Institute for the Study of Asymmetric Conflict

Incident and Casualty Databases as a Tool for Understanding Low-Intensity Conflicts

Abstract: Today's "low-intensity" conflicts do not involve victory or defeat in the conventional sense; instead, each side attempts to achieve a psycho-political victory by influencing people's thoughts and feelings about the issues in dispute. Casualty statistics are an important element in forming these thoughts and feelings; in turn, a robust incident and casualty database can be an important tool in coming to an accurate understanding of complex conflicts with multiple actors and incident types. For a casualty database to produce meaningful, informative, and accurate results, it must have a rich array of well-defined categories to which to assign incidents and casualty data. It must also be conceived, designed, and administered with a strict adherence to accuracy rather than advocacy as a primary goal.


Don Radlauer

1 The Need for Casualty Data in Low-Intensity Conflicts

In traditional warfare, victory or defeat is ultimately a matter of physical realities: at some point, one side is no longer able to continue fighting, and either negotiates an end to the conflict or surrenders unconditionally. In this type of conflict, accurate casualty figures are useful to generals and political leaders who need to assess the situation – how much damage each side has suffered and how much fighting capacity each side retains – but it is clear that these statistics (as opposed to the lives they represent) are not in themselves a vital objective of the combatants. They are important only in that they clarify the situation “on the ground”.

In contrast, today’s “low-intensity” conflicts do not involve victory or defeat in the conventional sense. While these conflicts often have a significant military component, they are more accurately described as “psycho-political” conflicts rather than military ones, in that each side’s primary goal is to influence the course of events by changing people’s thoughts and feelings regarding the issues and parties involved in the dispute. In such conflicts, casualty statistics have become an important lever for gaining political influence, quite apart from any strictly military significance they may have; in fact, actors in low-intensity conflicts often pursue policies that would at first glance appear antithetical to their own interests, simply because of the political value (positive or negative) of casualty statistics.

While adversaries in low-intensity conflicts might wish to achieve a conventional victory, the nature of low-intensity conflict itself precludes such an accomplishment. Non-state actors in these conflicts lack the resources, infrastructure, and (in general) disciplined troops to achieve anything more than abject defeat in traditional, formation-against-formation warfare against a competent regular military; and while terror and guerilla attacks against a state actor can exact a heavy price, they do not constitute an existential threat. But at the same time – and paradoxically – the advantages of even the best-trained and equipped army are of little value in defeating non-state opponents.[1] Since non-state militants routinely shelter among non-combatant civilians, and since some of the most deadly weaponry they use (including suicide bombs and fertilizer-based car bombs, for example) can easily be constructed using simple “civilian” ingredients and techniques, only the most draconian measures taken by a state actor can entirely prevent terrorist and guerilla attacks. Such measures carry a very high political price, and in any case seem to have only local and temporary effect.

Neither side, then, can win (or can afford to win) a physical victory in a low-intensity conflict; instead each side must instead struggle to achieve psychological and political victory, or at least to avoid psycho-political defeat. Non-state actors seek to influence decision-making and alter the political climate of the state actor, to gain the attention and sympathy of non-participant states and groups, and gain support and recruits from their own social milieu. State actors seek to pursue their policies without internal or external interference. Victory can be achieved only in the hearts and minds of the antagonist societies themselves, as well as in the hearts and minds of influential non-participants.

In turn, people’s thoughts and feelings are largely a product of the information they receive about the course of the conflict. Casualty statistics are a major component of this information: They have huge emotional impact, as they represent human lives lost and damaged; and yet, as numbers, they carry the intellectual gravitas of mathematical factuality. In their most over-simplified form, they are also short and memorable, making them an easily remembered “sound bite” in an age of instant – and often superficially reported – news. But despite their importance, these statistics have not in general been subjected to the level of analysis they deserve.[2]

Low-intensity conflicts tend to be highly complex. Each side consists of multiple elements – some civilian, some military or quasi-military, and others somewhere in between. The non-state side in a low-intensity conflict may not even be a “side” in any conventional sense – it is perhaps better to view non-state actors as collections of more-or-less allied “sides” acting semi-independently, rather than as unitary entities. Even state actors can behave in fragmentary ways at times, with civilian groups acting as terrorists or vigilantes.

