A Review of Spatiotemporal Models for Count Data in R Packages. A Case Study of COVID-19 Data

Several R-packages to deal with spatiotemporal count areal data are currently available. Each package is based on different models and/or statistical methodologies. Unfortunately, results generated by these models are rarely comparable due to differences in notation and methods which ultimately hinder the robustness and effectiveness of such models. These models have become crucial nowadays, because their ability to analyze COVID-19 related data. This prompted us to revisit three spatiotemporal count models and monitor their performance under the same conditions (on a same data set and under a consensus notation). The three different models assessed are representative of the main two extended methodologies used to analyse spatiotemporal count data i.e., the classical approach (based on Penalised Likelihood meanwhile and based to Estimating Equations) and the Bayesian point of view. As a case study, we have used these packages to analyze COVID-19 data corresponding to 24 health regions within the Valencian Community, Spain. In particular, daily COVID-19 positive individuals are used to predict daily hospitalisations. Data corresponds to the period from 28 of June to 13 of December 2020.