In a recent blog Prof. Marijke Verpoorten at the University of Antwerp (which also houses her colleague Filip Reyntjens) raised some interesting questions regarding the lower range of GenoDynamics' Tutsi casualty estimation, which was approximately 200,000. To begin, Verpoorten does not disagree with our higher or median values - just the lower one. Indeed, as I will discuss further below, she actually uses our estimations quite prominently in her research (2012).
Verpoorten's contribution to the discussion is interesting and, indeed, useful because it highlights some important factors regarding casualty estimation. It also identifies some of the weaknesses. Verpoorten's piece is useful for it identifies something that is essential for all efforts at casualty estimation:
Many of these themes were addressed in my book Media Bias, Perspective and State Repression (2010).
With these ideas in mind, Verpoorten raised concerns about the lower range of our estimation. Essentially, to generate this figure, we:
We thus trusted that the government of Belgium to conduct a census in Rwanda - the place where governed for a while and the one where they assisted in solidifying the differences between ethnic groups in the first place. Given the precarious nature of the Tutsi-led government, who governed for the Belgians and were in the numerical minority, it is clear that the Belgians had a vested interest in identifying everyone by ethnicity in order to get a better understanding of what their situation was. These actually end up overlapping with population estimates that use 1978 and 1991. In addition to this, we trusted the UN and we trust UNHCR. In part, this is because of the second and third reasons above (i.e., discussions of methodology/transparency and this is one of the tasks that they have undertaken for decades).
Before moving forward, it must be noted that there are some factual errors in Verpoorten's contribution regarding our work. For example, the census we used was from 1958 not 1952; and we have a negative growth rate following 1990/1 not a positive one, which is standard within conflict situations (i.e., interstate war between the Rwandan Patriotic Front who were invading from Uganda and the then Rwandan government). There are also some differences that we have regarding the number of Tutsi saved: Verpoorten tends to believe a figure of 100-150,000 saved taken from some unnamed "Aid organizations" while we believe that a figure around 300,000 is more accurate, taken from a Tutsi survivor's association (Ibuka) and the Rwandan government's survey of genocide survivors.
It is also worth noting how puzzling it is that Verpoorten would express problems with our work for in her own research and in direct correspondence with GenoDynamics she expressed nothing but respectful collegiality and appreciation for our willingness to share data as well as discuss the topic. Indeed, in her 2012 piece ("Detecting Hidden Violence: The spatial distribution of excess mortality in Rwanda") Verpoorten relies upon our casualty estimate of all violence in order to compare against the measurement she created, which as was correctly identified was one of the only estimates produced by some rigorous and transparent method. As a result, there is no fundamental problem that Verpooten has with our work up until the date of the recent blog and, in fact, she presents a rather straightforward account of our research when she states that
Davenport and Stam (2009) apply a Bayesian latent variable model to the data in order to obtain
estimates on the number of killings taking place in 164 administrative communes during AprileJune
1994. The sum of victims in communes under the jurisdiction of the FAR is estimated at
approximately 890,000, which adds to the evidence of a high death toll under the jurisdiction of the
genocidal regime. Davenport & Stam (2009) use the same approach to estimate the number of
killings that took place in 1994 within the zones under RPF controland the zones contested by the
RPF and FAR.. They arrive at estimates of respectively 80,000 and 90,000, which confirm previously
made allegations of 1994 human rights violations, in particular reprisal killings, at the address of the
RPF (Des Forges, 1999; Prunier, 1998, the Gersony report).
That was all she said about it. There was no fatal flaw to our operationalization of total casualty figures. There were differences but nothing that would result in casting the measure aside.
As for the particular comments made in the recent blog, it appears that Verpoorten's problem lies with the lower range of our estimation regarding how many Tutsi were killed in the territory under the control of the FAR. For the record, this includes genocide but it also includes other forms of violence as well - something Verpoorten's (2012) article explicitly investigates. The higher range is fine and indeed overlaps with the estimates that she provides (i.e., our estimation ranges overlap).