Further, low-intensity conflicts consist of many different types of incident, with conventional battle being only one of the less-common possibilities. Fatalities can be the result of terror attacks, guerilla attacks, counter-terror operations, arrest attempts, internecine violence, riots, “work accidents”, and so on. With all the different possible types of incident, all the different groups participating in these incidents, and of course innocent victims on both sides, the simplistic body-counts usually used to describe low-intensity conflicts completely fail to convey an accurate picture of what is really going on.

1.1 Competitive victimization

As stated above, many actors in low-intensity conflicts realize the political importance of fatality statistics, and have begun to craft policy accordingly. The most obvious “value” of fatalities is to promote a sense of victimization, and thus to gain the sympathy and support of well-meaning people in non-participant societies and among the opposing side; this sense of victimization has also been used by some actors to maintain enthusiasm for continued fighting among their own population. This political use of fatality statistics is seen in several current low-intensity conflicts, notably the Israeli-Palestinian “al-Aqsa Intifada”.

The asymmetry of low-intensity conflicts applies to the political use of fatality statistics as well as to the combatant forces themselves. It is common for advocates of the state actor in a low-intensity conflict to wish to emphasize the extent of their own side’s victimization by the non-state side; but in fact this strategy, while it works well for the underdog (virtually always the non-state actor), can have a boomerang effect on a state actor. States benefit from an atmosphere of stability: They profit from trade, encourage capital investment, and benefit from tourism. By emphasizing the extent to which a state is victimized by terrorism and other violence associated with low-intensity conflict, the state’s partisans are likely to do the state more harm than good.

2 Extracting Clarity from Chaos

It is clear, then, that while casualty statistics have become centrally important in low-intensity conflicts, the extreme complexity of the conflicts themselves, in addition to the intense emotional and political impact of casualty data, makes it quite difficult to arrive at an accurate and comprehensive picture of the deaths and injuries incurred. However, a robustly-constructed and carefully-administered database system can do a great deal to clarify the history of a low-intensity conflict; and perhaps, by improving the understanding of the nature of the casualties incurred and exposing the extent to which lives are being sacrificed to achieve political “points”, such a database may help to reduce the human cost on both sides.

In order to achieve these ends, a database for low-intensity conflict must provide features not found in traditional terror-incident databases. In particular, it must include detailed data for each fatality, since the usual practice of including only aggregate casualty figures does not allow for demographic and other analyses of those killed.

2.1 Basic structure and potential accomplishments of a casualty database

In order to improve our understanding of a low-intensity conflict, a database system must go far beyond the usual reporting of each side’s total body count. Both incidents and victims must be classified according to as rich a set of criteria as is realistically possible. Some of these criteria and their utility will be discussed in this section; the next section will cover in greater detail some of the “real world” difficulties and dilemmas that must be confronted in applying such a classification scheme.

While this paper is intended to address low-intensity conflicts as a generality, specific reference will be made to the International Policy Institute’s “al-Aqsa Intifada” Database to demonstrate the potential accomplishments of a low-intensity-conflict database application. Examples will be provided showing some of the significant findings of the “al-Aqsa Intifada” Database Project; but these are included only to demonstrate the methodology used. Discussion of these findings in relation to the issues involved in the Palestinian-Israeli conflict itself is outside the scope of this paper.

For reasons that will be discussed in the next section, it has proven impractical to deal directly with injury statistics in our work. Instead, we have chosen to focus almost exclusively on fatalities, on the assumption that deaths, as a subset of total casualties, can be used as a representative sample for purposes of analysis. Discussion of various aspects of our database will reflect the fact that while its design permits tracking of injuries as well as of fatalities, in practice we work only with the latter.

2.2 Combatant levels

One of the most significant – and contentious – categories for understanding casualty statistics is the combatant status of those killed and injured. The most emotionally significant casualty in any conflict is the innocent victim: While most people take the death or injury of an armed soldier or militant more or less in stride, almost everyone regrets and condemns the killing of the defenseless and innocent civilian. It is important, then, to determine how many of those killed and injured on each side were combatants. In order to enable detailed and flexible reporting of the combatant status of those killed in the conflict, we have divided fatalities into nine categories:

  • Full Combatant A “full combatant” is a soldier on active duty, an active member of a terrorist group, a civilian independently choosing to perpetrate an armed attack against the opposing side, or someone using a “hot” weapon to defend against an attack. In general, rock-throwers are not considered to be combatants; an exception to this generalization would be, for example, someone dropping large rocks from a bridge onto fast-moving traffic. A rioter throwing "Molotov cocktails", grenades, or the like can be considered a full combatant.

