Testing the Advocacy Coalition Framework by Document Analysis (the Case of Russian Civil Society Policy)
Gregory Khvatsky, Dmitry Zaytsev, Valentina Kuskova, Polina LushnikovaWe present a methodology for identification and classification of policy actors. We used network analysis and deep learning-based named entity recognition on a computational cluster for actor identification, and Speaker-listener Label Propagation Algorithm to identify probable coalitions (and overlaps between them) among the identified actors. We test this methodology on the case of Russian policy towards civil society. The theory we have chosen is the Advocacy Coalition Framework, which is a public policy theory aimed at explaining the long-term policy change by understanding how and why people engage in policy-making. One of the key ideas of the theory is that people participate in policy to translate their beliefs into action, and then gather into advocacy coalitions based on the shared beliefs system. Identification of actors is one of the most fundamental issues in political science, as it is often the first step in the research process. Another problem of interest is the classification of the actors based on latent characteristics such as shared beliefs. Most of research papers apply qualitative methodology for both of the steps. In this paper, we try to do both of these steps using quantitative (computational) methods.