Last week the website for the Global Registry of Violent Deaths (GReVD) was launched. GReVD will be a single catalogue of every violent death coded by time and location, led by a consortium of prominent data providers and experts and convened by the Stockholm International Peace Research Institute (SIPRI) and the Brookings Institution. The mission for GReVD is born out of the United Nations Agenda 2030, and more specifically, UN Sustainable Development Goal 16.1: ‘significantly reduce all forms of violence and related death rates everywhere’. The main objective of GReVD is to create a registry for every violent death and thereby improve the precision of the current reported numbers of violent deaths—which are all estimates. This blog post emphasizes the importance of counting violent deaths and outlines some of the main challenges to do so.
Why count violent deaths?
The rationale for counting violent deaths is simple: if rates of violence and death are accurately recorded, then resources can be targeted at priority areas to reduce violence and death rates and improve the quality of life. Moreover, the violent death of a human being is the most extreme consequence of insecurity and therefore, violent deaths are a useful proxy measure for insecurity in both conflict and non-conflict settings. Fundamentally, counting violent deaths is a moral, legal and humanitarian imperative. However, while the rationale for counting violent deaths is clear, the actual counting is a hugely complex task riddled with difficult challenges at every stage of the process.
Definitional challenges
Broadly speaking, the challenges of counting violent deaths can be grouped into definitional, political, technical, and data processing challenges. Definitional challenges concern establishing what is meant by ‘violent deaths’. This is not as straightforward as it might sound. Once we begin to unpack the concept of violent deaths it becomes clear that there are no clear distinctions. Separating violent deaths from non-violent deaths is not an exact science. Rather, there are many types, patterns and dynamics of violence. To most people, violent deaths would include—at a minimum—intentional homicides and direct conflict deaths. Many would also regard assassinations, extra-judicial killings, and deaths caused by landmines and explosive remnants of war to be violent deaths. Some might even include the use of deadly force by police and legitimate acts of self-defence as violent deaths. However, should capital punishment, unsolved political disappearances, prison deaths or indirect deaths as a result of war also be included in the definition? Definitions matter and reaching a common understanding of what is meant by ‘violent deaths’ is essential. The international community requires consistent and comparable data over time to make evidence-based decisions and to measure the impact of interventions aimed at reducing violence.
Political challenges and biases
Violent deaths are recorded in two principal sources: official records (often referred to as administrative data) and media reports. The accurate recording and reporting of violent deaths, however, is affected by numerous political challenges and biases. Counting and classifying violent deaths is not a politically impartial exercise. Choosing what to count inevitably makes a political claim about what not to count. Statistics are highly influential, and there can be strong politically vested interests in manipulating data by, for example, inflating, downplaying or censoring numbers to generate certain narratives. Notable examples include the United States Department of Defense using a highly inflated figure to represent the number of communist fighters killed in the 1955–75 Vietnam War as a measure of their success. Battlefield reporting in the 1990–91 Gulf War, on the other hand, was largely censored in an attempt to prevent the recording of civilian casualties which might otherwise negatively impact the legitimacy of the intervention.
Official records are the most common sources of homicides. The Igarapé Institute plays an important role in collating official data produced by countries worldwide into a homicide monitor. Alternatively, media sources are the most common sources for conflict datasets, such as those produced by the Armed Conflict Location and Event Database (ACLED), the Global Terrorism Database (GTD) and the Uppsala Conflict Data Program (UCDP). Media reporting suffers from its own sets of biases as well as being affected by censorship. In general, media coverage increases when an event is large-scale, especially violent and when there are no competing major news stories. Likewise, violent deaths that occur in urban areas are more frequently reported than those in remote, rural areas. While local news outlets, in general, have greater coverage of rural and hard-to-reach areas than international media, the use of such sources for the monitoring of violent deaths is often restricted due to language barriers and the high cost of collection.
Technical challenges
The ability to capture events reported in local languages is a key technical challenge. Other technical challenges include reconciling different standards and counting rules used by actors that record violent deaths. Different approaches to police killings are a pertinent example of this. Countries differ in if and how they report police killings. This matters for counting violent deaths since—in some countries and jurisdictions—killings by police account for a considerable proportion of total killings. For example, in Brazil, the Dominican Republic and Jamaica between 10 and 20 per cent of violent deaths are caused by the police even though these are not captured in homicide statistics. A UN special rapporteur concluded that in Kenya, such killings are systematic, widespread, carefully planned and committed with utter impunity. In the USA, law enforcement agencies are not required to report on police killings. This means that significant numbers of what might arguably be considered violent deaths are excluded from official figures.
