The chapter, prepared in conjunction with the Science and Human Rights Program of the American Association for the Advancement of Science (AAAS), is the first large-scale data project conducted by Human Rights Watch.1 It hopefully contributes to the growing field of human rights and statistical analysis.2

The interviews were conducted by Human Rights Watch researchers, usually with a interpreter, in Albania and Macedonia between March 28 and June 12, 1999, and in Kosovo between June 12 and December 31, 1999. Interviewees were selected for their knowledge of specific abuses inside the province. All interviews were conducted with a view to eliciting open narratives of what the interviewee had seen or experienced inside Kosovo between March 20 and June 12, 1999, rather than through standardized questionnaires. On return to New York, the interview documents were coded by trained volunteers for violation types, time and place of violations, victims, and perpetrators. A database was created, which is available for public use at http://hrdata.aaas.org.
 

Limitations of the Data

The statistics presented in this chapter shed light on the nature of war crimes in Kosovo, but they do have limitations. Most importantly, Human Rights Watch did not randomly sample the interviewees. On the contrary, researchers purposefully sought out not only the victims and witnesses of violations, but specifically those with knowledge of the most serious violations, such as torture, sexual violence, and executions. Therefore, these data cannot be extrapolated to general findings for Kosovo as a whole. Nor can these data provide information about the total number of persons killed, or give a complete picture of violations throughout the province.

Human Rights Watch activities were largely in the municipalities in Kosovo that were known to have been hardest hit by the war, such as Glogovac, Orahovac, Djakovica, Prizren, and Srbica. However, additional focus was directed to municipalities in the southwest, such as Djakovica and Orahovac, because researchers in North Albania during the war documented a heavy flow of refugees from those areas. Some northeastern and central municipalities, specifically Podujevo and Kosovo Polje, where many killings took place, are under-reported in this chapter and in the report as a whole.

Lastly, the process of coding and database creation is imperfect, particularly when dealing with complex narratives, as is the case with many war crimes in Kosovo. In complicated scenarios, it is not always easy to prepare statistical data that records accurately what occurred, where, and when. Kosovo is further complicated by confusion and ambiguity concerning place names in Serbian and Albanian, as well the fact that some of the same village names appear in up to four different municipalities.

Because of concerns like these, the Human Rights Watch Kosovo database was repeatedly checked and adjusted to eliminate errors, a process that is ongoing. To reduce errors, all instances for which the number of execution victims was over ten but considered imprecise were dropped from the total number of reported executions. In addition, the top five municipalities for executions, as well as some of the other municipalities, were carefully reviewed an additional time to eliminate faulty entries or records that counted the same execution violation more than once.

Like the report in general, these data only deal with violations committed between March 20 (when the OSCE withdrew from Kosovo) and June 12, 1999 (when NATO entered Kosovo). Note that for clarity, all percentages have been rounded to the nearest integer.

General Findings

From the large body of Human Rights Watch interviews, 577 interviews were coded because the interviewee had direct knowledge of a human rights or humanitarian law violation. From these 577 interviews, Human Rights Watch recorded more than 35,000 unduplicated violations, although many were suffered in succession by the same individuals.3 It must be noted that a violation may involve one or more victims. In many cases, for instance, the populations of whole villages or cities were expelled, such as the village of Ade or Pec city, or entire households were killed, like the Berisha household in Suva Reka. Again, the number of violations reported to Human Rights Watch cannot be extrapolated to suggest how many violations were committed in Kosovo as a whole.

The main violations reported to Human Rights Watch are depicted in Graph 1 below. Only those violations reported more than one hundred times are shown.

Note: This graph should not be construed to suggext the frequency or relative frequency 
of violations in Kosovo, since Human Rights Watch tried to document the worst of abuses. 
Executions, for example, are likely to be overrepresented in comparison to indiscriminate shelling 
since researchers purposefully sought out evidence of individual cases of such killings. 
sep - separations of men, women, and children; disp - forced displacement; det - detentions;
exec - executions; beat - beatings; hars - harassment; rob - robbery; shell - indisriminate shelling
prop - private property destruction; miss - missing persons; lab - forced labor; atex - attempted execution

With 5,122 reported violations, the forced separation of men, women and children4 was the most commonly reported violation. Displacement,5 with 4,485 reported violations, was the second most common, which is understandable given that it was a dominant violation of the conflict-more than 850,000 Kosovar Albanians were expelled from Kosovo, according to UNHCR, and thousands more were internally displaced. The third most frequent violation was detention6 with 3,478 reported violations, followed by extrajudicial executions with 3,453 violations.

