Multi-scale Modeling of Spatio-Temporal COVID-19 Mortality: County and State Level in the US
TL;DR
The results indicate that joint models which have linked scale components are to be preferred in terms of goodness of fit compared to separate models, and suggest that basing policy decisions on models which disregard linkage across different scale levels could be biased and incorrectly estimated.
Abstract
Infectious disease spread occurs both temporally and spatially at a variety of scales. These scales could be temporal and/or spatial. In the case of the COVID-19 pandemic, much effort has been made to consider temporal modeling and prediction of mortality and case incidence in the time domain. Meanwhile the spatial structure of the epidemic spread has had less attention, and in particular the explicit linkage between multiple spatial scales, has received comparatively less attention. We have addressed this evaluation based on CDC data archived from the COVID-19 dashboard for the weeks of January 2020 to June 2023, available for the counties and state level of South Carolina. Our results indicate that joint models which have linked scale components are to be preferred in terms of goodness of fit compared to separate models. Multi-scale components also benefit for separate models, where time series models applied solely to the state level do not fit as well as models, which exploit linkages with case counts and cumulative counts of disease. Spatial correlation effects are supported in multi-scale models but are not favored when separate models are considered. These findings suggest that basing policy decisions on models which disregard linkage across different scale levels could be biased and incorrectly estimated.