***
(0.071)
-0.0024 **
(0.0009)
-0.1114
(0.330)
0.1253 ***
(0.0438)
-10.722***
(2.394)
0.0059
(0.009)
2.9967 ***
(0.975)
0.1649 **
(0.672)
0.0232 *
(0.137)
-5.4994
(4.437)
-5.084 ***
(1.437)
0.2471 ***
(0.069)
-0.0032 **
(0.0009)
-0.2485
(0.3402)
0.0194
(0.0042)
-0.0002
(0.0006)
1.2605 ***
(0.459)
0.4544 *
(0.2587)
2.6052 ***
(0.815)
0.1478
(0.074)
0.0358
(0.020)
0.2054
(0.069)
0.0019
(0.0109)
1.9884 **
(1.015)
0.1243 *
(0.070)
0.0121
(0.014)
-2.1397
(4.588)
-10.71 ***
(2.69)
0.1568
(0.081)
0.0098
(0.005)
24
N
Log Likelihood
Pseudo R2
402
-140.25
0.14
402
-141.47
0.13
402
-136.03
0.17
*** significant at the 1% level, ** significant at the 5% level,
* significant at the 10% level; standard errors in parenthesis;
marginal effects only shown for variables with statistically significant effects
We ran three regressions, of which the third is the most complete and for which we also
computed marginal effects. Age and age squared are significant in all three, meaning that the
probability to become a perpetrator increases with age, tops at a certain age (calculated as 38)
and then declines again. As the age variable captures an individual’s place in history, it is clear
that the fate of these individuals whether or not to participate in the genocide was partly
determined by (their) history: the timing of their birth made them the children of the Hutu
Revolution (1959-1962). In 1962, most perpetrators in the 1994 genocide were small children
or were not even born. As such they have only known an independent and republican
Rwanda. They grew up as children or young adults during Habyarimana’s reign. The
alfabetisation variable, a dummy variable capturing the capability of writing and reading proves
insignificant in the regression. Being literate does not reduce the probability of participation.
The amount of personally earned off-farm income has a significant effect on the probability of
becoming a perpetrator in the second regression. In the first regression there may be multicollinearity going on as the percentage of income from off-farm activities at the household
level picks up some of the effects of individual off-arm income. We introduced the individual
level variable in order to demonstrate that off-farm income earned is not just a household
level variable, it is a certain member that earns this income, a member that can be identified.
The higher the amount of one’s off-farm income (be it earned in part-time or full-time
employment) the higher one’s chances of participation in the genocide. The effect of off-farm
incomes was already observed in the descriptive statistics in previous sections. In Rwanda, offfarm income was the most important source of monetary income. In the first regression, offfarm income (as percentage of household gross income) proves very significant. As a
consequence, and to avoid multi-collinearity, we have left individual off-farm income out in
18
the third regression.
At the household level, the sex of the head of the household and the value of gross income
play a significant role. Males (since we have excluded women) of households headed by
18
The inclusion of a dummy variable for off-farm work (y/n) in stead of the level of off-farm income
yields a significant result, but is ultimately less informative then the level of income of off-farm work.
A variable discriminating between agricultural and non-agricultural off-farm work was not significant.
25
women had a higher chance of becoming perpetrators. In practice, the oldest male member is
the de facto head of the household in this kind of household. As the descriptive statistics of
households with perpetrators showed, often the heads of households or the (oldest) sons in
female headed households have participated. The gross income of the household also had a
significant and positive effect on the probability of participation, indicating again that income
poverty is not the explanation for participation, on the contrary one would say. Care should be
given to the relative weakness of the marginal effect of this variable, we should not overstate
its explanatory power. If income raises by 100%, the probability to participate raises by 3%.
The landholding variable is not significant in the regressions. This is a surprising result. Given
the importance of land in Rwanda, one could expected the size of landholdings (in adult
equivalents) to have a significant effect on participation. The land question, however, is more
complicated. It is not the lack of land to cultivate per se that is important in the profile of the
perpetrators in our sample, it is the status of that land. The rented land variable is highly
significant and it has a strong marginal effect. An increase of 1% in the size of rented land
relative to owned land, increases the probability of participation by 15%. This suggests that
people who are active in the land market, be it out of land scarcity (for quasi-landless people)
or out of opportunity (for landlords) had a higher probability of becoming perpetrators.
Another land characteristic, anti-erosion investments in the land, also proves significant (but a
weak marginal effect). Members of households who invested a lot of effort in anti-erosion
measures on their farms (measured by the number of meters of anti-erosion ditches per are)
had a higher probability of becoming perpetrators compared to farmers who invested less in
anti-erosion measures. We will return to the interpretation of this effect in section 7.
