The Montreal evidence for cycle track-to-street safety

A recent paper on Montreal cycle tracks has drawn a lot of publicity from the official news sources and blogs without an appropriate amount of skepticism for the results. 

Briefly, the authors compare six cycle tracks to reference streets to compare the relative risk of cycling on those streets.  Exposure is determined by a two-hour simultaneous count on 2009 on the streets.  The number of injuries is tabulated from emergency medical reports (EMR) from 1999 though 2008 during the cycling season; but are subsequently adjusted by estimating the number of intersection injuries from bicycle/vehicle collision police reports.  Aggregating over all six cycle tracks they conclude that the relative risk of cycle track streets to the reference streets is approximately 0.72.  Moreover their two-hour simultaneous count estimated 2.5 times more cyclists on cycle track streets.

Fundamentally, the relative risk measure is only valid if the cycle track and reference street have similar qualities and would face similar risks if not for the cycle track.  Below are person-views -- roughly at the beginning of the focus area -- with maps of the first comparison in Table 1 of the paper.


St. Denis ... the reference street


Brebeuf ... the cycle track street

Personally, I think a picture is worth a thousand words.  In the interest of completing the point, a cyclist traveling northbound on St. Denis has to deal with several more lanes of motor vehicle traffic including opposing direction traffic -- left turns into driveways, u-turns, buses and so on -- in a commercial area.  Moreover, it is unclear to me what type of injuries are in the EMR data; but if pedestrians collisions are included then the St. Denis cyclist would have people shopping and dining in the area.  Wider roads also tend to have faster motor vehicles which increase the likelihood of a serious injury.  As one might expect, Table 3 show that the relative risk to motor vehicle occupants is significantly different from 1.  In short, to find no statistical difference in relative risk would be far more surprising than their estimate.  Identifying any effect from the cycle track is highly unlikely, in my opinion.  Just to be clear, I picked this comparison because it was listed first.  One could grind through the other cycle track-reference street comparisons to be more thorough but I suspect that the point has already been made.  Others have done this precise task.  If and when they write up a paper or webpage, I will be sure to link it.

In large, we can probably stop here with regards to raising a serious level of skepticism about the study.  However, there are also some serious questions about the data underlying the relative risk metric.  The exposure data was collected during a two-hour period on a day with mild weather.  That data was used to compute the exposure ratio for the cycle track and reference street during a 10-year period of cycling seasons.  Now, if we ignore the huge difference in reference periods, we should consider whether the population of cyclists on the cycle track and reference street are similar enough that this remains a good estimate subject to changes in weather, time of day, and other random perturbations over time.  Again a picture is worth a thousand words.  Does anyone expect similar cyclists on the two pictured streets above?

As I mention above, the EMR injury data that occurs at intersections is adjusted for the expected number of cyclists originating from the cross street instead of traveling on the streets of interest.  The authors estimated the fraction from the cross street based on tabulations of police reports which overlap but do not fully cover the same reference period.  But do EMR and police bicycle/motor-vehicle collision reports have the same meaning?  Certainly they are correlated, but simply based on anecdotes in the US, when a police report is filed seems to vary considerably and could differ from say whether an ambulance is sent to the scene.  For instance, until a recently, an Arlington police report would only be filed if the cyclist in a collision with a motor vehicle went to the emergency room despite emergency personnel being sent to the scene.   Note that part of this uncertainty is about my limited understanding of Montreal/Canadian laws and data: It could be the case that the relationship is well understood.  But the short blurb in the paper omitted a thorough discussion and cited a paper about New Zealand.  Consequently, I couldn't tell whether their conclusion was based on their own tabulations/local knowledge or the New Zealand paper.  Nonetheless, besides the potential bias regarding the share of intersection EMRs originating from the cross street, modeling this component also induces a bias in relative risk due to the variance of the estimated shares.  At the very least, it would have been prudent to do some robustness estimates by using police report collision data only to compute the relative risk ratios.  There are 340 and 191 EMR injuries for cycle track and reference streets using the authors' method.  If the overlap is high then one could probably get good estimates to make comparisons.

In the interest of brevity, I want to avoid nit-picking another dozen points on the paper and instead make a few general points relevant to cycling advocacy.  One, many discussions about safety conveniently ignore notions of convenience and speed.  Pro-facilities people gloss over increased conflicts at intersections and slower travel.  Vehicular cyclists (VC) tend to ignore the ability for skilled cyclists to use something like a bike lane to pass motor vehicles stuck in a traffic jam which might be safer than filtering up in an alternative environment.  By simply looking at collisions, fatalities, or whatever, we often forget that these things are an imperfect proxy for the true object of interest: getting around quickly and conveniently with a low amount of risk and effort.  My point is that as advocates, we need to mentally process the statistics and research into something more appropriate for our goals.

Two, the concerns that AASHTO and many traffic engineers have with cycle tracks ( 1, 2)  and similar side paths that have a theoretical and empirical base.  And there are certainly lots of anecdotes from experienced cyclists -- not necessarily VC -- that echo the concerns.  So when we read that there is research concluding the cycle tracks makes cycling safer in an all-things-equal context -- for contradictory evidence read this paper on Copenhagen -- it deserves a lot of skepticism and/or it needs further consideration with respect to my first point regardless of one's beliefs about the world.  Given the strong feelings among certain groups, I want to emphasize that bike lanes, wide-outside lanes, cycle tracks or whatever should still be under consideration for some set of circumstances.  Instead, I want to emphasize that one should be much more careful about implementation than the current euphoria over cycle tracks.

Lastly, in the context of building livable neighborhoods, I share many goals with cycling/walking advocates and other "smart growth" people.  But part of smart growth means being good scientists, properly evaluating all evidence that either confirms or rejects our priors, and giving it the appropriate weight when making decisions and choosing what to advocate.  Most importantly, this includes considering much more difficult topics regarding causality and what we don't know.  

P.S.  M. Kary, a statistician living in Montreal, wrote a response to the paper here.  Thanks to John Allen for pointing out the paper and adapting it for the Internet.