The Regional Plan Association released a new report, called New York Penn Station: Constraints and Considerations for Meeting Future Demand. It argues against through-running on the grounds that it would reduce capacity, and asserts that any solution to station capacity after the Gateway tunnel, the Hudson Tunnel Project, opens in the late 2030s must include “station expansion or some other form of system expansion.” There’s something disappointing about this report, not because it’s wrong (although it is) but because it seems to just rehash past arguments without seriously addressing past criticisms of the $17 billion Penn expansion.

The lead author Rachel Weinberger and I talked about Penn Station capacity issues in 2024, after I wrote these two blog posts attacking Amtrak for its assumptions that imply additional tracks are required at Penn Station. Since then, the Effective Transit Alliance has done additional work, modeling the interfaces between the trains, the Penn Station platforms, and the concourses to arrive at feasible dwell times and capacities. Unfortunately, the RPA isn’t really engaging with any of that work, and retreats to just asserting that through-running would reduce capacity at Penn Station.

Capacity and dwell times

The most advanced technical work on the subject of capacity at Penn Station, the question being whether the bare Hudson Tunnel Project can perhaps with minor facelifts provide enough capacity to run 48 trains per hour between New Jersey and New York and the answer being yes, was done in the above-linked ETA report. We found the source of past claims by Amtrak that very long dwell times are necessary and deconstructed their assumptions, and modeled based on the current (post-Moynihan Station project) capacity of vertical circulation elements how long it would take passengers to clear each platform under rush hour assumptions.

The interface between the train and the platform itself is not the limiting factor; my two above linked blog posts from 2024 go over this and find a very short minimum dwell time, of 2 minutes or even less. The limiting factor is vertical circulation between the infamously narrow platforms and the concourses. Nonetheless, passengers can clear these in about 3 minutes. Notably, the narrowest of the platforms, platform 9, is compliant with NFPA 130 and its requirement that passengers be able to clear the platform in case of a fire in 4 minutes; the other platforms (except the wide platform 10) are not compliant, because they have two adjacent tracks and NFPA 130 assumes both tracks can be occupied, in which case the load doubles. Across all platforms, the one with the highest clearing time, platform 4, can clear a 1,620 passenger NJT bilevel in 4.83 minutes. Writing timetables to cycle trains between the platforms so that no platform comes close to having queuing between trains, is a routine exercise.

The report instead asserts that,

The time that the train dwells at the platform is 7 minutes in reasonable operating conditions: 5 minutes to unload passengers, 1 minute to check for stray passengers and then 1 minute for schedule recovery. There’s an additional 2.5 minutes to clear the interlockings and for train safety separations.

Accounting for all these aspects, under reasonable operating conditions, each track can accommodate 6 trains inbound per hour using drop-and-go service.

There is no citation for the model used, nor justification for either the 5 minute figure or for why an additional minute for schedule recovery is required. There is no explanation for why the 2.5 minutes to clear the interlockings matter to the capacity of each platform – once a train is past the platform, it’s gone and the capacity to measure is that of running track, not the platform. On outbound trains, it ominously says that,

The process of bringing an empty train into Penn Station, loading passengers and then departing takes 9 minutes in reasonable operating conditions during the PM peak period. This includes 7 minutes for passengers to hear the track announcement, descend to the platform and load into the train and 2 minutes for schedule recovery.

“Hear the track announcement” is an Americanism. In Germany, the track numbers are printed on one’s ticket, even on intercity trains, even at capacity-constrained stations with track shortages like Köln Hbf. The track for each fixed branch should be scheduled months in advance and known by regular passengers. Precisely because demand is asymmetric – toward Manhattan in the morning, away from Manhattan in the afternoon – the on and off peaks do not coincide at all, and encouraging passengers to get to the platform even before the train arrives would not overload the platform or the access and egress points.

The report completely missed the consequence of the asymmetry of demand when it finally asserted, based on modeling for which the report provides no details, that,

It would take an estimated 6 minutes for passengers to deboard, clear the platform and reach the concourse and another 4 minutes for passengers to descend to the platform and board the train during the AM peak. 2 additional minutes must be added to provide a buffer that ensures schedule adherence. The trains also require 2.5 minutes for clearing interlockings and safety separation, resulting in an hourly capacity of 4 trains per track and potentially 8 per platform.

