A longitudinal study of bike infrastructure impact on bikesharing system performance in New York City

Abstract

The sustainability of bikesharing systems depends on the bicycle network connectivity and accessibility. However, studies of bike lane infrastructure investment impacts on bikesharing ridership have been limited to station-level ridership to support bikeshare operators’ resource allocation decisions. City planning agencies also need to forecast and analyze bikeshare demand in order to make investment decisions of bike facilities over time. To measure the marginal effect of building bike lanes on bikeshare demand at a network-wide level over time, an autoregressive conditional heteroscedasticity (ARCH) model with autoregressive (AR) disturbance is proposed to capture system-wide bike ridership. The model is applied to investigate the relationship between bikeshare average daily trip counts and the total length of bike lanes in New York City (NYC) regardless of specific locations. Our results show that the installation of one additional mile of bike lanes in NYC led to an average increase of 102 bikesharing daily trips. Spatial heterogeneity is addressed through two market segments by borough (Manhattan vs. Non-Manhattan), which indicates that the improvement of bike lanes had a significant impact on bikesharing ridership in Manhattan, generating 285 more trips with one mile of bike lane built in Manhattan, but was not as effective outside of Manhattan. In addition, model results show that there were 135 and 13 more trips generated per day when one more bikesharing sharing station was added in Manhattan and Non-Manhattan, respectively. The marginal effect could vary due to the specific implemented location of a bike lane. We demonstrate that this model, as opposed to the previous studies in the literature developed at the station-level, can provide new insights into system-level causality and temporal lag characteristics. When considering whether a city agency should install a set of bike lanes immediately or to stage them over multiple weeks, the “discount rate” for ridership benefits is estimated via a proposed model for different types of investment. The results of investment scenario analysis show that there is a higher effective “discount rate” for ridership benefits from front-loading all of the investment.