How Replaceable are Seattle's Transit Routes?

Ridership-based route productivity measurements—those which compare a route’s average number of riders to its cost of operation—are common within transit planning. Unproductive routes will be the first considered for elimination when funding is tight. Those that perform well may merit additional service when the budget supports expansion. I have previously argued that ridership-based measurements are, at best, incomplete for these purposes. There are factors beyond ridership that determine how much impact the elimination of, or investment in, a route will have. Transit routes have varying amounts of capability, regardless of how many people choose to take advantage of it. The transit agencies that I’m familiar with entirely neglect this factor, though, perhaps because they lack a way to quantify it. Measuring access offers a solution. While this style of analysis has become increasingly common at the network level, using it to compare the contributions of routes within a network has not.

In early 2024, I began a series of posts looking at route-level measurements of access in an attempt to quantify this capability. I defined five measurements that could serve as access-based assessments of routes. I then selected thematically similar King County Metro bus routes and compared their performance in Seattle. Ultimately, I ended the series prematurely due to a computation error that resulted in invalid results for a small number of routes. I was also unhappy with the cost of computation, and dissatisfied with the format of the series, which made it difficult to determine where routes analyzed earliest in the series fit into the overall ranking.

I revisited one of the measurements in this project at the end of 2024, when I published a revised ranking of transit routes in Seattle by journeys per in-service second. This route-based access measurement counts the appearance of each route in the fastest path for each completed journey—a combination of origin sector, destination sector, and time of day that constitute a trip that can be made in at most 30-minutes using transit and walking—and divides this count by the number of seconds that every transit vehicle operating the route is in service. Routes that rank highly in this measurement are those which make the most of their service investment by getting riders to a broad variety of destinations at many times of day. Routes that rank low struggle to create access to opportunities beyond what walking and other routes already accomplish.

For the goal of making less harmful deletion decisions by understanding the unique capability that each route adds to the network, journeys per-service second provides some insight, but is insufficient. A route that appears most often in the fastest paths seems like it should have the biggest impact when deleted, but there are circumstances where this won’t be the case. A route may be part of the fastest path for many journeys, but, in the absolute worse case, it may only represent a second of savings in every single one of them. What is a superlative route from a journeys per in-service second standpoint could in fact be eliminated without any meaningful impact on riders. On the other hand, a route that has a geographically constrained walkshed could look poor in terms of journeys per in-service second, but eliminating it could leave its riders with no viable alternative. The only way to assess the impact of deleting a route is to delete it and see how the rest of the network adapts. For a given route r, quantifying the impact starts with two numbers that can be extracted from the access analysis that yielded the journeys per in-service second. They are the overall number of completed journeys, J, and the number of journeys in which r is part of the fastest path, J[r]. Recomputing the access with r deleted from the network yields a new total number of completed journeys, K. The number of lost journeys L is J - K. The percent replaceable for r—the percentage of completed journeys in which r formed part of the fastest path, but is not in fact critical for making the trip within the time budget—is thus 100 * (1 - (L/J[r])). Lower numbers indicate a route that is harder to replace, and thus more valuable.

Computing percent replaceable gets expensive. It requires a new access computation for each route. I’ve managed to reduce the full-day calculation of 30-minute access in Seattle to a few minutes of computation, costing a few dollars of AWS Lambda runtime, but there are 91 transit routes that stop at least twice in Seattle, and appear in the fastest path of at least one completed journey. For a transit agency or grant-funded academic project, this would not be an unreasonable expense; for an independent researcher who has yet to find paying customers for this service, it’s more than I’d like to spend.

I’ve reduced the cost somewhat by recently developing a means to “patch” existing access calculations with a route deletion. Since computing the journeys per in-service second already requires keeping track of which routes comprise the fastest path for each journey, the system can use this information to avoid recomputing the access centered at sectors where the route was never part of a fastest path. While this is helpful, the savings is not that large in practice. Transit routes are surprisingly broad in their geographic impact (so much so that I’m planning a follow-up post to discuss this), and the sectors that need to be recomputed tend to me the most expensive ones to compute. The table below is the result of these calculations, for transit service within Seattle only, on a typical weekday, and using a 30-minute time budget.

I’ll be highlighting individual routes in subsequent posts, but I want to make some preliminary comments on the range of percent replaceable values. It is quite large, spanning from 31.9% to 93.3%. But the routes at these endpoints are ones that few regular transit riders in Seattle will ever take. They are both “Custom Bus” routes that King County Metro operates on the behalf of Seattle-area private schools. Their intended riders are students who are sold passes, though, ostensibly the service is open to the public. Route 981 makes two stops in Seattle on its single trip before getting on the highway. Three minutes are allocated to connecting two locations that are approximately five minutes apart when walking. It’s not surprising that there are few journeys where those two minutes of savings are critical to remaining within the time budget. On the other end of the spectrum, route 986’s routing inadvertently provides unique express service between various points in Northeast Seattle, including a valuable transfer point at the University of Washington Link Station. To consider these routes’ scores as the true extent of the percent replaceable is somewhat disingenuous, though. They are technically fixed-route public transit service, but function as something else in practice.

Excluding the Custom Bus routes, the high end of the replaceability spectrum becomes route 193’s 92.2%. This route provides peak service between Federal Way and First Hill. The service pattern within the latter—consisting of a looping path with several stops—is what is getting measured in this analysis. The intention of this service isn’t to enable trips within Seattle, though, but to give those working at various points in First Hill a one seat ride to suburban Park and Rides, so, again, using this route’s score as an endpoint for the measurement feels somewhat strained. Many other commute-oriented routes serving downtown and the University District have similar properties, and consequently a high percent replaceable is well.

When considering only routes that primarily serve destinations within Seattle, the South Lake Union Streetcar represents the upper end of percent replaceable at 84.4%. Transit mode itself doesn’t impact this measurement (unless different modes always imply different scheduling or stop spacing choices), but if choosing to ignore the streetcars, route 10 is worst at 74.7%. On the other end, the least replaceable route outside of Custom Bus service is the Link 1 Line at 33.1%. When considering only Metro’s bus routes, route 75 is not far behind at 34.5%.

Why is it important to know that there is a wide range in percent replaceable among King County Metro’s bus routes that serve Seattle? As part of its System Evaluation process, Metro uses a set of ridership-based productivity measurements to rank its routes. To make decisions solely on the basis of this, routes would have to be identical in the ways that ridership doesn’t measure. But Metro knows that this isn’t true, so it supplements each route’s productivity measurements with an equity score to disfavor reductions that impact transit-dependent populations. Targeting routes with low ridership, after all, may confine any impact to the fewest individuals, but says nothing about how much the actual impact will hurt each individual. Supplementing ridership with need still doesn’t account for every way in which the consequences of deleting a route differ, though. Routes exist in the context of a network; other routes in the network can partially mitigate the loss of a deleted route. Metro’s System Evaluation process does not account for this at all; neither ridership nor equity measurements capture it. Therefore, Metro must be assuming the network’s ability to compensate for a lost route is either immeasurable, or roughly identical across routes. The wide range of percent replaceable among Metro’s routes indicates that this just isn’t true, and it is signaling a missing component in Metro’s System Evaluation process. If Metro—or the many other transit agencies that evaluate their networks in largely the same way—is willing to introduce access measurement into this process, that omission can now be corrected.