Spatial Analysis of the 2018 Logistics Performance Index Using Multivariate Kernel Function to Improve Geographically Weighted Regression Models

This paper analyzes the 2018 Logistics Performance Index (LPI) from the World Bank to determine the spatial effects of countries’ logistics performance. Although the standardized ordinary least square (OLS) models show good results, the spatial lags and Moran’s I of LPI suggest the OLS assumptions are violated. Consequently, an improved geographically weighted regression (IGWR) model using multivariate kernel functions (MKF) is implemented. Through the analysis of the Moran scatter plot, the authors identified the countries that have different logistics performance development trends in the four quadrants representing the relationship between the spatial lags and the LPI. Using trade activity (i.e., import/export) in the MKF, the authors compared different MKF types and bandwidths to ensure the model’s predictability and accuracy and found that the adaptive Gaussian MKF is suitable. Finally, the IGWR model indicates both positive and negative influencing factors on LPI overall score. Specifically, the improvements of LPI are more associated to economic variables in mid- and low-income countries around the world, and are more related to import of construction equipment in the Middle East. Also, business environment is more important in Latin America and the Pacific. European countries are more sensitive to customs efficiency, whereas Pacific-Asian countries are more sensitive to quality of infrastructure and have higher coefficients than African and American countries. This spatial heterogeneity is related to the specific factors that promote the development of their logistics performance.