Some Estimation Methods for Spatio-Temporal Data in Spatial Statistics: (A Comparative Study)

Abstract

Traffic accident fatality is a pressing public health issue that demonstrates significant spatio-temporal heterogeneity and would benefit from a statistical approach that can appropriately account two concurrently. In the field of spatial statistics, spatio-temporal regression models and kriging methods are used for estimation and prediction, but their relative performance varies largely according to data structure and the availability of explanatory covariates. In addition, there is quite limited empirical evidence that compares these two approaches to their application over real spatio-temporal traffic fatality data, specifically for developing countries, and in contexts where multiple covariates are available. A major aim of this study is to compare the efficiency of estimation using spatio-temporal linear regression and spatio-temporal ordinary kriging to model the total number of traffic accident fatalities in 15 provinces in Iraq during the period 2020–2023. The results demonstrate the superiority of the spatio-temporal regression model over kriging in prediction error and explanatory power by effectively integrating road, vehicle, driver, and pedestrian variables. The work we presented provides a focused, data-rich comparison showing that for fatality data with relevant covariates, regression based spatio-temporal modeling outperforms dependence on spatial temporal dependence. The observed patterns indicate that traffic safety analysis and policy should focus on models that incorporate explanatory variables, while future methodological decisions should be based on the underlying concept, data availability, and the strength of covariates of interest.