Why do governments kill their citizens? Ghengis Khan oversaw massive state massacres of civilians, as did Emperor Chang Hsien-chung(17th century China), the Revolutionary Tribunal in France (1793-94) and countless other kings, sultans, emperors, and other political leaders. The many terrible episodes of contemporary state violence against civilians in such states as Indonesia, Kenya, Guatemala, Argentina, and others underscores the continued popularity of such domestic policies. But comparative research on state formation and consolidation between 1500 – 1900 suggests that rulers who provided for external defense and internal stability with less intrusive state behavior (including less arbitrary violence against civilians) and more secure property rights ultimately out-competed their less-restrained rivals. By the mid-1900s, rulers in most developed states did not arbitrarily kill their civilians as a matter of policy. In fact, these states have pressured member states of the international community into developing domestic and international laws which unequivocally require a state to assume responsibility for protecting its civilians from arbitrary, extrajudicial state violence. State violence is certainly not as acceptable as it once was, and most states have (at least on paper) foresworn use of violence against their civilian populations.
Nevertheless, some states continue to use violence, including extrajudicial executions, disappearances, and massacres, against their civilians. So, why do governments kill their civilians? This question has been inadequately addressed by existing research, largely due to data limitations. Data on civilian deaths caused by governments are hard to precisely compile because of inherent logistical difficulties in sensitive data collection and obstacles erected by governments themselves. In this paper, I first discuss measurement issues that structure how we count civilian deaths caused by government violence. Next, a brief review of existing datasets includes information about the definition and operationalization of relevant variables, the years covered and the countries included. These datasets have strongly influenced and shaped research into violent internal conflicts, with the current result being that we need better data sources on state violence against civilians in order to deepen our understanding of such state policies. Two new data sources, namely the Human Security Report and State Violence Against Civilians Dataset, have been designed to remedy this deficiency. To conclude, I briefly describe an on-going research project and its use of the State Violence Against Civilians Dataset.
2. Measuring Deaths
Civilian deaths caused by organized violence tend to be difficult to accurately count. Civilian deaths caused by state-organized violence often are impossible to count. Few organizations wish to publicize their atrocities and jeopardize their existence. Leaders of killing organizations, especially governments, have every incentive to make all evidence of civilian deaths disappear as quickly and permanently as possible. States that readily used violent policies against their civilian populations during most of the 20th century were often able to hide, destroy and deny reports of such violence. Poor government record-keeping efforts, tremendous distances from concerned groups (often located in other countries), and inefficient communication links aided and abetted killing governments. Even international politics, especially during the post-WW II era, helped convince involved and observer parties to sweep concerns about “internal affairs” off official agendas.
Until the 1980s, information about state-sponsored murder of civilians was disseminated primarily through existing state-to-state channels. News wire archives contain shockingly few reports about even major episodes of government violence. State leaders explained away state massacres as simply a necessary part of state consolidation rather than brutal attacks against civilian populations, and major international powers accepted these explanations. During the 1980s, non-governmental agencies began to play a far more important role in publicizing and documenting such violence, and refused to accept feeble explanations justifying state atrocities against civilians. Today, groups such as Amnesty International, Human Rights Watch, and the International Crisis Group generally provide very good evidence of state atrocities, especially when detailing major episodes of state violence. News wire services provide fairly wide-ranging daily updates on small and large episodes of state violence, although their findings cannot always be independently confirmed and often include contradictory information. Academic researchers are also an invaluable resource, especially those individuals who spend considerable time conducting research in countries that have or continue to experience state violence. By systematically working their way through primary documents in local languages, researchers make available important information about historical and contemporary events.
The quality of data on state violence for the 20th century and beyond is very uneven. While we may not be able to gather data on state violence between 1900 and 1980 that matches the quality of data we have for 1980 to the present, we can make sure that all data are properly specified. In some ways, measurement issues have created significant problems for research into state violence. The following section discusses some of these issues.
2.1 Measuring “Civilians”
Identifying a civilian (i.e. noncombatant) from a soldier (i.e. combatant) was once a relatively straight-forward endeavor. Today, the task can be near impossible when there is an internal conflict between the government and at least one organized rebel group that relies heavily, if not entirely, on local communities for resources to fight the conflict. Rebels often continue to live in their local communities and may work civilian jobs concurrently with rebel duties. There are many documented cases of members of government security forces working as rebels during the night and weekends. Even in refugee camps, both within and outside state boundaries, rebels have melted into refugee populations, buying time and resources needed to resume their military campaign. Given the liberal use of disinformation by both sides in an internal conflict, as well as the secrecy surrounding activities (especially those that result in civilian fatalities) that take place within civilian areas, properly discerning between civilian and soldier can truly be a tricky process.
For this project, the term “civilian” is synonymous with “noncombatant” and refers to any unarmed person who is not a member of a professional or guerrilla military group and who does not actively participate in hostilities by intending to cause physical harm to enemy personnel or property (Valentino et al, 2001). Furthermore, as Valentino et al (2001) note,
“. . . simply associating with combatants, providing food or other non-lethal military supplies to them, or participating in non-violent political activities in support of armed forces does not convert a noncombatant to a combatant. Because these activities pose no immediate threat of physical harm to combatants, those engaged in such activities deserve protection from killing – although they may be subject to judicial punishment (8).”
