Dilemmas and trade-offs in the diffusion of conventions

TL;DR

The paper identifies three core dilemmas shaping the diffusion of conventions: (i) a triad of social, sequential, and contextual consistency; (ii) the balance between local network-driven coordination and global cultural or institutional coordination; and (iii) the trade-off between achieving optimal collective outcomes and incurring decision costs during conflict resolution. It introduces a broadly applicable statistical-physics framework, notably an inverse Ising approach and simulation-based inference, to recover underlying coordination structures, networks, and mechanisms of preference formation from empirical data on a sign convention in physics. Applying this to metric-signature usage among physicists, the study finds sequential consistency often dominates, with both local and global coordination contributing to partial alignment, and leadership (especially last-author influence) playing a key role in resolving conflicts, sometimes suboptimally. The work generalizes Lewis' account of conventions and provides a versatile toolkit for diagnosing why conventions fail to become universal norms in naturalistic settings, with implications for understanding epistemic cultures, disciplinary matrices, and the diffusion of scientific conventions.

Abstract

Outside ideal settings, conventions are shaped by competing processes that can challenge the emergence of norms. This paper identifies three trade-offs challenging the diffusion of conventions: (I) the trade-off between the imperatives of social, sequential, and contextual consistency that individuals balance when choosing between conventions; (II) the competition between local and global coordination, depending on whether individuals coordinate their behavior via interactions throughout a social network or external factors transcending the network; and (III) the balance between decision optimality (e.g., collective satisfaction) and decision costs when collectives with conflicting preferences choose a convention. We develop a broadly applicable statistical physics framework for measuring each of these trade-offs, which we then apply to a sign convention in physics. Our method can recover the structure of the underlying coordination game, the networks of social interactions involved, and the processes through which conflicts are resolved in collaborations. We find that the purpose of conventions may exceed coordination, and that individual preferences towards conventions are concurrently shaped by cultural factors and multiple social networks. Additionally, we reveal the role of leadership in the resolution of conflicts. Finally, this work provides a generalization of Lewis' account of conventions.

Figures (12)

• Figure 1: Left. Local coordination: nodes align to their neighbors through pairwise interactions. They may get stuck in a Nash equilibrium. Right. Global coordination: nodes are coordinated by a common cause transcending the graph structure (with some possible noise). Local and global processes generally predict different patterns of coordination, which means their contribution can be inferred from behavioral data. In each of these toy examples of local and global coordination, the Ising model correctly identifies the most likely process.
• Figure 2: Three trade-offs affecting conventions and their relationships.
• Figure 3: Importance of sequential and contextual consistency in scientists' behavior.
• Figure 4: Metric signature preferences in the co-author network. Each node is an author. Edges represent co-authorship relationships between authors. Nodes' colors indicate authors' preferences (pink for $-1$, green for $+1$). Only the largest connected component is shown.
• Figure 5: Illustration of local coordination in multilayered social networks. Nodes can be connected through different kinds of relationships (for instance, authors can be related via collaborations ($G$) or citations ($G^{\text{cit}}$)). In this diagram, patterns of coordination are better explained by the directed graph at the top ($G^{\text{cit}}$): (1,2) have imitated (4), and (3) has imitated (5). The Ising model correctly identifies the relevant social structure.