Endogenous Coalition Formation in Policy Debates

TL;DR

This paper addresses how coalitions form endogenously in policy debates by proposing three micro-level learning mechanisms—bonding, bridging, and repulsion—that shape how actors adopt beliefs from others. It formalizes these mechanisms within a dynamic relational-event framework and tests them on a nine-year German pension-debate dataset using a past-event sequence $E_{t-1}$ and a temporally decayed weight $w(t_e, t, T_{1/2})$. Empirically, all three mechanisms are positively and significantly associated with new statements, supporting a theory of endogenous coalition formation that explains both convergence within coalitions and polarization across them, with bridging acting as a counterbalance. The approach yields notable out-of-sample predictive power, illustrating how micro-level learning processes can generate macro-level discourse structures, and it opens avenues for applying such models to other policy domains and online deliberation contexts.

Abstract

Political actors form coalitions around their joint normative beliefs in order to influence the policy process on contentious issues such as climate change or population ageing. Policy process theory maintains that learning within and across coalitions is a central predictor of coalition formation and policy change but has yet to explain how policy learning works. The present article explains the formation and maintenance of coalitions by focusing on the ways actors adopt policy beliefs from other actors in policy debates. A policy debate is a complex social system in which temporal network dependence guides how actors contribute ideological statements to the debate. Belief adoption matters in three complementary ways: bonding, which exploits cues within coalitions; bridging, which explores new beliefs outside one's perimeter in the debate; and repulsion, which reinforces polarization between coalitions and cements their belief systems. We formalize this theory of endogenous coalition formation in policy debates and test it on a micro-level empirical dataset using statistical network analysis and event history analysis.

Figures (10)

• Figure 1: Illustration of policy debates as an event sequence of statements in news media articles. Each news article contains a number of text portions in which an actor expresses a policy belief, either by supporting or rejecting the belief. Different actors, beliefs, and signs are represented by colors. The actor--belief--agreement statement tuples can be put on a timeline, which forms an event sequence for statistical analysis.
• Figure 2: Illustration of three mechanisms of endogenous coalition formation. Bonding: An actor has a higher probability to adopt a specific policy belief with a specific agreement pattern if another actor who shared many belief--agreement combinations in the past event sequence also adopted the same belief with the same agreement pattern. Bridging: An actor has a higher probability of adopting a belief if many other actors with whom the actor shared at least minimal overlap in belief--agreement combinations in the past also adopted the same belief with the same agreement pattern. Repulsion: An actor is more likely to adopt a belief with a specific agreement pattern if the belief was previously adopted with a conflicting agreement pattern by many other actors who previously adopted conflicting belief--agreement combinations in the past event sequence. The three mechanisms together shape the formation and maintenance of coalitions.
• Figure 3: Frequency of statements per week. Election campaign highlighted.
• Figure 4: Estimates from the stratified Cox model. The basic (main) model in green; model with GOV interactions in blue. Bonding, bridging, and repulsion all have positive and significant coefficients. Note: Table \ref{['tab_rem']} in the Appendix reports the accompanying regression table.
• Figure 5: Marginal effects for interaction terms.