Population  2005/4

Population

2005/4 (Vol. 60)

pages200

doi 10.3917/popu.504.0401

publisher I.N.E.D

E-mail Alerts

Please provide your e-mail address to receive a Table Of Content alert every time a new issue is published.


you're reading

The Death Toll of the Rwandan Genocide: A Detailed Analysis for Gikongoro Province

byMarijke Verpoorten[*]By the same author

Marijke Verpoorten, Catholic University of Leuven, Research scholar of the Fund for Scientific Research–Flanders (FWO-Vlaanderen, Belgium).

E-mail : marijke.verpoorten.at.econ.kuleuven.be

Page 331-367
1

Though counting the number of victims of a genocide is a historical and political necessity, it is always an extremely hazardous exercise. Regarding the Rwandan genocide of 1994, none of the estimates produced until now were based on records held by the communes. Focusing on Gikongoro Prefecture, Marijke Verpoorten not only evaluates the number and the profile of genocide victims, but also describes the geographical pattern of killings. This difficult assessment is based on local population registers and census data and applies a rigorous approach which starts with a detailed analysis of the data sources. It confirms the highest death toll estimates.

2

Unrest started in Rwanda at the end of 1990, when the RPF (Rwandan Patriotic Front) started launching attacks from Uganda. Intermittent hostilities and negotiations resulted in a power-sharing agreement between the government and the RPF. But on 6 April 1994 the plane carrying president Habyarimana was shot down. Thereafter, Rwanda descended into chaos. Within hours, the military, administrators, the Interahamwe militia [1]  Interahamwe literally means “those who stand together”...[1] and ordinary people started to kill Tutsi, moderate Hutu and Hutu leaders from political parties rival to the president’s party, the MRND (Mouvement révolutionnaire national pour le développement). Simultaneously, the war between the Rwandan army and the RPF resumed. A large percentage of the population took refuge in neighbouring countries. In July 1994 the massive killings came to an end [2]  For a discussion of the causes of war and genocide...[2].

3

During and after the genocide, many tried to estimate the number of casualties, first through body counting, later using demographic data. About 20 days after the start of the genocide, Human Rights Watch reported 100,000 casualties. Just a few days later, Médecins Sans Frontière (MSF) doubled this estimate. In May 1994, Radio Muhabura, the RPF radio, talked about 500,000 persons killed, and adjusted the figure downwards to 300,000 several days later. These figures were guesses rather than estimates, since they were not based on any systematic counting (Prunier, 1998, p. 262). After the genocide, the accuracy of the estimates did not improve. The UN report of November 1994 (UN 1994) took a safe range between 500,000 and one million victims. These figures include both Tutsi and Hutu.

4

Demographic data should provide a means to estimate the Tutsi death toll more accurately. The last population census prior to the genocide was conducted in 1991. This census reported 596,400 Tutsi living in Rwanda, representing 8.4% of the population. Based on an annual population growth of 3%, the number of Tutsi would have been 650,900 at the end of July 1994 (under the no-genocide scenario) [3]  PJuly 1994 = (1 + 0.03)time span × PAugust 1991; time...[3]. The next step is to obtain an estimate of surviving Tutsi. At the end of July 1994, head counting in refugee camps resulted in an estimated 105,000 Tutsi survivors. According to Prunier (1998, p. 265) 25,000 survivors who did not go to camps should also be added. Human Rights Watch (HRW, 1999, p. 15) adds another 20,000 surviving Tutsi in Zaire (now Democratic Republic of the Congo) and Tanzania. This gives a total of 150,000 Tutsi survivors. By subtracting the number of survivors from the estimated Tutsi population under the no-genocide scenario, we obtain an estimate of 500,900 Tutsi killed in the genocide, a loss of 77.0% of the Tutsi population of Rwanda.

5

Many readers may question the estimates regarding the number of survivors. Indeed, we will probably never really know how many Tutsi managed to survive without seeking refuge in camps, and counting in the camps was also prone to error. Moreover, since it is no longer politically correct in Rwanda to talk about ethnicity, the latest census of 2002 does not provide information on the current size of the Tutsi population. Another problem is the reliability of the 1991 census. Two criticisms have been put forward. First, to avoid discrimination, an undetermined number of Tutsi registered as Hutu. Second, the Habyarimana regime is said to have deliberately under-reported the number of Tutsi in order to keep their school enrolment and public employment quotas low. Until now, this allegation has not been documented. However, this strong suspicion alone led G. Prunier (1998) to estimate the actual Tutsi population in 1994 at 12% of the total population instead of 8%, the estimate put forward by the Habyarimana regime. If we repeat the same exercise as above with this larger proportion, the death toll of the genocide increases from around 500,000 to some 800,000 Tutsi killed (Prunier, 1998, p. 264), representing the annihilation of about 84% of the Tutsi population in 1994.

6

A number of Hutu also lost their lives during the genocide and its aftermath. Moderate Hutu were killed, especially in the first few weeks of the genocide. They were often adversaries of the MRND and the CDR (Coalition pour la défense de la république), the two political parties that most fanatically supported the idea of “Hutu Power”. No figure has yet been put forward to quantify these killings, but the order of magnitude is likely to be small because, from 12 April onwards, political leaders of the MRND and the CDR tried to unify all Hutu in a common fight against Tutsi. Once the Tutsi population had been defined as the common target, murders of political opponents and moderate Hutu slowed down (HRW, 1999, p. 201). However, many Hutu died in refugee camps in Zaire, Tanzania and Burundi. The cholera epidemic in Goma is believed to have taken around 30,000 lives (Prunier, 1998, p. 303). Also, an unknown number of Hutu were killed by the RPF, both in the course of combat and in acts of revenge after the genocide. Estimates range between 6,000 and 60,000, but there is no data to confirm these figures (Prunier, 1998, p. 324; HRW, 1999, p. 16) [4]  Using population data from various sources, Reyntjens...[4].

7

In short, estimating the victims of the war and the genocide in Rwanda is a hazardous enterprise. The counting of Hutu and Tutsi casualties during the war and genocide was not systematic and the analysis of demographic data is hampered by two unanswered questions. First, how many Tutsi lived in Rwanda prior to the genocide? Second, how many Tutsi survived? The present study does not answer these questions, but it contributes to the debate on the death toll of the genocide in two ways. A first objective is to provide evidence for the under-reporting of Tutsi in the 1991 census. A second objective is to analyse the geographical pattern of killings during the genocide and estimate the survival chance of Tutsi in Gikongoro, a prefecture located in southern Rwanda (Map 1, Appendix I).

8

The first contribution is made by comparing the 1991 census data with population data of 1990 from the local administration of Gikongoro. The second contribution takes these latter data as a starting point. The 1990 local population data provide information on the proportion of Tutsi in 117 administrative sectors of Gikongoro Prefecture. These data are compared with local population data of 2002 for the same 117 administrative sectors. Since the local administration stopped recording ethnic identity after 1994, the death toll among Tutsi living in these sectors cannot be derived from a mere comparison between the number of Tutsi in 1990 and 2002. However, much can be revealed from a study of the sector’s population growth between 1990 and 2002, and its sex ratio in 2002.

9

First, we present the local population data of Gikongoro Prefecture, discuss their reliability, and compare them with the census data of 1991. Next, a nonlinear least square (NLS) regression analysis is used to explain population growth and changing sex ratios across the 117 sectors of Gikongoro, and estimate the proportion of Tutsi killed during the genocide. We will examine what factors explain the unequal impact of the genocide in Gikongoro. Lastly, we present in our conclusion a new and better substantiated estimate of the number of Tutsi who were killed during the genocide.

I - Evidence for under-reporting of Tutsi in the 1991 census

1 - The 1991 population census

10

Since its independence in 1962, Rwanda has organized three population censuses: in 1978, 1991 and in 2002. The first two censuses recorded the ethnic identity of the population. After the genocide this was no longer done. Since the census of 1991 was the last census prior to the genocide, the information it provided concerning the size of the Tutsi population in Rwanda has been used to estimate the death toll of the genocide (Prunier, 1998, p. 265; HRW, 1999, p. 15). According to this information, shown in Table 1, the Tutsi population amounted to 8.4% of the total de jure population of Rwanda.

