A1231
Title: Modelling count spatio-temporal data
Authors: Dimitris Karlis - RC Athens University of Economics and Business (Greece) [presenting]
Abstract: Spatio-temporal data that are counts are common in epidemiology, where disease occurrences are observed across regions over time, criminology, where crime counts are analyzed within specific jurisdictions across successive periods, and many other disciplines. A class of models designed to account for the discrete nature of these observations explicitly is introduced. Such models offer better forecasting accuracy and provide better insights compared to existing continuous approximations. A general spatio-temporal framework based on the idea of thinning operations that preserves the discrete nature of the data is defined, and its fundamental statistical properties are derived. For parameter estimation, a conditional maximum likelihood approach facilitated by an Expectation-Maximization (EM) algorithm is employed. Forecasting procedures are developed and the model performance is validated through extensive Monte Carlo simulation experiments. The practical utility of the proposed framework is demonstrated using a real-world application: the spatio-temporal progression of the COVID-19 epidemic in the Netherlands. Possible extensions are also discussed.
