Large Scale Agent-based Model Simulating Information Diffusion of Covid-19 Vaccines in New York State

 In the past, we have built an agent-based model to simulate individuals' Covid-19 vaccines uptakes in a rural county - Chautauqua (New York). Recently, at the 7th ACM SIGSPATIAL International Workshop on Geospatial Simulation (GeoSim 2024), we (Na (Richard) Jiang, Andrew Crooks, and Lucie Laurian) have extended that model to the entire New York State to explore the diffusion of Covid-19 opinions in a hybrid context (e.g., physical, relational and cyber spaces) and its impacts on people's attitudes which are reflected in their vaccination behaviors. 


This project has significantly extended the previous work in a number of ways. 

  • First, we move the model from a single rural county to the entire NYS that has 62 counties which differ substantially in socioeconomic status and move from a small population of 120,000 to over 20 million agents. By doing so, it allows us to compare vaccination uptakes in different areas (e.g., urban versus rural communities, second home destinations versus college towns). 
  • This paper also uses different parameters to initialize hybrid spaces for urban and rural populations to understand how individuals’ preferences on hybrid spaces affect information diffusion and vaccination rates at a macro level. 
  • Last, we update the decision-making rules for minors (ages under 18) that allows us to better simulate young population groups.


This project is the first one that has successfully scaled up an agent-based model from a local rural area (i.e., Chautauqua county) with a small population of 120,000 to the entire region (i.e., NYS) with over 20 million agents. Our modeling outputs can match the ground truth vaccination rates with small errors (MAE=6.93 and RMSE=8.87) for the entire NYS.


By giving urban and rural agents different weighting schemes of hybrid spaces, the final weighting scheme "urban, 1:3:3 - rural, 3:3:1" outperforms others in capturing the real-world vaccination rates. The realistic meaning of this weighting scheme is that in our model, urban agents give more trust to the information they have received from relational and cyber spaces, while rural agents place more emphasis on physical and relational spaces. By using this weighting scheme "urban, 1:3:3 - rural, 3:3:1", the model can capture the general trend of vaccination progress for 50 out of 62 counties (81%) in New York State with small errors (MAE less than 10). Additionally, for a half of NYS counties (n=31), this weighting scheme can produce almost a perfect match (MAE less than 6) with their real-world vaccination progress. For counties such as Erie, Ontario and Niagara, the differences between simulated and ground truth vaccination rate is around 2% on a daily basis.


This large-scale agent-based model uses a detailed geographically-explicit synthetic population dataset to create individual agents and a python-based ABM framework (i.e., Mesa) to code agents' behaviors. The simulation lasts for 500 days (from January 2021 until May 2022). For interested readers, we are sharing our modeling scripts, input data and results at https://osf.io/3khyq/ to promote the findable, accessible, interoperable and reusable research community. 


Abstract: 

During the COVID-19 pandemic, social media become an important hub for public discussions on vaccination. However, it is unclear how the rise of cyber space (i.e., social media) combined with traditional relational spaces (i.e., social circles), and physical space (i.e., spatial proximity) together affect the diffusion of vaccination opinions and produce different impacts on urban and rural population's vaccination uptake. This research builds an agent-based model utilizing the Mesa framework to simulate individuals' opinion dynamics towards COVID-19 vaccines, their vaccination uptake and the emergent vaccination rates at a macro level for New York State (NYS). By using a spatially explicit synthetic population, our model can accurately simulate the vaccination rates for NYS (mean absolute error=6.93) and for the majority of counties within it (81%). This research contributes to the modeling literature by simulating individuals vaccination behaviors which are important for disease spread and transmission studies. Our study extends geo-simulations into hybrid-space settings (i.e., physical, relational, and cyber spaces).


Full References:

Yin, F., Jiang, N., Crooks, A., & Laurian, L. (2024, October). Agent-based modeling of COVID-19 vaccine uptake in New York State: Information diffusion in hybrid spaces. In Proceedings of the 7th ACM SIGSPATIAL International Workshop on GeoSpatial Simulation (pp. 11-20). (pdf)



Schematic representation of hybrid spaces. Physical space includes family and group quarter network. Relational space represents people’s social circles in work, school and daycare. Cyber space is a social media network. This figure only display 2% of total population in NYS (around 200,000 agents) for visualization process


Modeling process and structure: from data to agent-behaviors.

Mapping the differences (i.e., MAE) in vaccination rate between simulated and ground truth data