Social norm dynamics in a behavioral epidemic model

TL;DR

This work addresses how social norms shape preventive vaccination behavior in epidemic settings by introducing a two-layer multiplex agent-based model that couples a $SIR$ epidemic on a physical network with norm dynamics on a social layer. Vaccination intentions $x_i(t)$ are driven by an Experience-Weighted Attraction (EWA) learning framework and are modulated by distinct normative inputs: descriptive expectations $\widetilde{x}_i$, injunctive expectations $\widetilde{y}_i$, and personal norms $y_i$, updated through cognitive dissonance, social projection, and consistency mechanisms; decisions depend on both empirical payoffs and normative cues via $\phi_i(t)$ and a safety factor $S_i(t)$. Key results show injunctive norms exert stronger and more persistent effects on vaccination uptake and infection levels than descriptive norms, and that external interventions targeting injunctive expectations reduce infections more reliably, though interventions on descriptive expectations can backfire under certain conditions. The findings suggest norm-based models, once empirically calibrated, offer a richer description of human behavior and can inform public-health strategies for pandemics and other collective-action problems beyond disease control.

Abstract

Understanding the social determinants of preventive behavior is vital for epidemic modelling and effective policy making. Traditional models emphasize imitation or rational trade-offs, but recent evidence highlights the role of social norms. We develop a behavioral epidemic model of seasonal disease on multilayer networks, where vaccination decisions combine learning from experience with coevolving social norms. The framework distinguishes descriptive norms (what others do) from injunctive norms (what others think ought to be done), while incorporating cognitive dissonance, social projection and logical consistency. Simulations show that norm dynamics yield markedly different vaccination uptake and infection levels compared to considering solely payoff-driven learning. Injunctive norms exert stronger and more persistent effects than descriptive norms. Interventions targeting injunctive expectations improve outcomes, while those on descriptive norms may be weaker or even counterproductive. Norm-based models, once empirically validated, can better capture human behavior and guide strategies for collective action problems even beyond pandemics.

Figures (6)

• Figure 1: Dynamics' algorithm, Two-layer network and schematic of dynamics. a.: Algorithm illustrating the system dynamics. b. Top: Vaccination game on a two-layer network. The physical layer hosts a seasonal SIR epidemic, while the social layer governs vaccination decisions based on information from both layers. The physical layer follows a small-world topology and the social layer a Klimek–Thurner network, with partial overlap reflecting that many physical contacts also serve as information links. Bottom: Schematic of the algorithm. Each season, the SIR dynamics run on the physical layer to estimate infection probabilities. Agents then decide whether to vaccinate, considering both payoff-based learning and normative factors, update their norms accordingly, and begin the next season.
• Figure 2: Decision-making process and social norm dynamics. a.: Schematic of the decision-making process. Vaccination intention combines learning (1), personal norms (3), and social norms (2) and (4), each weighted by its relative influence. Agents rely more on empirical factors when they feel unsafe (low $S_i(t)$) or when their environment is unstable and heterogeneous, and more on normative factors under stable conditions. b.: Schematic of social-norm dynamics. Each agent updates norms based on external cues (with probability $\gamma_i$), personal beliefs, and peers’ actions. In unstable and highly divided environments, agents rely more on internal factors than on social conformity.
• Figure 3: Role of social norms in epidemic outcomes.a. Infected fraction versus infectivity rate $\beta$. Infection increases with $\beta$ and saturates at high values, while longer memory reduces outbreak size. For fixed $m=4$, adding social norm dynamics further lowers infections. b. Vaccination coverage versus $\beta$. Coverage stabilizes near 0.5 for large $\beta$ due to panic effects, but with norms, even small $\beta$ sustain $\sim20\%$ coverage as heterogeneous initial norms persist. c. Mean values of $y$, $\widetilde{y}$, $\widetilde{x}$, and $x$. All converge in the long run, indicating alignment between norms and vaccination intention.
• Figure 4: Role of external factors on social norms.a. Infected fraction versus infectivity rate $\beta$ when an external factor influences $y_i$, $\widetilde{y}_i$, or $\widetilde{x}_i$. The factor reduces perceived disease severity, driving norms toward $G_i = 0.6$, below the equilibrium value $0.7$ in Fig. \ref{['fig:Fig3']}. It increases infections when applied to $y_i$ or $\widetilde{y}_i$, but has negligible effect on $\widetilde{x}_i$. b. Vaccination coverage versus $\beta$ for the same cases. Parameters: $\gamma_i = 1.0$, $G_i = 0.6$.
• Figure 5: Role of the external factor’s strength $\gamma$ and target value $G$.a. Outbreak size versus coupling strength $\gamma$ when the external factor acts on $y_i$, $\widetilde{y}_i$, or $\widetilde{x}_i$, with targets $G = 0.6$ (lower severity perception) and $G = 0.8$ (higher). Driving $y_i$ or $\widetilde{y}_i$ toward $G = 0.6$ increases infections, whereas for strong coupling most other cases reduce outbreak size. Driving $\widetilde{x}_i$ toward $G = 0.6$ has little effect. b. Outbreak size versus $G$ for $\gamma = 0.5$ and $\gamma = 1$. The red dashed line marks the baseline without intervention. For large $G$ (nudging toward vaccination), infection approaches zero; for small $G$, outbreaks grow. Intermediate $G$ values are most effective via $\widetilde{y}$, while for $G \approx 0.6$–$0.8$, interventions on $\widetilde{x}_i$ buffer against external influence.