Development of Hierarchical Bayesian Statistical Model For Prediction of Multidimensional Poverty Patterns: Application of Spatial-Temporal Analysis to Disadvantaged Village Data in Eastern Indonesia

Abstract

Multidimensional poverty in Eastern Indonesia is still a serious problem that is not only influenced by economic factors, but also by educational, health, infrastructure, and complex spatial conditions. Disadvantaged villages in the region face high development inequality, while the approaches to poverty measurement and prediction used so far are still conventional and less adaptive to spatial and temporal variations. This study aims to develop a Hierarchical Bayesian statistical model based on spatial-temporal analysis to predict multidimensional poverty patterns more accurately and contextually. The method used was a quantitative approach with spatial-temporal hierarchical Bayesian modeling, using multivariate panel data from 450 disadvantaged villages in East Nusa Tenggara, Maluku, and West Papua during the period 2015–2022. The model was analyzed using Markov Chain Monte Carlo (MCMC) and Integrated Nested Laplace Approximation (INLA) techniques for parameter estimation and risk prediction. The results show that the model is able to map poverty risk clusters spatially with high accuracy and capture significant temporal dynamics, especially during the pandemic. The largest contribution comes from indicators of sanitation and access to clean water. This model generates predictive risk and trend maps that can be used to support microdata-based development policies, as well as strengthen the accuracy of interventions in high-risk villages more effectively.