Category: Transportation

The Yamanote Line: a Ring or a Radial?

Note: I am going to take some suggestions for post topics in the future. This post comes from a Twitter poll I ran the day before yesterday.

The Yamanote Line in Tokyo is a ring. Trains go around the ring as on any other circular rail line. However, the line is not truly circumferential, since it serves as a north-south trunk through Central Tokyo. In that way, it contrasts with fully circumferential rings, such as the Moscow Circle Line, Seoul Metro Line 2 (see update below), and the under-construction Paris Metro Line 15. It’s really a hybrid of radial and circumferential transit, despite the on-paper circular layout. In previous posts I’ve attacked one kind of mixed line and given criteria for when another kind of mixed line can work. In this post, I’m going to discuss the kind of mixed line Yamanote is: why it works, and in what circumstances other cities can replicate it.

Consider the following diagram:

The red and blue lines are radial. The other three are hybrids. The yellow line is radial, mostly, but skirts city center and acts as a circumferential to its west; this kind of hybrid is nearly always a bad idea. The pink line is radial, but at the eastern end bends to act as a circumferential at the eastern end; this kind of hybrid is uncommon but can work in special cases, for example if Second Avenue Subway in New York is extended west under 125th Street. The green line is a Yamanote-style ring, offering radial service through city center but also circumferential service to the south and west.

On this map, the green line ensures there is circumferential service connecting what are hopefully the major nodes just west and south of city center. It doesn’t do anything for areas north and east of it. This means that this line works better if there is inherently more demand to the west and south than to the east and north. In Tokyo, this is indeed the case: the Yamanote ring offers north-south circumferential service west of Central Tokyo, through what are now the high-density secondary business districts of Ikebukuro, Shinjuku, and Shibuya. East of Central Tokyo, the only really compelling destinations, judging by subway ridership, are Oshiage and Asakusa, and neither is as big as Ikebukuro, Shinjuku, or Shibuya. Toyosu has high subway ridership, but is close enough to the water that it’s hard to build a circumferential through it.

Such a mixed line also becomes more useful if the radial component is better. The radial line can’t extend very far out, since the line needs to form a ring, so it should connect to very high-density neighborhoods just a few stops outside city center, or else provide additional service on an overloaded radial trunk. The Yamanote Line benefits from looking less like a perfect circle and more like upside-down egg, with two elongated north-south legs and two short (one very short) east-west legs; it extends its radial segment slightly farther out than it would otherwise be. In Tokyo, of course, all rail lines serving the center are beyond capacity, so the Yamanote Line’s extra two tracks certainly help; in fact, the two radial lines going north and south of Tokyo Station on parallel tracks, the Tohoku and Tokaido Lines, are two of the three most overcrowded in the city. (The third is the Chuo Line.) There’s even a dedicated local line, Keihin-Tohoku, covering the inner segments of both lines, making the same stops as Yamanote where they are parallel, in addition to the more express, longer-distance Tokaido and Tohoku Main Line trains.

Finally, there should not be radials that miss the mixed line; this is always a danger with subway lines that are neither pure radials nor pure circumferentials. Yamanote avoids this problem because it’s so close to the water at Shimbashi that the north-south subway lines all curve to the west as they go south, intersecting the ring. It’s actually the east-west lines that cross the Yamanote Line without transfers, like Tozai and Hanzomon; the north-south lines intersect the line with transfers.

The obvious caveat here is that while the Yamanote Line functions very well today, historically it did not originate as a circumferential in an area that needed extra service. It was built as a bypass around Central Tokyo, connecting the Tokaido and Tohoku Line at a time when Tokaido still terminated at Shimbashi and Tohoku at Ueno. Tokyo Station only opened 30 years later, and the ring was only completed another 10 years after that. Shinjuku only grew in the first place as the junction between the Yamanote and Chuo Lines, and Ikebukuro and Shibuya grew as the terminals of interwar private suburban railways. When the line opened, in 1885, Tokyo had 1.1 million people; today, the city proper has 9.5 million and the metro area has 38 million. The early rail lines shaped the city as much as it shaped them.

Nonetheless, with the economic geography of Tokyo today, the Yamanote Line works. Even though the history is different, it’s a useful tool for mature cities seeking to build up their rail networks. Provided the principles that make for the Yamanote Line’s success apply – stronger demand for circumferential service on one side of city center than on the others, demand for supplemental inner radial service, and good connections to other lines – this layout can succeed elsewhere.

Waterfront cities should take especial note, since they naturally have one side that potentially has high travel demand and one side that has fish. In those cities, there may be value in running the radial closest to the shoreline in a ring with an inland line.

This does not mean that every waterfront city should consider such a line. On the contrary: non-examples outnumber examples.

In Toronto, using two mainline tracks and connecting them to a ring to provide subway relief could have worked, but there are no good north-south corridors for such a ring (especially on the west), and the only good east-west corridor is Eglinton, which is being built incompatible with mainline rail (and has too much independent value to be closed down and replaced with a mainline link).

In Chicago, the grid makes it hard to branch lines properly: for example, a ring leaving the Red Line heading west at Belmont would necessary have to branch before Belmont Station, cutting frequency to the busiest station in the area. Plans for a circle line from last decade also faced limited demand along individual segments, such as the north-south segment of the Pink Line parallel to Ashland; ultimately, the planned line had too small a radius, with a circumference of 16 km, compared with 34.5 for Yamanote.

In Tel Aviv, there just isn’t any compelling north-south corridor outside the center. There are some strong destinations just east of Ayalon, like the Diamond Exchange and HaTikva, but those are already served by mainline rail. Beyond that, the next batch of strong destinations, just past Highway 4, is so far from Central Tel Aviv that the line would really be two radials connected by a short circumferential, more the London Circle Line when it was a full circle than the Yamanote Line, which is just one radial.

So where would a Yamanote-style circle be useful outside Tokyo? There are semi-plausible examples in New York and Boston.

In New York, it’s at the very least plausible to cut the G off the South Brooklyn Line, and have it enter Manhattan via the Rutgers Street Tunnel, as a branch of the F, replacing the current M train. There is no track connection enabling such service, but it could be constructed just west of Hoyt-Schermerhorn; consult Vanshnookraggen’s new track map. This new G still shouldn’t form a perfect circle (there’s far too much radial demand along the Queens Boulevard Line), but there are plausible arguments why it should, with a short tunnel just west of Court Square: namely, it would provide a faster way into Midtown from Williamsburg and Greenpoint than the overcrowded L.

In Boston, there is a circumferential alignment, from Harvard to JFK-UMass via Brookline, that can get a subway, in what was called the Urban Ring project before it was downgraded to buses. Two of the busiest buses in the region, the 1 and 66, go along or near the route. An extension from Harvard east into Sullivan and Charlestown is pretty straightforward, too. Beyond Charlestown, there are three options, all with costs and benefits: keep the line a semicricle, complete the circle via East Boston and the airport, and complete the circle via the North End and Aquarium. The second option is a pure circumferential, in which South Boston, lying between East Boston and JFK-UMass, would get better service north and south than west to Downtown. The third option cuts off East Boston, the lowest-ridership of the radial legs of the subway, and offers a way into the center from South Boston and Charlestown.

Of note, neither New York nor Boston is a clear example of good use of the Yamanote-style ring. This style of mixed line is rare, depending on the existence of unusually strong circumferential demand on just one side (west in Boston, east in New York), and on the water making it hard to build regular circles. It’s an edge case; but good transit planning revolves around understanding when a city’s circumstances produce an edge case, in which the simplest principles of transit planning (“every subway line should be radial or circumferential”) do not apply.

Update 5/16: commenter Threestationsquare reminds me that Seoul Metro Line 2 is the same kind of ring as Yamanote. The north leg passes through City Hall, near the northern end of the Seoul CBD, providing radial east-west service. The south leg serves a busy secondary commercial core in Gangnam, Tehran Avenue; Gangnam Station itself is the busiest in Seoul, and has sprouted a large secondary CBD.

Slotting Intercity Trains on Regional Lines

In 2011, Clem Tillier and Richard Mlynarik put out sample schedules for modernized Caltrain service, with an applet anyone could use to construct their own timetables. I played with it, and one of the schedules I made, a trollish one, had room for local and express regional trains, but not intercity trains; intercity trains would be slotted with express regionals, and make the same stops. This was a curious exercise: intercity trains would be high-speed rail, which should not slow down to make every express regional stop. But more recently, as I’ve worked on schedules for Boston and New York, I’ve realized that when the regional trains are fast, there is merit to slotting legacy (but not high-speed) intercity trains together with them.

The origin of this pattern is the problem of slotting trains on busy railroads. There are many lines that are not really at capacity, but cannot easily combine trains that run at different speeds. One solution to the problem is to build extra tracks and give the intercity trains a dedicated pathway. This works when there is heavy intercity traffic as well as heavy regional traffic, but four-tracking a long line is expensive; Caltrain and California HSR ended up rejecting full four-tracking.

Another solution, favored for Caltrain today instead of full four-tracking, is timed overtakes. I have argued in its favor for Boston-Providence and Trenton-Stamford for high-speed rail, but it requires more timetable discipline and makes it easier for delays on one train to propagate to other trains. It should be reserved for the busiest lines, where there is still not enough traffic to justify long segments with additional tracks (that would be four tracking Boston-Providence and six-tracking Stamford-New Rochelle and Rahway-New Brunswick), but there is enough to justify doing what is required to run trains on a tight overtake schedule. It is especially useful for high-speed trains, which tend to be the most punctual, since they use the most reliable equipment and have few stops.

But on lower-ridership intercity routes, the best solution may be to force them to slow down to the speed of the fastest regional train that uses the line. On the timetable, the intercity train is treated as a regional train that goes beyond the usual outer terminal. This option is the cheapest, since no additional infrastructure is required. It also boosts frequency, relative to any solution in which the intercity train does not make regional stops: since the intercity train is using up slots, it might as well provide some local frequency when necessary. These two benefits together suggest a list of guidelines for when this pattern is the most useful:

  1. The intercity line shouldn’t be so busy that a slowdown of 10 or 15 minutes makes a big difference to ridership relative to the cost of overtakes. Nor should it be especially fast.
  2. The regional line, or the most express pattern on the regional line if it has its own local and express trains, should have wide stop spacing, such that the speed benefit of running nonstop is reduced.
  3. The regional line should connect long-distance destinations in their own right, and not just suburbs, so that there is some merit to connecting them to the intercity line. These destinations may include secondary cities, airports, and universities (but airports would probably be intercity stops under any pattern).
  4. The regional and intercity lines should be compatible in equipment, which in practice means either both should run EMUs or both should run DMUs (locomotives are obsolete for passenger services).

Both Switzerland and Japan employ this method. In Switzerland, the fastest intercity trains in the Zurich/Basel/Bern triangle run nonstop. But intercity trains going north or east of Zurich stop at the airport, interlining with regional trains to create a clockface pattern of trains going nonstop between the airport and the city.

