Empirically Predicting International Diplomatic Decisions Regarding the Libyan Civil War

The evolution of international political systems is heavily influenced by the choices made by diplomatic decision makers. Improved predictive modeling of these decisions could significantly contribute to improved prediction of political as well as military processes. Much of the conventional work on political prediction frames political data in the form of events sorted into a taxonomy of types. In this work the type of event I focus on recognition of the Libyan National Transitional Council (NTC) by members of the international community. The NTC was the main opposition authority in the 2011 Libyan Civil War. The recognition of authorities like the NTC can have both political and economic impact on the course of the conflict and its aftermath. The specific predictive task I focus on is that of estimating the time to recognition for a selection of countries. In contrast, existing work has mainly focused on predicting the occurrence events within a fixed time horizon. I experimented with both parametric and non–parametric approaches using both country–specific data (like indicators of economic development), general temporal data (like text features from international news), and dyadic relationships between countries (like trade relationships). Overall, I found predictive power through the influence of dyadic relationships between countries. Specifically, certain models based on dyadic relationships predict recognition with about the same accuracy as some baseline models. The models with dyadic relationships can ultimately be considered superior because they do not presuppose knowledge of the underlying distribution of event times as the baseline models do. However, once reframed as a classification problem rather than a regression, the effect of the dyadic relationships becomes even more salient. The results strongly suggest that while events like recognition are difficult to predict because they depend on human agents, each diplomatic decision is heavily influenced by pre–existing diplomatic relationships. The results also point to the ontological heterogeneity of the data that describes sociopolitical processes and the resulting difficulties of handling it in a classically inductive manner. Thus, the present work provides strong motivation for exploring alternative methods, such as those of Statistical Relational Learning, to make inferences on such structurally complex datasets.