Estimating Censored Spatial-Temporal Demand with Applications to Shared Micromobility
share
In shared micromobility networks, such as bike-share and scooter-share networks, operators and city planners are interested in understanding user demand. However, observed trips do not equate directly to demand. The distribution of available bikes affects the distribution of observed trips both through the censoring of potential users who cannot find a nearby bike and the spatial dependence between where a user originates and where a trip begins. The ability to use trip data to accurately estimate demand in both docked and dockless systems is key to analyze the number of dissatisfied users, operational costs, and equity in access, especially for city officials. In this paper, we present a flexible and interpretable framework to estimate spatial-temporal demand by explicitly modeling how users interact with the system. This choice model and algorithm was informed by our collaboration with city planners from Providence, RI, and we demonstrate our algorithm on data from Providence's dockless scooter-share network. Our estimation algorithm is publicly available to use through an efficient and user-friendly application designed for other city planners and organizations to help inform system planning.
READ FULL TEXT