An additional technical hurdle is the fact that many countries have low statistical capacities which may impair their ability to capture violent deaths. According to the UN Office on Drugs and Crime, in 70 countries, neither the criminal justice system nor the public health system—the two main sources for official records of (violent) deaths—produce statistical information on homicides. Administrative data is particularly problematic in conflict-affected countries. Armed conflict can lead to a breakdown of the mechanisms that normally make death notifications available to a country’s statistical authority as well as weaken or disrupt public health and criminal justice institutions. Consequently, official data in conflict-affected countries is often missing.
Data processing challenges
Technical challengeslike these can have a considerable impact on data processing, including the coding of violent deaths by the different databases that collect official data and media reports. Paradoxically, despite the problems of a large number of unreported deaths, the sheer number of reported deaths create equally difficult data processing challenges. News reports, in particular, contain far more mentions of violet deaths than could be accurately monitored at the global scale by human researchers alone. To capture data at this scale requires machine reading and coding, but both methods come with their own difficulties. These include challenges around retrieving the reports that describe events into a research database; detecting and deduplicating copies of roughly the same report within the research database; and correctly extracting the sentences containing the events and the salient event attributes in each individual news report. Important work in this area is being done by the Cline Center for Advanced Social Research. Finally, there are numerous issues tied to integrating multiple databases, which often contain the same incident coded in different ways and with varying degrees of precision. Expertise to overcome this challenge is being led by the Center for International Development and Conflict Management.
Combined, these challenges can lead to overestimating or underestimating violent deaths. The best estimates currently available on the number of violent deaths are those produced by the Small Arms Survey’sGlobal Violent Deaths (GVD) project. These estimates are used by the GReVD consortium as a conservative estimate. GReVD consortium members are actively grappling with these challenges and the consortium—as a collective—aims to improve data on violent deaths, in alignment with UN Agenda 2030. GReVD will do this by applying new technologies to counting deaths, establishing standards and a common ontology to support better coding to improve human and machine reading methods. As methods improve, so will precision which will lead to better estimates. Ultimately, the success of the project will enable resources to be better targeted to reduce violence and death rates everywhere.
ABOUT THE AUTHOR(S)
Dr Caroline Delgado is a Senior Researcher and Director of the Food, Peace and Security Programme at SIPRI.
Last week the website for the Global Registry of Violent Deaths (GReVD) was launched. GReVD will be a single catalogue of every violent death coded by time and location, led by a consortium of prominent data providers and experts and convened by the Stockholm International Peace Research Institute (SIPRI) and the Brookings Institution. The mission for GReVD is born out of the United Nations Agenda 2030, and more specifically, UN Sustainable Development Goal 16.1: ‘significantly reduce all forms of violence and related death rates everywhere’. The main objective of GReVD is to create a registry for every violent death and thereby improve the precision of the current reported numbers of violent deaths—which are all estimates. This blog post emphasizes the importance of counting violent deaths and outlines some of the main challenges to do so.
Why count violent deaths?
The rationale for counting violent deaths is simple: if rates of violence and death are accurately recorded, then resources can be targeted at priority areas to reduce violence and death rates and improve the quality of life. Moreover, the violent death of a human being is the most extreme consequence of insecurity and therefore, violent deaths are a useful proxy measure for insecurity in both conflict and non-conflict settings. Fundamentally, counting violent deaths is a moral, legal and humanitarian imperative. However, while the rationale for counting violent deaths is clear, the actual counting is a hugely complex task riddled with difficult challenges at every stage of the process.
Definitional challenges
Broadly speaking, the challenges of counting violent deaths can be grouped into definitional, political, technical, and data processing challenges. Definitional challenges concern establishing what is meant by ‘violent deaths’. This is not as straightforward as it might sound. Once we begin to unpack the concept of violent deaths it becomes clear that there are no clear distinctions. Separating violent deaths from non-violent deaths is not an exact science. Rather, there are many types, patterns and dynamics of violence. To most people, violent deaths would include—at a minimum—intentional homicides and direct conflict deaths. Many would also regard assassinations, extra-judicial killings, and deaths caused by landmines and explosive remnants of war to be violent deaths. Some might even include the use of deadly force by police and legitimate acts of self-defence as violent deaths. However, should capital punishment, unsolved political disappearances, prison deaths or indirect deaths as a result of war also be included in the definition? Definitions matter and reaching a common understanding of what is meant by ‘violent deaths’ is essential. The international community requires consistent and comparable data over time to make evidence-based decisions and to measure the impact of interventions aimed at reducing violence.
