A Better Billion and the Cost Model versus the 125th Street Subway Extension

We released a new report called A Better Billion. It was covered rather positively in the New York Times yesterday, with quotes from other transit advocacy groups. The idea for our report is that Zohran Mamdani promised free buses in his successful primary campaign, and promised free and fast buses in his successful general election campaign for mayor, so let’s take the $1 billion a year this could cost in forgone revenue and see how to spend it on subway expansion instead.

There’s been a lot of discussion in the article and on social media about the idea of free buses, but instead I want to talk about our proposal’s cost model, in the context of a rather incompetent plan the MTA released recently for a subway extension of Second Avenue Subway under 125th Street, at twice the per-km cost of Second Avenue Subway Phases 1 and 2, and twice the cost we project. Our model is not based on non-Anglo costs, but rather on real New York costs, modified to incorporate the one major cost saving coming from our previous reports, namely, shrinking station size. Based on everything combined, we came up with the following medium cost model:

ItemCost (2025 prices)
Tunnel (1 km)$530 million
Tunnel, underwater (1 km)$1,050 million
El or trench (1 km)$260 million
Station, cut-and-cover$510 million
Station, mined$770 million
Station, el or trench$240 million

These costs include apportioned soft costs and not just hard costs. Altogether, an extension of Second Avenue Subway from Park Avenue to Broadway, a distance of 2 km with three mined stations at the intersections with the north-south subway lines, should cost $3.4 billion. This is not much less per kilometer than Second Avenue Subway Phases 1 and 2, which can be explained by the denser stop spacing and the need for mined stations at the undercrossings. If everything else were done in the right way rather than the American way, the low cost model would apply and costs would be reduced further by a factor of about 3, but the per-km cost would remain one of the highest outside the Anglosphere for those geotechnical reasons.

But the MTA and its consultants, in this case AECOM, project $7.7 billion, not $3.4 billion. Why?

Worse project delivery

We’ve assumed the existing project delivery systems the MTA is familiar with. However, what doesn’t move forward moves backward, and the procurement strategy at the MTA is moving backward rapidly, for which the primary culprit is Janno Lieber, first in his role at MTA Capital Construction (now Construction and Development), and then in his role as MTA head, pushing alternative delivery methods, especially design-build and increasingly progressive design-build (unfortunately legalized in New York last year). Such methods add to the procurement costs and especially to the soft costs. Second Avenue Subway Phase 1 had an overall soft cost multiplier of about 1.5: the total cost including soft costs was 1.5 times the hard costs (Italy: 1.2-1.25 times). This proposal, in contrast, has a multiplier of 1.75: the hard costs are estimated at $4.4 billion, and the total costs are 75% higher, technically including rolling stock except rolling stock at current New York costs is $80 million.

Contingency

The soft costs include a federally mandated 40% contingency. The FTA mandates excessive contingencies – the norm in low-cost countries is 10-20%, and anything more than that is just wasted. The contingency figure varies by phase of design and decreases as it advances, but in the earliest phase it is 40%, and it’s in that phase that budgeting is done. However, 40% is only required over hard costs based on standardized cost categories (SCCs), and not over past ex post costs that incorporated contingency themselves. In effect, the estimation method the MTA and AECOM prefer bakes in a 40% overrun at each stage, letting project delivery get worse over time as the globalized system of procurement takes deeper roots in New York.

Overdesign and overbuilding

Based on our recommendations, the MTA shrank the station overages in Second Avenue Subway Phase 2. Phase 1 had station digs 100% longer than the platforms, based on standards that were both extravagant to the taxpayer and spartan to the end user – the extra space is not usable by passengers but instead for unnecessary break rooms, separated by department. By Phase 2, this was reduced to a 50% overage, and we hoped that proactive design around best practices would reduce this further.

Unfortunately, the overages are still substantial, 50% at St. Nicholas and 25% at the other two stations (Italy, Sweden, France, Germany, China: 3-20%). Moreover, the stations still have full-length mezzanines. This a longstanding New York tradition, going back to the 1930s with the opening of the IND lines starting with the A on Eighth Avenue in 1933. And like all other New York subway building traditions that conflict with how things are done in more advanced, non-English speaking countries, it belongs in the ashbin of history. Mined stations’ costs are sensitive to dig volume, and there is little need for such additional circulation space, for passenger comfort or fire safety. Mezzanines are essentially free if the stations are built cut-and-cover, in which case they are used for back-of-the-house space in advanced countries, but not if the stations are mined, in which case the best place for break rooms is under stairs and escalators.