Future reports by Human Rights Watch and others may focus on the patterns of these violations: when and where they occurred over time, and in what circumstances. Also of interest is whether certain violations tended to take place in isolation or together with other violations. For the sake of simplicity, however, this report focuses on only one of the violation types, albeit the most serious of the crimes: executions.

An Analysis of Extrajudicial Executions

It should be noted that extrajudicial executions by state actors-deliberate killings with no judicial process-may be over-reported in relation to other violations in this chapter since Human Rights Watch researchers actively sought to document such deliberate killings as a priority. At the same time, many extrajudicial executions committed in Kosovo are clearly not included in the 3,453 cases; just as an example, information on large-scale killings in Beleg, Goden, Kacanik, and Podujevo were not included in these data or the geographic chapters. In addition, the bodies of some people reported missing to Human Rights Watch during the data collection period have since been discovered. Despite these concerns, the body of information on executions collected by Human Rights Watch is large enough to draw some significant conclusions about the pattern of killings by Serbian and Yugoslav forces.

In the 3,453 documented executions, Human Rights Watch obtained the names of 916 people, or 27 percent of the victims. The rest of the victims were unidentified by witnesses.

The Gender of Execution Victims

As is clear from the cases documented in other chapters of this report, Serbian and Yugoslav forces summarily executed males at a much higher rate than females.

Of the 3,453 execution victims reported to Human Rights Watch, the gender of the victim was known for 2,232 people (65 percent). Of these 2,232 victims, 2,055 of the people were male (92 percent) and 177 were female (8 percent). This breakdown is depicted in Graph 2 at right.

These findings would be expected if the data dealt with deaths in combat or even summary executions of combatants, since the KLA's forces were predominantly male. But, as the case studies in other chapters make clear, the vast majority of summary execution victims were civilians who did not participate in combat. Take, for example, the killings of approximately ninety prisoners in the Dubrava prison or the roughly 300 men taken from refugee convoys and killed in Meja.

Clearly this represents a targeting of Kosovar Albanian males. This finding is reinforced by the fact that, during the NATO bombing, many males were either in hiding within Kosovo, fighting with the KLA, or living abroad, while women were more likely to have stayed at home during the war, where they were susceptible to abuse.

The Ages of Execution Victims

Of the 3,453 known victims of summary execution, Human Rights Watch obtained the age of 630 people (18 percent). Of the victims for whom age was known, 530 were males (84 percent) and one hundred were females (16 percent). The fact that the age of victims was known in only 18 percent of the cases should be considered when conducting an analysis, since bias may have been introduced. Witnesses might have only provided ages for the youngest or oldest of the victims, for example, in order to emphasize the seriousness of the crimes. Even given this possibility, however, these data reflect some interesting results that deserve mention.

Notably, the ages of summary execution victims differ for men and women. For male execution victims, the average age was 40.3 years. For female victims, the average age was 32.7 years. Graphs 3 and 4 depict the age distributions for male and female summary execution victims.

It is clear that the pattern of violation is different for male and female. The killings of men and boys tended to target equally males between the ages of 10 and 70, with a falloff at higher ages. The summary executions of females were high for ages 10-30, then fell off to a uniform level for those over 30.

In both cases, the murder of children below ten were lower. However, female children in that age group were proportionately more likely to be killed than males. These qualitative comments based on Graphs 3 and 4 are reflected in the summary comparison statistics in Figure 1 below.

As Figure 1 shows, 75 percent of the male execution victims were below 56 years of age, while 75 percent of female execution victims were under 50. The "average" age for males was 40.3 and for females was 32.7. Similarly, 25 percent of the female victims were below 14.5, whereas 25 percent of the male victims were below 22.

At first glance, the results are counter-intuitive. Most notably, based on the case studies, one would expect to see a rise in executions of military age men, who were targeted for killing during village sweeps, such as in the villages of Cuska, Bela Crkva, and Meja. In numerous cases, men between the ages of 18 and 50 were separated from women and children and killed. However, there are a number of plausible explanations for the discrepancies in the victims' ages.

