Those E-Scooters Might Not Be as Dangerous as You Think
A civil engineer looks at traffic data to improve road safety
E-scooters continue to intrigue us. They’re new and unfamiliar, and they’re also everywhere. Perhaps this explains the sensationalized media coverage on e-scooters, much of which is driven by anecdotes of accidents. But an army of researchers has been itching to unveil empirical evidence to augment the e-scooter dialogue.
I am a soldier in this army.
At the center of the e-scooter controversy is safety. Let’s look at some recent headlines:
- “8 Deaths Now Tied to E-Scooters”
- “Electric scooter injuries jumped 222% over the past four years”
- “Electric scooters were to blame for at least 1,500 injuries and deaths in the U.S. last year”
Now, those headlines have to instill some fear. But road safety is not a simple matter that can be summarized in a headline. This is especially true for e-scooters because we don’t know much about them yet.
I started my PhD journey with a mission of answering the question “Are e-scooters safe?” Fast forward a year, and I am now convinced that the more appropriate question may be “Are our streets safe?” Let me tell you why.
The little we know about e-scooters
Katie Harmon and Laura Sandt at the University of North Carolina Highway Safety Research Center have been tracking fatal crashes involving e-scooters in the U.S. since late 2018. Of the 18 fatal crashes in 2019, 16 of them involved e-scooter riders killed from being struck by motor vehicles. One e-scooter fatality was a result of an e-scooter rider colliding with another rider. The remaining fatality resulted from a single-vehicle event where the e-scooter rider crashed into a tree. From these statistics, it’s difficult to decipher how much to attribute these terrible events to e-scooters and their riders versus how much to attribute to the general lack of safety for vulnerable road users.
The universal metric for road safety has been the number of crashes, which requires a three- to five-year observation period. Statistically speaking, crashes are rare and random, and with shorter observation periods, all data is zero-inflated. Even “dangerous” intersections may only accumulate several crashes over five years. This makes it difficult to get any significant results in crash modeling for engineers, and, unfortunately, modern e-scooters haven’t been in wide use for that long. Through the lens of traditional road safety engineering, we are only seeing a few data points in a comprehensive e-scooter safety story that has yet to be written, let alone be told.
We don’t have to wait for more e-scooter crashes
Road safety is a topic that is near and dear to my heart. I started my career in technical assistance and monitoring and evaluation of road safety in emerging economies, which put me knee-deep in the problem of crash data scarcity. This is when I grew fond of surrogate measures of safety, which are objective indicators that either resemble crashes or have strong relationships with crashes. A common one is “traffic conflicts,” which a layperson might refer to as “close calls.”
In our everyday lives, we don’t think much about almost getting into an accident. But for safety researchers, this data can be a gold mine. Christer Hyden, known as the godfather of traffic conflicts, introduced the safety pyramid below to illustrate the continuum of traffic event types. The idea is to study severe traffic conflicts as they greatly resemble actual crashes and occur more frequently. This way, we overcome the paradox of road safety engineers having to wait for crashes to occur in order to understand how to prevent them.
The beginnings of surrogate safety research involved humans physically observing and recording indicators at sites for days or weeks. You can imagine why this was not scalable. With recent advances in computer vision and machine learning, however, we are able to automate this process. We can feed traffic video data into a machine-learning algorithm and automatically detect and classify road users and calculate indicators of interest like spatial and/or temporal proximity indicators such as speed, traffic conflicts, and evasive maneuvers. Video data may be available through traffic cameras, or researchers could collect their own data. Several years ago, I wandered the streets of Accra and Bogota to install and babysit my GoPros.
The research presented hereafter is a lightly edited section of my PhD dissertation at McGill University with guidance from my PhD adviser, Luis Miranda-Moreno. Preliminary findings of this research were presented at the 2020 Transportation Research Board annual meeting.
A proactive approach to e-scooter safety
Last May, I took to the streets of Washington, D.C., where I currently call home, to understand e-scooter safety. I chose the intersection of 11th Street and R Street NW for its unique geometry that combines shared and bike lanes and has high volumes of bikes, e-bikes, and e-scooters. I collected 196 hours of video data through the month.
I observed 110,582 cars, 19,802 pedestrians, 8,639 pedal-only bikes and e-bikes, and 1,960 electric and non-electric scooters. It’s important to note that the video analytics I used was great, but it wasn’t magic. I was unable to differentiate whether the bikes and scooters were electric or solely human-powered nor whether they were shared or privately owned. Also, the algorithm didn’t know what electric standing scooters looked like. A limited army of research assistants and I manually validated all scooters, most of which were originally misclassified as funny pedestrians and bikes.
