Statistical Analysis of the Data
The data for this study were compiled by the American Bar Association Central and East European Law Initiative (ABA/CEELI) and the American Association for the Advancement of Science (AAAS) with the cooperation and support of other human rights NGOs, including The Center for Peace Through Justice (Center), Physicians for Human Rights (PHR), and Human Rights Watch (HRW). Each of these organizations conducted extensive interviewing of Kosovars concerning what happened during the conflict, and each generously agreed to contribute their raw data to this project.22 In total, there were 3,353 interviews included in this study.
This combined dataset consists of three subsets. The first source is the PHR survey of 1,180 Kosovar Albanian households conducted between April and May of 1999. The second subset is from an HRW survey of 591 Kosovar Albanians from March to September 1999. The third data subset is from the ABA/CEELI-Center interviews of 1,582 Kosovar Albanian refugees from May to September 1999. Combined, these three projects interviewed Kosovar Albanian refugees in diverse contexts. Interviews were collected in refugee camps in Albania, Macedonia, Poland, and the United States; at border crossing points in Albania, Macedonia, and Montengero; and in villages among Kosovar Albanians after refugees began returning to Kosova/Kosovo.
The statistical projection of the total number killed should be contrasted with the descriptive information that follows the estimate. The descriptive information is drawn directly from the cases documented in the various interviews. This type of information can be used to present a picture of the incidents reported, but it cannot support generalizations about the entire population. As discussed below, a larger dataset including more diverse and extensive lists of Kosovar Albanians killed in the conflict would help to extend the population-level generalizations to more precise analyses of particular villages and municipalities and specific times. The authors of this report hope that these promising initial results will encourage other organizations to join in exactly such an expanded, collaborative effort.
Estimates of the Total Number Killed
In this study, ABA/CEELI and AAAS generate estimates of the number of killings that occurred between March 20 and June 12, 1999, dates that correspond generally with the period of the NATO air campaign. AAAS statisticians estimate that approximately 10,500 Kosovar Albanians were killed during this period. This estimate tracks closely the early numbers suggested by the U.S. Department of State and the ICTY. Furthermore, it is based upon statistical methods that would be defensible in a court of law.
Figure 1: Estimated Number of Kosovars Killed23
Because the estimate of 10,500 killed was generated from samples of the population and not the entire population itself, a range must be computed that represents a margin of error for the estimate due to the sampling methods and the estimation technique. Using a 95 percent confidence interval, AAAS statisticians estimated the number of Kosovar Albanians who were killed during this time period to fall between 7,449 and 13,627 (see Figure 1). This confidence interval indicates that if this study were repeated 100 times using different but independent lists of data, one would expect that in 95 of the 100 studies, the estimate would fall within the range of 7,449 to 13,627 killings.
This confidence interval is most significant because it establishes that the estimate of 10,500 killed is consistent with the findings of other scientific estimates of the number of killings (see Figure 1). In September 1999, the Center for Disease Control (CDC) conducted a two-stage cluster survey among the Kosovar Albanian population in Kosova/Kosovo. They collected retrospective mortality data, including cause of death, for the period from February 1998 to June 1999. Their report concluded that approximately 12,000 Kosovars had died in the conflict with a confidence interval of 5,500 to 18,300.24 An independent PHR study25 estimated that there were 9,269 Kosovar Albanians killed in the year preceding the interviews (the majority of these killings occurring in 1999). A 95 percent confidence interval for the PHR estimate results in a range of killings between 6,911 and 11,627.
The bar graph in Figure 1 illustrates the different estimates of killings and their confidence intervals, including an estimate and confidence interval generated in this study. The first bar represents estimates of the number of killings between March 20 and June 12, 1999, generated from the combined HRW, PHR and ABA/CEELI-Center data.26 It shows 10,538 killings, with a corresponding 95 percent confidence interval ranging between 7,449 and 13,627 (see Section II on Statistical Methodology for further discussion of these techniques).
Thus, the estimate from these analyses yields an estimate of approximately 10,500 and a 95 percent confidence interval with an approximate range between 7,500 and 13,750 individuals killed between March 20 and June 12, 1999 (bar one). This figure further illustrates a general convergence of our estimate with other scientific estimates, as shown in bars two and three in Figure 1. Bar two indicates the PHR estimate of 9,269, with the confidence interval ranging between 6,911 and 11,627 killings. Bar threerepresents the CDC estimate of 12,000, with a range between 5,500 and 18,300 killed. Note that the estimate of 10,500 fits within the confidence intervals from the PHR and CDC studies, and that both the PHR and CDC estimates fit within this study’s confidence interval.
With this comparison, it is possible to assert that there is strong scientific evidence to support the early ICTY and U.S. Department of State estimates. Further data could refine the overall estimate, narrowing the confidence interval. ABA/CEELI and AAAS speculate that the resulting estimate could increase slightly with additional data, rather than decrease. Thus, the 10,500 figure may be viewed in terms of a minimum total number estimated to have been killed.