First, as mentioned above, there were relatively fewer fighting age males in the villages during the NATO bombing. Many men between the ages of 20 and 50 were either hiding in the hills (fearful of being targeted), fighting with the insurgency, or living abroad. In many villages, women and children were left behind with a smaller number of older men. So, while the case studies provide testimonial evidence that military age males were targeted for execution, this is not reflected in these data since military age men were relatively less present in the areas susceptible to attack.

Second, the case studies show how men were frequently killed by government security forces after having been separated from women and children, such as in Bela Crkva or Izbica. Executions of females, however, more often took place in group killings (such as the execution of an entire family) and not from the deliberate targeting of women. The killing of twelve members of the Gerxhaliu family on May 31 in Gornja Sudimlja7 or the Berisha family in Suva Reka8 on March 25 are examples where a family-men, women and children-was killed together. In other words, women were more likely to be killed in groups for which the killers did not distinguish between gender or ages, thereby including some younger female victims.

A third possibility is that executions of women were related to sexual violence which involved younger female victims. Unfortunately, it is impossible to prove this theory with the data collected. Although Human Rights Watch coded for rape and sexual violence in the database, which would theoretically allow an analysis of whether female executions and sexual violence tended to occur at the same time, the sensitive nature of sexual violence in Kosovar Albanian society rendered the data on those crimes unreliable, in the opinion of Human Rights Watch. In other words, sexual violence was under-reported both in testimony and in the database.

One point to consider with these data on ages is the rate of killing. For example, it looks as though government forces were not summarily executing older men and women at a high rate. However, given the fact that there are fewer older people in the population than middle-aged or younger adults, then the elderly were being killed at a relatively higher rate. The opposite is true for children under age ten. Since it is likely that there are more children of that age in the population than adults or elderly, then the rate of children being killed is less than that of adults.9

Executions Over Time

Plotting the total extrajudicial executions reported to Human Rights Watch over time reveals a great deal about the systematic and coordinated nature of the violations in Kosovo. As Graph 5 demonstrates, total summary executions took place in three very distinct waves.

From the beginning of the offensive on March 20, there was a clear and rapid spike in extrajudicial executions, culminating around March 25-27, just after the commencement of NATO bombing. This was followed by a significant drop off, with a low point around April 5-6. A second wave of extrajudical executions peaked around April 27-28. A third but significantly smaller wave of executions reached its zenith around May 10-11, followed by a precipitous drop-off that peters out by June 12, with the exception of a minor bump around May 30-31.

The three distinct surges in executions suggest that the killings were not the result of random violence by government forces. Rather, that data supports other evidence that they were carefully planned and implemented strikes that fit into the government's larger strategic aims.

Of course, Graph 5 summarizes only those executions reported to Human Rights Watch and not the total number of executions committed in Kosovo. However, the pronounced nature of the three waves, based on 3,453 executions, strongly suggests the purposeful and coordinated nature of the violations. Although not all executions are represented, the findings based on partial data are strong and clear.

Executions by Municipality

Of the 3,453 extrajudicial executions reported to Human Rights Watch, 66 percent of the executions for which we have municipality identification occurred in the following five municipalities: Djakovica, Orahovac, Srbica, Glogovac, and Suva Reka. With the exception of Podujevo, where Human Rights Watch conducted little research, this clearly reflects the municipalities that were most impacted by the war and associated security operations from 1998 to 1999. Thirty-five percent of the reported executions took place in Djakovica and Orahovac municipalities alone, as shown in Figure 2 below.

Again, Figure 2 should not be interpreted as a representation of total extrajudical executions in the municipalities or in Kosovo as a whole since it reflects only those executions reported to Human Rights Watch. Two municipalities in particular stand out as having been undercounted due to only partial research in those areas: Podujevo and Kosovo Polje. With these notable exceptions, the data can be taken as a relatively fair reflection since Human Rights Watch documented a high percentage of the major killing sites across Kosovo. These data were collected in Albania and Macedonia during the war, as well as inside Kosovo after the war, so that all geographic areas were covered.10

The coordinated nature of extrajudicial executions in Kosovo is further revealed when the killings are examined by municipality. As the graphs below reveal, intense killing "sprees" tended to occur in municipalities over short periods of times, suggesting a strategic order to commit executions in certain areas or, in the least, the deployment of forces known for brutality and disregard with orders to terrorize the civilian population without legal constraints. More sporadic executions in the municipalities may not be reflected in the graphs since Human Rights Watch tended to focus on the larger-scale incidents.

