Category: Urban Transit

I Saw a Stampede on the Metro

France won the World Cup. Once the final ended, people all over Paris went out to the streets to celebrate. At Nation I saw impromptu dancing, drivers waving tricolore flags, and car passengers climbing out of their cars to wave their own flags. But the real celebration was elsewhere, on Champs-Elysees in the central business district. This was well covered in the media; the Guardian cites an estimate of one million people going to Champs-Elysees to celebrate, and ESPN reports riots (which I didn’t witness but can easily believe happened given the general conduct I did see) and 110,000 police and gendarmerie officers.

The sidewalks were crowded and it was difficult to move; there were too few street closures, so pedestrians were confined to narrow zones for the most part. But the crowding was worst at the Metro stations, and RATP should learn from this example and do better next time there are large celebrations, perhaps next Bastille Day.

The problem is cascading closures. In London, where the Underground platforms are narrower and have fewer cross-passageways than the Metro platforms here, closures are routine at Bank because often the passageways get dangerously overcrowded. These closures cascade: once Bank is closed to limit crowding, passengers swarm the adjacent stations, such as Moorgate and London Bridge, which are not built to handle the typical Bank crowds, forcing TfL to close them as well.

France won the game around 7 in the evening Paris time. By 8, some stations on Champs-Elysees were closed, and as I sat on my severely delayed Metro Line 1 train, with passengers banging on the train’s walls and ceiling, I heard that they were closing more, ultimately going express from Palais-Royal to Argentine and skipping all the CBD stations, including Etoile. I got off at Argentine, as did practically the entire train. Not designed to handle the crowds of the entire CBD at once, Argentine’s platform was jammed. I spent maybe ten minutes trying to make my way from where I got off to the front end of the platform, where the only exits were, and failed, and at a few points the mass of passengers was such that I thought a stampede was likely. The only reason nobody fell onto the tracks was the platform edge doors, installed during the automation of Line 1.

Trains kept serving the station, dumping more and more people. The only mechanism preventing more passengers from getting on was that the crowding was so intolerable that some people started getting back onto the trains, including eventually me. I couldn’t even get off at the next stop, Porte Maillot – the platform was fine but the train was too crowded – so I got off in the suburbs, at Les Sablons, and walked back east.

Perhaps RATP did eventually close Argentine. But both RATP and the city made crucial mistakes that evening, which they should fix in the future.

First, they should have made the trains free to improve passenger circulation. Paying at the turnstiles takes time. This is especially bad in Paris, where there are separate gates for entry (which are turnstiles) and exit (which are one-way doors), unlike the two-way turnstiles of New York. Moreover, unlike New York, Paris has no large emergency doors that can be opened. All passengers were going in one direction – out – so RATP should have propped the exit doors open to let passengers out more smoothly.

Free transit for special events is routine in Paris. The trains are free around New Year’s, in order to encourage people to take the train rather than add to car traffic and pollution (and perhaps drunk driving). Bastille Day celebrations and any future victory at the World Cup or Euro Cup should be added to the list of free transit events, not to discourage people from driving but to prevent stampedes.

And second, the city should have closed the surrounding area to non-emergency car traffic. Champs-Elysees was closed, but there wasn’t much place to spill over; the side street I took once I tried leaving had a narrow sidewalk, and police cars were parked in a way to restrict people to a constrained exit path. There is no parallel street that can act as a spillover route, and between the Rond-Point and Etoile there is only one crossing street wider than about 25 meters, Avenue George V on the south side (whereas almost all rail alternatives to the Metro Line 1 are on the north side). With narrow side streets, it’s especially important to dedicate space to pedestrians and emergency vehicles and not to cars. This was as far as I can tell not done, making it hard for people to leave the most crowded areas. In contrast, Etoile itself, with twelve avenues radiating from its circle, was not so crowded, as people had escape routes.

World Cup victories are rare enough that cities understandably don’t design their entire layout based on them. But when they do happen, it’s critical to have a plan, and the same is true of other big celebrations, which often occur annually on national days. If passengers are overwhelming the subway, it’s critical to quickly do whatever the agency can to increase throughput at station passageways as well as on the tracks. And if pedestrians are overwhelming the streets above ground, it’s critical to give them more street space, including for entry and exit.

Bus Branching

There are two standard reasons why public transit should limit branching. The first is that it reduces frequency on the branches; this is Jarrett Walker’s reason, and distantly the reason why New York doesn’t interline more than two subway services anywhere except 60th Street Tunnel. The second is that it makes schedules more fragile, first because services have to be scheduled more precisely to alternate among branches, and second because delays on one branch propagate to the others. And yet, rail and bus networks still employ branching, due to benefits including better coverage and focusing frequency where demand is the highest. This is especially common on regional rail, where all services are scheduled and often interact with the mainline network, so the second problem of branching is present no matter what. Metro systems instead have less branching, often because they only serve dense areas so that the main benefits of branching are absent. But what about buses?

I posit that bus branching is more valuable in low-density areas than in high-density areas. If an area only has demand for a bus every 30 minutes, and some farther-out places only have demand for an hourly bus, then it’s fine to branch the route in two. The bus would only be useful with some timed transfers at the inner end – maybe it’s feeding a regional train station with a train every half hour – but the Zurich suburbs have half-hourly clockface schedules with timed bus/rail connections and maintain high mode share for how low their density is.

In the other direction, look at Manhattan specifically. I’ve been looking at its bus network even though I’m only supposed to redesign Brooklyn’s. I’ve mentioned before that my epistemology is that if the presence of factor A makes solution B better, then the absence of factor A should make solution B worse. I noticed that the Brooklyn bus network has very little branching: the only route numbers that branch are the B41 and B38, and the only routes with different numbers that share the majority of their lengths are the B67 and B69 (which reverse-branch). However, Manhattan has extensive branching: the M1/2/3/4 share the Madison and Fifth Avenue one-way pair, and the M101/102/103 share the Third and Lexington one-way pair. Understanding why would be useful even if I only care about Brooklyn: if there is a good reason for Manhattan buses to branch then I should consider adding branching in Brooklyn where appropriate, and even if it’s inappropriate, it’s useful to understand what special circumstances make branching good in Manhattan but not in Brooklyn.

As it is, I don’t believe the branching in Manhattan is useful for Brooklyn. This comes from several reasons, at least one of which implies it’s not really useful for Manhattan either, and by extension for other high-density regions.

Base frequency

You can run a bus that comes every half hour on a schedule, making it possible to interline two hourly routes evenly. With some discipline you can go down to 15 minutes, or possibly even 10: Vancouver runs 12-minute limited buses on 4th Avenue on a clockface schedule with on-board fare collection and shared lanes, but there is signal priority at nearly all intersections and relatively little car traffic since the West Side’s street network is rich in arterial roads and distributes cars across other routes (i.e. Broadway, 12th, and 16th Avenues).

In contrast, it’s not really feasible to run buses on a schedule when they come every 5 minutes. There can be a printed schedule, but buses won’t follow it reliably. Once frequency hits about once every 3 minutes, regular street buses bunch so much that adding more buses doesn’t increase passenger capacity, but even in the 5-10 minute range, schedules are less important than headway management, unless the bus has extensive BRT treatments reducing schedule variance. This means that if a bus comes every 10 minutes and is scheduled on headway management, then branching the route means each branch gets service every 20 minutes scheduled on headway management as well. Few passengers would want to ride such a route. This is the worst region for branching, the 7.5-15 minute range in which branches force passengers to use buses that are both infrequent and irregular.

The highest-frequency routes can branch with less risk. If a 5-minute bus branches in two, then each branch gets 10-minute service, at which point reliable schedules are still desirable but not absolutely necessary. How much service do the Manhattan bus trunks run? In the following scheme, peak means the busiest hour in the morning in the peak direction, and off-peak means the lowest frequency between the morning and afternoon peaks, which is usually around 11 am.

M1: 13 buses per hour peak (8 limited, 5 local), 5 off-peak (all local)
M2: 9 peak, 4 off-peak
M3: 6 peak, 6 off-peak
M4: 12 peak (5 limited, 7 local), 6 off-peak (all local)

M101: 6 peak, 6 off-peak (8 in the busiest off-peak hour, 2-3 pm)
M102: 5 peak, 4 off-peak
M103: 5 peak, 4 off-peak

What we see is that Manhattan branches precisely in the worst frequency range. The buses are frequent enough that it’s not possible to run them on a timetable without either much better segregation from traffic than is feasible (even waving away politics) or massive schedule padding, but they still require passengers in Upper Manhattan to wait 10-15 minutes for their specific branch. One might expect that Bus Time would make it easier on passengers by telling them where the bus is, but no, ridership has actually fallen since apps were introduced (and this fall predates the entry of app-hailed TNCs into the city). It turns out passengers like being able to rely on easily memorable clockface schedules, or else on frequencies so high that they only need to wait 5 minutes, not 15.

The street network

Even one-time visitors to New York notice that the avenues in Manhattan are all one-way. This features prominently in the Manhattan bus network, which employs consistent one-way pairs on First/Second, Third/Lex, Madison/Fifth, and Ninth/Tenth. Moreover, again as every visitor to New York knows, Central Park occupies a large blob of land in the middle, interrupting Sixth and Seventh Avenues.

The upshot is that there are more north-south routes north of 110th Street than south of it. This is roughly the branch point on the three trunks that branch (First/Second only carries the M15). In Harlem, there’s demand for buses on Lenox (i.e. Sixth) and Seventh, both of which are two-way there. There’s also commerce on an interpolating route, Manhattan/St. Nicholas, which is effectively 8.5th Avenue in most of Harlem. Farther west, Ninth/Columbus is no longer a useful through-route north of 110th, but instead Tenth/Amsterdam is two-way, and one of the two buses using the Columbus/Amsterdam one-way pair on the Upper West Side, the M11, indeed goes two-way on Amsterdam north of 110th.

This situation occurs very frequently in cities without gridded street networks. One trunk route will split in two, heading to different former villages that were incorporated into the city as it industrialized and grew. Manhattan is unusual among gridded cities in that its avenues are one-way, forcing buses into one-way pairs south of Harlem that, together with Central Park, ensure there are more useful routes north of 110th than south of it. But among cities without a planned street network this is typical.

