What Makes Discourse Coalitions Coherent? (T)ERGM Evidence from the Nord Stream 2 Debate in Germany

Research on coalition-building in politicized policy debates often explains alignment through preferences, institutions, or framing, yet the structural foundations of coalition cohesion are frequently treated descriptively rather than tested. Discourse coalition approaches are useful for mapping who aligns with whom around shared frames and arguments, but they rarely evaluate which endogenous network dependencies and which exogenous affinities jointly generate cohesive coalitions, or whether these drivers change as phases of contestation shift. This paper asks what makes discourse coalitions structurally cohesive and whether the relative importance of actor-based affinities (including actor type and policy position) and endogenous structural dependence differs across phases of the debate, using the German Nord Stream 2 debate as an empirical setting. The analysis combines discourse network analysis with exponential random graph models (ERGMs) and a temporal extension (TERGM). Temporally ordered public statements are coded into an actor–concept network and projected into a sequence of one-mode actor–actor congruence networks with valued, symmetric ties capturing discursive alignment. Congruence ties are constructed as similarity in actors’ concept endorsements, allowing discursive alignment to be analyzed as a weighted network while retaining temporal ordering. Phase-specific ERGMs (and a TERGM across phases) model coalition structure under tie interdependence and test mechanisms central to network theories of coalition formation: (1) assortativity/homophily by actor attributes and policy position, (2) triadic closure capturing clustering and coalition reinforcement, and (3) degree heterogeneity capturing hub formation and unequal visibility. The temporal component targets timing by assessing whether these effects are stable or phase-contingent across three analytically defined debate phases. Robustness checks address operationalization-driven artifacts via alternative similarity specifications and phase scaling choices for valued congruence ties. The empirical focus is the Nord Stream 2 debate in Germany, linking energy security, EU market regulation, and environmental transition in a single infrastructure project. The dataset comprises an originally coded set of temporally ordered statements in German public debate on Nord Stream 2 between 2015 and 2022, drawn from four high-profile newspapers: Frankfurter Allgemeine Zeitung (FAZ), Süddeutsche Zeitung (SZ), Die Welt, and Handelsblatt (approximately 680 coded statements by 80+ actors). Three phases correspond to the lifespan of the pipeline: EU-level regulatory contestation, US sanctions, and domestic German certification. Actors include political parties, government-related institutions, business and industry organizations, and civil society groups. Preliminary descriptive analyses indicate persistent polarization between pro- and anti-Nord Stream 2 camps alongside a small set of structurally central actors. Initial model fits are suggestive that closure and degree heterogeneity improve fit beyond attribute-based mixing, with indications of phase-contingent variation as the debate shifts from regulation-focused contestation to more security-salient framing. By combining discourse network analysis with ERGM/TERGM inference on valued alignment ties, the paper moves beyond descriptive coalition mapping to test structural mechanisms and their phase-contingent variation in a high-salience energy policy debate.