This distinction is important because in many states, governments wage violent campaigns against civilians who are merely suspected of being sympathetic to an enemy’s cause. Governments and rebels will go to great lengths to classify their victims as combatants, rather than civilians, even in the most messy internal conflicts like in Indonesia, Angola, Burma, Colombia, and many African states. The verbal gymnastics played by opposing sides in these conflicts is almost comical. Why would these sides care about how their victims are classified?
Due to the tremendous number of international treaties and conventions concerning the rights of civilians, both during peace and armed conflict, killing organizations might be concerned about potential punitive actions by the international community in response to human rights violations. Certainly the trials of former Yugoslavian and Rwandan government officials might provide some food for thought. But rebels and governments probably play these word games for a far more immediate audience. Leaders of successful organizations tend to have a relatively sophisticated understanding of how to use and abuse publicity, especially at the international level. At times, a violent organization may want to publicize its atrocities, perhaps to grab the international community’s attention and thereby gain a stronger domestic bargaining position within the conflict. Other times, the violent organization must downplay its atrocities, and justify its violence. Manipulating the media is just another military tactic to be used for a group’s advantage. Since killing organizations are so cognizant of differences between a “civilian” and a “combatant”, they will target and kill these people for different reasons. In order to understand the differences behind such a strategic choice, we must have data the separates civilians from combatants.
2.2 Categorizing Killing Methods
Besides sorting out civilians from soldiers, must victims be counted according to how they were killed? If the end result is death, does it really matter how a civilian was killed? Some researchers have argued that the end result, not the process, is the relevant piece of information. But the choice between policy options is a strategic decision, and each policy is chosen from very specific reasons. All policies that cause civilian deaths are not substitutable. Policies of forced migration and massacring an entire village may both produce very high body counts, but these policies require very different types of institutional support, can produce very different reactions from the international community, and will affect the internal political balance in very different ways.
Killing methods must be assessed according to two criteria. First, are civilians the actual target to be killed or are they killed as “collateral damage,” caught in the cross-fire of two warring sides? Second, if civilians are actually targeted, are they to be killed directly and immediately, or indirectly and over time? Answers to these questions are important. They help us understand very different strategic policy decisions that do produce the same end, but are most certainly carefully sorted through by policy makers.
2.2.1 Targets or Collateral Damage?
Not every civilian death is equal, if we are most concerned with assessing direct responsibility and explaining state policies that violently target civilians. Although any and every civilian death is a tragedy of a wasted life, the reasons for a person’s death are very different depending on whether that person is the direct target or simply caught in the cross-fire of two warring groups. For example, in the Aceh conflict in Indonesia, civilians are regularly killed as “collateral damage” – i.e. they become the unlucky victims of cross-fire between Indonesian security forces and GAM rebels. They are not the actual targets, but they pay a heavy price for being in the wrong place at the wrong time. But, in this same conflict, there are civilians who are targeted by Indonesian security forces for death. For these individuals, despite their status as non-combatants, their deaths are not the result of bad timing. Instead, they are a matter of state-approved policy implemented by the government’s security forces. Counting civilians killed as “collateral damage” in the same category as civilians killed as direct targets greatly misrepresents the reasons for these two different types of civilian casualties. In order to explain civilian deaths, we must first separate targeted from non-targeted victims.
Recently, some researchers and NGOs have focused attention on tactics by governments and rebels that directly target civilians. Oxfam argues that “[T]oday, in armed conflicts around the world: violence is directed overwhelmingly against civilians; violence against civilians is a deliberate strategy, not an accidental side-effect; the frequent purpose is to kill or expel civilians of another group (Cairns 1997, 6).” Azam and Hoeffler (2002) conclude that violence against civilians is motivated by military objectives. Hoffman (2003) agrees, and argues further that groups that do target civilians will specialize in tactics, so that specific violent tactics (e.g. amputations, family massacres, etc) become the “signature” tactic for which a group is well known. In order to properly understand the decision-making process behind state policies that directly target civilians, we must have data that separates all civilian deaths into the proper categories.
2.2.2 Death Through Direct or Indirect Methods?
Civilians have been killed many different ways. Most simply, we can say that civilians may be killed directly or indirectly. When civilians are killed indirectly, state policies may be the initial step onto the causal path leading to death, but they can rarely be cleanly attributed as the sole cause. Although millions of civilians have perished in labor camps, as a result of forced migration, or state-induced famines, there continues to be much argument over the state’s responsibility for these deaths. When civilians are killed directly by government forces, the causal path is much simpler. Time and time again, direct killings have proven much more difficult for governments to deny or explain away than have indirect killings. If for this reason alone, we must be willing to consider that leaders choose between these policies for very different reasons. In order to explain these decisions, we must divide direct from indirect killings.
2.3 Assessing Responsibility for Civilian Deaths
Governments can dedicate many different agencies to killing civilians. The following is a list of possible agents authorized by governments to kill civilians.