Central to Verpoorten's critique is the issue of the 1991 census that was conducted by the Rwandan government. Why do we keep discussing this census? Well, partly this is because we cannot go forward. Citing fears of divisionism, ethnicity (accept for explicit discussion of the "genocide of the Tutsi") all references to ethnicity have been banned (including the census). We also keep going back to this source because to date there has been no list produced regarding victims of the mass killing and without a list with all the potential victims it is hard to understand what took place as well as why. Genocide, for example, requires ethnic targeting and the census identifies ethnicity thus making it a natural location to look. Lists were read over the hate radio stations but the sheer number of names read did not come close to the number commonly believed to be killed. Lists were believed to be handed out to perpetrators but none have been recovered. The census or something like it thus proves crucial to those of us trying to understand what took place in Rwanda during 1994. For example, with a good census and the identification of everyone by ethnicity, then this information would be used to find and kill everyone. If the census was good though, then the casualty count for Tutsi would be low according to Verpoorten, relative to the number she believes to be present. But, if the census was bad and it did not identify everyone by ethnicity properly, then the high casualty count based on ethnicity identification is problematic. That is, unless there was another source that was used.
Verpoorten's (2005) work on one geographic location (GIkongoro) is the source she trusts to make her case. This was a population estimate provided by local authorities whom she trusts were more accurate in identifying individuals ethnicities than authorities from the national government office. If this information is good, then this would account for a larger number of Tutsi deaths. If the information is bad, however, then this suggests that the killing was unclear. Some interesting assumptions exist here.
First, it is presumed that the list created by the local administrators regarding who should be targeted was the one that was used when people were being killed. No one has provided evidence to this effect and there are numerous reports of people coming from the outside a community to kill people. What list would they have used? Were lists discarded and people in the community pointed out those to be targeted? To answer these questions, we need detailed information about who did what to whom in each locale, not simply population figures taken before and after overt violence.
Second, the approach Verpoorten uses assumes that the situation in Gikongoro applies to ALL of Rwanda or at least parts of it. Indeed, she explicitly states that this is problematic when she remarks that "Admittedly, there is no good reason to assume that the under-reporting occurred on exactly the same scale in all prefectures" (2005: 341).
Third, the approach Verpoorten uses (in 2005 and in her blog) assumes that individuals were killed where they came from. But, as Verpoorten acknowledges in her 2012 piece
The refugee crisis was nationwide, but more intense at the borders with RD Congo and Tanzania,
two neighboring countries that hosted an estimated 1,200,000 and 580,000 refugees, respectively.
Burundi also hosted a considerable number of refugees (270,000) while Uganda hosted no more
than 10,000 refugees (UNHCR, 2000, chap. 10). Both the proximity to a refugee camp and its size
are likely to affect excess mortality (51: fn 3).
In addition to this, there is the issue of internally displaced persons (IDPs) which might have reached another 2 million, if not more. This means that somewhere between a half or more of the population was on the move during the violence of 1994.
Movement is an especially problematic issue in the calculation of casualty estimation for it raises some questions that are rarely asked in Rwanda. For example,
As a community of people interested in Rwanda and the violence of 1994, we have very little systematic answers to these questions but they are crucial for movement tends to undermine and invalidate many of the discussions that take place (something also noted in Verpoorten's 2012 study as she attempts to figure out how many died in the different forms of killing that occurred: civil war, genocide, (counter)insurgency and the refugee crisis.
Related to this, movement is problematic because it makes one think a bit harder about how people were killed and this is a level of depth that most do not wish to engage in (for obvious reasons). When one does this, however, it is clear that most assume that killers are traveling around with lists of people to kill most likely derived from the census or local administration. But, this requires a high degree of coordination, literacy and copying machines - all of which are in short supply, especially as the existing government were getting destroyed on the battlefield and the Rwandan population was fleeing. What good is a list if people are running and hiding?
All of this discussion places a tremendous amount of confidence in Rwandan abilities to identify ethnic others. But, this is not clear. Indeed, this might explain why people had id cards. If ethnicity was so obvious, then these would be unnecessary. But, if you were not sure who somebody was, then you would need some id. Historical researchers note the significant amount of intermarriage that took place between Tutsi and Hutu. What was the child of a Hutu man and Tutsi female? What was the child of a Tutsi man and Hutu female? Were all Hutus identified as Hutus? Were all Tutsis identified as Tutsis? Do answers to these questions vary across time and place? Were mistakes made and under what circumstances? These issues clearly complicate the identification and targeting of ethnic victims as the violence takes on an air of variability across individuals and contexts. This complexity appears to be evident in eyewitness testimony from the ICTR regarding what took place in 1994: 1 and 2 (these are two examples from around 7000 that were compiled by the ICTR).
