The Straw that Broke the Camel’s Back: Separating Close Calls from True Onsets in Conflict Prediction

The goal of conflict forecasting is to correctly anticipate events of armed violence. While the field made significant advancement, differentiating between close calls of conflict and true onsets has so far been overlooked despite its potential to improve our ability to predict new outbreaks of war. In this paper, we investigate why some countries with a high predicted risk of conflict experience limited violence only. Against this backdrop, we introduce the Averted Onset finder, a methodological innovation of the Shape finder approach, aimed at capturing the underlying processes of stalled escalation, where low-intensity violence does not surpass a certain level. We argue that the processes stalling the transition into more extreme conflict states, including conflict onset above a certain intensity, display recurring patterns that can be leveraged to enhance conflict prediction. Combining UCDP GED data with raw data on low-intensity violence, we apply a cascaded approach, training the Averted Onset finder on country-months with a high predicted risk of conflict escalation. Our preliminary results show a marginal improvement in out-of-sample performance. The significant reduction of false positives likewise reduces the model’s ability to anticipate true onsets. Nevertheless, we find regional differences in terms of where conflicts appear less likely to transition into higher intensity. This suggests that patterns from stalled conflicts can improve our ability to correctly predict true onsets, highlighting a gateway to explore, theoretically and empirically, what makes high-risk cases resilient against war.