Military (but most commonly, Army troops)
directly, openly linked to a government
secretly, indirectly linked to government
Violent Arm of a Political Party
Research into how governments make assignments for implementing violent policies between possible agents is quite limited. James Ron (2003) argues that governments tend to assign the military to operations in “frontiers”, or peripheral areas, while regular police troops operate in “ghettos”, or center areas. In frontier regions, the military seems to have a freer, more brutal hand, whereas police tend to work with greater limits on the amount and type of their violent activities. Ron’s observations from Serbia and Israel correspond nicely with empirical observations of state activities in Indonesia and other southeast Asian countries. Extensive testing of such hypotheses is needed in order to tease out the factors that influence how governments choose between agents to implement violent policies.
The civilian population may have very different ties to each type of potential killing organization. In some states, the military is considered more trustworthy than the police (e.g. Mexico, Thailand), where as in other countries, the military is one of the most feared organizations (e.g. Indonesia). Still other governments prefer to farm out their violent policies to somewhat ad-hoc organizations, such as paramilitary groups (e.g. Kenya, Colombia), death squads and an armed faction of political parties (e.g. Zimbabwe). Often when the civilian government’s greatest threat is the military, the government will use another killing agent to complete its business while the military will conduct its own violence for its own purposes. The relationship between killing agent and civilian population requires further study. Additional factors that might influence the civilian-potential killing agent relationship include:
Level of organizational involvement in illicit business activities, organized crime and black market operations;
Level of organizational corruption and demands made to civilian population for daily transactions;
Killing tactics used by the killing agent, e.g. killing by machine-gun fire, machete, bombing, etc.
Gender dimension to the killing, i.e. are men generally killed while women are raped and maimed, but left to live?
In places where NGOs have replaced the state in providing many day-to-day basic living necessities, does this automatically cast government security forces into an adversarial position? What about situations in which there are peacekeepers, especially if they are there to protect the civilian population from state security forces or because state security forces cannot protect the civilian population from rebel security forces?
What are the demographic characteristics, e.g. ethnicity, socio-economic background, education level, etc, of members in killings organizations? Do they vary greatly from the civilian population in which the members work?
Unfortunately only limited research into these issues has been completed, partly due to data limitations on state violence against civilians.
3. Existing Datasets: A Brief Survey
What are the existing measures of civilian deaths caused by their own governments? One of the most widely cited resources is also one of the most underutilized resources. Rummel’s (1997) Statistics of Democide: Genocide and Mass Murder since 1900 purports to contain estimates, sources and calculations for democide committed by 141 state governments, as well as 73 quasi-state regimes and rebel groups from 1900 – 1987. Much of his data is available at http://www.hawaii.edu/powerkills/welcome.html. Rummel’s book includes a vast amount of information, but none is in a user-friendly format and has therefore not been widely used by other researchers. He is concerned with “democide” – a term that refers to the killing of civilians by an organized group, whether that be a state, quasi-state or rebel group. Democide, as a category, includes deaths that result from the following activities:
Prison/camps – People murdered in or dying as a result of incarceration in prisons or concentration/forced labor camps;
Forced labor – People who were murdered during or dying as a result of forced labor (includes that in forced labor camps);
Terror – Murder of specific individuals (unlike massacres);
Famine/disease – Regime intended deaths from starvation and disease;
Deportation – Murder during or deaths from deportation or explusion;
Genocide – People killed because of their religion, ethnicity, race, language, nationality, or other social group membership;
Prisoners of war – Prisoners of war murdered or dying in camps for lack of care;
Bombing – Noncombatants killed indiscriminately by bombing, shelling, torpedoing, germ warfare, or defoliation (1997, 343).
Rummel does attempt to break his collected numbers into these sub-categories, but the end result is imprecise and suspect. He provides a low, medium and high estimate for episodes of democide by state, but since each episode varies in length, it is impossible to construct precise, comparable time-series, cross-sectional data from his data.
Benjamin Valentino and Paul Huth assembled a new data set on state-sponsored mass killings, first presented in their article (co-written with Dylan Balch-Lindsay) “Draining the Sea: Mass Killing, Genocide and Guerrilla Warfare” (available at http://www.cbrss.harvard.edu/programs/hsecurity/papers/june/Valentino.doc) and now undergoing back-dating and revision. Mass killing is defined as “the intentional killing of a massive number of noncombatants (2001, 5).” To be included in the data set, the killing must be intentional, not accidental. Civilians must be the direct object of a policy that results in widespread death immediately and over time. There must be at least 50,000 intentional deaths over the course of 5 years or less. The victims must be noncombatants. According to this criterion, they identified 41 cases of state-sponsored mass killing from 1945-2000. For each incident of state-sponsored mass killing, they give a start and end year.
The State Failure Task Force’s project has collected data on revolutionary wars, ethnic wars, adverse regime changes and genocides/politicides from 1955 – 2001 (see http://www.cidcm.umd.edu/inscr/stfail/ for all data and codebooks). Data on genocides and politicides are most relevant here. The operational definition for genocides and politicides is
“Mass murder against unarmed members of a rebellious communal group. The promotion, execution, and/or implied consent of sustained policies by government elites or their agents that results in the death of a substantial portion of a communal group or a politicized non-communal group. In genocides the victimized groups are defined primarily in terms of their communal (ethnolinguistic, religious) characteristics. In politicides groups are defined primarily in terms of their political opposition to the regime and dominant groups. Victims are unarmed civilians, not combatants.”
