The Subway’s Broken Schedules

Every now and then, someone on Twitter posts an image of a subway countdown clock showing an extraordinarily long wait for the next train. Sometimes, that headway is the hard reality of our transit system, beset as it is by crew shortages, long off-peak headways, and disruptive maintenance. But sometimes, that seemingly distant train is actually a lot closer than the clock says. Why, you ask? Because over the last four years, New York City Transit has been slowly and subtly changing the way it handles the impacts of maintenance on subway service. This nearly invisible process has not only degraded the accuracy of countdown clocks, but also has contributed to increasing operating costs, falling service reliability, and an erosion of scheduling’s power as an organizing tool for transit.

An R train enters 9th Avenue during a weekend diversion in 2019. Image credit: Uday Schultz

Backstory

As transit riders, we often come to think of schedules as being a guide to our local transit network, or perhaps as a yardstick by which to measure its performance. But schedules are much more than that. Fundamentally, they’re the organizational system of transit networks — they make sure that vehicles, crews, and support staff are where they need to be when they need to be there. The New York subway runs on two sets of schedules. One is the base schedule, which lays out how a normal day’s service should work, and which establishes regular jobs for the system’s crews. The other is what’s called a supplement schedule. Supplements modify the base schedule to account for transient changes to service. These can be anything from minor schedule adjustments for ‘leaf season‘ on the Q line to the complete shutdown of a Manhattan trunk over multiple weekends. Critically, supplements are not blank slate rewrites of the base schedule for their applicable time period. They must dovetail with base service at each end. They also need to respect the base’s crewing fabric. Though supplements can change the movement of trains at will, they must pay regularly assigned crews for the travel time between their normal and supplement-schedule start locations, and must pay overtime for any shift backwards from normal start times. Any additional crews required then have to be woven into this highly constrained fabric, adding an additional layer of complexity to the process. Supplements, in other words, are organizational capacity saps: not only do they require a considerable commitment of scheduling resources to write, but they also must converse with a much broader infrastructure responsible for coordinating track access and diversions to become useful planning implements.

The supplement creation workflow. It’s complicated!
Acronym translations: MOW is maintenance of way, CPM is capital program management, RTO is rapid transit operations, and Division C is the department responsible for coordinating and operating work trains. Image credit (pdf download)

Given how resource-intensive the creation and operation of supplement schedules tends to be, one might imagine that they are used sparingly, acting as an institutional incentive for diversion planners to minimize the number of outages, maximize repeatability, and strive for precise estimates of service impacts. In reality, this could not be further from the truth. Responding to the subway’s 2016-2018 service crisis, agency leaders and politicians developed a series of initiatives to re-focus the agency on basic operations (SPEED), intensify maintenance (Subway Action Plan), and accelerate system upgrades (Fast Forward). SPEED aside, these corrective actions were not free lunches. In the absence of efforts to remedy NYCT’s maintenance productivity and service management problems, more work meant more service disruptions (during all off-peak hours, but especially on weekends, which will be the focus of this post), and more disruptions meant more supplements and variability.

The resulting scheduling problems have two components: one of scheduling policy, and the other of schedule quality. Increasing work volumes led to changed supplement scheduling practices that have degraded the accuracy of the system’s schedules and contributed to rising operating costs. At the same time, the sheer volume of supplement schedule production demanded by the agency’s work regime has collided with the dire effects of an agency-wide hiring freeze and longstanding underinvestment in scheduling capacity to seriously erode the quality of the schedules written by the agency. The net result: far from being an efficient scheduled transit service, the subway’s weekend reality today is chaotic, as reasonable management of costs and rider experience has become increasingly impossible in a time of imprecise schedules.

