SABCoM: A Spatial Agent-Based COVID-19 Model

Abstract

This paper studies the effects of social learning on the transmission of COVID-19 in a network model. We calibrate our model to detailed data for Cape Town, South Africa and show that the inclusion of social learning improves the prediction of excess fatalities, reducing the best-fit squared difference from 20.06 to 11.28. The inclusion of social learning both flattens and shortens the curves for infections, hospitalizations, and excess fatalities. This result is qualitatively different from {\em flattening the curve} by reducing transmission probability through non-pharmaceutical interventions. While social learning reduces infections, this alone is not sufficient to curb the spread of the virus because learning is slower than the disease spreads. We use our model to study the efficacy of different vaccination strategies and find that a risk-based vaccination strategy--vaccinating vulnerable groups first--leads to a 50% reduction in fatalities and 5% increase in total infections compared to a random-order benchmark. By contrast, using a contact-based vaccination strategy reduces infections by 9% but results in 64% more fatalities relative to the benchmark.