Empirical Best Linear Unbiased Prediction Method with K-Medoids Cluster for Estimate Per Capita Expenditure of Sub-District Level

One of the data needed is per capita expenditure of sub-district level. The National Socio-Economic Survey (Susenas) obtain per capita expenditure data. However, due to limited samples, Susenas cannot get statistical information down to the sub-district or village level from this survey. Therefore, a model that includes additional data (census data) to estimate sub-district levels uses Small Area Estimation (SAE) modelling. In this study, the Empirical Best Linear Unbiased Prediction (EBLUP) method estimates the per capita expenditure of sub-district levels in Jambi Province in 2018. The EBLUP modelling also applies cluster information to estimate per capita expenditure in the sub-districts that are not surveyed (non-sampled area). The K-Medoids Cluster method is used to get sub-district clusters based on their characteristics. The mean of area random effects per cluster adds to the prediction model for the non-sampled area. This study aims to compare and evaluate the direct estimation method and EBLUP method to estimate sub-district level per capita expenditure based on Relative Root Mean Square Error (RRMSE). The result shows that EBLUP estimation produces more accurate estimates than direct estimation. This research also estimates the per capita expenditure of non-sampled areas using the EBLUP method by applying K-Medoids Cluster information.