The impact of digital media-driven affective polarisation on epidemic dynamics

TL;DR

This work addresses how digital-media–driven affective polarisation shapes epidemic dynamics by coupling information diffusion with disease spread on a multiplex network. It develops a model where information and epidemic layers interact through awareness and infection risk, with polarisation measured by $\psi$ and digital-media influence by $\gamma$, and where awareness dampens transmission via $\beta_A = \epsilon\beta_U$. Simulations show that $\psi$ increases with $\gamma$ beyond a threshold, and that the effect of polarisation on $\rho^I$ reverses with infection strength: a negative association at low $\beta$ and a positive one at high $\beta$, mediated by the aware population. The results highlight potential feedback loops where digital-media–driven polarisation can either suppress or exacerbate outbreaks depending on transmission conditions, informing public-health strategies to mitigate infodemic and polarisation risks.

Abstract

While prior studies have examined the influence of information diffusion on epidemic dynamics, the role of affective polarisation--driven by digital media usage--remains less understood. This study introduces a mathematical framework to quantify the interplay between affective polarisation and epidemic spread, revealing contrasting effects depending on transmission rates. The model demonstrates that greater digital media influence leads to increased polarisation. Notably, the results reveal opposing trends: a negative correlation between polarisation and the infected population is observed when transmission rates are low, whereas a positive correlation emerges in high-transmission scenarios. These findings provide a quantitative foundation for assessing how digital media-driven polarisation may exacerbate health crises, informing future public health strategies.

Figures (5)

• Figure 1: A schematic representation of the proposed model at a given time step. The central part of the figure represents the network that simulates real-world social interactions, where each node belongs to one of two partisan groups, denoted by red and blue colours. The left panel illustrates the epidemic layer, where each node is in one of two states: $S$ (Susceptible) or $I$ (Infected), represented by red and green nodes, respectively. The right panel depicts the information layer, where each node holds opinions on five topics. The yellow and light green colours represent awareness states regarding the epidemic: $U$ (Unaware) and $A$ (Aware).
• Figure 2: The relationship between the probability $\gamma$ of interacting with nodes other than adjacent nodes during information updates and the level of polarisation $\psi$. $\gamma$ is varied from 0 to 1 in increments of 0.02, resulting in 51 patterns. For each value of $\gamma$, 100 simulations are performed, and the $\psi$ values after 50,000 steps are plotted in light blue, with the average value plotted in dark blue. As $\gamma$ increases, $\psi$ also increases. All other parameter values follow Table \ref{['tab:parameters']}.
• Figure 3: Comparison of scatter plots of the level of polarisation $\psi$ and the proportion of infected individuals $\rho^I$. The colour differences represent the digital media influence $\gamma$. As in Figure \ref{['fig:scatter']}, for each value of $\gamma$, the $\psi$ values after 50000 steps are calculated, and the average over 100 simulations is plotted. In Figure \ref{['fig:low_rho_I']}, a downward trend is observed, while in Figure \ref{['fig:high_rho_I']}, an upward trend is obtained.
• Figure 4: Heatmap of $\rho^I$ when $\beta$ and $\gamma$ are varied. $\beta$ was varied in 50 steps from 0.001 to 0.05 in increments of 0.001, and $\gamma$ was varied in 11 steps from 0 to 1 in increments of 0.1, resulting in 550 combinations. For each combination, 10 simulations were run, and the average $\rho^I$ after 5000 steps was plotted. $\mu = 0.1$ and other parameters follow those in Table \ref{['tab:parameters']}.
• Figure 5: Scatter plots of the level of polarisation $\psi$ versus the proportion of aware individuals $\rho^A$, along with scatter plots of $\rho^I$ plotted against the residual $\psi_{\text{res}}$. The top row corresponds to the case of $\beta = 0.005$, $\mu = 0.1$, and the bottom row to $\beta = 0.05$, $\mu = 0.01$. Each data point represents the average over 100 simulations for each $\gamma$, calculated after 50000 steps. The other parameter values follow Table \ref{['tab:parameters']}.