Table 1 –  Distribution of the population of Rwanda(a) by prefecture and ethnic group in 1991
11

However, many scholars have raised and repeated the allegation that this percentage is unreliable (Prunier, 1998, p. 265; HRW, 1999, p. 15). In the early seventies, the Hutu regime introduced an ethnic quota system for the appointment of civil servants and for school enrolments. This policy required that places be distributed between Hutu and Tutsi according to their proportion in the total population (Reyntjens, 1985, pp. 501-503). As the government wanted to restrict the power held by the Tutsi, they had a motive for keeping quotas for Tutsi in education and administration low, and to this end, may have under-reported the proportion of Tutsi. This accusation has often been levelled, but never proved. In addition, there are many stories of Tutsi reporting as Hutu in the population census in order to avoid discrimination. Therefore, even the census report itself urged caution with respect to the figures on ethnicity (SNR, April 1994, p. 114):

“Given that the declaration of the ethnic group is not a simple matter in Rwanda, one may wonder whether the current proportion of Hutu is not overestimated. Indeed, some members of other ethnic groups report that they are Hutu. This practice has existed since the end of the Tutsi Monarchy (1961)”

The same reasons for caution apply to the 1978 census, according to which Tutsi made up 9.7% of the total population (SNR, 1982). But the colonial rulers had no reason to misreport the relative size of ethnic groups. From 1930 onwards, the Belgian colonial administration counted the population almost every year using sample techniques. For three years, 1933, 1952 and 1956, data on ethnicity are available. These data show much higher proportions of Tutsi than under the Hutu regime: respectively 15.3%, 17.5% and 16.6% (Mamdani, 2001, p. 98; HRW, 1999, p. 40; Reyntjens, 1985, p. 28; Inforcongo, 1959, pp. 31-40). However, this hardly constitutes evidence for the under-reporting of Tutsi in the 1978 and 1990 census. The percentages reported in 1978 and 1991 even seem very plausible if we take into account the number of Tutsi who left the country between 1959 and 1990, or who were killed during this period [5]  Von Meijenfeldt (1995) mentions that in early 1994,...[5]. Then again, this extrapolation is not very convincing since the exact number of Tutsi exiles and casualties during this period is unknown. In addition, prior to the Hutu revolution of 1959, a number of wealthy Hutu registered as Tutsi (SNR, 1994, p. 110).

2 - The population data of the local administration

12

In the administrative subdivision of Rwanda, “prefectures” are followed by “communes”, “sectors” and “cells”, which are the smallest codified administrative units, including on average 150 households. Even within the cell, households live in organized units of ten households, “nyumbakumi” or ten-house agglomerations. The structure is marked by the dominance of the national government all the way from the centre via prefectures, communes, sectors and cells to the nyumbakumi [6]  The nyumbakumi, the fifth administrative level at the...[6]. Few African countries are so well organized and use existing structures so extensively as Rwanda [7]  As a grim example of this, the organizers of the genocide...[7].

13

One of the tasks of the local administration of each commune is to file reports on its population four times a year. A personal file recording all vital events and migration on each individual aged 17 and over is kept at the commune. The nyumbakumi leaders report to the leaders of the cells, who report to the leaders of the sector, and so on. This dense administrative structure records all births, deaths, immigrations and emigrations. Prior to the genocide, in addition to the age and sex, the ethnic identity of every individual was recorded [8]  Children from an inter-ethnic marriage took the ethnicity...[8].

14

In contrast to the doubts concerning the reliability of the population censuses, there is no reason to suspect any systematic under-reporting of Tutsi in the local population data. Since everybody knows each other at the local level, it would have been almost impossible for Tutsi to deceive the local administration about their true ethnic identity. Whereas census interviewers sent by the central government were not allowed to ask for the respondent’s identity card (on which ethnic identity was mentioned), no such restriction was imposed on the local administration (SNR, 1994, p. 109). In addition, the local administration had no incentive to misreport the size of the local Tutsi population. The quota policy was centralized and implemented at the national level. Even if the local administration had been inclined to influence this policy, it could not have done so by modifying its reports. Each of the reports made by the communes was filed to the Interior Ministry in Kigali, but none of these reports were ever made public, and the quota policy remained solely based on census reports as published. By comparing census data with local population data, we could assess the extent of under-reporting of Tutsi in the population census. Unfortunately, local population data are not readily available. The present study uses local population data of only one single prefecture, Gikongoro [9]  If authorization could be obtained from the Rwandan...[9].

3 - The population data of the local administration of Gikongoro Prefecture

15

Before the administrative reorganization of Rwanda, Gikongoro was one of the country’s eleven prefectures [10]  During the administrative reform of the Rwandan territory...[10]. At the time of the genocide, Gikongoro counted 13 communes and 125 sectors (see Map 2 in Appendix I). Authorized by the Rwandan government to do research on economic issues, we were able to get local population data for 1990 as well as for 2002 from the Statistical Unit of Gikongoro Prefecture.

16

The data for 1990 concerned the de facto population, distinguished by sex and ethnic group and were complete for 117 out of the 125 sectors. The eight missing sectors were those of Rwamiko Commune. A separate page with aggregated information by commune did provide some summary statistics for Rwamiko Commune, presenting population by sex and age group, but not by ethnicity. The data of the 117 sectors were listed by commune, each time on a separate page entitled “Répartition de la population par sexe, ethnie et par secteur administratif au 31/12/1990 [11]  “Population distribution according to sex, ethnicity...[11]. Each page had a similar typewritten table. An example of such a table is given in Appendix II [12]  On eight of the twelve pages the population figures...[12]. The local population data for 2002 show the de facto population by sector. They were complete for all 125 sectors and had a similar format as the 1990 data. These data included information by sex but not by ethnic group.

17

The information derived from the local population data is summarized in Table 2. Using data at the sector level, we obtained, at the commune level, the proportion of Tutsi in 1990, the total population in 1990 and 2002, the sex ratio in both of these years and the average annual population growth between 1990 and 2002. For now, only ethnicity data listed in the first column are discussed. The other data presented in Table 2 will be analysed later.

Table 2 –  Ethnicity, population growth and sex ratio for the 13 communes of Gikongoro prefecture, Rwanda in 1990 and 2002
18

According to the local population data, 17.5% of the population of Gikongoro was Tutsi in 1990. The geographical distribution of Tutsi in Gikongoro was very unequal. The commune with the highest proportion of Tutsi was Mubuga, with around 43%, followed by Nyamagabe and Muko. Map 2 in Appendix I illustrates the ethnic distribution in more detail at the sector level. The western part of the prefecture counted many sectors with almost no Tutsi, whereas Tutsi were very well represented in the northern and eastern parts of Gikongoro. In five sectors a majority of the population was Tutsi [13]  In Yonde, Gitondorero, Rususa, Muganza and Nyaruso...[13]. In fact, the government created the prefecture of Gikongoro shortly after independence. The southern and western outskirts of the Nyanza region were attached to what is now the eastern part of Gikongoro, a highland area inhabited largely by Hutu. The aim was to weaken Tutsi influence around the former royal capital Nyanza, located in the north-western corner of Butare (HRW, 1999, p. 303).

