Latent Estimation of Civil War Exposure from Conflict Event Data

• In Event: New Methods in the Study of Conflict Dynamics

Abstract

The paper presents a novel methodology to estimate subnational conflict exposure or intensity from conflict data using spatial and temporal weights. The analysis of event data from sources such as the Georeferenced Event Dataset (GED), the Global Terrorism Database (GTD), or the Armed Conflict Location Event Data Project (ACLED) has become an integral part of quantitative conflict researchers’ tool kit. These point-pattern event data are typically aggregated to larger polygons, because many covariates of interest such as indicators of economic development or the ethnic composition of the population operate at the level of subnational geographical areas such as districts, municipalities, or grid cells.

The current standard practice of summing all events that discretely fall within the borders of a polygon is highly problematic. The practice assumes that the impact of conflict events stops at the chosen geographic boundaries. In addition, the standard approach causes conflict exposure values to be highly sensitive to the exact choice of boundaries in particular when working with grid cells—the ubiquitous modifiable areal unit problem that is often mentioned but rarely tackled.

To address these shortcomings, I employ weights that allow the influence of a single conflict event to dissipate continuously from its occurrence over space and time. The speed of the dissipation is estimated via a latent variable model. Subnational conflict exposure is then computed as the sum of all weighted event counts in each month or year. The resulting estimates resemble a continuous heat map of conflict exposure that can be further disaggregated into the desired subnational units. The method is easy to implement and does not require extensive computational resources. The corresponding R code will be made freely available and allows for the custom adjustment of the speed of dissipation. The replication of seminal work using grid cell-level event counts illustrates the importance of the new continuous measure of conflict exposure for future research on conflict processes.

Author

• Therese Anders, University of Southern California