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.
- Bike Trail: 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.
- 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.
- 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.
- 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.
- Road: 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.
- 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.
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 62,161 reports, from January 15, 2008 to January 3, 2019. Since 2012, when LPD
started widespread use of computer-written accident
reports, the dataset includes 54,308 reports, 868 (1.6%) 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
(352/868) 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 54.1% (470) of the collisions
indicate a pattern of riding on sidewalks, while merely
22.0%
(191) 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
Bike Trail Crossing
Intersection
Sidewalk
Bike Trail
Bike Lane
Unknown
Crosswalk
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 while 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 5.8 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=0.84), 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
It's hard to tell how much of the fluctuation here is
just randomness and how much is a trend; the changes are
rapid and fairly small. We had been on an upward trend
from 2012 through 2016, but we only have bike traffic
count data from 2014 onward so it's hard to tell if that
upward trend is due to increased ridership, or to other
factors. There appears to be a downward trend since 2016,
which could be due to projects like the N street bikeway,
or could just be pure chance.
Mean collisions per year is 122.40
and median collisions is 126.
Collisions per month
January 2010, February 2015, January 2017, January 2019, February 2019, March 2019, April 2019, May 2019 are the only months
without a recorded collision. The most dangerous months in
the data set, with 24, are September 2008, September 2011, September 2016.
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=0.89), 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 correlates quite
strongly with the hours of daylight per day (r=0.88), 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 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
8 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 31% of people hit on bikes in Lincoln
are children.
We only have age data for cyclists injured since 2011,
but given that 93.9% 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. These line
up with national figures on ridership; People For Bikes
reports that
about
24% of bike trips nationally are taken by women
(based on a 2009 study). Compared to other small cities,
though, if this is a chart of ridership numbers,
Lincoln seems to have a fairly high number of women
riding.
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 one of the highest injury rates, but
also the most non-disabling injuries. Collisions
occurring in dedicated bike infrastructure -- bike lanes
and trail crossings -- are the least likely to lead to
injuries. Collisions on the road and in intersections
generally show more disabling injuries. (Collisions on
bike trails *away from* intersections also show a lot of
disabling injuries, but that's due to the extremely
small sample size.)
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.
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 868 collisions,
162 (18.7%) were categorized as
hit-and-runs by LPD. In 10 cases, either both the
driver and cyclist left the scene, or it couldn't be
easily determined who left the scene. In the remaining
152 cases, the driver fled the scene the vast
majority of times -- 128
times, or 84.2% of remaining cases.
Overall, in 14.7% 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."
Or, say, B8-038226, in
which the "Driver of V1 did not exit the vehicle and
drove off without stopping," and at the time of the
report the vehicle (and, presumably, its driver) was
still "unknown" -- but somehow was also not a
hit-and-run.
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.
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.
Qualitative Results
A number of interesting trends revealed themselves in
reading through the reports. Most disturbingly, at least
some members of LPD have a significant misunderstanding of
the law, claiming in the
accident
report that
cyclists are required to
walk their bike across intersections, and
in some cases specifically
noting that they
"lectured" the cyclist.
Even after LB716 took effect, cops were still
lecturing cyclists about
walking across
sidewalks, and noting in
reports that the cyclist did
not "dismount from the bicycle". (In
that report, the officer
also coded "No improper driving" for driver error and
"Improper crossing" for cyclist error.) One officer even
cited an unrelated state
statute that he apparently believed prevented cyclists
from riding through an intersection.
In other (fewer) cases,
the officer had a correct
understanding of the law, and
"educate" the cyclist; and
in some cases the driver of the
car was ticketed for
failure to yield to a
cyclist in a crosswalk.
In one case, it's not
clear who knows the law and who doesn't. That case report
notes that "the bicyclist thought all bicycles had the
right of way [when] crossing in the non marked crosswalk
area," and she was correct. It seems that the only reason
the officer would mention that is if they disagreed, which
seems to be the case as there were "no citations
issued."
The law for crossing private drives is less clear, but
at least one cop thinks
that bikes on sidewalks must yield to cars in driveways,
while another one
disagrees. One cop lectured
a cyclist who collided with a car in a private drive,
even though the accident report states that the view was
blocked. I guess you should leave the Oakleys at home and
wear your X-ray glasses.
Some officers may not really have a good understanding of
right of way at all;
in one case, a motorist was
cited for running a stop sign and hitting a child on a
bike, but the cyclist was "educated/lectured about
yielding right of way," which is apparently something that
all cyclists owe all drivers regardless of the actual
rules of the road.
Perhaps ironically, in one
case a cyclist who had dismounted to walk his bike
across the intersection was hit while on foot.
If you get right-hooked by a driver who just can't wait
three seconds for a parking space, you might just
wind up with a ticket
yourself for "violating general traffic laws." Which
laws? Well, general ones, of course.
In some cases, the accident report
mentions that the cyclist
was not wearing a
helmet, despite the fact
that this has absolutely no legal bearing, nor does it
have any bearing on whether or not a cyclist is hit. It's
hard to see this as anything other than editorial
victim-blaming. In another case the fact that cyclists
*were* wearing helmets and
reflective clothing is mentioned; although the
reflective clothing has a clear bearing on their
visibility, helmet use again seems to be mentioned simply
as part of a false correlation with "proper" cycling.
A number of accident
reports reverse the agents, claiming that
a bike hit a car in cases
where the car clearly hit the bike. In some cases, it's
described both ways. It's
not clear why this inversion of causality occurs.
Some people, upon hitting a ten-year-old child, can think
of nothing better to do than
yell at the kid about
right-of-way and drive away.
In some cases, bicyclists
"suddenly appear". Stop
apparating in public or the
muggles will catch on! (Strangely, the accident reports
where cyclists appear show
the bike moving through space normally, not
teleporting.)
The only place two
"elsewhere" collisions
occurred is on 27th street, just south of Highway 2, where
there's a cutout of the median to let train tracks cross
27th street. People are crossing a busy arterial street
with no crosswalk along train tracks -- with a pedestrian
bridge overhead. Please don't be dumb.
Don't ride like a dingus.
Please,
get information from a
driver who
hits
you. Call the police.
File a report.
If you don't have brakes on
your bike, get some. If you can't afford them,
visit the
incredible Lincoln
Bike Kitchen. If you can afford an extra pair,
please donate
to the Bike Kitchen.
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.
Media coverage of this report
Local Bicycling Organizations
Please consider joining or supporting one or more of
these organizations to help advance bicycle advocacy.