To summarise, six variables were found to significantly explain the event of becoming
a perpetrator during the Rwandan genocide. Three of the six have strong marginal
effects and the other three have weak marginal effects. It should be added that being
male is not included in these six but its relevance is self-evident. Hence we have
- being male;
- (strong marginal effects) living in a household with a high percentage of income
earned from off-farm activities, renting a lot of land for cultivation relative to its
own landholdings, having a female head of the household;
- (weak marginal effects) having a high gross income at the household level, being a
child of the Hutu revolution (middle aged) and having invested a lot in antierosion ditches on one’s land.
26
In the final part of this paper, we will try to interpret these findings.
7. POLITICAL ECONOMY AND THE PROFILE OF INDIVIDUAL
PERPETRATORS
7.1 Three rural socio-economic categories
Evidence presented in my analysis and in research by other scholars suggests that two groups
of households (and their members) began to lose their peasant condition in Rwanda before
19
the genocide. These two groups are active in the rural labour and land markets. The first are
land-poor wage workers in agriculture or low skilled jobs and the second are land-rich
employers who hire in wage workers. Table 12 provides an overview. A household is
registered in one of the three categories when at least two out of three conditions are met.
These conditions are based on the value of three variables, landholdings, percentage land
rented and off-farm income. With this proceeding, we were able to place 275 of the 281 Hutu
households with full information. The remaining six households were registered according to
the size of their landholding only.
What distinguishes the first group (quasi-landless peasants, employees) from the second group
(middle-sized farmers) is their very low size of landholdings, which forces them to earn an
income outside the family farm. Middle-sized farmers, as we defined the second group, have
enough land to produce their own food (which does not mean that they only produce for
subsistence). The difference with the third group (landlords, employers) is that the first group
only has low skilled, low paid jobs, whereas employers have highly paid jobs outside the family
farm. In a 1989 paper on inequality and off-farm work, Clay documents the emergence of a
small group of landowners and a large group of nearly landless in Rwanda.20 He writes (p. 8)
that households with small holdings tend to work off the family farm as agricultural wage
labourers, while those with larger landholdings are more likely to hold jobs as functionaries or
in commerce. André and Platteau also find this (p. 14-17) when they write that the
landholdings of households with access to regular off-farm activities was significantly
19
20
Relevant literature used to develop the arguments in this section is found in Clay, D., Kampayana,
T., Kayitsinga, J., Inequality and the Emergence of Non-Farm Employment in Rwanda, paper
presented at the Annual Meetings of the Rural Sociological Society, Seattle, 1989 – André., C and
Platteau, J. Ph., Land relations under unbearable stress: Rwanda caught in the Malthusian trap,
Journal of Economic Behaviour and Organisation, vol 34; 1998 – De Lame, D., Une Colline entre mille
ou me calme avant la tempete, Transformations et Blocages du Rwanda Rural, Musée Royale de l’
Afrique Centrale, Tervuren, 1996
Clay, D., ibidem, 1989
27
larger compared to households without such access. These authors have documented rising
Table 12 : Socio-economic Differentiation of Hutu peasants in rural Rwanda
quasi-landless middle-sized landlords
peasants,
farmers
employers
employees (peasant condition)
Three conditions
Land owned (hect.)
% land rented
off-farm income
< 0.5
>10%
1000
0.5
<10%
<1000
>1
0
>10,000
Observations
number in the sample
percent in the sample
91
32%
151
54%
39
14%
Perpetrators
number of hh with at least one perpetrator 21
% of hh with at least one perpetrator
23.0%
20
13.2%
11
28.2%
Test of statistical significance of the difference in participation between classes
Pearson Chi-Square
Number of valid cases
Value
df
6.465
281
2
p
0.039**
** significant at the 5% level
land inequality over a period of five years, with a major cause being the capacity of wealthy
off-farm income earners to purchase land. De Lame (p. 296) considers land disputes to be at
the origin of most conflicts between households, whereas access to off-farm employment
signals the existence of a powerful relation, mostly a protection offered by the local political or
commercial elite.
From Table 12, it is clear that perpetrators are over-represented in the first and third group.
The difference between classes is statistically significant at the 5% level. A household’s activity
in the labour and land market before the genocide is thus a pretty good predictor of
participation of a male household member in the genocide. If should be added that only one
Tutsi family belonged to the land-rich, off-farm non-agricultural income earners in our data.
