Last updated: 2017-08-07 11:08

In March 2015, I was hit by a car while on my bike in a bike path crosswalk, and was ticketed for failure to yield the right of way. I got interested in bicycle safety and, as a software engineer by trade, I decided to download, parse, and quantify as much collision data as possible from the Lincoln Police Department.

Interpreting These Data In the data below, bike-car collisions are categorized based on where they happened. The following locations are tracked: Bike Lane : Collision happened while a person on a bicycle was riding in a bike lane, including in an intersection.

: Collision happened while a person on a bicycle was riding in a bike lane, including in an intersection. Bike Trail : Collision happened while a person on a bicycle was riding on a bike trail, not in an intersection.

: Collision happened while a person on a bicycle was riding on a bike trail, not in an intersection. Bike Trail Crossing : Collision happened while a person on a bicycle was using a crosswalk between two bike trail segments, or continuing past the end of a bike trail over a crosswalk to the start of a sidewalk. This does not include cyclists crossing perpendicularly to a bike trail, unless they are continuing on another bike trail on the other side.

: Collision happened while a person on a bicycle was using a crosswalk between two bike trail segments, or continuing past the end of a bike trail over a crosswalk to the start of a sidewalk. This does not include cyclists crossing perpendicularly to a bike trail, unless they are continuing on another bike trail on the other side. Crosswalk : Collision happened while a person on a bicycle was using a crosswalk between two sidewalks; that is, attempting to cross a road or intersection either in a painted crosswalk that is not part of a bike trail, or crossing an intersection from one sidewalk to another. This does not include cyclists who are riding on the street and pass through the intersection without using a crosswalk, nor cyclists in a crosswalk that is part of a bike trail.

: Collision happened while a person on a bicycle was using a crosswalk between two sidewalks; that is, attempting to cross a road or intersection either in a painted crosswalk that is not part of a bike trail, or crossing an intersection from one sidewalk to another. This does not include cyclists who are riding on the street and pass through the intersection without using a crosswalk, nor cyclists in a crosswalk that is part of a bike trail. Elsewhere : Collision happened elsewhere (alleyways, parking lots, etc.). This also includes ride-outs at mid-block and other collisions that happened on the road, but where the cyclist was not riding on the road as such.

: Collision happened elsewhere (alleyways, parking lots, etc.). This also includes ride-outs at mid-block and other collisions that happened on the road, but where the cyclist was not riding on the road as such. Intersection : Collision happened while a person on a bicycle was riding through an intersection on the road, not using a crosswalk. Does not include cyclists who entered the intersection from a bike lane.

: Collision happened while a person on a bicycle was riding through an intersection on the road, not using a crosswalk. Does not include cyclists who entered the intersection from a bike lane. Road : Collision happened while a person on a bicycle was riding on the road, excluding intersections.

: Collision happened while a person on a bicycle was riding on the road, excluding intersections. Sidewalk : Collision happened while a person on a bicycle was riding on a sidewalk that is not a bike trail. For instance, a car entering or leaving a private driveway or, in rare situations, a car that jumps the curb.

: Collision happened while a person on a bicycle was riding on a sidewalk that is not a bike trail. For instance, a car entering or leaving a private driveway or, in rare situations, a car that jumps the curb. Unknown: The police report contained insufficient information to determine where the collision happened; or the collision record came from NDOR and no police report could be located. In the data below, bike-car collisions are categorized based on where they happened. The following locations are tracked: There is some inherent difficulty in categorizing certain types of collisions; for instance, if a cyclist enters a crosswalk from a sidewalk but, midway through the intersection leaves the crosswalk, it's impossible to know if the cyclist was intending to enter the roadway, avoid an obstacle and continue in the crosswalk, or so on.

Quantitative Results The current dataset includes 48,211 reports, from January 15, 2008 to August 5, 2017. Since 2012, when LPD started widespread use of computer-written accident reports, the dataset includes 40,358 reports, 756 (1.9%) of which involved a person on a bicycle. This number is almost certainly an underestimate, for several reasons: Not every accident is reported. Non-injury accidents in particular may not be; anecdotally, I was initially advised not to file an accident report when I was hit, since my injuries were very minor.

Accident reports are not written for bicycle-bicycle or bicycle-pedestrian collisions. Nearly all of the post-2011 reports are from LPD, but 494 reports came from the Nebraska Department of Roads [sic] via Lacey Matney's 2014 study. All of the data from 2008-2011 is from NDOR. Because the first four years of data in this dataset is from NDOR, and includes less data, some of the reports below only include data from 2012 onward, while others include data from 2008. I've noted this where it's relevant. 7369 reports could not be parsed automatically, largely because they were handwritten. The years 2012-2014 have been cross-checked against the NDOR data (which includes handwritten reports), and the handful of handwritten reports from 2015 have been manually parsed.