Moreover, as we will explain soon at the Effective Transit Alliance, mined stations and bored tunnel require a minimum spacing from the street and from other tunnels – but the proposal includes much more space than necessary, forcing the stations to be deeper, more expensive, and less convenient as it takes a full five minutes to transfer between platforms or to get from the platform to the street. It’s possible to ge even shallower with shoring techniques used in China to reduce tunnel and station depth in complex urban undergrounds.

Proactive and reactive cost control

When the MTA announced cost savings and station size shrinkage in Phase 2, we were excited. But on hindsight, costs in effect fell from $7 billion to $7 billion. The savings were entirely reactive, designed to limit further cost overruns, and are not proactively incorporated into further projects.

No doubt, if a $7.7 billion project is approved against any honest benefit-cost analysis (which is not required in American law), then shrinkage in station footprint and reduction in mezzanine length will be found to be saving money in 2032, and the successor of Lieber, hired from the same pipeline of people whose takes on other countries are “I had a kid who did a semester abroad in Stockholm,” will be proud of reducing costs from $7.7 billion to $7.7 billion.

The path forward must instead incorporate cost savings proactively. There’s a way of building subway stations cost-effectively, and instead of quarter-measures, the MTA should adopt it; we have blueprints from a growing selection of examples, all in places that have avoided the destruction of subway building capacity infecting the entire English-dominant world in the last 25 years. The MTA can even hire people with direct transport official-to-transport official communication with peers at other agencies (for example, through COMET) and with the language skills to read documents produced in lower-cost countries, instead of people whose best skill is giving interviews to softball interviewers and talking about sports.

British Construction Costs and Centralization

There’s an ongoing conversation in the United Kingdom right about state capacity and centralization. The United Kingdom is notable for how centralized it is compared with peer first-world democracies of similar size. It also has weak state capacity on matters including infrastructure construction, which quite a lot of analysts and thinktanks assume is connected. Most recently, I’ve seen conversations on Bluesky in British media, talking about how the loss of state capacity in the UK in the last 45 years has really been about centralization of functions that local and county governments used to do and therefore the solution is to devolve some functions in England to counties or regions. Much of this discourse is by people I deeply respect, like the Financial Times’ Stephen Bush, pointing out sundry services Greater London ran before Margaret Thatcher’s anti-local government reforms in the 1980s. And yet none of this is relevant to infrastructure construction costs and the country’s inability to build more than a half-phase high-speed line at costs that would be high for a subway.

English centralization…

England has been very centralized for a long time, since the Early Middle Ages. It never really had anything like the German or Italian states, or even French provinces, not have its reforms to local government established anything like the French regions. Scotland, Wales, and Northern Ireland have extensive devolution, but 85% of the population lives in England, and the overall character of the state is never driven by peripheral regions comprising 15% of it. Indeed, in the OECD, by one measure, the United Kingdom has the single largest share of taxes going to the central government, and the second largest share of spending decided by the central government behind New Zealand. The other reasonable measure of centralization would be to do the same but assign social security to the central government, since it invariably either is national or comes with extensive equalization payments; then a few small countries like Israel and Ireland end up more centralized, but none of the European countries of comparable population.

How to fund and what to devolve to local government in England has been a complex issue over the last few generations. Local taxation is weak, and there is nothing like the German system in which the federal government, which collects 95% of taxes, distributes the taxes to the states by formula to do with as they please. The succession of local council tax programs to fund such governments culminated in Margaret Thatcher’s community charge, better known as the poll tax, which was so unpopular it led to her overthrow in a palace coup; at the time, Labour was polling 20% ahead due to backlash against the proposal.

Thatcher herself worked to disempower regional governance that had been established in the two decades before she took office. She aimed to destroy three institutions that she believed were keeping Britain socialist and thus backward: unions, public-sector bureaucracies, and regional governments. On the last point, she led a reform that eliminated regional governance in the Metropolitan Counties, creating a unique situation in which the constituent municipalities of these counties are single-tier municipalities with no level of governance between them and the state. Anything else would permit powerful Labour regions to challenge state privatization and deunionization schemes. Indeed, reforms by the New Tories to undo this and devolve some functions to the Met Counties created elected mayors, and now the mayor of Greater Manchester, Andy Burnham, has arisen as a powerful Labour politician who is mooted by many (including the Green Party) as a potential replacement for Keir Starmer in an intra-party palace coup.