Shooting Blind

As maintenance volumes escalated, especially following the July 2017 announcement of the Subway Action Plan, the system slowed down. After a series of worker fatalities in 2006, NYCT adopted a stringent set of roadway worker protection rules which mandate that all trains passing alongside or through work zones slow to 10 miles per hour. While lighter work volumes had previously meant that those delays could be handled within the basic running time adjustments programmed into diversion schedules (eg. the added time that comes with an express schedule revised to run local), heavier workloads created an operating environment in which weekend trains would often run very late. Poor schedule adherence meant that weekend trains would arrive at merge points well off of schedule, exponentiating delays from work projects. As these work and merge delays propagated across a line and through the day, slow service interfered with the overall fabric of the system’s timetable. Frequencies fell below scheduled levels as trains and crews failed to cycle quickly enough along lines, and as congestion around work zones reduced throughput. Between February 2017 and February 2018, the fraction of weekend trains actually run fell from 98 percent to 94 percent as diversion-driven cancellations cut into service levels on lines with heavy workloads. This was not a tenable situation, yet was one which seemed to becoming the norm for New York’s weekends.

How NYCT’s supplement scheduling policies changed in 2018. Credit: me.

The agency’s fix for this set of service problems was to adjust their supplement scheduling practices to accommodate the system’s changing realities. There were essentially two prongs of this effort. The first involved reducing scheduled frequencies to align train volumes with line capacities; the second involved adding extra running time to ensure on-time arrivals at termini. The shadow service cuts involved in weekend work commanded more initial attention (they have been baked into base timetables since then), but the crudely implemented spate of weekend running time policy adjustments have been more lasting and harmful. Since the fall of 2018, the agency’s supplement schedules have included 3-8 minutes of additional running time for each work zone or diversion through which a line passes. These values are not tailored to expected work intensities along the line, but rather are standardized for each type of diversion (eg. when Es, Fs, and Rs sharing a track along Queens Boulevard, their schedules get an extra 8 minutes).

To make matters worse, the insertion of this time into schedules is imprecise. Rather than extending schedules where trains will pass through work areas, running time additions are sometimes implemented as long scheduled stops at stations remote from the actual work zone (this method of addition is also what lies behind the countdown clock issue — the clocks think trains will actually obey those holds). This approach means, for example, that adjustments to F schedules for trackwork in Queens might show up as an 8 minute long scheduled stop in South Brooklyn. That is trivial for line-level on-time performance calculations, but it is in fact very important for the operational integrity of a service. With its compensatory running time additions near the end of the line, an F delayed by work in Queens will run late into its merge with the G at Bergen St, as well as any other merge created by service changes routing trains onto the F’s tracks in Manhattan.

A case study in the implementation of running time additions on the F line. Note the placement of additional running time in South Brooklyn, far from the relevant service change (Fs running local in Queens). How to read this chart: the horizontal axis is time; the vertical is distance. Each line is a train’s schedule. Credit: me, with data from NYCT GTFS feeds.

These modifications consequently have a mixed effect on weekend service. While they solve equipment cycle time problems, merges remain poorly coordinated, and with standards-based (rather than condition-based) time additions to schedules, timetables continue to diverge from realities on the ground. To take two examples from the last weekend in August: service changes ran E trains over the F in Manhattan, for which the F received some running time adds in Brooklyn. Those, along with padding baked into the base schedule, provided the F with a cushion well in excess of what was actually required, so F trains stacked up around Coney Island running down the clock until their scheduled slot into the terminal.

F train service this past weekend. Note how trains ran slightly behind schedule through most of the line, only to end up well ahead of schedule through the last few stops where considerable running time had been added (I’ve highlighted one trip for clarity). Also note that the service change which ran Fs express from 18 Av to Kings Highway did not get programmed into the supplement schedule — these sorts of schedule-less service changes have become increasingly common of late. Credit: me, with data from NYCT GTFS feeds and from Philippe Vibien’s fantastic stringline website.

Over on 4th Avenue, schedules erred in the opposite direction. D, N and R trains were all running express towards Manhattan to allow work on the local track. Supplements provided 5 minutes of additional running time, inserted as an extended stop at 36 St (a great example of a running time increase implemented at the work zone). These 5 minutes were far from enough; D trains left the far end of the service change running 10-15 minutes behind schedule. When those Ds reached Manhattan, they entered a second diversion, running alongside the A and C on Central Park West’s local track — where their unplanned arrivals interacted with similarly off-schedule A and C arrivals to create relatively frequent merge delays at 59 St. Each of these mismatches is costly. Overlong F schedules inflate operating costs and crew requirements in the face of a budget and crew headcount crisis; inadequately padded D schedules cause rippling delays at merge points and force shadow service cuts as train and crew cycle times shoot up.