As a check, let’s look at the bus networks in two ungridded American cities: Boston and Providence. Do they have a lot of interlining, involving one trunk route splitting in two farther out? Yes, they do!

Here is Providence. Going west of Downcity, there are two major routes to Olneyville, Westminster and Broadway, but beyond Olneyville there are four main streets, so each of the two inner corridors carries two bus routes, and one of these four routes even splits in two farther out. Going north, Charles Street carries four routes, branching off at various locations. Going east there’s a bus tunnel to College Hill carrying many routes, but even outside the tunnel, the one-way pair on Angell and Waterman carries three buses, which split in East Providence. And going south and southwest, Broad Street carries multiple routes, and one of its branches, Elmwood, carries two, splitting farther south.

Here is Boston. Unlike in Providence, buses don’t converge on city center, but on subway stations, so the map is much less clean. However, we see the same pattern of trunk routes splitting into branches. For example, going south of Ruggles, many routes go southeast to Dudley and then south on Warren Street, splitting to various destinations in Dorchester, Mattapan, and Hyde Park on the way. Going southwest of Forest Hills we see many routes use Washington Street, some staying on it and branching in Dedham and some veering west to West Roxbury and branching there. Elsewhere in the system we see the same pattern going north of Maverick and Oak Grove, northeast of Malden, west of Harvard (briefly on Mount Auburn), and northwest of Alewife.

One-seat rides and reverse-branching

I have repeatedly criticized the practice of reverse-branching on subway networks, especially New York, in which two train routes share tracks in an outlying area (such as Queens Boulevard) and then split heading into the center (such as Eighth Avenue on the E versus Sixth Avenue on the F). I did so on the same grounds that any branching is suspect: it reduces frequency on specific routes, and makes the schedule more fragile as delays propagate to more of the network. Moreover, the issue of schedule fragility gets worse if many routes share tracks at some point during their journey, whereas with conventional branching there are only two or three branches per trunk and the trunks form self-contained systems. Finally, reverse-branching lacks the main benefit of conventional branching, as it does not concentrate traffic in the core, where there’s most demand.

These issues are present on bus networks, with two modifications:

  1. The value of one-seat rides is somewhat higher. Transferring between buses is less nice than transferring between subways: in a Dutch study about location decisions, people’s disutility of out-of-vehicle time on buses was 1.5 times as high as on trains.
  2. Buses can overtake each other and, even without overtakes, run much closer together than trains. The limiting factor to capacity on buses is schedule fragility and bunching and not stopping distances. This means that reverse-branching is less likely to lead to cascading delays – buses do not have a 2-minute exclusion zone behind them in which no buses may enter.

This means that reverse-branching is more defensible on buses than on trains. However, even then, I don’t think it’s a good idea. At least in Manhattan, reverse-branching consists of avenues in Upper Manhattan that have buses going to both the East Side and the West Side: the M7 (serving the Ninth/Tenth pair) and the M102 both run on Lenox, and the M4 and M104 (running on Broadway to Midtown) both run on Broadway in Morningside Heights. These splits both reduce the frequency available to bus riders and should be eliminated. East-west service should be provided with high-quality bus routes on the main streets, especially 125th (which needs a full subway) but also 116th, 135th, 145th, and 155th.

The snag is that grids don’t work well unless they are complete. The Manhattan grid isn’t complete through Upper Manhattan, because 116th and 135th are discontinuous, without a direct connection from Central Harlem to Morningside Heights and West Harlem. However, the M7 route duplicates the 2 and 3 trains, so it’s not necessary for east-west connectivity. The M4 route doesn’t duplicate the subway, but does duplicate the M101, which runs on 125th Street and Amsterdam (and isn’t a reverse-branch because the M11 terminates shortly after 125th), so it’s not useful by itself.

Should buses branch?

There is one solid reason for buses to branch: if the street network has more major routes closer to the center than in outlying areas, then buses running on the outer arterials should come together close to the core. This is common enough on cities with haphazard street networks. It may also be reinforced if there are weak circumferential streets (Sydney is one such example). In contrast, cities with gridded street plans, even broken grids like those of Brooklyn and Tel Aviv, should have little to no bus branching.

If a bus does branch, it should ideally be extremely frequent on the trunk, so that even the branches have decent headway-based service. I’m not willing to commit to a maximum headway, but Barcelona and Toronto both have at worst 8-minute headways on their bus grids, so if that is indeed the maximum then a bus shouldn’t branch if its off-peak frequency is worse than every 4 minutes and better than every 10-20 (the more reliable the timetable is, the lower the upper limit is, since it’s possible to run on a timetable at higher frequency). In my case of interest, Brooklyn, there is exactly one bus route that comes at least every 4 minutes off-peak: the B46 on Utica runs 16 buses per hour in each direction, counting both local and limited (SBS) routes.

The area in which buses absolutely should not branch – strong interconnected networks of arterials (not necessarily grids – Paris’s network counts too), running buses every 5-15 minutes off-peak – is exactly where most strong bus networks are. It’s rare to have a bus that has extremely high frequency all day, because in most functional city such a bus would be a subway already; as it is, Utica has long been New York’s second priority for subway service, after Second Avenue. So for the most part, the places where buses are the strongest are precisely those where branching is the most deleterious. Low-frequency networks, perhaps connecting to a suburban train station with a timed transfer, should add bus branching to their planning toolkit, but high-frequency urban networks should not.

Why is Tramlink So Weak?

I’ve mentioned on Twitter that I’m visiting London. I’m taking a lot of railfan trips, one of which was on Tramlink, London’s circumferential light rail service. Tramlink runs in South London, from Wimbledon in the west to Croydon in the east and thence along several branches to southeastern outer neighborhoods. Much of the route uses former mainline rail rights-of-way that were only partly grade-separated. The trains satisfy all of TransitCenter’s principles for good light rail operating practices, but their ridership is lackluster by the standards of Paris, TransitCenter’s comparison city. Tramlink has 30 million annual riders on 28 km of route, or about 3,500 per km per weekday; Ile-de-France’s system had 900,000 daily riders in 2015 on about 100 km route, or 9,000 per km. My goal is to explain why. One reason involves route choice, but the main reason is lack of development; this problem is very common in other cities, and must be added to the other pitfalls that TransitCenter mentions.

The operating practices on Tramlink are not bad. The frequency is high: every 5 minute off-peak. There’s no fare integration with the proper rail network (including the Underground), but the buses in London have no fare integration with the trains either and still have high ridership. The connections with radial train lines are decent, though there’s one big miss (with the Northern line) and one smaller one (with West Croydon, which points to train stations that are served by other lines that do get interchanges); the two most important transfers, Wimbledon and East Croydon, require relatively little walking between platforms. The right-of-way quality is high by light rail standards, mostly in a private right-of-way with only a small extent of street running within Croydon; the average speed is 21 km/h (higher than the Parisian tramways – T3 averages 18 km/h). And yet, ridership is not so strong. London is a big city with high rail ridership, so it’s not a matter of a small city underperforming Paris on raw ridership; something deeper is wrong with Tramlink.

Part of the problem has to involve route layout. East of East Croydon, the route has three branches. Two, heading to Elmers End and Beckenham Junction, keep the route’s circumferential character; in theory it should be faster to take mainline rail and change trains than to ride Tramlink, but in reality the mainline routes that would be used have missed connections and therefore are not useful for diagonal trips. Each of these two branches runs every ten minutes, interlining to a train every five minutes between East Croydon and Wimbledon. However, a third route connects East Croydon and New Addington, a radial line, running every 7.5 minutes. This route does not run through to Wimbledon (which would be a radial-circumferential mix) and exists as an orphaned feeder line, sharing tracks with the two main branches just east of East Croydon (thus, creating schedule conflict due to the uneven frequency on the shared trunk).

But the main difference between Tramlink and the Parisian tramways is adjacent density. London is generally a less dense city than Paris. London has two- and three-story rowhouses with back gardens where Paris has five- to nine-story buildings with high lot coverage. The Tramlink route itself is even less dense, passing through suburbia, industrial sites, and golf courses. The Parisian tramways are all in the suburbs (except for T3), but serve high-density clusters, surrounded by a mixture of mid-rise buildings and social housing towers. This is especially true on the workhorse Parisian routes – T1, T2, and T3, which collectively have about three quarters of the system’s total ridership – but even the other routes, while much weaker than the main three, serve denser areas than Tramlink and get higher ridership per kilometer.

Here is a randomly-selected station on T2, Meudon-sur-Seine:

Compare it with Mitcham, one of the more populated stations on Tramlink between Wimbledon and Croydon:

Also compare both with the site of the missed connection with the Northern line, Morden Road:

I want to make it very clear that the two satellite maps of Mitcham and Morden Road are not representative of all of South London, certainly not when weighting by population. East Croydon is full of mid- and high-rise TOD, and to some extent so is Wimbledon; the two stations rank fifth and sixth in ridership in London excluding the Central London terminals. The problem is that a circumferential line is rarely used over a long stretch. The longer the angle subtended on a circumferential line, the more favorable it is to take the radials and transfer.

London in particular has four-track mainlines on most rail routes, including the London and South Western Main (serving Wimbledon) and the Brighton Main (serving East Croydon), making it easy to run express routes. Every hour, there are 9 trains running nonstop between East Croydon and Clapham Junction, and 16 trains running between Wimbledon and Clapham Junction with two intermediate stops. The diagonal commuter rail trip is faster than Tramlink, even counting transfer time at Clapham Junction.

Paris is full of express trains, represented by the RER. But T3 misses nearly all of the RER connections, which weakens the route but also means that there is no express alternative on the outer margin of Paris; but one would still not take it all the way, especially since there is a forced transfer at Porte de Vincennes. But T1 and T2 have better RER and Transilien connections. The high density all along these routes, and not just at widely-separated key junctions, ensures that there is high demand even on short segments.

In fact, there is circumstantial evidence that T1, T2, and T3 have extensive short-range ridership: their ridership levels per kilometer are very high (respectively 11,000, 12,000, and 15,000 per weekday), and if they had low turnover they would not have capacity for such high ridership. New York has 15,000 weekday subway riders per route-km, and this is with long trains, extensive four-tracking, and higher peak frequency than on the Parisian tramways. It’s hard to imagine comparable ridership levels on a surface tramway without very high turnover, which I have in fact observed riding T3.

