M. Kary, who prepared a review of the Lusk et al Montreal study, has had a preliminary look at the Monsere, Dill et al. study of barrier-separated on-street bikeways (“cycle tracks”) which the bicycle industry lobby PeopleforBikes is promoting as demonstrating their safety. Dr. Kary has given me permission to publish his comments here.
An Introduction To and Overview Of:
Monsere C, Dill J, et al. (2014) Lessons From The Green Lanes: Evaluating Protected Bike Lanes In The U.S. Final Report, NITC-RR-583
To begin with a platitude: traffic accidents are rare events. The totals are large only because the overall volumes of exposure are huge. Therefore, if considering safety in terms of outcomes rather than the underlying mechanisms of operation, any facility, no matter how poorly designed, will appear safe if examined over a short period of time.
But collecting data over a long period of time has its disadvantages too: not just cost and delay, but also the averaging, and therefore blurring, of the effects of various changing causes and circumstances. Nor does it work at all for facilities that are yet to be built. In response to these problems, engineers developed the methods of traffic conflict analysis. They can be seen as based on the following logical and kinematic necessities. First, in order for a collision to occur, the vehicles involved must eventually get on a collision course. Second, in order to get on a collision course, they must first get on a near-collision course. On the other hand, not all vehicles once on collision or near-collision course do end up colliding: their operators make course corrections and avoid that outcome. Such potentially dangerous but often ultimately safe trajectories, i.e. traffic conflicts, occur much more frequently than actual collisions, deaths, or injuries. If there exists a suitable relationship between the former and the latter, then conflict analysis can be used to study road safety at reduced cost, with better timing, and even via simulation modelling of facilities that have been designed but not yet built.
The theory and practice of conflict analysis for motor vehicles has been developed over something like a half a century of research. This has evolved to quantitative methods using not just traffic cameras, but also instrumented vehicles, automated data extraction, and theoretical concepts such as time to collision, gap time, gap acceptance, post-encroachment time, and many others. There is no such corresponding body of research for bicycles. Even if there were, it could never be as important to bicycle or pedestrian deaths and injuries as it is for the occupants of cars and trucks: for example, the latter vehicles never topple over at stops or just slip and fall, so that their occupants fracture an arm or strike their heads on a curb. In fact the majority of bicyclist injuries, even those requiring hospitalization, apparently involve only the bicyclist, making conflict analysis entirely or at least largely irrelevant to them.
On the other hand collisions with motor vehicles are major factors in cyclist deaths and injuries, and they are what cyclists worry most about. And even apparently bicycle-only crashes can be provoked by e.g. general fears or specific intimidations, or avoidance manoeuvres leading to loss of control. Thus there are also dimensions of traffic conflicts applicable to bicycling, but either inapplicable or less so to motor vehicle-only conflicts. Nor is every conflict visible or strictly kinematic: consider for example the effects of sudden and loud horn honking or engine revving.
With these fundamental limitations in mind, obviously traffic conflict analysis is a promising method for investigating important aspects of bicycling safety. The theory needs to be developed, so we can figure out what constitutes a high or low rate of conflicts, what types of conflicts figure what way into which accident types, and how vehicle operators and pedestrians cope with them, such as through hypervigilance, or avoidance of the area and thus diversion of problems to a different one.
Not only does the theory need to be developed, but also the methods of data extraction and analysis: the subjective review of traffic camera recordings, typically of low quality, is a mind-numbingly tedious, labour-intensive and error-prone task, that does not scale well.
The work of Monsere et al. (2014), Lessons From The Green Lanes: Evaluating Protected Bike Lanes In The U.S., should be considered a pilot project in this effort, although the authors themselves do not describe it as such.
Monsere et al. aimed to address six questions:
- Do the facilities attract more cyclists?
- How well do the design features of the facilities work? In particular, do both the users of the protected bicycle facility and adjacent travel lanes understand the design intents of the facility, especially unique or experimental treatments at intersections?
- Do the protected lanes improve users’ perceptions of safety?
- What are the perceptions of nearby residents?
- How attractive are the protected lanes to different groups of people?
- Is the installation of the lanes associated with measureable increases in economic activity?
Apart from noting that, as with most sociological research, their survey response rates were dismally low (23-33% overall, counting even only partially completed surveys as full responses), to produce a socioeconomically skewed sample (e.g. the bicyclists being 89% white, 68% male, 82% having at least a four-year college degree, and 48% with annual incomes over $100,000)— this overview of their work considers only the first part of their question No. 2.
