Construction Costs and Experience

The most persistent criticism I have heard of my writings on construction costs, coming from YIMBY Princeton, is the importance of gradual expertise and experience. Against my claim that Americans build subways for higher costs than the rest of the world due to poor management practices, regulations, and procurement, and scope creep, YIMBY Princeton says that high costs are a result and not a cause of the rarity of American subway investment. I believe that high US costs are endogenous and therefore the US is reluctant to fund rail transit; he believes that disinterest in transit is endogenous and if the US were willing to build more rail lines, then construction costs would naturally go down through economies of scale and steady accumulation of project management expertise. I promised last year that I would go over his argument more carefully, and am going to do so in this post.

The obvious difficulty with this debate is that we agree there is negative correlation between construction costs and the extent of construction, and disagree on causation. Wikipedia lists 55 countries with metro systems, with a handful more with under-construction metros, but this is not enough of a dataset for large-n studies. There are too many control variables – for example, it’s easier to build the first subway line than the tenth, which reduces the proper comparisons for New York to a handful of large cities. Instead, the only real way to figure out what causes what is to rely on a handful of natural experiments.

I can come up with a number of natural experiments. One is ambiguous about causation: the role of poor project management in the high construction costs in Boston and Paris. The others all suggest that high costs are endogenous to the US, rather than unwillingness to build subway tunnels. These include the history of construction costs in New York, London, and Paris in the 1930s; the construction costs in London today; and the history of construction costs in Seoul from the 1970s to the present.

Project management

I know of two overpriced rail extensions blamed explicitly on poor project management: the Green Line Extension in Boston, and Grand Paris Express. As I explained in CityLab a few months ago, the GLX was budgeted at $3 billion for just 7.5 km of light rail trench in preexisting open cuts, but the MBTA cut this to $1.1 billion in actual construction plus $1.2 billion in rolling stock and sunk costs through hiring a more experienced project manager. In Paris, one of the reasons the Cour des Comptes cited in its report about the cost overrun is lack of experience in managing such a large project; as a result, the 200 km system, with 160 km underground, is now up to €35 billion.

The problem is that even with better cost control, Boston’s construction remains pricey. At $150 million per km, GLX is expensive for a line in a preexisting right-of-way, and not far behind GPX’s $220 million per km for an 80% underground network. While both Boston and Paris can expect future construction to be cheaper if they apply the lessons of GLX and GPX cost overruns, their absolute costs remain different, with Boston spending more per unit than Paris. At best this is neutral between my explanation and that of YIMBY Princeton.

Construction costs, dieselpunk edition

New York’s subway construction costs have risen since the start. In 1900-4, the First Subway cost $32 million for 22 km of subway and 1 km of viaduct (namely, the 125th Street viaduct on today’s 1 train), and another $3 million for 9 km. In today’s money, this is $39 million per km underground and $9 million per km elevated. JRTR has some statistics for the Dual Contracts, built in the 1910s and early 20s, and the IND, built in the 1930s. The Dual Contracts cost $366 million, equivalent to around $8 billion today; the total route-length added was about 180 km, of which 70 km was underground, consistent with a cost of about $80 million per underground km. Then the IND cost $815 million, equivalent to about $14 billion today, for 97 route-km, practically all underground, or about $140 million per km.

The projected cost of Second Avenue Subway kept rising. In 1929 the projection was $86 million, or about $1.2 billion in today’s money, or $90 million per km; this was before the IND cost overruns materialized (at a time of general deflation). In 1939 it was up to $249 million, or $4.2 billion today, about $320 million per km; by 1949, it had crept up to $500 million, or $5 billion today and $390 million per km. Put another way, WW2-era America, a country that had just built massive public works in the Depression as well as the war, including the IND and Chicago’s two subway lines through the Loop, was already projecting a higher per km cost than is routine in nearly the entire world today. Moreover, the plan was to build Second Avenue Subway cut-and-cover, a technique that is cheaper today than the deep boring typical of comparable infill subways in the first world.

I have less data than I’d like for other cities’ construction costs in the interwar era, but where they exist, they are a fraction of New York’s. The London Underground extension to Cockfosters in the Depression cost £4m for 12 km, 60% underground, per Wikipedia. In today’s money it’s $45 million per underground km. In Paris, there was little growth in real costs between 1913 and 1930: according to a presentation by Pascal Désabres, construction costs in today’s money in both 1913 and 1930 were about €23 million per km, or about $29 million per km.

London’s mounting costs

In the 1930s, London built an Underground extension for $45 million per km. After the war, it could no longer do so. According to numbers in the Financial Times, the Victoria line cost £4.5 million per km in the 1960s, all underground, which is about $110 million per km in today’s money, while the Jubilee line, built in the 1970s, cost about $250 million per km.

The Victoria and Jubilee lines were more complex projects than the Cockfosters extension, going under older Underground lines. The Jubilee line also included the construction of a transfer station with the Northern and Bakerloo lines at Charing Cross, whereas previously they only connected at the next two stations, Embankment and Waterloo. However, the construction technique, the tunnel-boring machine, is one that is supposed to have a much smaller city-center premium over outlying construction, since there is no surface disruption.

But whereas the Victoria and Jubilee lines had excuses for their high costs, more recent Underground extensions do not. In the 1990s, the Jubilee line extension cost around $500 million per km in today’s money, going under a few older Underground lines and crossing the Thames four times (in an environment with not much underwater premium) but mostly extending the system to the east, to previously underserved areas like Canary Wharf. The under-construction Battersea extension, crossing under one older line and serving a relatively undeveloped area at Battersea Power Station, is about $550 million per km. The next Underground extension under discussion today, that of the Bakerloo line to Lewisham, is budgeted at £3.1 billion over 7 km, or about $620 million per km, crossing under no Underground lines and largely following a wide road.

Under YIMBY Princeton’s theory, London’s construction costs should be decreasing as it obtains more experience tunneling in a constrained urban area with millennium-era sensitivity to environmental impact like noise. But on the contrary, costs keep growing.

Seoul’s low costs

If London is the expensive city that should under YIMBY Princeton’s theory get cheaper but isn’t, Seoul is the cheap city that should have been expensive in the 1980s but wasn’t. JRTR has data from the 1970s to the 1990s: after an increase at the beginning, Seoul’s construction costs stabilized in the 1980s and 90s at about $80 million per km in constant dollars in today’s money. These costs seem to persist today, judging by the Sin-Bundang Line, which cost 1,169 billion won for 18 km, converting to about $90 million per km in PPP dollars.

Seoul is consistent with the theory that costs are endogenous to a city or country. There is high correlation between the construction costs of different lines within the same city: having set non-US records with the Jubilee line extension, London keeps building very expensive Underground extension of the Northern and Bakerloo lines; Paris is spending around $250 million per underground km on a number of Metro extensions; Seoul keeps building subways at just under $100 million per km.

Missed Connections

I’ve discussed before the topic of missed connections on subway systems, both here and on City Metric. I’ve for the most part taken it for granted that on a rapid transit network, it’s important to ensure that whenever two lines intersect, they offer a transfer. This seems like common sense. The point of this post is not to argue for this principle, but to distinguish two different kinds of missed connections: city center misses, and outlying misses. Both are bad; if I had to say which is worse I’d say it’s the city center miss, but city center misses and outlying misses are bad for distinct reasons.

A useful principle is that every pair of rapid transit lines should intersect, unless one is a shuttle, or both are circumferential. If the city is so large that it has multiple circular lines at different radii (Beijing has two, and London vaguely has two as well depending on how one counts the Overground), then they shouldn’t intersect, but rapid transit networks should be radial, and every radial line should connect to every other line, with all radial-radial transfers ideally located within the center. City center misses weaken the network by making some radials not connect, or perhaps connect at an inconvenient spot. Outlying misses often permit more central transfers, and their problem is that they make it harder to transfer to the better or less crowded radial on the way to the center. London supplies a wealth of examples of the latter without the former.

What counts as a missed connection?

Fundamentally, the following picture is a missed subway connection:

The red and blue lines intersect without a transfer. Even if a few stations later there is a transfer, this is a miss. In contrast, the following picture is not a missed connection:

It might be faster for riders to transfer between the southern and western leg if there were a station at the exact physical intersection point, but as long as the next station on the red line has a transfer to the blue line it counts, even if the blue line has one (or more) stations in the middle. Washington supplies an example of this non-miss: it frustrates riders that there’s no connection between Farragut West and Farragut North, but at the next station south from the intersection on the Red Line, Metro Center, there is a transfer to the Blue and Orange Lines. London supplies another pair of examples: the Northern line and the Waterloo and City line appear to intersect the District line without a transfer, but their next station north from the physical intersection point, Bank, has an in-system transfer to Monument on the District.

There are still a few judgment calls in this system. One is what to do at the end of the line. In this case, I rule it a missed connection if the terminal clearly has an intersection without a transfer; if the terminal is roughly between the two stations on the through-line, it doesn’t count. Another is what to do about two lines that intersect twice in close succession, such as the Bakerloo and Hammersmith and City lines in London, and Metro Lines 4 and 10 in Paris. In such cases, I rule that, if there’s just one station on the wrong side (Paddington on Bakerloo, Mabillon on M10) then I rule it a single intersection and allow transfers at the next station over, by which standard London has a missed connection (Edgware Road has no Bakerloo/H&C transfer) and Paris doesn’t (Odeon has an M4/M10 transfer).