Mere possession of a weapon does not imply combatant status. A civilian driving with a weapon in his/her car, or a pedestrian with a holstered pistol, is normally considered a non-combatant. However, a civilian who encounters a terror attack in progress and draws his/her weapon in an attempt to stop or prevent the attack is a combatant once the weapon is out of its holster.

  • Probable Combatant A “probable combatant” is someone killed at a location and at a time during which an armed confrontation was going on, who appears most likely – but not certain – to have been an active participant in the fighting. For example, in many cases where an “Intifada” incident has resulted in a large number of Palestinian casualties, the only information available is that an individual was killed when Israeli soldiers returned fire in response to shots fired from a particular location. While it is possible that the person killed had not been active in the fighting and just happened to be in the vicinity of people who were shooting, it is reasonable to assume that the number of such coincidental deaths is not particularly high. Where the accounts of an incident appear to support such a coincidence, the individual casualty has been given the benefit of the doubt, and assigned a non-combatant status.
  • Violent Protester A “violent protester” may not be a full-time militant, but has taken an active and violent part in rioting or vigilante activity – such as throwing incendiary devices. This category is a subset of “full combatants”, created to enable more specific reporting of riot-related fatalities.

Full Combatants, Probable Combatants, and Violent Protestors are normally aggregated into a total “Combatant” figure; all the remaining categories are considered “Non-combatants”.

  • Non-Combatant A non-combatant is an innocent bystander – a person whose death or injury has no justification other than nationality or ethnicity. This category is used as a “catch-all” for those non-combatants who do not fall into one of the more specific non-combatant categories.
  • Health Related A “health related” fatality is someone who died from a cause only indirectly related to violence – for example, due to a heart attack following an incident, tear-gas inhalation, or a roadblock delay that prevented an ill person from receiving medical treatment in a timely manner.
  • Uniformed Non-Combatant A “uniformed non-combatant” is a non-civilian, but is not actively involved in the conflict. This category can include civil police as well as soldiers in uniform but not on active duty.
  • Protestor Unknown A “Protestor Unknown” is anyone who was killed during a protest for whom information as to violent behavior is unavailable.
  • Suspected Collaborator This is a special category for people targeted by militants of their own nationality who suspect them of aiding the enemy.
  • Unknown In some cases, the information at hand may be insufficient to decide the circumstances of death for a given casualty. In our experience, this was especially true in the early days of the “Intifada”, when many Palestinians were reported as having been killed, but with minimal information as to the circumstances surrounding their death. Even when detailed reports are available, however, there are cases in which it is impossible to draw a conclusion as to the combatant status of those killed – especially when each side’s report of a given incident is radically different from the other side’s, with both versions appearing reasonably credible.

The fact that all “Unknowns” are classified as Non-combatants serves as a “safety factor” in our analysis: Since a large number of Palestinian fatalities are classified as “Unknown” compared to a tiny number of Israelis, and since it is reasonable to assume that a substantial proportion of these “Unknowns” were in fact combatants, we can state with high confidence that our figures for total combatants among Palestinians are not inflated.

As an example of the results of our combatant-level classification system, the following two pie charts show the combatant status of Israeli and Palestinian fatalities:

Palestinian Fatalities by Combatant Level
Fig. 1. Breakdown of Palestinian fatalities by Combatant Level

Israeli Fatalities by Combatant Level
Fig. 2. Breakdown of Israeli fatalities by Combatant Level

2.3 Incident Types

As mentioned above, low-intensity conflicts are typically a mixture of many different types of incident. In order to make some sense of the confusion, we have categorized incidents at two levels: Incident Type and Attack Type. Incident Type refers to the general nature of what happened, and includes the following possibilities:

Terror Attack

Guerilla Attack

Attempted Infiltration

Crossfire

Violent Clash

Riot/Violent Demonstration

Roadblock Confrontation

Counter-Terror Operation/Interception

Initiated Military Operation

Targeted Killing

Work Accident

Internecine Violence

Unclaimed Killing

Unrelated Accident

Unknown

Attack Types are used to provide more detail as to exactly what form of violence took place, and are in turn divided into several “meta-attack-types”:

Bombings (non-suicide)

  • Bombing
  • Car Bomb
  • Letter Bomb
  • Bomb Threat

 

Suicide Bombings

  • Suicide Bomb
  • Suicide Car Bomb

 

Armed Assaults

  • Shooting
  • Ground-to-Ground Missile
  • Artillery
  • Vehicle Attack
  • Hand Grenade/RPG
  • Infiltration
  • Incendiary Device
  • Mortar Attack
  • Helicopter Missile

Kidnappings

  • Kidnapping
  • Hostage-taking
  • Hijacking
  • Lynching

 

Unknown/Other Attacks

  • Chemical Attack
  • Unknown

 

“Cold Weapon” Attacks

  • Knife Attack
  • Stoning
  • Arson
  • Vandalism

Incident Type and Attack Type classifications are highly useful in answering “nuts and bolts” questions about a low-intensity conflict, such as “How many people were killed in suicide bombings?” and the like. Combined with other criteria, such as demographic categories (see below), they can provide some highly interesting and unexpected results.

2.4 Demographics

The age and gender of each person killed have turned out to be two of the most useful pieces of information in our database. Not only are these data easily obtained and uncontroversial (unlike Combatant Level, for example); they have yielded some very interesting results.

We treat victims’ ages in three ways: For most of our demographic analyses, we use five-year age categories – i.e. 0-4 Years, 5-9 Years, and so on. For our analysis of the deaths of young people, we need a more detailed view – so we look at age year-by-year. And to compare the impact of the conflict on specific population sectors, it has proven useful to work with specific age categories:

  • Children: Ages 0 through 11
  • Adolescents: Ages 12 through 17
  • Young Adults: Ages 18 through 29
  • Adults: Ages 30 through 44
  • Mature Adults: Age 45 and over*
  • Adolescents Plus Young Adults: Ages 12 through 29

Age and gender analysis has yielded some of the most surprising results for the “al-Aqsa Intifada”. For example, age distributions comparing combatants with non-combatants are very different for the two sides:

Age Distribution of Israeli Fatalities

Fig. 3. Israeli combatant and noncombatant fatalities, broken down by five-year age categories. Note the very regular narrow distribution of ages of combatants, compared with a broader and “sloppier” spread for the ages of noncombatants

 

Age Distribution of Palestinian Fatalities

Fig. 4. Age distribution of Palestinian combatant and noncombatant fatalities. Note that while the spread of ages for Palestinian combatants is broader than for Israeli combatants, the age profile of Palestinian noncombatant fatalities is much narrower and more regular than is the case for Israeli noncombatants.

The detailed breakdown of childhood and adolescent fatalities by age and gender again showed patterns that we had not expected beforehand:

Young Non-Combatants Killed by Opposite Side
 

Fig. 5. Young victims of the conflict, arranged by age and gender. Note that fatalities are relatively uncommon among young children on both sides and of both genders. Israeli fatalities begin to increase at around age 13, with an essentially even balance between males and females. Palestinian fatalities show a tremendous increase between the ages of 10 and 13 among boys, but no increase with age among girls

One of our earliest findings was that Palestinian fatalities, including those classified as non-combatant, were overwhelmingly male, while Israel fatalities were much more balanced – about sixty per cent male:

Palestinian Noncombatants Killed by Israel

Fig. 6. Age distribution (in five-year brackets) of Palestinian noncombatants killed by Israel, with males and females separated. The preponderance of males is obvious; in addition, males show a very regular age distribution, while females do not

Israeli Noncombatants Killed by Palestinians

Fig. 7. Age distribution of Israeli noncombatants killed by Palestinians. The distribution is broad and “sloppy” for both genders; fatalities are essentially equally balanced between males and females, except for those between 20 and 54 years old

The combination of demographic information with Incident Type and Attack Type shows considerable promise as an analytical approach. While we have only begun to explore the possibilities of this type of analysis, our initial attempt yielded some surprising results regarding the demographics of Palestinians unintentionally killed in Israeli “targeted killings” compared to the demographics of Palestinian non-combatants killed in other types of incident:

 
  All Palestinian Noncombatants Killed by Israel

Fig. 8. The prevalence of “collateral victims” of Israeli “targeted killings” among Palestinian noncombatants killed by Israel

