Moran’s Eigenvector Spatial Filtering Regression Modeling of Interacting Disasters
Abstract: This study aims to quantify spatiotemporal relationships among hurricane impacts, social vulnerability, and community resilience with COVID-19 deaths per capita. Data used in this study included the Southeastern US with Hospital Service Areas (HSAs) and Hospital Referral Regions (HRRs) instead of county-level data. The Moran eigenvector spatial filtering (MESF) modeling method was employed to index and filter spatial autocorrelation (SA) from the residuals. SA was visualized to understand better the spatial structure meaningful covariates contain. The MESF model outperformed the linear model, not accounting for SA. Adjusted R-squared was 49.8% higher, and residual SE decreased 29.3% when comparing a 4-nearest-neighbor HRR to the linear model not considering SA. Analyzing cumulative COVID-19 deaths with this methodology is more effective at the HRR than at the HSA level. Additionally, vulnerability and resilience indicators were more substantial than hurricane impacts.
This image captures the 12 Moran eigenvector spatial filtering (MESF) models fit.
I have included the introduction, results and conclusion sections of the thesis below.