Quantifying Transit Reliability using Accessibility Indicators

Anson Stewart
Published in
3 min readOct 26, 2018


Bunched buses stuck in traffic, as seen from Conveyal’s Boston office

The purpose of public transport is to provide access to opportunities, such as destinations where employment, health care, and other services are located. Accessibility indicators combine transport network and spatial opportunity data, making them holistic characterizations of metropolitan regions that are appealing from a policy perspective and a focus of Conveyal’s work.

Accessibility-based characterizations of public transport, such as the number of jobs reachable within 30 minutes from a given location, have traditionally been calculated using deterministic, scheduled travel times between origins and destinations. This approach is suitable for long-range scenario planning, but it typically does not incorporate how accessibility may be affected by operational unreliability (e.g. varying travel times attributable to traffic, bus bunching, etc.).

Recent research, using data sources such as roadway sensor or automatic vehicle location (AVL) systems, has begun to explore the impact of using actual travel times to compute accessibility indicators. For example, Cui and Levinson (2016) assess the variability of accessibility to jobs by car in the Twin Cities. In a transit-focused example, Wessel, Allen, and Farber (2017) create a “routable retrospective transit timetable” from archived AVL data and use it to compute the percent difference between actual and scheduled accessibility to jobs in Toronto.

Conveyal Analysis is built on a probabilistic approach to accessibility, making it a great tool for calculating similar accessibility-based reliability indicators. As part of my dissertation research, I created an experimental retrospective real-time GTFS feed based on stop arrival and departure times for MBTA service in Greater Boston including:

  • Green Line light rail (see Fabian (2017))
  • Red, Orange, and Blue Line subways, using track-circuit train-location data
  • Bus routes, using a stop-events ODX input table

Loading this GTFS feed, representing two weeks of service in 2016, into Conveyal Analysis allows for some interesting comparisons. In the visualization below, each small dot on the map is one worker’s home location — a potential employee. According to scheduled weekday service, transit should provide employers in Copley Square with access to 413,000 potential employees within 45 minutes, corresponding to the reachable area (isochrone) shown in red.

But because of disruptions, slow speeds, uneven headways, and dropped trips, Copley Square actually had access to many fewer employees within 45 minutes. This unreliability is depicted with the fluctuating blue bar and isochrones in the visualization below. On one particularly bad weekday in the sample, these employers would have access to approximately 43% fewer potential employees!

Area and corresponding number of workers’ homes reachable by transit from Copley Square within 45 minutes, departing between 5 and 6 PM, according to MBTA schedules (red) and vehicle location data from 10 weekdays (blue).

Similar calculations can be repeated for every origin in the region. In the example below, the number of average number of jobs reachable in 30 minutes based on actual service over 10 weekday mornings was compared to the number reachable based on scheduled service. If you live in one of the red areas below, you might have high job access on paper, but have 40% fewer jobs reachable within 30 minutes of actual commuting time (including unreliability), at least on the days in this sample.

Percent difference in number of jobs reachable within 30 minutes, comparing scheduled and average actual accessibility over 10 weekdays

This case study was the basis of a presentation in Sydney at the 7th International Symposium on Transport Network Reliability. The presentation and follow-on discussion covered a number of ideas for refining the analysis, including how such indicators might be correlated with other measures of interest such as unemployment and missed healthcare appointments. One important caveat is these indicators assume travelers have perfect foresight about the network and adjust their itineraries to avoid disruptions.

As more agencies publish GTFS-RT data, we’re considering building pipelines to bring archived real-time data directly into Conveyal Analysis to calculate similar accessibility-based reliability indicators. If you’d like to collaborate on this work, please let us know.



Anson Stewart

Analysis and Research, @conveyal | PhD in Transportation, @MIT | '10 TJ Watson Fellow + @SwatAlum | Californian in exile on East Coast