In contrast, I saw relatively little turnover between Wimbledon and East Croydon on Tramlink. I saw some, generally involving a small net decrease in passengers on the tram at the first few stations past Wimbledon, but a large proportion of passengers who got on at Wimbledon stayed on until Croydon. To them, the tram is perhaps a slower but cheaper alternative to mainline rail. Some would also ride until one or two stations before East Croydon, within the built-up cluster of Croydon; perhaps their exact destination was closer to one of these tram stations than to East Croydon, where the tram loses a lot of time due to circuitous street running.

Reinforcing the importance of turnover, the tram was crowded. I took it at 4:30 in the afternoon, on the shoulders of rush hour, and it was standing-room only for my entire trip, with considerable crowding among the standees for the first few stations. And yet, despite the crowding, ridership per kilometer is a fraction of that achieved by Paris’s top three tramways, which do not appear more crowded.

I wrote about turnover in the context of Vancouver buses, talking about patterns of development along north-south arterials (Main, Fraser, and Commercial) versus east-west ones (King Edward and 49th). Here we see how it interacts with development on a circumferential tramway within the context of a rapid transit network with fast radial lines. It’s common in a large city to have strong demand for circumferential transit but not so much that full rapid transit is justifiable, leading to tramway networks such as Tramlink and the tramways of Ile-de-France. In this context, it’s important to attract short-hop ridership and not just end-to-end ridership, where the tramway would struggle with the radial rapid transit network. This in turn requires the region to ensure that the intermediate stops generate ample ridership, which requires either uniformly high density (as is the case in and around Paris) or a deliberate effort at TOD in the middle.

This is true for more than just tramways. The fundamental fact about Tramlink and the Ile-de-France tramways is that they are slower than their respective cities’ radial rail networks. The same fundamental fact is true of circumferential buses, even in cities where the radial rail network is light rail rather than rapid transit. In theory this could even happen in an all-bus city, provided the buses’ right-of-way quality were such that the radials were faster than the circumferentials; but in reality this is hard to arrange, since buses get stuck in traffic even when they’re BRT, and there’s more traffic near city center than outside.

Focus on What’s Common to Good Transit Cities, not on Differences

Successful transit cities are not alike. There are large differences in how the most expansive transit networks are laid out. It takes multiple series of posts across several blogs (not just mine but also Human Transit and others) covering just one of them, for example stop spacing or how construction contracts are let. With so much variation, it’s easy to get caught up in details that differentiate the best systems. After all, the deepest communities of railfans tend to sprout in the cities with the largest rail networks; arguing with railfans with experience with London, Tokyo, or Paris is difficult because they know intricate details of how their systems work that I am catching up on but only know in the same depth for New York. Add in the fact that London and Paris view each other as peer cities and from there the route to arguing minutiae about two cities that by most standards have good public transit is short.

But what if this is wrong? What if, instead of or in addition to figuring out differences among the top transit cities, it’s useful to also figure out what these transit cities have in common that differentiates them from auto-oriented cities? After all, in other aspects of development or best practices this is well-understood: for example, a developing country can choose to aim to be hyper-capitalist like Singapore or the US or social democratic like Sweden or France, but it had better develop the institutions that those four countries have in common that differentiate them from the third world.

Unfortunately, before discussing what the common institutions to transit cities are, it’s necessary to discuss things that may be common but don’t really matter.

The US as a confounding factor

The biggest problem with figuring out things all good transit cities have in common is that in the developed world, the US (and to some extent Canada and Australia) is unique in having bad transit. Frequent commenter Threestationsquare has a list of cities by annual rapid transit ridership (counting BRT but not infrequent commuter rail, which lowballs parts of the US); New York is near the top, but the second highest in the US, a near-tie between Boston, Chicago, and Washington, would rank #22 in Europe. As a result, some social, political, and technical features that appear to differentiate good and bad transit are not really about transit but about the US and must be discarded as confounding factors. Fortunately, most of these confounding factors are easy to dispose of since they also occur in New York.

The more difficult question concerns factors that are distantly related to the weakness of US transit but are not direct explanations. I wrote about racism as such a factor a few months ago, arguing that high US construction costs come from weak civil service, which in turn comes from the way American segregation works. The US is not uniquely racist or even uniquely segregated; the unique aspect is that it a) has a long-settled oppressed minority and not just immigrants who arrived after the characteristic of the state was established, and b) has segregation within metro areas (unlike Singapore, which has social but not spatial segregation) but not between them (unlike Israel, where the built-up area of Tel Aviv has very few Arabs). But while this can explain why institutions developed in a way that’s hostile to transit, it’s not a direct explanation for poor US transit except in Atlanta, where the white state underinvests in the black city. White people in Boston, Los Angeles, Houston, and other cities with little to no public transit do not avoid the bus or the train out of stereotypes that match typical American racial stereotypes, such as crime; they avoid the train because it doesn’t go where they’re going and the bus because it is slow and unreliable.

There are two ways to avoid confounding factors. The first is the sanity check, where available: if some feature of transit exists across major transit cities but is absent in auto-oriented cities not just in the US but also in Canada, Australia, New Zealand, Israel, and Italy, then it’s likely to be relevant. Unfortunately, clean examples are rare. The second and more difficult method is to have theoretical understanding of what matters.

Size artifacts

London and Paris are transit cities. So are Prague and Stockholm. I’ve stressed the importance of scale-variance before: features that work in larger cities may fail in smaller ones and vice versa. Thus, it’s best to look at common features of successful transit cities within each size class separately.

In fact, one way cities can fail is by adopting transit features from cities of the wrong size class. China is making the mistake in one direction: Beijing and Shanghai have no express subway trains or frequent regional rail services acting as express urban rail, and as a result, all urban travel has to slow down to an average speed of about 35 km/h, whereas Tokyo has express regional lines averaging 60 km/h. China’s subway design standards worked well for how big its cities were when those standards were developed from the 1970s to the 1990s, but are too small for the country’s megacities today.

In contrast, in the developed world, the megacities with good public transit all have frequent express trains: Tokyo and Osaka have four-track (or even eight-track!) regional lines, Paris has the RER, New York has express subways (and the premium-price LIRR trains from Jamaica to Penn Station), London has fast regional rail lines and Thameslink and will soon have Crossrail, Seoul has a regional rail network with express trains on Subway Line 1, and Moscow stands alone with a strictly two-track system but has such wide stop spacing that the average speed on the Metro is 41 km/h. Smaller transit cities sometimes have frequent express trains (e.g. Zurich and Stockholm) and sometimes don’t (e.g. Prague), but it’s less important for them because their urban extent is such that a two-track subway line can connect the center with the edge of the built-up area in a reasonable amount of time.

And if China failed by adopting design standards fitting smaller cities than it has today, the US fails in the other direction, by adopting design standards fitting huge megacities, i.e. New York. Small cities cannot hope to have lines with the crowding levels of the Lexington Avenue Line. This has several implications. First, they need to scale their operating costs down, by using proof of payment ticketing and unstaffed stations, which features are common to most European transit cities below London and Paris’s size class. Second, they need to worry about train frequency, since it’s easy to get to the point where the frequency that matches some crowding guideline is so low that it discourages riders. And third, they need to maximize network effects, since there isn’t room for several competing operations, which means ensuring buses and trains work together and do not split the market between them.

The best example of an American city that fails in all three aspects above is Washington. While railfans in Washington lament the lack of express tracks like those of New York, the city’s problems are the exact opposite: it copied aspects of New York that only succeed in a dense megacity. With interlining and reverse-branching, Washington has low frequency on each service, down to 12 minutes off-peak. The stations are staffed and faregated, raising operating costs. And there is no fare integration between Metro and the buses, splitting the market in areas with price-sensitive riders (i.e. poor people) like Anacostia.

The political situation

While I’ve written before about what I think good metro design standards are, these standards themselves cannot separate the major transit cities from cities like Los Angeles (which has about two and a half rail trunks in a metro area larger than that of London or Paris) or Tel Aviv (which has no metro at all). Instead, it’s worth asking why these cities have no large subway systems to begin with.

In the case of Tel Aviv, Israel has had an official policy of population dispersal since independence. After independence the North and South of the country had Arab majorities, and the government wished to encourage Jews to settle there to weaken any Palestinian claims to these areas. As a result, Prime Minister David Ben Gurion rejected a plan to develop an urban rail network centered on Tel Aviv and instead encouraged low-income Jewish immigrants to move far away, either to depopulated Arab towns or to new towns (“development towns”) built at strategic points for national geopolitics. Decentralization was national policy, and with it came auto-oriented urbanism. A less harsh but equally politicized environment led to Malaysia’s auto-centric layout: Paul Barter’s thesis outlines how Malaysia choked informal transit and encouraged auto-oriented suburbanization in order to create an internal market for state-owned automakers.

In the case of the US, the situation is more complex, since there were several distinct political trends in different eras favoring cars. In postwar suburbia (and in Los Angeles going back to the 1920s) it was the association of cars with middle-class normality, and in California also with freedom from hated railroads; it’s related to the fact that American suburbanization was led by the middle class rather than by the working class as with more recent exurbanization. In Israel suburbanization was led by the working class, but the deliberate government policy of decentralization meant that the urban middle class’s demands for better transportation were ignored until the 1990s.

Without enough of an urban middle class to advocate for more transit, US transit withered. New cities in the Sunbelt had little demand for public transit, and in the older cities the middle class cared little for any transit that wasn’t a peak-only commuter train from the suburbs to the CBD. Moreover, in existing transit cities the middle class demanded that the urban layout change to fit its suburban living situation, leading to extensive job sprawl into office parks that are difficult to serve on transit. This paralleled trends in Canada, Australia, and New Zealand; Sydney in particular saw middle-class suburbanization early, like Los Angeles.

The political situation changed in the 1970s, 80s, and 90s, but by then high construction costs, NIMBYism constraining the extent of TOD (unlike in Canada), and indifference to leveraging regional rail for urban transit (as in Canada and until recently Israel but unlike in Australia) made it difficult to build more public transit lines.

