Modeling the number of confirmed and suspected cases of covid-19 in east Java using bi-response negative binomial regression based on local linear estimator

Abstract

The number of confirmed and suspected cases of Covid-19 are type of count data and they correlate each other. A popular regression model of two response variables for count data is bi-response Poisson regression. However, assumptions violation of Poisson regression that frequently occurs is over-dispersion. Negative bmomial regression can overcome this over-dispersion case. The goal of this research is to model the number of confirmed and suspected Covid-19 cases affected by population density using bi-response negative binomial regression based on local linear estnnator. The proposed method gave the optnnal bandwidth of 609 based on maximum likelihood cross validation criterion, with deviance value of 1.537 which is less than 27.083 of the parametric regression approach. It means that the estimated model of the number of confirmed and suspected cases of Covid-19 affected by population density using bi-response negative binomial regression based on local lmear estnnator is better than the parametric model approach.