Big Data, Selection Bias, and the Statistical Patterns of Mortality in Conflict

TLDR

Big‑data analytics assume near‑complete data, yet in policy and social contexts such assumptions are rarely satisfied. The study investigates how conflict‑killings data are generated and how this process shapes observed statistical patterns, concluding with policy recommendations. The authors use case studies from Syria and Iraq to illustrate how data‑generation processes bias observed patterns. They find that bias in conflict‑killings data can mislead policy decisions.

Abstract

The notion of “big data” implies very specific technical assumptions. The tools that have made big data immensely powerful in the private sector depend on having all (or nearly all) of the possible data. In our experience, these technical assumptions are rarely met with data about the policy and social world. This paper explores how information is generated about killings in conflict, and how the process of information generation shapes the statistical patterns in the observed data. Using case studies from Syria and Iraq, we highlight the ways in which bias in the observed data could mislead policy. The paper closes with recommendations about the use of data and analysis in the development of policy.