Collision locations The majority of collisions involving a bicycle (399/756) happened in a crosswalk. Large numbers of collisions also involve people riding in the street, and people riding in sidewalks. (Most sidewalk accidents involve cars turning into private drives, but a few featured cars jumping the curb.) A smattering involves people riding elsewhere, which includes parking lots, or crossing streets in the middle of blocks (i.e., away from crosswalks). Collisions in bike lanes are very rare indeed, but that likely reflects the rarity of bike lanes more than anything. Clearly, crosswalks are far and away the most dangerous place for cyclists. No matter what other riding patterns are, cyclists almost certainly spend orders of magnitude more time out of crosswalks, so more than half the accidents happening in crosswalks is pretty clearly damning. Unfortunately clear data on the number of miles ridden on streets as compared to sidewalks are not available, so it's not possible to extrapolate from the data shown above to conclude whether it's more dangerous to ride in the street or on the sidewalk. It is worth noting, though, that 68.5% (518) of the collisions indicate a pattern of riding on sidewalks, while merely 20.2% (153) indicate a pattern of riding in the street. Unsurprisingly, no collisions were reported on off-street bike paths. (Recall that these data only include car-bike accidents; bike-bike and bike-pedestrian accidents certainly do occur on the off-street bike paths.)

Locations by age Road Crosswalk Intersection Sidewalk Bike lane Elsewhere The location of collisions varies slightly by the age of the cyclist. Again, though, without data on the number of miles ridden in various scenarios, it's hard to make much sense of these data -- especially with such variable-sized (and often small) data sets. The 51-to-60 set seems to be doing something interesting, though, with relatively more collisions in intersections and fewer in crosswalks than most other groups. Perhaps these cyclists ride on the street more often than cyclists of other ages.

Monthly collision rate Collisions Per Month Monthly Rock Island Riders Collisions Per 1000 Monthly Rock Island Riders For this chart I've invented a couple of metrics. One is Monthly Rock Island Riders, or MRIR. The city has had a bike traffic counter installed on the Rock Island Trail since July 2014, and it provides our best estimates of how many people are out riding at any given time. So it's a slightly arbitrary number, we expect that the number of riders on the Rock Island Trail roughly mirrors the total number of riders in Lincoln during each month. Combining the MRIR with the average number of collisions per month we can see a reasonable estimate of the collision rate during any given month, measured in Collisions Per Monthly Rock Island Rider (CPMRIR). Although the line looks fairly flat, this is a logarithmic scale so it's anything but -- you're 4.8 times more likely to be hit in November as in March. It's not immediately obvious why this tremendous difference -- or why there's such a peak in November. One possibility might be more darkness during the commute due to the end of Daylight Savings Time. Interestingly, the increase in ridership during the spring months doesn't track CPMRIR, either. That is, it doesn't appear that when more cyclists appear on the streets they're more likely to be hit; instead there seems to be a novelty effect, where March is quite safe (relatively speaking) and then collisions go up in April once drivers are more accustomed to seeing cyclists on the streets. Ultimately, for all the fluctuations of the CPMRIR, collisions per month correlates quite closely with MRIR (r=0.86), suggesting that our transportation system is just plain designed to bring cyclists into inevitable conflict with drivers, no matter how many of us are on the roads or what time of year it is.

Average monthly collisions Predictably, most accidents by sheer numbers happen during the summer months, simply because so many people are riding. January has the fewest collisions, making it one of the safer months.

Collisions per year The decrease from 2012 to 2013 is real, and very good news -- it's a huge drop. Since then things have been mostly flat, with the exception of a small decrease in 2015. Unfortunately it looks like 2012 was just an exceedingly dangerous year, and we're on a very slow upward trend in general. We only have bike traffic count data from 2014 onward, so it's too soon to tell if that upward trend is due to increased ridership, or to other factors, or (hopefully) if ridership increases outpace collision increases. Mean collisions per year is 123.78 and median collisions is 127.

Collisions per month January 2010, February 2015 and January 2017 are the only months without a recorded collision. August 2012 is still the most dangerous month in the data set.

