Temporal and Spatial Pattern of Shared Bike Trips --- An Empirical Study of New York City

Abstract

Public bike share system wins its global popularity due to inherit merits. A key problem of expanding as well as maintaining the system is predicting the demand of bike trips temporally and spatially, either for expanding the system or for balancing the bikes and docks. In this paper, we investigate the contributing factors in daily trips of Citi Bike, the bike share system in New York City from the city level and census tract level. Factors coming from demographics, employment, weather, land use, subway, and taxi trips are investigated temporally by a simple multiple linear model and spatially by spatial filtered negative binomial (NB) regression models. The results of multiple linear model show the daily shared bike trip counts of the whole city are significantly dependent on weather, day type, and taxi trips. The Moran’s I statistic confirms the spatial correlation in ordinary NB regression when modeling average daily bike trips in each census tract on both weekday and weekends. With additional spatial filters, the improved NB regression models can better model the spatial autocorrelation and show a better fitting. Meanwhile, spatial filtered NB regression models confirm the significance from explanatory variables of population, demographic data, land use, etc. The results are helpful in obtaining potential locations of bike stations for the expansion of shared bike system and arranging available bikes and docks temporally and spatially for daily management.