…is not the reason for high costs

The truth is that while the United Kingdom is atypically centralized, at least in England, the exact same problems are seen across the Anglosphere, with roughly the same origin (except in the United States, whose problems have a different origin). British costs exploded in the 1990s, and Canadian and Australian costs followed suit in the 2000s and 2010s, imitating bad British practices. Australia and Canada both have some of the fiscally strongest states/provinces in the OECD, the exact opposite of the United Kingdom.

Across Europe, subtracting the United Kingdom, there isn’t an obvious relationship between centralization and poor state capacity. The Nordic countries are both rather devolved for small unitary states, managing health care and education subnationally. For example, in Finland it’s done in 21 health care regions, with ongoing debates over reducing the number of regions, but no attempt to eliminate devolution and make it a national system, in a country that after all has about the same population as Scotland. On the other hand, Italy is not much less centralized than the United Kingdom, and its infrastructure construction program is excellent, limited by money and uncertain growth prospects but perfectly capable of building a national high-speed rail network for less than half the budget of the 225 km High Speed 2. In democratic Europe, the strongest correlate of high costs is exposure to the United Kingdom and its way of doing things, with the Netherlands having the worst costs and most compromised infrastructure construction.

The privatization of state planning

At the Transit Costs Project, we are putting together a cost report on London, largely about Crossrail but also other recent urban rail expansion including the Docklands Light Railway and the Northern line extension. And what comes out of this history is that it’s not really about centralization. Rather, the United Kingdom invented what I called in the Stockholm report the globalized system, in which planning functions are privatized to large consultant firms while the role of the state is reduced to at most light oversight.

The explosion in costs in the 1990s, producing the Jubilee line extension at nearly four times the real per-km cost of the original Jubilee line, was part of this transition. It was easy to miss the first time we looked because the telltale signs of the globalized system, like design-build contracts, weren’t there yet. If anything, DLR was more privatized in its project delivery, and had reasonable costs until the Bank extension. And yet, delving more deeply, we (by which I mean Borners) found that beneath the surface it did have quite a lot of those negative features.

For example, while the Jubilee line extension was designed with in-house planning, it was understood that future projects would transition to more privatized planning. Thus, there was no expectation that the knowledge gained while building the extension would stick around, and at any rate, there was pressure to build like in Hong Kong, where pro-privatization British consultants cut their teeth in the 1980s and early 1990s. In effect, while the extension was designed in-house, it had all the features of a special purpose delivery vehicle (SPDV, or SPV), the preferred British and increasingly pan-Anglosphere way of delivering large projects: each project’s team is specific to the project itself and after completion the employees, drawn from a mix of private consultancies and public agencies, scatter and are not reassembled as a team for future projects. DLR was if anything the opposite: it was nominally private in its delivery but the same designers were involved throughout, moving between private employers and functions, so it was not de facto an SPDV, whereas the Jubilee line extension de facto was one.

Why do Brits blame centralization?

High costs in the United Kingdom cannot have much to do with the extent of devolution. Australia and Canada have adopted the same way of planning, after all. The British system in which big decisions can only be made by the minister and not by senior, let alone mid-level, civil servants, can be implemented regardless of scale, and Canadian provinces and Australian states are thus without exception not capable of building infrastructure for costs that were routine as late as 20 years ago.

But it can look like centralization, for all of the following reasons:

  • The United Kingdom really does have issues with overcentralization and underempowerment of regional governments, though the latter is being fixed to some extent, at least for the Met Counties. It is natural to see two governance problems that cooccur and assume they’re related.
  • The origin of the British cost explosion is, ideologically, the same process that also disempowered regional governments. It took some work to figure out the exact process, which was not at all the same as the conversion of the Met Counties to single-tier authorities. Indeed, in Canada the disempowerment of local or regional authorities never happened, but the transition to privatized planning with political rather than civil service oversight did happen, with large design-build contracts with more consultant involvement.
  • The implementation of centralization in the United Kingdom has relied on ministerial approvals, and those genuinely bottleneck the state. A better system of centralization, such as in Italy, relies on trust in the civil service. But a prime minister whose favorite television show was Yes, Minister would never have produced such a system.