Only three Tutsi families are land-poor, off-farm non-agricultural income earners.21 Most
21
Both when we use the mean and the median as categorisation variable.
28
Tutsi in the sample are land-poor or middle-sized families earning an income exclusively from
agriculture (and a little of livestock). When the Tutsi in the sample are land-rich farmers,
earning their income from agriculture, they are not employing workers on their farm.
Employees (off-farm workers who are land-poor) and employers (off-farm workers who are
land-rich) have in common that they lost (or were in the process of losing) their peasant
identity or peasant condition. Only farmers with access to off-farm labour could keep or
expand their land. This means that Rwandan society during the Habyarimana regime went
through a process whereby a sizeable number of households and their members left the
second category, either to join the landed or the landless socio-economic category.
These findings suggest that the interests for members of both these groups to participate in
the genocide is to be found in their respective relation to the land and labour markets. The
landlords or employers had “something to defend”, meaning their job, their land, their farm or
farm output and their overall privileged position in Rwandan society. The poor, landless group
on the other hand, whose livelihood crucially depends on the availability of off-farm low
skilled jobs (mostly working on someone else’s farm) and/or the chance to land rent from a
landlord, were in a very vulnerable position. They could expect to gain from participation: it has been
widely documented that a large number of participants, mainly the rank and file among the
perpetrators were very interested in the property of the murdered Tutsi. Among the property,
land was a much desired asset. In order to document this, we quote a document from an
22
official meeting in the commune of Bwakira (Kibuye prefecture) during the genocide. From
the documents it is clear that the Burgomaster of Bwakira commune had to devote all his time
during the meetings throughout the genocide discussing two issues: “security”, meaning the
advancement of the genocide in the different sectors and cells of the commune and
“property” left behind by the victims. At a meeting on 5 May 1994 the Burgomaster says,
“I asked the conseillers to give lists of those who died. Only…have submitted these reports. In meetings, I said
that lands must be guarded by members of cell committees. People who want to cultivate may ask permission to
do so (lending). After six months, the lands will become the commune’s property again. No one should take
those lands for theirs, or add them to their own lands. Those who cultivate must not give any money, because it
is not a rent. The crucial problem is that there is still sorghum and banana plantations in some fields. He
wonders how people can use them.”
These quasi-landless households not only expected to gain from participation, but because of
the vulnerability of their position, they also needed to protect the few things they had. They
22
Copies of these documents were made available by Alison Desforges from Human Rights Watch. She
and her colleagues found and traced these kinds of documents after the genocide in Rwanda.
29
could be deprived of their land, houses or even lives by decisions by the powerful people who
wanted to carry out the genocide. As Alison Desforges puts it “during this period when the
guy with the gun was the one who gave the orders, the poor and weak – who had no way to
23
get a gun – had precarious little means of defence except to join the strong.
The interpretation of “something to defend” for the local elite and “everything to gain” combined
with “economic and social vulnerability” for the poor must be confronted with the “obeying the
24
regime” interpretation. The government indeed demanded a high degree of conformity and
obeisance of its population. The regime forced farmers to dig anti-erosion ditches. Since this
programme had to be implemented nationwide without taking account of local conditions, it
was resented by farmers (Guichaoua, 1991, p. 562). As with coffee policy, however, farmers
who did not believe in governmental programmes refused to take part in them (especially
from 1988 onwards), or even destroyed previous achievements. It is therefore reasonable to
assume that
-
since constructing anti-erosion ditches is very difficult and hard work and, depending on
local conditions of field and soil quality, they had something to defend after they build them;
farmers who did implement anti-erosion measures may also have performed this task to
avoid administrative fines.
Hence, peasant participation in genocide may be understood as complex behaviour whereby
poor people expected to gain something, but in addition hoped to preserve what they already
had. This behaviour is very similar to the performance of Umuganda, the digging of antierosion ditches and the cultivation of coffee: peasants hoped to benefit from their
participation in these activities and at the same time they wanted to avoid sanctions for
refusing to participate.
7.2 The importance of the rural off-farm labour and the land market
As to why access to off-farm labour is, together with access to land, according to my research,
key to the explanation of participation in genocide in our sample, we have to look at the
importance of off-farm labour in a poor rural society as Rwanda in the early nineties. The
23
24
Desforges, A., personal communication, December 2002. In the author’s doctoral dissertation, he tries
to understand the logic of participation in genocide, using recent developments in the theory of political
economy such as developed in Bardhan, P.(1997), Method in the Madness, a political economy analysis
of ethnic conflicts in less developed countries, World Development, vol. 25, no. 9.