To be included in the dataset, authorities’ complicity in mass murder that targets a group must be established. Only sustained episodes that last six months or more are included. The numbers of death are given as an annual number, coded as a scale from 0 – 256,000 deaths or more per year. The dataset includes a start and end date for each episode, the country, the scaled death magnitude and a brief description of the episode.
The Civil War Termination (CWT) dataset contains information on 83 civil wars between 1927-1992 within the international system (see http://www.rci.rutgers.edu/~licklide/). Cases are included only if they meet three criteria, including a 1,000 minimum battle death threshold per year. For each civil war, there is information on the start and end dates, length, termination mode, mediation variables, goals of fighters, genocide-politicde type, target population of genocide, low and high estimates of geno/politicde deaths, estimated war deaths, source of war death estimate, civil war after five years from settlement, issue of war, result of war and years of peace before next war. Much of the genocide and politicide information comes from the State Failure Task Force dataset.
Peter Brecke designed the Conflict Catalog (http://www.inta.gatech.edu/peter/) to be a comprehensive collection of violent conflict between 1400 AD – present. Violent conflict is defined as “[A]n occurrence of purposive and lethal violence among two or more social groups pursuing conflicting political goals that results in fatalities, with at least one belligerent group organized under the command of authoritative leadership. The state does not have to be an actor. Data can include massacres of unarmed civilians or territorial conflicts between warlords.” A violent conflict must result in a minimum of 32 deaths to be included. For each conflict, information on who, when, where and (whenever possible) the number of military and civilian fatalities is given.
The yearly publication World Military and Social Expenditures includes information on interstate warfare in the international system between 1900 and 1995. To be included, a war must produce an average of more than 1,000 deaths per year. For each war, the location and identification of conflict, dates (by year), and number of deaths, divided into civilian, military, and total, are listed.
Matthew White’s web site includes information about state-sponsored mass killings, available at http://users.erols.com/mwhite28/20centry.htm#war. Rather than presenting results from an original data collection effort, White instead simply lists estimates of death tolls according to the original source. He includes information about death toll estimates for both inter- and intrastate conflicts during the 20th century. For the most part, the estimates are left in aggregate form (i.e. grouping battle-deaths with civilian deaths), unless the original source disaggregated the numbers. He groups the conflicts according to number of estimated casualties, so the “big” conflicts are well represented but smaller killing events are not as detailed.
Jorge Restrepo, Mchael Spagat, and Juan F. Vargas compiled a new data on civilian fatalities from the Colombian civil conflict for 1988-2002 and introduce it in their article “The Dynamics of the Colombian Civil Conflict: A New Data Set” (available at http://personal.rhul.ac.uk/pkte/126/). Using daily articles from Colombian sources, especially Justicia y Paz and Noche y Niebla, information has been collected according to each violent event that meets specific criteria of motivation and group action (9-11). As a result, the dataset includes only “. . . those violent episodes that are part of the Colombian internal conflict and not cases of ordinary crime (12).” For each event, the information gathered includes: intensity (as measured by the number of individuals killed, injured, detained and retained), date, location (township and department), whether or not there was a clash, and if so, the groups involved; whether or not there was an attack, and if so, the type of attack and the group(s) responsible; killings and injuries (13).
Mohammad Zulfan Tadjoeddin generated a database on social violence in Indonesia for 1990-2001 (available at http://www.communalconflict.com). The main sources used were Antara, the national news agency, and the national daily, Kompas. The database covers 1990 – 2001, and 26 Indonesian provinces. However, violence in Aceh during the DOM (military operation zone) era (1989-1998) and violence in East Timor are not included (Tadjoeddin 2002, 2). The unit of observation is incident of violence, and he grouped incidents according to the type of violence. Only two categories deal with government/group violence, i.e. separatist violence and state-community violence. Separatist violence is “social violence between the state and the people of certain area, which is rooted in regional separatism, i.e. a movement motivated by the desire of people in certain areas to separate from Indonesia as a country. This social violence refers to the current conflicts in Aceh and Papua (Tadjoeddin 2002, 4).” State-community violence is “violence between the state and the community who are expressing protests and dissatisfaction against state institutions without any separatist motives (Tadjoeddin 2002, 4).” For each incidence of violence, the dataset provides information on the date of the violence, the location, the number of deaths and the material damages caused by the violence.
Malcolm Sutton first published his data on deaths caused by the Northern Ireland conflict in his book Bear in Mind these Dead . . . An Index of Deaths from the Conflict in Ireland 1969-1993. His data has been updated to include all deaths between 14 July 1969 and 31 December 2001 and is available at http://cain.ulst.ac.uk/sutton/index.html. The unit of observation is the person killed, and for each person, the dataset provides information on the date of death, status of person killed, organization responsible for the death, description of the incident and geographical location where incident took place.
Youcef Bedjoui, Abbas Aroua, and Meziane Ait-Larbi (eds) provide a dataset on massacre victims between 1992-1998 in their book An Inquiry into the Algerian Massacres. The raw data came from international and national news media, human rights NGO reports and interviews with victims (1999, 17). They organized their data around massacre events, with massacres defined as “victimization episodes where a number of unresisting civilians are killed indiscriminately (1999, 16).” Massacres are further categorized according to “selective mass victimization” (SMV) events in where there is random victimization within selective sub-groups of the civilian population, and “random mass victimization” (RMV) events in which there is random mass killing of people belonging to random sub-groups of the population (1999, 16). Perpetrator identity is not assigned; indeed the authors discuss difficulties in correctly identifying perpetrators and assessing responsibility (1999, 136-141). The dataset gives the date of the massacre, the location, the number of deaths and the number of injured.