How many missed connections are there?

In Paris, there are three missed connections on the Metro there is one missed connection on the Metro (update: see comments below): M9/M12, M5/M14, M9/M14. As I discuss on City Metric, it’s no coincidence that two of these misses involve this miss involves Line 14, which has wide stop spacing. Narrow stop spacing makes it easier to connect within line-dense city centers, and Paris famously has the densest stop spacing of any major metro system. M9/M12 and M9/M14 morally should connect at Saint-Augustin and Saint-Lazare, but in fact there is no in-system transfer. M5/M14 should connect at Gare de Lyon, but when M5 was built it was not possible to get the line to the station underground and then have it cross the Seine above-ground, so instead it meets M1 at Bastille, while M14 doesn’t serve since it expresses from Gare de Lyon to Chatelet. A fourth missed connection is under construction: the extension of M14 to the north misses M2 at Rome, prioritizing long stop spacing over the connection to the M2/M6 circumferential.

In Tokyo, there are many misses. I am not sure why this is, but judging by line layout, Tokyo Metro and Toei try to stick to major roads whenever possible, to avoid tunneling under private property, and this constrains the ability of newer lines to hit station locations on older lines. If I understand this map correctly, there are 19 missed connections: Ginza/Hibiya (Toranomon and Kasumigaseki should connect), Ginza/Mita, Ginza/Yurakucho, Ginza/Shinjuku, Marunouchi/Mita (Ginza and Hibiya should connect), Marunouchi/Yurakucho, Asakusa/Yurakucho, Asakusa/Hanzomon, Hibiya/Namboku, Hibiya/Yurakucho (Tsukiji and Shintomicho should connect), Hibiya/Hanzomon, Hibiya/Shinjuku, Hibiya/Oedo, Tozai/Oedo, Tozai/Fukutoshin, Mita/Oedo, Chiyoda/Oedo twice, and Oedo/Fukutoshin. Oedo is particularly notable for being a circumferential line that misses a large number of transfers.

In New York, there are even more misses. Here the culprit is clear: the two older layers of the subway, the IRT and BMT, have just two missed connections. One, 3/L at Junius Street and Livonia Avenue, is an outlying miss. The other is central: Bowling Green on the 4-5 and Whitehall on the R-W should connect. But the newer layer, the IND, was built to drive the IRT and BMT into bankruptcy through competition rather than to complement them, and has a brutal number of misses: ABCD/2-3, ACE/1-2-3, AC-F/2-3-4-5, AC-G/2-3-4-5-BQ-DNR, BD/NQRW, BDFM/NQRW, BD/JZ, E/1, E/F, M/NW, R/7, F/BD-NQ, F/NRW, F-Q/4-5-6, F/NW, G/7, G/JMZ. Counting individual track pairs, this is 46 misses, for a total of 48 including the two IRT/BMT misses; I’m excluding local-only transfers, such as Columbus Circle and 53rd/Lex, and counting the 42nd Street Shuttle as an express version of the 7, so it doesn’t miss the BDFM transfer.

Finally, London only has eight misses. In Central London there are three: the Metropolitan or Hammersmith and City line misses the Bakerloo line as discussed above, and also the Victoria line and Charing Cross branch of the Northern line at Euston. The other five are outlying: the Central line misses the Hammersmith and City line at Wood Lane/White City, and its branches miss the Piccadilly line’s Uxbridge branch three times; the fifth miss is Metropolitan/Bakerloo. But one more miss is under construction: the Battersea extension of the Northern line is going to intersect the Victoria line without a transfer.

The difference between the two kinds of miss

Many misses are located just a few stations away from a transfer. In New York, some misses are just a station away from a transfer, including the G/7 miss in Long Island City, the E/1 miss between 50th Street and 59th Street, and several more are a few stations away, such as the various BDFM/NQRW misses. In London, these include two of the three Central London transfers: there is an H&C/Bakerloo transfer at Baker Street and an H&C-Met/Victoria transfer at King’s Cross-St. Pancras. As a result, not counting the Waterloo and City line, only two trunk lines in the system do not have any transfer: the Charing Cross branch of the Northern line and the Metropolitan/H&C line.

On a radial network, if two lines don’t have any transfer, then the network is degraded, since passengers can’t easily connect. In New York, this is a huge problem: some station pairs even within the inner networks require two transfers, or even three counting a cross-platform local/express transfer. My interest in subway networks and how they function came about when I lived in Morningside Heights on the 1 and tried socializing with bloggers in Williamsburg near the JMZ.

In Paris the three misses are also a problem. Line 4 is the only with a transfer to every other main line. Line 9 intersects every other line, and Line 14 will when its northern extension opens, but both miss connections, requiring some passengers to take three-seat rides, in a city infamous for its labyrinthine transfer stations. Fundamentally, the problem is that the Paris Metro is less radial than it should be: some lines are laid out as grid routes, including Lines 3, 5, and 10; moreover, Lines 8, 9, 12, and 13 are radial but oriented toward a different center from Lines 1, 4, 7, and 11.

In London, in contrast, there is almost no pair of stations that require a three-seat ride. The Charing Cross branch of the Northern line doesn’t make any stop that passengers from the H&C or Met line can’t get to from another line with one interchange (Goodge Street is walking distance to Warren Street). A bigger problem is the lack of interchange to the Central line on the west, which makes the connections between the H&C stations on the west and some Central line stations awkward, but it’s still only a small number of stations on each line. So the problem in London is not network robustness.

Rather, the problem in London is severe capacity limitations on some lines. Without good outlying interchanges, passengers who want to get between two lines need to ride all the way to the center. Most likely, passengers between the Piccadilly and Central line branches to the west end up driving, as car ownership in West London is relatively high. Passengers without a car have to instead overload the Central line trunk.

The same problem applies to misses that are strictly speaking not missed connections because the two lines do not actually intersect. In Paris, this occurs on Line 7, which swings by the Opera but doesn’t go far enough west to meet Lines 12 and 13. In London, the best example is Hammersmith station: the H&C and District lines have separate stations without an interchange, but they do not intersect since it’s the terminus of the H&C line and therefore I don’t count it as a miss. But morally it’s an outlying miss, preventing District line riders from changing to the H&C line to reach key destinations like Euston, King’s Cross, and Moorgate without overloading the Victoria or Northern line.

In New York this problem is much less acute. The only outlying misses are the 3/L and the ABCD/2-3; the 3/L connects two very low-ridership tails, so the only serious miss is on the Upper West Side. There, passengers originating in Harlem can walk to either line, since the two trunks are two long blocks apart, and passengers originating in Washington Heights can transfer from the A-C to the 1 at 168th Street; at the other end, passengers bound for Midtown can transfer at Columbus Circle, using the underfull 1 rather than the overcrowded 2 and 3.

The role of circumferential lines

Outlying transfers are useful in distributing passengers better to avoid capacity crunches, but they are incidental. They occur when formerly competing suburban lines get shoehorned into the same subway network, or when two straight roads intersect, as in Queens. But the task of distributing passengers between radial lines remains important and requires good connections between as many pairs of radials as possible.

The usual solution to this is a circumferential line. In Moscow, there are several missed connections in the center (Lines 3/6, 3/7, 6/9) and one more planned (8/9), but the Circle Line helps tie in nearly all the radii together, with just one missed connection (to Line 10 to the north) and one more under construction (to Line 8 to the west). The point of the Circle Line is to allow riders to connect between two outlying legs without congesting the center. This is especially important in the context of Moscow, where there are only a handful of interchange stations in the center, most of which connect more than two lines.

In London, the Overground is supposed to play this role. However, the connections between the Underground and Overground are weak. From Highbury and Islington clockwise, the Overground misses connections to the Central line, the Victoria line, the main line of the District line, the Piccadilly line, the Hammersmith and City line, both branches of the Northern line, and the Piccadilly line (it also misses the Metropolitan line, but that’s on a four-track stretch where it is express and local service is provided by the Jubilee line, with which there is a transfer). Much of this is an unforced error, since the Underground lines are often above-ground this far out, and stations could be moved to be better located for transfers.

In New York, the only circumferential line is the G train, which has uniquely bad transfers, legacy of the IND’s unwillingness to build a system working together with the older subways. Triboro RX (in the original version, not the more recent version) would play this role better: with very little tunneling, it could connect to every subway line going counterclockwise from the R in Bay Ridge to the B-D and 4 at Yankee Stadium. On the way, it would connect to some major intermediate centers, including Brooklyn College and Jackson Heights, but the point is not just to connect to these destinations in the circumferential direction but also to facilitate transfers between different lines.

Going forward, cities with large metro network should aim to construct transfers where feasible. In New York there are perennial proposals to connect the 3 and L trains; these should be implemented. In London, the missed outlying transfers involve above-ground stations, which can be moved. The most important miss, White City/Wood Lane, is already indicated as an interchange on the map, but does not to my understanding have an in-system transfer; this should be fixed.