19

This proportion of 17.5% is calculated without the data for Rwamiko. There are three reasons to suspect that the missing Rwamiko data cause an underestimation rather than an overestimation of the proportion of Tutsi in Gikongoro. First, Rwamiko was surrounded by areas with a large Tutsi population. Indeed, Rwamiko Commune borders Butare Prefecture (Map 1), the prefecture with the highest proportion of Tutsi (17.3% according to the 1991 census) (Map 2 in Appendix I). Moreover, the populations of most sectors surrounding Rwamiko were more than 40% Tutsi [14]  For example: Rususa, Buremera, Kibeho, Nyarushishi...[14]. Second, the population decline of Rwamiko Commune is striking, with a decrease from 28,240 inhabitants in 1990 to 22,929 in 2002, corresponding to an annual decline of 1.8%. Given that women have six children on average (ONAPO, 2001), a decline of this size can only be explained by exceptionally high emigration or by a high death toll. A third piece of evidence for a high pre-genocide proportion of Tutsi in Rwamiko is provided by the change in the sex ratio. For Gikongoro, on average, the sex ratio decreased from 92.5 in 1990 to 88.6 in 2002 [15]  The decrease in the sex ratio may be explained both...[15]. The decline was much more pronounced in Rwamiko, where it fell from 92.5 to 81.5 (Table 2). These three facts suggest that the proportion of Tutsi in Gikongoro would have been even higher if the population of Rwamiko had been included.

4 - Evidence for the under-reporting of Tutsi in the 1991 census

20

According to the population census of 1991, the proportion of Tutsi in Gikongoro was 12.8% (Table 1). The local population data of 1990 revealed a much higher proportion of at least 17.5%. This suggests that the under-reporting of Tutsi in the 1991 census was quite high. However, to provide solid support for this argument, we must ensure that the local population data is reliable. To this end, we compared other aspects, besides ethnicity, such as the number of men and women and the total population, with census data. If the local population data are of good quality, we should find a close resemblance with census data.

21

Table 3 shows figures for the total population, the male and female population, and the proportion of Tutsi for 117 sectors of Gikongoro. The first and second columns list the figures based on respectively the census of August 1991 and the local population data of December 1990. The third column provides an estimate of the population in August 1991 based on the local data of December 1990. In the calculations, we use an average annual growth rate of 1.8% which corresponds to the growth rate for Gikongoro Prefecture between 1978 and 1991. This rate is quite low compared to the national average of 3.1% during the same period since Gikongoro is a net-emigration region in Rwanda [16]  For more details on the choice of the annual population...[16].

Table 3 –  Local population data of 1990 versus census data of 1991 for the 117 sectors of the prefecture of Gikongoro(a)
22

The last column of Table 3 shows by what percentage the prediction based on the local population data deviates above (+) or below (–) the census data of 1991. The deviation is less than 1% for the figures in the first three rows. Thus, local administration data for total population and population by sex match very well with those from the census [17]  After this exercise of comparison we checked the local...[17]. In sharp contrast, the proportion of Tutsi differs strongly and the deviation amounts to 37%. Since the local authorities had no reason to over-report the proportion of Tutsi in their population, this is convincing evidence that the proportion of Tutsi in the census was under-reported.

23

The question remains as to whether the government under-reported to restrict the quota for Tutsi in education and civil service or whether Tutsi themselves registered as Hutu in order to avoid discrimination. It is still also unclear whether the under-reporting was systematic. It would greatly help if we could obtain local population data for other prefectures [18]  HRW has local population data by ethnicity for some...[18]. There is at least one argument supporting the idea that the under-reporting was not limited to Gikongoro Prefecture alone. In Rwanda, it is generally accepted that Butare is the rural prefecture with the highest proportion of Tutsi [19]  It is believed that Tutsi started to settle in Rwanda...[19]. Therefore it is no surprise that in the census report Butare leads with 17.3% Tutsi (see Table 1). But, having found that the proportion of Tutsi in Gikongoro is actually at least 17.5% instead of 12.8%, the ranking would be reversed unless the figures of Butare and other prefectures were also misreported.

5 - Implications for the estimation of the genocide death toll

24

In sum, the large gap between the officially reported proportion of Tutsi in Gikongoro (12.8%) and the estimate based on local population data (17.5%) confirms the suspicion that the 1991 census under-reported the proportion of Tutsi in the population. Admittedly, there is no good reason to assume that the under-reporting occurred on exactly the same scale in all prefectures. But interestingly, if the scale of under-reporting at the national level was the same as that found in Gikongoro — i.e., about 40% [20]  For Gikongoro, we found a deviation of 37% (Table 3),...[20] — the calculated death toll would be close to 800,000, in the order of magnitude put forward by Gérard Prunier, taking account of speculations about the systematic under-reporting of Tutsi in the 1991 census report (Prunier 1998, p. 265).

25

Indeed, with 40% under-reporting at the national level, the proportion of Tutsi at the national level would have been 11.79% instead of 8.4%. This would mean that there were 837,100 Tutsi in Rwanda in 1991 instead of 596,400, and, under the no-genocide scenario, 913,600 Tutsi at the end of July 1994. By subtracting the estimated 150,000 Tutsi survivors (see above), we obtain a death toll of 763,600 Tutsi, representing about 83.6% of the Tutsi population. However, there are several possible scenarios. Appendix III summarizes the results of several different under-reporting scenarios. For example, one alternative scenario would be that there was only misreporting for the prefectures that had a relatively large proportion of Tutsi. If we assume, for example, that besides the figures for Gikongoro, only the figures of the prefectures with an even higher proportion of Tutsi were under-reported by 40% (Butare, Kibuye and Kigali City), this scenario leaves the ranking of prefectures by ethnicity intact. Rwanda would count 717,300 Tutsi in 1991, i.e. 10.1% of the population. The death toll would be estimated at 632,900 Tutsi, a loss of about 80.8% of the Tutsi population in 1994.

26

The evidence presented here is far from able to answer the two questions needed to make a sound estimate of the genocide death toll. How many Tutsi lived in Rwanda prior to the genocide? How many Tutsi survived? Regarding the first question we can now safely say that the use of data from the 1991 census report leads to an underestimation of the death toll. With respect to the numbers of survivors, the estimation of an absolute number for the death toll is beyond the range of our data. However, the next section shows how the local population data can be used to estimate the average survival chance of Tutsi in Gikongoro and to study the geographical pattern of killings in this prefecture.

II - The pattern of killings in Gikongoro Prefecture

27

As a consequence of war and genocide, Rwanda has experienced population decline and a decrease in the sex ratio. For Rwanda as a whole, the annual population growth slowed down from 3.1% in 1978-1991, to 1.2% in 1991-2002. Comparing the same periods, the sex ratio decreased from 95.1 to 91.3 (SNR, 2003). These demographic changes vary across as well as within prefectures, at the commune level and at the sector level. Since Tutsi were targeted during the genocide, we expect to find that the larger the proportion of Tutsi in a sector in 1990, the stronger the sector’s population decline and the lower its sex ratio in 2002. The local population data presented in the previous section can be used to assess this relationship for the sectors of Gikongoro Prefecture. We will now study the effect, in Gikongoro, of the proportion of Tutsi in a sector in 1990 on respectively population growth between 1990 and 2002 and the sex ratio in 2002. Three questions will be addressed: what was the average survival chance of Tutsi (men/women) in Gikongoro? which sectors deviated significantly from this average? and why did Tutsi (men/women) in these sectors have a higher or lower chance of survival?

1 - The relationship between the proportion of Tutsi and population decline

Preliminary evidence

28

The local population data of Gikongoro reveal that the annual population growth rate for 1990-2002 varied across communes and sectors. For example, Table 2 shows that the communes of Mubuga, Rukondo and Rwamiko experienced a sharp population decrease during this period, corresponding to a yearly decline of respectively 4.74%, 1.16% and 1.80%. At the sector level, the variation is even more pronounced. Figure 1 shows the estimated kernel density function of the annual population growth rate for 117 sectors of Gikongoro [21]  Kernel density estimators approximate the density function...[21]. The range between the minimum and maximum annual population growth is very large. Moreover, we observe that the curve is asymmetrical.

Kernel density of average annual population growth between 1990 and 2002, for 117 sectors in Gikongoro Prefecture

Reading of y: the estimated probability of observing an annual population growth of x

Sources: Local population data (provided by the author) for 117 sectors of Gikongoro, for 1990 and 2002; SNR (2003) for the average growth rate 1978-1991
29

The shape of the density function can largely be explained by three facts. First, during the 1990-2002 period, quite a few sectors had annual population growth close to 1.8%. Second, only a handful of sectors had an annual population growth above 5%. This stems from the fact that Gikongoro has few trading centres and only one small urban hub, Gikongoro City. Finally, there were 37 sectors with a negative population growth, including six with an annual population decrease of more than 5%. This explains the bump in the left tail of the density function.