Patrick Bell compiled data on deaths in Guatemala between 1960-1996 (the data is available at http://shr.aaas.org/guatemala/ciidh/data.html). He includes information on the number of victims, the location of the violence, the perpetrator (e.g. civilian, army, PAC, police, paramilitary group, URNG), and the type of human rights violation (e.g. tortured, kidnapped, killed, injured, disappeared)
4. How do these data shape our understanding of internal violent conflicts?
Internal violent conflict research projects that use these data sources share some important similarities. Since most of the data has been collected with reference to some sort of event (international war, civil war, insurgency movements, etc), most studies have tended to conceptualize violence against civilians as something that happens during a violent conflict. While this is certainly true, it is also clear that one-sided violence against civilians happens during times that, for one reason or another, do not qualify as an “event” and therefore are not included in many datasets. By constantly thinking about internal violent conflict this way, most researchers have conceptualized one-sided violence as a result of another type of conflict rather than as a state policy option in and of itself. There is very little discussion on the actual types of violent state policies available to leaders, why they are chosen and if they are substitutable.
Existing data also tends to be skewed towards episodes of state violence that resulted in huge numbers of civilian victims. Oftentimes, these episodes are well documented because immediately after the violence, there was relatively open access to primary sources (such as victims’ families or witnesses of one-sided violence), an existing media infrastructure to transmit information about the violence, and some sort of records or archives concerning basic demographic information such as birth and death rates, immigration and emigration rates, etc. While some mass-murdering governments (such as Nazi Germany) kept detailed records of their crimes, the prevalent tendency amongst such governments has been denial, denial, denial. Even in well-established, uncontested governments such as in Argentina, plausible deniability required that the secrets of the disappearances be buried away from most governmental branches. In newly independent states, unstable states and failing states, governments most generally have few if any records that are useful for reconstructing a state’s violence against its civilians. Since some of the best data on state violence concerns “mass murdering” governments, research has tended to conceptualize violent state policies as aberrations of normal state policies, rather than policy options that leaders can regularly compare against less-violent policy options.
5. New Data Sources
Despite the number of existing data sources that contain some information about directly targeted civilian casualties of state violence, there are major limitations on how we can use these data. Definitional differences and threshold exclusions preclude direct comparisons between many of the civilian death counts. In 1999, I began working on a data project designed to overcome some of these differences and combine existing data into a useable data source on government violence against civilians. In 2001, researchers with the Human Security Report (HSR), headed by Professor Andrew Mack, also turned their attention to remedying deficiencies in our information about government violence against civilians. Rather than relying on existing datasets, HSR researchers were able to initiate a new data collection effort. Thanks to an invitation from Kristine Eck, the HSR project leader, I began working with the HSR project in June 2004. Although we are working together on coding the data, we will use the raw data in different ways. The end products will complement one another, but available data will not be identical.
5.1 Human Security Report
Data collection for the Human Security Report is an extension of the Uppsala Conflict Data Project (UCDP). Whereas UCDP data focused exclusively on state conflict (i.e. at least one party to a conflict must be a state government), the Human Security Report reports data on non-state conflict and one-sided violence. The primary source for the data collection is Factiva, an online wire service that provides access to articles from such sources as Reuters, Agence France Presse, Xinhua, etc. The Human Security Report Project commissioned Doug Bond to adapt his automated data retrieval software (called IDEA) to meet its specific needs. In effect, the IDEA software runs targeted searches through all Factiva files and identifies articles that may have relevant information about incidents of non-state and one-sided violence with key-word searches. Once all articles published during a calendar year have been searched, the tagged articles are compiled into a single report file. HSR coders then manually read through each tagged article and record all relevant information about incidents of non-state and one-sided violence.
During 2003, HSR researchers coded all violence reported during 2002. Coding will continue indefinitely so that data on non-state and one-sided violence should be released along with state-conflict data each year. In order to harmonize the non-state and one-sided violence data with the state-conflict data, we began back-coding the data in June 2004. We plan to have the data for 1989-2001 coded completely by the end of 2004.
According to the HSR project, non-state conflict is “ the use of armed force between two organized groups, neither of which is the government of a state, which results in at least 25 battle-related deaths.” The operational definition is as follows:
1. Use of armed force: use of arms in order to promote the parties’ general position in the conflict, resulting in deaths.
1.1 Arms: any material means, e.g. manufactured weapons but also sticks, stones, fire, water, etc.
2. 25 deaths: a minimum of 25 battle-related deaths per year and per warring dyad.
2.1 Battle-related deaths: deaths directly related to combat between the warring parties.
3. Organized groups: must be either
3.1 a formally organized group; any non-governmental group of people having announced a name for their group and using armed force, or
3.2 any group who does not have an announced name, but who uses armed force and whose violent activity meets at least one of the following organizational requirements:
3.2.a there must be a clear pattern of incidents which are connected, or
3.2.b there must be evidence that violence was planned in advance
4.2 an internationally unrecognized government controlling a specified territory whose sovereignty is not disputed by another internationally recognized sovereign government previously controlling the same territory.