Moreover, it is especially important to have transfers from the radial lines to the circumferential ones. These improve network connectivity by allowing passengers to change direction (from radial to circumferential, e.g. from east-west to north-south within Queens), but also help passengers avoid congested city centers like outlying radial-radial transfers. Where circumferential lines don’t exist, they should be constructed, including Triboro in New York and Line 15 in Paris; where they do, it’s important to ensure they don’t miss connections the way the Overground does.

The MTA Genius Challenge is as Bad as Expected

A year ago, Governor Andrew Cuomo declared a competitive $2.5 million grant, to be disbursed by what he dubbed the Genius Challenge. I wrote about it at the time, expressing skepticism that it would lead to anything useful. The panel of eight judges had only one person with background in the transportation industry, a former FRA administrator. The word “genius” itself is a tech mainstay that to me mostly means “I don’t know any Fields Medalists.” And the topics within the scope of the grant seemed more about what the tech industry thought were the most pressing issues and not what the lowest-hanging actually were. I had very low expectations, and the announcement of the winning entries met them.

The grant has three topics: signaling, rolling stock (interpreted broadly), and underground mobile or wireless service. The last three is by far the least important; it also got only half a million dollars, whereas each of the other two got a full million. Each of the two main ideas shows how weak the very concept of the genius grant is, but they do so in dramatically different ways.

The rolling stock winners included a vendor asking for a grant for New York to use its rolling stock (CRRC); the problems with that idea are more akin to those of the signaling section, so I will cover them there. A second rolling stock winner was a proposal to use better data collection to facilitate preventive maintenance; this idea may or may not work, it’s hard to tell from layers of obfuscating business language. It’s the third idea that deserves the most attention, and the most scorn: lengthening trains but not platforms.

The crank Idea: lengthening trains

The genius competition gave a $330,000 grant for the idea of lengthening trains from 10 to 14 cars without lengthening the platforms. Trains would alternate between only berthing the first 10 cars and only berthing the last 10. Transit Twitter has already dumped on this idea, and for good reason: the proposal reads like a crank paper purporting to prove the Riemann hypothesis or another famous result, starting with a lot of trivial observations and then making a leap of logic buried somewhere in the middle.

The basic problem with running trains that are longer than the platforms is that passengers need to be able to move to the correct car, which takes time. The report says that this is done on the London Underground, which is true, but only at outlying stations – as is the case on the subway in New York. The conductor announcement “only the first five cars will open” is familiar to anyone riding the 3 train and was familiar to anyone riding the 1 train before the new South Ferry station opened. This is fine as long as the station in question is low-volume enough that the extra dwell time does not interfere with operations. Lengthening trains beyond the platforms at busier stations than Harlem-148th Street 145th Street or South Ferry would result in a shuffle forcing passengers to scramble within the train (if moving between cars is possible) or on the platform (if it isn’t). The dwell times would be brutal and would almost certainly reduce capacity measured in passengers per hour.

The proposal handwaves this critical flaw by saying that dwell times would decrease because crowding would decrease. This assumes that dwell times are a function exclusively of on-train crowding, rather than of the number of passengers getting on or off the train. The same number of passengers would have the same platform space, but would actually only be able to use a fraction of it: many would only be able to use the 6 cars that go to their chosen destination, and at those cars, the volume of passengers per unit of platform length would rise.

The second handwave is unlimited stations, with longer platforms. Acknowledging that the busiest stations should have all doors open, the proposal says,

[P. 20] Third, 18.5% of rides occur through just 10 stations in Manhattan. In the medium term, the platforms can, and should, be extended at these 10 stations to enable customers that embark and disembark at them to use any car at both ends of their trip. Accordingly, 9.25% of the customers that presently need to use the middle cars could instead use the end cars.

This is the equivalent of the logical leap from trivial to wrong in a crank paper. First, the number of central stations that would need to be lengthened is much more than 10, including some key origins (86th/Lex, Jamaica Center, etc.) and transfer points (West 4th, Canal, 96th/Broadway). And second and more importantly, the busiest stations are multilevel complexes, where just adding more pedestrian circulation is hard; London is spending a considerable amount of money on that at Bank. Lengthening platforms at these stations is prohibitively expensive. This problem is discussed in cities with constrained underground platforms in the CBD, such as Vancouver, where nearly all Expo Line stations are above-ground (thus, relatively easy to lengthen), but the most crowded in Downtown Vancouver are in a tunnel, where platform reconstruction costs too much to be economic.

The bigger question is why the judges did not catch the error. The proposal brings up London as an example, which serves to bring the magic of the foreign to people who are unfamiliar with best industry practices. Saying that New York does the same is equally true, and in a way more relevant to the proposal (since New York doesn’t let people move between cars, making this more challenging than in London), but would raise questions like “can the dwell times of relatively light stations like South Ferry or Harlem-145th be replicated at the top 40 stations?”. London is Anglophone and some reformist New Yorkers have used it as a source of foreign ideas the way they wouldn’t use non-Anglophone cities. But the judges didn’t do the basic due diligence of checking whether London really implements the idea as widely as the proposal implies, and if not, then why not.

The rent-seeking idea: CBTC by another name

New York State awarded four applicants $250,000 each for ideas about signaling. All four ideas boiled down to the same thing: introducing new technology for communication between trains permitting the functional equivalent of moving-block signaling, at a lower cost than preexisting communication-based train control (CBTC) installations.

The grantees all have experience in the transportation industry. Rail signaling vendors Thales and Ansaldo propose to use cameras to read automated signals; train sensor provider Metrom Rail and veteran rail manager and consultant Robert James propose ultra-wide broadband to improve train location precision. There’s nothing obviously wrong about their proposals. Nor is there anything outlandish, which is why each of the two technologies has two independent applicants behind it. Thales and Ansaldo in particular have experience in advanced signaling – Thales supplied CBTC to the L 7 train in New York and to Metro Line 13 in Paris, and Ansaldo supplied rail automation to Copenhagen and CBTC to a number of Paris Metro lines.

Even then, questions about cost remain. Robert James’ and Metrom’s proposals leave a bad taste in my mouth for their cost estimates. James has a systemwide cost estimate somewhat less than $200 million, not much more than $500,000 per km; Metrom says its system costs “$3 million per mile” and compares itself positively with legacy CBTC systems at $20 million per mile. Actual costs of CBTC without automation in Paris on Line 13 were about 5 million euros per km according to Wikipedia, and this includes modification of the railyards and not just the signaling system. So the Metrom system’s claimed figure is still cheaper, but not by quite as much. Metrom also complains that in Boston, CBTC would not improve capacity much because it would prohibit double-berthing, an issue that is only relevant to a subway-surface system and not to a full metro.

The broader problem with this part of the grant is that if the MTA put out an RFP about CBTC on the subway, it would get bids from Ansaldo, Thales, and Metrom, and James might well bid or consult for a bidder. It would be able to judge the technical merit of each proposal in much closer detail than given in the competition. Instead, the state is paying vendors to market their technology to the public, which would influence future procurement.

While the grant asks about whether the technology is proprietary, it makes no attempt at establishing a multi-vendor standards. Such standards exist: Thales and Ansaldo are both listed as ERTMS vendors. In France there’s already a discussion in the trade press about whether using ERTMS is better than using CBTC; the discussion specifically mentions New York’s uniqueness as a network with connected rather than isolated lines, and says CBTC is designed for isolated lines whereas ERTMS is designed for shared lines, such as the RER system. European experts might well recommend that New York use ERTMS for the subway, even though it’s a system originally designed for mainline rail.

New York’s highly-branched system means it must be more conservative with new technology – there’s nowhere to test it, now that the L and 7 already have CBTC. The shuttles might be useful test cases, or the 1 and 6 trains on weekdays, but without isolated lines, the cost of a mistake in procurement or technological failure is much higher. This suggests the MTA should try to reduce the complexity of branching (which is what I would’ve proposed if it had been within the grant’s scope), and until then concentrate on imitating proven technology rather than innovating. This is especially important given the potential for rent-seeking, in which the vendors use the grant to market themselves to the state over competitors selling similar product.

The judges don’t know any better

Would a panel of judges with more familiarity with metro operations around the first world have come to better decisions? Probably. Through blogs, railfan forums, and comments, I know people with great knowledge of existing operations in a number of cities in the first world, and for the most part they think highly enough of their local systems that they’d ask of any innovation, “why hasn’t it been implemented here already?”.

I wrote in 2011 that people in the US who make technical arguments in favor of public transit tend to be skeptical of many proposals, to the point of finding existing US agencies incompetent. This is US-specific: London Reconnections is a technical blog but it tends to support Transport for London’s process, Swiss and Japanese railfans seem to trust their local rail operators, and even Transport Paris is more positive about STIF’s capital investment than New York-based blogs are about the MTA’s. Experts (and not just bloggers like me) could point out innovations their cities have that can be imported into New York, as well as shoot down bad ideas for which “why doesn’t London/Paris/Tokyo do it?” is a useful sanity check.

Note that sometimes there is a legitimate reason to do something that nobody else has tried. New York’s highly branched network makes ERTMS a better deal there than on other metro systems, and an RFI would be prudent. But because the details of implementation matter more than the idea of innovative genius, it has to go through the regular procurement process.