But that it may be reasonable for a Brit who doesn’t look too closely at how Canada and Australia fail in parallel ways to assume that it’s about centralization does not mean that it is in fact about centralization. Centralized states that don’t speak English routinely build infrastructure efficiently and highly devolved ones that do are incapable of relieving the most important bottlenecks in their city centers.

Fare Practices

Here’s a table of urban public transport fares for various cities, covering the United States, Canada, parts of Europe, Turkey, and Japan. Included are single fares, multi-ride discounts, day passes, weeklies, and monthlies, with the last three shown with their ratios to single fares. As far as possible we’ve tried doing fares as of 2026, but it’s possible a few numbers are not updated and depict 2025 figures.

The thing to note is that in Continental Europe, there are steeply discounted monthlies – only two cities in the table charge for a monthly more than for 30 single-trips (Paris at 35.5, Bari at 35). Most Italian cities cluster around 20, and Barcelona, Lisbon, and especially Porto are even lower. Berlin used to have a multiplier of 32 before the 9€ monthly and the subsequent Deutschlandticket but the current multiplier is 15.75 within the city. Stockholm has a monthly multiplier of 24.7. Prague’s multiplier is 12.

Japanese monthly fares are strange by Western standards, in the sense that they are station-to-station, with subsegments allowed but no trips outside the segment; subject to this constraint the multiplier is 30-40, with small additional discount for buying 3-6 months in advance, but the unrestricted monthly fare is very high. London and Istanbul functionally do not have monthlies, in the sense that the multiplier is so high (78.5 Istanbul-wide, and it’s not truly unlimited but is capped at 180 trips/month) that except for trips within Central London it might as well not exist.

American and Canadian monthly fares are usually higher than in Continental Western Europe, with multipliers in the 30s. New York’s multiplier was especially high, about 46, and the MTA has just abolished the monthly fare entirely and phased out the MetroCard (as of the new year, starting in two hours), making people use the weekly cap with OMNY instead, which has a multiplier of 11.7 and, over a 30-day month, forces a monthly multiplier of 50. Toronto has a very high monthly multiplier as well, 46.6. This is bad practice: a high monthly discount functions as a technologically simple off-peak discount (indeed, London pairs its stingy monthly discount with a substantial off-peak discount), and OMNY itself is buggy to the point that fare inspectors on the buses can’t tell if someone has actually paid except by looking at debit card statements, which do not show one as having paid if one has a valid transfer or has reached the weekly cap (and not tapping in this case is still illegal fare dodging in New York law).

The practice of the cap, increasingly popular in the US under London influence, is rare as well. London’s fare cap originates in its complex zone system: the Underground has nine zones with zone 1 only covering Central London so that passengers taking multiple trips per day can expect to take trips across different zones that they may not be familiar with; there isn’t fare integration, but rather there’s a special surcharge on some commuter train trips and a discount on buses; peak and off-peak fares are different. Thus, the calculation for the passenger of whether to buy tickets one at a time or get a pass is difficult, so Oyster does this calculation automatically to give the most advantageous fare. In a Continental city where fares are either flat regionwide or have zones with limited granularity (often the entire metro is in the innermost zone) and monthly discounts are steep, the calculation is simple: an even semi-regular rider should always get a monthly.

American and Canadian cities typically have flat fares or a simple zone system, good fare integration between buses and the subway or light rail, and commuter rail that’s functionally unusable for urban trips rather than resembling the subway with a $2 surcharge. The use case of London does not apply to such cities. New York should not have a fare cap, but a heavily surcharged single trip, perhaps $5, and an attractive flat monthly fare, perhaps $130. This system ensures passengers are incentivized to pay and there is little opportunistic fare dodging as the user has already prepaid for the entire month, so it pairs well with proof-of-payment fare collection, common in many of the European examples (though metro systems outside Germany and its immediate vicinity do have faregates).

The overall level of the fare is determined by the willingness of the government at various levels to subsidize public transport; the table can be used to compare these at PPP rates as well. However, the distribution of fares across different products and distances is not a matter of subsidy but a matter of good and bad industry practices, and the best practice for simple fare collection is to offer a prepaid monthly at a heavy discount compared with the single ride.

