BikewaySim: Using Changes in Revealed Impedance to Assess Perceived Safety Benefits of New Cycling Infrastructure 

Abstract
According to the 2017 National Household Travel Survey (NHTS), about 46% of all single occupancy motor vehicle trips in the U.S. are 3 miles or less, yet the NHTS estimates that only 1% of all trips are accomplished by bicycle (1). This mismatch is suspected to be primarily due to the real and perceived danger imposed on cyclists by drivers when they share the same right-of-way (2). However, with new bicycle design guidelines from organizations like NACTO, the resources are available to design high quality, separated bike infrastructure to reduce the perceived danger of cycling. However, engineers and planners need to know what type of cycling infrastructure is needed and where to best place it to make cycling more accessible.

This research aims to develop a tool called BikewaySim to analyze how bicycle infrastructure contributes to decreasing traffic stress and increasing bikeability (i.e. the accessibility of cycling), so that alternatives that stand to have the greatest impact are selected over those that would be less effective. Utilizing the findings of stated and revealed cycling route choice and bicycle facility preference research, we are developing impedance functions that serve as a proxy for measuring bikeability. These functions will be used in three ways: routing individual cycling trips, quantifying aggregated zone-to-zone impedance, and delineating impedance adjusted bikesheds.

Our initial work has been focused on a comprehensive literature review and network development. The literature review focusses on past studies that have utilized GPS-recorded cycling trips to investigate cyclists’ revealed preferences for road attributes, such as the presence of a bike lane, presence of on street parking, the speed limit, and the number of lanes (3, 4, 5). Collectively, these studies show that cyclists choose a route that’s further than the shortest distance route by around 8-18% on average (6). The studies also show that cyclists prefer routes that are shorter, have a higher percentage of bicycle infrastructure (both separated infrastructure and traditional bike lanes), fewer turns, and fewer steep hills. The relative value that cyclists have for these attributes is usually interpreted in terms of marginal rates of substitution (MRS) of distance (4).

For network development, we looked at three networks available in our study area of Atlanta, GA: HERE, OpenStreetMap, and the Atlanta Regional Commission’s ABM network. We found that while OpenStreetMap had the most coverage, HERE and the ABM network had a more complete set of link attributes that would be necessary for impedance functions. These networks were conflated to yield a preliminary network with 7,138 links and 5,666 nodes in a 12 square mile study area. Lastly, we performed shortest path routing using 25,000 TAZ pair combinations to understand how shortest path routing changes when modifying link costs and adding new links. Our results show that shifting from a simplified travel demand model network to our conflated network led to a 30% reduction in trip miles on the simplified travel demand model links.