Regional rail and TOD

The largest transit cities in the rich and middle-income world all make extensive use of regional rail, with the aforementioned exception of Chinese cities, where the lack of regional rail is creating serious travel pain, and New York, where the city itself is transit-oriented but its suburbs are not. Smaller transit cities usually make use of regional rail as well, but this isn’t universal, and to my understanding is uncommon in Eastern Europe (e.g. Kyiv has one semi-frequent ring line) even in cities with very high metro and tramway usage.

However, smaller transit cities that do not have much regional rail have full metro systems and not just tramways, let alone BRT. Curitiba and Bogota are famous for their BRT-only transit networks, but both instituted their systems in a context with low labor costs and both are building metro systems right now.

The other common element to transit cities is TOD. Here, we must distinguish old cities like London, Paris, Berlin, and Vienna, whose urban layout is TOD because it was laid out decades before mass motorization, and newer cities like Stockholm, Tokyo, and every city in Eastern Europe or the East Asian tiger states. The latter set of cities built housing on top of train stations, often public housing (as in the communist world or in Stockholm) but not always (as in Tokyo and to some extent Hong Kong), in an era when the global symbol of prosperity was still the American car-owning middle class.

The importance of TOD grows if we compare countries with relatively similar histories, namely, the US and Canada. Neither country does much regional rail, both have had extensive middle-class suburbanization (though Canada’s major cities have maintained bigger inner-urban middle classes than the US’s), and English Canada’s cities came into the 1970s with low urban density. The difference is that Canada has engaged in far more TOD. Calgary built up a large CBD for how small the city is, without much parking; Vancouver built up Downtown as well as transit-oriented centers such as Metrotown, New Westminster, Lougheed, and Whalley, all on top of the Expo Line. Nowhere in the US did such TOD happen. Moreover, American examples of partial TOD, including Arlington on top of the Washington Metro and this decade’s fast growth in Seattle, have led to somewhat less awful transit usage than in the rest of the country.

Most cities in the developed world are replete with legacy rail networks that can be leveraged for high-quality public transit. We see cities that aim at transit revival start with regional rail modernization, including Auckland and to some extent Tel Aviv (which is electrifying its rail network and building new commuter lines, but they run in freeway medians due to poor planning). Moreover, we see cities that are interested in transit build up high-rise CBDs in their centers and high- and mid-rise residential development near outlying train stations.

“Regional rail and TOD” is not a perfect formula; it elides a lot of details and a lot of historical factors that are hard to replicate. But both regional rail and TOD have been major elements in the construction of transit cities over the last 60 years, and while they both have exceptions, they don’t have many exceptions. In the other direction, I don’t know of examples of failed TOD – that is, of auto-oriented cities that aggressively built TOD on top of new or existing rail lines but didn’t manage to grow their transit ridership. I do know some examples of failed regional rail, but usually they make glaring mistakes in design standards, especially frequency but also station siting and fare integration.

At a closer in level of zoom, it’s worthwhile to talk about the unique features of each transit city. But when looking at the big picture, it’s better to talk about what all transit cities of a particular size class have in common that auto-oriented cities don’t. Only this way can an auto-oriented city figure out what it absolutely must do if it wants to have better public transit and what are just tools in its kit for achieving that goal.

Reverse-Branching Does not Save You the Transfer

I wrote a detailed proposal about why New York should deinterline, and how. I got a lot of supportive comments (in the transit blogging sense, i.e. nitpicking), but also some pushback, arguing that people like their one-seat rides, and making them transfer under a more coherent system would make their riding experience worse. I could go on about how London is facing the same problem and is choosing to invest a lot of money into deinterlining in order to increase train capacity, but in the case of New York, there’s a blunter answer: what one-seat ride? The extent of reverse-branching on the subway does not really give people one-seat rides, and New York City Transit is making service decisions that do not maximize one-seat rides even when doing so would be relatively painless.

Outer branches

Most outer branches with just one route naturally offer direct service to the route’s trunk line. Let’s look at the current subway_map, and compare it with my proposed deinterlining, which is again this:

Today, riders on the West End Line only have service on the D, so they only have a one-seat ride to Sixth Avenue. Riders on the Sea Beach Line only have the N, and riders on the local Brighton Line trains only have the Q, so they only have one-seat rides to the Broadway express trains, and if they want to travel to Prince Street or 8th Street-NYU on the R they have to change trains at Canal, which is not a cross-platform transfer. Only a handful of stations get genuine choice between the two trunk lines: 36th Street on the D and N, and the inner few express stops on the B and Q, say up to Newkirk Avenue. These are express stops, with more ridership than the locals, but they’re not the majority of ridership on the subway in Southern Brooklyn. The majority of riders have to deal with the drawbacks of both reverse-branching (slow, infrequent trains) and coherent service (fewer one-seat rides).

Queens Boulevard has the same situation: local and express patterns mix up in a way that makes the choice of one-seat rides much weaker than it appears on the map. Riders at the local stations can choose between the M and the R, two trains that are never more than a few blocks apart in Midtown; only one station on either line is inconvenient to access from the other, 57th Street/7th Avenue, the least busy stop on the Broadway Line in Midtown on a passengers per platform basis (49th and 5th have less ridership but have two platform tracks and no Q service). The express stops get more serious choice, between the E and F, but those are just three stations: Jackson Heights-Roosevelt Avenue, Forest Hills-71st Avenue, and Kew Gardens-Union Turnpike. Queens Plaza has E, M, and R service, but passengers actually getting on at Queens Plaza can equally get on at Queensboro Plaza and ride the N, W, or 7.

Genuine choice between two relatively widely-separated trunk lines on the same trunk only exists in two and a half places in New York: the Central Park West Line offers a choice between the B and C trains, the Nostrand Avenue Line offers a choice between the 2 and 5 trains, and the inner half of the White Plains Line offers a choice between the 2 and 5 trains off-peak (at the peak the 5 runs express, so local stations only get the 2).

Cross-platform transfers

New York is blessed with cross-platform interchanges, usually between local and express trains on the same line. Riders on the 1 train are used to transferring to the 2 and 3 trains cross-platform at 96th Street; in the morning, the 1 train’s busiest point is actually from 103th Street to 96th, and not heading into Midtown. With 170,000 boardings at its stations north of 96th per weekday, the 1 is much busier than Nostrand (with 60,000 weekday boardings) or the combined total of local Central Park West stations from 72nd to 116th (with 65,000 boardings). It’s also slightly busier than the White Plains Road Line, let alone the inner segment with both 2 and 5 service (which has 95,000 boardings).

In Queens, a similar situation occurs on the 7. The stations east of Queensboro Plaza, excluding 74th Street-Broadway (where the transfer to the Queens Boulevard Line is), have a total of 215,000 weekday boardings. The trains fill at the outer end and then discharge at 74th Street as most passengers transfer, not cross-platform, to the faster Queens Boulevard Line; then they fill again at the stations to the west and discharge at Queensboro Plaza, which has a cross-platform transfer to the N and W.

This is relevant to some of the few segments of the subway where reverse-branching offers choice between different trunk lines. Passengers on the Nostrand Avenue Line could transfer cross-platform at Franklin Avenue, where the platforms aren’t much narrower than at 96th Street and Broadway, where passenger volumes are almost three times as high. Similarly, passengers on the Central Park West Line and its branches to Washington Heights and Grand Concourse could transfer cross-platform at 125th Street or at Columbus Circle; Columbus Circle is extremely busy already with origin-and-destination traffic, and the interchanges between local and express passengers could not possibly overwhelm it.

Only one place has a difficult connection: 149th Street-Grand Concourse, the interchange between the 2, 4, and 5 trains. This also happens to be the most difficult deinterlining project in general, because of the merger of the 2 and 3 further south; it requires either closing the northernmost two stations on the 3, or opening up a few blocks of Lenox Avenue to construct a pocket track. Because of the disruption involved, this project can be left for last, and come equipped with more passageways at 149th Street, just as London is first deinterlining the Northern line to the south (raising peak capacity on the Bank branch from 26 trains per hour to 32) and leaving the north for later (which would raise capacity further to 36 tph).

NYCT has deinterlined in the past

Upper Manhattan witnessed two deinterlinings in the second half of the 20th century, one in the 1950s and another in the 1990s. The service NYCT inherited from its three predecessor networks had systematic route nomenclature taking into account conventional and reverse branching.

On the IRT, West Side trains were numbered 1 (to Van Cortlandt Park), 2 (to the White Plains Road Line), and 3 (to Harlem-148th Street), and Lexington trains were numbered 4 (to the Jerome Avenue Line), 5 (to the White Plains Road Line), and 6 (to the Pelham Line); 2, 4, and 5 trains ran express, 3 and 6 trains ran local, and 1 trains could be either local or express. In the 1950s, NYCT changed this system on the West Side so that all 1 trains became local and all 3 trains became express. This was the result of track layout: the junction at 96th Street is flat if 3 trains have to cross over to the local tracks and 1 trains have to cross over to the express tracks, but under today’s present service pattern there are no at-grade conflicts. NYCT chose capacity and reliability over offering one-seat rides from West Harlem and Washington Heights to the express tracks.

On the IND, trains were identified by letters. A, C, and E trains ran on Eighth Avenue and B, D, and F trains on Sixth Avenue; A and B trains went to Washington Heights, C and D trains to Grand Concourse, and E and F trains to the Queens Boulevard Line. Local and express trains were identified using letter doubling: a single letter denoted an express train, a doubled one (e.g. AA) a local. The single vs. double letter system ended up discontinued as few trains consistently run express (just the A and D) and several run a combination of local and express (the B, E, F, N, and Q), and NYCT slowly consolidated the trains on Eighth and Sixth Avenue until there were only seven services between them. Eventually the B and C switched northern terminals, so that now the C runs as the local version of the A and the B as something like the local version of the D. Passengers in Washington Heights who wish to use Sixth Avenue Line have to transfer.

The situation on the IND wasn’t as clean as the deinterlining on the IRT. But it shows two important things. First, changes in train service have made the original reverse-branching less tenable from an operational perspective. And second, the value of a one-seat ride from Washington Heights or Central Harlem to local tracks is limited, since everyone takes the express train and transfers at Columbus Circle.