Hourly collision rate Annual Collisions Per Hour Hourly Rock Island Riders Collisions Per Hourly Rock Island Rider Similar to the Monthly Collision Rate graph above, I've invented a few metrics. First is Hourly Rock Island Riders, which is an average count of the number of cyclists on the Rock Island Trail during any given one-hour period. The other new metric is Annual Collisions Per Hour, or ACPH. That's the average number of collisions in a given one-hour time slot that have happened per year since January 15, 2008. From these two metrics, we can see that the most dangerous times in terms of Annual Collisions Per Hourly Rock Island Rider (ACPHRIR) are the dark hours, followed distantly by the morning commute. In particular, the hours between midnight and 4 am are incredibly dangerous, possibly because the extremely low ridership numbers then make even the small numbers of collisions inflate the ACPHRIR. Although the ACPHRIR rate is lower, most collisions (by aggregate count) happen during commutes. It's interesting that more people are hit during the evening commute -- not the presumably darker morning commute. There are fare more collisions during the evening commute, but the morning commute is actually more dangerous because so many more people are out riding in the evening -- nearly twice as many as in the morning. It's also interesting that we see a steady stream of cyclists all day long, but collisions drop off sharply outside of commutes. It seems likely that this is due to the number of drivers out (and possibly their perceived urgency). Initial results (below) indicate that darkness has a limited effect on the number of collisions. Despite the spikiness of the ACPH, it correlates quite well with HRIR (r=0.90), so the number of cyclists out has a significant effect on the number of collisions. This is a stronger correlation than we see on the Monthly Collision Rate graph, so time of day has even less to do with the rate of collisions than time of year. Note that this graph uses a logarithmic (base 2) Y axis.

Collision rate by position of sun Here we see that the collision rate is *much* higher at night that during other times -- and that it's much lower during times of dim light. This was pretty unexpected, given how closely collision rates track ridership in other cases; I didn't expect nighttime to be that much more dangerous. It's probably a bit misleading, due to the presence of the incredibly low dusk and dawn bars; ultimately riding at night is only 1.8 times more dangerous than riding during the day. Dusk and dawn are also pretty interesting. It turns out that dusk is safe because there are huge numbers of cyclists out -- that evening commute is incredibly busy. So even though it has a much higher number of collisions than dawn, it's the safest phase of the day. Dusk and dawn may also be thrown off a bit by the limited amounts of data available; they're each only about 30 minutes long, while the other two phases of the day make up the remainder, so we have much more data about day and night.

Proportion of collisions by position of sun The vast majority of collisions happen during the day, simply because so many riders are out then.

Collisions by position of sun and by month Night Dawn Dusk Day Hours of daylight More collisions happen at night during the winter months, but that's probably because there's just more night. The rate of daytime collisions loosely correlates with the hours of daylight per day (r=0.83), but not as strongly as with the number of cyclists out. In other words, lack of visibility is a factor in the collision rate, but a much greater factor is the simple presence of bikes. This aligns with the earlier finding that riding at night is more dangerous than during the day, but not a whole lot. Drivers are almost as good at not seeing people on bikes in the middle of the day as they are in the darkest night. (Or maybe our transportation systems are designed to make collisions inevitable, regardless of whether it's day or night.)

Collisions by age of cyclist Just so we're clear, that's about 11 children under 11 hit by cars every year in Lincoln alone. I suspect that this may just be a graph of ridership numbers. There's a dropoff at 16, when kids get their driver's license, but it goes back up for the college age crowd; then a slow descent to 40, after which it picks up again, presumably with the MAMIL crowd. But seriously, over 33% of people hit on bikes in Lincoln are children. We only have age data for cyclists injured since 2011, but given that 91.1% of cyclists in collisions are injured, it's still a pretty broad sample.

Collisions by gender of cyclist Men are apparently over-represented, but this again might just be a graph of ridership numbers. This uses LPD's reporting of the gender of injured cyclists; the gender of uninjured cyclists is not reliably recorded. The "Unknown" category represents only those cyclists who were uninjured. LPD uses a purely bivalent gendering system, so trans and nonbinary people may be misgendered in this graph.

Injuries by collision location Interestingly, different locations have different injury rates. Sidewalks have the highest injury rate, but also the most non-disabling injuries. Collisions occurring on the road are the least likely to lead to injuries, but when they do, the injuries are more likely to be disabling. Intersections also show more disabling injuries. Overall, a collision while riding in the street (whether in an intersection or not) is about 2% less likely to result in injury, but about 12% more likely to result in disabling injury. Note that these data are based entirely on accident reports, and so do not take into account cyclists who are hit and later killed; the "Killed" category only represents cyclists who were dead at the scene. Non-injury collisions are the most likely collisions to not be reported, so all of the injury rates are hopefully overestimates. Bike lane collisions are so rare that the data here aren't really reliable. Killed Disabling Visible but not disabling Possible but not visible

Injury severities Overall, while disabling injuries are scary, they represent a fairly small percentage of collisions. That said, it's still worrisome that a cyclist is about as likely to escape a collision unscathed as to suffer disabling injury.