Cuomo attempted to inject the inventions of the American tech industry into the subway. Instead, he created space for cranks to promulgate their ideas and for vendors to have a leg up over their competitors in any future bid. In effect, his attempt to improve the economic productivity of the public sector to be more in line with that of the American tech industry is going to make the public sector less productive, through weaker institutions (namely, a less robust CBTC bid) and distraction (namely, the useless train lengthening idea).

New Hudson Tunnels Are Canceled. Again.

Amtrak’s Gateway project, spending $30 billion on new tunnels from New Jersey to Penn Station, just got its federal funding yanked. Previously the agreement was to split funding as 25% New York, 25% New Jersey, 50% federal; the states had committed to $5.5 billion, which with a federal match would build the bare tunnels but not some of the ancillary infrastructure (some useful, some not).

When Chris Christie canceled ARC in 2010, then estimated at $10-13 billion, I cheered. I linked to a YouTube video of the song Celebration in Aaron Renn‘s comments. ARC was a bad project, and at the beginning Gateway seemed better, in the sense that it connected the new tunnels to the existing station tracks and not to a deep cavern. But some elements (namely, Penn Station South) were questionable from the start, and the cost estimate was even then higher than that of ARC, which I attributed to both Amtrak’s incompetence and likely cost overruns on ARC independent of who managed it.

But I’m of two minds about to what extent good transit advocates should cheer Gateway’s impending demise. The argument for cheering is a straightforward cost-benefit calculation. The extra ridership coming from Gateway absent regional rail modernization is probably around 170,000 per weekday, a first-order estimate based on doubling current New Jersey Transit ridership into Penn Station. At $40,000 per weekday rider, this justifies $7 billion in construction costs, maybe a little more if Gateway makes it cheaper to do maintenance on the old tunnels. Gateway is $30 billion, so the cost is too high and the tunnel should not be built.

Moreover, it’s difficult to raise the benefits of Gateway using regional rail modernization. On the New Jersey side, population density thins fast, so the benefits of regional rail that do not rely on through-running (high frequency, fare integration, etc.) are limited. The main benefits require through-running, to improve access on Newark-Queens and other through-Manhattan origin-destination pairs. Gateway doesn’t include provisions for through-running – Penn Station South involves demolishing a Manhattan block to add terminal tracks. Even within the existing Penn Station footprint, constructing a new tunnel eastward to allow through-running becomes much harder if the New Jersey Transit tracks have heavy terminating traffic, which means Gateway would make future through-running tunnels more expensive.

But on the other hand, the bare tunnels are not a bad project in the sense of building along the wrong alignment or using the wrong techniques. They’re just extremely expensive: counting minor shoring up on the old tunnels, they cost $13 billion for 5 km of tunnel. Moreover, sequencing Gateway to start with the tunnels alone allows dropping Penn South, and might make it possible to add a new tunnel for through-running mid-project. So it’s really a question of how to reduce costs.

The underground tunneling portion of Second Avenue Subway is $150 million per km, and that of East Side Access is $200 million (link, PDF-p. 7). Both figures exclude systems, which add $110 million per km on Second Avenue Subway, and overheads, which add 37%. These are all high figures – in Paris tunneling is $90 million per km, systems $35 million, and overhead a premium of 18% – but added up they remain affordable. A station-free tunnel should cost $350 million per km, which has implications to the cost of connecting Penn Station with Grand Central. Gateway is instead around $2 billion per km.

Is Gateway expensive because it’s underwater? The answer is probably negative. Gateway is only one third underwater. If its underwater character alone justifies a factor of six cost premium over Second Avenue Subway, then other underwater tunnels should also exhibit very high costs by local standards. There aren’t a lot of examples of urban rail tunnels going under a body of water as wide as the Hudson, but there are enough to know that there is not such a large cost premium.

In the 1960s, one source, giving construction costs per track-foot, finds that the Transbay Tube cost 40% more than the bored segments of BART; including systems and overheads, which the source excludes, BART’s history gives a cost of $180 million, equivalent to $1.38 billion today, or $230 million per km. The Transbay Tube is an immersed tube and not a bored tunnel, and immersed tubes are overall cheaper, but a report by Transport Scotland says on p. 12 that immersed tubes actually cost more per linear meter and are only overall cheaper because they require shorter approaches, which suggests their overall cost advantage is small.

Today, Stockholm is extending the T-bana outward in three direction. A cost breakdown per line extension is available: excluding the depot and rolling stock, the suburban tunnel to Barkarby is $100 million per km, the outer-urban tunnel to Arenastaden in Solna is $138 million per km, and the part-inner urban, part-suburban tunnel to Nacka is $150 million per km. The tunnel to Nacka is a total of 11.5 km, of which about 1 is underwater, broken down into chunks using Skeppsholmen, with the longest continuous underwater segment about 650 meters long. A 9% underwater line with 9% cost premium over an underground line is not by itself proof of much, but it does indicate that the underwater premium is most likely low.

Based on the suggestive evidence of BART and the T-bana, proposing that bare Hudson tunnels should cost about $2-2.5 billion is not preposterous. With an onward connection to Grand Central, the total cost should be $2.5-3 billion. Note that this cost figure does not assume that New York can build anything as cheaply as Stockholm, only that it can build Gateway for the same unit cost as Second Avenue Subway. The project management does not have to be good – it merely has to be as bad as that of Second Avenue Subway, rather than far worse, most likely due to the influence of Amtrak.

The best scenario coming out of canceling Gateway is to attempt a third tunnel project, this time under a government agency that is not poisoned by the existing problems of either Amtrak or Port Authority. The MTA could potentially do it; among the agencies building things in the New York area it seems by far the least incompetent.

If Gateway stays as is, just without federal funding, then the solution is for Amtrak to invest in its own project management capacity. The cost of the Green Line Extension in Boston was reduced from $3 billion to $2.3 billion, of which only $1.1 billion is actual construction and the rest is a combination of equipment and sunk cost on the botched start of the project; MBTA insiders attribute this to the hiring of a new, more experienced project manager. If Gateway can be built for even the same unit cost as Second Avenue Subway, then the existing state commitments are enough to build it to Grand Central and still have about half the budget left for additional tunnels.

Urbanism and Standards of Respectability

Alex Baca wrote a thread on Twitter a week ago, talking about cities and normativity. The key tweet is,

The discourse about ~cities~ is, to me, a big fight against a normative, hegemonic, mostly white, mostly straight dominant culture, which has very clearly not made physical space for people who don’t fit that profile. It sucks, and we’re seeing the effects.

A few days later, I saw an unrelated meme, attacking Scott Wiener, the state senator representing San Francisco, who supports the YIMBY movement and introduced a zoning preemption bill that would permit mid-rise construction and abolish parking minimums near public transit. The meme consisted of two photoshops:

My first reaction to seeing the first picture was “I live in a building just like this.” I lived adjacent to buildings like the ones in the second picture in Singapore (where they’re actually taller) and the French Riviera. I’d already been thinking of different standards of middle-class respectability, in large part thanks to Alex’s thread, but the memes crystallized this so perfectly.

There’s a certain American standard of middle-class normality. A detached house for a nuclear family, with a backyard for the children to play in, and a garage that fits a car per adult. A school that has as few black students as possible without making the white middle class feel too guilty. Social engagements and hobbies that are so common, like a knitting circle, that a suburb of 10,000 can support a group. Everything else is deviant and embarrassing.

This is not the only middle-class respectability standard out there. There’s a competing standard, common in countries without a history of white flight. In cities with good transit, like Stockholm and Paris, generations can grow up and live in the city in apartment buildings and not own cars. Car ownership is still higher in richer areas – transit ridership is very far from universal in Paris, even intra muros – but it’s not mandatory. Detached housing is also less common even among the most comfortable segments of the middle class. The Singaporean dream is to own a car and live in a condo. The Israeli dream always includes a car but runs the gamut on density, from detached houses through mid-rises to high-rise condos in Tel Aviv.

Academia has lower car ownership than other professions of equivalent income, but still exhibits the same difference in mentality. My former postdoc advisor at KTH, a tenured professor, biked to work. At my last math conference, in Basel, one professor complained that when they tried biking to work when on sabbatical at the University of Michigan they got strange looks from the rest of the department. “Ann Arbor is a left-wing city and they still drive,” the professor said. In the United States, math postdocs usually don’t own cars but tenured professors do and often live in the suburbs, even at Columbia.

New York is supposed to have good transit and urban amenities, but for the most part the middle class treats it as part of a life cycle in which families live in the suburbs, rather than as a stable place. The city’s poverty rate is 20.3% per the 2012-6 American Community Survey, but among children age 5-17 it’s 29.3% (and among under-5 children it’s 27.7%). In my largely middle-class American social circle there are a number of New Yorkers, but also a number of people whose parents moved from New York to segregated suburbs when they were born or when they were about to start school. Even on the level of urban layout, there’s a stereotype that the city is not a good place to raise a family, e.g. because the apartments are too small (in fact they’re larger than Parisian ones).

It’s not just a matter of different tastes – some people think respectability means the Mad Men lifestyle, some think it means living in a walkable city. The walkable city is capable of containing more than one standard of respectability, because it arranges itself to let people access more potential friends, who could form different social networks. In theory it’s possible to drive an hour in some suburbs and meet many people, but in practice it’s uncommon, for two reasons. First, it requires one car per adult, which is expensive and produces class stratification even in social groups that could be cross-class. And second, in practice the social identity of suburbs in the Northern US is local more than regional – for example, they tend to have smaller, more local schools.