How Deinterlining Can Improve New York City Transit

New York is unique among the major subways of the world in the extent of interlining its network has. All routes share tracks with other routes for part of the way, except the 1, 6, 7, and L. The advantage of this system is that it permits more one-seat rides. But the disadvantages are numerous, starting with the fact that delays on one line can propagate to nearly the entire system, and the fragile timetables lead to slower trains and lower capacity. New NYCT chief Andy Byford just released a plan calling for investment in capacity, called Fast Forward, focusing on accessibility and improved signaling, but also mentioning reducing interlining as a possibility to increase throughput.

I covered the interlining issue more generally in my article about reverse branching, but now I want to explain exactly what it means, having learned more about this issue in London as well as about the specifics of how it applies to New York. In short, New York needs to reduce the extent of reverse branching as much as possible to increase train speed and capacity, and can expect serious gains in maximum throughput if it does so. It should ultimately have a subway map looking something like this:

Lessons from London

In London, there is extensive interlining on the Underground, but less so than in New York. The subsurface lines form a complex interconnected system, which also shares tracks with one branch of the Piccadilly line, but the Northern, Central, Victoria, Jubilee, and Waterloo and City lines form closed systems (and the Bakerloo line shares tracks with one Overground line). The Northern line reverse-branches: it has two central trunks, one through Bank and one through Charing Cross; one southern segment, with through-trains to both trunks; and two northern branches, each sending half its trains to each trunk. The other closed systems have just one trunk each, and as a result are easier to schedule and have higher capacity.

As the Underground moves to install the same high-capacity signaling on more and more lines, we can see what the outer limit of throughput is on each system. The Northern line’s new moving-block signaling permits 26 trains per hour on the Bank trunk and 22 on the Charing Cross trunk. When the Battersea extension opens, reverse branching on the south will end, pairing the older line to Morden with Bank and the new extension with Charing Cross, and capacity will rise to 32 tph per trunk. Planned improvements to transfer capacity at Camden Town, the northern branch point, will enable TfL to permanently pair each northern branch with one central trunk, raising capacity to 36 tph per trunk. Moreover, TfL expects moving block signaling to raise District line capacity from 24 tph to 32, keeping the current reverse branching. The Victoria line already runs 36 tph and the Jubilee line soon will too, while the Central line runs 35 tph. So 36 vs. 32 seems like the difference coming from the final elimination of reverse branching, while more extensive reverse branching reduces capacity further.

The reason complex branching reduces capacity is that, as delays propagate, the schedule needs to incorporate a greater margin of error to recover from unexpected incidents. It also slows down the trains, since the trains are frequently held at merge points. The general rule is that anything that increases precision increases capacity (such as automation and moving block signaling) and anything that reduces precision reduces capacity; reverse branching reduces timetable precision, because each train can be delayed by incidents on more than one line, making delays more common.

What deinterlining in New York entails

NYCT has its work cut out for it when it comes to deinterlining. There are eight different points in the system where reverse-branching occurs – that is, where lines that do not share track in Manhattan (or on the G trunk outside Manhattan) share tracks elsewhere.

  1. The 2 and 5 trains share tracks on the Nostrand Avenue Line.
  2. The 2 and 5 also share tracks in the Bronx.
  3. The A and C trains share tracks on the two-track narrows through Lower Manhattan and Downtown Brooklyn.
  4. The A and D share the express tracks on Central Park West while the B and C share the local tracks.
  5. The E and F share the express tracks on Queens Boulevard, while the M and R share the local tracks, the E and M share the tunnel from Queens to Manhattan, and the N, R, and W share a different tunnel from Queens to Manhattan.
  6. The B and Q share tracks in Brooklyn, as do the D and N.
  7. The F and G share tracks in South Brooklyn.
  8. The M shares the Williamsburg Bridge tracks with the J/Z but runs in Manhattan on the same tracks as the F.

NYCT should work to eliminate all of the above reverse branches. The easiest to start with is #6: the junction at DeKalb Avenue should be set to keep the B and D trains together and to keep the N and Q together rather than to mix them so that the B shifts to the Q tracks and the D to the N tracks. This requires no changes in physical infrastructure, and has especially high benefits as the junction delays trains by several minutes in each direction. Moreover, the loss of one-seat rides is minimal: the BDFM and NQRW run closely parallel in Manhattan and intersect with a transfer at Herald Square in addition to the inconveniently long BQ/DNR transfer at Atlantic Avenue.

Another relatively easy reverse branch to eliminate is #8, a recent introduction from the 2010 service cuts. Previously, today’s M route in Queens and Manhattan was covered by the V train, which turned on the Lower East Side, while the M ran the same route as the J/Z, merging onto the R and thence the N in Brooklyn at rush hour. Today’s route is thus an M-V merger, which railfans including myself hoped would help decongest the L by creating an alternative route from Williamsburg to Midtown. Unfortunately, such decongestion has not happened, perhaps because gentrification in Williamsburg clusters near the L and not near the J or M.

Harsh decisions

Fixing the reverse-branching at DeKalb Avenue and on the Williamsburg Bridge is painless. The other reverse branches require a combination of hard decisions and new infrastructure.

Fixing reverse branches #3 and #4 requires no capital investment, just political will. Reverse branch #4 is there because there’s demand for two routes’ worth of capacity in the tunnel from Brooklyn but there’s only one express line on Eighth Avenue, and that in turn is the result of reverse branch #3; thus these two issues should be tackled together.

NYCT should decide between having the A and C trains run express between 145th and 59th Streets and the B and D trains run local, or the other way around. This is not an easy decision: either Washington Heights or Grand Concourse would get consigned to local trains. North of 145th the total number of boardings is 102,000 at B/D stations compared with 79,000 on the A/C, but conversely Concourse riders can change to the express 4 trains whereas Washington Heights’ only alternative is the local 1. However the A and C run, express or local, the E should run the opposite in Manhattan – it can merge to either the local or express tracks – and the express trains should continue to Brooklyn. The map I made doesn’t distinguish local from express service, but my suspicion is that Washington Heights should get express trains, on account of its long commutes and lack of fast alternatives.

The same problem of harshness occurs in reverse-branch #5. In theory, it’s an easy fix: there are three track pairs in Queens (Astoria, Queens Boulevard local, Queens Boulevard express) feeding three tunnels to Manhattan (63rd, 60th, and 53rd Streets). In practice, the three Manhattan trunks have astonishingly poor transfers between them in Midtown. Nonetheless, if it does nothing else, NYCT should remove the R from Queens Boulevard and route all 60th Street Tunnel trains to Astoria; together with fixing DeKalb Avenue, this would separate the lettered lines into two closed systems, inherited from the BMT and IND.

However, undoing the connection between the BMT and the IND probably requires constructing a transfer station in Long Island City between Queensboro Plaza and Queens Plaza, which involves a few hundred meters of underground walkway. Even then, the connection cannot possibly be convenient. The saving grace is that Eighth Avenue, Sixth Avenue, and Broadway are close enough to one another that passengers can walk to most destinations from any line.

Subsequently, NYCT should make a decision about whether to send express Queens Boulevard trains to 63rd Street and Sixth Avenue and local trains to 53rd and Eighth, as depicted in the above map, or the other way around. The problem is that the merge point between 53rd and 63rd Street Tunnels is one station east of Queens Plaza, at a local station, and thus the true transfer point is Roosevelt Avenue, far to the east. Riders on the local stations west of Roosevelt would get no choice where to go (though they get little choice today – both the E and M serve 53rd Street, not 63rd). The argument to do things as I depict them is to give the local stations access to 53rd Street; the argument to switch the lines is that there is more demand on 53rd than 63rd and also more demand on the express tracks than the local tracks, so the busiest lines should be paired. However, this in turn runs into turnback capacity limitations on the E in Manhattan, at the World Trade Center bumper tracks.

Potentially, NYCT could try to convert 36th Street into an express station, so that passengers could connect cross-platform. But such a dig would be costly and disruptive to operations. There were plans to do this at 59th Street on the 1/2/3 a few decades ago, for the transfer to the A/B/C/D, but nothing came of them.

Where new infrastructure is needed

The remaining reverse branches get increasingly more difficult. Already #7 requires new turnouts. The South Brooklyn trunk line has four tracks, but there’s not enough demand (or space in Manhattan) to fill them, so only the local tracks are used. There are occasional railfan calls for express service using the F, but it’s better to instead use the express tracks to segregate the G from the F. The G could be turned at Bergen Street on the local tracks, while the F could use the express tracks and then transition to the locals on new turnouts to be constructed at Carroll Street.

Also in the category of requiring new turnouts is #1: Rogers Avenue Junction is set up in a way that briefly forces the 2, 3, and 5 trains to share a short segment of track, limiting capacity. This can be resolved with new turnouts just east of the junction, pairing Nostrand Avenue Line with the local tracks and the West Side and the portion of the Eastern Parkway Line east of Nostrand with the express tracks and the Lexington Avenue Line. Trains on Eastern Parkway could either all go local, or keep the current mixture of local trains to New Lots (currently the 3) and express trains to Utica (currently the 4), skipping a total of two stations. This fix also reduces passengers’ access to one-seat rides, but at least there is a reasonable cross-platform transfer at Franklin Avenue, unlike on Queens Boulevard or at 145th Street and St. Nicholas.

And then there is #2, by far the most difficult fix. Demand on the White Plains Road branch in the Bronx is too strong to be a mere branch: the combined number of boardings at all stations is 166,000 per weekday, and besides, the line branches to Dyre Avenue near its outer end and thus needs the frequency of either a trunk or two branches to ensure adequate service to each other branch. This is why it gets both the 2 and 5 trains. There is unfortunately no infrastructure supporting a switch eliminating this track sharing: the 4 and 5 trains could both use this line, but then the 2 has no way of connecting to Jerome Avenue Line without new tunnels.

On the map, I propose the most obtrusive method of fixing this problem: cutting the 3 trains to a shuttle, with a new pocket track at 135th Street, letting passengers transfer to the 2, ideally cross-platform. With all through-trains running on the 2, there is no need (or space) for the 5 on White Plains Road, and instead the 5 should help boost the frequency on Jerome Avenue. In addition, some work is required at Woodlawn, which currently has bumper tracks, fine for a single branch but not for a non-branching trunk line (the bumper tracks on the L limit throughput to 26 tph no matter how good the electronics are).