Body region of injuries Predictably, injuries are most common to the extremities, particularly the lower legs.

Hit-and-runs Of the 756 collisions, 119 (15.7%) were categorized as hit-and-runs by LPD. In 9 cases, either both the driver and cyclist left the scene, or it couldn't be easily determined who left the scene. In the remaining 110 cases, the driver fled the scene the vast majority of times -- 94 times, or 85.5% of remaining cases. Overall, in 12.4% of all bike-car collisions the driver fled the scene. If you are hit, make sure to immediately get license plate numbers! I'm not sure what LPD counts as a "hit-and-run"; for example, case no. B6-118033 sure sounds like a hit-and-run to me, but LPD didn't count it as such despite the fact that a driver hit a cyclist, left the scene, wasn't contacted by LPD until the following day, and the officer "explained to [the driver] that if that happens again to just stay at the scene."

Heatmap of Collisions Luckily, there's not too much red on this map. While we're unlikely to avoid any particular intersection based simply on the number of collisions there, we can be more careful at the more dangerous intersections. Some of the most dangerous intersections are at or near bike path underpasses or overpasses, so you can avoid crossing 27th at Vine, for instance, on the Mopac Trail. Some intersections may seem more dangerous solely because they are more heavily trafficked, not because they are actually more dangerous. That said, a few corridors seem particularly dangerous, with 27th street from Vine to Holdrege leading the pack. Capital Parkway/Normal Blvd. is a close second; the intersections with 27th, 33rd, and South are all among the most dangerous in the city. As an aside, can you imagine what an on-street bike lane would do to help out north 27th street and South street between 13th and 27th? Both are struggling but viable commercial districts that are already lined with local businesses, and they have terrific numbers of car-bike collisions.

Map of Collisions All locations should be considered approximations, and there are almost certainly mistakes in places. Crosswalk Sidewalk Road Intersection Bike Lane Elsewhere Unknown This gives us some granular detail to see where collisions are happening around town. For instance, we knew Capitol Parkway and 27th had lots collisions, but here we can see that they were all in crosswalks. In general, collisions increase towards downtown and diminish as you get further out. There are two notable exceptions: The entire length of 84th street (with nearly twice as many crosswalk collisions as 70th and 56th streets); and Pine Lake near 27th. These two areas have something in common: Bike paths adjacent to streets, where cyclists must still cross many side streets. Superior street, which also has a street-adjacent bike path, has no nearby analogue to compare it to, but it has as many crosswalk collisions as Havelock, Fremont, and Holdrege combined. A significant number of crosswalk and sidewalk collisions occur in the downtown exclusion area. Stop riding on the sidewalks downtown already! It's not even safer! (It also could get you a ticket.) Road collisions, predictably, are focused downtown. But at least riding on the streets is legal.

Methodology Accident reports are downloaded automatically from the Lincoln Police Department's website. The reports are parsed automatically and selected data are extracted. The data are available as CSV files at data/db_dump. These reports are then curated by hand to determine where the collision happened, as described above. This depends on the accuracy of the accident report; anecdotally, several cyclists have reported to me that their accident reports were not completely accurate. I found one accident report where the drawing does not look like the actual intersection where the collision happened, so there are definitely at least some inaccuracies. These inaccuracies are not expected to be significant, but there's no obvious way to test this with the given data set. Much of the methodology is described in more detail in the README. All of the code used to generate this report is free and open source under the GPLv2. I have made no effort at all to determine fault, as that process would be fraught beyond any semblance of reason. The Lincoln Police Department deserves special thanks for their open records. In the other two largest cities in Nebraska, the Omaha and Grand Island Police Departments both charge for accident reports ($5 and $6, respectively), and the Omaha Police Department does not allow listing accident reports. Lincoln's open records made this project possible.

Further Study Classifying each collision according to the U.S. DOT classification system would let us compare data with other cities that have published their data, and see how Lincoln stacks up. It could be interesting to look at how tickets are assigned during collisions and see how likely drivers and/or cyclists are likely to be ticketed. Current traffic data is based only on the Rock Island Trail, but the city has other traffic counters and they've generously shared the data with me. I need to find a way to normalize it all to get a general traffic rate for the city based on more than just the Rock Island Trail.