[Paper Review] Strategic tradeoffs in competitor dynamics on adaptive networks

Laurent Hébert‐Dufresne, Antoine Allard+3 more|arXiv (Cornell University)|Jul 15, 2016

Evolutionary Game Theory and Cooperation1 citations

TL;DR

This paper introduces a strategic tradeoff model in adaptive networks where competitors balance aggression (targeting opponents) and defense (targeting allies), leading to paradoxical dynamics like non-transitive competition. It demonstrates that in multi-player systems, no optimal strategy exists due to sensitivity to extreme strategies, and maps network structures to game-theoretic payoff matrices, offering a network-based lens for analyzing online political debates.

ABSTRACT

Recent empirical work highlights the heterogeneity of social competitions such as political campaigns: proponents of some ideologies seek debate and conversation, others create echo chambers. While symmetric and static network structure is typically used as a substrate to study such competitor dynamics, network structure can instead be interpreted as a signature of the competitor strategies, yielding competition dynamics on adaptive networks. Here we demonstrate that tradeoffs between aggressiveness and defensiveness (i.e., targeting adversaries vs. targeting like-minded individuals) creates paradoxical behaviour such as non-transitive dynamics. And while there is an optimal strategy in a two competitor system, three competitor systems have no such solution; the introduction of extreme strategies can easily affect the outcome of a competition, even if the extreme strategies have no chance of winning. Not only are these results reminiscent of classic paradoxical results from evolutionary game theory, but the structure of social networks created by our model can be mapped to particular forms of payoff matrices. Consequently, social structure can act as a measurable metric for social games which in turn allows us to provide a game theoretical perspective on online political debates.

Motivation & Objective

• To model competitor dynamics in social networks where strategies adapt based on network structure and node states.
• To investigate how tradeoffs between aggression (targeting opponents) and defense (targeting allies) shape competition outcomes.
• To explore the absence of optimal strategies in three-competitor systems and the influence of extreme strategies.
• To map emergent network structures to payoff matrices in evolutionary game theory, enabling inference of strategic dynamics from empirical data.
• To provide a framework for interpreting real-world online political debates through network-structural signatures of competitor strategies.

Proposed method

• Uses a directed stochastic block model (SBM) to represent strategies via a g×g density matrix P, where pij defines the probability of a directed link from state i to state j.
• Extends the voter model and Moran process to adaptive networks, where node state changes trigger link rewiring according to P.
• Imposes a strategic tradeoff constraint: pii + pij = 1 for all j ≠ i, forcing competitors to choose between targeting allies (defense) or opponents (aggression).
• Derives analytical solutions for fixed points in the prevalence simplex using matrix inversion and the Sherman-Morrison formula.
• Applies the model to empirical Twitter data by estimating user ideologies and constructing retweet matrices, then normalizing to infer strategy matrices.
• Uses coarse-graining to reduce multi-ideology systems to 3-competitor dynamics and simulates flows in the prevalence space to visualize long-term behavior.

Experimental results

Research questions

• RQ1 How do strategic tradeoffs between aggression and defense in adaptive networks lead to non-transitive or paradoxical competition dynamics?
• RQ2 Does an optimal strategy exist in a three-competitor system, and how does the presence of extreme strategies affect outcome stability?
• RQ3 Can the structure of social networks emerging from strategic interactions be mapped to specific payoff matrices in evolutionary game theory?
• RQ4 To what extent can observed network structures in online political debates be used to infer the underlying strategic behaviors of competing ideologies?
• RQ5 How do different topics (political vs. non-political) shape the resulting network structure and strategic dynamics in online discussions?

Key findings

• In two-competitor systems, a unique optimal strategy exists under the tradeoff constraint, but in three-competitor systems, no such optimal strategy exists due to non-transitive dynamics.
• The introduction of extreme strategies—even those with no chance of winning—can drastically alter competition outcomes, demonstrating high sensitivity to initial conditions.
• Non-transitive dynamics emerge, where A beats B, B beats C, and C beats A, resembling classic paradoxes in evolutionary game theory.
• The model’s fixed points can be analytically determined using matrix inversion, with a condition for degeneracy (fixed points on simplex edges) derived as ∑j pk/(2pj−1) = ∑j pj/(2pj−1) for all k.
• Empirical analysis of Twitter data shows that political topics (e.g., budget, marriage equality) generate strong homophily and echo chambers, while non-political topics (e.g., Winter Olympics) show minimal polarization.
• The resulting network structures from real data—such as core-periphery or fuzzy multipartite patterns—can be directly mapped to specific game-theoretic payoff structures, validating the model’s interpretability.