Additional infrastructure suggested by deinterlining

Deinterlining is a service increase. Lines that today only get half or two-thirds service would get full trunk frequency: Second Avenue Subway would get the equivalent of the Q and N trains, the Astoria Line would get the entirety of 60th Street Tunnel’s capacity, Eighth Avenue would get more express trains, and 63rd Street Line and the South Brooklyn Line would get more than just the half-service of the F. With six track pairs on the lettered lines through Midtown and six north and east (Central Park West*2, Second Avenue, Astoria, Queens Boulevard*2), there is no room for natural branching to give more service to busy areas than to less busy ones.

One solution to this situation is targeted development on weak lines. Even with half service, South Brooklyn has underfull F trains, and Southern Brooklyn’s B, D, N, and Q trains aren’t much busier, making this entire area an attractive target for upzoning.

However, it’s more interesting to look at lines with extensions that suggest themselves. Second Avenue Subway has the obvious extension north to Harlem in phase 2, and a potential subsequent extension under 125th Street to Broadway, which is less obvious but popular among most area railfans (including one of my inside sources at NYCT). The Astoria Line has a natural extension to LaGuardia, ideally elevated over Ditmars to capture local ridership on Astoria as well as airport traffic; while the steel el structures in New York are soundboxes, it is possible to build quieter els using concrete (as on the 7 el over Queens Boulevard in Sunnyside) or a mixture of concrete columns and a steel structure (as on the Metro 2 and 6 els here in Paris).

The Nostrand Avenue Line provides an especially interesting example of a subway extension suggested by deinterlining. The terminal, Flatbush Avenue, was intended to be temporary, and as a result has limited turnback capacity. To prevent it from constraining the entire 2 route, which became the city’s most crowded even before Second Avenue Subway began decongesting the 4 and 5, it would be prudent to extend the line to Sheepshead Bay as the city intended when it planned the line in the 1910s.

In contrast, a subway extension under Utica, a stronger bus corridor than Nostrand with a strong outer anchor at Kings Plaza, loses value under deinterlining. It could only get a branch and thus have lower capacity than Nostrand. It would also definitely force branching on the express trains, whereas without such an extension they could run as a single unbranched line between Woodlawn and New Lots Avenue via Jerome Avenue, Lexington Avenue, and Eastern Parkway. A Utica subway should wait until there is political will to fund an entirely new crossing into Manhattan, presumably via Williamsburg to help decongest the L.

A second extension I have occasionally mooted, a subway under Northern, loses value even more. Such a subway would be a fourth trunk line in Queens and have to come at the expense of capacity on Queens Boulevard. It is only supportable if there is an entirely new tunnel to Midtown, passing under the mess that is the tracks in Long Island City.

A deinterlined New York City Subway

Fast Forward proposes moving block signaling on the most crowded subway segments, but typically only on trunks, not branches, and in some cases not even entire trunks. But in the long term, New York should transition to moving blocks and automation on all lines – at the very least the highly automated system used on the L, but ideally fully driverless operation, in recognition that wages are going up with economic growth but driver productivity isn’t. Simultaneously, deinterlined operations should allow tph counts in the mid 30s or even more (Metro 13 here runs 38 tph and Metro 14 runs 42).

Instead of the potpourri of lines offered today, there would be fewer, more intense lines. Nomenclature would presumably change to deal with the elimination of some services, clarifying the nature of the subway as a nine-line system in which five lines have four tracks and four lines have two. Each of these fourteen track pairs should be able to support a train every 90-105 seconds at the peak; not all lines have the demand for such frequency, and some have capacity-limiting bumper tracks that aren’t worth fixing (e.g. the 1 at Van Cortlandt Park), but many lines have the infrastructure and the demand for such capacity, including the express lines entering Midtown from Uptown or Queens.

Off-peak, too, service would improve. There is ample capacity outside rush hour, but turning the system into one of lines arriving every 4-5 minutes with strategic transfers rather than every 8-10 minutes would encourage people to take the trains more often. The trains would simultaneously be faster and more reliable, since incidents on one line would wreck service on other trains on the same line but leave the rest of the network unaffected.

With service improvements both during and outside rush hour, New York could expect to see substantial increases in ridership. Raising peak frequency from the current 24 tph to 36 tph on the busiest lines (today’s 2/3, 4/5, A/D, and E/F) is equivalent to building an entirely new four-track subway trunk line, and can be expected to produce similar benefits for passengers. The passups that have become all too familiar for riders on the 4, L, and other busy trains would become a thing of the past unless ridership rose 50% to match the increase in capacity.

Which Older Lines Should Express Rail Have Transfers to?

In my writings about metro network design I’ve emphasized the importance of making sure every pair of intersecting lines have a transfer. Moreover, I’ve argued that missed connections often come from having very wide stop spacing, because large metro networks have very closely-spaced lines in the core, and if the stop spacing in the core is too wide, as in Moscow, then lines will frequently cross without transfers. In contrast, in Paris, where the Metro has very closely-spaced stops, there is only one missed connection on the Metro, between Lines 5 and 14. However, what’s missing from this discussion is what to do on lines that, due to network design, have to run express and miss some connections. This question mattered to most RER lines and currently matters to Crossrail and Crossrail 2, and will be critical in any New York regional rail plan.

I claim that the most important connections to prioritize should be to,

  1. The busiest lines.
  2. Lines that are orthogonal to the newly-built express lines.

But before explaining this, I’d like to go over the scale of the underlying problem of prioritizing transfers. For a start, look at the Underground in Central London:

Crossrail is the dashed gray line. Between Paddington and Liverpool Street, it intersects seven north-south lines, including five in rapid succession on the West End; stopping at all of Bond Street, Oxford Circus, Tottenham Court Road, and Holborn would slow down too much what’s intended to be an express relief line to the Central line.

Stopping between two stations and having transfers to both is possible – look at Farringdon-Barbican and at Moorgate-Liverpool Street – but results in very long transfer times. The RER has opted for this solution at Auber, which is located between the Opera and Saint-Lazare, with a transfer stretching over three successive stations on Line 3, leading to legendarily labyrinthine transfers between the RER and the Metro:

Observe that in contrast with the RER A’s convoluted transfer at Auber, the RER B simply expresses between Chatelet-Les Halles and Gare du Nord, missing the connection to the east-west Lines 3, 8, and 9 and the north-south Line 7, and only connecting to the circumferential Line 2 via a long underground passageway. The reason for this is that a transfer station at Bonne Nouvelle or Sentier would be very expensive to construct; the RER’s stations were all extremely costly, and the RER A’s record of $750 million per km for the Nation-Auber segment remains unbroken outside the Anglosphere. On Crossrail (the recordholder in cost per km outside the US, soon to be overtaken by Crossrail 2), it’s the stations that drive up costs as well, and the same problem is even more acute in New York.

The tension is then between the network effects of including more transfer points, and the costs and slowdowns induced by stopping more often. The first point in my claim at the beginning of this post follows immediately: it’s more valuable to stop at transfer points to busier lines. The RER A misses Line 5 entirely, as does the express Line 14, because Line 5 is so weak that it’s not worth it to detour from Gare de Lyon through Bastille to connect to it; the oldest plans for the RER A had a stop at Bastille and not at Gare de Lyon, but under SNCF’s influence the system was redesigned to connect to the train stations better and thus Bastille was replaced.

Whereas the RER A in theory connects to every north-south one except the weakest (although the second strongest after Line 4, Line 13, has an even longer connection than at Chatelet), Crossrail does the opposite. The busiest station in London excluding mainline stations is Oxford Circus, thanks to the three-way transfer of the Bakerloo, Victoria, and Central lines; the Victoria line is the busiest in the system per km (although the longer Northern and Central lines have more riders), and it’s certainly the busiest north-south trunk line. However, plans to have a transfer to both Bond Street and Oxford Circus were rejected in favor of a connection to Bond Street alone. The reason is that London’s low-capacity passageways get congested, and TfL’s hamfisted solution is to omit critical transfers, a decision also made at the Battersea extension of the Northern line, which will miss a connection to the Victoria line at Vauxhall.

This brings me to the second transfer priority: it’s the most important to connect to orthogonal lines. The reason is that parallel lines, especially closely parallel lines, are less likely to generate transfers. New York’s four-track subway lines have very high volumes of local-express transfers, because those are easy cross-platform interchanges; as soon as any walking between platforms is required (for example, on the Lexington Avenue Line at 59th and 86th Streets), transfer volumes fall dramatically. In Paris, transfers between Line 1 and the RER A happen, but usually for longer-distance travel; I find it faster to take Line 1 from Nation to Chatelet than to take the RER A, even without any transfer, purely because it’s easier to get between the street and the Metro platforms at both ends.

This issue was never really in contention when Paris built the original RER system. The one place where the RER prioritized a transfer to a same-direction Metro line over an orthogonal one, Gare du Nord, is such an important destination for commuter and intercity trains that it’s obviously justified to prioritize it over an easier connection to Line 2. However, more recently, the RER E has seen this issue surface with the location of the infill Rosa Parks station. The RER E could have sited a station at the intersection with Line 5, but Line 5 goes northeast and serves much the same area as the RER E, so the network effects from an interchange would be weak. Instead, the station is sited to interchange with the circumferential T3 tramway, which opens up a connection toward Nation and eventually toward Porte d’Asnieres.

In London, the same question is critical to the central route of Crossrail 2. The current plan has three Central London stops: Victoria, Tottenham Court Road (with a transfer to Crossrail), and Euston-St. Pancras. But Victoria itself is not much of a destination, and of the two lines served, the District and the Victoria, the Victoria line is parallel to Crossrail 2 rather than orthogonal to it. The purpose of Crossrail 2 is to add north-south capacity through the West End to decongest the Victoria line and reduce the shuffle at Victoria station between mainline trains and the Underground; to this end, there’s no need to stop at Victoria station itself.

