ABSTRACT
As already discussed, for the measurement of the amount of one-sided violence perpetrated by the selected rebel groups, I use the UCDP One-sided Violence Dataset version 1.3-2010b and 1.3-2011 (covering the period 1989-2010) (see Eck and Hultman 2007). According to these datasets there are four armed groups that were selected that did not killed any civilians intentionally: the Forces Armées de la République Démocratique du Congo (FARDC), the Mau Mau, the Mouvement des Nigériens pour la Justice (MNJ), and the Ethiopian rebel group Oromo People’s Congress (OPC). These groups were included on the basis of the pretests, during which some experts indicated that they would like me to include particular armed groups. Consequently, they received a value of 0. Note, however, that the experts indicated in the survey that at least two of these groups (the FARDC and the Mau Mau) have often targeted civilians intentionally. On average, the sampled rebel groups killed more than 1,200 civilians in the period of 1989-2010. Some of the groups are, however, more violent than others. At the extreme, one can find the Congolese AFDL. Although this group was only active for one year, the dataset showed that this particular group killed more than 35,000 civilians. Also the Rassemblement Congolais pour la Démocratie (RCD) and the Rwandan Forces Démocratiques de Libération du Rwanda (FDLR), both active in the Congo, are known for targeting civilians on a large scale. The RCD has targeted close to 6,500 civilians, while the FDLR over 5,000. Other rebel groups are known to have shown restraint: the South African Afrikaner Weerstandsbeweging (AWB) has killed “only” 26 civilians in there separatist struggle. Also the APLA, and the Burmese Karen National Union (KNU) are now for their human rights abuses against the civilian population. Both groups have “only” killed 27 civilians in their quest. It is important to note that the information on the number of civilians killed is left-skewed; there are many armed groups that “just” killed a few civilians, and there are just a few armed groups that killed many noncombatants. This is very normal for count data. Consequently, scholars working with this kind of data are often advised to use methods based on the Poisson or the Negative Binominal distribution. However, this variable does not follow these distributions perfectly. There are many armed groups that have perpetrated a low level of civilian killings and few armed groups that have perpetrated an enormous amount of civilian killings. However, there are some groups that perpetrated more killings than
expected according to this distribution. This is often the case with count data in the field of international relations, such as battle death rates or the number of terror attacks. In these cases, running a Poisson model or a Negative Binominal model (if the variance is greater than the mean) might result in wrong standard deviations. To avoid this problem, I transformed the count data into a categorical variable with 5 categories, ranging from 0 (low level of civilian killings) to 4 (high level of civilian killings). Although this data transformation results in information loss, I no longer risk the possibility of drawing wrong conclusions. Table 5.2 shows the distribution of this variable.