In theory, conservative lifestyles could also be more supported in cities. Haredi Jews are very urban: they need certain community amenities like a kosher supermarket and a mikveh and have to live within walking distance of synagogue; when they suburbanize, it’s en bloc, like the Satmar move to Kiryas Joel or master-planned Haredi cities in Israel, often in the settlements.

In practice, this doesn’t happen. Haredi Jews are notable in being an oppressed minority in Israel as well as around New York. But traditional groups that view themselves as part of the majority end up wanting everyone to live like they do. When the Progressive Movement created the idea of suburbia around 1900, it came out of an explicit desire to assimilate immigrants into what it viewed as proper American values. The Historic American Engineering Record gives background about the politics of the subway both before and after construction. The point of suburbia from the start was to make it impossible to form any culture except the dominant WASP culture.

Not for nothing, urbanism in the United States tends to disproportionately feature people who have other reasons to be dissatisfied with traditional culture. Foremost among these are queers, who led gentrification in the 1970s and 80s, when they weren’t safe in most suburbs and small cities; even this decade, a genderqueer Canadian acquaintance told me that there are parts of the US that they’re scared to be in. Without outing people, I believe that between one third and one half of the people who write online about public transportation or urbanism in the US are queer.

In order to reinforce the notion that only single-family suburbs are the respectable way to live, American society has to denigrate everything else as ridiculous. Parisian apartment buildings feature a hammer and sickle and defenestration; Mediterranean apartment buildings feature gay flamboyance. Were the US more willing to admit that there are educated professionals in some countries that do not need a car, it would need to find ways to accommodate professionals with the same preferences domestically, and that would lead to accidentally accommodating people who are not in the social or cultural mainstream.

Boston Regional Rail

At TransitMatters, we have finally released our regional rail paper, recommending improvements to the MBTA that regular readers of this blog are probably familiar with. Alert readers might even want to probe which parts were written by me and which by others; the main document underwent several edits but some stylistic differences might persist, and the appendices were mostly written individually. We are suggesting the following two-step process:

1. Modernize the system based on best industry practices. This includes full electrification and fleet replacement with electric multiple units (and not electric locomotives), high platforms at all stations, and high frequency all day, every half hour on every branch interlining to support a train every 10-15 minutes on urban trunk lines. In some areas, such as Revere, there should also be infill stops. The capital cost, excluding fleet replacement, should be on the order of $2-3 billion, but the first priority, the Providence Line, is maybe $100 million excluding rolling stock, mostly going to high platforms.

2. Build the North-South Rail Link, with four tracks connecting the South Station and North Station systems. This takes longer than electrification, so planning should start immediately, with the intention of opening somewhat after the entire system is wired. The capital cost should be $4-6 billion, per a study that we’re referencing in our report.

In my mind, regional rail serves three main markets:

1. Local trips on trunk lines, connecting to urban neighborhoods and subway transfer points. The main benefit of regional rail is that it provides an express subway at very high frequency, just as I use the RER to get to Western Paris faster than I would on the Metro. In Boston, areas that would benefit include Forest Hills, Allston and Brighton, Hyde Park, Dorchester and Mattapan along the Fairmount Line, Chelsea, Revere, and Porter Square. Residents of these neighborhoods are likely to travel to other neighborhoods and not just to Downtown Boston.

2. Suburban trips, which are dominated by peak commutes; I complained here that US commuter rail demand is peaky, with 67-69% of suburban trips on the LIRR and Metro-North and 80% on the MBTA occurring in the morning peak compared with around 47% on Transilien, but this is in large part about land use and not just frequency. We’re calling for replacing park-and-rides with town center stations in the report, but absent extensive transit-oriented development, suburban trips are likely to remain peaky and CBD-bound. This is the only market North American commuter rail serves, and its users are territorial about what they view as their trains. However, electrification would speed up these trips materially (the Sharon-South Station trip time would go from 35 to 23 minutes), and the North-South Rail Link would offer North Side suburbs access to the CBD, which is too far from North Station.

3. Intercity trips, which are not peaky except insofar as some people commute. Those tend to dominate off-peak ridership today: per a CTPS study from 2012, about half of the Providence Line’s off-peak ridership originates in Providence itself, which also accords with my observations taking the line on weekends. These trips gain less from high frequency, but need a consistent frequency all day, every day, at worst every 30 minutes, ideally every 15 or 20. Regional rail modernization also speeds these trips the most.

Bear in mind that even though the report just came out, the actual writing was for the most part done in November. This means that the technical aspects of scheduling reflect my thinking in November and not now. At the time, I hadn’t thought about peak-to-base ratios systematically, so my sample schedule for the Providence Line has a train every 15 minutes on each branch (Providence and Stoughton) at the peak and a train every 30 minutes off-peak. I had been assuming a peak-to-base ratio of 2 would be appropriate, by comparison with schedules in Tokyo and on the RER here in Paris. I knew that the ratio was lower in some other places I think highly of, including London and the German-speaking world, but my assumption had been that demand would be so peaky that the maximum acceptable peak-to-base ratio was the correct one.

I’ve argued before that the peak-to-reverse-peak ratio must be 1 or as close to it as practical, in order to avoid parking trains in city center midday. The capacity problems at South Station, which averages a train arrival per platform track per 35 minutes at the peak even though the system is capable of 10-minute turnaround times, come from trains going from the platform tracks to the layover yard during the peak, crossing the station throat at-grade and delaying peak arrivals.

But recently, I started thinking more carefully about operating costs, and wrote this post about peak-to-base ratios. I no longer think peak-to-base frequency ratios higher than 1 are supportable. The marginal labor cost of midday service when there’s a prominent peak is very low, since the railroad would be replacing split shifts with regular shifts, and this encourages running the same frequency during rush hour and midday, if not during the evening and on weekends. And as I explain in the linked post, the cost of rolling stock purchase and maintenance encourages running trains as often as possible. Only energy costs scale linearly with service-km, and those are low: at New England’s current electricity rates, it costs $180 to run a 320-ton 8-car EMU between Providence and Boston each way, and at current fares, inducing 16 extra passengers from the extra frequency is enough to make this pay.

In the report, we talk about American commuter rail operating costs, mostly because that’s what’s available. SEPTA’s are $311/car-hour, whereas those of the LIRR, Metro-North, New Jersey Transit, Metra, and the MBTA are $500-600/car-hour. Per car-km, SEPTA costs about $9 to operate. But a system built around cost minimization, with a peak-to-base ratio of 1 (thus, relatively empty off-peak trains), can get this down to about $2/car-km, or about $180/car-hour.

The reason I think the MBTA could run modern regional rail for $2/car-km, where the RER costs $6/car-km and the Singapore MRT $4-5/car-km, is that the schedule is faster. The costs of rolling stock and labor are based on time rather than distance, and the regional rail system we’re proposing has aggressive schedules, averaging 90 km/h between Boston and Providence. Even energy costs can be contained, since a fast schedule implies relatively few stops. For the same reason it’s easier to make a profit on high-speed rail averaging 200 km/h than on low-speed rail, it’s easier to make a profit on a 90 km/h train at the boundary between regional and intercity scale than on a 40 km/h local train.

In general, I believe that transit planning has to be opportunistic: no city is perfect, so it’s always necessary to find workarounds for some local misfeatures, or ways to turn them into positives. In Boston, the misfeature is very low suburban density, making intense regional service modeled after the RER less useful. The opportunity lies in retooling lines that serve low-density suburbs as intercity lines, connecting Boston with Worcester, Providence, Lowell, Nashua, and Hyannis. With the exception of Worcester, which is on a curvy line, these cities can be connected to Boston at an average speed of 90 km/h or so: the stop spacing is so sparse, and the lines are so straight, that long stretches of 160 km/h are feasible.

But none of this can happen under the present-day operating paradigm. The opportunity I’m describing relies on postwar travel patterns and to some extent even on 21st-century ones (namely, university travel between Providence and Cambridge), which requires reforming frequencies, rolling stock, and infrastructure decisions to incorporate best industry practices that emerged from the 1970s onward. The MBTA can offer a fast, affordable, frequent regional transportation system from as far north as Manchester to as far south as Providence, but for this it needs to implement the regional rail improvements we’re proposing.

Why Northeast Corridor Privatization is Doomed to Failure

Alex Armlovich asked me whether it’s possible to design a public-private partnership on the Northeast Corridor (NEC) to build high-speed rail. I took it to a Patreon poll, in which it prevailed over three other options (why land value taxation is overrated, why community groups oppose upzoning, and what examples of transit success there are in autocracies). On social media I gave a brief explanation for why such a privatization scheme would fail: the NEC has many users sharing tracks, requiring coordination of schedules and infrastructure, and privatizing one component would create incentives for rent-seeking rather than good work. In this post I am going to explain this more carefully.

Conceptually, the impetus for privatization is that the public sector cannot provide certain things successfully because it is politically controlled. For example, political control of infrastructure tends to lead to spreading investment around across a number of regions rather than where it is most needed; when Japan National Railways was broken up and privatized, the new companies let go of many lightly-used rural lines and focused on the urban commuter rail networks and the Shinkansen. Political control may also make it harder to keep down headcounts or wages. A competent government that recognizes that it will always be subject to political decisionmaking about services that should not be political will aim to devolve control of these services to the private sector.