Quick Note: The Importance of Penn Station Access West to Through-Running

A video by the Joint Transit Association talks at length about through-running in New York – which lines are easier and which are harder, what some of the tradeoffs are, what sequencing works best with ongoing infrastructure plans starting with the Gateway tunnel. It’s a good video and I recommend watching – and not just because it gets a lot of its ideas from ETA reports but also because of its own analysis and own points (about, for example, Mott Haven Junction) – but it has one miss that I’d like to highlight: it neglects Penn Station Access West, the proposal to connect the Hudson Line to Penn Station via the Empire Connection.

The issue is that without the realignment, too many trains would be going into Grand Central – all preexisting Metro-North service minus trains diverted to Penn Station Access. We expect all this through-running infrastructure to add to peak demand substantially. Today it fills about 50 peak trains per hour, which a four-track trunk line would struggle with (Metro-North runs trains three-and-one at the peak). Even with diversion of 6-10 trains to Penn Station Access, the extra demand would saturate the line. Penn Station Access West is important in reducing this capacity crunch.

The realignment is both important and cheap. The Empire Connection exists and the tunnel has room for two tracks; it needs a short realignment to reach the right part of Penn Station – the high-numbered northern tracks as in the image, where today there is a single-track link from the Connection proper to the low-numbered tracks – but that realignment is much cheaper than a full through-tunnel such as between Penn Station and Grand Central or the various lines to Lower Manhattan mooted for longer-term plans.

The total capacity produced should be every train that doesn’t have to go to Grand Central. It’s hard to exactly say what the split should be – there should be a minimum of a train every 10 minutes to each destination, if only to serve the inner stations that are (or would be infill) on the lower Hudson Line or the Empire Connection before the two routes meet at Spuyten Duyvil. Beyond that it’s a matter of measuring demand and seeing what the limit of timed connections are; ideally there should be 12 peak trains per hour on Penn Station Access West and only 6 on the preexisting route, up from 14 total on the Hudson Line today due to service improvements brought by through-running and related upgrades. This is necessary to create the capacity to run more service on the other lines – today the Harlem Line peaks at 16 trains per hour and the New Haven Line at 20, but these upgrades would create a lot more demand and my assumption in sketching through-running tunnels is that the Harlem Line would need 24 and the New Haven Line would need 18 to Grand Central and 6 on Penn Station Access.

It’s not the Baumol Effect

The Baumol effect is an observation that wages in an economy rise based on its average productivity, and therefore the wages also rise in sectors with low or no productivity growth, increasing their real costs. The classical example is that it takes the same number of people to play an opera today as in the 19th century, but wages have to be competitive with the 21st-century economy, and therefore opera tickets cost in real terms more than they did in the 19th century. People from time to time invoke this effect to explain rising infrastructure construction costs in the United States, relating it to a broader cost disease affecting health care and education. But it’s not really a correct explanation for what’s going on at global scale. High American construction costs are not downstream of high incomes, but rather of poor governance leading to low labor efficiency, poor procurement practices, nonstandard systems, and overbuilding.

At global scale, there is no significant correlation between GDP per capita and tunneling construction costs. There’s a significant but weak correlation between GDP per capita and metro construction costs, but it comes from the fact that developing countries like India build mostly elevated systems. Adjusted for the ratio between subway and elevated cost, which is about 2 in both India and China, the correlation is reduced to insignificance.

There is extensive temporal correlation between GDP per capita and costs, in the sense that the US, UK, and France were all capable of tunneling for around $40 million/km in today’s prices in the early 1900s, and aren’t capable of doing so today. But then countries with the GDP per capita of early-1900s America, like India, build subways for maybe $400 million/km. The techniques used in the early 1900s were labor-intensive, with workers digging up streets by hand. These techniques are not used today in low-income countries, which instead use capital-intensive techniques learned from Western countries, Japan, or increasingly China, and which rarely have the mass industrial working class that characterized rich cities around 1900. That is not Baumol; that is a transition to capital-intensive techniques that are then applied where it’s inappropriate.