 

 

 

 

 

 


Figs. 9-10. The prevalence of “collaterals” among Palestinian females and young children killed by Israel. While Figure 8 showed that “collaterals” amount to just over five percent of all Palestinian noncombatants killed by Israel, these figures show that they include a disproportionate share of female and young victims

 

 

 


Figs. 1 1-12. Another look at the comparative demographics of “collaterals” versus other noncombatant Palestinians killed by Israel. Among unintended victims of “targeted killings”, some 40% were female – comparable to the percentage of females among Israeli noncombatants killed by Palestinians. But among Palestinian noncombatants killed in other types of incident, the number of females is much lower

 

Figs. 13-14 . The same type of comparison gives a similar result for young children: they make up a large proportion of the unintended victims of “targeted killings”, but a very small proportion of Palestinian noncombatants killed in other types of incident.

2.5 Time

Another basic area of analysis is the way in which a low-intensity conflict can change over time. It is quite common to talk and write about these conflicts as if they were single, unchanging events; but in fact they show significant change over time, and it is quite typical for them to consist of a number of fairly distinct phases, with different rates of death and injury, and even differences in the demographic characteristics of the victims. These phases are signaled by various events: mediation efforts, truces, major attacks or incursions, or even significant outside events such as the 9/11 attacks. (The latter attacks, for example, marked the onset of the third phase of the “al-Aqsa Intifada”, the most chaotic nine months of the conflict to date.) Of course, nobody in a position of authority officially announces that Phase X of a given conflict has ended and Phase Y has begun; only somewhat after the fact is it apparent that a particular event has actually changed the nature of the conflict.


Fig. 15. Change in death rates of noncombatants on each side, separated by phase and normalized per standard 30-day month


Fig. 16. Change in the percentage of young males among noncombatants killed on each side, separated by phase

3 Classification of Victims and Incidents – Issues, Dilemmas, and Solutions

3.1 Avoiding bias

In order for the results of a low-intensity-conflict database to be meaningful, data must be gathered and classified in ways that minimize bias and inaccuracy. This is not always as easy and straightforward as it might seem: The information available on incidents and casualties is often unreliable, incomplete, and sometimes contradictory; and some of the classification criteria involve the exercise of judgment, which invites unconscious bias even if those performing the classification are attempting to be scrupulously fair. It would perhaps be best if these databases could be created and maintained by completely disinterested outsiders; but, sadly, the disinterested are generally uninterested in committing the requisite time and effort to investigating someone else’s conflict. (In any case, the truly disinterested investigator is probably mythical.)

While the problem of bias is not completely solvable, there are a number of steps that can be taken to minimize the possibility of biased or unreliable results:

  • From the outset, it must be made very clear to everyone who works on a low-intensity-conflict database that the goal of the project is an accurate picture of the conflict, rather than to prove one side or another “correct”.
  • Data must be gathered from a variety of sources, including “enemy” media, non-governmental organizations, and so on, as well as “friendly” media. Reports in “friendly” media that favor the investigators’ side in the conflict must be treated with an extra degree of skepticism, since the human tendency is to grant extra credibility to favorable news.
  • Data should be gathered, entered into the database, and categorized by one team, and a separate team should perform all analysis. By separating these functions, one avoids contaminating the judgment of the person entering data with thoughts of “How will this look on my graphs?”
  • Categories should be defined clearly and specifically, to leave as little as possible to the discretion of the person assigning categories. We have found it helpful to test our criteria on difficult or strange hypothetical cases, to help establish the boundary conditions for category assignment. *
  • Categorization should as much as possible be based on physical rather than ethical or political criteria. As discussed below, while we do track incident-level responsibility (i.e. who started any particular incident), virtually all our analysis is based on casualty-level responsibility – meaning that we track responsibility for fatalities based purely upon who killed whom, not upon why a particular person was killed. This approach avoids the infinite regress of “who started first” – where each side claims that its own actions are a just and proper response to the other side’s previous actions.
  • “Safety reserves” (such as our categorization of all Palestinians of “Unknown” combatant level as non-combatants) should be created wherever categories unavoidably involve the use of judgment.
  • The more that can be done with the least controversial data – such as the age and gender of victims – the better. To the extent that we can learn interesting things from simple demographic data, we significantly reduce the possibility that bias will color our results.
  • Raw data and categorizations of each incident and casualty should be made available to the general public. By making a database inquiry function available on a website, the investigator makes it plain that the general public does not have to take the published aggregate statistical results on faith. In addition, skeptical (and energetic) members of the public may find and report genuine problems with the categorization of specific incidents and casualties, improving the quality of the data. (Sadly, our experience has been that while many people accuse us of bias simply because we are Israeli, virtually none of them make the effort to check the accuracy of our work.)
3.2 Civilians and non-combatants