To this effect, Martha Dosztal proposes moving Crossrail 2 to Westminster or possibly Charing Cross. Instead of spending $2 billion on a station at Victoria, London would need to spend probably a comparable amount on a station that interchanges with lines that go northwest-southeast like Jubilee or Bakerloo rather than on the parallel Victoria line; moreover, Westminster and Charing Cross both have connections to the District line, so Crossrail 2 would still connect to all three east-west Underground lines.

Finally, the application to New York is the most delicate. New York’s scores of missed connections come from deliberate indifference on the IND’s part to transfers with the older lines rather than any systematic attempt at prioritizing important interchanges; the older IRT and BMT systems have between them just two missed connections (3/L in Brooklyn, 4-5/R-W in Lower Manhattan). But including better connections in the event the city builds more rail lines remains critical. Second Avenue Subway gets this right by having a cross-platform transfer to the east-west F; there’s no transfer to the north-south Lexington Line, but this is less important given Second Avenue’s role as a Lexington relief line.

Regional rail transfers are especially circumscribed in New York given the system’s nature as a few short tunnels: new tunnels across the Hudson, and ideally a connection between Penn Station and Grand Central. This is why there is little room for improving connectivity between the subway and what I call Lines 1-3 of New York regional rail. However, the priorities I’m advocating in this post suggest two important things about Penn Station: first, it’s important to reopen passageways to Sixth Avenue to allow connections to the NQRW and BDFM trains; and second, it’s not important to have a connection to the 7 at Hudson Yards, as IRUM proposes.

On more speculative lines involving longer tunnels, the same priorities point to my proposed stopping pattern in and around Lower Manhattan. What I call Line 4, a north-south line from Grand Central to Staten Island stopping at Union Square and Fulton Street would intersect the east-west subways: the 7 at Grand Central, the L at Union Square, and PATH and most Brooklyn-bound trains at Fulton Street. The only missed subways – the F/M at Houston Street and the N/Q at Canal – go mostly north-south (except the M, which has a same-platform transfer with the J/Z, connecting at Fulton). Likewise, what I call Line 5, connecting from Pavonia to Atlantic Terminal, would connect to most north-south subways at Fulton Street.

Ideally, it’s better to make every interchange, and subway builders around the world should aim for very long-term planning in order to prevent missed connections in the future. However, when the inevitable changes happen and missed connections are unavoidable, there are emergent rules for which are more important: busier lines are more important than less busy lines, and less obviously, lines that are orthogonal to the new line are more important than ones that are parallel. These priorities make it possible to build express lines that maximize regional connectivity with minimal loss of travel time due to making local stops.

Little Things That Matter: Circulation at Transfer Stations

I’ve written before about some problems of metro network design in large cities. In brief, it’s important to maximize network effects in a multi-line system, which means offering plenty of transfers between lines. The perfect network should have every pair of lines intersecting in the center with a transfer, with possible additional intersections outside the center, again with transfers. In practice, it never works quite this way. There are always compromises, based on particular historical and geographical details of city layout. But probably the single biggest contributor to the issue is transfer capacity. This issue also has independent interest, but the two worst examples I know of involve the central transfer points of London and Paris, where many lines converge.

For a start, it’s worth asking why even have multiple stations. Why not just build a perfect star-shaped system? Two-line subway networks usually just cross once in city center. Three-line networks can intersect at one point (as in Stockholm and the first three lines of Moscow), but more commonly they intersect in a triangle of three city center stations. Central transfer points go way beyond three lines, though: Otemachi has five subway lines; Tokyo Station has a subway line and six independent JR East commuter lines; Chatelet-Les Halles has five Metro lines and two and a half RER lines; Bank and Monument together have four Underground lines and the Docklands Light Railway; Times Square has five subway lines, three of which are four-track. Why not just add more lines to the same central station? There are three distinct answers.

Coverage

Transit networks aren’t just about connecting large neighborhoods (“Upper West Side”) to a nebulously defined city center. They’re about specific connections. City centers are larger than a single subway stop, and much larger than a single subway stop in any city that has any business building four or more subway lines. In Stockholm, where three lines is about right, the CBD extends about two stops heading north and east of T-Centralen. New York, Paris, Tokyo, and London all have CBDs several square kilometers in area, so it makes sense to route lines in such a way that it’s easy to reach many points within the CBD from all directions.

Concretely, take Times Square and Grand Central. It’s useful to serve both of them on multiple subway lines, but a north-south line can only serve one. Thus, the 4/5/6 serve Grand Central, and the A/C/E, 1/2/3, and N/Q/R/W serve Times Square. The same process repeats itself at a number of nodes within Midtown, and within CBDs of other large cities.

Construction difficulties

Independently of the value of having extensive service in multiple directions from multiple points in the CBD, there is the cost of bringing lines together. In large cities, the biggest source of missed connections to begin with is that the street available for line 6 may happen to pass right between two widely-spaced stops on line 1, which never had a stop at this street because line 6 was not in the planning stages yet.

For the same reason, urban street networks make it difficult to serve one point from more than a few directions. Even lines bored deep under the surface, without regard for the street network, would find it difficult to go on level -9, beneath eight older lines. The stations with the largest number of independent lines all have tricks to make this work. Times Square has three north-south subways that don’t physically intersect (the 1/2/3 is always to the west of the N/Q/R/W and the A/C/E well to the west of both), a stub-ending east-west subway (the shuttle), and one deep-bored east-west subway (the 7). At Tokyo Station four of the six commuter lines are elevated at the same level. At Otemachi the lines form a square, with one side consisting of two lines that were built together at the same time. Chatelet-Les Halles has platforms that do not intersect, and the most difficult retrofit, the addition of the RER, was a massive excavation project that cost billions of euros.

The same construction difficulties are also relevant to small transfer stations. In Paris, transfer stations try to avoid superimposing one line’s platforms on top of the other, just because it’s hard to build. As a result, transfers often involve long walks; transfers at Chatelet are particularly labyrinthine.

Transfer capacity

The biggest problem is not coverage, and only partly related to construction difficulties. At the busiest stations, pedestrian circulation between platforms can be a challenge. London Reconnections has a fourpart series about Bank, where three deep-level Underground lines meet, all having built in the late 1890s, when expected ridership was far lower than it is today. Circulation is so obstructed that at rush hour TfL occasionally has to close the station for safety reasons, or else passengers would fall onto the tracks. Retrofitting the station with additional connections between lines as well as from the platforms to the street has been a daunting task, since the most logical places for escalators to one line would often pass through the platforms of another.

At Chatelet-Les Halles, the same problem occurs, if not so acutely that trains need to skip the station at rush hour. The passageways between the Metro platforms and between the Metro and the RER are long and narrow, and barely adequate to handle the large volume of passengers.

The simplest way to prevent this problem from occurring at a particular station is to make sure to design enough room for the transfer. The simplest way to do that, in turn, is to ensure transfers are cross-platform. The RER has such cross-platform transfers at the station, pairing westbound RER A trains with northbound RER B trains and eastbound RER A trains with southbound RER B trains. But even the wrong-way RER A-B transfers and the transfers involving the RER D are fine: the station’s extreme cost paid for a full-length, full-width mezzanine. The London Underground, too, retrofitted these onto some older lines when it built the Victoria line, which has cross-platform transfers at such key stations as Oxford Circus (the busiest in London not counting mainline stations – Bank is only the second busiest) and Euston.

However, cross-platform transfers connect two lines. I know of one place where they connect three: Jamaica Station, where one track in each direction has platforms on both sides, and passengers on the trains on the opposite sides of these platforms are sometimes told to walk through the train to transfer. This is a unique feature of regional rail, with its timed connections; subways with a train every 2-3 minutes can’t realistically time the connections, and without timing the connections, passengers are better off walking up and down to the other platform than waiting for a train to come in for a purely horizontal transfer.

The need for coordinated planning

The ultimate problem with using more cross-platform transfers if that they require a great deal of foresight. Retrofitting them is not always possible, and costs money in modification of existing stations. Hong Kong, Singapore, and Taipei all use these transfers extensively, but only on lines that were planned together, such as the first two lines in Singapore; Singapore’s newer lines have long (though spacious) transfer corridors, and Hong Kong’s lines inherited from the original MTR and from mainline rail have poor transfers.

With relatively limited opportunities to have high-capacity, high-quality transfers, it’s no wonder that most cities that build rapid transit try to avoid four- and five-way transfers when possible. Complex transfers like this can arise by accident, over several layers of planning – in the case of Paris, Line 11 was planned and built a generation later than Lines 1, 4, and 7; the RER was built a generation later than Line 1; and Line 14 was built a generation later than the RER, and indeed was designed as a relief line to the RER A.

Ultimately, the best way to prevent a situation like Chatelet or Bank from occurring is to know in advance where every line will go. However, this is necessarily a hard task. In the 1890s, London was a city of 6 million, with a large number of poor people living in overcrowded condition in East and South London; a planner could guess how the city would grow and suburbanize in the 20th century but would not be able to predict this with any certainty. Paris, the capital of a then-poorer and far less industrialized country, has grown even more tremendously – in 1901 Ile-de-France had 4.7 million people, not all living in the built-up area of the capital.

In very large third-world cities, the task of predicting future growth is somewhat easier, but only because they’re already very large and have informal transit pointing the way to the major corridors. I can draw a semi-serious Lagos metro proposal based on the city’s urban layout today and expect much of its future growth to come from increasing building heights so that the same density can be accommodated with less overcrowding, but I can’t meaningfully say which future areas will become hotspots that must be served from all directions or how far the suburban sprawl will go.

Why Free Trade in Rolling Stock is Good

Classical economics asserts that if two countries freely trade, then both gain relative to a baseline in which they don’t trade. The classical theory of comparative advantage hinges on reciprocal free trade. But more recently, economists have begun to push for entirely domestic support for free trade, arguing that reducing trade barriers is good even without reciprocation. The arguments involve corruption and misallocation of capital coming from protectionism. Whatever criticism there may be of this neoliberal conception of trade, rolling stock appears to be an example in which this conception is right.

I have previously criticized informal French protectionism in high-prestige procurement for blowing up Parisian rolling stock costs by a factor of almost 2. In Paris, my example of what could be done with the money Ile-de-France Mobilités is wasting on rolling stock was infrastructure construction, justified by the city’s very low construction costs relative to ridership (if not relative to route-length). But there’s an even better set of examples of high costs in the United States, justified on labor grounds and yet involving wastes of money disproportionate to the number of jobs created.