30

We expect to find that the sectors whose rates of increase are shown in the left tail of the figure were those with the largest proportion of Tutsi in 1990. A comparison between Maps 2 and 3 in Appendix I provides us with preliminary evidence. The south-eastern part of Gikongoro is an especially striking illustration of the relationship between the proportion of Tutsi in 1990 and population decline. This relationship is somewhat less evident in the centre of the prefecture, around Gikongoro City, but it is again pronounced in the north-eastern part of Gikongoro. This visual impression is confirmed by Table 4. Here the sectors were divided into five categories according to the proportion of Tutsi in the sector. The categories are the same as those used to colour the sectors in Map 2. The mean annual population growth rate clearly declines across the five categories, from 2.02% for sectors with less than 5% Tutsi to –1.10% for sectors with more than 40% Tutsi.

Table 4 –  Relationship between the proportion of Tutsi in 1990 and the mean annual population growth between 1990 and 2002 for 117 sectors of Gikongoro

Estimation of the survival chance of Tutsi in Gikongoro

31

The relationship between the proportion of Tutsi in 1990 in a sector and population growth can be used to estimate the survival chance of Tutsi in Gikongoro Prefecture. The population in sector i in 2002 can be expressed as follows:

32
33

with

34

P2002,i: the population of sector i in June 2002

35

P1990,i: the population of sector i in December 1990

36

Ti: the ratio of Tutsi to the total population in sector i in 1990

37

δi: the proportion of Tutsi killed in sector i in 1994

38

gi: birth rate (BIRi) – net emigration rate (neMIGi) – mortality rate (MORi) (besides the mortality rate of Tutsi during the genocide) in sector i

39

11.5 years: time span between December 1990 and June 2002

40

Taking logarithms and rearranging [I.1] we obtain:

41
42

with

43

the annual population growth rate in sector i.

44

Since we do not know BIRi, neMIGi, and MORi, we cannot directly derive δi from equation [I.2]. However, we can obtain an estimated value for the average proportion of Tutsi killed (δi) by estimating the following equation with nonlinear least squares (NLS):

45
46

with α a constant, equal to the average value of ln (1 + gi) over the 117 sectors. The NLS estimator for δ is consistent if the disturbances εi are independently and identically distributed with expected value zero and finite variance σ2 (Malinvaud, 1970). This condition implies that the proportion of Tutsi in 1990 in sector i is totally uncorrelated with the net immigration rate, the birth rate and the non-genocide mortality rate. Formally:

47
48

In what follows, we proceed in two steps. First, we discuss the plausibility of assumptions [A1], [A2] and [A3] and present variables that may proxy neMIGi, BIRi and MORi. If these variables are good proxies, the three assumptions can be relaxed. Second, we estimate equation [I.3] to obtain the survival rate for Tutsi in Gikongoro.

49

Four different forms of migration can be distinguished: migration in search of cultivatable land, migration to urban areas, female exogamy and distress migration in response to insecurity. The latter form may cause correlation between Ti and neMIGi. It is believed that two million people from all parts of Rwanda were displaced during the genocide. The vast majority of them were Hutu who fled in fear of the advancing RPF army. Most returned to their homes after spending some months or even years in refugee camps. But, some, especially Hutu men who took part in the genocide, did not return. If these men came from sectors with a high Tutsi population prior to 1994, Cov {Ti, neMIGi} > 0. In addition, after the genocide, and upon the death of their husband, widows may have chosen to move back to their native sector. This choice might have been motivated by the desire to move away from their assailants or the place of terror, or by the search for land [22]  Especially if a widow has no sons recognized by her...[22]. Also, upon the destruction of their houses, many Tutsi survivors were left without shelter. They might have moved to so-called imidugudu, houses provided under a government housing project. Again, the result would be that Cov {Ti, neMIGi} > 0. On the other hand, the death toll of the genocide resulted in a decline in the labour-land ratio in sectors with a high ratio of Tutsi (Ti). This may have attracted farmers from other sectors, such that Cov {TineMIGi} < 0.

50

The birth rate is the ratio of live births to the population. According to the 1991 census report, Tutsi women gave birth to fewer children (TFR = 5.82) than Hutu women (TFR = 7.40) (SNR, 1994), such that Cov {Ti, BIRi} < 0. Considering that both ethnic groups have shared a common history and lifestyle for many centuries, this difference seems quite large. Rather than stemming from different social values and cultural norms, it may result from under-reporting of births by Tutsi parents (cf. discussion in Appendix III). Ideally, one should control for the number of Hutu casualties in a sector. A relatively small number of Hutu were killed because they opposed the genocide. In addition, the RPF killed an unknown number of Hutu in the course of combat and in acts of revenge after the genocide. Moreover, acts of sexual violence were frequent during the genocide. Therefore, AIDS may have increased mortality rates after 1994, such that Cov {Ti, MORi} > 0.

51

If the three assumptions are violated, the estimate for δ (proportion of Tutsi killed during the genocide) will be inconsistent. Such omitted variable bias can be reduced by including proxies for neMIGi, BIRi and MORi. However, data at the sector level are limited. Besides Ti (the ratio of Tutsi in 1990), the available variables are: total population in 1990 and 2002, the sex ratio in 1990, the sex ratio in 1990 and 2002, and the area (km2) of each sector [23]  In 2002, the country’s surface was re-estimated with...[23]. To control for emigration trends in the sectors prior to the genocide, we will use three indicators: population density in each sector in 1990, the square of this density and the sex ratio in 1990. Density represents pressure on land and may capture both migration in search of cultivatable land and rural-urban migration in the absence of war. The sex ratio in 1990 may also be informative about future migration: a high proportion of women may be linked to a continued tradition of male emigration which gives rise to female exogamy.

52

Based on the previous discussion, we estimated several models based on equation [I.3], including the population density in 1990, the squared population density in 1990 and the sex ratio in 1990. The results of model A are presented in the first column of Table 5 [24]  We experimented with the functional form of δ in model...[24]. The estimate for the coefficient δ is 0.745. This means that about three out of four Tutsi were killed in April 1994, and the t-test shows that the coefficient is significantly different from zero.

Table 5 –  Proportion of Tutsi killed during the genocide estimated from a NLS(a) regression on the average annual growth rate (1990-2002) for 117 sectors of Gikongoro (model coefficients)

The spatial pattern of killings in Gikongoro

53

For a more in-depth analysis of the geographical pattern of killings, we plotted the 117 data points in Figure 2. The horizontal axis measures the proportion of Tutsi in 1990 (Ti). The curve represents the estimated relationship between Ti and Gi:

54

. Its slope is with δ = 0.75.

55

The dots around the curve are the error terms of equation [I.3], i.e., the deviation from the predicted annual population growth. The curve can be used as a benchmark to identify in which sectors the extent of killings significantly deviated from the mean. Especially at the right of the figure, the curve does not fit well to the data points. This indicates that the genocide was considerably more deadly in some places than in others.

56

How can the unequal impact of the genocide be explained? One answer may be that in some sectors with a large Tutsi population, the Tutsi succeeded in protecting themselves by grouping together to outnumber the attackers (as happened for example at the Catholic diocese of Kabgayi in Gitarama Prefecture). Where this strategy failed, the genocide had devastating consequences (e.g. Kibeho and Kaduha parish in Gikongoro). A second reasoning is that local authorities might have succeeded in some places in resisting the genocide, or at least in slowing it down or weakening its effectiveness. In other sectors and communes, local authorities were actively involved in organizing the killings. The 800-page report by Human Rights Watch (HRW 1999) details the events of the genocide. A comparison of their findings with our data sheds more light on the unequal impact of the genocide. For the purposes of comparison, we numbered the observation points in Figure 2, each number corresponding to a commune.