Media reports frequently discuss the casualties of low-intensity conflict in terms of the number of “civilian fatalities” on each side. We have deliberately avoided this usage. In any conflict between a country with conventionally-organized military and police forces and an opposing force mostly composed of non-uniformed “irregulars”, the state actor’s forces cannot avoid killing a disproportionate number of “civilians” – since even their most deadly opponents are usually not members of an official military or security force, and in many cases have perfectly respectable “day jobs”. Further, people on either side of a low-intensity conflict may act in different capacities at different times – for example, many members of Palestinian security forces combine their official service with membership in one or more unofficial groups such as Hamas or the various arms of Fatah. When Palestinians in this situation have killed Israelis, they have generally done so in their “civilian” capacity.

At first glance, it should be easier to determine which state-actor fatalities are “civilians”. However, even here the distinction between “civilians” and members of official security forces paints a somewhat distorted picture. A substantial number of Israeli fatalities, especially those killed inside “Israel proper”, have been members of the civil police, or noncombatant members of the Israel Defense Forces such as office workers and mechanics. By most internationally accepted definitions, such individuals are considered to be non-combatants. (See, for example, the U.S. State Department’s definition of “noncombatants” in their “Patterns of Global Terrorism” reports. It is worth noting, however, that we differ from the State Department in our definition of terrorism: They define terrorism as attacks against non-combatant targets, while we define it as attacks against civilian targets.)

As a result of all these factors, dividing a low-intensity conflict's fatalities into “civilians” and “non-civilians” over-emphasizes the “civilian” status of many of the non-state side’s victims, and to a degree distorts the significance of the state actor’s fatalities as well. At best, such categorization paints an inaccurate picture of the conflict; and in some instances, those who use these categories are clearly being disingenuous in claiming the deaths of active militants as “civilian casualties”.

For this reason, we chose to classify those killed by their actual combatant status, according to the criteria laid out in the “Combatant Level” section above. While this method requires a degree of judgment in categorizing those killed, it offers some hope of making sense of an asymmetrical conflict; whereas the alternative system, while easier to apply, cannot provide meaningful results.

3.3 When a terrorist’s not engaged in his employment

The purpose of using our “combatant/non-combatant” criteria is, in essence, to produce a fair picture of the extent to which each side is limiting its attacks to legitimate targets – that is, targets that are properly considered military or quasi-military. Thus, as mentioned above, an armed civilian is considered a non-combatant – until s/he draws a weapon; and an armed soldier on duty is a combatant, even if he was asleep at his post and never fired a shot. But terrorists (or guerilla fighters – in the “Intifada” both types of attack are carried out by the same organizations) present a problem: Are they combatants when they are not actively involved in carrying out an attack?

This question triggered an internal debate. If an army general were killed in his home, we would normally classify him as a Uniformed Non-Combatant. But if the head of a terrorist organization were killed in his home, we were classifying him as a Full Combatant. This struck some of us as fundamentally unfair. Finally, we decided that our initial approach was correct – or at least more correct than any other approach we could think of – because it came closest to meeting our fairness criteria. As with so many other problems in dealing with low-intensity conflict, the difficulty here stems from the conflict’s asymmetry. IDF generals spend a large amount of their time at their job, just as other soldiers do. During these periods, they are considered combatants even if they are not actively fighting. In essence, the nature of being a soldier is that one spends a large percentage of one’s time being a legitimate military target. Terrorists and guerilla fighters – and even more so the leaders of terrorist/guerilla groups – do not “report for duty” as soldiers do; they perform no “routine patrols”, and in general do not allocate a regulated portion of their time to being official “military targets”. Since so little of their time is spent in actual “military” activity (and none at all for the leaders of terrorist groups), we have to consider them as essentially “full-time combatants”. (This is similar to the status of undercover espionage agents in enemy territory, who are legally considered to be full-time combatants even when they are doing nothing overt to harm their target country.)

