Theory-based Measurement of Varieties of Power Using a Novel Semi-supervised IRT Model

Duke University, Durham NC

Investigators

Abstract

The concept of state power has been invoked to explain everything from decisions to begin or end wars to trade policy to decisions nations make within their own borders. Yet, despite the prevalence and importance of the concept, it has been challenging to measure power in any consistent manner. What set of characteristics might compel a nation to go along with the preferences of other nations? And can we measure those characteristics ahead of time, in order to better predict the behavior of states as they respond to their own and other states’ absolute and relative power? Such predictions are essential to securing the national defense. This project will produce a theoretically-informed set of measures of power by using new semi-supervised machine learning techniques on a wide range of data sources gathered for the purpose. That set of measures will aid policy-makers in assessing the power dynamics between states, helping decision-making and enhancing security. In both scientific and policy circles, a wide range of theoretical conceptions of power are employed, most notably hard and soft power. Measurements of power, however, typically only capture some of its theoretical uses, while at the same time including aspects that might be unrelated to the use of power in question. When international relations scholars discuss “power,” for instance, they could be referring to very specific types of power, and their conclusions, measures, and models may not carry over to other types. For example, military power has been operationalized by the size of the army and the Composite Index of National Capability (CINC) score, but neither are likely useful as measures when elements of soft power are in play. This project offers several improvements in the ability to measure power. One, it provides a series of theoretically-informed measures of types of power. Two, it gathers a large dataset of sources that inform those different measures of power. Three, it provides a flexible tool—a semi-supervised machine learning algorithm making use of a modified Bayesian Item Response Theory model—to transparently construct alternate measures of power. That opens the door to studies that consistently assess across contexts the role of different types of power in international relations and allows stakeholders to communicate clearly about the aspects of power most relevant to them. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.