Clustering Methods for the Discovery of Advocacy Coalitions

Abstract

Network analysis provides a powerful toolkit to operationalize, measure, and model core theoretical concepts in research on the policy process. Prominent among these concepts is that of an advocacy coalition, or a group of actors who coordinate with one another to pursue common policy goals. How to identify advocacy coalitions is an important empirical question, and crucial to the testing of several theoretical frameworks including the Advocacy Coalition Framework (ACF). However, the best way to detect advocacy coalitions is still an open question in the literature. Based on most conceptual definitions, we argue that advocacy coalitions are defined not only by clustered attributes (such as shared incentives, beliefs, or policy preferences), but also by clustered link structures since common attributes raise the probability of coordinated action in the network. We propose how one specific clustering method, k-means clustering, may be used to identify clusters of collaborative networks that are relatively homogenous in terms of policy beliefs, and illustrate the importance of clustering simultaneously on collaborative ties and belief systems. This method is used to identify advocacy coalitions within the climate change policy subsystem in Switzerland (N = 70). Results suggest that the delineation of coalition membership is very sensitive to whether researchers choose to cluster on attributes, link structures, or the union of the two.