Monsere et al. installed video cameras along short bicycle sidepaths (“protected lanes”, “cycle tracks”) constructed between approximately the summer of 2012 and the early summer of 2013 as part of the Green Lanes Project. These were in four U.S. cities, San Francisco (two 0.3 mile paths), Portland (one 0.8 mile path), Chicago (0.8 and 1.2 mile paths) and Washington (a 1.12 mile path; no cameras were installed in Austin, although sociological surveys were conducted there). They did their video recording chiefly at intersections, six in these four cities in the summer and fall of 2013. This was then presumably while the users were still in a cautious or exploratory state, as they got used to the new facilities.
Only 12-18, or in one case 20, independent hours of video were analyzed from each intersection. As each intersection examined was given a unique treatment, results cannot easily be pooled. These are very small numbers.
(This makes for substantially less than 120 hours total. The authors seem to say they analyzed 144 hours of video at intersections. This would mean that some of this total came from multiple cameras examining the same intersection at the same time. The authors do show frame captures from some of their cameras. This observer would find it difficult to correctly identify the conflicts from the views on display.)
As noted following the opening platitude, any facility, no matter how poorly designed, will appear safe if examined over a short enough period of time.
The six facilities examined were all so new (less than or little more than a calendar year old) that there were no injury or death data available for them. (For comparison, the entire city and island of Montreal, with all its thousands of intersections, averages of late about five cyclist deaths and 25-50 police-recorded serious cycling injuries per year.) Thus, there would not have been a way to use even many more hours of recording to examine for any relationship between the surrogate outcomes (conflicts, violations or errant behaviours) and the outcomes of most interest, deaths and injuries.
Further, as this was neither a before-after study nor a comparison with standard intersections, there is no way to know whether the numbers of observed conflicts, violations, or errant behaviours, were themselves high or low.
As to the actual results from this pilot project, the much touted headline was that there were only six minor conflicts found, out of nearly 12,900 bicycle movements through intersections. The most basic problems with this headline are:
1. It is the wrong comparison. The conflict rate has to be the number of conflicts divided by the number of occasions where at least two users capable of conflicting are present, e.g. a bicycle and at least one other bicycle, pedestrian, or motor vehicle. Thus the authors give figures of 7574 turning motor vehicles, but only 1997 turning motor vehicles with bicycles present. The corresponding conflict rates (which they normalize by the products of bicycle and motor vehicle movements, not by the numbers of bicycle movements alone) they give for the individual intersections therefore vary by factors of approximately 3 to 10, depending on which figures are used.
2. Six is the total of observed “minor” conflicts, not the total number of observed conflicts. There were also 379 “precautionary” conflicts with motor vehicles, 216 with pedestrians, and 70 with other bicycles.
3. Besides conflicts, there were numerous violations or other errant behaviours: e.g. 9-70% of bicycles and 7-52% of turning motor vehicles in the various intersection designs used the lanes incorrectly, 1-18% of turning motor vehicles in the various mixing zone designs turned from the wrong lane, 5-10% of motorists turned illegally on red arrows at intersections with bicycle-specific signals, and 7-23% of bicyclists disobeyed their signals.
4. Without any theory or model of how any of these occurrences or their frequencies relate to death, injury, or property damage, and without any before-after or non-sidepath comparison data— not to mention, with the very small numbers of observation hours— there are almost no safety implications, positive or negative. The only concrete result is that one of the local authorities apparently deemed the problem of motor vehicles turning from the wrong lane (18%), straddling lanes (another 17%), or entering the turn lane early (15%) to be so severe that they later removed the intersection treatment and replaced it with another design (at Fell and Baker in San Francisco).
5. The sociological surveys tell another story: one-third of all bicyclists surveyed said they had been involved in at least one near collision on the paths, while 2% experienced an actual collision. 23% had a near collision with turning cars, 1.8% an actual collision with turning cars; 19% a near collision with a pedestrian, and 0.4% an actual collision with a pedestrian.
In short: this is an interesting pilot project, whose methods are impractical for the amount of data collection needed for meaningful safety results. Even with better methods, conflicts are only one facet of the bicycling, and overall safety picture; while road designers and road users, whether bicyclists or motorists, have to consider more than just safety. Convenience, transit time, cost, and greenhouse gas emissions also matter. A cycle track that, like the downtown de Maisonneuve track in Montreal, lies largely dormant in the winter, but delays motor vehicle traffic in the winter and ties it up spring, summer and fall, will be of no help in reducing CO2 emissions. The much touted headline results from this study are selective, overblown, and misleading. Any facility will appear safe if examined over a short enough period of time, and surely 12 to 20 hours each is short enough