Quantitative studies of international conflict face challenges with inconsistent findings and poor forecasting. This article proposes a conjecture that causes of conflict are typically small effects, but large when the probability is high. We test this using a neural network model designed to capture these features.

Neural Network Approach:

The authors developed a novel statistical model based on their conjecture about international conflict causality. The model specifically incorporates key elements identified in their hypothesis regarding conflict causes being small and ephemeral except when ex-ante probability is high.

Key Findings & Implications:

* Our neural network model significantly outperforms previous forecasting methods, providing more accurate predictions of international conflict.

* This improvement occurs with minimal trade-off compared to simpler standard models used in the field.

* The conjecture explains several observed patterns in existing literature without contradicting empirical evidence.

Real-World Significance:

The article demonstrates a practical application for improving predictive capabilities in forecasting major international conflicts, offering researchers and policymakers more reliable insights into conflict dynamics.