This is not a perfect solution to the problem, of course. It illustrates the fact that creating a “level playing field” for analyzing a fundamentally asymmetrical conflict is a goal to which we aspire, but which we can never fully achieve.

3.4 Assigning responsibility

As mentioned above, we assign responsibility at both an incident and casualty level. In both cases, we use “responsibility” in a strictly physical sense. Thus, an arrest attempt is the responsibility of the security forces making the attempt, without regard to the guilt or innocence of the person they are attempting to arrest.

Incident-level responsibility can present some interesting borderline cases, particularly in regard to attacks that are foiled before they can be carried out. Once a terrorist or guerilla reaches his target and begins his attack, the incident is clearly “his”. On the other hand, if security forces receive advance warning of an attack and are able to intercept the attacker before he has reached his target, responsibility for the incident rests with the security forces. What happens, though, when an attacker reaches the area of his target, but is intercepted before he can succeed in carrying out his attack?

Our practice here is to judge based on whether security forces intercepted the attacker based upon advance warning of the attack, or whether they were alerted by the actual approach of the attacker (who may have set off alarms cutting through a fence, for example). If security forces were responding to an alarm the attacker set off, the incident-level responsibility rests with the attacker even if he never succeeded in firing a shot or triggering an explosive charge.

Our statistical analysis of the “al-Aqsa Intifada” has so far made use almost entirely of casualty-level responsibility: who killed whom, without regard for who was responsible at the incident level. This approach enables a less “politicized” approach, particularly in analyzing situations, such as violence at riots, in which it is impossible to determine who is responsible for escalating the incident to the point where lives were lost.

4 Technical Issues – Basics, Bells and Whistles

4.1 Platform choice

We chose to build our database using standard, “off-the-shelf” tools. The database runs on Microsoft Access; but while data-entry screens have been built using Access-specific features, all the database tools for analysis have been built using SQL commands. This approach has the advantage that it enables future portability to other platforms with minimal inconvenience, since SQL – unlike Microsoft Access’s application-building tools – is a standardized, multi-platform language. In future, we are likely to transfer the database to Microsoft SQL Server, with Access retained as a data-entry front end if possible.

A large suite of SQL queries (on the order of 50 queries) was created to make data available for graphing and analysis. A series of four Microsoft Excel spreadsheets extract data from the database using these queries, and then process the query results in preparation for graphing. Another set of Excel spreadsheets performs the actual graphing; this division avoids problems we experienced with Excel crashes due, apparently, to the complexity and size of the spreadsheets. Separate graphing spreadsheets also make it easier to produce differently formatted graphs and charts based on the same data and computations.

Our website is based on Cold Fusion; this allows the entire website, including almost all article text, to be database-driven. Cold Fusion allows us to make on-line database query functions available to the public, as well as a summary Breakdown of Fatalities screen which presents the most recent figures from our on-line database.

4.2 Database structure

The “foundation” tables in our database are Incidents and Casualties. In order to facilitate the required range and complexity of queries we require for analysis, we have added a rather large number of “support” tables:

  • Age (in five-year brackets)
  • AgeBrackets (e.g. Adolescents, Adults)
  • AgeHR (“high-resolution” – that is, one row in the table for each year of age)
  • AttackTypes and MetaAttackType (described above, under “Incident Types”)
  • CasualtyTypes (various levels of injury, from “lightly injured” to “killed”)
  • CombatantLevel (described at length above)
  • ConfidenceLevels (1 = extremely low, 3 = questionable, 5 = extremely high)

· Countries

  • Gender (male / female / unknown)
  • IncidentTypes (described above)
  • Months (one row for each month of the conflict; used by time-series SQL queries)
  • Organizations (shared with our other terrorism databases)
  • Side (Israel / Probably Israel / Palestinian / Probably Palestinian / Unclear / None. This table is used in assigning responsibility to incidents and casualties. In practice, we treat the “probable” assignments of responsibility the same as the definite ones.
  • Targets and MetaTarget (respectively, specific targets such as Bank or Marketplace, and general target categories such as Transportation, Civilian Personnel, and Military)

A few relatively unimportant tables have been left out of this list.

