That question has no single right answer. As purveyors of early warning, however, we have to choose a definition in order to conduct and describe our risk assessments. If we aren’t very specific about what we’re attempting to foresee, we can’t build sound statistical models, and we can’t properly direct the attention of the experts in our opinion pool or assess the accuracy of their forecasts. So, following trends in scholarship on the topic, we have found our way to a definition of mass killing that focuses on deliberate and sustained attacks on noncombatant civilians who belong to a certain communal or political group.
One aspect of our definition that is not self-evident is our decision to restrict our attention to violence within countries and not between them. Our statistical risk assessments focus on state-led mass killings, which we understand as situations in which state security forces or closely allied domestic militias do the killing on the government’s behalf. Many of our opinion-pool questions widen that aperture to consider the risk that rebel groups and other non-state actors might be the perpetrators. Even then, however, we guide forecasters not to consider the risk that foreign militaries will perpetrate atrocities within the borders of the country in question. In neither case does our early warning system consider the risk that national military forces will perpetrate mass killings against citizens of another state on that other state’s territory.
This decision does not involve a value judgment about the moral or practical significance of atrocities perpetrated during wars between states, foreign occupations, and other international military operations. The Holocaust spanned many countries in Europe, and those murders were no less horrible when committed outside of Germany than in it, by Nazi forces or their local collaborators.
Instead, this decision was driven by practical motivations. The Early Warning Project has limited bandwidth and high standards. We cannot effectively assess risks of every variety of deadly atrocity in all places all the time. So, we have chosen to focus our efforts, and our expert forecasters’ attention, on the form of mass killing that has been the most common by far in the post-World War II era. Since 1945, foreign invasions and wars between states have become extremely rare, making statistical modeling of mass killings in those conflicts difficult. Wars within states have been far more common. Intra-state wars usually have some international dimensions to them, but state security forces or other armed groups operating on their own territory do most of the fighting and killing.
Under these conditions, we have chosen to restrict our view to the prospect of mass killings attached to conflicts within states and not between them. If and as the project grows, better data on become available, and—we hope not—conflict between states becomes more common, we may reconsider that choice. In the meantime, we thought it important to clarify this aspect of our assessments and to explain its origins.