Major delays on the D. Credit: me, with data from NYCT GTFS feeds and from Philippe Vibien’s fantastic stringline website

At this point, you might be wondering why NYCT does not use a more precise methodology for timetabling. The short answer? Predictability. Within each weekend, running times are only a bit more inconsistent than they are on weekdays; when trains are slow, they’re consistently slow, and vice versa. However, between weekends, running times vary widely. To give one example: through the fall of 2019, NYCT frequently ran all service on the Queens Boulevard line local to accommodate signal modernization work. On some weekends, service ran essentially as it did on weekdays. On others, trains blew through their 8 minutes of padding as the Queens corridor turned to molasses. Moreover, the localization of these delays varied extensively between weekends. On one, delays were concentrated around 63rd Drive; on another, they fell near 36th Street and Queens Plaza. It is not enough, then, to simply know that there will be work on a line. Writing accurate schedules requires precise knowledge of what work is going in each diversion, and where that work will impact service.

Weekend and weekday runtime variability on the A. While weekends can be more variable than weekdays, the dominant trend is heterogeneity in running time between weekends, not within them. Credit: me, with data from NYCT GTFS feeds and Philippe Vibien’s archive of NYCT’s GTFS-RT feeds.
An illustration of variable running times across 4 weekends of all-local service on Queens Boulevard. Credit: me, with data from NYCT GTFS feeds and Philippe Vibien’s archive of NYCT’s GTFS-RT feeds.

That information is difficult to provide. Thanks to their complexity (and to staffing and infrastructure issues I’ll discuss in a moment), supplement schedules take time to write; the agency generally plans them on six-week time horizon. While the development process for diversions and their supplements obviously involves an understanding of what work needs to happen during outages, a six-week projection for capital or maintenance projects may not be able to accurately predict the exact locations and intensities of work on a given day of a diversion, rendering precise estimates of schedule impacts impossible. This is not to say that the agency’s current practices have no scope to change — there may be some potential to begin tuning time additions to the number of different projects in each work zone, or the like — but in a project management environment where delays and cost overruns are common, it is understandable that planners might take long-run fine-grained work forecasts with a grain of salt. Writing better supplements will either require a complete overhaul of the agency’s work planning and asset monitoring processes to allow high-precision planning, or shorter supplement schedule lead times. The reason NYCT have not acted on one of these fronts yet? Institutional capacity.

Mixed Signals

NYCT’s struggles with its scheduling capabilities is most visible in the quality of the timetables it produces. The reality of the system today is that even if maintenance predictability problems were to disappear, the schedules put out by the supplement scheduling process would still likely be imperfect. Especially since the beginning of the pandemic, the agency has been producing increasingly inoperable timetables. In these, trains on the same track might be scheduled to pass each other, while others run at impossible 0 second headways, while even others run down varying amounts of padding at the same platform, resulting in as many as 4 trains being scheduled to be in the same place at the same time. Perhaps most concerningly (as noted in the caption to in the F line chart above), some service changes are now being done entirely without supplements, relying on dispatchers and padding built into the base timetable to manage the impacts of the reroute. The impact of these scheduling failures on delivered service is attenuated by the fact that weekend trains tend to have poor schedule adherence in general, but nevertheless present a roadblock to improvement. They all but guarantee that trains running on time will become late, which is hardly a desirable outcome for a schedule — these are supposed to be planning instruments which make the system work.

A particularly egregious schedule from November 2020. Credit: me, with data from NYCT GTFS feeds.

