Linear and nonlinear effects explaining the risk of Covid-19 infection: an empirical analysis on real data from the USA

Abstract

Using data from 3142 counties in the United States and a fully nonparametric variable selection approach for high-dimensional models, we identify predictor variables (among social, behavioral, economic, political, regulatory, demographic, and health characteristics) and discriminate against them between linear and nonlinear, depending on their effect on the risk of Severe Acute Respiratory Syndrome Coronavirus 2 infection. The data refer to the period from January to December 2020. We use a nonparametric and non-additive screening selection approach, the Derivative Empirical Likelihood Sure Independent Screening (DELSIS), in combination with a subsample technique. The results show that the relevant variables are different between counties with “large” and “small” populations. Furthermore, predictors such as mask wearing, age levels, ethnicity and poor health conditions are the main relevant variables for predicting the risk of infection, but with some differences over time.