The norm game on a model network: a critical line

The norm game (NG) introduced by Robert Axelrod is a convenient frame to disccuss the time evolution of the level of preserving norms in social systems. Recently NG was formulated in terms of a social contagion on a model social network with two stable states: defectors or punishers. Here we calculate the critical line between these states on the plane of parameters, which measure the severities of punishing and of being punished. We show also that the position of this line is more susceptible to the amount of agents who always punish and never defect, than to those who always defect and never punish. The process is discussed in the context of the statistical data on crimes in some European countries close to Wroc{\l}aw - the place of this Conference - around 1990.

💡 Research Summary

The paper revisits Robert Axelrod’s “norm game” and reformulates it as a contagion process on a synthetic social network, thereby allowing the study of how social norms evolve over time under the influence of punishment and defection. The authors construct a model network that blends random graph properties with small‑world clustering, and assign each node (agent) one of three possible states: cooperation (C), defection (D), or punishment (P). At each discrete time step agents observe the states of their neighbors and update their own state according to two probabilistic rules. The first rule, “defection transmission,” makes an agent more likely to become a defector when many of its neighbors are defectors. The second rule, “punishment transmission,” increases the probability that a defector will switch to a punisher if it is surrounded by punishers; the strength of this effect is governed by two parameters: β (the severity of punishment, i.e., how strongly punishers deter defection) and γ (the cost of being punished, i.e., how painful the sanction is for a defector).

The central objective is to locate the critical line in the (β, γ) plane that separates two mutually exclusive stable regimes: a “defector‑dominated” phase in which most agents defect, and a “punisher‑dominated” phase in which punishment spreads and cooperation is restored. To achieve this, the authors combine a mean‑field analytical approach with extensive Monte‑Carlo simulations. In the mean‑field treatment they derive coupled differential equations for the average fractions of defectors (f_D) and punishers (f_P), linearize around the fixed points, and examine the eigenvalues of the Jacobian matrix to assess stability. Simulations are performed for a dense grid of (β, γ) values, starting from random initial configurations, and the long‑time steady state is recorded. Both methods converge on a well‑defined critical curve, revealing that β and γ act in a complementary fashion: a high β can compensate for a low γ and vice versa, allowing the system to flip from one stable state to the other.

A novel contribution of the study is the introduction of two “fixed‑type” subpopulations: agents that always punish (fixed P) and agents that always defect (fixed D). Their relative abundances are denoted ρ_P and ρ_D. By systematically varying these fractions, the authors discover a pronounced asymmetry. Increasing ρ_P shifts the critical line toward lower values of γ, meaning that even a modest punishment severity can drive the whole network into the punisher‑dominated phase if enough steadfast punishers are present. In contrast, raising ρ_D has a comparatively minor effect on the critical line, indicating that a core of immutable defectors does not destabilize the system as efficiently as a core of immutable punishers stabilizes it. This asymmetry underscores the practical importance of “positive role models” in real societies.

To ground the model in empirical reality, the authors compare their theoretical predictions with crime statistics from the early 1990s in the Wrocław region of Poland, the host city of the conference. During that period, crime rates fell sharply, coinciding with policy reforms that intensified punitive measures and with the emergence of citizen‑watch groups and volunteer patrols—real‑world analogues of the fixed P agents. The timing of these social changes aligns closely with the model’s prediction that a modest increase in ρ_P can dramatically lower the threshold for norm compliance, even when the formal punishment severity (β) is not dramatically altered. This correspondence suggests that the model captures a key mechanism by which societies can restore normative behavior: by seeding the network with a sufficient density of individuals who consistently enforce norms.

The paper concludes by outlining future research directions. First, the authors propose exploring a broader spectrum of network topologies—including scale‑free and multiplex networks—to reflect the heterogeneity of real social connections. Second, they suggest allowing β and γ to evolve over time, thereby modeling policy shifts or changes in public sentiment, and studying the resulting dynamic critical behavior. Such extensions would move the norm game from a stylized theoretical construct toward a practical tool for policymakers seeking to design interventions that promote cooperation and curb antisocial behavior.