Exploring urban taxi ridership and local associated factors using GPS data and geographically weighted regression

• Published in: Cities 2019-04, Vol.87, p.68-86
• Main Authors: Li, Bozhao, Cai, Zhongliang, Jiang, Lili, Su, Shiliang, Huang, Xinran
• Format: Article
• Language: English
• Subjects: Access; Business hours; Cardinality; Clinical assessment; Clustering; GWR; Land; Mobility; Passengers; Regression analysis; Taxi ridership; Taxi trajectory data; Taxicabs; Traffic; Traffic source and sink places; Transportation; Travel; Urban areas; Urban mobility; Variability; Variables
• ISSN: 0264-2751; 1873-6084
• DOI: 10.1016/j.cities.2018.12.033

Taxi is a core component of urban transit systems. Since they can provide more time-saving and convenient service than many other transit options, taxis have a certain passenger base. The analysis of taxi ridership can be used to better understand the travel mobility of passengers and the traffic structure of urban areas. In previous studies, taxi trajectory data have been widely used, especially in exploring taxi ridership, and point-of-interest (POI) data are usually used to evaluate the land-use type of a certain sub-district. On the basis of preceding research, this paper uses taxi trajectory data within the long time scale of one week. Five traffic factors are taken into consideration: pick-ups, drop-offs, and the ratio of pick-ups to drop-offs, pick-up probability and drop-off probability. The research model is divided into weekdays and weekends. For the calculation of probabilities, an index termed the Area Crossing Index is proposed to reflect the taxi cardinality and accessibility of a region. At the same time, POI and demographic data are used as explanatory variables. In this study, we also take the business hours of POIs into consideration. In order to explore the ridership in each hour, hierarchical clustering is used to determine the similarity characteristics of hourly dependent variables. Then, stepwise linear regression is used to screen and evaluate coefficients without collinearity. Finally, geographically weighted regression is adopted to evaluate spatial variability, and the coefficients of common explanatory variables on weekdays and weekends are examined. At the end of this paper, the causes of common explanatory factors on weekdays and weekends for each traffic factor are discussed. This paper also analyzes ridership by combining all the results of dependent variables and proposes some suggestions for taxi scheduling. •The Area Crossing Index is proposed to reflect the taxi cardinality and accessibility of a region.•Five indicators are used to analysis urban taxi ridership from different angles.•Full-day ridership patterns research through POI data with business time and demographic data•Analyze the taxi activities and taxi spatial-temporal distribution rules•Discuss urban traffic source and sink place, and propose some suggestions for taxi scheduling