The notion of vulnerability can also be invoked to interpret the findings on the effect of sex of the
household head. Young males in female headed households had no elder to protect them and hence
were more subject to promises or pressure from powerful persons in the community.
30
average household in the early nineties had very few sources of monetary income: the sale of
crops and banana beer, the occasional sale of livestock, the coffee harvest and income from
off-farm work. In each of the Gitarama, Gikongoro and Kibuye prefectures, off-farm income
is the most important contributor to monetary income, followed by (in that order) crop sales
(including coffee), beer sales and livestock sales. From the data it is very clear that income
from coffee declined dramatically in all three prefectures between 1989 and 1991, namely by
50%. This situation did not improve in 1992 and 1993; on the contrary. This decline has an
effect (ceteris paribus) on the demand for banana beer, which has to be paid for in cash. As a
result, banana beer sales decline, also depriving households of this source of income.
As in other poor economies, jobs offering high-income security depend on political loyalty.
This linkage has been mentioned often by scholars of Rwanda (and other African countries).
It is not a coincidence then that in our data, of the three male Hutu who have a full-time offfarm, non-agricultural occupation, two participated in the genocide (these three of course also
owned land, otherwise they would not be included in the sample). Granted that our absolute
numbers are very small, this relative number (66%) is very high. We recall that we draw our
data from the rural areas where full-time off-farm non-agricultural jobs are in short supply.
We also recall that this data is the result of a random sample of farm households in rural
Rwanda, drawn before the genocide. Moreover, the three jobs we are dealing with involve two
businessmen and a policeman. The policeman and one of the businessmen participated in the
killings of Tutsi civilians in their communes. This evidence is corroborated with other data
from people holding part-time off-farm jobs. They too show higher-than-average participation
figures. In effect, it is not exaggerated to speak of a labour market for participation in
genocide. Other scholars, such as Des Forges (1999) have shown that the local elite (a sociopolitical description of off-farm, landed, non-agricultural income earners) was particularly
active in the genocide. She even writes, very revealingly, that one of the first decisions of the
interim-government (set-up after the murder of president Habyarimana) was to pay the wages
of the heads of cellules (the lowest administrative unit) that had not seen their wage for several
months. This payment is a clear political indication that wages (off-farm non-agricultural
incomes) play an important role in perpetrator profiles. These incomes namely tie one’s
economic well-being to the source of that income, which, in Rwanda, was very often a
political or a powerful source.
One should add to this that the elite is very much aware of the political basis of its income
sources. Political activities do produce public goods, but they also secure positions of power,
giving the local elite access to interesting economic opportunities. In this sense, the situation
of the early nineties, both politically and economically, did threaten the economic well-being
31
of the local elite. Participation in genocide of this local elite thus had a very clear political
economy element. Other authors as well have highlighted the importance of off-farm incomes
in Rwanda. Marijsse (1994) has argued that the monetary incomes of a rural population in
Butare has declined by 35% between 1990 and 199225. Guichaoua, in his well-known twovolume book on agrarian issues in Rwanda, explains the importance of off-farm income in
terms of the access it offers to money. Indeed, in a poor rural economy, where subsistence
production accounts for a large part of consumption, monetary income is the only way to
consume items that are not produced on the farm: bottled beer, radios, bicycles, Europeanstyle clothes, etc.). A household also needs money to pay taxes, to send children to school and
to pay for house repairs26. A decrease in off-farm monetary income necessarily implies a
reduction in the expenses for bottled beers, other “luxury” items and possibly education and
medicine.
According to Jean-Paul Kimonyo (2000), the Habyarimana regime, in time of economic crisis,
succeeded in the preservation of the incomes (wages) of the employees of the administration27.
Kimonyo argues that this is evidence that the elite did not experience economic stress. We
believe this interpretation of the effect of economics is not correct. Firstly, communal reports
show that communes had to dismiss workers because of lack of funds, which makes it
possible to keep the wages of the remaining employees at the same level as before the budget
cut. Secondly, the local off-farm working elite may not have been hit hard itself by the
economic crisis in the 1988-1994 period and the civil war period 1990-1994, but this had a lot
28
to do with generous foreign aid. In order to benefit from this, one had to have very good
connections with political power holders.
In this and previous sections of the paper, we have argued that “economic” explanations of
participation, both of landless or poor and of landed or rich people should be regarded in a
political (or better political economy) framework. If, on the one hand wages for government
job holders did not decline during the period, the “ something to defend thesis “ which is one way
to understand an economic explanation of participation, holds true. Pointing out the relevance
25
Marysse, S. (a.o), Rwanda, Apprauvissement et Ajustement structurel, Cahiers Africains, CEDAF, 1994, p. 43.