In Japan, high-speed services run on their own dedicated tracks, with separate track gauge from the legacy network, but legacy intercity services are integrated with express regional trains. An intercity trip out of Tokyo on the Chuo Line starts out as a regular express commuter train, making the same stops as the fastest express trains: starting from Shinjuku, the Azusa sometimes stops at Mitaka, skips Kokubunji, and stops at Tachikawa and Hachijoji. Beyond Hachijoji, some trains make regional express stops, others run nonstop to well beyond the Tokyo commuter belt. On the Tokaido Line, the intercity trains (the Odoriko) skip stops that every regional train makes, but they still stop at Shinagawa and Yokohama, and sometimes in some Yokohama-area suburbs.

In North America, there are opportunities to use this scheduling pattern in New York, Boston, and Toronto; arguably some shorter-range intercity lines out of Philadelphia and Chicago, such as to Reading and Rockford, would also count, but right now no service runs to these cities.

In Toronto, GO Transit already runs service to Kitchener, 100 kilometers from Union Station. For reasons I don’t understand, service to Kitchener (and to Hamilton, a secondary industrial city 60 km from Toronto) is only offered at rush hour; in the off-peak, commuter trains only run closer in, even though usually intercity lines are less peaky than commuter lines. There is also seasonal service to Niagara Falls, 130 km from Toronto. As Metrolinx electrifies the network, higher frequency is likely, at least to Hamilton, and these trains will then become intercity trains running on a regional schedule. This works because GO Transit has very wide stop spacing, even with proposed infill stops. Niagara Falls is a leisure destination, with visitors from all over the Greater Toronto Area and not just from Downtown, so the extra stops in the Toronto suburbs are justified. Right now, Niagara Falls trains make limited stops, about the same number in the built-up area as the express trains to Hamilton but on a different pattern.

There are no infill stops planned on Lakeshore West, the commuter line to Hamilton and Niagara Falls. It is likely that future electrification and fare integration will create demand for some, slowing down trains. The line has three to four tracks (with a right-of-way wide enough for four) and is perfectly straight, so as demand grows with Toronto’s in-progress RER plan, there may be justification for local and express trains; express trains would make somewhat fewer stops than trains do today, local trains would stop every 1-2 km in the city and in Mississauga. Intercity trains could then easily fit into the express commuter slots; potential destinations include not just Hamilton and Niagara Falls, but also London.

This is unfriendly to high-speed trains. However, Canada is not building high-speed rail anytime soon; if it were, it would connect Toronto with Montreal, using Lakeshore East, and not with points west, i.e. London and Windsor. London and Windsor are small, and a high-speed connection to Toronto would be financially marginal, even with potential onward connections to Detroit and Chicago. A Toronto-Niagara Falls-Buffalo-New York route is more promising, but dicey as well. Probably the best compromise in such case is to run trains on a four-tracked Lakeshore West line at 250 km/h; the speed difference with nonstop trains running at 160 km/h allows 15-minute frequency on each pattern without overtakes, and almost allows 12 minutes. Alternatively, express trains could use the local tracks to make stops, as I’ve recommended for some difficult mixtures of local, express, and intercity trains on the Northeast Corridor in New York.

In Boston, the Northeast Corridor is of course too important as an intercity line to be slowed down by regional trains. Thus, even though in other respects it would be great for merging intercity and regional service, in practice, overtakes or four tracks are required.

However, all other intercity-range commuter lines in Boston should consider running as regular commuter trains (electrified, of course) once they enter MBTA territory. These include potential trains to Hyannis on Cape Cod, 128 km from South Station; Manchester, 91 km from North Station; and Springfield, 158 km from South Station; as well as existing trains to Portland, 187 km from North Station. Hyannis, Manchester, and Portland all feed into very fast regional lines: my sample schedule and map have trains to Hyannis averaging 107 km/h and trains to Manchester averaging 97 km/h. Trains to Haverhill, the farthest point on the line to Portland with any Boston-bound commuter traffic, average 88 km/h.

Springfield is more difficult. The Worcester Line is slower, partly because of curves, partly because of very tight stop spacing in the core built-up area. Once under-construction infill is complete, Auburndale, 17 km out of South Station, will be the 7th station out, and another infill station (Newton Corner) is perennially planned; my schedule assumes 3 additional stations, making Auburndale the 11th station out. On the line to Hyannis, the 11th station out, Buzzards Bay, is at the Cape Cod Canal, 88 km out. There is room for four tracks for a short segment in Allston, but in the suburbs there is no room until past Auburndale, which constrains any future high-speed rail plan to Albany. Low-speed intercity trains would have to slow down to match commuter rail speed, because the alternative is to run commuter rail too infrequently for the needs of the line. Average speed from South Station to Worcester is 70 km/h, even with express diesels today, so it’s not awful, but here, slowing down intercity trains is a less bad option rather than a good one.

In New York, as in Boston, intercity trains fit in regional slots away from the Northeast Corridor. Already today there are intercity trains running on the LIRR, to the eastern edge of Long Island, much too distant from the city for commuter traffic. Those trains run nonstop or almost nonstop, and are infrequent; if the entire LIRR were electrified, and express trains were eliminated, locals could match the express speed today thanks to reduced schedule padding, and then some trains could continue to Greenport and Montauk providing perhaps hourly service. Service to Danbury and Waterbury on Metro-North is of similar characteristics.

The New Jersey end is more interesting. Right now, there is no significant intercity service there, unless you count the Port Jervis Line. However, New Jersey Transit is currently restoring service on the Lackawanna Cutoff as far as Andover, and there remain proposals to run trains farther, to Delaware Water Gap and Scranton. Those would be regular express diesel trains on the Morris and Essex Lines, presumably stopping not just at Hoboken but also at important intermediate stations like Newark Broad Street, Summit, and Morristown.

If service were electrified, those trains could run, again on the same pattern as the fastest trains that can fit the Morristown Line (where I don’t think there should be any express trains), going to New York and onward to whichever destination is paired with the shorter-range commuter trains on the line. The same is true of other potential extensions, such as to Allentown, or, the favorite of Adirondacker in comments, a line to West Trenton and onward to Philadelphia via the West Trenton SEPTA line. There’s not much development between the edge of the built-up suburban area at Raritan and either Allentown or the Philadelphia suburbs; but intercity trains, averaging around 90 km/h, could succeed in connecting New York with Allentown or with the northern suburbs of Philadelphia, where a direct train doing the trip in an hour and a half would be competitive with a train down to 30th Street Station with a high-speed rail connection.

The characteristics of intercity lines that favor such integration with regional lines vary. In all cases, these are not the most important intercity lines, or else they would get dedicated tracks, or overtakes prioritizing their speed over that of commuter trains. Beyond that, it depends on the details of intercity and regional demand. But by default, if an intercity line is relatively short (say, under 200 km), and not so high-demand that 200+ km/h top speeds would be useful, then planners should attempt to treat it as a regional line that continues beyond the usual terminus. Alternatively, the commuter line could be thought of as a short-turning version of the intercity line. Planners and good transit advocates should include this kind of timetabling in their toolbox for constructing integrated regional rail schedules.

Too Many Branches, Too Few Trunks

A recent discussion on Twitter about the through-running plan offered by ReThinkNYC got me thinking about an aspect American through-running crayonistas neglect on their maps: the branch-to-trunk ratio. It’s so easy to draw many branches converging on one trunk: crayon depicts a map and not a schedule, so the effects on branch frequency and reliability are hard to see.

In contrast with crayonista practice, let us look at the branch-to-trunk ratio on existing through-running commuter networks around the developed world:


The RER has 5 lines, of which 4 are double-ended and 1 (the E) is single-ended, terminating in the Paris CBD awaiting an extension to the other side. They have the following numbers of branches:

RER A: 3 western branches, 2 eastern branches.
RER B: 2 northern branches, 2 southern branches; on both sides, one of the two branches gets 2/3 of off-peak traffic, with half the trains running local and half running express.
RER C: 3 western branches, 4 eastern branches; one of the eastern branches, which loops around as a circumferential to Versailles, is planned to be closed and downgraded to a tram-train.
RER D: 1 northern branch, 3 southern branches; the map depicts 4 southern branches, but only 3 run through, and the fourth terminates at either Juvisy or Gare de Lyon.
RER E: 2 eastern branches; the ongoing western extension does not branch, but is only planned to run 6 trains per hour at the peak, so some branching may happen in the future.

The RER B and D share tracks between Chatelet-Les Halles and Gare du Nord, but do not share station platforms.


Thameslink has 3 southern branches. To the north it doesn’t currently branch, but there is ongoing construction connecting it to more mainlines, and next year it will gain 2 new northern branches, for a total of 3. Crossrail will have 2 eastern branches and 2 western branches. Crossrail 2 is currently planned to have 3 northern branches and 4 southern branches.


Berlin has 2 radial trunk routes: the east-west Stadtbahn, and the North-South Tunnel. The Stadtbahn has three S-Bahn routes: S5, S7, S75. The North-South Tunnel also has three: S1, S2, S25. Each of these individual routes combines one branch on each side, except the S75, which short-turns and doesn’t go all the way to the west.

Berlin also has the Ringbahn. The Ringbahn’s situation is more delicate: S41 and S42 run the entire ring (one clockwise, one counterclockwise), but many routes run on subsegments of the ring, with extensive reverse-branching. At two points, three services in addition to the core S41-42 use the Ringbahn: S45, S46, and S47 on the south, and S8, S85, and S9 on the east.


There is a two-track central tunnel, combining seven distinct branches (S1-8, omitting S5). S1 and S2 further branch in two on the west.

The excessive ratio of branches to trunks has created a serious capacity problem in the central tunnel, leading to plans to build a second tunnel parallel to the existing one. This project has been delayed for over ten years, with mounting construction costs, but is finally planned to begin construction in 2 days, with expected completion date 2026. At more than €500 million per underground kilometer, the second tunnel is the most expensive rail project built outside the Anglosphere; were costs lower, it would have been built already.


The Tokyo rail network is highly branched, and many lines reverse-branch using the subway. However, most core JR East lines have little branching. The three local lines (Yamanote, Chuo-Sobu, Keihin-Tohoku) don’t branch at all. Of the rapid lines, Chuo has two branches, and Tokaido and Yokosuka don’t branch. Moreover, the Chuo branch point, Tachikawa, is 37 km from Tokyo.

The northern and eastern lines branch more, but the effective branch-to-trunk ratio is reduced via reverse-branching. To the east, the Sobu Line has 5 branches, but they only split at Chiba, 39 km east of Tokyo. The Keiyo Line has 3 branches: the Musashino outer ring, and two eastern branches that also host some Sobu Line trains. The services to the north running through to Tokaido via the Tokyo-Ueno Line have 3 branches – the Utsunomiya, Takasaki, and Joban Lines – but some trains terminate at Ueno because there’s no room on the Tokyo-Ueno trunk for them. The services using the Yamanote Freight Line (Saikyo and Shonan-Shinjuku) have 2 southern branches (Yokosuka and Tokaido) and 3 northern ones (Utsunomiya, Takasaki, and a third Saikyo-only branch).