Nor has there been an explosion of costs since the 1970s globally. The US has gone from high to very high costs, and the UK from medium to very high ones. But German construction costs are barely higher now than then. Italian ones have if anything fallen a bit, due to anti-corruption laws in the 1990s. If anything, Germany is seeing an increase in construction costs now, with rather high NBS construction costs even without tunneling, at a time in which economic growth is weak. The weak economic growth here – Germany’s GDP per capita has been essentially constant since 2019 – combined with fast economic growth in the United States means that German elites are starting to imitate American procurement practices, with Deutsche Bahn starting to use previously unheard of design-build contracts and public-private partnerships, with the attendant costs.

In the UK, similarly, high costs interact with weak growth, in that weak growth leads to cancellation of infrastructure projects like High Speed 2 north of Birmingham, which cancellation then leads to orphaned designs. There’s been growing discourse in the UK about the problem of feast-or-famine projects, with rail electrification proceeding in waves rather than at a constant rate as at Continental European comparanda like Italy. Italy is hardly posting Polish economic growth rates, but in the UK the origin of the feast-or-famine problem is in the cycle of top-down infrastructure plans and cancellations.

In truth, while the US has had higher economic growth than nearly all of Western Europe since 2019, GDP per hour remains barely above the weighted average of the Germanic-majority Continental countries and France. This is not why the US is expensive; poor project delivery is.

New York Isn’t Special

A week ago, we published a short note on driver-only metro trains, known in New York as one-person train operation or OPTO. New York is nearly unique globally in running metro trains with both a driver and a conductor, and from time to time reformers have suggested switching to OPTO, so far only succeeding in edge cases such as a few short off-peak trains. A bill passed the state legislature banning OPTO nearly unanimously, but the governor has so far neither signed nor vetoed it. The New York Times covered our report rather favorably, and the usual suspects, in this case union leadership, are pissed. Transportation Workers Union head John Samuelsen made the usual argument, but highlighted how special New York is.

“Academics think working people are stupid,” [Samuelsen] said. “They can make data lie for them. They conducted a study of subway systems worldwide. But there’s no subway system in the world like the NYC subway system.”

Our report was short and didn’t go into all the ways New York isn’t special, so let me elaborate here:

  • On pre-corona numbers, New York’s urban rail network ranked 12th in the world in ridership, and that’s with a lot of London commuter rail ridership excluded, including which would likely put London ahead and New York 13th.
  • New York was among the first cities in the world to open its subway – but London, Budapest, Chicago (dating from the electrification and opening of the Loop in 1897), Boston, Paris, and Berlin all opened earlier.
  • New York has some tight curves on its tracks, but the minimum curve radius on Paris Métro Line 1, 40 meters, is comparable to the New York City Subway’s.
  • The trains on the New York City Subway are atypically long for a metro system, at 151 meters on most of the A division and 183 on most of the B division, but trains on some metro systems are even longer (Tokyo has some 200 m trains, Shanghai 180 m trains) and so are trains on commuter rail systems like the RER (204 m on the B, 220 m on the A), Munich S-Bahn (201 m), and Elizabeth line (205 m, extendable to 240).
  • New York has crowded trains at rush hour, with pre-Second Avenue Subway trains peaking at 4 standees per square meter, but London peaks at 5/m^2 and trains in Tokyo and the bigger Chinese cities at more than that. Overall ridership, irrespective of crowding, peaked around 30,000 passengers per direction per hour on the 4 and 5 trains in New York, compared with 55,000 on the RER A.

New York is not special, not in 2025, when it’s one of many megacities with large subway systems. It’s just solipsistic, run by managers and labor leaders who are used to denigrating cities that are superior to New York in every way they run their metro systems as mere villages unworthy of their attention. Both groups are overpaid: management is hired from pipelines that expect master-of-the-universe pay and think Sweden is a lower-wage society, and labor faces such hurdles with the seniority system that new hires get bad shifts and to get enough workers New York City Transit has had to pay $85,000 at start, compared with, in PPP terms, around $63,000 in Munich after recent negotiations. The incentive in New York should be to automate aggressively, and look for ways to increase worker churn and not to turn people who earn 2050s wages for 1950s productivity be a veto point to anything.

High Speed Rail-Airport Links

As somewhat of a followup to my last post on how successful high-speed rail isn’t really made for tourists, I’d like to talk about the issue of air-rail links. Those are beloved by both foreign tourists and domestic residents using them to travel abroad, and American high-speed rail planning has on occasion tried focusing on them. This has always been awkward for both environmental and ridership goals. Such links are not inherently bad, but they are often overrated in planning, especially at the level of public advocacy and shadow planning agencies, which reproduce the biases of frequent fliers.

