The politics of descriptive inference: contested concepts in conflict data

Descriptive research is sometimes understood as simply compiling and presenting objective facts, or 'telling it like it is.' We challenge this understanding, arguing that description involves a series of subjective, value-laden decisions that may reflect, reinforce, or alternatively undermine, existing narratives and power structures; accordingly, description is fundamentally, and unavoidably, political. We illustrate this argument with respect to descriptive research on violence against civilians by comparing how three descriptive research outputs-the Uppsala Conflict Data Program's One-Sided Violence, the Political Instability Task Force's Genocide and Politicide, and the Targeted Mass Killings datasets-define contested concepts relating to the distinction between combatants and civilians, identification of state actors, and intent. We demonstrate how differences in these definitions manifest in different descriptive inferences about violence in Burundi in 1993, and we discuss how an understanding of description as political relates to researchers' responsibilities as compilers and users of descriptive data.