A multivariate generalized logistic approach with spatially varying nonlinear components for modeling epidemic data

This work considers the joint analysis of time series for epidemiological count data of neighboring regions. The joint analysis involves parameter estimation and prediction of future outcomes. The literature concentrated on imposing similarities on components of the linear predictor for the mean. However, some hierarchical model specifications for the mean contain non-linear components with similar behavior over neighboring regions. This paper proposes the use of spatial specification for these components. Parametric forms based on a data-driven approach are assumed for the waves of epidemic counts, and multiple waves are considered. The resulting model is tested in simulation studies and applied to real data. Model evaluation is based on the fitting and prediction capabilities. An illustration is provided by the analysis of counts of COVID19 cases, and it compares favorably against alternative models. Finally, the paper concludes with a discussion of the proposed methodology.