A Few Variables Capture over 70% of State-Level Variance in U.S. COVID Deaths

A U.S. state-level analysis of factors associated with COVID-19 deaths reveals inequality (as defined by the Gini coefficient) to be far and away the strongest single-variable predictor, capturing 40% of vari ance in COVID deaths and 49% of variance in all-cause excess deaths since the start of the pandemic. A linear regression model with 5 independent variables accounts for over 70% of variation in COVID deaths,as does a 4-variable linear regression model for all-cause excess deaths. Similar models for COVID and all-cause excess deaths since October 1achieve similar results. Coefficients are highly significant (p <0.01) in almost all cases. A consistent finding across all 4 models is that a state’s relative humidity is strongly associated with fewer deaths after controlling for other factors. Lockdown stringency (as measured by the Oxford stringency index) is also strongly associated with fewer deaths for all models. Other significant factors for some models include population density, nursing home resident density, voting patterns shifting towards Donald Trump vs. prevous Republican candidates, and share of population under 18 years old. The models pass various robustness checks. The results are reproducible via an open-access data repository and Python notebook made available online.