Conversely, all of these lines mix local and express trains on two tracks, with timed overtakes, except for the three non-branching local lines. The upper limit, beyond which JR East only runs local trains, appears to be 19 or 20 trains per hour, and near this limit local trains are consistently delayed 4 minutes at a time for overtakes.

Implications for Through-Running: Boston

In Boston, there are 7 or 8 useful southern branches: Worcester, Providence, Stoughton, Fairmount, the three Old Colony Lines, and Franklin if it’s separate from Fairmount. The Stoughton Line is planned to be extended to New Bedford and Fall River, making 8 or 9 branches, but the intercity character of the extension and the low commute volumes make it possible to treat this as one branch for scheduling purposes. To the north, there are 5 branches today (Fitchburg, Lowell, Haverhill, Newburyport, Rockport), but there are 2 decent candidates for service restoration (Peabody and Woburn).

The North-South Rail Link proposal has four-tracks, so the effective branch-to-trunk ratio is 3.5. It is not hard to run service every 15 minutes peak and every 30 off-peak with this amount of branching, and there’s even room for additional short-turn service on urban lines like Fairmount or inner Worcester and Fitchburg. But this comes from the fact that ultimately, Boston regional rail modernization would create an RER C and not an RER A, using my typology as explained on City Metric and here.

There are several good corridors for an RER A-type service in Boston, but those have had subway extensions instead: the Red Line to Braintree, the Orange Line to Malden, and now the Green Line Extension to Tufts. The remaining corridors could live with double service on an RER C-type service, that is, service every 7.5 minutes at the peak and every 15 off-peak. For this reason, and only for this reason, as many as 4 branches per trunk are acceptable in Boston.

Implications for Through-Running: New York

Let us go back to the original purpose of this discussion: New York through-running crayon. I have previously criticized plans that use the name Crossrail because it sounds modern but only provide a Thameslink or RER C. Independently of other factors, the ReThinkNYC plan has the same issues. It attempts to craft a sleek, modern regional rail system exclusively out of the existing Penn Station access tunnels plus a future tunnel across the Hudson.

Where Boston has about 7 commuter rail branches on each side, New York has 9 on Long Island (10 counting the Central Branch), 6 in Metro-North territory east of the Hudson, and 9 in New Jersey (11 counting the Northern Branch and West Shore Railroad). Moreover, one branch, the Hudson Line, has a reverse branch; where the Keiyo/Sobu reverse-branching in Tokyo and the Grand Central/Penn Station Access reverse-branching on the New Haven Line offer an opportunity to provide more service to a highly-branched line, the Hudson Line is a single line without branches.

The upshot is that even a four-track trunk, like the one proposed by both the RPA’s Crossrail NY/NJ plan and ReThinkNYC, cannot possibly take over all commuter lines. The frequency on each branch would be laughable. This is especially bad on the LIRR, where the branch point is relatively early (at Jamaica). The schedule would be an awkward mix of trains bound for the through-running system, East Side Access, and perhaps Downtown Brooklyn, if the LIRR doesn’t go through with its plan to cut off the Atlantic Branch from through-service and send all LIRR trains to Midtown Manhattan. Schedules would be too dependent between trains to each destination, and reliability would be low. ReThinkNYC makes this problem even worse by trying to shoehorn all of Metro-North, even the Harlem and Hudson Lines, into the same system, with short tunneled connections to the Northeast Corridor.

On the New Jersey side, the situation is easier. This is because two of the key branch points – Rahway and Summit – are pretty far out, respectively 33 and 37 km from Penn Station. The population density on branches farther out is lower, which means a train every 20 or 30 minutes off-peak is not the end of the world.

The big problem is the attempt to link the Erie lines into the same system. This makes too many branches, not to mention that the Secaucus loop between the Erie lines and the Northeast Corridor is circuitous. The original impetus behind my crayon connecting the South Side LIRR at Flatbush with the Erie lines via Lower Manhattan is that the Erie lines point naturally toward Lower Manhattan, and not toward Midtown. But this is also an attempt to keep the branch-to-trunk ratio reasonable.

The first time I drew New York regional rail crayon, I aimed at a coherent-looking system. The Hudson Line reverse-branched, and I was still thinking in terms of peak trains-per-hour count rather than in terms of a consistent frequency, but the inner lines looked like a coherent RER-style network. But the Hoboken-Flatbush tunnel still had 5 branches on the west, and the Morris and Essex-LIRR line, without a dedicated tunnel, had 4 to the east. My more recent crayon drops the West Shore Line, since it has the most freight traffic, leaving 4 branches, of which 1 (Bergen County) can easily be demoted to a shuttle off-peak, keeping base frequency on all branches acceptable without overserving the trunk; by my most recent crayon, there are still 4 branches, but there’s a note suggesting a way to cut this to 3 branches by building a new trunk. Moreover, several branches are reduced to shuttles (Oyster Bay, Waterbury) or circumferential tram-trains (West Hempstead) to avoid overloading the trunks. There’s a method behind the madness: in normal circumstances, there should not be more than 3 branches per double-track trunk.

I am not demanding that the RPA or ReThinkNYC put forth maps with multiple new trunk lines. The current political discussion is about Gateway, which is just 1 trunk line; it’s possible to also include what I call line 3 (i.e. the Empire Connection), which just requires a short realignment of an access track to Penn Station, but the lines to Lower Manhattan still look fanciful. New York has high construction costs, and the main purpose of my maps is to show what is possible at normal construction costs. But it would be useful for the studios to understand issues of frequency, reliability, and network coherence. This means no Secaucus loop, no attempt to build one trunk line covering all or almost all commuter lines, and not too many branches per trunk.

New York is an enormous city. It has 14 subway trunk lines, and many are full all day and overcrowded at rush hour. That, alone, suggests it should have multiple commuter rail trunk lines supplementing the subway at longer-range scale. It’s fine to build one trunk line at a time, as London is doing – these aren’t small projects, and there isn’t always the money for an entire network. But it’s important to resist the temptation to make the one line look more revolutionary than it is.

New York Regional Rail: Scheduling Trains of Different Speeds

The simplest train schedules are when every train makes every stop. This means there are no required overtakes, and no need for elaborate track construction except for reasons of capacity. In nearly all cities in the world, double-track mainlines with flying junctions for branches are enough for regional rail. Schedule complexity comes from branching and short-turns, and from the decision which lines to join together, but it’s then possible to run independently-scheduled lines, in which delays don’t propagate. I have worked on a map as part of a proposal for Boston, and there, the only real difficulty is how to optimize turnaround times..

But then there’s New York. New York is big enough that some trunk lines have and need four tracks, introducing local and express patterns. It also has reverse-branching on some lines: the Hudson Line and New Haven Line can serve either Penn Station or Grand Central, and there are key urban stations on the connections from either station to either line. The presence of Jamaica Station makes it tempting to reverse-branch the LIRR. Everything together makes for a complex map. I talked in 2014 about a five- or six-line system, and even there, without the local/express artifacts, the map looks complicated. Key decisions turn out to depend on rolling stock, on scheduling, and on decisions made about intercity rail fares.

Here is what I drew last week. It’s a six-line map: lines 1 and 2 connect the Northeast Corridor on both sides plus logical branches and the Port Washington Branch of the LIRR, line 3 connects Hempstead with the Empire Corridor, line 4 connects the Harlem Line with the Staten Island Railway as a north-south trunk, line 5 connects the Erie Lines with the South Side LIRR lines, line 6 connects the Morris and Essex Lines with the LIRR Main Line.

As I indicated in the map’s text, there are extra possible lines, going up to 9; if I revised the map to include one line, call it line 7, I’d connect the Northern Branch and West Shore Railroad to a separate tunnel under 43rd Street, going east and taking over the LIRR portions of line 3; then the new line 3 would connect the Hudson Line with the Montauk Line (both Lower Montauk and the Babylon Branch) via an East River Tunnel extension. The other options are at this point too speculative even for me; I’m not even certain about line 6, let alone line 7, let alone anything else.

But the real difficulty isn’t how to add lines, if at all. It’s the reverse branch of lines 1 and 2. These two lines mostly go together in New Jersey and on the New Haven Line, but then take two different routes to Manhattan. The difficulty is how to assign local and express trains. The map has all line 1 trains going local: New Brunswick-Port Washington, or Long Branch-Stamford. Line 2 trains are a mix of local and express. This is a difficult decision, and I don’t know that this is the right choice. Several different scheduling constraints exist:

  1. Intercity trains should use line 1 and not line 2. This is for two reasons: the curve radius between Penn Station and Grand Central might be too tight for Shinkansen trains; and the Metro-North trunk north of Grand Central has no room for extra tracks, so that the speed difference between intercity and regional trains (e.g. no stop at Harlem-125th) would limit capacity. For the same reason, line 1 only has a peak of 6 trains per hour on the Northeast Corridor east of where the Port Washington Branch splits.
  2. Since not many regional trains can go between New Rochelle and Penn Station on the Northeast Corridor, they should provide local service – express service should all go via Grand Central.
  3. There are long segments with only four tracks, requiring track sharing between intercity trains and express regional trains. These occur between New Rochelle and Rye, and between the end of six-tracking in Rahway and New Brunswick. See details and a sample schedule without new Hudson tunnels here. This encourages breaking service so that in the Manhattan core, it’s the local trains that share tunnel tracks with intercity trains, while express trains, which share tracks farther out, are less constrained.
  4. Express trains on the New Jersey side should stay express on the New Haven Line, to provide fast service on some plausible station pairs like Newark-Stamford or New Rochelle-New Brunswick. Flipping local and express service through Manhattan means through-riders would have to transfer at Secaucus (which is plausible) or Penn Station (which is a bad idea no matter how the station is configured).
  5. There should be infill stops in Hudson County: at Bergenline Avenue for bus connections and the high local population density, and just outside the portal, at the intersection with the Northern Branch. These stops should be on line 2 (where they can be built new) and not line 1 (where the tunnels would need to be retrofitted), and trains cannot skip them, so the line that gets these stops should run locals.

It is not possible to satisfy all constraints simultaneously. Constraint 5 means that in New Jersey, line 2 should be local and line 1 should be express. Constraint 4 means the same should be true on the Metro-North side. But then constraints 2 and 3 encourage making line 1 local, especially on the Metro-North side. Something has to give.

On the map, the compromise is that there’s an infill stop at Bergenline but not at the intersection with the Northern Branch (which further encourages detaching the Northern Branch from line 5 and making it part of a Midtown-serving line 7). So the line 2 express trains are one stop slower than the line 1 locals between Newark and New York, which is not a huge problem.

The scheduling is still a problem, The four-track segment through Elizabeth between the six-track segments around Newark Airport and in Linden and Rahway has to be widened to six tracks; the four-track segment between the split with the North Jersey Coast Line and Jersey Avenue can mix three speed classes, with some express trains sharing tracks with intercity trains and others with local trains, but it’s not easy. At least on the Connecticut side, any high-speed rail service requires so many bypasses along I-95 that those bypasses can be used for overtakes.