Relationship between the proportion of Tutsi in 1990 and the annual population growth rate between 1990 and 2002, estimated for 117 sectors of Gikongoro

Note: The prediction (curve) is based on the following equation: (1/11.5) × ln(1 – Tiδ) with δ (average proportion of Tutsi killed) = 0.75 and Ti = the ratio of Tutsi in 1990. The dots are the residuals of regression A (See Table 5). The numbers correspond to the sectors of the following communes: Muko (11), Musange (12), Musebeya (13), Karambo (14), Rukondo (15), Mudasomwa (16), Nyamagabe (17), Karama (18), Kinyamakara (19), Kivu (20), Rwamiko (21), Mubuga (22), Nshili (23).

Sources: Local population data (provided by the author) for 117 sectors of Gikongoro, for 1990 and 2002.
57

In Figure 2, the observations for Mubuga Commune (22) are practically all located well below the line of our benchmark, implying that the Tutsi in Mubuga Commune had little chance of surviving. The commune counted 43% Tutsi among its population and recorded an annual population growth of –4.74%. The northern part of the commune was most severely hit [25]  Kibeho, Nyarushishi, Nyarusovu, Nyarwumba and Kabi...[25]. This result is supported by the fact that assailants from Rwamiko carried the violence into Mubuga and that a massacre took place at Kibeho parish and college (HRW, 1999, p. 313). Another commune with a large proportion of Tutsi (43%) is the partly urban commune of Nyamagabe (17). Despite its high proportion of Tutsi, Nyamagabe had an average population growth of 1.0%. At first sight, one might think that urban growth in Nyamagabe would have flattened out the relationship between the genocide and population decline. But, even after controlling for rural-urban migration with the square of population density in model A, the sectors of Nyamagabe are characterized by higher growth figures than those estimated by the model [26]  However, this result depends on the square of population...[26]. Also in Muko Commune (11) a considerable number of Tutsi seem to have escaped the killings. It recorded 2.83% growth despite its large Tutsi population (35%). The three sectors that recorded the lowest population growth were those in the east, close to Musange [27]  Sovu, Gitondorero and Yonde[27]. This might have been because the Tutsi of these sectors tried to take refuge in the church of Kaduha and were slaughtered while on their way or later in the massacre at the church (HRW, 1999, p. 336). Although Rukondo (15) was largely Hutu (87%), its population declined on average by 1.16%. The sectors that recorded lower than expected population growth were those close to the border with Karambo and Musange [28]  Gikoni, Mbazi and Remera[28]. A large number of Tutsi in these sectors may have been victims of the killings at Kaduha parish in Karambo.

58

The communes discussed till now, Mubuga, Nyamagabe, Muko and Rukondo, show a large difference with respect to the benchmark. For Mubuga and Rukondo this difference points to lower survival chances for Tutsi compared to the benchmark. The opposite is true for Nyamagabe and Muko. For the other communes, the difference is not pronounced or only noteworthy for one or two sectors of the commune [29]  All sectors of Karambo closely follow the fitted line...[29]. A new model was estimated, including dummy variables for the communes (Model B, Table 5). The reference group was Karambo Commune. The estimated coefficients for Muko and Nyamagabe are positive and significant; those for Mubuga and Rukondo are negative and significant. This confirms the analysis based on Figure 2.

59

While discussing Figure 2, we pointed to some explanations for the diversity of survival rates across communes and sectors. These explanations can be studied more explicitly. There were three communes where the mayor opposed the genocide (HRW, 1999): Kinyamakara, Kivu and Musebeya. However, the mayor from Kinyamakara only opposed initially, and later even helped organize the killings in his commune (HRW, 1999, p. 340). In model C we added a dummy variable that equals one for the sectors of Kivu and Musebeya (where the mayor did not change his mind about opposing the genocide). The dummies “near Kaduha massacre” and “near Kibeho massacre” equal one when a sector is near Kaduha Parish and Kibeho Parish respectively [30]  Sectors qualified as “near” are those bordering Kibeho...[30]. We included these two locations since they were the sites of the two worst massacres in Gikongoro Prefecture, each with several thousands of dead. The dummy for the opposition of a mayor has the expected sign, but is not significant. The estimated effect of the massacres is significant and highly negative. HRW (1999, p. 338) reports the killings in Kaduha as follows:

60

“As the attacks expanded from one hill to the next and from one commune to another, Tutsi found it impossible to stay in their homes and increasingly difficult to hide with Hutu neighbours. Assailants in Muko, for example, were threatening to make Hutu protectors kill any Tutsi whom they had sheltered. First hundreds, then thousands of people from Musebeya, Muko, Karambo, and Musange Communes gathered at Kaduha parish center, in the church itself, in the adjoining schools, in the health center and in all the spaces in between. Tutsi from more distant regions, like parts of Muko, came first. Tutsi in the immediate vicinity of the church moved there only about April 14, when they were threatened with attacks by Hutu from the hills. Many Tutsi had come on their own, but some had come with the help of local officials, like those transported from Musebeya. In Muko, and perhaps elsewhere, the burgomaster had at first refused to help Tutsi flee to Kaduha, but later changed his position and began encouraging them to go there. Some survivors believe that authorities decided at a meeting at the sub-prefecture to attract Tutsi to Kaduha for one enormous massacre rather than to continue killing them in smaller numbers throughout the area. Such a decision would have been consistent with the pattern of killings elsewhere in the country.”

Conclusion

61

The genocide was severe in Gikongoro. We calculated that Tutsi in Gikongoro had on average 25% chance of surviving the genocide. Some communes significantly deviated from this benchmark. In several cases, the results for the sectors within a single commune varied considerably. We could explain part of the variation by their proximity to the location of massacres, such as those of Kaduha and Kibeho, or by the spreading of violence from one commune to another [31]  From Rwamiko to Mubuga, from Mubuga to Kivu, and from...[31]. Very few mayors opposed the genocide in Gikongoro. Those who opposed were eliminated, constrained or started to fear for their own safety (HRW, 1999, p. 303-352). Although there is no doubt that their opposition saved a number of lives, it did not succeed in preventing the killings. By contrast, the strategy used by the assailants had a strong impact. The genocide was much worse in sectors where assailants convinced Tutsi to gather in Kibeho or Kaduha Parish, after which thousands were massacred in these announced places of safety.

62

Finally, our finding of an extremely deadly genocide in Gikongoro corresponds to previous research. One of the rare studies that use household data to examine the effect of genocide found a survival percentage of only 10.7% in Gikongoro (Verwimp, 2003). This percentage is based on a very small sample of 28 Tutsi, located in five different sectors [32]  These sectors are Bitandara, Kibirizi, Gasaka, Kamegeri,...[32]. Moreover, the HRW report (1999, p. 303) based on testimonies and interviews, summarizes the genocide in Gikongoro as follows:

63

“Some of the earliest attacks as well as some of the worst massacres of the genocide took place in Gikongoro. MRDN supporters launched the violence at three points and from there spread it into adjacent areas, much as they expanded disorder outward from Kigali and its vicinity into the prefecture of Gitarama. In some communes, like Musebeya, Kivu and Kinyamakara, administrators opposed the genocide and initially drew strength from the people in their communes who refused to kill. But as prefectural authorities failed to act against the violence and national authorities pressured for more and faster slaughter, they lost power to local rivals who saw the killing campaign as an opportunity to establish or re-establish their power. The dissenters judged continuing opposition futile and dangerous and either withdrew into passivity or themselves took up the role of killers.”

2 - The relationship between the proportion of Tutsi and the sex ratio

Preliminary evidence

64

In this sub-section, the geographical pattern of sex ratios in Gikongoro are mapped and explained. Figure 3 presents the kernel density function of the sex ratio in 2002 for 117 sectors of Gikongoro. The vertical line at 92.5 represents the average sex ratio in 1990. We note from the density function that by 2002, the mean of the density function had shifted to the left, to 88.6. The density function is quite symmetrical, with a slightly larger frequency of below-average sex ratios. It spans a wide range, from below 75 to above 100. This wide range may stem from the existence of sectors with a high out-migration of men and others with a high in-migration of men, such as sectors with tea plantations. The low mean and thick left tail may be the consequences of war and genocide or of continued male emigration. From the information in Maps 2 and 4 of Appendix I, we notice that, at first sight, the effect of the genocide on the sex ratio is less clear than it was for the population growth. We observe low sex ratios in Mubuga, Rwamiko and in some parts of Rukondo, Karambo and Muko. From Table 6 we note a rather weak negative relationship between the proportion of Tutsi in a sector and the sex ratio in 2002. For a more detailed analysis, we now proceed as in the previous section, with a regression analysis and a plot of the observation points.