The Rwandan government has not helped with assisting individuals in identifying which specific groups were killed. Indeed, the research conducted by the government on the topic reveals the difficult in obtaining an actual count of casualties by ethnicity. According to the Ministry of Local Governments own survey of genocide victims, the complexity identified above is quite clear. As they state in their report: "genocide victims are all persons which have been killed in the period between 10/1/1990 and 12/31/1994
This is a rather broad conception of victims and one that likely pushes beyond the traditional definition of genocide. Indeed, Article II of the Convention on the Prevention and Punishment of the Crime of Genocide defines the relevant political violence as any of the following acts committed with the intent to destroy, in whole or in part, a national, ethnical, racial or religious group, as such:
Clearly, the categories of genocide victim used by the Rwandan Ministry above only partially fits the definition for genocide identified. Being a "friend" or having an "affinity with" someone from a group is not the same thing as being from the group. Actually, being a "friend" or having an "affinity with" likely represents Hutus for if they were Tutsi then they would have been classified as ethnic Tutsi, which is the first category. Having a "lineage with a Tutsi" presumes that this is the reason why someone would be targeted and that the perpetrators know this. It is possible that someone might see themselves and be perceived by others around them as Hutu but through lineage they would later be counted as Tutsi. Finally, there are those that because of their belief against divisionism, hatred and violent policies would be placed in the category of genocide victim. As this concerns political beliefs, this clearly moves past the strict categorization of the term genocide. Again, this also includes Hutus.
All of this raises questions regarding the identification of Tutsi victims. It may be that the research from GenoDynamics is capturing 200,000 ethnic Tutsis that are clearly and consistently identified as such and that the other victims we are identifying are those fitting into the other categories highlighted by the Rwandan government which includes both Tutsi as well as Hutu victims. Rather than compel a reduced conversation, therefore, our estimation and the issues identified above compel a deeper, wider and more thorough conversation about who died, why and who was responsible that allow us to differentiate between the categories of victims identified above.
Verpoorten's contribution to the discussion is interesting and, indeed, useful because it highlights some important factors regarding casualty estimation. It also identifies some of the weaknesses. Verpoorten's piece is useful for it identifies something that is essential for all efforts at casualty estimation:
- all estimation must make numerous assumptions.
- good casualty estimation will discuss these assumptions, why they were made, run alternative scenarios, discuss which values are most reasonable and make all data behind these estimations available (if this is not done, then don't bother reading further).
- all casualty estimation must put their trust in some source (invariably all data come from somewhere and to continue on the topic, one will have to trust that a source did something appropriately).
Many of these themes were addressed in my book Media Bias, Perspective and State Repression (2010).
With these ideas in mind, Verpoorten raised concerns about the lower range of our estimation. Essentially, to generate this figure, we:
- established a baseline for the Rwandan population divided by ethnic groups, which was undertaken by the colonial powers at the time along with the Tutsi-dominated government that administered the colony (This was from the census of 1958 which was believed to be the least politicized of Rwanda's censi);
- we selected a reasonable growth rate (from the UN); and,
- we subtracted the number of refugees (from UNHCR).
We thus trusted that the government of Belgium to conduct a census in Rwanda - the place where governed for a while and the one where they assisted in solidifying the differences between ethnic groups in the first place. Given the precarious nature of the Tutsi-led government, who governed for the Belgians and were in the numerical minority, it is clear that the Belgians had a vested interest in identifying everyone by ethnicity in order to get a better understanding of what their situation was. These actually end up overlapping with population estimates that use 1978 and 1991. In addition to this, we trusted the UN and we trust UNHCR. In part, this is because of the second and third reasons above (i.e., discussions of methodology/transparency and this is one of the tasks that they have undertaken for decades).
Before moving forward, it must be noted that there are some factual errors in Verpoorten's contribution regarding our work. For example, the census we used was from 1958 not 1952; and we have a negative growth rate following 1990/1 not a positive one, which is standard within conflict situations (i.e., interstate war between the Rwandan Patriotic Front who were invading from Uganda and the then Rwandan government). There are also some differences that we have regarding the number of Tutsi saved: Verpoorten tends to believe a figure of 100-150,000 saved taken from some unnamed "Aid organizations" while we believe that a figure around 300,000 is more accurate, taken from a Tutsi survivor's association (Ibuka) and the Rwandan government's survey of genocide survivors.
