Abstract
Prioritization of networkwide bicycle investments is data limited in the United States. The framework proposed in this paper addresses the temporal factoring of sparse bicycle counts through Markov chain Monte Carlo sampling and introduces a novel spatial factoring method to expand estimates of bicycle usage to all network edges. Bicycle usage varies widely on the basis of weather, infrastructure, trip origin and destination, and cultural expectations, and this variability necessitates more-detailed volume models than those that suffice for automobile use. A multilevel temporal model that includes hourly, weather-related, and commute-day factors maximizes the information obtained from sparse count observations. Spatial factoring then extends these data to cover unobserved streets through Bayesian updating of prior estimates from a regional travel demand model informed by an edge correlation matrix. For a small city in the United States with some manual volunteer bicycle counts and no permanent counting infrastructure, the proposed framework was able to estimate an edge-specific bicycle usage networkwide reasonably and, unlike typical factoring methods, as distributions rather than single values. This rigorous characterization of parameter variance allows planners and software to interpret results appropriately and to avoid the common misconception that all model outputs are equally valid. The framework is globally applicable because it is based on open-source tools and data and will be used in the upcoming long-range plan for the study region. By providing comprehensive safety exposure data, the framework enables networkwide safety prioritization with empirical Bayes methods to allocate scarce funds.
Get full access to this article
View all access options for this article.