Kernel density function of the sex ratio in 2002, for 117 sectors in Gikongoro Prefecture

Reading of y: the estimated probability of observing a sex ratio of x

Sources: Local population data (provided by the author) for 117 sectors of Gikongoro, for 1990 and 2002.
Table 6 –  Relationship between the proportion of Tutsi (1990) and the sex ratio (2002) for 117 sectors of Gikongoro

Estimation of the survival chance of Tutsi men and women in Gikongoro

65

The sex ratio in 2002 can be written as follows:

66
67

with

68

SR2002,i: the sex ratio in June 2002 in sector i

69

PM1990,i: number of men in sector i in December 1990

70

PW1990,i: number of women in sector i in December 1990

71

gmi: male birth rate – net emigration rate – mortality rate (excluding the mortality rate of Tutsi men during the genocide) in sector i.

72

gwi: female birth rate – net emigration rate – mortality rate (excluding the mortality rate of Tutsi women during the genocide) in sector i.

73

Tmi: the ratio of Tutsi men to total men in sector i in 1990

74

Twi: the ratio of Tutsi women to total women in sector i in 1990

75

δmi: the proportion of Tutsi men killed in sector i in 1994

76

δwi: the proportion of Tutsi women killed in sector i in 1994

77

11.5 years: time span between December 1990 and June 2002

78

Taking logarithms and rearranging [II.1], we obtain:

79
80

Again, we do not have sufficient information (about gmi and gwi) to directly derive δmi and δwi from [II.2]. We therefore estimate the following equation [33]  The equation [II.3] is estimated under the constraint...[33] with NLS:

81
82

with β the average difference between 11.5 × ln (1 + gmi) and 11.5 × ln (1 + gwi) for the 117 sectors of Gikongoro. A necessary condition for consistency of the NLS estimate for δm and δw is that Ti does not correlate with neMIGi,men/women and MORi,men/women, i.e. the ratio of the net-emigration of men to the net-emigration of women, and the ratio of the mortality of men to the mortality of women. Formally:

83
84

These assumptions might be violated. On the one hand, if Ti (ratio of Tutsi in 1990) is strongly correlated with casualties of Hutu men, then Cov {Ti, MORi,men/women} > 0. On the other hand, if sexual violence during 1994 resulted in HIV/AIDS infections and other sexually transmitted diseases, then Cov {Ti, MORi,men/women} < 0. Furthermore, if many Tutsi widows moved out of their sector after the genocide, Cov {Ti, neMIGi,men/women} < 0. However, if alleged (mostly male) perpetrators of the genocide took refuge abroad and did not return, the reverse is true: Cov {Ti, neMIGi,men/women} > 0.

85

To reduce omitted variable bias, we added the population density in 1990 and the squared population density in 1990 to equation [II.3]. The latter variable may capture rural-urban migration of men. The sex ratio of 1990, which is included in [II.3] controls for inertia in the sex ratio, and may capture migration. Table 7 shows the results. In model D the estimated values for the parameters δm and δw are respectively 0.791 and 0.710 and they are significant. The link-test fails to reject the null hypothesis that the model is correctly specified. We may conclude that whereas Tutsi men had a 21% chance on average of surviving the genocide, the survival rate for Tutsi women was not much higher, only 29%. This small difference is in line with the observation of Human Rights Watch (HRW 1999, p. 296) that in 1994 assailants did not often spare the lives of Tutsi women.

86

“In many communities women and children who had survived the first weeks of the genocide were slain in mid-May. In the past Rwandans had not usually killed women in conflicts and at the beginning of the genocide assailants often spared them. When militia had wanted to kill women during an attack in Kigali in late April, for example, Renzaho (Colonel, and prefect of Kigali) had intervened to stop it. Killers in Gikongoro told a woman that she was safe because ‘sex has no ethnic group’. The number of attacks against women, all about at the same time, indicates that a decision to kill women had been made at the national level and was implemented in local communities. Women who had been living on their own as well as those who had been kept alive to serve the sexual demands of their captors were slaughtered.”

Table 7 –  Proportion of Tutsi men and women killed during the genocide estimated from a NLS(a) regression on the (log) sex ratio in 2002 for 117 sectors of Gikongoro, (model coefficients)

The spatial pattern of survival chances of Tutsi men and women in Gikongoro

87

In order to explain the geographical pattern of sex ratios, we plotted the observation points in Figure 4. The horizontal axis measures the proportion of Tutsi in a sector. The curve gives the estimated relationship between Ti and SR2002,i: , with δm = 0.79 and δw = 0.71. The dots are the error terms from equation [II.3], i.e., the deviation from the predicted sex ratio. Figure 4 shows considerable variation around the curve, with the sectors of Nyamagabe and Muko Commune lying both far below and far above the curve. Only the sectors of Mubuga Commune show a clear pattern below the benchmark. In regression E, Mubuga has a coefficient that is highly negative and significantly different from zero. This corresponds to our earlier discussion where we found that the genocide was particularly severe in Mubuga Commune. In regression F, the coefficient for the dummy of the Kibeho massacre in Mubuga Commune is significantly negative.

Relationship between the proportion of Tutsi in 1990 and the sex ratio in 2002, estimated for 117 sectors of Gikongoro

Note: The prediction (curve) is based on the following equation: (1 – Tmiδm)/(1 – Tmiδw), with δm (average proportion of Tutsi men killed) = 0.79 and δw (average proportion of Tutsi women killed) = 0.71, Tmi = the ratio of Tutsi men in 1990 and Twi = the ratio of Tutsi women in 1990. The dots are the residuals of regression D (See Table 7). The numbers correspond to the sectors of the following communes: Muko (11), Musange (12), Musebeya (13), Karambo (14), Rukondo (15), Mudasomwa (16), Nyamagabe (17), Karama (18), Kinyamakara (19), Kivu (20), Rwamiko (21), Mubuga (22), Nshili (23).

Sources: Local population data (provided by the author) for 117 sectors of Gikongoro, for 1990 and 2002.

Conclusion

88

This paper provides evidence for the allegation that the number of Tutsi in the 1991 population census was under-reported. This allegation has often been made on the basis of two arguments. On the one hand, because of its ethnic quota policy, the Rwandan government had a motive for under-reporting the proportion of Tutsi in the population. On the other hand, Tutsi themselves, trying to avoid discrimination, had a motive to register as Hutu. The evidence presented stems from a comparison between the 1991 census data and the 1990 population data from the local administration of Gikongoro Prefecture. The former reported 12.8% Tutsi in the population of Gikongoro, while the latter reported a much higher proportion of Tutsi, 17.5%. By contrast, other population characteristics, such as population size and the sex ratio hardly differ across both data sources. If the extent of under-reporting was similar in other prefectures, the number of Tutsi at the outbreak of the genocide would have been about 40% higher than the number extrapolated from the 1991 census.

89

After the genocide, the Rwandan government no longer reported ethnic identity in the population statistics. This complicates the estimate of Tutsi survivors. Using the local population data of Gikongoro Prefecture, we estimate that only 25% of Tutsi in Gikongoro survived. Some communes in Gikongoro clearly deviated from this benchmark. Especially in Mubuga and Rukondo the genocide was much more deadly. The opposite was true for Nyamagabe and Muko. The location of huge massacres and the way violence spread across communes were strong determinants of the survival chance of Tutsi in a sector. The opposition of local authorities in Gikongoro did not make a significant difference. The lives of Tutsi women in Gikongoro were not spared: it is estimated that Tutsi women had 29% chance of surviving, compared with a 21% chance for Tutsi men.

