Scooters have rapidly emerged as a part of Paris’ transportation network, with modal share suggesting rates of initial growth twice that experienced by the Velib’ system in its first year. This has come with some teething pains, with the city of Paris exploring how best to integrate the potential micromobility presents into its streets. The city of Paris has clearly defined goals for reducing clutter and promoting proper scooter parking. The first is to ensure that riders don’t park on the city’s narrow sidewalks. The second is that riders end their rides in “preferred parking” spots: new on-street corrals the city is building especially for scooters, as well as existing parking locations traditionally used by two-wheelers (that is, mopeds and motorcycles).
To help the city achieve those goals, Bird introduced a parking “preferred product” which uses visual reference points in the app, real-time navigation, and GPS enabled alerts to make it easy for riders to end their ride within an approved parking space. To encourage riders to park correctly, those who parked in the designated zones were rewarded with a discount on their next ride. Going forward, the city is shifting to “mandatory parking”, and Bird will introduce graduated punishments for parking inconsistent with the city’s requirements.
We’ve spent some time taking stock of our riders’ parking behavior in the city and the ways in which we can improve compliance with city rules. Using two weeks of anonymized ridership data from August 2019, as well as information made available by the city via Paris Open Data and Open Street Map, we conducted a deep dive into rider parking activities, and how Bird’s new product impacted behavior.
How did we do it?
Bird’s preferred parking spots consist of locations provided to us by the city of Paris, as well as other spots identified by our team which adhere to the city’s parking rules. By mapping these preferred parking spots onto the Paris landscape, we are able to see how they fit into the urban design of the city.
The map below shows the footprint of the buildings (such as the Palais Garnier) in the Opéra neighborhood of the 9th arrondissement. Bird preferred parking spots as of August 2019 are shown in black; for this example, we’ll focus on the starred parking spot at the center.
For each preferred parking spot, we can use a Voronoi partition to define a corresponding “catchment area” — the region of the city for which that parking spot is the closest spot. This allows us to explore the density of parking spots across the city.
Then, for each preferred parking spot, we can examine the rides that fall within its catchment area; an illustrative example (with hypothetical data) is shown below. We can calculate the percent of rides in the catchment area that did or did not end at a Bird preferred parking spot. Scooter location reports are somewhat noisy, due to limitations on the accuracy of GPS positioning; we estimate that they are within 10-12 meters of the scooter’s actual location. Therefore, we draw a radius around each parking spot in order to determine whether preferred parking was likely used.
We can also add features to each parking spot, so as to determine whether and how such features predict the use of the parking locations. From Open Street Map, we are able to add information on the number of nearby points of interest, such as cafes, restaurants, schools, and libraries. We also collected information on each spot’s street, such as whether it is a one-way street or a main road. Using data from the city of Paris, we were also able to add information on how far the parking spot was from the nearest Velib and metro stations. We also calculated whether the preferred parking spot fell within one of the officially designated parking spots of Paris; if so, we also added features like the type of vehicles allowed in the parking spot, its surface area, and its design (e.g. whether it has parallel, perpendicular, or angled parking).
Finally, we tested features about the (anonymized) user and ride stored in our own system, such as whether the user typically rides on weekends, whether the user is a high-frequency rider, the version of the Bird app used, and the duration of the trip.
At the ride level, we fit regression models to predict whether or not a user parked in a preferred parking spot according to these different sets of features. We control for the hour of day, day of the week, and the geohash in which a ride ends, in order to account for the natural variation in the data over space and time.
So what did we learn?
Our early results suggest that…
Preferred parking use is correlated with both the characteristics of the parking spot and the trip.
- Rides that start in a preferred parking spot are 7.9% more likely to end in a preferred spot. It’s possible that seeing a scooter parked in a preferred location reminds a rider that they should also use preferred parking.
- The use of preferred parking is lower for longer trips. It’s possible that riders are less likely to follow the rules when they are tired.
- Unsurprisingly, the use of preferred parking is lower where parking spots are sparse. People park in preferred spots more often when spots are closer to each other, and when they have smaller catchment areas.
- As parking spot capacity (the maximum number of vehicles per spot) increases, preferred parking use goes up. For every additional scooter that a spot can accommodate, we estimate that preferred parking rates in its catchment area increase by 0.3%.
- The use of preferred parking is higher when the parking spots are near more bicycle and motorcycle parking spots. It is possible that having many two-wheel spots nearby acts as a visual reminder to use preferred parking.
- The use of preferred parking is higher as the number of nearby schools increases. For each additional school within 200m of a parking spot, we estimate a 0.5% increase in the use of preferred parking. It looks like Parisians are setting a good example for their kids!
What’s next?
Here at Bird we’ve borrowed the idea from Lessig (2006) that user behavior is governed by four broad forces: laws, markets, architecture or code, and norms. Paris has introduced law in the form of parking rules, and we’ve tested market forces by offering small incentives to users who park in preferred spots. By introducing a parking map in our app (and playing with some of the design features of that map and those spaces), and by studying the placement of parking spaces, we’ve explored some possible architectural solutions to encourage preferred parking use. A next step is to think about how we can change the norms around parking.
Our first campaign was the installation in July of 450 strategically placed billboards around the city to remind riders to park their scooters off the sidewalk and in the places set out by the city. We are excited about the potential of also using unobtrusive in-app messages to try and remind users to park in preferred spots. For example, we might want to encourage users to do the right thing and “return to the nest” by bringing their Bird back to a preferred spot. Or, we might ask them to copy others and “follow the flock” by using preferred parking. We hope that these small nudges will help educate users about parking, teaching riders about what they should do, and about what many other riders already do. We have also reminded riders in app and via email to park in designated spaces, off the sidewalk.
All new technologies undergo a period of experimentation and learning. Scooters provided by companies like Bird are no different. By studying how to build better parking spaces and how to encourage riders to use them, we reinforce our commitment to the safe use of our scooters, and to educating riders to use Bird in a way that can benefit them and their communities.
This post was written by Katherine Hoffmann Pham, a doctoral candidate in Information Systems at NYU Stern, and Laurence Wilse-Samson, senior manager for Policy Research at Bird.
- The US Government GPS standard implies that in best-case scenarios the telemetry provided is accurate within a 5m radius, and this accuracy declines around buildings and in congested urban areas.
- We tested separate models for trip duration, a parking spot’s distance to its nearest neighbor, the size of the parking spot’s catchment area, and the spot’s capacity. While these factors are correlated, our findings are consistent when we also test a single model pooling all of these factors, suggesting that each is an important predictor of preferred parking use.
