Too many options: How to identify coalitions in a policy network?

For different currents in policy analysis as policy networks and the Advocacy Coalition Framework (ACF), identifying coalitions from policy beliefs and coordination between actors is crucial to a precise understanding of a policy process. Focusing particularly the relational dimension of ACF approaches linked with policy network analysis, determining policy subsystems from the actor collaborations and exchanges has recently begun offering fertile links with the network analysis. Studies in this way frequently apply Block Modeling and Community Detection (BMCD) strategies to define homogeneous political groups. However, the BMCD literature is growing quickly, using a wide variety of algorithms and interesting selection methods that are much more diverse than those used in the policy network analysis and particularly the ACF when this current focused on the collaboration networks before or after regarding the belief distance between actors. Identifying the best methodological option in a specific context can therefore be difficult and few ACF studies give an explicit justification. On the other hand, few BMCD publications offer a systematic comparison of real social networks and they are never applied to policy network datasets. This paper offers a new, relevant 5-Step selection method to reconcile advances in both the policy networks/ACF and BMCD. Using an application based on original African policy network data collected in Madagascar and Niger, we provide a useful set of practical recommendations for future ACF studies using policy network analysis: (i) the density and size of the policy network affect the identification process, (ii) the ''best algorithm'' can be rigorously determined by maximizing a novel indicator based on convergence and homogeneity between algorithm results, (iii) researchers need to be careful with missing data: they affect the results and imputation does not solve the problem.