Descriptive Information on Killings by Time and Municipality
While the data are sufficient to estimate the total number of killings in the population as a whole, it is more difficult to generate detailed population estimates by dates and locations of killings. Although this information was collected, the coverage for each location and date is not extensive enough to allow for statistical inference. That is, these data are insufficient to make estimates of the total number of people killed in each municipality for each sub-period of the conflict. For this reason, the discussion that follows turns from population estimates to descriptive data from the three data sources. The information presented below does not represent estimates of the total number of people killed. There are killings that were not reported to the ABA/CEELI-Center, HRW or PHR researchers. Because it is not currently possible to say with certainty how they are distributed across time and space, estimates of these unreported killings are not included in the results below.
While an accurate estimate of the total number killed is an important issue with geopolitical consequences, an equally compelling point of inquiry is an examination of the timing of the killings. These findings reveal that a majority of the documented killings occurred between late March and mid-April. This timing correlates to a substantial degree with refugee flows. Building upon past AAAS research on this topic, the results of the ABA/CEELI-AAAS study support the proposition that there was a systematic campaign conducted against the Kosovar Albanian population.
An analysis of the documented killings by time can be seen in the lower portion of the line graph in Figure 2. The data indicate that there was a peak in killing in late March, followed by another in mid-April. Reports also indicate a smaller, but sustained peak in late April to mid-May, after which the number of documented killings tapered off. This pattern of peaks corresponds with the pattern of refugee flows that occurred during these times.
Figure 2: Documented Killings and Estimated Expulsions by Time
In the AAAS study Policy or Panic,27 refugee flows out of Kosova/Kosovo are described as having occurred in three distinct phases: March 24 - April 6, April 7 - 23, and April 24 - May 11. During the beginning of each phase, the flow of refugees was relatively light. The number of refugees leaving Kosova/Kosovo would rise to a high point (a peak, group of peaks, or plateau) during the middle of the phase, before tapering off toward the end of the phase. These estimates of refugee flows are depicted in the upper segment of the line graph in Figure 2, with the three phases noted by the shaded portions of the figure.
By comparing the estimated numbers of people who left each municipality over time to the times when NATO airstrikes occurred, the AAAS study concludes that only a small fraction of Kosovar Albanians fled Kosova/Kosovo as a direct result of NATO bombing raids. It also concludes that the mass exodus of refugees from Kosova/Kosovo occurred in patterns so regular that they must have been coordinated. In the context of descriptive accounts given by refugees, the most likely explanation for the migration is the implementation of a centrally-organized campaign to clear at least certain regions of ethnic Albanians.
This last proposition is supported by the timing of reported killings seen in the analyses presented here. As is evident from comparing the upper and lower segments in the line graph in Figure 2, killing patterns mirror refugee flows extremely closely. If killings are used as a means of intimidation to facilitate mass, forced evictions, then refugee flows and killings would logically occur together.28 The close correspondence between the rise and fall of numbers of refugees leaving their homes and reported killings is wholly consistent with that postulate. Furthermore, the fact that the number of reported killings fluctuates in unison with refugee flows is consistent with the proposition that there was a centrally-organized campaign targeting ethnic Albanians.
Figure 3 shows the overall distribution by municipality of documented killings in Kosova/Kosovo from March 20 - June 12, 1999. The municipalities of Skenderaj/Srbica and Rrahovec/Orahovac reflect a substantially higher concentration of killings than the others. These findings can be attributed to the fact that those two municipalities historically had been deeply-infiltrated by the UÇK and were therefore the site of heavy fighting. However, there were also killings in places with little or no reported UÇK activity, for example Gjilan/Gnjilane. With the low levels of UÇK activity, it is difficult to explain these killings as casualties of fighting between Serbian and UÇK forces. This wide dispersal of killings across municipalities with low UÇK activity lends support to the conclusion that the deaths were mainly a result of a centrally-organized campaign that targeted civilians. Moreover, the wide dispersal of killings provides an explanation for why the number of bodies discovered in ICTY exhumations of mass graves to date do not match public expectations.29 Bodies resulting from dispersed killings would likely be buried or disposed of in a similarly dispersed pattern. Thus, this dispersal pattern is consistent with current ICTY findings.
Figure 3: Documented Killings by Municipality
As with the patterns by time, the pattern of killings by municipality closely follows that of refugee flows. Figure 4 presents a comparison of killings and expulsions throughout Kosova/Kosovo during each of the three phases defined in Policy or Panic. On the right side of the figure are the maps of expulsions for each of the three phases. During the first phase (March 24 - April 6), most of the refugees came from western and southwestern Kosova/Kosovo. In the second phase (April 7 - 23), most of the refugees left their homes in the northern and central municipalities. During the final phase (April 24 - May 11), refugees came largely from the western and southern municipalities. The patterns of expulsions are again replicated by the documented reports of killings found in this study, as can be seen by comparing the expulsion maps with the corresponding maps of killings on the left side of Figure 4.