The intensity of executions in any given municipality over a short period suggests that, as in Graph 5, the killings were not random events. Rather, there were distinct periods when killings were intense, suggesting they were the result of a premeditated and coordinated policy of violence. Our anecdotal research also supports the conclusion that executions in each municipality were specific and purposeful.

Executions and Expulsions: a Correlation

Evidence of a centrally coordinated attack on Kosovar Albanians is strengthened by another statistical study on the outflow of refugees from Kosovo to Albania. The April 2000 study conducted by the AAAS found that the refugee flows into Albania occurred in three separate waves.11 From late March to late May 1999, the report said, 95 percent of the Kosovar Albanian refugees who entered Albania did so during one of three "distinct phases," as shown in Graph 11.

To explain the graph, the report concluded:

It is our conclusion that the evictions were not spontaneous: mass migration on this scale and in this pattern could only have been driven by a centralized policy, not by individual decisions or emotions of either Kosovar Albanians or local Yugoslav military or police officials . . .

The coherence of the phases, and their apparent coordination across broad regions of Kosovo suggests that Yugoslav authorities devised and implemented a policy to attempt to clear at least certain regions of ethnic Albanians.12
 

As may be evident from Graph 11, the timing of the three refugee waves to Albania documented by AAAS coincides closely with the three waves of executions documented by Human Rights Watch (Graph 5).13 This is made even more clear when the AAAS data on expulsions and the Human Rights Watch data on executions are compared more directly in Graph 12.

As Graph 12 shows, the peaks and valleys of the three phases, and even the final bump, closely match for both executions reported to Human Rights Watch and refugee outflows to Albania. In other words, the executions in Kosovo over time appear to parallel expulsions.14 The difference in magnitude for the second wave could be attributable to the fact that most of the executions documented by Human Rights Watch in that time frame occurred in the north and central municipalities, particularly Srbica and Glogovac. A large percentage of those expelled from these municipalities went to Macedonia, where they would not have been picked up by the AAAS data.

This strong relationship further suggests that there was a centrally devised and implemented strategy to target Kosovar Albanians. The three phases of killings and expulsions seem tied to the strategic objectives of the military and political leadership in Belgrade.

One explanation is that government forces committed executions in order to expedite the expulsions-a theory that is supported by some case studies. In many villages documented in this report, such as Celina and Korenica, police, army, or paramilitary forces committed executions before, or in the process of, expelling the civilian population from a village or city. It is also understandable that killings would rise along with executions since government forces were unleashed on an area to be "cleansed." Often these were areas of KLA activity where policemen and soldiers had been killed, giving the government forces a justification, in their own mind, for violence.

The three waves of expulsions and executions can be further analyzed by municipality. AAAS found that the three phases of expulsions (Graph 11) related to different regions of Kosovo. Specifically, in the first phase of expulsions, most of the refugees came from western and southwestern Kosovo. In the second phase, most of the refugees came from the northern and central municipalities. In the final phase, refugees came largely from the western and southern municipalities. This geographic distribution is represented in Graph 13, taken from the AAAS report, which shows the proportion of refugees to Albania that came from the southwestern Kosovo municipalities (Suva Reka, Orahovac, Prizren, and Djakovica).

The Human Rights Watch data shown in Graph 6 through 10 are consistent with these findings. Namely, the municipalities in Kosovo's southwest, like Djakovica, Orahovac, Suva Reka and Prizren (see Graph 14), have large numbers of killings in the first time period. The northern municipalities like Glogovac were more likely to see executions in the second phase. In the third phase, the executions reported to Human Rights Watch were again mostly in the southwestern municipalities of Djakovica and Prizren.

Naturally, there are some exceptions. The killings in the north-central municipality of Srbica, for example (Graph 8), fall more neatly into the first and third phases rather than the second. The first phase is explained by the March 28 killing of between 146 and 166 men in Izbica, a former stronghold of the KLA that was attacked early on by government forces. The third phase surge is due to the killings in Rezala and Cirez as part of the government's offensive in Drenica.

Likewise, the killings in Pec (see Graph 15) tend to mirror the first and second phase. The surge around May 14 represents the killing of seventy people in the villages of Cuska, Zahac and Pavlan. As mentioned in the detailed section on these villages in the chapter on Pec, it remains unclear why these three villages were attacked at this time, since they had remained intact throughout the war and were apparently devoid of any KLA presence. Possibilities range from revenge (KLA General Agim Ceku's family is from Cuska) to local paramilitaries plundering the three untouched villages in the area.

This correlation between executions and expulsions was also studied by the AAAS, which conducted a second study on killings in Kosovo, Political Killings in Kosova/Kosovo, published in September 2000, in conjunction with the American Bar Associations's Central and East European Law Initiative (ABA/CEELI). The report's analysis of killings across Kosovo was based on 3,353 interviews collected by Human Rights Watch, ABA/CEELI, the Center for Peace Through Justice, and Physicians for Human Rights. The study concluded that approximately 10,500 Kosovar Albanians were killed between March 20 and June 12, 1999, with a 95 percent confidence interval from 7,449 to 13,627.15

The study compared the executions documented by these four organizations with the original AAAS report on expulsions and found very similar results to those presented by Human Rights Watch in Graph 12; namely, the three phases of expulsions closely match the three phases of executions. Similarly, as the AAAS-ABA/CEELI report states, "the pattern of killings by municipality closely follows that of refugee flows."

The strikingly similar conclusions reached by Human Rights Watch and the AAAS-ABA/CEELI report is in part due to the fact that Human Rights Watch provided its interview data for the AAAS-ABA/CEELI report. However, Human Rights Watch interviews accounted for only 577 of the 3,353 total interviews (17 percent). The similar findings should, therefore, be taken as independent confirmation of the results.

Perpetrators of Executions

Human Rights Watch asked all witnesses and victims of violations whether they could identify the type of perpetrator involved in the abuse: Serbian police, Yugoslav Army, paramilitary, or "other," for example, local Serbs, NATO, or the KLA. The results for the perpetrators of executions are presented below, but they must be taken only as an indication of perpetrator trends rather than definitive statements.

The main reason for this was Kosovar Albanians' difficulty in identifying Serbian and Yugoslav forces. While some witnesses and victims were confident in their identifications, many others, due to lack of knowledge about the forces and the generally stressful environment, were unable to distinguish between the police, army, and paramilitaries.

This was made more difficult by the large array of government forces used in the campaign, such as military police in the army, special antiterrorist forces in the police, paramilitaries, and local armed groups (see Forces of the Conflict). There were few standard uniforms and badges and insignia were not always displayed.

In addition, the scenarios in which these abuses took place were complex: one type of force might have shelled a village, another invaded it, and a third committed executions. Human Rights Watch asked witnesses which type of government force was "present" at the scene of a violation. This does not necessarily mean that it was that government force which actually committed the particular killings, but it can corroborate the testimonial evidence that most large-scale operations involved combined military and police or paramilitary forces.

Of the 3,453 extrajudicial executions reported to Human Rights Watch, witnesses claimed to have identified the Serbian police in 1,768 executions (51 percent), the Yugoslav Army in 1,173 cases (34 percent), and paramilitaries in 1,154 cases (33 percent). More than one perpetrator type may have been present at any execution.

The results are counterintuitive since the narrative chapters in this report suggest that paramilitaries were responsible for much of the worst killing, although the police and army were hardly exempt. Again, the fact that witnesses had difficulty identifying the different forces and that larger operations often involved a mix of forces probably account for the contradictory results.

When identifying perpetrators, it is easier to identify those with command responsibility for a notorious unit or a region where largescale killings took place. Given the intensity of the deliberate and unlawful killings in certain areas of Kosovo over short periods of time, as depicted in Graphs 6 through 10, as well as Graphs 14 and 15, it is highly likely that the various commanders in charge of the given municipalities were aware of the killings that took place in their respective areas of responsibility. Despite this, there is no evidence that military or political leaders took any steps to punish those responsible for the killings, or to minimize further such killings taking place as the conflict continued.
 