At this point, it stops being purely about regional rail scheduling. The question of intercity rail fares becomes relevant: can people take intercity trains within the metro area with no or limited surcharge over regional trains? If so, then constraint 4 is no longer relevant: nobody would take regional trains on any segment served by intercity trains. In turn, there would be demand for local intercity trains, stopping not just at New Haven, New York, Newark, and Philadelphia, but also at Stamford, New Rochelle, perhaps Metropark (on new express platforms), and Trenton. In that case, the simplest solution is to flip lines 1 and 2 in New Jersey: line 1 gets the express trains to Trenton and the trains going all the way to Bay Head, line 2 gets the locals to Jersey Avenue, the Raritan Valley Line trains, and the Long Branch short-turns.

This, in turn, depends on rolling stock. Non-tilting high-speed trains could easily permit passengers with unreserved seats to pay commuter rail fare. On tilting trains, this is dicier. In Germany, tilting trains with unreserved tickets (ICE-T) have a computer constantly checking whether the train is light enough to be allowed to tilt, and if it is too heavy, it shuts down the tilt mechanism. This should not be acceptable for the Northeast Corridor. This might not be necessary for tilting Shinkansen (which are so light to begin with this isn’t a problem, and they do sell unreserved tickets in Japan), but it’s necessary for Pendolinos and for the Avelias that Amtrak just ordered. Selling reserved tickets at commuter rail fares is another option, but it might not be plausible given peak demand into New York.

The point of this exercise is that the best transit planning requires integrating all aspects: rolling stock, timetable, infrastructure, and even pricing. Questions like “can intercity trains charge people commuter rail fares for unreserved tickets?” affect express regional service, which in turn affects which branch connects to which trunk line.

Ultimately, this is the reason I draw expansive maps like this one. Piecemeal planning, line by line, leads to kludges, which are rarely optimized for interconnected service. New York is full of examples of poor planning coming from disintegrated planning, especially on Long Island. I contend that the fact that, for all of the Gateway project’s scope creep and cost escalations, there’s no proposed stop at Bergenline Avenue, is a prime example of this planning by kludge. To build the optimal line 2, the region really needs to know where lines 3-6 should go, and right now, there’s simply none of this long-term planning.

Amtrak Defrauds the Public on Gateway Benefits

A stenographer at Bloomberg is reporting an Amtrak study that says the social benefit-cost ratio of the Gateway program is about 4. Gateway, the project to quadruple the double-track line from New York to Newark, including most important the tunnel across the Hudson, is now estimated to cost $25 billion. Cost overruns have been constant and severe: it was $3 billion in the ARC era in 2003, $9 billion when Governor Chris Christie canceled it in 2010, and $13.5 billion when Amtrak took over in 2011 and renamed it Gateway. And now Amtrak is claiming that the net present value of Gateway approaches $100 billion; in a presentation from late 2016, it claims that at a 3% discount rate the benefit-cost ratio is 3.87, and compares it positively with Crossrail and California HSR. This is incorrect, and almost certainly deliberate fraud. Let me explain why.

First, the comparison with Crossrail should give everyone pause. Crossrail costs around the same as the current projection for Gateway: about $21 billion in purchasing power parity terms, but future inflation means that the $25 billion for Gateway is very close to $21 billion for Crossrail, built between 2009 and 2018. Per Amtrak, the benefit-cost ratio of Crossrail as 3.64 at the upper end – in other words, the benefits of Crossrail and Gateway should be similar. They are clearly not.

The projection for Crossrail is that it will fill as soon as it opens, with 200 million annual passengers. There is no chance Gateway as currently planned can reach that ridership level. New Jersey Transit has about 90 million annual rail riders, and NJT considers itself at capacity. This number could be raised significantly if NJT were run in such a way as to encourage off-peak ridership (see my writeup on Metro-North and the LIRR, for which I have time-of-day data), but Gateway includes none of the required operational modernization. Even doubling NJT’s ridership out of Gateway is unlikely, since a lot of ridership is Hoboken-bound today because of capacity limits on the way to New York, and Gateway would cannibalize it; only about 60 million NJT riders are taking a train to or from New York, so a more realistic projection is 60 million and not 90 million. Some additional ridership coming out of Amtrak is likely, but is unlikely to be high given Amtrak’s short trains, hauled by a locomotive so that only 5-7 cars have seats. Amtrak has an asterisk in its comparison saying the benefit-cost ratios for Crossrail and Gateway were computed by different methodologies, and apparently the methodologies differ by a factor of 3 on the value of a single rider.

That, by itself, does not suggest fraud. What does suggest fraud is the history of cost overruns. The benefits of Gateway have not materially increased in the last decade and a half. If Gateway is worth $100 billion today, it was worth $100 billion in 2011, and in 2003.

One change since 2011 is Hurricane Sandy, which filled the existing North River Tunnels with corrosive saltwater. A study on repairs recommended long-term closure, one tube at a time. But the difference is still small compared to how much Amtrak thinks Gateway is worth. The study does not claim long-term closure is necessary. Right now, crews repair the tunnels over weekends, with weekend closures, since weekend frequency is so poor it can fit on single track. The study does not say how much money could be saved with long-term closures, but the cost it cites for repairs with long-term closures is $350 million, and the cost under the current regime of weekend closures cannot be several billion dollars more expensive. The extra benefit of Gateway coming from Sandy is perhaps $1 billion, a far cry from the almost $100 billion projected by Amtrak for Gateway’s worth.

What this means is that, if Gateway really has a benefit-cost ratio approaching 4 today, then it had a benefit-cost ratio of about 7 in 2011. Amtrak did not cite any such figure at the time. In 2003 it would have have had a benefit-cost ratio approaching 25, even taking into account inflation artifacts. None of the studies claimed such a high figure. Nor did any of the elected or appointed officials in charge of the project act like it was so valuable. Construction was not rushed as it would have if the benefit-cost ratio was so high that a few years’ acceleration would have noticeable long-term consequences.

The scope of the project did not suggest an extreme benefit-cost ratio, either. ARC, then Gateway, was always just two tracks. If a two-track tunnel has a benefit-cost ratio higher than 20, then it’s very likely the next two-track tunnel has a high benefit-cost ratio as well. Even a benefit-cost ratio of 4 would lead to further plans: evidently, Transport for London is planning Crossrail 2, a northeast-southwest tunnel complementing the east-west Crossrail and north-south Thameslink. Perhaps in 2003 Port Authority thought it could not get money for two tunnels, but it still could have planned some as future phases, just as Second Avenue Subway was planned as a full line even when there was only enough money for Phase 1.

The plans for ARC included the awkward Secaucus loop bringing in trains from the Erie lines into Penn Station, with dual-mode diesel/electric locomotives. This is a kludge that makes sense for a marginal project that needs to save every penny, not for one where benefits exceed costs by more than an order of magnitude. For such a strong project, it’s better to spend more money to get it right, for example by electrifying everything. It would also have been better to avoid the loop kludge and send Erie trains to Lower Manhattan and Brooklyn, as I have proposed in various iterations of my regional rail plan.

All of this together suggests that in 2003, nobody in charge of ARC thought it was worth $70 billion in 2003 dollars, or around $100 billion in 2017 dollars. Even in 2011, Amtrak did not think the project was worth $85 billion in 2011 dollars. It’s theoretically possible that some new analysis proves that old estimates of the project’s benefits were too low, but it’s unlikely. If such revisions were common, we would see upward and downward revisions independent of cost overruns. Some rail projects with stable costs would see their benefit-cost ratios shoot up to well more than 10. Others might be revised down below 1.

What we actually see is different. Megaprojects have official estimates on their benefit-cost ratios in a narrow band: never less than 1 or else they wouldn’t be built, never more than 4 or 5 or else people might disbelieve the numbers. In an environment of stable costs, this would make a lot of sense: all the 10+ projects have been built a long time ago, so the rail extensions on the table today are more marginal. But in an environment of rapid cost escalation, the fact that benefits seem to grow with the costs is not consistent with any honest explanation. The best explanation for this is that, desperate for money for its scheme to build Gateway, Amtrak is defrauding the public about the project’s benefits.

Where is Electrification Warranted?

At the beginning of the month, I published a piece in Voice of San Diego calling for medium-speed rail investment in the Los Angeles-San Diego corridor, centering electrification. This was discussed in a 500-comment thread on California HSR Blog, in which area rail activist Paul Dyson ripped into my plan, arguing (among other things) that electrification is costlier and less useful than I think. Instead of reopening the debate on that particular corridor, I want to discuss a more general set of guidelines to when rail lines should be electrified.

I haven’t said so in these exact words, but I think North American rail authorities and activists underrate electrification. As a result, I find myself persistently prescribing electrification and defending it when it’s already on the table, even as I attack other rail investments as wasteful. On social media and in blog comments I find myself having to constantly explain to people that no, a $20 billion New York regional rail plan should not use dual-mode locomotives but rather spend $250 million on New Jersey-side electrification.

A year and a half ago I wrote about why small, dense countries should fully electrify. The reasons laid out in that post are included in the guidelines below, but there are some additional circumstances justifying electrification.

Narrow stop spacing

Each train has a stop penalty – a total amount of time it loses to making each stop. The penalty is based on dwell time, line speed, and train acceleration and braking performance. If the line speed is 130 km/h, then the penalty excluding dwell time is about 35 seconds for a FLIRT and 80 seconds for a diesel GTW. This 45-second difference per stop is the same if there is a stop every 3 km or if there is a stop every 50 km.

Stop spacing is narrower on commuter lines than on intercity lines, so electrification usually starts from commuter rail. The first mainline electrification in the world was in Paris on the commuter lines serving Gare d’Orsay; subsequently the commuter lines in Paris, London, Tokyo, Berlin, New York, Philadelphia, and other major cities were wired. In many of these cases, commuter rail was electrified decades before intercity mainlines: for example, Japan started electrifying Tokyo’s innermost commuter lines in the 1900s and completed them in the 1920s and early 30s, but took until 1956 to electrify the first intercity line, the Tokaido Line.

However, in some dense regions, even the intercity lines have many stops. Cities in Israel, Belgium, the Netherlands, and Switzerland are just not very far apart, which blurs the distinction between regional and intercity lines somewhat. Switzerland is all-electrified, and my post from 2015 argued that the first three should be, too. In the US, there are specific regions where continuous sprawl has led to the same blurring: the Northeast Corridor, Southern California, Central and South Florida, New England. All are characterized by high population density. New England has closely spaced cities, whereas the LA-San Diego corridor and corridors within Florida have so much sprawl that there have to be several stations per metro area to collect people, reducing stop spacing.

Frequent sharp curves between long straight segments

Electric multiple units (EMUs) can make use of their high acceleration not at stations, but also at slow restrictions due to curves. They are also capable of higher cant deficiency than top-heavy diesel locomotives, since they have low center of gravity, but the difference for non-tilting trains is not so big. A uniformly curvy line does not offer EMUs much advantage, since all trains are slow – if anything, the lower the top speed, the less relevant acceleration is.

