Development and aplication of Bayesian spatial analysis on poverty data in East Java, Indonesia

This paper is extracted from the research educational process of M.S level at Bogor Agricultural University, Indonesia. We first created a research group named Geoinformatics on Peverty and then invited some students to join and do research related to spatial statistics and poverty. Poverty is one of the crucial problems in Indonesia which is not easy to be solved. Following a survey held by the Statistics Indonesia in March 2011 showed that there were 30.02 million people or 12.49% of total Indonesian are considered poor. From the statistical point of view, poverty is an interesting topic because there are many problems of spatial data, i.e. spatial autocorrelation, error variance heterogenity, spatial interaction, and other statistical issues. The main objective of the geoinformatics group is to compile and develop spatial statistics applied on poverty allowing spatial effects. Those are GSM (General Spatial Model), SAR (Simultan Autoregressive), CAR (Conditional Autoregressive), SEM (Spatial Error Model), Bayesian SAR, and SAE (Small Area Estimation). The models are implemented on poverty data in East Java, Indonesia. The general results show that the poverty in East Java are related to low education, poor access to clean water, government pro poor programs (health insurance for the poor, subsidized rice for the poor, and poor letter), and improper housing. Statistically, Bayesian SAR slightly perform better than the other spatial regressions. In terms of small area estimation, spatial empirical best linear unbiased prediction (SEBLUP) and empirical Bayesian methods are more efficient than empirical BLUP (EBLUP).