Track: Statistics and Empirical Research

Abstract

The availability of spatial and spatio-temporal data has widely increased and allow researcher to describe potential geographical pattern, including information about space and time (and its interraction) in many scientific fields. Bayesian method to deal with spatial and spatio-temporal data extensively approach with Markov Chain Monte Carlo (MCMC), however, when models are complex and designed with hierarchical fashion, MCMC algorithms may be extremely slow and even become computationally unfeasible. The Integrated Nested Laplace Approximation (INLA) algorithm is current development in R-INLA package in R, designed to deal with fundamental limitation of MCMC computation. This paper purpose to investigate how the socioeconomic information (i.e. population density, expectation years of schooling and construction overpriced index) effect the number of poor people in East Java province, Indonesia, using Bayes spatial model. Investigation result that expectation years of schooling has greatest effect in reducing number of poor people. Not only on its spatial pattern, we also investigate time dependency of poor people data from years 2012 to 2016 using classical, dynamic and space-time interaction of Bayes spatio-temporal models. In this paper, the computational aspect is efficiently solved with R-INLA, resulting dynamic Bayes Spatio-temporal is the best model based on the smallest Deviance Information Criteria.