The problem with this story is that privatization itself is a public program. This means that the government needs to be in good enough shape to write a PPP that encourages good service and discourages rent-seeking. Such a government entity does not exist in the realm of American public transportation. This doesn’t mean that all privatization deals are bad, but it means that only the simplest deals have any chance of success, and those deals in turn have the least impact.

When it comes to HSR, private operations work provided there is no or almost no need to coordinate schedules and fares with anyone else. One example is Texas, which has no commuter rail between Dallas and Houston nor any good reason to ever run such service. In California, this is also more or less the case: Caltrain-HSR compatibility is needed, but that’s a small portion of the line and could be resolved relatively easily.

In the Northeast, where there is extensive commuter rail, such coordination is indispensable. Without it, any operator has an incentive to make life miserable for the commuter rail operators and then demand state subsidies to allow regional trains on the track. Amtrak is already screwing other NEC users by charging high rates for electricity (which is supposedly the reason Conrail deelectrified, having previously run freight service on the NEC with electric locomotives) and by coming up with infrastructure plans that make regional rail modernization harder and demanding state money for them. If anything, the political control makes things less bad, because congressional representatives can yell at Amtrak; they will have less leverage over a private operator. In the other direction, Metro-North is slowing down Amtrak between New Rochelle and New Haven for the convenience of its own dispatching, and is likely to keep doing so under any PPP deal.

I have written many posts about what it would take to institute HSR on the NEC at the lowest possible cost. All of these make the same point, from many angles: organization – that is, improving timetabling – is vastly cheaper than pouring concrete and building bypass tracks. In chronological order, I’ve written,

Privatization is supposed to solve the problems of an incompetent public sector. But Amtrak’s incompetence is not really about wages or staffing; NEC trains are overstaffed relative to Shinkansen trains, but not relative to TGVs. Nor is it about unprofitable branch lines, not when the proposal is to privatize the NEC alone, rather than the entirety of Amtrak so that the private operator could shut down the long-distance trains. Some of the incompetence involves politicized procurement, but this is not the dominant source of high NEC costs. No: the incompetence manifests itself first of all in poor coordination between the various users of the NEC. Given better coordination, Amtrak could shave a substantial portion of its New York-New Haven runtime, perhaps by 10-20 minutes without any bridge replacements, and reduce schedule padding elsewhere.

To fix this situation, some organization would need to determine the timetables up and down the line and handle dispatching and train priority. In the presence of such an organization (which could well be Amtrak itself given top-to-bottom changes in management), a PPP is of limited benefit, because the private operator would be running on a schedule set publicly. Absent such an organization, privatization would make the agency turf battles that plague the entire NEC even worse than they are today.

In 2009, SNCF proposed to develop HSR in four places in the US: California, Texas, Florida, and the Midwest. The NEC, with its existing public intercity and regional rail operations, was not on its map. More recently, Texas Central is a private Japanese initiative to build HSR between Dallas and Houston. On the NEC the only Japanese initiative involved maglev between Washington and Baltimore, a mode of transportation that doesn’t fit the NEC’s context but is guaranteed to not share tracks with any state-owned commuter rail operation.

The invention of HSR itself was not privatized, and the European privatization paradigm involves public control of track infrastructure. Competing operators (some public, some private) can access tracks by paying a track charge, set equally across all operators. But even then, the track infrastructure owner has some decisions to make about design speed – mixing slower and faster trains reduces capacity, so if there’s a mixture of both, does the infrastructure owner assume the design speed is high and charge slower trains extra for taking high-speed slots or does it assume the design speed is low and charge faster trains extra? So far the public rail infrastructure operators have swept this question under the rug, relying on the fact that on high-speed tracks all trains go fast and on low-speed ones few HSR services go faster than an express regional train.

Unfortunately, the NEC requires large speed differences on the same route to avoid excessive tunneling. This complicates the EU’s attempts at a relatively hands-off approach to rail competition in two ways. First, it’s no longer possible to ignore the design speed question, not when regional trains should be connecting Boston and Providence in 51 minutes and high-speed trains in 20 minutes, on shared tracks with strategic overtakes. And second, the overtakes must be timed more precisely, which means whoever controls the tracks needs to also take an active hand in planning the schedules.

Handwaving the problems of the public sector using privatization works in some circumstances, such as those of Japan National Railways, but could never work on the NEC. The problems a PPP could fix, including labor and rolling stock procurement, are peripheral; the problems it would exacerbate, i.e. integrating infrastructure and schedule planning, are the central issues facing the NEC. There is no alternative to a better-run, better-managed state-owned rail planning apparatus.

Transit and Scale Variance Part 3: Grids

This is my third post about scale variance in transit planning; see parts 1 and 2. In part 1, I discussed how good bus networks exist at a certain scale, which can’t easily be replicated at larger scale (where the slowness of city buses makes them less useful). In part 2, I went over a subway planning feature, especially common in the communist bloc, that again works only at a specific scale, namely cities with enough population for 3-4 subway lines; it gets more complex in larger cities, and cannot be imported to bus networks with 3-4 lines. In this post, I will focus on one scale-variant feature of surface transit: the grid.

The grid works only for surface transit and not for rapid transit, and only at a specific scale, so constrained as to never be maximally useful in an entire city, only in a section of a city. This contrasts with what Jarrett Walker claims about grids. Per Jarrett, grids are the perfect form of a transit network and are for the most part scale-invariant (except in very small networks). One of the impetuses for this post is to push back against this: grids are the most useful at the scale of part of a transit city.

Grid Networks Versus Radial Networks

I’ve written a few posts exhorting subway planners to build their networks in a certain way, which, in the most perfect form, is radial. In particular, tangential subway lines, such as the G train in New York (especially when it ran to Forest Hills), Line 10 in Paris, and Lines 3 and 6 in Shanghai, are weak. When the G train was running to Forest Hills, most local passengers would switch from it to the next Manhattan-bound train, leading New York City Transit to send more Manhattan-bound local subways to Forest Hills and eventually to cut back the G to Long Island City. Based on these examples, I contend that on a subway network, every line should be either radial, serving the CBD, or circumferential, going around the CBD.

My post about New York light rail proposes a network with some lines that are neither: in the Bronx, my proposal is essentially a grid, with north-south routes (Grand Concourse, Webster, 3rd) and east-west ones (161st, Tremont, Fordham) and one that combines both (145th-Southern). Regular commenter NewtonMARunner criticized me for this on Twitter. I answered that the lines in my proposal are based on the busiest buses in the Bronx, but this simply shifts the locus of the question to the existing network: if transit lines should be radial or circumferential, then why are the tangential Bx19 bus (145th-Southern) or the Bx40/42 and Bx36 (Tremont, with a long radial eastern tail) so successful?

To answer this requires thinking more carefully about the role of circumferential routes, which by definition don’t serve the most intensely-used nodes. In Paris, Lines 2 and 6 form a ring that misses five out of six train stations and passes just outside the CBD, and yet they are both busy lines, ranking fourth and fifth in ridership per km. The reason is that they are useful for connecting to radial Metro lines and to some RER lines (namely, the RER A and the southern half of the RER B). Tangential lines miss connections much more easily: in the west, Line 10 here has a decent transfer to Line 9 and a somewhat decent one to Line 8, but to Lines 12 and 13 it’s already not very direct. The G train in New York has the same problem to the south – few connections to lines that actually do go into Manhattan.

Consider the following three possible networks:

The radial network is a typical subway network. The full grid lets you go from everywhere to everywhere with just one transfer, at the cost of having far more route length than the radial network. The partial grid no longer lets you go from everywhere to everywhere easily, and has the outer two lines in each service direction missing city center, but still has more overall route-length than the radial network. The principle here is that a grid plan is useful only if the grid can be complete.

The scale, then, is that rapid transit is so expensive that there’s no money for a complete grid, making a radial plan more appropriate. But surface transit, especially by bus, can be spread across a grid more readily. The Bronx’s size, density, and bus ridership patterns are such that a mostly complete grid is feasible within the western two-thirds of the borough, supplemented by the subway. In this environment, a tangential route is fine because it hits all the radial routes it could, and could provide useful two-seat rides to a large variety of destinations.

Are Grids Really Grids?

Chicago has a relentless bus grid. The three busiest north-south routes are the tangential 8 (Halsted), 9 (Ashland), and 49 (Western), which are 22, 29, and 26 km long respectively. None enters the Loop; Halsted, the easternmost, is at the closest approach 800 meters from the Loop, across a freeway. The two busiest east-west routes, the 77 (Belmont) and 79 (79th), are also far from the Loop.

However, I contend that these routes don’t really form a grid, at least not in the sense that passengers ride between two arbitrary points in Chicago by riding a north-south bus and connecting to an east-west bus. Instead, their outer ends form tails, which people ride to the L, while their inner ends are standard circumferentials, linking two L branches. The L in turn is purely radial and doesn’t follow the Chicago grid, with the Blue Line’s O’Hare Branch, the Orange Line, and the Brown Line all running diagonally.

