Beyond univariate measurement of spatial autocorrelation

Abstract

Most studies that incorporate spatial effects use a very limited number of spatial variables in the growth model, e.g. growth spillovers or infrastructure impacts of neighbouring regions. This article innovates on previous work in spatial econometrics by differentiating among spatial contributions to economic development; e.g. infrastructure, capital, human capital, land and labour. We explore whether including more spatial effects can improve the viability of a growth model, and also test the hypothesis that the differential spatial effects of various important predictor variables are discernible for Indonesia. We develop two econometric estimations based on the Durbin representation of the Spatial Error Model. We take advantage of a panel data set spanning Indonesia's post-decentralization years, 2003-2008. The first model uses a modified fixed effects formulation, and the second uses a maximum likelihood estimator. The two sets of models are reported together to serve as a check to the robustness of the results. Multiple estimation methods were attempted, including (to control for potential endogeneity) two-stage least squares (2SLS) and generalized method of moments (GMM). The findings suggest that various types of spillover effects affect a place by different processes, and accounting for this variety of processes in growth models improves the efficacy of those models. Our findings suggest that, for Indonesian districts, the influence of neighbours extends beyond GRDP per capita levels and growth, and also includes demographics, human capital and infrastructure components. We also demonstrate empirically that accounting for spatial effects in analysis of GRDP per capita can improve growth-model estimations. Show more