The big opportunity to accelerate is then when a mostly straight line is punctured by short, sharp curves. Slowing briefly from 130 km/h to 70 km/h and then speeding back up costs a FLIRT on the order of 15 seconds. A diesel train, whether powered by a locomotive or by diesel multiple units (DMUs), can’t hope to have the required power-to-weight ratio for such performance.


EMUs’ better acceleration profile makes them better-suited for climbing hills and mountains. Modern EMUs, especially low- and medium-speed ones optimized for high acceleration, can effortlessly climb 4% grades, at which point DMUs strain and diesel locomotives require helper engines. When the terrain is so mountainous that tunnels are unavoidable, electric trains do not require ventilation in their tunnels. As a result, some long rail tunnels were electrified from the start. The combination of uphill climbs and tunnels is literally toxic with diesels.

Cheap, clean electricity

Electrification has lower operating costs and lower greenhouse gas emissions in areas where the electricity is powered by cheap hydro or geothermal power than in areas where it is powered by fossil fuels. Switzerland became the only country with 100% rail electrification because it had extensive hydro power in the middle of the 20th century and was worried about relying on coal shipments from Nazi Germany during the war.

This is especially useful in far northern countries, like Sweden and Canada, which have low population density and little evaporation, leading to extensive hydro potential per capita. Despite its low density, Sweden has electrified about two thirds of its rail network. In the US, this is the most relevant to the Pacific Northwest.

But in the future, the falling cost of solar power means that clean electricity is becoming more affordable, fast. This favors electrification in more places, starting from sunny regions like most of the US.

Small installed diesel base

A rich or middle-income country building railroads for the first time, or expanding a small system, needs to build new yards, train maintenance crews, and procure spare parts. It should consider electrifying from the start in order to leapfrog diesel technology, in the same manner many developing countries today leapfrog obsolete technologies like landline phones. In contrast, a larger installed base means electrification has to clear a higher bar to be successful, which is why Japan, France, and other major core networks do not fully electrify.

The US situation is dicey in that it does have a lot of diesel equipment. However, this equipment is substandard: reliability is low, with mean distance between failures (MDBF) of about 45,000 km on the LIRR compared with 680,000 on new EMUs (source, pp. 30-31); the trains are very heavy, due to past FRA regulations; and the equipment is almost universally diesel locomotives rather than DMUs, which makes the acceleration problem even worse than it is for GTWs. Total acceleration and deceleration penalty on American diesel locomotives is not 80 seconds but 2-2.5 minutes.

Because North America underrates electrification, some people who self-identify as forward-thinking propose DMUs. Those require new maintenance regimes and facilities, creating an entire installed base from scratch instead of moving forward to EMUs.

Globally, the installed diesel base for high-performance lines is vanishingly small. The technology exists to run diesel trains at more than 200 km/h, but it’s limited in scope and the market for it is thin.

Through-service to electric lines

Whenever a diesel line is planned to run through to an electric line, it should be a prime candidate for electrification. Dual-mode locomotives exist, but are heavy and expensive; dual-mode multiple units are lighter, but are still boutique products.

This is especially true for the two biggest investments a network can make in passenger rail: RER tunnels, and HSR. RER tunnels involve expensive urban tunneling. When a kilometer of urban subway costs $250 million and a kilometer of catenary costs $2 million, the economics of the latter become stronger. Not to mention that RERs are typically short-hop commuter rail, with frequent stops even on the branches. HSR is a different beast, since it’s intercity, but the equipment is entirely electric. Running through to a diesel branch means towing the train behind a diesel locomotive, which means the expensive HSR traction equipment is idle for long periods of time while towed; this is at best an interim solution while the connecting legacy line is wired, as in the line to Sables d’Olonne.

Nearly complete electrification

Areas where the rail network is almost completely electrified benefit from finishing the job, even if individually the diesel lines are marginal candidates for electrification. This is because in such areas, there is a very large installed electric base, and a smaller diesel base. In small countries the remaining diesel base is small, and there are efficiencies to be had from getting rid of it entirely. This is why the Netherlands and Belgium should finish electrification, and so should Denmark and Israel, which are electrifying their main lines.

This is somewhat less applicable to larger countries, such as Sweden, Poland, and especially Japan. However, India is aggressively electrifying its rail network and planning even more. Note that since networks electrify their highest-trafficked lines first, the traffic can be almost completely electrified even if the trackage is not. For example, Russia is about 50% electrified, but 86% of freight tonnage is carried on electric trains, and the share of ton-km is likely higher since the Trans-Siberian Railway is electrified.

This also applies to networks smaller than an entire country. Commuter rail systems that are mostly electrified, such as the LIRR, should complete electrification for the same reason that mostly electrified countries should. In New England and Southern California, regional rail electrification is desirable purely because of the acceleration potential, and this also makes full electrification desirable, on the principle that a large majority of those two regions’ networks have enough potential traffic to justifying being wired without considering network effects.


Every place – a country, an isolated state or province, a commuter rail system – that is at least 50-60% electrified should consider fully electrifying. The majority of the world that is below that threshold should still wire the most important lines, especially regional lines. Capital-centric countries like Britain and France often get this wrong and focus on the intercity lines serving the capital, but there are low-hanging fruit in the provincial cities. For example, the commuter rail networks in Marseille, Lyon, and Bordeaux are almost entirely electrified, but have a few diesel lines; those should be wired.

In North America, electrification is especially underrated. Entire commuter rail networks – the MBTA, Metra, Metrolink, MARC/VRE, GO Transit, AMT, tails on the New York systems – need to be wired. This is also true of short-range intercity lines, including LA-San Diego, Chicago-Milwaukee, Boston-Portland, Toronto-Niagara Falls, and future New York-Scranton. It is important that good transit activists in those regions push back and support rail electrification, explaining its extensive benefits in terms of reliability and performance and its low installation cost.

What the RER A vs. C Contrast Means for New York Regional Rail

A few weeks ago, I published a piece in City Metric contrasting two ways of through-running regional rail, which I identify with the RER A and C in Paris. The RER C (or Thameslink) way is to minimally connect two stub-end terminals pointing in opposite directions. The RER A (or Crossrail) way is to build long city-center tunnels based on urban service demand but then connect to legacy commuter lines to go into the suburbs. Crossrail and the RER A are the two most expensive rail tunnels ever built outside New York, but the result is coherent east-west regional lines, whereas the RER C is considerably more awkward. In this post I’d like to explain what this means for New York.

As I said in the City Metric piece, the current plans for through-running in New York are strictly RER C-style. There’s an RPA project called Crossrail New York-New Jersey, but the only thing it shared with Crossrail is the name. The plan involves new Hudson tunnels, but service would still use the Northeast Corridor and LIRR as they are (with an obligatory JFK connection to get the politicians interested). I alluded in the piece to RER A-like improvements that can be done in New York, but here I want to go into more detail into what the region should do.

Regional rail to Lower Manhattan

Regional rail in New York should serve not just Midtown but also Lower Manhattan. Owing to Lower Manhattan’s intense development in the early 20th century already, no full-size train stations were built there in the era of great urban stations. It got ample subway infrastructure, including by the Hudson Tubes (now PATH), but nothing that could be turned into regional rail. Therefore, regional rail plans today, which try to avoid tunneling, ignore Lower Manhattan entirely.

The Institute for Rational Urban Mobility, longtime opponent of the original ARC project and supporter of through-running, even calls for new tunnels between Hoboken and Midtown, and not between Hoboken and Lower Manhattan. I went to an IRUM meeting in 2009 or 2010, when Chris Christie had just gotten elected and it was not clear what he’d do about ARC, and when people pitched the idea, I asked why not go Hoboken-Lower Manhattan. The reply was that it was beyond the scope of “must connect to Penn Station” and at any rate Lower Manhattan wasn’t important.

In reality, while Midtown is indeed a bigger business district than Lower Manhattan, the job density in Lower Manhattan is still very high: 320,000 people working south of Worth Street in 1.9 km^2, compared with 800,000 in 4 km^2 in Midtown. Nothing in Ile-de-France is this dense – La Defense has 180,000 jobs and is said to have “over 800 jobs/ha” (link, PDF-p. 20), and it’s important enough that the RER A was built specifically to serve it and SNCF is planning a TGV station there.

Regional trains to Lower Manhattan are compelled to be more RER A-style. More tunnels are needed than at Penn Station, and the most logical lines to connect create long urban trunks. In a post from two years ago, I consistently numbered the regional lines in New York 1-5 with a non-through-running line 6:

  1. The legacy Northeast Corridor plus the Port Washington Branch, via the existing Hudson tunnels.
  2. More lines in New Jersey (some Northeast Corridor, some Morris and Essex) going to the New Haven Line via new Hudson tunnels and Grand Central.
  3. Some North Side LIRR lines (presumably just Hempstead and the Central Branch) to the Hudson Line via Penn Station and the Empire Connection; some LIRR trains should terminate at Penn Station, since the Hudson Line can’t support as much traffic.
  4. The Harlem Line connecting to the Staten Island Railway via Lower Manhattan and a Staten Island-Manhattan tunnel, the most controversial piece of the plan judging by comments.
  5. The New Jersey lines inherited from the Erie Railroad (including the Northern Branch) to the South Side LIRR (to Far Rockaway, Long Beach, and Babylon) via Lower Manhattan.
  6. More North Side LIRR lines (probably the Ronkonkoma and Port Jefferson branches) to Grand Central via East Side Access.

The Lower Manhattan lines, numbered 4 and 5, have long trunks. Line 4 is a basic north-south regional line; it’s possible some trains should branch to the Hudson Line, but most would stay on the Harlem Line, and it’s equally possible that the Hudson Line trains to Grand Central should all use line 2. Either configuration creates very high all-day frequency between White Plains and St. George, and still high frequency to both Staten Island branches, with many intermediate stations, including urban stops. Line 5 goes northwest-southeast, and has to have, at a minimum, stops at Pavonia, Lower Manhattan, Downtown Brooklyn, and then all the LIRR Atlantic Branch stops to and beyond Jamaica.

More stops within new tunnels

Even new tunnels to Midtown can be built with the RER A concept in mind. This means more stations, for good connections to existing subway and bus lines. This is not superficially obvious from the maps of the RER A and C: if anything, the RER C has more closely-spaced stops within Paris proper, while the RER A happily expresses from La Defense to Etoile and beyond, and completely misses Metro 5 and 8. Crossrail similarly isn’t going to have a transfer to every Underground line – it’s going to miss the Victoria and Piccadilly lines, since connecting to them would have required it to make every Central line stop in the center of London, slowing it down too much.

However, the important feature of the RER A is the construction of new stations in the new tunnels – six of them, from La Defense to Nation. The RER C was built without any new stations, except (later) infill at Saint-Michel, for the transfer to the RER B. The RER C’s urban stations are all inherited legacy stations, even when underground (as some on the Petite Ceinture branch to Pontoise are), since the line was built relatively cheaply, without the RER A’s caverns. This is why in my City Metric piece, I refer to the RER B as a hybrid of the RER A and C approaches: it is a coherent north-south line, but every station except Saint-Michel is a legacy station (Chatelet-Les Halles is shared with the RER A, Gare du Nord is an existing station with new underground platforms).

