Fig 1.
Schematic of the two-layer networked simulation model.
Each individual agent (A, B,..., etc.) has attribute values that define its age, resource level, opinions, behavior, and status throughout a simulation run. Their opinions are updated through interactions in the opinion network layer via the relative agreement model. There are two broadcasting channels that transmit pro-intervention and anti-intervention messages, respectively. Agents make decisions regarding their NPI compliance and vaccination willingness based on the given situation and their opinions. This decision-making process is represented by the DM (Decision-Making) boxes in the figure. The infectious disease spreads throughout the contact network at the same time. Individual decisions regarding self-protection affect the spread of the disease.
Fig 2.
Simulation procedure flow for a single instance.
Fig 3.
Decision trees for individual agents.
Individual agents make decisions through three decision trees. An agent’s compliance with NPIs, such as wearing a mask or maintaining social distancing, is determined through decision trees (A1) and (A2), depending on its vaccination status. If an agent is eligible for vaccination, it decides on its willingness to get vaccinated using the decision tree (B). Only two probability values (highlighted) are based on an agent’s perception that changes over time: and
, while other probability values are fixed at a certain value for simplification. This is to minimize the complexity of the model and have only one dynamic opinion value closely related to each of the two self-protection behaviors: NPI compliance and vaccination willingness.
Table 1.
State transitions and expected binomial transition sizes for internal disease dynamics of standalone midge habitat agents.
Table 2.
Dataset descriptions and their sources.
Fig 4.
Two datasets incorporated in the model for generating the population characteristics.
Plot (a) visualizes the U.S. Census Bureau dataset about the joint distribution of income and age, and plot (b) presents a heatmap of contact frequency matrix between age groups.
Fig 5.
Simulation results of the calibrated baseline model: average NPI compliance level trend over time (top), trend (%) of the population vaccinated (middle), and daily new case trend (bottom).
The simulation results based on the calibrated parameter tuples are presented (210 instances in total for visualization). The four vertical dotted lines mark significant events of the baseline scenario.
Table 3.
Alternative scenario table.
Fig 6.
Simulation results of the calibrated baseline model and three alternative scenarios: (1) no NPI mandate, (2) no vaccine, and (3) no NPI mandate and no vaccine at the same time.
Since the random seeds are strictly managed over simulation runs, we can observe the clear bifurcating branches between scenarios around day 100 and day 250 in the first and third rows.
Fig 7.
Simulation results (1) with different scenarios across each decile group.
From the plots, we can observe that both the NPI mandate and vaccine introduction significantly reduced disease spread by comparing the daily new cases across decile groups in different scenarios (1st row, lower is better). The plots in the 2nd, 3rd, and 4th rows (higher is better) show how agents make decisions regarding self-protection against infectious diseases based on their characteristics or conditions.
Fig 8.
Simulation results for scenarios in which a specific age demographic shows an inclination toward anti-intervention messages.
The solid curve shows the simulation results of the baseline scenario, in which none of the agents are selected to have an altered response setting to media channels’ messages. Three inserted bar charts show the percentage changes in measures caused by different setups compared to the baseline model. The later part (beyond Day 200) of the NPI compliance plot and the early part (before Day 250) of the vaccination plot are omitted, as they remain the same level in those periods. The plots show that the inclination toward the anti-intervention messages among the younger generation (Q1) causes the greatest change in disease spread.
Fig 9.
Simulation results for scenarios in which the inclination toward anti-intervention messages is concentrated within specific quartiles of the population by resource availability.
The inserted bar charts and omitted sections follow the same structure as Fig 8. The plots show that the inclination toward the anti-intervention messages among the middle-resource groups (Q2 and Q3) causes significant changes in NPI compliance behavior and leads to a greater impact on disease spread.
Fig 10.
Simulation results comparing the outcomes of the baseline scenario (blue curve) with those of alternative vaccination policies: open eligibility (top), high-exposure priority (middle), and low-resource priority (bottom).
The results show that, under our simulation settings, alternative scenarios outperform age-based vaccine prioritization by enabling higher vaccine administration during the early stage. This leads to a smaller peak in new cases. Although the alternative policies delay vaccination for populations at higher risk of severe outcomes, they more effectively reduce overall disease spread by mitigating transmission among high-exposure populations and those with limited resources, who may have lower motivation for self-protection. Notably, the impact is limited to the faster distribution of vaccines to the willing population and does not change the total number of individuals willing to be vaccinated. This leads to convergence in vaccination trends around t = 400.