Modelling the effect of spatial determinants on freight (trip) attraction: A spatially autoregressive geographically weighted regression approach

This paper investigates the effect of spatial and locational characteristics of establishments on freight (trip) attraction (FA/FTA) models. The authors estimated econometric models of FA and FTA as a function of the establishment attributes as well as the spatial and locational determinant variables, using establishment-level data collected from Addis Ababa City, Ethiopia. The interconnected issues of spatial dependency and spatial heterogeneity, together with nonlinear specifications, were incorporated with the application of spatial techniques, including spatial error models (SEM), spatial autoregressive model (SAR), geographically weighted regression (GWR), multiscale-GWR (MGWR), and the combination GWR-SAR/MGWR-SAR. Regarding the explanatory variables, the empirical results revealed that firms in the manufacturing, wholesale and retail sectors located on the wider streets tend to receive more FA and FTA. The closeness to the primary road network and the city entry gate influences the FTA of manufacturing and construction firms. Moreover, retail establishments near the major market tend to receive more tonnage. The models also confirm that FA and FTA are the results of two different processes. Overall, the use of spatial regression techniques improves the accuracy of both FA and FTA models. MGWR-SAR exhibits superior performance by jointly addressing spatial dependency and heterogeneity. The MGWR-SAR model also uncovers the local variability of the variables representing the spatial and locational effects on freight attraction. The methodological analysis and empirical findings of the study could provide useful insights to support urban freight modelling, planning, and decision-making.

Automated summary

This paper investigates the effect of spatial and locational characteristics on freight attraction models. The authors estimated econometric models of attraction and the amount of freight attracted using establishment-level data from Addis Ababa City, Ethiopia. They incorporated spatial techniques, including SEM, SAR, GWR, MGWR, and the combination GWR-SAR/MGWR-SAR, to address spatial dependency, spatial heterogeneity, and nonlinear specifications. The empirical results show that firms in the manufacturing, wholesale, and retail sectors located on wider streets tend to receive more freight attraction. The models also confirm that FA and FTA are the results of two different processes. Overall, the use of spatial regression techniques improves the accuracy of both FA and FTA models.