Bicycling is generally regarded as a recreational activity, especially since there are few well-developed networks of bicycle paths in urban areas. However, bicyclists are making their presence known. Those living in the Washington, D.C., metro area are requesting that officials develop plans for new bicycle lanes that will ultimately create a network of connectivity between neighborhoods and commercial and business centers.

It’s interesting to see which street characteristics influence most bikers’ perception of level of service (LOS). At this week’s annual Transportation Research Board (TRB) meeting, I presented on using the cumulative logistic regression model for evaluating bicycle facilities on urban arterials—which is also a paper written by myself, Asma Ali, and Aimee Flannery.

Results of the model indicate that the presence of a bicycle lane or shoulder increases the LOS rating, as expected; however, an interesting finding is that bicycle lane width did not matter as much. On roadways with speed limits greater than 30 mph, as well as having more than one travel lane, and for streets with numerous unsignalized intersections or driveway entrances, the LOS rating reduces. Interestingly enough, the posted speed limit is a factor influencing bicyclist’s perception of LOS more than the presence or absence of a bicycle lane.

The model uses 1,400 data points gathered through a previous study to determine LOS and estimates the entire distribution of bicycle LOS rating probabilities, versus a mean probability. Four variables that significantly impact bicycle LOS on urban arterials were identified; opposed to 13 found through other studies.

The Results

In summary, the Cumulative Logistic Regression Model provides statistical evidence for practitioners that four geometric and operational street characteristics are sufficient to analyze the perceived LOS on bicycle facilities:

• Presence of a bicycle lane or shoulder
• Posted speed limit
• Number of travel lanes
• Unsignalized conflicts per mile

All these characteristics are usually readily available for practitioners and require a minimum amount of effort. The modeling technique provides a distribution of the estimated probabilities of ratings instead of one mean value, providing a better understanding of the data. The model can be applied to data pertaining to pedestrians and automobiles in a similar approach.