Exploring nonlinear effects of the built environment on ridesplitting: Evidence from Chengdu

Highlights

• Built environment factors have strong non-linear effects on the ridesplitting ratio.

• Gradient Boosting Decision Trees model is used to analyze the effects.

• Distance to the city center is the most important among built environment factors.

• Land use diversity and road density also have strong impacts on ridesplitting ratios.

• Non-linear thresholds of built environment factors are identified to guide the planning.

Abstract

Ridesplitting, a form of ridesourcing services that matches riders with similar routes to the same driver, is a high occupancy travel mode that can bring considerable benefits. However, the current ratio of ridesplitting in the ridesourcing services is relatively low and its influencing factors remain unrevealed. Therefore, this paper uses a machine learning method, gradient boosting decision tree (GBDT) model, to explore the nonlinear effects of built environment on the ridesplitting ratio of origin–destination pairs (census tract to census tract). The GBDT model also provides the relative importance ranking of all the built environment factors. The results indicate that distance to city center, land use diversity and road density are the key influencing factors of ridesplitting ratio. In addition, the non-linear thresholds of built environment factors are identified based on partial dependence plots, which could provide policy implications for the government and transportation network companies to promote ridesplitting.

Introduction

On-demand ridesourcing services, operated by transportation network companies, match passengers and drivers through intelligent mobile phone applications (Rayle et al., 2016). Ridesourcing services have become increasingly popular due to their convenience. As of 2020, Uber has provided services in over 890 cities in 71 countries (Wyatt, 2020). As of 2018, Lyft has provided services in 200 cities in the US with over 315,000 drivers (Lyft, 2020). DiDi Chuxing has provided mobility services to more than 550 million users all over China. On average, there are more than 30 million travel orders on the DiDi Chuxing platform per day, representing more than 350 million vehicle kilometers per day nationwide (DiDi Chuxing, 2018).

Ridesplitting, one form of ridesourcing services, matches riders with similar routes to the same driver, such as “UberPool” and “Lyft Line” (Shaheen et al., 2016a). There are some debates claiming that on-demand ridesourcing services could increase congestion and pollution (Anair et al., 2020). However, ridesplitting, a new shared mobility service, is a more sustainable travel mode for improving traffic efficiency and reducing traffic congestion and air pollution problems. Policy makers have realized the benefits of ridesplitting services and encouraged ridesourcing companies to launch ridesplitting services for passengers. For DiDi Chuxing, the number of passengers who choose ridesplitting services has continued to increase since its launch at the end of 2015. By the end of 2019, the cumulative number of users reached 2.9 billion, with a compound annual growth rate of 143.3%. The passenger volume using ridesplitting services was equivalent to 1.2 times the civil aviation passenger volume in 2019 (DiDi Chuxing, 2020).

However, the current ridesplitting ratio in key ridesourcing services provided in the city of Chengdu is low, only 7% (Li et al., 2019). The average time usage rate of private cars in China is only 7% (Accenture, 2016), while the rate is only 4% in the United Kingdom (Bates & Leibling, 2012) and 17% in the United States (Santos et al, 2011). This low utilization rate leads to a serious waste of resources. With the development of technology for autonomous driving, idle private cars could be used for autonomous ridesourcing services or ridesplitting services in the future. Tu et al. (2019) showed that such ridesplitting services could potentially have great benefits in the city of Chengdu, specifically by improving the percentage of cost savings and time savings to 17% and 23%, respectively.

The existing literature has proven that the built environment has a strong relationship with travel behaviors (Ding et al., 2018, Ding et al., 2019, Durning and Townsend, 2015, Tao et al., 2020). However, to the best of our knowledge, no existing research has investigated how the built environment impacts ridesplitting services. To fill this gap, the aim of this study is to explore the relationship between the built environment and origin–destination (OD) ridesplitting ratio using observed ridesourcing data of Chengdu, China. Specifically, this study involves three main tasks. First, we use a machine learning method, gradient boosting decision tree (GBDT) model, to identify the important features of the built environment at the origins and destinations of ridesplitting services. Second, we explore the nonlinear relationship between the ridesplitting service and key explanatory variables by creating partial dependence plots. Third, we provide insightful results that can help transportation network companies improve existing ridesplitting services and have policy implications for urban planners seeking to better understand how the built environment, demographic factors and travel time impact ridesplitting services.

The remainder of this paper is organized as follows. Section 2 summarizes the related research. Section 3 presents the data and variables used in this study. The GBDT model is described in Section 4. Section 5 presents the results in detail and discusses the related insights regarding how the built environment features, demographic factors and travel time impact ridesplitting services. The final section summarizes the main findings, presents some policy implications and proposes future research directions.