Figure 4: Documented Killings and Estimated Expulsions, by Phase and Municipality
|
Killings
|
Expulsions
|
|
 |
 |
Phase 1 |
 |
 |
Phase 2 7 April-23 April |
 |
 |
Phase 3 24 April-11 May |
 |
 |
During phase one of the conflict, while killings occurred region-wide, the majority were concentrated in the southwestern municipalities, particularly Peje/Pec, Gjakove/Djakovica, Rrahovec/Orahovac, Suhareke/Suva Reka, and Prizren/Prizren. In phase two, the concentration of killings in central Kosova/Kosovo increased, particularly in Istog/Istok, Gllogovc/Glogovac, and Lipjan/Lipljan. As Figure 4 shows, this pattern mirrors the increase in flow of refugees from these municipalities during the same time period. With the onset of phase three, the concentration of killings shifted to the north-eastern section of Kosovo. In particular Leposaviq/Leposavic, Zubin Potok/Zubin Potok, Vushtrri/Vucitrn, and Podujeve/Podujevo experienced a marked increase in killings from the previous phase.
In the third phase, the pattern of killings differs from the pattern of refugee flows. This variance could be explained by the lack of refugee flow data collected on people who fled to Macedonia or Montenegro. Given the transportation infrastructure in Kosovo, those residing in the northern municipalities most likely would not have left via Albania, which is where most of the data on refugee flows were collected. A more complete picture of the events in Kosovo, particularly in the third phase of the conflict, could be gained by further collaboration among NGOs with relevant data.
Previous work has concluded that the geography and timing of Kosovar Albanians’ mass departure from their homes suggests that there was an organized campaign to clear ethnic Albanians from parts of Kosovo. The patterns of people killed in Kosova/Kosovo over time and across space are similar to the migration patterns and also imply coordination. Narrative reports in the interviews in these datasets attributed the vast majority of the killings to Serbian forces. This claim is consistent with the information collected by other organizations such as the International Crisis Group30 and OSCE.31 In light of the descriptive sources’ attribution of human rights violations to Serbian forces and given the conclusion that the patterns of killings suggest central coordination, this report concludes that Serbian forces were responsible for conducting a coordinated campaign of killings against the Kosovar Albanians.
22 This sharing of data was undertaken with the mutually-accepted understanding that the identity of those interviewed would be protected.
23 Bar 1: This estimate follows Marks, Seltzer and Krotki, equation 7.118, with the error computed via jackknifing. E.S. Marks, W. Selzer, & K.J. Krotki, Population Growth Estimation: A Handbook of Vital Statistics Measurement (1974). See Section II for a detailed methodological explanation. Bar 2: PHR’s population inference, Physicians for Human Rights, War Crimes in Kosovo 45 (1999), depends on an assumption that there are 1.8 million Kosovar Albanians. Their estimate includes the period between April 1998 and May 1999. Bar 3: Spiegel and Salama’s (2000) estimate includes the period between February 1998 and June 1999. Paul B. Spiegel and Peter Salama, War and Mortality in Kosovo, 1998-1999: An Epidemiological Testimony 355 Lancet 2204 (2000).
24 Paul B. Spiegel and Peter Salama, War and Mortality in Kosovo, 1998-1999: An Epidemiological Testimony 355 Lancet 2204 (2000). The Spiegel and Salama study included 1,197 households comprising 8,605 people. From February 1998 through June 1999, 67 (64%) of 105 deaths in the sample were attributed to war-related trauma, corresponding to 12,000 (95% CI 5,500-18,300) deaths in the total population. The crude mortality rate increased 2·3 times from the pre-conflict level to 0·72 per 1,000 a month. Mortality rates peaked in April 1999 at 3·25 per 1,000 a month, coinciding with an intensification of the Serbian campaign of “ethnic cleansing.” Men of military age (15-49 years) and men 50 years and older had the highest age-specific mortality rates from war-related trauma. However, the latter group were more than three times as likely to die of war-related trauma than were men of military age (relative risk 3·2). Id.
25 The data in the PHR Study are distinguishable from the HRW and ABA/CEELI-Center data in that they were collected using probability sampling methods and closed-ended questions. The HRW and ABA/CEELI-Center data were collected in narrative form and without a sampling methodology. Consequently, the narratives in the ABA/CEELI-Center and HRW interviews had to be “cleaned” and coded to be used in this study.
27 Patrick Ball, Policy or Panic? The Flight of Ethnic Albanians from Kosovo, March-May 1999 (2000).
28 Médecins Sans Frontières came to a very similar conclusion in their 1999 report: “The population is not fleeing armed confrontations: they are being forced to leave their city or village under the threat of death.” Médecins Sans Frontières, supra note 2, at 2.
29 Jonathan Steele, Figures Put on Serb Killings Too High, The Guardian, Aug. 18, 2000, http://www.guardianunlimited.co.uk/Kosovo/Story/0,2763,355781,00.html.
30 See, International Crisis Group, Reality Demands: Documenting Violations of International Humanitarian Law in Kosovo 1999 (2000).
 |
|
|
|
 |
 |
 |
|