Multi-Scale Geographically Weighted Elasticity Regression Model to Explore the Elastic Effects of the Built Environment on Ride-Hailing Ridership

Understanding the relationship between the built environment and the ride-hailing ridership is crucial to the prediction of the demand for ride-hailing and the formulation of the strategy for upgrading the built environment. However, the existing studies on ride-hailing ignore the scale effect and zone effect of the modifiable area unit problem (MAUP), and show a lack of consideration for the elastic relationship with spatial heterogeneity between built environment variables and ride-hailing ridership. Taking Chengdu as an example, this paper selects 12 independent variables based on the “5Ds” (density, diversity, design, destination accessibility and distance to transit) of the built environment, the dependent variables are the density of ride-hailing pick-ups in the morning and evening peak hours, and 11 spatial units are proposed according to different scales and zoning methods for the aggregation of built environment variables and ride-hailing pick-ups. With the goal of global optimal goodness-of-fit, we determined the optimal spatial unit by using the log-linear Ordinary Least-Squares (OLS) model. A multi-scale geographically weighted elastic regression (MGWER) model is formulated to explore the relative effect of the built environment on the ride-hailing ridership and spatial heterogeneity. The average value of positive elastic local regression coefficient of different variables is used to measure the relative positive impact of built environment factors, and the absolute value of the average value of negative elastic local regression coefficient is used to measure the relative negative impact of built environment factors. The results show that: (1) The MGWER model under the community unit division has the best global goodness-of-fit. (2) Different built environment variables have different elastic impacts on the demand for ride-hailing. For the morning peak hours and evening peak hours, the top three built environment factors with positive impacts are ranked as follows: commercial POI density > average house price > population density, and distance to CBD has the highest negative impacts on pick-up ridership. (3) The different local elasticity coefficients of the built environment factors at different stations are discussed, which indicate the spatial heterogeneity of the ride-hailing ridership. The optimal community zoning method can provide a basis for the zoning and scheduling management of ride-hailing. The results of the built environment variables with greater impact are conducive to the formulation of targeted urban renewal strategies in the process of adjusting the ridership of ride-hailing.