Proposed robust estimators for the Poisson panel regression model: application to COVID-19 deaths in Europe

Abstract

In regression panel data analysis, the maximum likelihood (ML) estimates of the Poisson model with fixed effects (FE) are affected by outliers. Thus, the ML estimation method will not be appropriate to solve the problem of outliers in the panel data. Therefore, we need robust estimation methods where the estimates of these methods are not much affected when the dataset contains outliers. This study aims to propose three robust estimation (M, S, and MM) methods that deal with panel datasets that contain outliers to enhance the accuracy of the results and provide good, stable, and more accurate predictions. For this purpose, these proposed robust methods were applied to coronavirus data for twelve high-income countries in Europe during the period from June 23, 2021, to January 21, 2022, to examine the performance and efficiency of these estimators in the presence of outliers. The results of COVID-19 indicated that the estimates of the classical ML estimation method are highly sensitive to outliers unlike proposed robust estimation methods, especially the MM robust estimation method, where the MM estimates are better than the other estimates.