How “dangerous” are e-scooters?
I don’t have a global answer. What I can say though is, at this time, my findings do not support the hypothesis that e-scooters are more dangerous than bikes. I used post-encroachment time (PET), a popular measure of conflict severity. It indicates the extent to which two road users missed each other in space and time. The lower the PET value, the more serious the conflict.
The threshold for PET can be arbitrary. Usually, a PET that is less than a second and half or two seconds is considered a serious conflict. I observed 27 conflicts involving scooters with PET of less than two seconds and 181 conflicts for bikes with the same PET. When normalized by the observed road user volumes, I found that 1.4% of scooterists were involved in serious conflicts with a PET of fewer than two seconds. Rates for pedestrians were a bit lower at 1.1%, and cyclists were higher at 2.1%. In the cumulative density plot below, we see the line for scooters growing slower than bikes at PET less than approximately 2.8 seconds. In other words, a greater proportion of observed conflicts were severe for bikes than for scooters.
Some of the most dangerous conflicts occurred when westbound cyclists ran red lights. Here is one example:
And another almost identical scenario:
I am in the process of manually validating the propensity of red-light violations for cyclists versus scooterists. Preliminary counts indicate that cyclists are much more likely to run red lights than people riding scooters. Some cyclists have rationalized this behavior with physical science — cyclists must generate their own kinetic energy when they come to a halt. This rationale may not resonate as much for e-scooterists, thanks to the throttle.
Should we be designing streets that make it easier for cyclists? Or should we discourage red-light violations by fierce enforcement?
Awkward left turns for scooterists
Both cyclists and scooterists struggled to turn left. The lack of appropriate infrastructure appeared to be the primary issue.
Even though scootering and cycling on sidewalks were allowed at this location, northbound scooterists frequently traveled on the shoulder of the car lane (which is technically a shared lane). To make a left, northbound e-scooters have two options: 1) move into the left lane and make a standard left turn like cars or 2) make a two-stage left turn, also known as a Copenhagen left turn. The scooterists in the images below attempt a two-stage left turn by scootering through the intersection, then idling until the cars pass, then scootering across at a red light. From my personal experience, it is not easy to make a standard left on a scooter (especially one that you don’t own) because scooter-car speed differentials are too large, and car drivers aren’t so patient.
A two-step turn queue box (with a “no turn on red” sign for westbound traffic) could facilitate a safe queuing area for left-turning, northbound scooterists and cyclists.
Cyclists struggle to make lefts too
The vast majority of westbound cyclists traveled on the bike lane located on the right side of the car lane. Cyclists were often found crossing over westbound through cars to achieve their left turns. Here is one example:
And another:
These clips make the civil engineer in me cringe. One potential solution is to add a bike box to increase visibility of cyclists and scooterists. A bike box that extends across all travel lanes could facilitate safe left-turn positioning for westbound cyclists and scooterists (examples 4 and 5). These are some initial street design changes that come to mind.
We are still learning about e-scooters
We are in the very early stages of road safety research on e-scooters. What I have learned so far is limited to my study site. This study suggests that e-scooters share similar vulnerabilities as cyclists, but they’re not necessarily more dangerous.
The scooter boom has allowed a rare alignment of interests to improve road safety for all vulnerable road users. Let’s recall the Three E’s of Traffic Safety: engineering, enforcement, and education. My background limits me to contributing to the first E, engineering.
So, I humbly invite my mentors and fellow and future engineers and planners: Let’s continue to work together to do our part in building an environment that is kind and forgiving to all vulnerable road users — and that includes e-scooters and their future cousins. And let’s all remember that traffic crashes are not “accidents” but preventable incidents.
Acknowledgments
Huge thanks to my advisor, Luis Miranda-Moreno for his unlimited support on my research endeavors and research assistants at IMaTS at McGill who spent evenings and weekends manually validating scooters with me — Bismarck, Paula, and Yousteena. Thanks to Brisk Synergies, especially Lana for coordinating the automated analysis. Thanks a million, Ten, for your efforts in data collection.
Disclaimer: The views and opinions expressed in this article are solely mine and do not reflect those of any organizations that I have been or am currently affiliated with.