Behind these scheduling failures are conjoint problems of technology, staffing, and workload. NYCT, like many other transit operators in the US, uses a software called HASTUS for its vehicle and crew scheduling. Originally developed for bus operations, NYCT customized the software to be a useful support for rail scheduling tasks as well. However, the agency’s HASTUS installation is out of date. It has now been eight years since the agency last upgraded HASTUS, which stands in contrast to the six year upgrade cycle seen at less operationally complex agencies such as Chicago’s CTA. Thankfully, there is now hope for a better future in this arena: in July 2022, the MTA board approved a procurement to obtain a newer and better-customized version of HASTUS in the coming months, hopefully paving the way for a more efficient and effective supplement scheduling process in the years to come.

Software is only one small part of the battle here, however. Few departments have been hit as hard by the MTA’s ill-advised and abortive ‘Transformation Plan‘ and hiring freeze as has Operations Planning, the division of the agency responsible for creating schedules. Between the 2016 and 2021, the department went from about 400 budgeted and 377 filled positions to 350 budgeted and 284 filled. While a small portion of the change in budgeted headcount is likely attributable to changing responsibilities (as you might have guessed from its headcount, Ops Planning does a lot more than just scheduling, and some of those other responsibilities have been shifted over the years), the steep drop in the size of the planning workforce in the face of fast-increasing planning and scheduling requirements is grounds for alarm. Scheduling (and transit planning more broadly) is complex and contextual; even when supported by up-to-date software, schedulers must have a deep understanding of the quirks of each agency’s rules and network to write timetables and use heavily customized software effectively. A wave of departures like this not only places impossible strain on the agency’s remaining schedulers, but also risks interrupting the smooth reproduction of this institutional knowledge. The end of the agency’s hiring freeze provides some hope for improvement, but with a fiscal crisis looming and staff departures continuing, the horizon is hardly clear of storms.

There is, of course, another basic reality driving the divergence between agency resources and the demands of its right-of-way work: NYCT’s patchwork maintenance model. With its multi-track main lines and flexible interlockings, New York’s subway is well designed for its current maintenance strategy, in which service is rerouted around work zones but continues to run through the line. While it provides continuous (albeit confusing and slow) service during disruptions, this maintenance paradigm requires an enormous amount of supporting infrastructure, whether that be customized supplement schedules, flaggers, or otherwise. Before the Subway Action Plan and accelerated system upgrade projects exponentiated disruption levels during off-peak hours, subway schedulers were writing about eighty individual supplement schedules each week. Today, despite staff departures, that value remains high. Service this past week (8/27-9/2) involved the use of about 70 different supplements — and by New York standards, this week was not that rich in service changes.

Summer RATP closure map, modeling the sort of full shutdown treatment New York might benefit from. Image credit

Concentrating work into periodic full line segment shutdowns (as is done in Paris and Madrid) might both provide more consistent and predictable service for riders at lower costs, and help reduce the scheduling pressure and unpredictability that comes with today’s constant slew of minor diversions. Doing these shutdowns well would require considerable investments in bus capacity, regional rail alternatives, maintenance equipment, and otherwise, yet they might provide a much improved service product and more sustainable set of institutions for our transit network. Make no mistake: the first step towards better and more manageable off-peak service is improving the productivity of work crews and existing track outages through coordination and process enhancement. Yet alongside those efforts, there needs to be a parallel conversation about how we structure track access, especially given the post-COVID environment of depressed weekday ridership and strong weekends.

Conclusion

New York City’s supplement scheduling problem is a problem with no simple answer. It is equal parts staffing, technology, process, and infrastructure, and is entangled with the agency’s struggles with maintenance productivity, off peak service, and staffing, making its resolution an all-or-nothing project of holistic and far-reaching reform. Yet ensuring that New York City’s transit network can operate with the reliability and efficiency the city deserves is a goal worthy of such extensive efforts. Especially as the peak — formerly the time when transit networks were strongest — wanes, reliable off peak and weekend service will be essential to growing transit ridership following the pandemic. However esoteric the subject might be, however complex and impervious the institutional dimensions of the problem might seem, it is a project of reform which simply must be done. New York’s future relies on it.

One thought on “The Subway’s Broken Schedules

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s