5. Government: the party controlling the capital of the state.
One-sided violence is “the use of armed force by the government of a state or by a formally organized group against civilians which results in at least 25 deaths. Extrajudicial killings in custody are excluded.” One-sided violence death totals only include civilian deaths, unlike the UCDP’s data on state conflict and the just-detailed non-state conflict. The operational definition of one-sided violence is as follows:
1. Use of armed force: use of arms in order to exert violent force, resulting in death.
1.1 Arms: any material means, e.g. manufactured weapons but also sticks, stones, fire, water, etc.
2. 25 deaths: a minimum of 25 civilian deaths per year and per actor.
3. Government: the party controlling the capital of the state.
4. Formally organized group: any non-governmental group of people having announced a name for their group and using armed force.
5. State: a state is
5.1 an internationally recognized sovereign government controlling a specified territory, or
5.2 an internationally unrecognized government controlling a specified territory whose sovereignty is not disputed by another internationally recognized sovereign government previously controlling the same territory.
6. Extrajudicial killings in custody: when the government of a state kills a person in its custody.
Custody: when the person is located in a prison or another type of governmental facility.
For the initial coding, information on any reported non-state and one-sided violence is recorded. The coders work with daily information, so there are often multiple reports, each filed by a different news service. The unit of observation is an incident of violence that produces at least one confirmed fatality. Each entry includes information on the following variables:
Date of incident
Date of report
Who committed the violence
Where the violence occurred
What happened (number of victims, how killed)
The raw data includes detailed information about the victims, the perpetrators, and the actual violence. For the Human Security Report project, much of this detail will not be included in the final product. The data will be published in the annual publication and will only include information about state and non-state actors who kill at least 25 civilians per year. For those groups who surpass this threshold, the HSR will give information about the location, parties, total fatalities and a description of the violence. The final dataset and conflict descriptions will be made available from the UCDP webpage, but the raw data will not be made public.
5.2 State Violence Against Civilians Dataset
The State Violence Against Civilians Dataset is designed to include information about all one-sided government violence that directly targeted and caused civilian deaths between 1946-2003. The data collection has proceeded in two phases. In the first phase, I collected data for 1946-1989 from three existing data sources: Rudoph J. Rummel’s Statistics of Democide (1998), the State Failure Task Force (SFTF) Genocide and Politicide data set (Harff 2004), and a data set on mass killing presented in Valentino et al’s article “Draining the Sea: Mass Killing, Genocide, and Guerilla Warfare” (2001) (see Appendices A, B and C for complete list of the countries and years included on each list). From these sources, I collected variables on the estimates for the number of civilians (non-combatants) killed directly by their own governments, such as during massacres, assassinations and death squads. I have also gathered information on where the killings took place and who carried out the killings (e.g. army, police, death squads, etc). Although I know that not every incident of state violence has been accounted for with these three sources, I believe that the three sources together represent a reasonable compilation of existing sources.
Unfortunately, a reasonable compilation is far from a perfect compilation. A quick comparison of the three sources will underscore why we must be skeptical of the data. Given the operational definitions of “genocide” and “mass killing” used by Valentino et al and SFTF, many cases included on Rummel’s list should not make either the genocide list or mass killing list. However, given Rummel’s inclusive definition of democide, all cases listed on the “genocide” and “mass killing” lists should be included in Rummel’s list. A comparison of these lists reveals that Rummel’s data do not include all known episodes of “genocide” and “mass killing” by state governments against their civilians. Valentino et al collected data on mass killing for 1946-1999. Standardizing the dates of the Rummel and Valentino data (i.e. 1946-1990 only), Valentino has 279 country years recorded with mass killing. Of those years, Rummel’s data includes only 163 of those country years, for an overlap of 58.4%. Since Rummel’s data should include information on each of the mass killing episodes included in Valentino et al’s list, the fact that the overlap is just over 50% does not inspire much confidence.
For the State Failure Task Force (SFTF), data on genocide and politicide episodes between 1950-2000 is included. After standardizing the dates again, the Rummel data is left with 636 country years with state killing while the SFTF data lists 214 country-years with a genocide/politicide event. Of these years, the two sources share 133 country-years, which indicates that Rummel’s data includes 62.1% of the country-years recorded by the SFTF data. By combining the three sources, the data for 1946-1989 can be used more confidently. Nevertheless, it is clear that this data may be problematic.
During the second phase, I am using the raw data from the initial HSR coding as the primary source for 1989-2003. If possible, given time and resource constraints, I plan to do a second round of coding with articles from 1980-1988. Although the quantity and (quite possibly) the quality of articles will diminish over time, the overlap will permit me to run important comparative tests between the data collected in phase I and that from phase II. Such testing could help demonstrate that, despite the many methodological questions about Rummel, SFTF and Valentino et al data (i.e. used in combination or considered alone), we can still use the data with confidence. This testing could also prove otherwise, but either way, such comparisons will tell us much about the comparability of the two data collection efforts.