With this in mind, there are several locations where new regional rail tunnels in New York could have new stations. I wrote two years ago about Bergenline Avenue, within the new Hudson tunnels. The avenue hosts very high bus and jitney frequency, and today Manhattan-bound commuters have to go through Port Authority, an obsolete structure with poor passenger experience.

Several more locations can be identified. Union Square for line 4 has been on the map since my first post on the subject. More stations on line 5 depend on the alignment; my assumption is that it should go via the approach tracks to the Erie’s Pavonia terminal, but if it goes via Hoboken then there should be a station in the Village close to West 4th Street, whereas if it goes via Exchange Place then there should be a station at Journal Square, which is PATH’s busiest New Jersey station.

On lines 4 and 5, there are a few additional locations where a station should be considered, but where there are strong arguments against, on the grounds of speed and construction cost: Brooklyn Heights, Chinatown (on line 5 via Erie, not 4), a second Lower Manhattan station on line 4 near South Ferry (especially if the main Lower Manhattan station is at City Hall rather than Fulton Street).

There are also good locations for more stations on the Metro-North Penn Station Access routes, both the New Haven Line (given to line 1) and the Hudson Line (given to line 3). Current plans for Penn Station Access for the New Haven Line have four stations in the Bronx, but no connection to Astoria, and a poor connection to the Bx12 buses on Fordham Road. A stop on Pelham Parkway would give a stronger connection to the Bx12 than the Coop City station, which the Bx12 reaches via a circuitous route passing through the 6 train’s northern terminus at Pelham Bay Parkway. Astoria has been studied and rejected on two grounds: one is construction difficulties, coming from the constrained location and the grade; the other is low projected ridership, since current plans involve premium fares, no fare integration with the subway and buses, and low off-peak frequency. The first problem may still be unsolvable, but the second problem is entirely the result of poor industry practices.

On the Empire Connection, there are plans for stops at West 62nd and West 125th Street. It is difficult to add more useful stations, since the line is buried under Riverside Park, far from Upper West Side and Washington Heights development. The 125th Street valley is one of few places where urban development reaches as far west as the Empire Connection. That said, Inwood is low-lying and it’s possible to add a station at Dyckman Street. In between, the only semi-plausible locations are 145th Street or 155th-158th (not both, they’re too close), and even those are marginal. All of these neighborhoods, from West Harlem north, have low incomes and long commutes, so if it’s possible to add stations, Metro-North should just do it, and of course make sure to have full fare integration with the subway and buses. The one extra complication is that there are intercity trains on this line and no room for four-tracking, which limits the number of infill stops that can support high frequency (at worst every 10 minutes).

Infill stops on existing lines

The existing regional lines in New York have very wide stop spacing within the city. It’s a general feature of North American commuter rail; I wrote about it 5 years ago in the context of Chicago, where Metra is even more focused on peak suburb-to-CBD commutes than the New York operators. In most North American cities I heartily endorse many infill stops on commuter rail. I have a fantasy map for Los Angeles in which the number of stops on inner commuter rail lines triples.

However, New York is more complicated, because of the express subway lines. In isolation, adding stops to the LIRR west of Jamaica and to Metro-North between Harlem and Grand Central would be a great idea. However, all three lines in question – Metro-North, the LIRR Main Line, and the Atlantic Branch – closely parallel subway lines with express tracks. It’s still possible to boost urban ridership by a little by having a commuter rail stop for each express subway stop, which would mean 86th and 59th Streets in Manhattan and Utica Avenue in Brooklyn, but the benefits are limited. For this reason, my proposed line 4 tunnel from Grand Central down to Lower Manhattan has never had intermediate stations beyond Union Square. For the same reason, while I still think the LIRR should build a Sunnyside Junction station, I do not endorse infill elsewhere on the Main Line.

That said, there are still some good candidates for infill. Between Broadway Junction and Jamaica, the LIRR parallels only a two-track subway line, the J/Z, which is slow, has poor connections to Midtown (it only goes into Lower Manhattan), and doesn’t directly connect Jamaica with Downtown Brooklyn. The strongest location for a stop is Woodhaven Boulevard, which has high bus ridership. Lefferts is also possible – it hosts the Q10 bus, one of the busiest in the borough and the single busiest in the MTA Bus system (most buses are in the New York City Transit bus division instead). It’s 4.7 km from Woodhaven to Broadway Junction, which makes a stop around Logan or Crescent feasible, but the J/Z is much closer to the LIRR west of Crescent Street than east of it, and the A/C are nearby as well.

Another LIRR line that’s not next to a four-track subway is the inner Port Washington Branch. There are no stops between the Mets and Woodside; there used to be several, but because the LIRR had high fares and low frequency, it could not compete once the subway opened, and those stations all closed. There already are plans to restore service to Elmhurst, the last of these stations to be closed, surviving until 1985. If fares and schedules are competitive, more stations are possible, at new rather than old locations: Queens Boulevard with a transfer to a Triboro RX passenger line, and two Corona stops, at Junction Boulevard and 108th Street. Since the Port Washington Branch is short, it’s fine to have more closely-spaced stops, since no outer suburbs would suffer from excessive commutes as a result.

Beyond Jamaica, it’s also possible to add LIRR stops to more neighborhoods. There, the goal is to reduce commute length, which requires both integrated fares (since Southeast Queens is lower middle-class) and more stops. However, the branches are long and the stop spacing is already not as wide as between Jamaica and Broadway Junction. The only really good infill location is Linden Boulevard on the Atlantic Branch; currently there’s only a stop on the Montauk Line, farther east.

In New Jersey, the situation is different. While the stop spacing east of Newark is absurdly long, this is an artifact of development patterns. The only location that doesn’t have a New Jersey Transit commuter rail stop that could even support one is Harrison, which has a PATH station. Additional stations are out of the question without plans for intense transit-oriented development replacing the warehouses that flank the line. A junction between the Northern Branch and line 2, called Tonnelle in my post on The Transport Politic from 2009, is still feasible; another stop, near the HBLR Tonnelle Avenue station, is feasible on the same grounds. But the entire inner Northern Branch passes through hostile land use, so non-junction stations are unlikely to get much ridership without TOD.

West or south of Newark, the land use improves, but the stop spacing is already quite close. Only two additional locations would work, one on the Northeast Corridor near South Street, and one on the Morris and Essex Lines at the Orange Street stop on the Newark Subway. South Newark is dense and used to have a train station, and some area activists have hoped that plans to extend PATH to the airport would come with a South Street stop for additional urban service. At Orange Street the land use isn’t great, since a highway passes directly overhead, but the Newark Subway connection makes a station useful.

Finally, in Manhattan, the East River Tunnels have four tracks, of which Amtrak only needs two. This suggests an infill East Side station for the LIRR. There are strong arguments against this – namely, cost, disruption to existing service, and the fact that East 33rd Street is not really a prime location (the only subway connection there is the 6). On the other hand, it is still far denser than anywhere in Brooklyn and Queens where infill stations are desirable, and the 6’s ridership at 33rd Street is higher than that of the entire Q10 or Bx12.


The RER A and Crossrail are not minimal tunnels connecting two rail terminals. They are true regional subways, and cost accordingly. Extracting maximum ridership from mainline rail in New York requires building more than just short connections like new Hudson tunnels or even a Penn Station-Grand Central connection.

While some cities are blessed with commuter rail infrastructure that allows for coherent through-service with little tunneling (like Boston) or no tunneling at all (like Toronto), New York has its work cut out for it if it wants to serve more of the city than just Jamaica and the eastern Bronx. The good news is that unlike Paris and London, it’s possible to use the existing approaches in Brooklyn and New Jersey. The bad news is that this still involves a total of 30 km of new tunnel, of which only about 7 are at Penn Station. Most of these new tunnels are in difficult locations – underwater, or under the Manhattan CBD – where even a city with reasonable construction costs like Paris could not build for $250 million per km. The RER A’s central segment, from Nation to Auber, was about $750 million/km, adjusted for inflation.

That said, the potential benefits are commensurate with the high expected costs. Entire swaths of the city that today have some of the longest commutes in the United States, such as Staten Island and Eastern Queens, would be put within a reasonable distance of Midtown. St. George would be 6 minutes from Lower Manhattan and perhaps 14 from Grand Central. Siting infill stations to intersect key bus routes like Bergenline, Woodhaven, and Fordham, and making sure fares were integrated, would offer relatively fast connections even in areas far from the rail lines.

The full potential of this system depends on how much TOD is forthcoming. Certainly it is easier to extract high ridership from rapid transit stations that look like Metrotown than from ones that look like typical suburban American commuter rail stops. Unfortunately, New York is one of the most NIMBY major cities in the first world, with low housing growth, and little interest in suburban TOD. Still, at some locations, far from existing residential development, TOD is quite likely. Within the city, there are new plans for TOD at Sunnyside Yards, just not for a train station there.

The biggest potential in the suburbs is at White Plains. Lying near the northern terminus for most line 4 trains, it would have very good transit access to the city and many rich suburbs in between. It’s too far away from Manhattan to be like La Defense (it’s 35 km from Grand Central, La Defense is 9 km from Chatelet-Les Halles), but it could be like Marne-la-Vallee, built in conjunction with the RER A.

Right now, the busiest commuter lines in New York – both halves of the Northeast Corridor and the LIRR Main Line – are practically intercity, with most ridership coming from far out. However, it’s the inner suburbs that have the most potential for additional ridership, and middle suburbs like White Plains, which is at such distance that it’s not really accurate to call it either inner or outer. The upper limit for a two-track linear route with long trains, high demand even in the off-peak hours, and high ridership out of both ends, is around a million riders per weekday; higher ridership than that is possible, but only at the levels of overcrowding typical of Tokyo or Shanghai. Such a figure is not out of the question for New York, where multiple subway lines are at capacity, especially for the more urban lines 4 and 5. Even with this more limited amount of development, very high ridership is quite likely if New York does commuter rail right.

Express Airport Connectors are a Scourge of Public Transit

Earlier this month, Andrew Cuomo unveiled a proposal to spend $10 billion on improvements to JFK Airport, including new terminals, highway expansion, and public transit access. I encourage readers to look at the plan: the section on highways proposes $1.5-2 billion in investment including adding lanes to the Van Wyck Expressway and to on-ramps, and has the cheek to say that this will reduce fuel consumption and greenhouse gas emissions. This while the section on mass transit gives it short shrift, only proposing superficial improvements to the AirTrain; in the unlikely the case that this is built, highway mode share will grow and transit mode share will fall. Put in plainer terms, the environmental case for the plan includes fraud.

However, this is not really the topic of this post. That Andrew Cuomo lies to the voters and doesn’t care about good transportation is by now a dog-bites-man story. Instead, I want to focus a little on a throwaway line in the plan, and more on the Regional Plan Association’s reaction. The throwaway line is that almost every major world airport has a one-seat train ride to city center, and by implication, so should JFK.

