Data on armed conflict reveals patterns in violent chaos

Chaos and uncertainty are hallmarks of armed conflict. But new research that ties together multiple aspects of political violence reveals universal dynamics in how conflicts emerge and expand. The work provides a statistical framework that could one day help anticipate deadly violence.

Cornell researchers analyzed more than 100,000 reports of armed conflict in Africa, compiled by the Armed Conflict Location and Event Data Project over 20 years. They studied various forms of armed conflict, from battles between nation-states and violence against civilians to riots and protests, in search of common themes in seemingly complex phenomena.

“Surprisingly, we found a universal pattern for how conflicts grow. Armed conflicts tend to follow remarkably simple statistical models,” said Eddie Lee, M.S. ’18, Ph.D. ’19, and lead author on the research. “People tend to think of conflict as very regional and context specific. We found emergent patterns that align with physical dimensions of space and time in a unified and predictable way.”

The paper, “A Scaling Theory for Armed Conflict Avalanches,” was published Oct. 28 in Physical Review E. Christopher Myers, M.S. ’88, Ph.D. ’91, adjunct professor of physics in the College of Arts and Sciences, is a co-author.

The researchers clustered together conflicts that occurred in close proximity and within a specific time frame. These clusters, called “conflict avalanches,” followed statistically relevant relationships between a few key variables: fatalities, duration, geographic spread and incident reports recorded in the Armed Conflict Location & Event Data Project (ACLED).

The datasets from ACLED, which tracks political violence and protests across the world, provide a granular look into deadly violence. The Cornell researchers found that conflicts of various scales in Africa followed consistent patterns such as power law scaling, in which a change in one quantity is related to a proportional change in another.

Conflicts are notoriously difficult to measure, especially their fatalities. The research reveals how some characteristics of armed conflict are unified within a mathematical framework. For example, if a conflict avalanche lasted 10 days and spread over 100 kilometers (more than 60 miles), one could use the framework to estimate the number of fatalities.

The average trajectory of conflicts revealed in the research differs from past theories that hypothesized conflicts spread like forest fires, where a long buildup, such as tinder accumulating on the forest floor, is followed by a blaze of activity that reaches an apex and then dies out. Instead, these avalanches are more mathematically “flat,” without the peaks of processes as in forest fires. In addition, such “forest-fire” models do not explain how regional variation in conflict intensity comes about and why such intensity would show the power law scaling that the researchers discovered.

According to Lee, robust conflict forecasting could eventually help inform public policy, leading to intervention strategies before conflicts gain traction.

“These regularities could help predict trajectories of ongoing conflicts or estimate important quantities like fatalities that are hard to measure,” Lee said. “Large datasets on human behavior have the power to revolutionize our understanding of society.”

The research was funded by grants from the National Science Foundation, the St. Andrews Foundation, the John Templeton Foundation and the Proteus Foundation.

Matt Hayes is associate director for communications for Global Development in the College of Agriculture and Life Sciences.

Read the story in the Cornell Chronicle.