Nonparametric Kernel Density Estimation of COVID-19 Incidence

Abstract

Several studies have been done to predict the course of the COVID-19 pandemic. However, it is critical to infer from observed data, the characteristics of COVID-19 incidence to improve both predictions, and interventional strategies by policy makers. We estimate the probability density of the daily COVID-19 incidence data for the country Zimbabwe in the first 150 days since the first case was recorded. We apply nonparametric kernel density estimation to obtain a suitable smoothed distribution fit for the observed COVID-19 incidence data from 20 March to 16 August 2020. The density of COVID-19 daily incidence in Zimbabwe is characterised by a sharp peak and a positive fatter tail with several jumps (shocks). The probability mass is concentrated on the tails, and the density is greatly influenced by the few jumps. The findings suggest a distribution of the Brownian motion type. Daily incidence, among other factors, is important in understanding the dynamics of COVID-19. A kernel density estimate of COVID-19 incidence has been obtained. In the study of COVID-19 dynamics, outliers (jumps or shocks) should not be excluded in models since they help explain the density of the observed data.