Tuesday, January 6, 2015

Bicycling: Safety in Numbers

This article was originally published to LinkedIn on November 11, 2015 and represents an abridged version of the term project for CEE 567 – Health and Transportation at the University of Washington.

Photo Credit - Steven Vance (CC BY-NC-SA 2.0)

Citing ‘soaring’ popularity in bicycle use in the Greater Boston area, the Boston Globe presented some collision statistics in their October 11 article with the striking headline, Bike fatalities rise in Greater Boston’s suburbs. A statewide bicycle commuter increase of 75% from 2000 to 2012 paired with a 13% increase in collisions involving bicycles from 2010 to 2012 prompted Concord police lieutenant Roy Mulcahy to publicly declare that, “With the sheer number of bikes being ridden, there are going to be more accidents. It’s a statistical fact.” That sentiment was echoed less convincingly by Massachusetts DOT bicycle and pedestrian coordinator Josh Lehman when he stated, “More exposure generally leads to greater crashes.” Lehman’s qualifier ‘generally’ and Mulcahy’s convinced statement of ‘fact’ invite a scientific inquiry into whether an increase in cyclists definitively results in a greater bicycle-motor vehicle collision count.

P.L. Jacobsen (2003)
In order to gain a more nuanced understanding of the relationship between cycling volumes and incidents with motor vehicles, as well as more thoughtful statistical presentation, we turn to the body of peer-reviewed literature on the topic. The leading contemporary explanation is advanced by Peter Jacobsen. His ‘safety in numbers’ theory notes a personal likelihood of being struck by a motorist that varies inversely with the amount of walking or bicycling. This does not say that the total number of bicycle-motor vehicle collisions is reduced, but rather that each individual cyclist is less likely to be involved in a crash according to a non-linear risk relationship. Jacobsen posits that an apparent adjustment of the behavior of motor vehicle operators in the presence of people walking and bicycling is responsible for this observation. The finding itself is statistically supported by all remaining literature reviewed which draws statistical conclusions, though a strong call for caution is made against causal inference in the statistical association as a counterpoint.

Answering the overarching question is far from cut and dry. There are several factors which can correlate with increased or decreased collision risk for vulnerable road users. Jacobsen’s original work carried additional weight because it included data from communities of varying size, different countries, and across time periods. Siddiqui et al. found a relationship between crashes and the length of road with certain posted speed limits (15 and 35 miles per hour). Most noteworthy is the fact that the length of road with a posted speed limit of 15 miles per hour is negatively correlated, meaning that a built environment variable can have a significant impact on cyclist safety either by discouraging traffic volume of by governing motorist behavior.

A roadway design aspect shown to have a strong effect on crash rate is the prevalence of right turn traffic, which can be amplified by the presence of a right-turn lane or an increased turning radius. Additionally, other design elements such as the presence of a bus stop located at the intersection, the crosswalk length, and whether a median is located in the road under analysis are shown to have significant impact of the likelihood of collision at a given intersection.

Finally, an increase in cycling is not always measured in terms of the number of individuals utilizing the mode. Both Jacobsen and Dorothy Robinson incorporate datasets into their analyses which look at the average daily distance cycled, another logical representation of risk exposure. These limited examples of alternate cycling quantification do not appear to affect the risk relationship curves.

Photo Credit -
Flickr User straightedge217 (CC BY-NC-ND 2.0)
Jacobsen’s research provided an important starting point when considering questions of road safety for more vulnerable users. Whether the measure of a particular data set is mode share, distance per day, or trips per day, the relative risk, injury rates, or fatality rates of both cyclists and pedestrians follow similar curves. These collision risk curves both take the form of a power curve where the total number of bicyclists struck by motorists varies with the 0.4 power of the amount of bicycling. Thus, a community doubling its cycling can expect a 32% increase in injuries while the personal injury risk of an individual cycling in a community with twice as much bicycle volume will reduce to 66% of the original value.

Consistent with safety in numbers principles, Robinson's Australian analysis found that states with greater levels of cycling had fewer injuries per kilometer cycled while individual injury rates dropped in Western Australia concurrent with an increase in regular cyclists. Tangentially, the introduction of mandatory helmet legislation in Victoria appears to have unintentionally led to an increased risk of injury due to a drop in levels of cycling. This outweighed any benefits of increased helmet wearing, further proving the importance of cyclist concentration as a safety factor. A more recent study by Nordback et al., focusing on Boulder, Colorado, further bolstered the argument by finding that bicyclist-motorist collisions at signalized intersections increase non-linearly with increasing bicyclist and motorist volumes, collisions per cyclist decrease with increasing cyclist volumes, and models indicate that intersections with fewer than 200 average daily cyclists have substantially higher collisions per cyclist.