 
 
 

1 Dr. Patrick Ball, Deputy Director of AAAS's Science and Human Rights Program, designed the statistical analysis. Rebecca Morgan, a Human Rights Watch consultant, coordinated the coding process. Dr. Herbert F. Spirer, Professor Emeritus at the University of Connecticut, Adjunct Professor of International Affairs at Columbia University and consultant to AAAS, conducted the statistical analysis and generated the graphs. Fred Abrahams from Human Rights Watch wrote the accompanying text. Outside reviews were conducted by Dr. Fritz Scheuren and Tom Jabine.
Human Rights Watch is grateful to Drs. Ball, Spirer, Jabine and Scheuren for their time and expertise, as well as to the many volunteers, mentioned in the acknowledgement section, who helped to code the data.
This chapter is a joint product of Human Rights Watch and the Science and Human Rights Program of American Association for the Advancement of Science, which operates under the oversight of the AAAS Committee on Scientific Freedom and Responsibility (CSFR). The CSFR, in accordance with its mandate and association policy, supports publication of this chapter as a scientific contribution to human rights. The interpretations and conclusions are those of the authors and do not purport to represent the views of the Board, the Council, the CSFR, or the members of the AAAS.
2 Five other statistical studies of war crimes in Kosovo have been conducted by other organizations: See Central and East European Law Initiative of the American Bar Association and the American Association for the Advancement of Science, Political Killings in Kosova/Kosovo, September 2000; American Association for the Advancement of Science, Policy or Panic? The Flight of Ethnic Albanians from Kosovo, March-May 1999, April 2000; Physicians for Human Rights, War Crimes in Kosovo-A Population-Based Assessment of Human Rights Violations Against Kosovar Albanians, August 1999; Doctors Without Borders/Médécins Sans Frontières, Kosovo: Accounts of a Deportation,, April 30, 1999; Paul B Spiegel and Peter Salama, War and Mortality in Kosovo, 1998-99: An Epidemiological Testimony, The Lancet, vol. 355, no. 9222, June 24, 2000.
Books on human rights and data analysis that addressed related methods include: Spirer and Spirer, Data Analysis for Monitoring Human Rights, Washington: AAAS (1993); Patrick Ball, Who Did What to Whom?, Washington, AAAS (1996); Patrick Ball, Herbert F. Spirer and Louise Spirer (eds.), Making the Case: Investigating Large Scale Human Rights Violations Using Information Systems and Data Analysis, Washington, AAAS (2000); Jabine and Claude (eds.), Human Rights and Statistics: Getting the Record Straight, Philadelphia, University of Pennsylvania Press (1992); and Patrick Ball, Paul Kobrak and Herbert F. Spirer, State Violence in Guatemala, 1960-1996: A Quantitative Reflection, AAAS, Washington, 1999.
3 Human Rights Watch coded for the following violations: robbery, execution, looting, destruction of non-military objects, harassment, displacement, detention, abduction, beating, rape, sexual assault, indiscriminate shelling, separation, "disappearance," forced labor, torture, and human shields.
4 Separation was defined as a case where men and women and children were separated and the fate of one group or another, at the time of the interview, was not known.
5 Displacement was defined as forced expulsion or displacement from an area.
6 Detention was defined as an arrest or imprisonment in which detainees were held in the custody of the state. This includes cases in which detainees were subsequently tortured, "disappeared," or summarily executed.
7 Four of the Gerxhaliu victims were female, aged eight, thirty-six, forty-five, and eighty-one. Three other male victims were under the age of fourteen.
8 At least seven boys and five girls seventeen years of age or younger were killed.
9 This is especially true given that Kosovar Albanians have the highest birthrate in Europe.
10 For a description of the areas covered by Human Rights Watch, see the section on Methodology in the Introduction.
11 American Association for the Advancement of Science, Policy or Panic? The Flight of Ethnic Albanians from Kosovo, March-May 1999, April 2000.
12 Ibid.
13 The impact of this comparison is muted slightly by the fact that the AAAS data was predominantly from refugees who entered Albania, while the Human Rights Watch data was not limited in this way. However, Appendix A of the AAAS report explains why their data may be generalized, within limits, to the entire population of Kosovar Albanian refugees during this period, i.e. those who exited to Macedonia, Montenegro or Bosnia-Herzegovina. One important exception mentioned in the report is those who were internally displaced within Kosovo throughout the NATO bombing, for whom there is no information.
14 It is important to note that the time intervals for the two graphs are different: the Human Rights Watch graph is plotted by week, while the AAAS data is plotted by two-day periods. This does not, however, minimize the impact of the comparison. On the contrary, the correlation between the three phases is strengthened by the fact that, using different time intervals, the three phases still match. This helps show that the data are, in statisticians' terms, "robust."
15 The 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 100 studies, the estimate would fall within the range of 7,449 and 13,627.