26
Guichaoua, A., Destin Paysan et Politiques Agraires en Afrique Centrale, Tome 1, 1989, pp. 84-87
27
Kimonyo, J.P., Revue critique des interprétations du conflit Rwandais, Cahier n° 1, Centre des
Gestion des Conflits, Université Nationale du Rwanda, 2000, p. 38.
28
It is generally accepted that foreign aid at the end of the eighties and in the early nineties increased
strongly and allowed the Rwandan elite to maintain the same lifestyle (see Uvin, 1998, p. 91)
32
of “economic” variables does not mean that one has to blame the poor. It means looking at
interests and incentives: to whom do I owe my current job, wealth and status and what kind of
behaviour is required to continue or even improve my relationship with this person or group
of persons. If, on the other hand, some people lost their job, they had nothing to lose and
everything to gain and may have acted out of grievance, a wish to restore a previously
favourable position.
29
29
At this point, talking about “wages” is too general to draw definite conclusions. More research into the
1992-1994 period could help us to find out what sections of society became poorer or richer.
33
BIBLIOGRAPHY
André, C and Platteau, J-Ph., (1998), Land Relations under unbearable stress : Rwanda caught
in the Malthusian Trap, Journal of Economic behaviour and Organisation, vol 34, 1998. First
published as a Cahiers de la Faculté des Sciences Economiques et Sociales, no 164,
January 1996
Bart, F., (1993), Montagnes d’ Afrique, terres paysannes, le cas du Rwanda, Bordeaux.
Byiringiro, F. and Reardon, T., (1996), Farm productivity in Rwanda: effects of farm size,
erosion, and soil conservation investments, Agricultural Economics, 15
Clay, D., Kampayana, T., Kayitsinga, J., (1989), Inequality and the Emergence of Non-Farm
Employment in Rwanda, paper presented at the Annual Meetings of the Rural Sociological
Society, Seattle.
De Lame, D., (1996), Une Colline entre mille ou me calme avant la tempete, Transformations
et Blocages du Rwanda Rural, Musée Royale de l’ Afrique Centrale, Tervuren.
Desforges, A., (1999), Leave None to Tell the Story, Human Rights Watch, New York.
Duncan, T., Frankenberg, E. and Smith, J., (1998), Lost but not Forgotten, Attrition in the
Indonesian Family Life Survey, Paper presented at the Conference on Data Quality in
Longitudinal Surveys, Institute for Social Research, University of Michigan, October.
Griffin, R., (1995), Fascism: a reader , Oxford University Press.
Guichaoua, A., (1989), Destin Paysan et Politiques Agraires en Afrique Centrale, Tome 1.
Guichaoua, A., (1991), Les Travaux Communautaires en Afrique Centrale, Revue Tiers
Monde, tome 32, n.127, Juillet-Septembre, pp. 551-573
Kimonyo, J.P., (2000), Revue critique des interprétations du conflit Rwandais, Cahier n° 1,
Centre des Gestion des Conflits, Université Nationale du Rwanda.
Liao, T.F., (1994), Interpreting probability models, Quantitative applications in the social
sciences series, Sage publications.
Longman, T., (1995), Genocide and Socio-Political Change : Massacres in two Rwandan
Villages, Issue: a Journal of Opinion, vol. XXIII/2.
Marysse, S. (1994), Rwanda, Apprauvissement et Ajustement structurel, Cahiers Africains,
CEDAF
Moore, B., (1993), Social Origins of Dictatorship and Democracy, Beacon Press.
34
Renton, D., (2001), The Agrarian Roots of Fascism : German Exceptionalism Revisited, The
Journal of Peasant Studies, Vol 28, No 4, July
Uvin, P. (1998), Aiding violence, The development enterprise in Rwanda, Kumarian Press.
Verwimp, Ph., (2000), Development Ideology, the Peasantry and Genocide, Rwanda
represented in Habyarimanan’s speeches, Journal of Genocide Research, November.
Verwimp, Ph., (2003), Development and Genocide in Rwanda: a political economy analysis of
peasants and Power under the Habyarimana Regime, doctoral dissertation, Leuven,
January
Vidal, C, (1991), Sociologie des Passions, Editions Karthala, Paris.
Vidal, C., (1998), Question sur le rôle des paysans durant le génocide des Rwandais tutsi,
Cahiers d’Eudes africaines.
35