Photo Credit -
Flickr User Jacob-uptown (CC BY-NC 2.0)
A pair of studies of signalized intersections in Montréal also found that bicycle injury does not increase proportionately to bicycle volume. By disaggregating cyclist and motor vehicle movement for each 15 minute interval and each intersection approach direction, Miranda-Moreno et al. found that the elasticity constant for injury rate given additional bicycle flow was found to be 0.44. A similar data set was employed in a Bayesian modeling approach by Strauss et al. with the stated goal of ranking bicycle corridors on the Montréal Island for safety risk. In the process, some insight was gleaned regarding the precise impacts of other built environment factors. The Bayesian model finds that cyclist volumes have a strong association with injury occurrence, the strongest association of all literature surveyed, with an elasticity of 0.87. This is still not a proportionate effect as corridors with high bicycle volumes have a higher number of injuries but tend to have a lower risk per cyclist.

The largest and most consequential analysis of the non-linearity of injury risks for pedestrians and cyclists was undertaken by Rune Elvik. What differentiates Elvik’s final results from the others is the notion of trip transfer. Trip transfer signifies a reduction in automobile trips is factored in as alternate transportation mode share increases. The Elvik model merged nine extant studies to determine whether a reduction in the total number of accidents is indeed possible. A conclusion is reached that in theory, the total number of accidents involving vulnerable road users and motor vehicles could go down if a substantial share of trips by motorized transport is transferred to walking or cycling. According to the data tables, the absolute requirement to realize such an outcome independent of baseline motorized traffic counts is a transfer of 25% of motor vehicle trips to the other modes. Doubling the number of pedestrians and cyclists, with an attendant decrease in motorists, will also achieve an absolute reduction in collisions if the baseline pedestrian/cyclist mode share is 5.7% or greater. Elvik notes that these necessary motor vehicle trip reductions may not be plausible to achieve in the near-term. However, he cites a recent Norwegian behavioral study which finds that almost half of car trips are shorter than a reasonable cycling distance and almost 20% within a feasible walking radius, implying that change on this order of magnitude is far from impossible.

It is important to consider that the results in such studies are heavily dependent on recording and reporting methods. Collision severity has a massive impact on the quality of collision databases. In general, most parties involved in no-injury accidents do not report the incident to the police. Chang and Wang report that in Taiwan, only 2% of bicycle accidents are listed as ‘no injury.’ which seems rather unlikely given the individual injury risk of cycling. Miranda-Moreno et al. and Strauss et al. chose ambulance data in order to document collisions, citing fewer reporting errors compared with police report data.

Photo Credit - Justin Swan (CC BY-NC-SA 2.0)
It should also be noted that root causation for the non-linear risk relationship is still currently logical speculation. Bhatia & Wier provide the most serious rebuttal to the ‘safety in numbers’ movement. They contend that it is dangerous to shift cause and responsibility for safety to vulnerable road users. This shift would ignore the confounding environmental factors previously discussed. Their critical assessment reiterates the still unproven nature of the idea that heavily travelled cycling and pedestrian routes are safe because more vulnerable road users are present. There is potential that a value preference for intrinsically or perceived safer facilities is reflected in the concentration of these behaviors. More recent research on the topic by Nordback et al. calls for collision data where infrastructure does not change, but bicycle volumes vary in order to help reveal the underlying factors.

In conclusion, an assessment of the contemporary literature supports the claims of Lt. Mulcahy and Mr. Lehman, generally confirming the conception that adding bicycle traffic to a corridor increases the absolute number of collisions with motor vehicles. However, individual risk is uniformly shown to be reduced over multiple studies due to a non-linear risk-exposure relationship. In addition, certain aggressive mode transfer scenarios exist that allow for an absolute reduction in collisions between bicycles and motor vehicles and thus, the total number of cyclist injuries. Finally, from a prescriptive standpoint, future roadway design should consider exterior relationships which contribute to safety risk including frequency right turn movements, presence of bus stops, crosswalk length (implying road width), and the presence of raised medians.

No comments: