There has been a noticeable development of logistics facilities along the newly constructed circumferential expressway in the Tokyo Metropolitan Area, suggesting a strong association between the access provided by the roadway system and the locations of logistics facilities. As such, the development of analytical tools that can capture the complex relationships between the logistics facility location choice and locational attributes including spatial dependency is an important research endeavor. This paper aims to examine the effectiveness of alternative formulations of discrete choice models, including spatial autoregressive (SAR) probit and mixed logit models, for the analysis of logistics facility locations in the Tokyo Metropolitan Area.
The data from the 2013 Tokyo Metropolitan Freight Survey (TMFS) were used to compare the characteristics of several models. To apply the model, the Tokyo area was partitioned into approximately 18,000 1km by 1km grids. The independent variables include population density, employment, land price, land use, access to interchanges, and others. We estimated the location choice model with spatial correlations including Spatial Lag Model (SLM) and Spatial Error Model (SEM) with both probit and mixed-logit approaches. We also investigated the effects of different formulations of the spatial weight matrix on the models' performances. Our investigation found that inclusion of the spatial correlation terms improves the fit of the location choice model considerably. The mixed-logit model that incorporates both spatial lag and spatial error terms produced the best results. However, we believe the amount of time required for the estimation of the model may discourage practical application. On the other hand, the spatial autoregressive probit model can be within the reach of advanced practitioners.
View full abstract