Abstract for: Extending Opinion Polls through the Combination of Analytical and Simulation Models

This paper describes a method for extending the accuracy of opinion polls by combining a simulation model with text analysis. Opinion polls are an important tool for gauging how societies interpret issues ranging from elections and policy to uprisings and regime change. While such polls produce detailed data, they are also costly and take time to field. The “Arab Spring” uprisings highlight the challenges this lag creates when governments collapse more quickly than the time required for polling and analysis. In relation to this rapid change is the rise of social networking and the instantaneous information it provides. Yet research indicates that opinions expressed in social media are often not representative of societies as a whole. To integrate these two very different data sources, we use events extracted from media to perturb a simulation of representative agents that were initialized using a prior poll. Agents update opinions using equations developed in a system dynamics model of social identity theory’s bounded confidence. We then evaluate the model’s performance using longitudinal opinion studies. Preliminary results suggest that the integrated results offer improvements over the sources used in isolation, which may help leaders do better at anticipating important societal changes.