TL;DR
This paper introduces a model of social coordination where agents can choose both their actions and their limited network connections. They find that when agents can only form a limited number of links, the efficient (payoff-dominant) coordination strategy is more likely to emerge, unlike in unconstrained scenarios where the risk-dominant strategy often prevails. This suggests that network constraints can be a powerful driver of efficient social outcomes.
What Problem Does This Paper Address?
The paper tackles a fundamental question in social and economic behavior: how do groups of individuals coordinate their actions when there are multiple possible outcomes, and how does the structure of their interactions influence this coordination? Specifically, it addresses the puzzle of why, in many coordination games with multiple Nash equilibria, societies often converge to inefficient outcomes (risk-dominant conventions) rather than efficient ones (payoff-dominant conventions). While existing models often assume agents can interact with anyone (unconstrained networks), this paper questions whether this assumption is realistic and explores how limiting the number of connections an agent can maintain (constrained interactions) might alter coordination dynamics.
Methods
The authors employ a co-evolutionary model combining a 2x2 coordination game with a network formation process. Agents simultaneously decide on their action (e.g., adopting a technology standard) and whom to form costly links with. A key innovation is the "constrained interaction" aspect, where each agent can support at most k links. They use a noisy best-response learning dynamic, where agents update their strategy (action and links) based on their perceived best response in the previous period, with a small probability of random mutation. To analyze the long-run outcomes, they identify absorbing sets (states from which the system cannot escape without mutations) and use concepts from stochastic stability (specifically, the radius-coradius theorem) to determine the set of Long Run Equilibria (LRE). They also analyze variations with convex linking costs, heterogeneous constraints, and frictions in link adaptation.
Datasets
Not discussed in the paper.
Key Insights and Ideas
- Constraints Drive Efficiency: The core insight is that limiting the number of connections an agent can maintain significantly shifts the likelihood of reaching efficient coordination outcomes. This is because a small group of agents adopting the efficient strategy can "pull" others towards it by becoming attractive connection targets.
- "Voting with Your Feet" in Networks: In constrained networks, agents can effectively "vote with their feet" by choosing to link up with those playing the efficient strategy. This is more potent than in unconstrained networks where the sheer number of potential connections can dilute the influence of a small efficient group.
- Cost Matters, But Constraints Matter More: While linking costs are important, the degree of constraint (k) plays a crucial role. Even with low linking costs, if k is sufficiently small, the payoff-dominant convention can be selected.
- Escape vs. Entry: Constrained interactions make it easier for agents to switch to the efficient convention (easier entry) but harder to leave it once established (harder escape), thus stabilizing efficient outcomes.
Key Results
The paper provides a detailed characterization of the long-run equilibria (LRE) based on linking costs (γ) and the maximum number of links (k). For a 2x2 coordination game with a payoff-dominant equilibrium (B,B) and a risk-dominant equilibrium (A,A), they find:
- Low Linking Costs (0 ≤ γ ≤ d): The LRE depends on thresholds for k. For small k, only the payoff-dominant convention (B,B) is selected. For intermediate k, both conventions can be LRE. For large k (approaching unconstrained), the risk-dominant convention (A,A) becomes more likely.
- Intermediate Linking Costs (d ≤ γ ≤ c): Similar patterns emerge, with thresholds shifting based on γ.
- High Linking Costs (c ≤ γ ≤ a): The payoff-dominant convention (B,B) is selected for a wider range of k values, and the risk-dominant convention is less likely to emerge.
- Generalization: The paper suggests that these findings extend to general mxm games with a payoff-dominant strategy, especially when k is less than half the population size.
Limitations and Future Work
- Passive Payoffs: The model assumes that passive payoffs (from being linked to) do not depend on the active player's action choice. The authors suggest extending the model to incorporate situations where passive payoffs do depend on the active player's action.
- Symmetric Links: The model focuses on directed (asymmetric) links (like Twitter). Future work could explore models with symmetric links, where consent from both parties is needed.
- Frictions in Link Adaptation: While the paper touches on frictions (agents not changing all links at once), a more in-depth analysis of how different levels of friction affect outcomes is suggested.
- Costly Passive Links: The current model assumes passive links are free. Incorporating costs for passive links is another avenue for future research.
- Heterogeneous Constraints: While explored, the impact of more complex distributions of link constraints on coordination could be further investigated.
Summary updated