One-ways, Open Jaws, Closed Loops, and more: A day of Uber journeys in Chicago
A while back, we noticed something interesting in Uber’s data: a lot of trips don’t connect to other trips. In other words, a rider will use Uber to go somewhere, but they don’t return to their origin (or continue onward to their next destination) with Uber. This caught our attention because such journeys are typically not possible with a personal automobile: if you drive your own car somewhere, you almost always have to drive it back.
We’ve long suspected that these “one-way trips” are parts of multimodal journeys — in other words, riders are combining Uber with other modes of transportation to suit their needs, schedules, and budgets. For example, a rail commuter who works late one night might opt to take Uber home when the trains are running less often, or someone might use Uber to meet up with friends and then check out a bikeshare to ride to the park afterwards. We’ve already seen plenty of evidence that riders combine their Uber trips with public transit, and one-way trips might provide yet another measurable view into multimodality.
To determine how often this happens, we developed a simple algorithm to automatically sift through Uber’s data and count one-way trips. It turns out that a substantial number indeed fit this pattern: 20–30% of trips in most US cities appear to have no obvious connection on either end. The diagram below illustrates the definition we currently use, along with one-way trip percentages in the 5 largest US cities:
Put simply, we define a one-way trip as a trip that’s isolated in either space or time from the same user’s other Uber trips. However, this method has a few drawbacks. First, to ensure we aren’t overstating the number of one-way trips, we made the algorithm highly selective: for instance, it doesn’t count any trip with a subsequent onward journey (even if that onward journey doesn’t return to the original departure point). Second, and more importantly, travel patterns are complex; by trying to reduce multimodal behavior to a single number, we’re almost certainly missing some rich insights in the data.
For a more detailed picture, we went back to the most basic data science technique: actually looking at the raw data. We picked 100 anonymous riders at random who took a ride in Chicago during a week in February, and tried to categorize and count the types of trips they took on a single day. To protect user privacy, and consistent with our privacy policy, neither the riders’ identities nor their specific trip origins and destinations were recorded (except when a trip went to or from Midway or O’Hare, in which case it was counted as an “airport trip”).
As we looked at the data, a striking pattern quickly emerged: despite the universe of possible human travel behavior, the vast majority of Uber trips seemed to fit into five fairly simple categories. The following are the categories we found; note that we’ve only counted a trip as “connecting” to another if there was a pickup-dropoff pair within 1/2 mile and with a time difference less than 24 hours:
- One way: Trip from point A to point B, but no trip back to A or connecting trip onward from B.
- Round trip: Trip from A to B, and another from B back to A.
- End-on-end: Trip from A to B, and then onward (in about the same direction) from B to C.
- Open jaw: Trip from A to B, and another (in approximately the reverse direction) from B to C, where A and C are more than 1/2 mile apart. Alternatively, trip from A to B, followed by another from C back to A (with B and C more than 1/2 mile apart).
- Loop: 3 or more trips connecting to each other, where the user ends up back at their starting point.
Finally, a handful of trips were either to or from airports (technically one-way trips, but less interesting from a multimodality perspective) or didn’t seem to fit in any of these classifications. Diagrams of these journey types, and the percentage in each category, are shown in the figure below.
The number of one-way trips counted manually (34%) is somewhat higher than the automatically-measured number (26%) the same month. Some of this difference is likely because the automatic method ignores any trip that connects to another — so if someone takes Uber from their home to work and back, followed by a one-way trip from home to a restaurant the same day, this last trip wouldn’t be counted as one-way because the algorithm “sees” it as a connection from the previous trip. These differences aside, this exercise revealed a variety of more complex journey types (like end-on-end and open-jaw trips) that require other forms of transportation to fill in the gaps, and therefore may also signify multimodal journeys. Leaving out the “airport and other” category, over half (56%) of Uber trips appear to reflect a combination of Uber with other transportation modes — about twice as many as the number we’ve measured algorithmically.
Now, a few important caveats: the taxonomy here is somewhat subjective, and a different person looking at these trips would likely disagree on how some trips should be classified. The sample size is also small compared to the billions of trips Uber riders have taken, so it’s more subject to variance than the algorithmic measurement. And when drawing conclusions about multimodal behavior, it’s important to note that we don’t actually know what users are doing when they aren’t taking Uber: it’s entirely possible that on some of these “one-way trips,” riders used taxi, Lyft, or a ride from a friend for the other leg. Conversely, a round-trip Uber ride connecting to a Metra station would in fact be part of a multimodal journey, but wouldn’t be counted as such in the above classification.
Whether measured automatically or manually, our data show that Uber by itself does not substitute for private car ownership — if it did, a lot more of our users’ journeys would be roundtrips or loops. But in urban environments with walkable, bikeable neighborhoods and healthy public transit infrastructure, the growth of shared mobility can enable residents to leave their cars at home (or give up car ownership altogether) and choose the transportation options that fit their needs at any time of day or night.