4.3 Summary statistic generation infrastructure

To enable rapid, automated, and flexible generation of aggregate and computed statistics for multiple time periods, the database includes a powerful table-configured summary statistic generator. This generator consists of a series of SQL queries that make use of the following tables:

  • Intervals : This table includes one row for each time interval for which summary statistics will be generated, such as the entire conflict, each calendar year, and the various phases that have been designated. Intervals may be flagged as “open”, meaning that their ending date is automatically extended each time summary statistics are generated.
  • SummaryInstructions : This table tells the SQL queries what calculations to perform. Up to 40 counter variables and 30 computed variables can be defined. For counter variables, we can specify the following criteria:
    • Variable Number (1-40) and Name
    • From Table (Casualties or Incidents)
    • Age Bracket (uses the brackets defined in the AgeBrackets table)
    • Gender (uses the genders defined in the Gender table)
    • Combatant (combatants, non-combatants, or all)
    • Side Responsible (casualty level responsibility, Israeli / Palestinian / Unknown)
    • Nationality (Israeli / Palestinian / Foreign)
    • Incident Type (from the IncidentType table)
    • Attack Type (from the AttackType table)
    • Organization (applied to the organization responsible for the incident when From Table is Incidents, or to the organization a casualty belonged to when From Table is Casualties)

For computed variables, we can select any two counter variables (“X” and “Y”) that we have defined using the criteria listed above, and perform one of the following operations on them:

  • Percentage computes X as a percentage of X plus Y.
  • Ratio computes X as a percentage of Y.
  • Daily computes X as a daily rate over the length of the time interval.
  • Monthly computes X as a monthly rate over the length of the time interval, normalized to 30 days per month.
  • Yearly computes X as an annual rate over the length of the time interval.

The daily, monthly, and annual rate computations allow conflict phases of different lengths to be compared on an equal basis.

When the summary statistic generator is triggered, all the variables defined in SummaryInstructions are generated for all the time periods defined in Intervals; the results are stored in the SummaryData table, with one row for each Interval.

5 Conclusion

A well-constructed database application is one of the best tools available for understanding the complexity of low-intensity conflicts. However, the most meaningful and accurate results can be achieved only if a number of problems are addressed, not all of them technical:

  • The database must be sufficiently “rich” in detail that a large number of different criteria can be recorded and analyzed.
  • A substantial numerical “back end” is needed to process values from the database in order to find meaningful trends and relations, and to provide clear graphical data displays.
  • Categories must be carefully and precisely defined, especially for politically sensitive issues like the combatant level of casualties.
  • Those designing the database, administering the project, and analyzing the data must maintain a strong and consistent commitment to accuracy – even when the results will make their own side in the conflict look less than perfect. If the results produced cannot withstand the most skeptical scrutiny, the entire exercise will be a waste of time and effort.


[1] Maj. Hong Kian Wah. “Low-Intensity Conflict,” Journal of the Singapore Armed Forces, Vol 26 n3. July-September 2000. Ministry of Defense, Singapore. This brief article gives a good summary of current thinking on the subject, emphasizing the limitations of conventional military approaches in fighting low-intensity conflicts.

[2] See, for example, Lara Sukhtian and Josef Federman, “Children under 17 caught in Mideast crossfire,” Associated Press. Houston Chronicle, 5 March, 2005. This piece, while far better than most media coverage of the Israeli-Palestinian “numbers game”, entirely neglects the issue of gender in analyzing “Intifada” mortality statistics.

* As the author would otherwise be entering this category during the course of the year 2005, it is possible that the minimum age for “Mature Adult” will be preemptively increased to forestall this eventuality.

* One rather gruesome hypothetical case was this: A pregnant woman attempts a suicide bombing, but is identified by security forces before she can set off her explosives. In order to prevent her from triggering her explosives, a policeman fires at her; the bullet sets off the explosive. Which side, then, is responsible for the death of her unborn child? After some deliberation, we decided that in this case the would-be suicide bomber is responsible, even though she did not in fact set off the explosives – since she placed the explosives in such a way that they would cause the unborn child’s death upon detonation, and one way or another they were almost certainly going to detonate. The security forces, however, would be responsible for the woman’s death, even though she was intending to blow herself up and thus is held responsible for the incident itself. Had the unborn child been killed by gunfire, rather than the explosion of the bomb, the security forces would be responsible for both deaths.