Abstract
This study examines COVID-19 transmission across 3,142 U.S. counties using a truncated dataset from March to September 2020. County-level factors include demographics, socioeconomic status, environmental conditions, and mobility patterns. Ordinary Least Squares regression establishes a baseline for analyzing COVID-19 confirm case counts for each county. We then use Moran's I to evaluate spatial clustering, prompting Spatial Autoregressive and Spatial Error Models when autocorrelation is significant. Notably, spatial models outperform the Ordinary Least Squares approach-R (2) rises from 0.4849 with Ordinary Least Squares to 0.6846 under Spatial Error Model, while RMSE decreases from 2.0891 to 1.642-demonstrating improved fit and more accurate spatial transmission dynamics. A multilevel framework further explores state-level policy variations. Finally, Geographically Weighted Regression captures spatial non-stationarity by mapping local coefficient differences; we visualized temperature, precipitation, and other key variables-revealing precipitation peaks near 110° W in the Southeast and Northeast and strong sensitivity to temperature. This integrated sequence of methods provides a comprehensive lens for studying epidemiological phenomena. While certain findings align with established research, other variables reveal unexpected patterns. The proposed framework offers a robust template for future investigations where spatial dependence and policy heterogeneity warrant close examination.