Political challenges and biases
Violent deaths are recorded in two principal sources: official records (often referred to as administrative data) and media reports. The accurate recording and reporting of violent deaths, however, is affected by numerous political challenges and biases. Counting and classifying violent deaths is not a politically impartial exercise. Choosing what to count inevitably makes a political claim about what not to count. Statistics are highly influential, and there can be strong politically vested interests in manipulating data by, for example, inflating, downplaying or censoring numbers to generate certain narratives. Notable examples include the United States Department of Defense using a highly inflated figure to represent the number of communist fighters killed in the 1955–75 Vietnam War as a measure of their success. Battlefield reporting in the 1990–91 Gulf War, on the other hand, was largely censored in an attempt to prevent the recording of civilian casualties which might otherwise negatively impact the legitimacy of the intervention.
Official records are the most common sources of homicides. The Igarapé Institute plays an important role in collating official data produced by countries worldwide into a homicide monitor. Alternatively, media sources are the most common sources for conflict datasets, such as those produced by the Armed Conflict Location and Event Database (ACLED), the Global Terrorism Database (GTD) and the Uppsala Conflict Data Program (UCDP). Media reporting suffers from its own sets of biases as well as being affected by censorship. In general, media coverage increases when an event is large-scale, especially violent and when there are no competing major news stories. Likewise, violent deaths that occur in urban areas are more frequently reported than those in remote, rural areas. While local news outlets, in general, have greater coverage of rural and hard-to-reach areas than international media, the use of such sources for the monitoring of violent deaths is often restricted due to language barriers and the high cost of collection.
Technical challenges
The ability to capture events reported in local languages is a key technical challenge. Other technical challenges include reconciling different standards and counting rules used by actors that record violent deaths. Different approaches to police killings are a pertinent example of this. Countries differ in if and how they report police killings. This matters for counting violent deaths since—in some countries and jurisdictions—killings by police account for a considerable proportion of total killings. For example, in Brazil, the Dominican Republic and Jamaica between 10 and 20 per cent of violent deaths are caused by the police even though these are not captured in homicide statistics. A UN special rapporteur concluded that in Kenya, such killings are systematic, widespread, carefully planned and committed with utter impunity. In the USA, law enforcement agencies are not required to report on police killings. This means that significant numbers of what might arguably be considered violent deaths are excluded from official figures.
An additional technical hurdle is the fact that many countries have low statistical capacities which may impair their ability to capture violent deaths. According to the UN Office on Drugs and Crime, in 70 countries, neither the criminal justice system nor the public health system—the two main sources for official records of (violent) deaths—produce statistical information on homicides. Administrative data is particularly problematic in conflict-affected countries. Armed conflict can lead to a breakdown of the mechanisms that normally make death notifications available to a country’s statistical authority as well as weaken or disrupt public health and criminal justice institutions. Consequently, official data in conflict-affected countries is often missing.
Data processing challenges
Technical challenges like these can have a considerable impact on data processing, including the coding of violent deaths by the different databases that collect official data and media reports. Paradoxically, despite the problems of a large number of unreported deaths, the sheer number of reported deaths create equally difficult data processing challenges. News reports, in particular, contain far more mentions of violet deaths than could be accurately monitored at the global scale by human researchers alone. To capture data at this scale requires machine reading and coding, but both methods come with their own difficulties. These include challenges around retrieving the reports that describe events into a research database; detecting and deduplicating copies of roughly the same report within the research database; and correctly extracting the sentences containing the events and the salient event attributes in each individual news report. Important work in this area is being done by the Cline Center for Advanced Social Research. Finally, there are numerous issues tied to integrating multiple databases, which often contain the same incident coded in different ways and with varying degrees of precision. Expertise to overcome this challenge is being led by the Center for International Development and Conflict Management.
Combined, these challenges can lead to overestimating or underestimating violent deaths. The best estimates currently available on the number of violent deaths are those produced by the Small Arms Survey’s Global Violent Deaths (GVD) project. These estimates are used by the GReVD consortium as a conservative estimate. GReVD consortium members are actively grappling with these challenges and the consortium—as a collective—aims to improve data on violent deaths, in alignment with UN Agenda 2030. GReVD will do this by applying new technologies to counting deaths, establishing standards and a common ontology to support better coding to improve human and machine reading methods. As methods improve, so will precision which will lead to better estimates. Ultimately, the success of the project will enable resources to be better targeted to reduce violence and death rates everywhere.
ABOUT THE AUTHOR(S)