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

TL;DR

This paper reviews various R packages for spatiotemporal count data modeling, compares their methodologies, and demonstrates their application through a COVID-19 hospitalizations case study in Spain.

Contribution

It provides a comprehensive comparison of R packages for spatiotemporal count data analysis using a unified dataset and highlights their performance in a real-world COVID-19 case study.

Findings

Bayesian and classical models effectively predict COVID-19 hospitalizations.

Different R packages offer comparable results despite methodological differences.

The review aids in selecting appropriate models for spatiotemporal count data analysis.

Abstract

Spatio-temporal models for count data are required in a wide range of scientific fields and they have become particularly crucial nowadays because of their ability to analyse COVID-19-related data. Models for count data are needed when the variable of interest take only non-negative integer values and these integers arise from counting occurrences. Several R-packages are currently available to deal with spatiotemporal areal count data. Each package focuses on different models and/or statistical methodologies. Unfortunately, the results generated by these models are rarely comparable due to differences in notation and methods. The main objective of this paper is to present a review describing the most important approaches that can be used to model and analyse count data when questions of scientific interest concern both their spatial and their temporal behaviour and we monitor their performance under the same data set. For this review, we focus on the three R-packages that can be used for this purpose and the different models assessed are representative of the two most widespread methodologies used to analyse spatiotemporal count data: the classical approach (based on Penalised Likelihood or Estimating Equations) and the Bayesian point of view. A case study is analysed as an illustration of these different methodologies. In this case study, these packages are used to model and predict daily hospitalisations from COVID-19 in 24 health regions within the Valencian Community (Spain), with data corresponding to the period from 28 June to 13 December 2020. Because of the current urgent need for monitoring and predicting data in the COVID-19 pandemic, this case study is, in itself, of particular importance and can be considered the secondary objective of this work. Satisfactory and promising results have been obtained in this second goal.