As an organization dedicated to environment-friendly public transit, the RPA should have made it very clear it opposes the plan due to its low overall transportation value and its favoring of highways over transit. Instead, the RPA immediately launched a brief detailing possible new airport connectors between JFK and Manhattan. The RPA has a lot of good technical people, and its list of the pros and cons of each option is solid. It correctly notes that using the LIRR and Rockaway Beach Branch would compete for traffic with LIRR trains serving Long Island, although it doesn’t mention associated problems like low frequency. The brief is based on prior RPA proposals, but the timing, just after Cuomo came out with his announcement, suggests an endorsement. There are several intertwined problems here:

There is no no-build option

A good study for public transit should not only consider different alignments and service patterns, but also question whether the project is necessary. The US requires environmental impact statements to include a no build option; European countries require a cost-benefit analysis, and will not fund projects with a benefit/cost ratio under 1.2, because of cost escalation risk.

The RPA study does not question whether a one-seat ride from JFK to Manhattan is necessary or useful. It assumes that it is. Everything else about the study follows from that parameter. Thus, it considers entirely express plans, such as the LIRR option, alongside local options. Everything is subsumed into the question of connecting JFK to Manhattan.

One of the alignments proposed is via the LIRR Atlantic Branch and Second Avenue Subway, which the RPA has long believed should be connected. The brief says that it would be slow because it would have to make many local stops; I’ll add that it would serve Midtown, where nearly all the hotels are, via a circuitous alignment. But with all these stops on the way, shouldn’t this be considered as primarily a new trunk line connecting Eastern Brooklyn with Second Avenue? The question of whether the eastern terminus should be Jamaica or JFK must be subsumed to a study of this specific line, which at any rate is unlikely to offer faster service to JFK than the existing AirTrain-to-E option. After all, the most optimistic ridership projection for a JFK connector is maybe 40,000 users per day, whereas the projection for the full Second Avenue Subway is 500,000. I don’t think a Second Avenue-Atlantic Branch connection is warranted, but if it is, the question of whether to serve JFK at the end is secondary.

Express airport connectors are a fetish

I lived in Stockholm for two years, where I went to the airport exclusively using the Arlanda Express, a premium express link running nonstop between the airport and city center. I imagine many visitors to Stockholm use it, are satisfied, and want to replicate it in their own cities.

Unfortunately, such replications miss something important: any air-rail link must go to the areas that people are likely to want to connect to. For locals who wish to travel to the airport, this means good connections to the local transit network, since they are likely to come from many neighborhoods. Not even a small city like Stockholm worries about providing rich areas like Vasastan and Roslag with a one-seat ride. For visitors, this means a one-seat ride to where the hotels are.

Stockholm is a largely monocentric city, with one city center where everything is. (It has an edge city in Kista, with more skyscrapers than Central Stockholm, but Kista can’t be reasonably connected to the airport). The situation in other cities is more complicated. And yet, express air links prioritize serving a big train station even if it’s poorly connected to the transit network and far from the hotels. Let us consider London and Paris.

In London, the five-star hotels cluster around the West End. Only two are at Paddington Station, and only a few more are an easy walking distance from it. This is where the Heathrow Express and the slower Heathrow mainline trains go. No wonder the Heathrow Express’s mode share, as of 2004, is 9%, whereas other Heathrow connections, mainly the Piccadilly line, total 27% (source, PDF-p. 28). The Piccadilly line beautifully passes through the parts of the West End with the largest concentration of hotels, and last time I was in London, I chose it as my Heathrow connection. Nonetheless, the government chose to build the Heathrow Express.

In Paris, the five-star hotels cluster in the west of the city as well, in the 8th arrondissement. The current airport connection is via the RER B, which offers express service in the off-peak when there’s capacity, but not in the peak, when there isn’t. Even so, it is a local commuter rail service, with good connections to the city transit system, and a two-seat ride to the 8th. Because of slow perceived speeds, the state is planning to build an express connector, originally planned to open in 2015 but since delayed to 2023. The express connector will dump passengers at Gare de l’Est, with no hotels within walking distance, no access to Metro lines serving the hotel clusters (Metro 7 does so peripherally, M4 and M5 not at all), and a long walk to the RER for passengers wishing to connect to longer-range destinations such as parts of the Left Bank.

I bring this up to show that the idea of the express air-rail link is a fetish rather than a transportation project, and by analogy, so is the one-seat ride. There is value in faster service and in minimizing the number of transfers, but express airport connectors attempt both even at the cost of building a line that doesn’t go where people want to go.

Ultimately, Cuomo doesn’t care about good transit

Cuomo has many concerns. The chief one is most likely winning the 2020 presidential primary. He has been running for president since the moment he was elected, and many of his policies – gay marriage, the feuds with Bill de Blasio, the desperate attempt to build shiny infrastructure with his name on it – are best viewed through that lens. To the extent that he is not running for president, he has attempted to cement absolute power within the state. He backed a palace coup in the State Senate that secured a Republican(-ish) majority even though the Democrats won most seats; a Democratic majority would be led by a different faction of the party, one more beholden to Democratic interest groups, and might send Cuomo bills that he would lose political capital if he either signed or vetoed them.

This is why I keep giving him as an example of an autocrat in various posts; here is the major takedown, but see also here. Autocrats are always bad for the areas that they govern, which as two separate implications. The first is that their choice of spending priorities is compromised by the need to expand their own power and glory: even if you believe that New York needs $1.5-2 billion in new highway spending, is the Van Wyck really the best place for it?

The second and worse implication is that it is hard for outside groups to convince autocrats to do better. Autocrats don’t have to listen; if they did, they would be democratic leaders. Cuomo happens to be an anti-transit autocrat, and this means that pro-transit groups in New York need to view him as an obstacle and work to weaken him, rather than to ask him to please consider their plans for an air-rail link.

The difficulty is that, precisely because local- and state-level democracy in the US is so weak, it is difficult for issue-oriented groups to go out and oppose the governor. Planners in Democratic cities are hesitant to attack budget-cutting Republican governors like Charlie Baker and Larry Hogan; attacking Democratic governors like Cuomo is a nonstarter. Nonetheless, the RPA needs to understand that it needs to oppose governments hostile to public transit rather than ask them to improve. When Cuomo proposes a bad transportation project, say “no” and move on to more important things; don’t try to work with him, because nothing good can come of that.

Boston NightBus: Planning Around Timed Connections

Over the last year, several people at the Boston advocacy group TransitMatters have been working on a plan to restore night bus service in the area, which is one of few big US cities with no transit between 1 and 5 am. See here for the original concept, from March of last year. The TransitMatters plan assumes limited financial resources, designing the plan around eight or nine routes, all running on an hourly takt schedule, meeting at one central location for a pulse, currently planned to be Copley Square. This seems fairly standard: in Vancouver, too, the daytime bus grid is replaced with a pulse-based system at night, with 30-minute headways on most lines.

So far, so good. The problem is that after additional work, including checking travel times on Google Maps but also some nighttime test drives, TransitMatters found that the original map would not work with an hourly takt. Hourly service with one vehicle per route requires one-way travel time to be 30 minutes minus turnaround time. Double-length routes, at one hour minus turnaround times, can also fit into the system, with two vehicles, but nothing in Boston is that long. Several of the routes turn out to be just a hair too long, and the plan evolved into one with 75-minute headways, too long and irregular for customers. In meetings with stakeholders, the relevant members of Transit Matters were told as much, that 75 minutes was too low a frequency.

I started doing work on this plan around then. Since I think a clockface schedule is important – especially if there’s money for more buses, because then the headways would be 30 minutes and not an awkward 37.5 minutes – I started to sketch ideas for how to reduce travel time. The revisions center the schedule, fitting route choices around the need for buses to complete the roundtrip in an hour minus two turnaround times; this is what I came up with. Time is saved by avoiding detours, even to relatively major destinations, and by not going as far as would be ideal if there were no need to maintain the takt. Many of the design principles are generally useful for designing takt-based schedules, including for commuter rail and for rail-bus connections.

Schedule padding should be based on expected punctuality

This is a point I’ve made before in talking about LIRR scheduling, where fragile timetabling contributes to high schedule padding. Overall, punctuality depends on the following possible attributes of transit services:

  • Rail is more punctual than buses, and electric service is more punctual than breakdown-prone diesels.
  • Grade-separated transit is more punctual than surface transit.
  • Services are more punctual when there are fewer riders, especially buses, which only stop when riders request it.
  • Surface transit is more punctual if it has dedicated lanes, or if (as on some Vancouver routes) it runs on a street with signal priority over intersecting traffic.
  • Surface transit is more punctual off-peak, especially at night, when there’s no congestion.
  • Transit service is more punctual the shorter the span is: a system that’s only supposed to run for 5 night hours has less room for schedule slips than one that’s supposed to run for 21 daytime hours. (This I credit to Ant6n.)

While NightBus involves surface buses running in shared traffic lanes using on-board fare collection, the expected traffic is so low that travel time is likely to be close to the travel time depicted on Google Maps without traffic, and significant variations are unlikely. This means it’s possible to get away with less schedule padding, even though the plan requires 8 routes to converge at one pulse point. The maximum one-way travel time should be taken to be around 26 minutes. 24 minutes is better, and ideally not all routes should be 26 (they’d wait for one another at the pulse point, so it matters how many routes are near the maximum and not just what the maximum is).

Routes should run as fast as necessary and as far as possible

Sometimes, the optimal routing is already the fastest – for example, maybe it really is optimal to link two nodes with a nonstop route. Usually, it is not: on rapid transit there are intermediate stops, on surface transit there are detours and slower segments when freeways are available. When the schedule is tight, there is a plethora of tradeoffs that must be made about travel time. A detour to a major destination, so important that in isolation it would improve service despite the slowdown for through-passengers, must be weighed against other detours. On fast commuter rail line, where there is a significant stop penalty, the equivalent is the intermediate stop; I discussed this 5 years ago in the context of the Lowell Line. The overall length of the route is also a variable: when possible, the outer end should be as far as possible while maintaining the takt.

In the context of NightBus, I used this rule for all routes:

  • The N17, running parallel to the Red Line to Ashmont, runs straight on Dorchester Avenue, whereas in the original plan it detoured to serve Kane Square; there is no time to detour to Kane Square, so in the revised plan it skips it, and passengers going there would need to walk 500 meters.
  • The N28, running on Washington Street and Blue Hill Avenue, terminates at the future Blue Hill Avenue commuter rail stop, and not the Mattapan trolley stop. At night the trolleys don’t run, so the connection isn’t important, and the few hundred meters cut from the route give the buses 2 crucial minutes with which to make the 26-minute one-way schedule.
  • The N32/39 cannot go on Huntington (N39) and thence to Hyde Park (N32); it can either go on Huntington to less valuable Roslindale or on a route parallel to the Orange Line to Hyde Park. I believe the latter option is better, but this is up for debate.
  • The N57 follows the Green Line B Branch to Boston College (taking 20 minutes), not the 57 into Watertown (which would take about 27); I think this is also the optimal decision independently of the need to make the pulse, but the pulse makes it far better. Note that this means the route would have to use unmarked bus stops, since in the daytime there is no bus paralleling the B Branch.
  • The N1 terminates at Davis Square, without going farther into Cambridge or into Arlington (as N77).
  • The N82 and N110 use Storrow Drive to skip Downtown Boston’s slow streets. The buses run on a pulse, so there is no need for more than one bus to serve the same route – they’d be scheduled to bunch, rather than overlying to provide higher frequency. The N111 to East Boston, Chelsea, and Revere serves Downtown Boston instead. This cuts service from Downtown to Malden and Medford, but Downtown is a 9-5 neighborhood, so there’s less need to connect it in every direction at night.
  • The N111 terminates in central Revere and not in North Revere.

