Bayesian hierarchical modelling for spatio-temporal interactions with road traffic accident data

The Bayesian hierarchical approach is widely recognised as a powerful method for formulating spatio-temporal models, offering stable estimates by incorporating information from neighbouring units, even in the presence of low-count data. This study proposes the development of Bayesian hierarchical spatio-temporal interaction models, with a case study on Oman’s road traffic accident (RTA) data from 2013 to 2017. The integrated nested Laplace approximation (INLA) technique is employed to fit the models and estimate both fixed and random effects parameters. Model selection criteria, including the deviance information criterion (DIC), Watanabe-Akaike information criterion (WAIC), and conditional predictive ordinate (CPO), are discussed. Among the various models, the spatio-temporal interaction Type II model, which incorporates interactions between unstructured spatial effects and a structured temporal effect modelled as a first-order random walk, is identified as the most appropriate for capturing endogenous factors influencing Oman’s road traffic accident data, alongside several other covariates.