Multi-Source Assessment of State Stability
The wave of revolutions in the Arab world, commonly referred to as the Arab Spring, took the world by surprise. To some extent the September 2012 consulate and embassy attacks were also unforeseen. Despite the rich literature on interstate conflict, state stability, revolution and regime change these events could not be predicted nor fully accounted for by the existing theoretical traditions in the social sciences. Clearly, social media was touted as critical to these revolutions. Traditional media also gave voice to public concerns and provided critical information. However the role of media in fostering or mitigating or even providing insight into issues related to state stability is unclear. These recent events raise a number of questions about how access and usage of social media in comparison to traditional media can be used to promote change.
Our primary objective is to understand the way in which media, social and traditional, can be used to effect state stability or instability by individuals, groups and corporations. This projects uses data from news sources, twitter, and blogs, in conjunction with other state level indicators to examine the relative ability of these different mediate to:
- enable the diffusion of new ideas and actions that inhibit or promote violence,
- support new agendas,
- maintain or forge new alliances,
- forge or break trust,
- stabilize or destabilize situations,
- alter lines of power, and
- change an actors influentialness.
Forecasting the Arab Spring with newspaper data and agent-based modeling
Joseph, K., Carley, K. M., Filonuk, D., Morgan, G. P., & Pfeffer, J. (2014). Arab Spring: from newspaper data to forecasting. Social Network Analysis and Mining, 4(1), 1-17. doi:10.1007/s13278-014-0177-5
Agent-based simulation models are an important methodology for explaining social behavior and forecasting social change. However, a major drawback to using such models is that they are difficult to instantiate for specific cases and so are rarely re-used. We describe a text-mining network analytic approach for rapidly instantiating a model for predicting the tendency toward revolution and violence based on social and cultural characteristics of a large collection of actors. We illustrate our approach using an agent-based dynamic-network framework, Construct, and newspaper data for the sixteen countries associated with the Arab Spring. We assess the overall accuracy of the base model across independent runs for twenty different months during the Arab Spring, observing that although predictions led to several false positives, the model is able to predict revolution before it occurs in three of the four nations in which the government was successfully overthrown.