Application of Small Area Estimation for Estimation of Sub-District Level Poverty in Bengkulu Province: Comparison of Empirical Best Linear Unbiased Prediction (EBLUP) and Hierarchical Bayesian (HB) Methods

Abstract

Poverty is an important problem facing the world. Various ways are done to eradicate poverty. In planning poverty alleviation, policy makers need detailed information down to the smallest area level that can be produced. Currently, the demand for estimation at the small area level is increasing, while the success of estimation using the indirect method in reducing the Relative Standard Error (RSE) is very dependent on data conditions and the selection of the right method. This study aims to compare the results of estimating the percentage of poor people using direct estimates with indirect estimates using the Small Area Estimation (SAE) technique such as Empirical Best Linear Unbiased Predictor (EBLUP) and Hierarchical Bayesian (HB) method using a case study of poverty data at the sub-district level of Bengkulu Province. The data used are from the Social and Economic Survey (Susenas) in March 2022 and the 2021 Village Potential Data Collection (Podes). There is one sub-district that was not sampled in the March 2022 Susenas. The average RSE value of the direct estimator is 47.014 and the average RSE of the EBLUP estimator is 39.40 and the HB estimator is 15.318. In addition, the SAE EBLUP and HB methods can reduce the mean and median values of RSE estimation results when compared with direct estimates. The RSE of the direct estimator is greater than the RSE of the indirect estimator.