Modelling the coevolution of opinion dynamics and decision making in social dilemmas

1. Problem Statement

The paper addresses the fundamental challenge of understanding how cooperation emerges and persists in social dilemmas, specifically within the context of collective action. While classical game theory suggests that "defection" (free-riding) is the rational strategy for selfish individuals, real-world observations show widespread cooperation.

Existing models often treat actions and opinions as decoupled or assume a one-way relationship (e.g., opinions dictate actions). The authors identify a gap in unified mathematical models that capture the interdependent coevolution of:

1. Actions: Strategic decisions to cooperate or defect in a Public Goods Game (PGG).
2. Opinions: Continuous beliefs or attitudes regarding the preference for cooperation.

The core problem is to model how these two variables evolve simultaneously, where an individual's action influences their opinion (via a desire for self-consistency) and their opinion influences their action, all within a networked social structure.

2. Methodology

A. The Coevolutionary Game Framework

The authors propose a novel game-theoretic model defined on a weighted network $G = (V, E, W)$.

• State Space: Each player $i$ has a strategy $z_i = (x_i, y_i)$, where:
  • $x_i \in \{0, 1\}$: The discrete action (0 = Defect, 1 = Cooperate).
  • $y_i \in [0, 1]$: The continuous opinion (0 = Strongly prefer Defection, 1 = Strongly prefer Cooperation).
• Payoff Function: The total payoff $\pi_i$ for player $i$ is a convex combination of three components:
  $$\pi_i(z_i, z_{-i}) = \alpha_i \pi^a_i(x_i, x_{-i}) + \beta_i \pi^o_i(y_i, y_{-i}) - \frac{1}{2}\lambda_i (x_i - y_i)^2$$
  1. PGG Payoff ($\pi^a$): Based on a standard Public Goods Game with multiplier $r$ ($1 < r < n$).
  2. Opinion Payoff ($\pi^o$): Modeled using the Friedkin–Johnsen (FJ) dynamics, where players minimize the squared distance between their opinion and their neighbors' opinions, weighted by their attachment to a constant prejudice $u_i$.
  3. Self-Consistency Penalty: A term penalizing the squared difference between action and opinion. This creates the feedback loop: acting against one's belief reduces utility.
• Parameters: $\alpha_i, \beta_i, \lambda_i \in [0, 1]$ represent the relative weights of the PGG, opinion dynamics, and self-consistency, respectively, with $\alpha_i + \beta_i + \lambda_i = 1$.

B. Dynamics and Update Rules

• Update Mechanism: Players update asynchronously at discrete time steps. Active players adopt a myopic best-response strategy to maximize their current payoff given the current state of the system.
• Best-Response Derivation: The authors derive an explicit discriminant quantity, $\delta_i(y)$, which determines the optimal action.
  • If $\delta_i(y) > 0$, the best response is to Cooperate ($x_i=1$).
  • If $\delta_i(y) \le 0$, the best response is to Defect ($x_i=0$).
  • Crucially, the optimal opinion $y_i$ is a convex combination of neighbors' opinions, personal prejudice, and the newly chosen action.

3. Key Contributions

1. Novel Coevolutionary Model: The paper introduces a unified framework distinguishing between discrete actions and continuous opinions, linked by a self-consistency term. This contrasts with previous models (like CODA) that quantize opinions directly into actions or models that treat them as one-way dependencies.
2. Explicit Best-Response Characterization: The authors provide a closed-form expression for the best-response strategy (Proposition 1), showing that the decision to cooperate depends on a discriminant $\delta_i$ that aggregates PGG parameters, social influence, and self-consistency weights.
3. Equilibrium Analysis:
   • Action-Opinion Alignment: They prove that in any equilibrium where the population reaches an action consensus (all defect or all cooperate), the population must also reach an opinion consensus (Proposition 3).
   • Existence Conditions: They derive sufficient conditions for the existence of the All-Defection and All-Cooperation consensus equilibria (Theorem 1).
4. Global Stability Proof: The paper establishes conditions under which the system globally converges to the all-defection equilibrium, regardless of initial conditions (Theorem 2).

4. Key Results

A. Equilibrium Conditions

Under the assumption of zero prejudice ($\gamma_i = 0$):

• All-Defection Equilibrium $(0,0)$: Always exists. It is the unique equilibrium if the "cost" of maintaining self-consistency and opinion dynamics outweighs the benefit of the PGG multiplier. Specifically, if:
  $$\frac{\beta_i \lambda_i}{\beta_i + \lambda_i} \le 2\alpha_i \left(1 - \frac{r}{n}\right)$$
  for all players $i$.
• All-Cooperation Equilibrium $(1,1)$: Exists if the self-consistency and opinion dynamics are strong enough to overcome the inherent advantage of defection in the PGG. This occurs if:
  $$\frac{\beta_i \lambda_i}{\beta_i + \lambda_i} > 2\alpha_i \left(1 - \frac{r}{n}\right)$$
  for all players $i$.

B. Global Convergence

The authors prove that if the condition for the unique all-defection equilibrium holds (Eq. 15), the system globally converges to the all-defection state $(0,0)$.

• Proof Mechanism: They utilize a potential function $\Phi(y)$. When the condition holds, the best-response dynamics ensure that the action vector $x(t)$ becomes all zeros in finite time. Once $x(t)=0$, the opinion dynamics reduce to a standard FJ model with a self-consistency term that drives all opinions to 0. The potential function is shown to be monotonically increasing and bounded, guaranteeing convergence to the unique maximum at $y=0$.

5. Significance and Implications

• Theoretical Insight: The model demonstrates that cooperation in social dilemmas is not solely determined by the game payoff (PGG parameters) but is heavily influenced by the psychological drive for self-consistency and the structure of social influence.
• Policy/Intervention: The results suggest that to sustain cooperation, interventions must strengthen the parameters $\beta_i$ (social influence) and $\lambda_i$ (desire for consistency) relative to $\alpha_i$ (pure material payoff). If the desire for consistency is too weak relative to the temptation of free-riding, the system inevitably collapses into universal defection.
• Methodological Advancement: By separating action and opinion, the model captures the "tension" observed in real-world protests and movements where individuals may hold pro-cooperation beliefs but defect due to immediate costs, or vice versa. The explicit derivation of the best-response provides a tractable tool for analyzing complex coevolutionary systems.

In summary, this paper provides a rigorous mathematical foundation for understanding how the interplay between strategic decision-making and social opinion formation dictates the fate of cooperation in social dilemmas, offering clear conditions for when cooperation can survive and when it is doomed to fail.