The major difference between the data collected during the two phases is how the data was structured. For 1946-1989, the data is structured according to an event (e.g. state-sponsored mass-killing, genocide, politicide, or democide). It is not recorded as daily data, which is a double-edged sword. While I am fairly confident that this data represents quite well the episodes of one-sided government violence that produced a large number of victims, I am just as confident that episodes of lower intensity and magnitude are not well represented. I have relied heavily on other sources (e.g. NGO reports, academic research, etc) to research the events included in the Rummel, SFTF and Valentino et al datasets for additional information. I can not find daily information for most of these events, but I haven been able to collection more information about the actual states policies (i.e. did they cause civilian deaths immediately or over time), if the casualty estimates include civilians killed as collateral damage or only those directly targeted, what agent did the actual killing, the location of the killing, and other details. Even with the additional information, the level of detail about victims, location and perpetrators is abysmally low for some events. For this data, the best variables I can hope to generate are:
Absence or presence of one-sided government violence during the year.
Estimate on number of civilian fatalities during the year. This number may simply be the estimate of civilian deaths during the entire violent episode, divided by the number of years of the entire episode.
What state agent committed the violence, i.e. the police, military, paramilitary, death squads, etc.
Where did most of the violence occur, e.g. rural versus urban location?
In order to properly use this data with the data coded for 1989-2003, I will have to aggregate the 1989 – 2003 data in a similar fashion so that the variables are structured identically.
For 1989-2003, I can generate the aggregate the data into the four variables listed above, as well as create several other interesting variables:
Number of civilians killed on a daily basis.
Distance from capital city/major commercial center.
Linkage between killing agent and top government leaders (e.g. in Zimbabwe, the war veterans/land squatters are known to specifically support President Robert Mugabe, and not his political party).
Method of killing (e.g. machine-gun fire, machete, bomb, etc).
Gender of victims.
If nothing else, I will be able to use the entire dataset to test for general trends over time, and then break the dataset into sub-sets to test more specific hypotheses over a more limited time period. While not perfect or even ideal, the State Violence Against Civilians dataset and the Human Security Report data do offer a major improvement over existing data sources. Researchers interested in understanding why governments violently target their civilians, whether during peace or a declared conflict, will find these to be valuable new resources.
6. Current Research
Once the State Violence Against Civilians dataset is ready, I will use it to test existing explanations, or “common wisdom,” about why governments kill their civilians. The following is a list of hypotheses derived arguments about state violence.
H1: Democratic institutions constrain the use of violence against a state’s civilians, and therefore fewer citizens are killed by democratic regimes than by non-democratic regimes.
H2: Higher levels of ethnic, religious and/or linguistic factionalization in a state will increase the probability of state killing because targeting members of specific groups will be easier.
H3: A larger population will be correlated with higher levels of state killing.
H4:.As per capita wealth increases, the probability of state killing decreases.
H5: If a state has recently fought a civil war, there is a greater probability for state killing.
H5.1: If a state is participating in an interstate war, there is a greater probability for state killing.
H5.2: If a state is participating in an internal war, there is a greater probability for state killing.
H5.3: If a state is participating in an internationalized internal war, there is a greater probability of state killing.
H5.4: If a state is host to any type of war (civil or international) within its boundaries, the probability of state killing increases.
H6: If a state has only recently gained its independence, there is a greater probability for state killing.
H7: As a state’s level of international interdependence increases, the probability of state killing decreases.
Using the new time-series, cross-sectional data on civilian deaths caused by state policies that directly target civilian victims, I can test these hypotheses for all states within the international system between 1946-2003. For example, I can begin by testing for the correlates to the on-set of state violence that kills civilians by dropping all subsequent years of violence. With this new dataset, I can empirically test explanations against the universe of cases that have been proposed and tested (most commonly) within a single or a few in-depth case studies.
I plan also to test for “neighborhood effects” to see if domestic policy decisions are influenced by neighboring domestic policies. With the new data, I can map incidents of one-sided government violence to see if there are observable clustering effects. Next, using a binary matrix of states’ geographical connectedness/location, I can produce estimates of what countries are like to employ one-sided government violence during a specific year. I can use those estimates to forecast state violence over subsequent years and then match the forecast to actual events.
Using the same methods, I can also test for the interaction of neighborhood effects and variables that produced significant results in the first round of testing for the correlates to the on-set of state violence. If the interaction of neighborhood effects and these variables yield significant results, they will support the earlier results and help to better explain what conditions are necessary and sufficient to encourage or dissuade against the choice of one-sided violence.
7. Bibliography Azam, Jean-Paul and Anke Hoeffler. 2002. “Violence Against Civilians in Civil Wars: Looting or Terror?” Journal of Peace Research 39(4): 461-485.
Ball, Patrick. 1999. AAAS/CIIDH Database of Human Rights Violations in Guatemala (ATV20.1). http://hrdata.aaas.org/ciidh/data.html (1 August 2004).
BedJaoui, Youcef, Abbas Aroua, Meziane Ait-Larbi (eds). 1999. An Inquiry into the Algerian Massacres. Geneva, Switzerland: Hoggar.
Cairns, Edmund. 1997. A Safer Future: Reducing the Human cost of War. Oxford: Oxfam Publication.
Harff, Barbara (Principal Investigator). 2004. Annual Data on Cases of Genocide and Politicide, 1955 – 2002. Compiled for the US Government’s State Failure Task Force. http://www.cidcm.umd.edu/inscr/genocide/ (1 August 2004).