Vancouver is similar. The north-south routes are radial, veering to enter Downtown. The east-west ones are more circumferential than tangential: they connect the Expo and Canada Lines, and most also connect to UBC. The Broadway buses (9 and 99) pass so close to Downtown Vancouver they’re more tangential, but they also offer the shortest path between the Expo and Canada Lines (making them a strong circumferential) while at the same time serving high job density on Central Broadway (giving them some characteristics of a radial).

In the absence of a radial rail network to connect to, long grid routes are less useful. Cities have a center and a periphery, and the center will always get more ridership, especially transit ridership. The outermost grid routes are often so weak that they should be pruned, but then they weaken the lines they connect to, making it necessary to prune even more lines until the grid is broken.

The Optimal Scale for a Grid

A strong transit grid will not form in a city too small for it. There needs to be a large enough center with enough demand for transit ridership to justify more than a purely radial bus network with a timed transfer. At the same time, the city cannot be too big, or else the arterial buses are too slow to be useful for ordinary work and leisure trips, as in Los Angeles.

What’s more, there is no Goldilocks zone, just right for a grid. Chicago is already too big for a bus grid without the radial rail layer. It’s also too big for what Jarrett calls grid accelerators – that is, rapid transit routes that replace bus grid lines: the Red Line is plausibly a grid accelerator, but the other lines in Chicago are not, and if there were L lines only at grid points, then the Red Line and the one east-west route would get overcrowded heading toward the Loop. Even Vancouver, a compact metro area hemmed by mountains and the ocean, relies on the diagonal Expo Line to serve Downtown and doesn’t really have a grid beyond city limits. A less dense city in the same land area could have a grid, but without much traffic or a strong CBD, cars would always beat transit on time and only the poor would ride the bus.

The scale in which grid networks work more or less on their own seems to be that of Vancouver proper, or that of the Bronx. Vancouver is 115 km^2 and the Bronx is 110 km^2; Vancouver’s bus grid spills over to Metrotown and the Bronx’s to Upper Manhattan, but in both cases these are small increases in the relevant land area.

Tellingly, Vancouver still relies on the bus network to feed SkyTrain; the Canada Line is a grid accelerator, but the Expo Line is not. The Bronx is the more interesting case, because it is not a city or even the center of the city, but rather a dense outlying portion of the city with an internal arterial grid. In both cases, the grid is supplementary to the radial rail core, even if the routes that use it have a lot of independent utility (Metro Vancouver has higher bus ridership than rail ridership, and the Bronx buses combined have slightly more ridership than the combined number of boardings on the Bronx subway stations).

Geographical constraints matter as well. The Bronx and Upper Manhattan are hemmed by water and by the administrative border of the city (which also includes a sharp density gradient), and Vancouver is hemmed by water and by a density gradient in the east. This makes it easier to equip both with grids that are close enough to the complete grid in the middle image above rather than the incomplete one in the third image. The Bronx’s lower-density eastern tails happen to meet up with those of Queens, forming circumferential routes, and also have enough north-south subway lines to feed that they remain useful.

In a transit city, the grid cannot come first. Even if there is a street grid, the spine of the network has to be radial as soon as there is demand for more than two rapid transit lines. The role of surface transit remains feeding rapid transit. Grids look attractive, but the optimal scale for them is awkward: large-scale surface transit grids are too slow, forcing the city to have a rapid transit backbone, and if the city is too small for that then the arterial grid provides too good auto access for public transit to be useful.

What the Spinetta Axe Reveals About Cost Control

The Macron administration commissioned a report about the future of SNCF by former Air France chief Jean-Cyril Spinetta. Spinetta released his report four days ago, making it clear that rail is growing in France but most of the network is unprofitable and should be shrunk. There is an overview of the report in English on Railway Gazette, and some more details in French media (La Tribune calls it “mind-blowing,” Les Echos “explosive”); the full proposal can be read here. Some of the recommendations in the Spinetta report concern governance, but the most radical one calls for pruning about 45% of SNCF’s network by length, which carries only 2% of passenger traffic. Given the extent of the proposed cut, it’s appropriate to refer to this report as the Spinetta Axe, in analogy with the Beeching Axe.

I wrote a mini-overview on Twitter, focusing on the content of the Axe. In this post I’m going to do more analysis of SNCF’s cost control problem and what we can learn from the report. The big takeaway is that cost control pressure is the highest on low-ridership lines, rather than on high-ridership lines. There is no attempt made to reduce SNCF’s operating costs in Ile-de-France or on the intercity main lines through better efficiency. To the British or American reader, it’s especially useful to read the report with a critical eye, since it is in some ways a better version of British and American discussions about efficiency that nonetheless accept high construction costs as a given.

SNCF is Losing Money

The major problem that the report begins with is that SNCF is losing money. It is not getting state subsidies, but instead it borrows to fund operating losses, to the tune of €2.8 billion in annual deficit (p. 28), of which €1.2-1.4 billion come from interest expenses on past debt and €1 billion come from taxes. Its situation is similar to that of Japan National Railways in the 1970s, which accumulated debt to fund operating losses, which the state ultimately wiped out in the restructuring and privatization of 1987. The report is aiming to find operating savings to put SNCF in the black without breaking up or privatizing the company; its proposed change to governance (turning SNCF into an SA) is entirely within the state-owned sector.

Unlike the Beeching report, the Spinetta report happens in a context of rising rail traffic. It opens up by making it clear that rail is not in decline in France, pointing out growth in both local and intercity ridership. However, SNCF is still losing money, because of the low financial performance of the legacy network and regional lines. The TGV network overall is profitable (though not every single train is profitable), but the TERs are big money pits. Annual regional contributions to the TER network total €3 billion, compared with just €1 billion in fare revenue (p. 30). The legacy intercity lines, which are rebranded every few years and are now called TETs, lose another €300 million. Some of the rising debt is just capital expenses that aren’t fully funded, including track renovation and new rolling stock; even in the Paris region, which has money, rolling stock purchase has only recently been devolved from SNCF to the regional transport association (p. 31).

In fact, the large monetary deficit is a recent phenomenon. In 2010, SNCF lost €600 million, but paid €1.2 billion in interest costs (p. 27); its operating margin was larger than its capital expenses. Capital expenses have risen due to increase in investment, while the operating margin has fallen due to an increase in operating costs. The report does not go into the history of fares (it says French rail fares are among Europe’s lowest, but its main comparisons are very high-fare networks like Switzerland’s, and in reality France is similar to Germany and Spain). But it says fares have not risen, for which SNCF’s attempt to provide deliberately uncomfortable lower-fare trains must share the blame.

The Spinetta Axe

The Spinetta report proposes multiple big changes; French media treats converting SNCF to an SA as a big deal. But in terms of the network, the biggest change is the cut to low-performing rail branches. The UIC categorizes rail lines based on traffic levels and required investment, from 1 (highest) to 9 (lowest). Categories 7-9 consist of 44% of route-km but only 9% of train-km (p. 48) and 2% of passengers (p. 51). Annual capital and operating spending on these lines is €1.7 billion (about €1 per passenger-km), and bringing them to a state of good repair would cost €5 billion. In contrast, closing these lines would save €1.2 billion a year.

But the report is not just cuts. Very little of SNCF’s operating expenditure is marginal: on p. 34 the report claims that marginal operating costs only add up to €1 billion a year, out of about €5.5 billion in total operating costs excluding any and all capital spending. As a result, alongside its recommendations to close low-ridership lines, it is suggesting increasing off-peak frequency on retained lines (p. 54, footnote 53).

There is no list of which lines should be closed; this is left for later. Page 50 has a map of category 7-9 lines, which are mostly rural branch lines, for example Nice-Breil-Tende. But a few are more intense regional lines, around Lyon, Toulouse, Rennes, Lille, and Strasbourg, and would presumably be kept and maintained to higher standards. Conversely, some category 5-6 lines could also be closed.

The report is equally harsh toward the TGV. While the TGV is overall profitable, not all parts of it are competitive. Per the report, the breakeven point with air travel, on both mode share and operating costs, is 3 hours one-way. At 3:30-4 hours one way, the report describes the situation for trains as “brutal,” with planes getting 80% mode share (p. 61). With TGV operating costs of €0.06/seat-km without capital (€0.07 with), it is uncompetitive on cost with low-cost airlines beyond 700 km, where EasyJet and Air France can keep costs down to €0.05/seat-km including capital and Ryanair to €0.04.

And this is where the report loses me. The TGV’s mode share versus air is consistently higher than that given in the report. One study imputes a breakeven point at nearly 4 hours. A study done for the LGV PACA, between Marseille and Nice, claims that as of 2009, the TGV had a 30% mode share on Paris-Nice, even including cars; its share of the air-rail market was 38%. This is a train that takes nearly 6 hours and was delayed three out of four times I took it, and the fourth time only made it on time because its timetable was unusually padded between Marseille and Paris. On Paris-Toulon, where the TGV takes about 4 hours, its mode share in 2009 was 54%, or 82% of the air-rail market.