It is also worth noting how puzzling it is that Verpoorten would express problems with our work for in her own research and in direct correspondence with GenoDynamics she expressed nothing but respectful collegiality and appreciation for our willingness to share data as well as discuss the topic. Indeed, in her 2012 piece ("Detecting Hidden Violence: The spatial distribution of excess mortality in Rwanda") Verpoorten relies upon our casualty estimate of all violence in order to compare against the measurement she created, which as was correctly identified was one of the only estimates produced by some rigorous and transparent method. As a result, there is no fundamental problem that Verpooten has with our work up until the date of the recent blog and, in fact, she presents a rather straightforward account of our research when she states that
Davenport and Stam (2009) apply a Bayesian latent variable model to the data in order to obtain
estimates on the number of killings taking place in 164 administrative communes during AprileJune
1994. The sum of victims in communes under the jurisdiction of the FAR is estimated at
approximately 890,000, which adds to the evidence of a high death toll under the jurisdiction of the
genocidal regime. Davenport & Stam (2009) use the same approach to estimate the number of
killings that took place in 1994 within the zones under RPF controland the zones contested by the
RPF and FAR.. They arrive at estimates of respectively 80,000 and 90,000, which confirm previously
made allegations of 1994 human rights violations, in particular reprisal killings, at the address of the
RPF (Des Forges, 1999; Prunier, 1998, the Gersony report).
That was all she said about it. There was no fatal flaw to our operationalization of total casualty figures. There were differences but nothing that would result in casting the measure aside.
As for the particular comments made in the recent blog, it appears that Verpoorten's problem lies with the lower range of our estimation regarding how many Tutsi were killed in the territory under the control of the FAR. For the record, this includes genocide but it also includes other forms of violence as well - something Verpoorten's (2012) article explicitly investigates. The higher range is fine and indeed overlaps with the estimates that she provides (i.e., our estimation ranges overlap).
Central to Verpoorten's critique is the issue of the 1991 census that was conducted by the Rwandan government. Why do we keep discussing this census? Well, partly this is because we cannot go forward. Citing fears of divisionism, ethnicity (accept for explicit discussion of the "genocide of the Tutsi") all references to ethnicity have been banned (including the census). We also keep going back to this source because to date there has been no list produced regarding victims of the mass killing and without a list with all the potential victims it is hard to understand what took place as well as why. Genocide, for example, requires ethnic targeting and the census identifies ethnicity thus making it a natural location to look. Lists were read over the hate radio stations but the sheer number of names read did not come close to the number commonly believed to be killed. Lists were believed to be handed out to perpetrators but none have been recovered. The census or something like it thus proves crucial to those of us trying to understand what took place in Rwanda during 1994. For example, with a good census and the identification of everyone by ethnicity, then this information would be used to find and kill everyone. If the census was good though, then the casualty count for Tutsi would be low according to Verpoorten, relative to the number she believes to be present. But, if the census was bad and it did not identify everyone by ethnicity properly, then the high casualty count based on ethnicity identification is problematic. That is, unless there was another source that was used.
Verpoorten's (2005) work on one geographic location (GIkongoro) is the source she trusts to make her case. This was a population estimate provided by local authorities whom she trusts were more accurate in identifying individuals ethnicities than authorities from the national government office. If this information is good, then this would account for a larger number of Tutsi deaths. If the information is bad, however, then this suggests that the killing was unclear. Some interesting assumptions exist here.
First, it is presumed that the list created by the local administrators regarding who should be targeted was the one that was used when people were being killed. No one has provided evidence to this effect and there are numerous reports of people coming from the outside a community to kill people. What list would they have used? Were lists discarded and people in the community pointed out those to be targeted? To answer these questions, we need detailed information about who did what to whom in each locale, not simply population figures taken before and after overt violence.
Second, the approach Verpoorten uses assumes that the situation in Gikongoro applies to ALL of Rwanda or at least parts of it. Indeed, she explicitly states that this is problematic when she remarks that "Admittedly, there is no good reason to assume that the under-reporting occurred on exactly the same scale in all prefectures" (2005: 341).
Third, the approach Verpoorten uses (in 2005 and in her blog) assumes that individuals were killed where they came from. But, as Verpoorten acknowledges in her 2012 piece
The refugee crisis was nationwide, but more intense at the borders with RD Congo and Tanzania,
two neighboring countries that hosted an estimated 1,200,000 and 580,000 refugees, respectively.
Burundi also hosted a considerable number of refugees (270,000) while Uganda hosted no more
than 10,000 refugees (UNHCR, 2000, chap. 10). Both the proximity to a refugee camp and its size
are likely to affect excess mortality (51: fn 3).
In addition to this, there is the issue of internally displaced persons (IDPs) which might have reached another 2 million, if not more. This means that somewhere between a half or more of the population was on the move during the violence of 1994.