90

Gikongoro is only one of the eleven prefectures in Rwanda, accounting for less than 7% of the total population and about 10% of the Tutsi population. Extrapolating evidence from Gikongoro to the whole of Rwanda cannot therefore provide one sound estimate for the overall death toll of the genocide. However, the case of Gikongoro brings the estimated death toll closer to its true value and narrows the interval around the estimate. Based on the presented evidence and a sensitivity analysis (Appendix III), we estimate that the number of Tutsi killed during the genocide lies between 600,000 and 800,000, and that only 25 to 30% of the Tutsi population survived the genocide of 1994.

Rwanda, administrative map of 1993
The proportion of Tutsi at the sector level in Gikongoro Prefecture, 1990
Source: Service national de recensement (SNR), January 2003
Population growth over 1990-2002 at the sector level in Gikongoro Prefecture
Source: Service national de recensement (SNR), January 2003
Sex ratio in 2002 at the sector-level in Gikongoro Prefecture
Source: Service national de recensement (SNR), January 2003

An example of the population data of the local administration of Gikongoro Prefecture

Table A –  Sensitivity analysis for estimating the genocide death toll
Table B –  Sensitivity analysis assuming an annual population growth rate of 3%
Table C –  Sensitivity analysis assuming an annual population growth rate of 2%
Acknowledgements:

I am grateful to two anonymous referees and an assistant editor from this journal for their useful comments. I thank Alison Des Forges, Florence Mukamugema, Filip Reyntjens, William Seltzer and André Twagiramungu for answering many of my questions. I also received support and comments from my colleagues at the department. Special thanks is owed to Romain Houssa, Pablo Rovira Kaltwasser, Jorre Renterghem, Filip Heremans, Herman Van Belle, Geert Dhaene and Lode Berlage. All opinions expressed and any errors remain my own.


REFERENCES

  • Baines Erin, 2003, “Body Politics and the Rwandan Crisis”, Third World Quarterly 24(3), pp. 479-493.
  • HRW, 1999, Leave None to Tell the Story: Genocide in Rwanda, Human Rights Watch, New York, 789 p.
  • Inforcongo, 1959, Office de l’information et des relations publiques pour le Congo belge et le Ruanda-Urundi, Brussels, 337 p.
  • Malinvaud, E., 1970, “The Consistency of Nonlinear Regressions”, The Annals of Mathematical Statistics, vol. 41(3), pp. 956-969.
  • Mamdani Mahmood, 2001, When Victims Become Killers, Princeton University Press, Princeton, 364 p.
  • Minecofin, 2002, Integrated Household Living Conditions Survey (2000-2001), Ministry of Finance and Economic Planning, Kigali, 168 p.
  • Newbury David, 1998, Understanding Genocide, African Studies Review, 41(1), pp. 73-97.
  • Onapo, 2001, Enquête démographique et de santé, République de Rwanda, ministère de la Santé, office national de la Population.
  • Prunier Gerard, 1998, The Rwanda Crisis: History of a Genocide, Third Impression, London, Hurst & Company, 424 p.
  • Reyntjens Filip, 1985, Pouvoir et droit au Rwanda, Musée royal de l’Afrique centrale, Tervuren, 508 p.
  • Reyntjens Filip, 1997, “Estimation de nombre de personnes tuées au Rwanda en 1994”, in S. Marysse, F. Reyntjenseds (eds), L’Afrique des grands lacs: annuaire 1996/1997, Paris, L’Harmattan, pp. 179-186.
  • Samset Ingrid, Dalby Orrvar, 2003, Rwanda: Presidential and Parliamentary Elections 2003, Nordem Report 12/2003, The Norwegian Institute of Human Rights, 59 p.
  • SNR, 1982, Recensement général de la population et de l’habitat, 1978. Résultats définitifs, service national de Recensement, ministère du Plan, Kigali.
  • SNR, 1994, Recensement général de la population et de l’habitat au 15 Août, 1991. Résultats définitifs, service national de Recensement, ministère du Plan, Kigali.
  • SNR, 2003, General census of population and housing of Rwanda, 16-30 August 2002: Report on the preliminary results, National Census Service, Ministry of Finance and Economic Planning, Kigali.
  • Uvin Peter, 1998, Aiding Violence: the Development Enterprise in Rwanda, West Hartford, Kumarian Press, 288 p.
  • UN, 25 November 1994, United Nations Commission of Experts Established to Security Council Resolution 935 on Rwanda, Final Report, Geneva.
  • Verwimp, Philip, 2003, “Testing the Double-Genocide Thesis for Central and Southern Rwanda”, The Journal of Conflict Resolution 47(4), pp. 423-442.
  • Von Meijenfeldt, Roel, 1995, “At the Frontline for Human Rights: Evaluation of European Union Participation in the Human Rights Field Operation in Rwanda”, on assignment of the Commission of the European Communities.

Notes

[*] Catholic University of Leuven, Research scholar of the Fund for Scientific Research–Flanders (FWO-Vlaanderen, Belgium)

[1] Interahamwe literally means “those who stand together” or “those who attack together”. This militia was formed by Habyarimana’s political party in 1992, when the party started giving military training to its youth.

[2] For a discussion of the causes of war and genocide in Rwanda, see Mamdani (2001); Prunier (1998); Newbury (1998); Uvin (1998); Verwimp (2003); Baines (2003).

[3] PJuly 1994 = (1 + 0.03)time span × PAugust 1991; time span = 2.96 years. A discussion on the assumption of 3% annual population growth can be found in Appendix III.

[4] Using population data from various sources, Reyntjens (1997) argues that the total number of Rwandans who “disappeared” from the population in 1994, whether Hutu, Tutsi or Twa, lies between 1,050,000 and 1,150,000. Subtracting the number of Tutsi casualties from this estimate would give a reasonable approximation for the number of Hutu who “disappeared”, since Twa accounted for only 0.4% of the population prior to April 1994. However, the disappearance of Hutu stems from several causes that cannot be disentangled: (1) killed by the perpetrators of the genocide, (2) killed by the RPF, (3) died from disease in refugee camps or, (4) fled to neighbouring countries (to escape prosecution).

[5] Von Meijenfeldt (1995) mentions that in early 1994, there were about 600,000 Rwandan Tutsi living in exile.

[6] The nyumbakumi, the fifth administrative level at the bottom of the chain, is not codified by the law however (Samset and Dalby, 2003).

[7] As a grim example of this, the organizers of the genocide exploited these structures to reach all Rwandans and to incite or force Hutu into participating in the slaughter. On the role of local authorities in the genocide and “Rwandan political tradition”, see, among others, Mamdani, (2001, p. 194, p. 199 and p. 218) and Prunier, (1998, p. 138).

[8] Children from an inter-ethnic marriage took the ethnicity of the father.

[9] If authorization could be obtained from the Rwandan authorities, it would be possible to put together much of the local population data by systematically checking the reports filed at the communes, prefectures and in the Interior Ministry in Kigali.

[10] During the administrative reform of the Rwandan territory in December 2000, Umutare was added as the twelfth prefecture. The terminology also changed. “Prefectures” became “provinces”, and “communes” became “districts”. In this paper, we use the old terminology and administrative subdivisions as they were in place at the time of the genocide.

[11] “Population distribution according to sex, ethnicity and administrative commune on 31 December 1990”

[12] On eight of the twelve pages the population figures were handwritten. The mayor signed eight out of the twelve pages, four had an official stamp of the commune and on seven pages the date of the signature had been added. Although the title on each page mentioned 31/12/1990 as the survey date, the dates of signature differed. The seven dates mentioned were 17 January 1991, 19 January 1991, 16 March 1991, 14 May 1991, 19 December 1991, 27 December 1991, and 3 February 1992. The dates of signature probably correspond to the time of sending the reports from the commune to the prefecture. The delays might result from the fact that the commune had to wait for the data from each sector, and had to summarize them on one page.

