Spatial regression model on gross regional domestic products in Indonesia with Eigenvector spatial filtering approach

Abstract

Gross Regional Domestic Product (GRDP) describes an important indicator to determine the economic development and structure of a region. Economic data which generally describes the value of an area contains spatial effects and can be overcome by using spatial modeling. Maximum likelihood (ML) is a method commonly used in estimating spatial dependency models. In Big Data, the ML method is not efficient, so eigenvector spatial filtering (ESF) is used. To overcome spatial dependence, ESF adding a linear combination of eigenvectors of spatial weighting matrix to regression model specification. The objective of this study is to determine the spatial regression model of GRDP data of regencies/cities in Indonesia in 2019 with the ESF approach. Estimation of RE-ESF model uses restricted maximum likelihood (REML). The results showed that the application of ESF approaches increased the accuracy of the model based on AIC and R-square values compared to the spatial autoregressive model (SAR) and spatial error model (SEM). Factors that influence GRDP Indonesia in 2019 are the original local government revenue, the number of workers, and the human development index. The RE-ESF improves the accuracy of regression coefficient estimation with more efficient computation time.