This article describes the process and methodology used to infer origin, destination, and transfer patterns using TriMet's internal datasets. The output generated through this methodology can be utilized to examine how the network is used over time.
The ODX data covers approximately 30% of TriMet rides taken by passengers who pay and tap via Hop: The Hop taps represents approximately 45% of unlinked trips on TriMet bus and Max. Of these, the ODX Model can infer a destination for 65% of those trips, equating to 30%.
Base Datasets
The Origin-Destination Transfer inference algorithm (ODX Methodology) depends on the following datasets, all sourced from TriMet's enterprise data warehouse:
HOP transaction journal: A journal of all the boarding taps made by riders using any fare payment method besides cash. Included are the date, time, stop ID, line ID, and anonymized card ID that allows different transactions to be tied to an anonymized individual rider.
Automated Vehicle Location (AVL): The observed stop times and stop IDs across all times of in-service TriMet vehicles. Covers both buses and light rail. A trip ID associates a set of stop times and IDs across a service date.
Static GTFS data: Attributes fields like the operating line ID to the AVL data.
Validation
A number of different methods were used to validate the results of the ODX inference:
Comparison with a human expert: Approximately 1000 randomly sampled journeys were presented to a human validator and achieved a 95% success rate.
Comparison with quarterly APC (Automated Passenger Counters): When accounting for the coverage of the HOP transaction journal and the inference success rate, quarterly APCs were within 4% of the inferred ODX counts.
Evaluation of synthetic journeys: The ODX methodology was applied to a set of boarding taps associated with 210 synthetic journeys. 85% of the inferred alighting locations were correctly inferred.
Main Methodology Assumptions
There are three main assumptions made in the ODX methodology. These are important to note.
First, the ODX methodology assumes that a rider's alighting location is close to their following boarding location. In the below figure, the dark gray circles indicate observed boarding events drawn from the HOP transaction journal. The light gray circles indicate inferred alightings, the target outcome of this process.
Second, the ODX methodology assumes that if a rider is inferred to have missed several boarding opportunities at a stop, then the location of their inferred alighting was, in fact, the end of their trip. In this case, the alighting is tagged as a destination, and the subsequent boarding is classified as an origin. All events (whether boarding or alighting) between an origin boarding or a destination alighting are tagged as mid-journey. The threshold for determining if an alighting is a destination depends on the headway observed at the next boarding stop/line pair. The threshold value follows a sigmoidal relationship bounded between one and three and centered on fifteen minutes. For low headways, riders must have above three boarding opportunities before alighting is tagged as a destination. Inversely, for high headways, a rider must take advantage of only one boarding opportunity for the alighting to be considered a destination.
Third, to select an inferred alighting location from the suite of possible candidates, a rider is assumed to minimize the time to the following boarding location. This time is composed of the ride time and the transfer time. The ride time is the time spent riding the transit vehicle to the alighting location, while the transfer time is the time taken to walk from the alighting location to the following boarding location. A walking speed of four feet per second estimates the transfer time. The below figure shows the ride times and transfer times given a candidate's alighting location.
Additional Notes on Methodology
The HOP Transaction Journal does not cover all rider events. For example, the transaction journal does not include riders who do not pay or pay through cash. The HOP transaction journal covers approximately 45% of riders.
Only journeys that exclusively use TriMet infrastructure are included, i.e., Portland Streetcar and C-TRAN are not included.
Journeys that end within 200 feet of where they began are considered loops and are removed from the dataset.
Journeys, where the alighting location is based on a boarding tap more than seven days in the future, are removed from the dataset.
The inference process successfully infers alightings for approximately 65% of boardings (including the removals noted in the previous two points).
Considering the coverage of the HOP transaction journal and the inference success rate, the ODX data cover approximately 45% of TriMet’s ridership.
The unique identifier per card is rotated once a month, meaning it is impossible to follow a card’s movement patterns across months. This causes the number of successful inferences to dip around the beginning/end of each month.
Have Questions?
In-app support is the best way to get help or answer questions regarding anything in the UrbanLogiq Platform (even questions regarding methodology). First, you can click on the chat bubble icon at the bottom left corner of the sidebar. Then you can start a new conversation in the chat box pop-up.
Furthermore, this is the best way to provide product feedback. We thrive and love hearing how our platform can be better. Giving us product feedback in the in-app chat requests is gathered and prioritized. This is the best way to give feedback. We love hearing feedback from our customers and love answering questions!
You can also call us (1-833-872-2647) or email us (customersuccess@urbanlogiq.com)
Happy Exploring!