[13] In Yonde, Gitondorero, Rususa, Muganza and Nyarusovu.

[14] For example: Rususa, Buremera, Kibeho, Nyarushishi and Muganza.

[15] The decrease in the sex ratio may be explained both by emigration and by casualties of the war and genocide. Gikongoro as a whole is a net-emigration region (MINECOFIN, 2002).

[16] For more details on the choice of the annual population growth rate used, see Appendix III.

[17] After this exercise of comparison we checked the local population data for outliers and found that 7 sectors had unusual sex ratios. We replaced these by the sex ratios of the census data. In two sectors, the number of inhabitants deviated strongly from the census. They were also replaced by figures from the census. Table 2 and the regression analysis of section 3 are based on the corrected data.

[18] HRW has local population data by ethnicity for some communes of Butare Prefecture and repeatedly requested other data from the Rwandan government. Since several communes of Butare are missing, HRW’s current data cannot be used for comparison with the census data of 1991 that reported the size of ethnic groups only at the prefecture level.

[19] It is believed that Tutsi started to settle in Rwanda from the 8th or 9th century onwards. Coming from the east, they first settled in the dry eastern parts of present day Rwanda, but they gradually moved westwards to the wetter highlands. In the 14th century the Tutsi feudal monarchy dominated the central parts of the country and Nyanza, in the north-western corner of Butare, became its royal capital.

[20] For Gikongoro, we found a deviation of 37% (Table 3), but we argued that the deviation would be even higher when including the sectors of Rwamiko. For example, if we assume that Rwamiko counted 28% Tutsi, Gikongoro would count 18% Tutsi and the deviation would increase from 37 to 40%.

[21] Kernel density estimators approximate the density function f(x) from observations on x. So, basically they give the probability of observing xi (population growth of sector i) in the sample. Consequently, the estimates of f(x) integrate to 1. As opposed to frequency tables, kernel density estimates have the advantage of being smooth and of being independent of the choice of origin. More formally, the data is divided into intervals and estimates of the density at the centre of the interval are produced. The intervals are allowed to overlap. The smoothness of the figure depends on the width of the interval chosen. In Figure 1, the width is set at 0.84.

[22] Especially if a widow has no sons recognized by her family-in-law, her customary right to land might be challenged.

[23] In 2002, the country’s surface was re-estimated with the help of the Information Management Unit of the United Nations Development Programme (IMU/UNDP). In December 2003, we received these data by e-mail from the SNR (Service National de Recensement du Rwanda).

[24] We experimented with the functional form of δ in model A, allowing for nonlinearities in the relationship between δ and Ti, but the alternative nonlinear models were rejected in favour of model A. Finally, we performed a model-specification link test for single-equation models. The test result fails to reject the assumption that the model is correctly specified.

[25] Kibeho, Nyarushishi, Nyarusovu, Nyarwumba and Kabirizi

[26] However, this result depends on the square of population density being a good proxy to capture migration. To verify this, we used the IHLCS (Integrated Household Living Conditions Survey) (MINECOFIN, 2002) to calculate immigration by commune for Gikongoro (The data cannot be used to calculate emigration at a communal level). The IHLCS has information on 2,461 individuals in Gikongoro. We calculated the proportion of immigrants among these individuals for each commune over the last ten years. It appears that Nyamagabe is indeed the commune with by far the largest proportion of immigrants; 9.4% of its population in 2002 were immigrants who settled in the commune between 1990 and 2002. But, even when introducing the proportion of immigrants into model A, the results for Nyamagabe remain unchanged (these results are not reported, but can be obtained from the author on request). This suggests that the square of population density does sufficiently capture immigration into Nyamagabe. We can thus be quite sure that the Tutsi in Nyamagabe had higher than average survival chances.

[27] Sovu, Gitondorero and Yonde

[28] Gikoni, Mbazi and Remera

[29] All sectors of Karambo closely follow the fitted line in Figure 2. This suggests that the genocide in Karambo was neither more nor less severe than the average of Gikongoro. The same applies to Karama. In Kinyamakara, the proportion of Tutsi was very small (3.2%), which makes it difficult to judge whether or not the genocide struck hard. The same is true for Musebeya (0.9%) and Nshili (3.7%). Tutsi were also a small minority in Mudasomwa, but they were almost all concentrated in two sectors bordering Nyamagabe (Nyamigina and Buhoro). Here, the killings appear to have been neither more nor less severe than the average of the prefecture. The eastern part of Kivu counted a sizeable Tutsi population, but the killings do not seem to have been as severe as in Mubuga. Mutovu in particular has much higher than expected population growth; maybe because it is somewhat more distant from Mubuga and Rwamiko, the communes that brought the violence into Kivu. The mayor of Kivu tried to oppose the genocide (HRW, 1999, p. 303) and he might have achieved some effective opposition in these communes. Finally, in Musange, the killings seem to have been in line with the average. One sector performing worse was Joma, at the crossroads between Karambo and Rukondo. This ties in with the fact that attacks in Musange were being launched from Karambo (HRW, 1999, p. 313).

[30] Sectors qualified as “near” are those bordering Kibeho and Nyakiza sector and those not further than one sector distant from Kibeho and Nyakiza sector.

[31] From Rwamiko to Mubuga, from Mubuga to Kivu, and from Karambo to Musange (HRW, 1999, p. 313).

[32] These sectors are Bitandara, Kibirizi, Gasaka, Kamegeri, Kigoma and Gorwe.

[33] The equation [II.3] is estimated under the constraint that δw = [δ – δm × (Tutsi men/total Tutsi)] / (Tutsi women/total Tutsi).

Abstract

English

The death toll of the Rwandan genocide remains highly debatable. The frequently quoted estimate of 500,000 Tutsi killed is based on the population census of 1991. However, two unanswered questions make this estimate unreliable. First, how many Tutsi lived in Rwanda prior to the genocide? Second, how many Tutsi survived? With respect to the first question, critics say that the proportion of Tutsi was under-reported in the 1991 census. By comparing the census data with population data of the local administration of Gikongoro Prefecture, we provide evidence for this allegation and study how the under-reporting may affect the estimate of the genocide death toll. We also use local population data for 117 administrative sectors within Gikongoro Prefecture to make a detailed analysis of the spatial pattern of killings in Gikongoro. We find that Tutsi in Gikongoro had, on average, a 25% chance of surviving the genocide. The survival rate for women was only slightly higher than for men: 29% versus 21%. The location of huge massacres and the way violence spread across sectors were more decisive for the Tutsi survival rate than whether or not local authorities opposed the genocide.

Outline

  1. Evidence for under-reporting of Tutsi in the 1991 census
    1. The 1991 population census
    2. The population data of the local administration
    3. The population data of the local administration of Gikongoro Prefecture
    4. Evidence for the under-reporting of Tutsi in the 1991 census
    5. Implications for the estimation of the genocide death toll
  2. The pattern of killings in Gikongoro Prefecture
    1. The relationship between the proportion of Tutsi and population decline
      1. Preliminary evidence
      2. Estimation of the survival chance of Tutsi in Gikongoro
      3. The spatial pattern of killings in Gikongoro
      4. Conclusion
    2. The relationship between the proportion of Tutsi and the sex ratio
      1. Preliminary evidence
      2. Estimation of the survival chance of Tutsi men and women in Gikongoro
      3. The spatial pattern of survival chances of Tutsi men and women in Gikongoro
  3. Conclusion

To cite this article

Marijke Verpoorten "Le coût en vies humaines du génocide rwandais : le cas de la province de Gikongoro", Population 4/2005 (Vol. 60).
URL  www.cairn.info/

Back to Top

Feedback

Why is this article not available in English?

Cairn International Edition is a service dedicated to helping a non–French–speaking readership to browse, read, and discover work published in French journals. You will find English full–text translations, in addition to French version already available on Cairn regular edition. Full text translations only exist for a selection of articles.

If you are interested in having this article translated into English, please enter your email address and you will receive an email alert when this article has been translated.

Message sent

Your email address has been saved.
We will notify you when this article becomes available in English.