The Boston Area Research Initiative recently released a dataset of all bicycle crashes from 2009-2012. There were several articles and analyses published, but the official City of Boston report did not consider the "impact adding bike lanes, shared lanes or cycletracks had on cyclist collisions", and "recommended that future analysis begin to look" into it.
One other factor the City of Boston report didn't consider in depth was the total traffic through an area. The number of collisions might be large along a particular street or intersection, but is that because the street is dangerous, or because there is simply a lot of traffic?
We combined the BARI dataset with data released for Hubhacks by Runkeeper, a smartphone application that tracks your runs and bicycle rides, to try to get a sense of the likelihood of collisions adjusted for traffic. And we used City of Boston bike lane data to determine how many of those collisions and rides occurred on bike lanes, to get a sense of how much they help.
The following graphs display, by street, traffic and collisions between May 2010 and December 2012, filtered down to streets with at least 10 collisions and 5 km worth of Runkeeper rides.
Across the board, collision rates are lower in bike lanes, which can be visualized by the difference in height on the orange collisions/distance graph. On average, combining data for all the above streets, we found that the collision rate was 42% higher off bike lanes than on them (160/132 collisions/km on and 322/186 collisions/km off bike lanes).
Both on and off bike lanes (and especially off), it has the lowest rate of collisions per biker. However, it's possible we are missing bike lane data, which would explain its high safety. Beacon Street has a bike lane now according to Google maps, but according to the City of Boston dataset, it didn't as of 2013. Regardless, the data indicates that it's one of the safest streets.
Commonwealth Ave has the most total collisions, but its collision-to-trip ratio is not very high compared to most other streets.
These streets have the worst collision-to-trip ratios, both off-bike lanes and in total.
Blue Hill Ave doesn't have as much Runkeeper data as the other streets, but overall it has the worst numbers, and the map detail below is telling. The street has gotten better since a bike lane was added to part of it (most of the collisions in that area occurred before the bike lane was added), but a dangerous area to the north remains.
Columbus Ave and Huntington Ave don't appear as bad, but they have more data supporting them, and Huntington Ave in particular appears to have been made safer in the areas where a bike lane was added.
Note that for Columbus Ave, there are multiple collision dots overlapping at the intersections with Centre St and Heath St, indicating that those spots might be good areas to focus on (to save more lives with relatively little effort).
The purpose of this section is to provide additional details about the data and to add caveats against overinterpreting it (but also justifications that it has some validity).
The main Hubhacks dataset we used was from Runkeeper, and the initial task was to correlate it with the collision dataset. The Runkeeper data has been filtered to only include cycling trips, and only the portions of those trips that are within Boston, since the collision dataset is Boston-only (not greater Boston area). In addition, both datasets have been filtered so that they span the same time range (they overlap from May 2010 to December 2012).
Within this time range, the Runkeeper data only spans 157 distinct biking trips (out of a total of 383 biking trips represented in the full data), but expanding those trips out into their full routes, we do have 34261 biker positions across 712 Boston streets, spanning a total of 2240 kilometers.
For the same time and space, we have data on 1340 biking collisions requiring EMS intervention across 430 Boston streets.
Both data sets are seasonal, and both vary across time of day and day of week:
Notice that collisions are concentrated around commuting hours and on workdays, with more of the collisions happening in the afternoon, while biking data is a little more evenly distributed throughout the afternoon and day of week. This may reflect a systematic bias in the Runkeeper data towards casual biking, although there is undoubtedly still commuting data present in the set.
Finally, we obtained data on bike lanes in Boston (along with approximately when they were added). These bike lanes span a total of 250 kilometers across 136 streets. One weakness of the bike lane dataset is that it only includes the year each segment of each bike lane was installed (and for lanes added before 2007, there's no year). That means, for collisions or Runkeeper trips that occur in the same year the bike lane was added, we don't know for sure whether that collision or trip point was on a bike lane. We assume yes for our visualizations, and even if we're wrong it shouldn't bias the results too much, but it would be useful to know exactly when the lane became active.