Not all transit services are meant for all social classes

At night, buses go at approximately the same speed as cars, provided cars can’t take freeways. If the cars are carrying multiple passengers, as ride-sharing counterproposals plan to, then they probably can’t take freeways. In theory, this means buses would be for everyone, since they were as fast as taxis. In practice, this is only true for people using one route – diagonal trips are still faster by taxi. But worst, the hourly frequency is brutal. People who can plan their night travel around the schedule would use the bus; so would people who can’t afford taxis. But people in the top two-thirds of the income distribution are unlikely to use NightBus, or any ride-sharing alternative (if ride-sharing can afford more vehicles for higher frequency, so can buses).

What this means is that the service needs to be designed around the needs of low-income riders. As a note of caution, in popular parlance there’s a tendency to conflate low-income riders with other groups, such as elderly riders, and pit their needs against good transit practices like wider stop spacing, off-board fare collection, frequent grids, and so on. Those practices are applicable to everyone, and if they appear to favor middle-class riders, it’s because when the buses are too slow, the middle class drives and the poor keep taking the bus, so faster buses have higher proportions of richer riders.

With that caveat in mind, what I mean when I talk about low-income riders is the distribution of origins and destinations. The various draft plans proposed by Transit Matters members all focused on serving lower-income neighborhoods. This is why it’s not such a problem that the N1 only goes as far as Davis Square: that is the favored quarter of the Boston area, and the areas cut off from service, such as Arlington, are rich enough that few would ride an hourly or even half-hourly bus. Additional decisions made based on this principle include,

  • The N32/N39 route serves Hyde Park and not Roslindale. At equal incomes, I’d probably suggest serving Roslindale, which makes for a shorter route, and allows the route to use the extra time gained to get to Forest Hills via a longer route on Huntington and pass near Longwood. But incomes are not equal: Roxbury is much poorer than Longwood and Jamaica Plain, and Hyde Park is poorer than Roslindale.
  • The N57 serves Boston College, which is middle-income but still poorer than Watertown.
  • The N111 serves Chelsea, and probably would regardless of average incomes, but it could instead go parallel to the Blue Line, serving somewhat less poor and less dense areas.

The schedule’s importance is higher at lower frequency

None of the above principles really matters to a subway with 2-minute peak headways and 4-minute off-peak headways. Some of these subways don’t even run on a fixed schedule: it’s more important to maintain even headways than to have trains come when the nominal schedule says they will.  The point where clockface scheduling starts to become important seems contentious among transit planners. Swiss planners use clockface schedules down to (at highest) 7.5-minute headways, and say that 11-minute headways are a recipe for low ridership. In Vienna and Berlin, timed transfers are offered on the U-Bahn on 5-minute trains. At the opposite end, hourly and even half-hourly services must be designed around a schedule with quick connections, to prevent passengers from having to wait the full headway.

In borderline cases – the 7.5-15 minute range – transfers can be timed, and at the less frequent end some overtakes, but there is no real need to design the rest of the schedule around the headway. The main reason to operate with tight turnarounds is to reduce fleet and crew requirements. Any looseness in the schedule, beyond the minimum required for punctuality and crew comfort, should be thought of as a waste. However, the waste is capped by the overall headway. Concretely, if your favorite transit route takes 31 minutes one-way after factoring in turnaround time and schedule padding, then it needs 2 vehicles to provide hourly service, lying idle half the time; to provide 10-minute service, it needs 7 vehicles, lying idle only 11% of the time. So if frequency is high enough, the route should be designed without regard to turnaround times, because the effect is reduced.

But NightBus is hourly; 30-minute service is aspirational. This means that the schedule is more important than anything else. Even if a single neighborhood feels genuinely screwed over by the decisions made to keep the routes at or under 26 minutes – for example, if Revere and Mattapan prefer service going farther out even at the cost of 70- or 75-minute frequency – good transit activists must think in systemwide terms. Maintaining the hourly takt throughout the service area is more important than North Revere and the last few hundred meters in Mattapan.

Ultimately, buses and trains are not all that different

There are major differences between buses and trains in capital costs, operating costs, reliability, and so on, leading to familiar tradeoffs. Even at medium-size transit systems such as the MBTA, frequent bus networks are convoluted and at times fully gridded, while rapid transit networks are invariably radial at least to some extent,. Buses also can’t consistently use timed transfers at high frequency.

However, there are many similarities, especially with small bus networks, which are designed around a pulse rather than a grid:

  • Public transit works with transfers and central dispatching. This makes it better at pulse-based network than any taxi (including ride-hailing apps) or ride-sharing service.
  • Vehicles are large – not to the same extent of course, but relatively speaking (trains in large cities, buses in small ones or at night). There’s less room for the everywhere-to-everywhere one-seat rides that taxis provide at higher cost. If there’s budget for more service-hours, it’s spent on higher frequency or longer routes and not on adding more one-seat rides.
  • Routes are centrally planned, with decisions made about one area affecting service in other areas. It is not possible for routes to evolve by private spontaneous action except in the thickest markets, far bigger than what small bus networks can support.
  • The importance of the schedule and of timed transfers is proportional to the headway, and inversely proportional to frequency.

This is good news, because it means that the large body of good industry practices for rail planning, inherited from such countries as Switzerland and Japan, can be adapted for buses, and vice versa. I did not invent the principle of running trains as fast as necessary; it’s a Swiss planning principle, which led the country to invest in rail just enough to enable trains to go between Zurich, Basel, and Bern in one hour minus turnaround and transfer time. Nor did I expect, when I started getting involved in Transit Matters, that this would be so helpful in designing a better bus plan.

Second Avenue Subway Phase 2 to Cost $6 Billion

Since the 2015-9 capital plan, the New York MTA had been including the second phase of Second Avenue Subway in its capital plan, without a clear estimate of its projected cost. The rumors said the cost would be about $5 billion. A new media story finally gives an official cost estimate: $6 billion. The total length of the project, from 96th Street and 2nd Avenue to 125th Street and Lexington, is about 2.7 km. At $2.2 billion per km, this sets a new world record for subway construction costs, breaking that of the first phase of the same line, which only cost $1.7 billion per km. See a compendium of past posts here to look how these projects stack up. For people not interested in combing through multiple old posts of mine, the short version is that outside the Anglosphere, subway tunnels typically cost $100-300 million per km, with outliers in both directions, but even inside the Anglosphere, costs are in the mid-to-high hundreds of million per km.

In some way, the high cost of SAS phase 2 is more frustrating than that of phase 1. This is because 1 km of the 2.7 km of route preexists. SAS construction began in the 1970s, but was halted due to New York’s financial crisis. In East Harlem, some actual tunnel segments were dug, roughly between the proposed station locations at 96th, 106th, 116th, and 125th Streets; Wikipedia has a more detailed list. Construction of phase 2 thus involves just the stations, plus a short bored segment under 125th Street to get from Second Avenue to Lexington, for a connection to the 4, 5, and 6 trains.

Not having to build tunnels between the stations is beneficial, not as a cost saver in itself but as a way to reduce station costs. In phase 1, it appears that most costs were associated with the stations themselves; if I remember correctly, the cost breakdown was 25% for each of three new stations, and 25% for the tunnels in between. The reason is that the stations are quite deep, while the tunneling in between is bored, to reduce surface disruption. Deep stations are more expensive because they require more excavation, while tunnel boring costs depend more on soil type and how much infrastructure is in the way than on depth. Counting the extra expense of stations, bored subways cost more per km than cut-and-cover subways, but create less surface disruption away from station sites, which is why this method was chosen for phase 1. In contrast, in phase 2, most construction is stations, which would favor a shallow cut-and-cover solution.

Unfortunately, according to rumors, it appears that the MTA now judges it impossible to use the preexisting tunnels in phase 2. If this is true, then this would explain the higher cost (though it would justify $400 million per km, not $2.2 billion): they’d have to build underneath those tunnels. But if this is true then it suggests severe incompetence in the planning stage, of the kind that should get senior employees fired and consultants blacklisted.

The reason is that Second Avenue Subway was planned as a single line. The Environmental Impact Statement was for the full line, including the proposed construction techniques. The phasing was agreed on by then; there was only enough state money for phase 1. This isn’t an unexpected change of plans. I’d understand if in the 2000s it was found that tunnels from the 1970s were not usable; this happened further south, in phase 4, where a preexisting tunnel under Chrystie Street was found to be difficult to use. But in the 2000s the SAS studies signed off on using the tunnels in Harlem, and what seems to be happening is that phase 1, built according to the specifications of the same study, is too deep for using the tunnels.

At $6 billion, this line shouldn’t be built. I know that it goes to a low-income, underserved neighborhood, one that I’ve attacked New York before for taking years to equip with bike lanes (scroll down to my comments here). But the ridership projection is 100,000 per weekday, and $60,000 per weekday rider is too much. Phase 1, providing an underrated east-west connection and serving a denser neighborhood, is projected to get 200,000, for a projection of around $25,000 per weekday rider, which isn’t terrible, so it’s a justified project even if the costs could be an order of magnitude lower.

Were costs lower, it would be possible to build subways to many more low-income neighborhoods in New York. A 125th Street crosstown line, extending phase 2 of SAS, would provide Harlem with crucial east-west connectivity. Subways under Nostrand and Utica Avenues would serve a mixture of working- and middle-class neighborhoods in Brooklyn. A subway under Northern Boulevard in Queens, connecting to phases 3 and 4 of SAS, would serve one of the poorest parts of Queens. A network of tramways would improve surface transit in the South Bronx. Triboro Line would connect poor areas like the South Bronx and East New York with richer ones like Astoria. New York could achieve a lot, especially for its most vulnerable residents, if it could construct subways affordably.

But in a world in which subways cost $60,000 per weekday rider and $2.2 billion per km, New York cannot extend the subway. If it has money in its budget for investment, it should look into things other than transportation, such as social housing or schools. Or it could not borrow money at all to pay for big projects, and in lieu of the money spent on interest, reduce taxes, or increase ongoing social spending.

Given persistent high costs, I would recommend shelving SAS and future rail extensions in New York, including the Gateway tunnel, until costs can be drastically cut. There’s no shortage of worthy priorities for scarce budget in New York, both city and state. Health care in the US is too expensive by a factor of 2, not 10, and transfer payments have near-100% efficiency no matter what; it’s possible to exhaust the tax capability of a state or city just on these two items. Perhaps the need to compete with other budget priorities would get the MTA to cut waste.