Last month, The American Prospect published an article about a union push to have more US rolling stock made in America, by unionized workers. The TAP article talks about a light rail vehicle order in Los Angeles for $890 million, for what the article says is 175 cars and what manufacturer Kinki Sharyo and other industry sources say is 235 cars, built at a dedicated factory in the Los Angeles exurbs. The purpose of the article is to advocate for more protectionism for the sake of American union members, so it details the wages the workers are making (about $20 an hour, up from $11 for unskilled jobs elsewhere) but does not delve into comparative costs. It’s worth asking if the costs are competitive, and the answer is that they are not.

The cost of LACMTA’s Kinki Sharyo order is $3.8 million per car; these cars are 27 meters long, so this translates to $140,000 per meter of train length. In contrast, the average cost in Europe appears to be just under $100,000 per meter, across a variety of cities and models:

The shortest trains on this list (the Citadis Compact orders, at 22-24 meters) are in the middle of the pack, so it’s unlikely there’s any nonlinearity in cost; moreover, the Compact is slightly shorter than the Kinki Sharyo trains, so no extrapolation is required, only interpolation.

The LACMTA order follows another premium-priced light rail order in the same state: as I wrote in the Bay City Beacon last year, Muni Metro’s Siemens LRV order cost about $4 million per 23-meter car, about $170,000 per meter of train length. The trains are being built at a new plant in Sacramento.

The United States has federal Buy America laws, requiring federally-funded contracts to buy domestic products provided they cost no more than 25% more than equivalent imports. However, there is no in-state purchase requirement. Owing to large New York City Subway orders, some vendors have long-established plants near New York (Kawasaki and Alstom are in-state, Bombardier is in Vermont). However, under informal pressure from activists within California to provide good local jobs, LACMTA asked bidders to open local factories. Moreover, Siemens most likely placed its plant in Sacramento rather than in lower-cost states in order to curry favor with state-funded orders.

We even see the same problem in Massachusetts, where CRRC opened a plant in Springfield for an MBTA Red and Orange Line car order. The order itself does not come at a premium – according to Metro Report the base order is about $100,000 per meter of train length and the option is $115,000, and the range of per-meter costs for subway trains is the same as that for LRVs – but it’s possibly a loss leader to help establish CRRC as a player in the American market. Even before Trump’s election, Congress investigated the order, which beat the competitors by a large margin; the competing bids were about $135,000 per meter for the base order. It says a lot about Massachusetts’ broken procurement that it takes a loss leader just to get costs down to their international levels. Nonetheless, the US premium does appear to be smaller for large subway orders than for small and medium-size LRV orders, since the extra costs of siting and setting up a factory are spread across more units.

The explicit goal of local content requirements is to create jobs. This is usually justified in terms of inequality and bleak prospects for unskilled workers. However, there is no cost-benefit calculation involved in this. According to TAP, the LACMTA order is creating 250 jobs manufacturing the trains; it doesn’t say how long they will last, but the duration of the contract is about 6 years. But the premium, about $300 million, works out to $1.2 million per job, a large multiple of total compensation to the workers. The Springfield plant has 200 jobs paying $50,000-60,000 per year, lasting 7 years across more than just the Boston contract; pro-rating to the Boston contract’s share of orders from the plant, the jobs will last around 5 years. Adding back the premium charged by the competing vendors raises the cost to $1 million per job, again a multiple of total working-class compensation.

There are two reasons why labor protectionism costs so much compared with its direct impact on working-class hiring. The first is leakage: much of the premium goes to management, including factory design and construction, or is just wasted on inefficiency (CRRC is opening a second American plant, in Chicago, instead of building everything at one plant). Some of the money goes to foreign consultants with the vendor and some stays domestic, but the domestic leakage goes to sitework and not to direct hiring.

The second reason is corruption and degradation of institutions. When the goal of public procurement is not just to buy the best product in terms of cost and quality, lobbyists make demands, like local hiring, that corrupt the process. A city that signals that the only things that matter are cost and quality will attract vendors who make the best bids in terms of cost and quality; a city that signals that the process depends on local political needs will attract vendors who make bids in order to satisfy local political actors, who as a rule don’t give a damn about good transit. Thus American agencies buy trains at a premium well beyond Buy America’s 25% limit, just because they think of cost and quality as just two of several political priorities and not as the sole legitimate bases of choosing a bidder.

The United States leads the world in higher education costs. The unsubsidized cost of a college degree at a good public university is about $100,000; at CUNY, which provides a good quality of degrees even if it’s so underfunded that classrooms aren’t supplied with chalk, it’s about $75,000. Stipends at the level of a good graduate program add another $30,000 or so per year. For around $200,000 per person, California could send low-income workers to college and pay for their living expenses for the duration of the degree, whereupon they will be able to get unsubsidized jobs paying much more than $20 per hour. For workers who can’t go to college, trade school is another option, offering decently-paying jobs for much lower cost since they take much less time. There is no need to lade the transit capital budget with what should be state or federal retraining grants; given the massive difference in cost, even the loss of matching funds (i.e. other people’s money) can leave the state or the city better off.

The problem is that there is no political incentive to think in such terms. Part of it is the corruption of institutions, as I mentioned already: labor groups see an opportunity to create jobs from a budget that from a local perspective is other people’s money. Another part is political prestige: romantics like old jobs (farmer, builder, truck driver, coal miner, baker, factory worker), which have had enough time to percolate into the national psyche, and since these jobs are old, they’re likely to be at the low end of the value-added ladder.

Absent very strong rules forbidding protectionism in procurement, this corruption will continue: evidently, Paris insists on buying expensive bespoke trains and somehow manages to get them manufactured within France, even though EU rules against interstate dumping are much stronger than US rules. Rules at the highest level are required to discourage such behavior (although Paris might still waste money on bespoke trains, just ones that can be made in Poland). Congress can and should stop funding any local or state agency that takes in-state content into account in procurement; the US is one democratic country, not fifty mercantile fiefdoms, and should use its status as a superstate with a large internal market to universalize good governance.

Why is Second Avenue Subway Phase 2 So Expensive?

I am only loosely following the news about the second phase of Second Avenue Subway. The project, running from 96th Street to 125th, with a short segment under 125th to Lexington, passing under the 4, 5, and 6 trains, is supposed to be cheap. In the 1970s, work began on Second Avenue Subway before the city went bankrupt, and there are extant tunnel segments built cut-and-cover in East Harlem between the station sites. The stations need to be dug, but the plan dating back to 2003 was to build them cut-and-cover as well, with local disruption for only a few blocks around 106th, 116th, and 125th Streets. Only one part would be difficult: going deep under 125th, under the preexisting subway. And yet, costs are very high, and the design seems to be taking a wrong turn.

In the early 2000s, the cost projections were $3.7 billion for phase 1 (actual cost: $5 billion, but much of the difference is inflation), and $3.3 billion for phase 2 (projected cost: at least $6 billion). Since then, there have been changes. For about a year I heard rumors that the preliminary engineering had been done wrong, and it was impossible to use the preexisting tunnel segments. Then I heard that no, it’s actually possible to use the existing tunnels. But a few days ago I heard that even though it’s possible, the MTA is now planning to demolish the existing tunnels and build the entire project deep underground using tunnel-boring machines.

With the information generally given out at community meetings, it’s hard to know what’s exactly going on. However, the fact that the MTA is talking about this suggests extreme disinterest in cost control. Cut-and-cover construction is cheaper than TBMs, per a 1994 paper looking at French urban rail costs since the 1970s. The tradeoff is that it forces rail lines to go underneath streets, which is disruptive to pedestrians and merchants, or demolish private property. Fortunately, Second Avenue is a wide, straight throughfare, and requires no such demolitions, while the disruption would be localized to areas that are scheduled to get subway stops as part of the project. Metro extensions here and in a number of other European cities are constructing stations cut-and-cover and the tunnels between them with TBMs; Metro Line 12’s online documents state that station construction involves just 18 months of digging.

It’s possible that the need to turn to 125th Street is messing up the plan to do everything cut-and-cover. While the turn itself can be done with minimal demolitions (the inside of the curve has a few small buildings, and there’s also an alignment slightly farther east that goes under vacant land while maintaining a 90-meter curve radius), going underneath the Lexington Avenue Line requires diving deep, and then there is no advantage to cut-and-cover. Building cut-and-cover under existing lines is difficult, and in that case, TBMs are warranted.

If the problem is 125th Street, then I would propose extending phase 2 and then breaking it apart into two subphases. Phase 2.0 would be cut-and-cover and open stations at 106th, 116th, and possibly 125th and 2nd temporarily. Phase 2.5 would involve driving TBMs under 125th Street all the way to Broadway; this could be done with a large-diameter TBM, with the platforms contained within the bore and vertical access dug so as to avoid the intersecting north-south subways. 125th Street has 30,000 crosstown bus boardings according to the MTA, which would make it the busiest bus corridor in the city per km: 10,000 per km, compared with 8,000 on the busiest single route, the M86. It is a priority for subway expansion, and if it’s for some reason not possible to easily build from 96th and 2nd to 125th and Lex in one go then the entire project should be extended to 125th and Broadway, at somewhat higher cost and far higher benefits.

The reason phase 1 was so expensive is that the stations were mined from small digs, rather than built cut-and-cover as is more usual. The idea was to limit street disruption; instead there was street disruption lasting 5 years rather than 1.5, just at small bore sites at 72nd and 86th rather than throughout the station boxes’ footprints. The TBM drive and systems cost together $260 million per kilometer, compared with $125 million on Paris’s Metro Line 1 extension, but the stations cost $750 million each, compared with $110 million.

It’s crucial that the MTA not repeat this mistake in phase 2, and it’s crucial that area transit activists hold the MTA’s feet to the fire and demand sharp cost control. Even taking the existing premiums as a given, cut-and-cover stations should not cost more than $200 million each, which means phase 2 as planned should cost $600 million for stations, about $330 million for systems, and another $350 million for overheads. At $1.3 billion this still represents high cost per kilometer, about $500 million, but it’s based on actual New York cost items, which means it’s plausible today. There is no excuse for $6 billion.