SNCF has some serious operating cost issues. For example, the conventional TGVs (i.e. not the low-cost OuiGo) have four conductors per 200-meter train; the Shinkansen has three conductors per 400-meter train. The operating costs imputed from the European and East Asian average in American studies are somewhat lower, about $0.05-6/seat-km, or about €0.04-5/seat-km, making HSR competitive with low-cost airlines at longer range. However, there is no attempt to investigate how these costs can be reduced. One possibility, not running expensive TGVs on legacy lines but only on high-speed lines, is explicitly rejected (p. 64), and rightly so – Rennes, Toulouse, Mulhouse, Toulon, Nice, and Nantes are all on legacy lines.

This is something SNCF is aware of; it’s trying to improve fleet utilization to reduce operating costs by 20-30%. With higher fleet utilization, it could withdraw most of its single-level trains and have a nearly all-bilevel fleet, with just one single-level class, simplifying maintenance and interchangeability in similar manner to low-cost carriers’ use of a single aircraft class. However, this drive is not mentioned at all in the report, which takes today’s high costs as a given.

Efficiencies not Mentioned

The biggest bombshell I saw in the report is not in the recommendations at all. It is not in the Spinetta Axe, but in a table on p. 21 comparing SNCF with DB. The two networks are of similar size, with DB slightly larger, 35,000 route-km and 52,000 track-km vs. 26,000 and 49,000 on SNCF. But DB’s annual track maintenance budget is €1.4 billion whereas SNCF’s is €2.28 billion. Nearly the entire primary deficit of SNCF could be closed just by reducing track maintenance costs to German levels, without cutting low-usage lines.

Nonetheless, there is no investigation of whether it’s possible to conduct track maintenance more efficiently. Here as with the TGV’s operating expenses, the report treats unit costs as a fixed constant, rather than as variables that depend on labor productivity and good management.

Nor is there any discussion of rolling stock costs. Paris’s bespoke RER D and E trains, funded locally on lines to be operated by SNCF, cost €4.7 million per 25 meters of train length, with 30% of this cost going to design and overheads and only 70% to actual manufacturing. In Sweden, the more standard KISS cost €2.9 million per 25-meter car.

Low-ridership dilapidated rural branch lines are not the only place in the network where it’s possible to reduce costs. Rolling stock in Paris costs too much, maintenance on the entire network costs too much, TGV operating costs are higher than they should be, and fleet utilization in the off-peak is very low. The average TGV runs for 8 hours a day, and SNCF hopes to expand this to 10.

The Impetus for Cost Control

The Beeching Axe came in the context of falling rail traffic. The Spinetta Axe comes in the context of rapidly growing SNCF operating costs, recommending things that could and probably should have been done ten years ago. But ten years ago, SNCF had a primary surplus and there was no pressure to contain costs. By the same token, the report is recommending pruning the weakest lines, but ignores efficiencies on the strong lines, on the “why mess with what works?” idea.

The same effect is seen regionally. French rolling stock costs do not seem unusually high outside Paris. But Ile-de-France has money to waste, so it’s spending far too much on designing new rolling stock that nobody else has any use for. This is true outside France as well: the high operating costs of the subway in New York are not a US-wide phenomenon, but rather are restricted to New York, Boston, and Los Angeles, while the rest of the country, facing bigger cost pressure than New York and Boston, is forced to run trains for the same cost as the major European cities. It is also likely that New York (and more recently London) allowed its construction costs to explode to extreme levels because, with enough money to splurge on high-use lines like 63rd Street Tunnel and Second Avenue Subway, it never paid attention to cost control.

This approach to cost control is entirely reactive. Places with high operating or capital costs don’t mind these costs when times are good, and then face crisis when times are bad, such as when the financial crisis led to stagnation in TGV revenue amidst continued growth in operating costs, or when costs explode to the point of making plans no longer affordable. In crisis mode, a gentle reduction in costs may not be possible technically or politically, given pressure to save money fast. Without time to develop alternative plans, or learn and adopt best industry practices, agencies (or private companies) turn to cuts and cancel investment plans.

A stronger approach must be proactive. This means looking for cost savings regardless of the current financial situation, in profitable as well as unprofitable areas. If anything, rich regions and companies are better placed for improving efficiency: they have deep enough pockets to finance the one-time cost of some reforms and to take their time to implement reforms correctly. SNCF is getting caught with its pants down, and as a result Spinetta is proposing cuts but nothing about reducing unit operating and maintenance costs. Under a proactive approach, the key is not to get caught with your pants down in the first place.

Reforming Pensions

I don’t usually write about labor issues or pensions specifically, but an interview I got with a clerical union representative a few months ago made me think about pensions and vesting. I’ve come to believe that, to improve labor productivity at US public agencies (including transit, but not just), it would be useful to reform pensions, keeping their current levels and defined-benefit nature, but changing them to vest annually. That is, a year of working for a public agency at salary X should entitle a worker to a retirement annuity equal to a fraction of X, depending on age and years of experience (the fraction should be higher at lower age, representing more time the pension fund has had to gain interest).

In contrast, state governments in the US today have a long-term vesting model. I talked to Tim Lasker, president of the local Office and Professional Employees International Union, representing MBTA clerical workers. I intended to ask about salary competitiveness and retention, but Lasker told me that the pension system represents a “golden handcuffs.” People who are on the job for 25 years have fully-vested pensions, but people who leave earlier leave a chunk of money on the table by going. As a result, people with 15 or so years of experience don’t leave.

Jamal Johnson, a labor relations analyst working for the state government of Pennsylvania, said that in Pennsylvania pensions take up to 30 years to fully vest, and take 10 years to even partially vest, so workers who leave earlier get nothing.

The result is that there isn’t much turnover among employees with defined-benefit pensions. This is not a good thing. Lasker told me there is a lot of burnout, and a lot of people who just show up to work but aren’t productive, and are just punching the clock every day until they vest. They’re not lazy, and probably would quit in frustration and leave to a place where they could be more productive if it didn’t come at an enormous cost. Among the people who don’t have defined-benefit pensions, such as assistant secretaries and M.B.A. hires, there is much more turnover – too much, per Lasker, with people typically staying 1-2 years.

In the private sector, the best practice seems to be letting stock options vest after a number of years, as in the tech industry. But this is in an environment with performance bonuses – the idea is that giving workers shares in the company that only vest in a few years will incentivize them to work toward the company’s bottom line. The pensions are defined-contributions, which means the company and employee set aside money for a savings account that the government undertaxes, rather than providing a real pension.

Calls for pension reform in the public sector have tried to look to private-sector models, hence calls on the right (e.g. by Nicole Gelinas) to give public employees tax-deferred defined-contribution pseudo-pensions rather than actual pensions. Unless the point is to surreptitiously cut pension obligations, there is little point in this. The difference between a defined-contribution and a defined-benefit pension is ultimately who bears the risk of the worker living longer than expected. Under defined-contribution, it’s the worker, who then has to save much more money to avoid being penniless at 90. Under defined-benefit, it’s the company, which can average the amount across a large number of employees. In effect, defined-benefit pensions work as free life insurance.

The problem with the defined-benefit model today is purely that it assumes people work for the same agency their entire life, and thus pensions take decades to vest. This might have made sense in the middle of the 20th century, but it doesn’t today. People burn out, at which point it’s mutually beneficial for both sides to have them look for a new job and for the agency to look for a new worker. With burned out employees, the effect of the golden handcuffs on loyalty is not positive: people who are just punching in and out have no reason to be especially loyal to an employer that they hate. And the effect on morale is destructive: Lasker did not tell me this, but I suspect that with so much resentment among middle management, new hires are inducted into a culture in which good work is not valued. Lasker blamed mismanagement, and it is likely that this mismanagement percolates down the food chain.

There’s also limited risk of revolving door with transit agencies. In regulatory agencies, limiting employee exchange with the private sector is desirable, because otherwise workers have an incentive to shirk their duties in exchange for later high-paying jobs at the private companies they’re supposed to regulate. Tax collection agencies, safety agencies (including the FRA), and antitrust regulators are all at risk of regulatory capture. But transit agencies exhibit little such risk, since they do operations in-house. Capital programs are more vulnerable, but there, the people who are most affected by the revolving door are at the very top, and they’re not relying on a union pension. So the biggest risk coming from encouraging turnover at other public agencies is limited when it comes to transit agencies.

With this in mind, it’s useful to come up with a model in which pensions work like defined-contribution pseudo-pensions – that is, a sum of money the employer gives the worker that’s earmarked for a tax-deferred savings account. However, to reduce the aforementioned risk of running out of money in old age, it should be defined-benefit. I’m not aware of an existing model that achieves this, but it doesn’t mean such a scheme is impossible: the numerical parameters of this scheme (retirement age, rate of return, etc.) are already set in union agreements. All that’s required is to break down the pension into fractional amounts, accumulated every year of work, and cut the overall levels slightly so as to account for people who leave before their money is vested.

The big drawback of this plan is that there’s no real political appetite in the United States for any benefit-neutral pension reform. The unions might be interested, but without small cuts to offset people who leave early, a small additional increase in spending is required; the effect on productivity should be more than high enough to justify it, but the current zeitgeist in rich American cities treats pensions as outdated for ideological reasons. These same reformers who, as a rule, are outsiders to the union and think in terms of defined-contribution pensions, aren’t especially interested in making defined-benefit pensions work. If they think in terms of attracting and retaining talent, they think in terms of higher base wages.