Hoffman, Danny. August 2003. “The Civilian Target in Sierra Leone and Liberia: Political Power, Military Strategy, and Humanitarian Intervention.” Unpublished paper presented at the American Political Science Association Annual Meeting, Philadelphia, PA.
Restrepo, Jorge, Michael Spagat, and Juan F. Vargas. 2003. “The Dynamics of the Colombian Civil Conflict: A New Dataset.” http://personal.rhul.ac.uk/pkte/126/ (1 August 2004).
Ron, James. 2003. Frontiers and Ghettos: State Violence in Serbia and Israel. Berkeley: University of California Press.
Rummel, Rudolph J. 1998. Statistics of Democide: Genocide and Mass Murder Since 1900. Munster: LIT.
Sutton, Malcolm. 1994. Bear in Mind these Dead . . . An Index of Deaths from the Conflict in Ireland 1969-1993. http://cain.ulst.ac.uk/sutton/index.html (1 August 2004).
Tadjoeddin, Mohammad Zulfan. 2002. “Database on Social Violence in Indonesia 1990-2001.” http://www.communalconflict.com (1 August 2004).
Valentino, Benjamin, Paul Huth and Dylan Balch-Lindsay. 2001 “”Draining the Sea:” Mass Killing, Genocide, and Guerrilla Warfare.” Unpublished paper available at www.cbrss.harvard.edu/programs/hsecurity/papers/june/Valentino.doc (1 August 2004).
Appendix A: Rummel’s Country-Year Observations of Democide, 1946 - 1990
Afghanistan 1978 - 1980
Albania 1946 - 1987
Angola 1975 - 1987
Argentina 1976 - 1982
Bangladesh 1972 - 1987
Brazil 1946 - 64
Brazil 1968 - 1972
Burundi 1965 - 1973
Cambodia 1953 - 1954
Cambodia 1967 - 1968
Cambodia 1970 - 1979
Cen African Rep 1979
Chad 1975 - 1983
Chile 1973 - 1974
Chile 1979 - 1987
Colombia 1948 - 1958
Colombia 1978 - 1979
Colombia 1981 - 1984
Congo Brazzaville 1964 - 1965
Congo Kinshasa 1977 - 1983
Cuba 1952 - 1978
Cyprus 1981 - 1985
Czechoslovakia 1948 - 1954
El Salvador 1975 - 1976
El Salvador 1979 - 1983
Equatorial Guinea 1968 - 1979
Ethiopia 1974 - 1987
France 1946 - 1948
Germany East 1946 - 1952
Guatemala 1954 - 1959
Guatemala 1966 - 1968
Guatemala 1977 - 1984
Guatemala 1987 - 1988
Guinea 1969 - 1976
Hungary 1946 - 1948
Hungary 1956 - 1957
India 1968 - 1985
Indonesia 1965 - 1966
Indonesia 1968 - 1969
Indonesia 1983 - 1984
Iran 1954 - 1976
Iran 1978 - 1987
Iraq 1958 - 1988
Korea North 1958
Korea South 1980
Laos 1960 - 1980
Mozambique 1975 - 1988 Netherlands 1946
Nicaragua 1946 - 1979
Nicaragua 1981 - 1983
Nigeria 1966 - 1970
Pakistan 1949 - 1956
Pakistan 1964 - 1965
Pakistan 1971 - 1972
Paraguay 1962 - 1980
Peru 1980 - 1987
Philippines 1972 - 1986
Poland 1948 - 1987 Portugal 1953
Portugal 1959 - 1962
Portugal 1971 - 1974
Romania 1946 - 1947
Romania 1966 - 1989
South Africa 1960
South Africa 1976 - 1977
South Africa 1985 - 1987
Sri Lanka 1983
Sri Lanka 1985
Sudan 1954 - 1972
Syria 1980 - 1982
Uganda 1971 - 1974
Uganda 1979 - 1986
United Kingdom 1948 - 1956
United Kingdom 1972
Uruguay 1973 - 1985
Vietnam 1976 - 1987
Vietnam North 1956 - 1976
Vietnam South 1955 - 1975
Yemen South 1970
Yugoslavia 1948 - 1950
Appendix B: The State Failure Task Force Genocide List, 1955-2000
Afghanistan 1978 - 1992
Angola 1975 - 1994
Angola 1998 - 2000
Argentina 1976 - 1980
Bosnia 1992 - 1995
Burundi 1965 - 1973
Cambodia 1975 - 1979
Chile 1973 - 1976
China 1966 - 1975
Congo Kinshasa 1964 - 1965
Congo Kinshasa 1977 - 1979
El Salvador 1980 - 1989
Ethiopia 1976 - 1979
Guatemala 1978 - 1990
Indonesia 1965 - 1966
Indonesia 1975 - 1992
Iran 1981 - 1992
Iraq 1963 - 1975
Iraq 1988 - 1991
Myanmar (Burma) 1978
Pakistan 1973 - 1977
Philippines 1972 - 1976
Rwanda 1963 - 1964
Somalia 1988 - 1991
Sri Lanka 1989 - 1990
Sudan 1956 - 1972
Sudan 1983 - 2000
Syria 1981 - 1982
Uganda 1971 - 1986
Vietnam South 1965 - 1975
Yugoslavia 1998 - 1999
Appendix C: Valentino et al Episodes of Mass Killing, 1946-2000