A good rule of thumb here is that if Munich manages to slot 7 numbered branches, rising to 9 when one includes sub-branches, through one central trunk tunnel with 30 peak trains per hour, and comparable ridership to all three New York commuter rail systems combined, then nobody needs to add 2 minutes to the dwell time for schedule recovery, or 2.5 minutes for interlockings. The RPA is welcome to release its model for why it should take 10 minutes for passengers to board and alight; ours is open for inspection on GitHub and finds that the busier of the two can be done in 3 with a bit of buffer time and the less busy of the two is essentially free since so few passengers ride reverse-peak, and the train can leave even if some passengers that got off it are still on the platform on their way to the escalators.

The Parisian issue

The RPA report doesn’t talk about Paris. It doesn’t rebut the point that the RER is a good comparison case for New York commuter rail capacity, but it’s clearly lurking in the background. It does mention the RER as an ill comparison for the benefits of through-running (see section below), on the grounds that “trains follow each other along the same paths,” whereas in New York there is more and, measured by number of stations if not distance, earlier branching. But it doesn’t really address the point that if the RER can run 24-27 trains per hour per track in each direction, and the Munich S-Bahn can run 30, then so can through-running paths at Penn Station.

Reverse-engineering from what the report does say and from what the biggest points of contention have been when I talked to the RPA on this, they clearly think it matters that the European comparison cases have multiple city center stations to spread the load. Penn Station, in contrast, is the single central business district station, in a high-kurtosis city with far higher job density within walking distance of the station than can be found in any European city.

And yet, as I explained in my second original post from 2024, the effect coming from New York’s single city center station versus Paris’s multiplicity thereof largely cancels out the effect of much higher overall ridership on the RER than on New York commuter rail. Averaged over the peak four hours, the highest resolution I have for Paris, the sum of peak boardings and alightings per train is actually a bit higher on the RER E at Haussmann-Saint-Lazare, as of the 2010s when it was still a terminal, than it was at Penn Station in 2019 just before corona. The same sum at each of Gare du Nord and Gare de Lyon on the RER D is respectively 20% and 15% less than at Penn Station; on the RER B at Gare du Nord, it’s also 20% less. The Gare du Nord numbers are unlinked, so passengers interchanging across the platform between the RER B and D are counted in both trains’ ridership, but from the perspective of the train-to-platform interface, this is still a flow that the train and its doors and platforms must accommodate. If that’s doable in a 55 second dwell time, then trains at Penn Station can unload in 2 minutes at the peak and the rest is just a matter of counting platform-to-concourse vertical circulation elements and adding up their capacities as in the model described in the above section.

The benefits of through-running

Through-running works ideally when there are multiple city center stations, allowing the trains to function as urban rail as well as distribute passengers across multiple destinations. This is well-known to any group pushing through-running in New York, which is why so many such groups advocated for a tunnel connecting Penn Station with Grand Central, the so-called Alternative G in the ARC era in the 2000s, and why Tri-State’s proposal showcases a trunk line from Newark Penn Station to Sunnyside, and why ReThink heavily markets Secaucus and Sunnyside as secondary business centers. Here’s what ETA produced in 2023:

The upshot of this is that even with the dominance of Manhattan, any reasonable through-running system cobbled from existing and under construction infrastructure would unlock commutes from east of the Hudson to Downtown Newark, and from west of the Hudson to Long Island City (the station labeled Queens Junction is essentially Sunnyside Junction, walkable to a large fraction of Long Island City jobs), Flushing, and Jamaica.

It’s important context to understand why the RPA graphics denigrating through-running as limited in use don’t work:

In truth, even setting aside a Penn-Grand Central tunnel (which is doable at MTA construction costs at a lower cost than Penn expansion), the destination in New Jersey is likely to be not just on the North Jersey Coast Line as depicted, but on any of the Northeast Corridor, North Jersey Coast, and Raritan Valley Lines. At present, the first two carry 14 trains per hour at the peak, and are decently likely to be one’s pair east of the Hudson. Similarly, one’s destination east of Penn Station is unlikely to be Long Island as commonly defined to exclude Brooklyn and Queens but rather to be on the trunk in Long Island City or on multiple branches within the city or perhaps on the New Haven Line.

The upshot is a hefty share of the total through-Manhattan market would in fact be served by a through-running system, usually with one-seat rides, or if not then with transfers at Sunnyside or Secaucus rather than New York. Through-running is not about those mythical trips from Oyster Bay to the Jersey Shore that Adirondacker in the comments section mocks, just as it isn’t about trips from Aulnay to IHES on the RER B or from the Marne-la-Vallée branch of the RER A to Saint-Germain-en-Laye. It’s about the overlapping near-center markets, individually small compared with commutes to the central business district and yet collectively significant.