A GPS-based bicycle route choice model for San Francisco, California

Recognizing the environmental and health benefits of cycling, cities around the world are promoting use of the bicycle for everyday transportation, but with limited information about the preferences of cyclists and the effectiveness of investments in bicycle infrastructure. To better understand the decision-making of cyclists, we estimated a route choice model with GPS data collected from smartphone users in San Francisco. Traces were automatically filtered for activities and mode transfers, and matched to a network model. Alternatives were extracted using repeated shortest path searches in which both link attributes and generalized cost coefficients were randomized. The prior distribution for the coefficients was calibrated automatically using only the network. A Path Size Multinomial Logit model revealed that bicycle lanes were preferred to other facility types, especially by infrequent cyclists. Steep slopes were disfavored, especially by women and during commutes. Other negative attributes included length and turns. Traffic volume, traffic speed, number of lanes, crime rates, and nightfall had no effect. Marginal rates of substitution imply a user benefit of bike lanes of $0.61 USD per km per trip. Coefficients were applied to a trip assignment model that will be used to evaluate prospective investments in bicycle infrastructure in San Francisco.