Movement is an especially problematic issue in the calculation of casualty estimation for it raises some questions that are rarely asked in Rwanda. For example,
- when did people move - exactly?
- how many moved and who moved (Tutsi alone, Hutu alone or Tutsi as well as Hutu)?
- how is ethnicity identified as people move around outside of their home communities?
- what does movement of the population mean for static pre/post evaluations?
As a community of people interested in Rwanda and the violence of 1994, we have very little systematic answers to these questions but they are crucial for movement tends to undermine and invalidate many of the discussions that take place (something also noted in Verpoorten's 2012 study as she attempts to figure out how many died in the different forms of killing that occurred: civil war, genocide, (counter)insurgency and the refugee crisis.
Related to this, movement is problematic because it makes one think a bit harder about how people were killed and this is a level of depth that most do not wish to engage in (for obvious reasons). When one does this, however, it is clear that most assume that killers are traveling around with lists of people to kill most likely derived from the census or local administration. But, this requires a high degree of coordination, literacy and copying machines - all of which are in short supply, especially as the existing government were getting destroyed on the battlefield and the Rwandan population was fleeing. What good is a list if people are running and hiding?
All of this discussion places a tremendous amount of confidence in Rwandan abilities to identify ethnic others. But, this is not clear. Indeed, this might explain why people had id cards. If ethnicity was so obvious, then these would be unnecessary. But, if you were not sure who somebody was, then you would need some id. Historical researchers note the significant amount of intermarriage that took place between Tutsi and Hutu. What was the child of a Hutu man and Tutsi female? What was the child of a Tutsi man and Hutu female? Were all Hutus identified as Hutus? Were all Tutsis identified as Tutsis? Do answers to these questions vary across time and place? Were mistakes made and under what circumstances? These issues clearly complicate the identification and targeting of ethnic victims as the violence takes on an air of variability across individuals and contexts. This complexity appears to be evident in eyewitness testimony from the ICTR regarding what took place in 1994: 1 and 2 (these are two examples from around 7000 that were compiled by the ICTR).
The Rwandan government has not helped with assisting individuals in identifying which specific groups were killed. Indeed, the research conducted by the government on the topic reveals the difficult in obtaining an actual count of casualties by ethnicity. According to the Ministry of Local Governments own survey of genocide victims, the complexity identified above is quite clear. As they state in their report: "genocide victims are all persons which have been killed in the period between 10/1/1990 and 12/31/1994
- because they are Tutsi;
- or they can be associated by lineage with a Tutsi;
- or they are friends with a Tutsi;
- or have a particular affinity with (a Tutsi);
- or they exhibit political thoughts and/or belong to a political party contrary to the ideology of the divisionist politics before 1994".
This is a rather broad conception of victims and one that likely pushes beyond the traditional definition of genocide. Indeed, Article II of the Convention on the Prevention and Punishment of the Crime of Genocide defines the relevant political violence as any of the following acts committed with the intent to destroy, in whole or in part, a national, ethnical, racial or religious group, as such:
- Killing members of the group;
- Causing serious bodily or mental harm to members of the group;
- Deliberately inflicting on the group conditions of life calculated to bring about its physical destruction in whole or in part;
- Imposing measures intended to prevent births within the group;
- Forcibly transferring children of the group to another group.
Clearly, the categories of genocide victim used by the Rwandan Ministry above only partially fits the definition for genocide identified. Being a "friend" or having an "affinity with" someone from a group is not the same thing as being from the group. Actually, being a "friend" or having an "affinity with" likely represents Hutus for if they were Tutsi then they would have been classified as ethnic Tutsi, which is the first category. Having a "lineage with a Tutsi" presumes that this is the reason why someone would be targeted and that the perpetrators know this. It is possible that someone might see themselves and be perceived by others around them as Hutu but through lineage they would later be counted as Tutsi. Finally, there are those that because of their belief against divisionism, hatred and violent policies would be placed in the category of genocide victim. As this concerns political beliefs, this clearly moves past the strict categorization of the term genocide. Again, this also includes Hutus.
All of this raises questions regarding the identification of Tutsi victims. It may be that the research from GenoDynamics is capturing 200,000 ethnic Tutsis that are clearly and consistently identified as such and that the other victims we are identifying are those fitting into the other categories highlighted by the Rwandan government which includes both Tutsi as well as Hutu victims. Rather than compel a reduced conversation, therefore, our estimation and the issues identified above compel a deeper, wider and more thorough conversation about who died, why and who was responsible that allow us to differentiate between the categories of victims identified above.
RSS Feed