Quantile regression for identifying latent structures in COVID-19 pandemic – Examples from Nepal

Abstract

During the COVID-19 pandemic, daily infections exhibited different pattern. It multiplied at an exponential rate, in the beginning. Due to physical restrictions imposed during the lockdown, this number stabilized to a certain value. During the relaxation of lockdowns, the pattern took another form. And after the distribution of three doses of vaccines, this number showed a different trend. In this paper, the path traced by the dependent variable Daily Infected, is explained using quantile values and quantile regression. The time period is from 26 February 2020 to 25 January 2023. Two quantile regression models are developed here. First, quantile regression of daily infection on daily discharged, phase and time of infection and phase time interaction is done for Nepal. Then, quantile regression of daily infection on Ratio 2, phase, time and phase and time interaction is constructed. Ratio 2, is the ratio of total new cases to total deaths, measuring the contribution of total deaths to total infected. The second model is tested for Nepal, India, Germany and the Netherlands. The behavior of the quantiles, before and after vaccination is compared. Here, Germany and the Netherlands are adjoining countries with good quality data. And Nepal and India are taken here as examples of neighboring countries with underreporting of daily infection and deaths. It is found that, quantiles and quantile regression are more robust with respect to underreporting. Thus, the latent behavior of daily incidence of COVID – 19 in different countries with different qualities of data is compared.