Skipping the airports in rich Asia

The Shinkansen does not serve Narita. There were plans for it to do so but they have not been implemented. Such service would require a dedicated line, since the Shinkansen is on a different gauge from the classical JR network and the standard-gauge link between the city and the airport is owned by private railway Keisei, and Narita itself is not important enough to drive such a line, not at the urban tunneling construction costs of Japan.

But the same lack of service to airports is seen in the two most Shinkansen-like systems outside Japan, Korea and especially Taiwan. The airport is not in Taipei but in Taoyuan, and is connected to the city by an express commuter train, the Taoyuan Airport MRT, but the Taiwan High-Speed Rail system does not serve it, instead having a different Taoyuan station on the Airport MRT. Even in Korea, which uses standard gauge and runs KTX trains through on classical lines in the French style, there is no KTX service to Incheon or to Gimpo.

The issue in all three countries is that the role of the capital’s international airport is to connect passengers between the capital region and the rest of the world. Tourists visiting the capital don’t need a train to secondary cities; in South Korea, last year, 66% of tourism by spending was in Seoul, and in Taiwan, 53% of tourism by occupied hotel nights was in Taipei, New Taipei, and Taoyuan (PDF-pp. 20-21 of the 2024 annual report). Domestic residents using the airport to travel abroad are a more serious use case, but far more residents of Busan or Kaohsiung are going to their respective country’s capital than abroad, and so the airport link is not a high priority for planning.

Serving the airports in Europe if they’re on the way

Three of the four busiest airports in the EU – CDG, Schiphol, and Frankfurt (the fourth is Barajas) – have high-speed rail links. However, in all cases, it’s because they’re on the way somewhere. CDG and Frankfurt are both on valuable bypass routes around the primary city with its terminal-only train stations, so they might as well be served. Schiphol is between Rotterdam and Amsterdam, but serving it involved high-cost tunneling, on a high-speed line, HSL Zuid, that has in retrospect been more a case of imitating the TGV than responding to Dutch intercity rail needs.

In all cases, the airport link is decidedly secondary to the network, and is not a major planning goal. There are intercity trains routed into Berlin-Brandenburg, but these are intended for long-distance regional use: the extensive rail tunneling to the new airport is for various regional express trains, with a 15-minute Takt to Berlin Hauptbahnhof and four hourly Takt trains to regional destinations starting next month and only one intercity train on a two-hour Takt between Berlin and Dresden. Munich has no ICE connection, and a proposal for one never got beyond the conceptual stage because the airport-city center connection was deemed a higher priority. It’s notable that even high-cost, high-prestige air-rail links here prioritize connections to city center, and not to the national network.

The awkward environmental politics of air-rail links

High-speed rail is justified on both economic and environmental grounds. But sometimes these different justifications end up conflicting. It’s noteworthy that in the United States, a common argument for high-speed rail in California and the Northeast has been that the airports are too clogged with short-haul regional flights and if high-speed trains replaced them then the gates and runway slots would be usable by long-haul flights. This argument is made at the same time as arguments about reducing greenhouse gas emissions – but long-haul flights contribute far more emissions than short-haul ones per unit of airport capacity consumed, airport capacity not particularly caring if you’re flying 700 km or 7,000.

It’s possible to ignore the environmental effects and just focus on the economic benefits; in Europe, the broad environmental movement is neutral or even hostile to high-speed rail, viewing it as inferior to running more night trains and regional trains. But then in Europe the economic-only planning for high-speed rail does not prioritize the air links, because they are fundamentally secondary. In a country like France, the demand for high-fare rail links to CDG is to the center of Paris, not Marseille.

High-Speed Rail is not for Tourists

Foreigners to a country often get a warped idea of what its infrastructure is like. Most infrastructure is used for day-to-day domestic travel, for commuting to work or school, for visits to family and friends, for social gatherings, for business travel within the national internal market. Foreign travelers make use of this infrastructure when they visit, but they use it differently, and can make erroneous assumptions about how locals use it and what it means for transportation in general. This has two policy implications: one concerns American misconceptions about European rail travel; the other concerns pan-European misconceptions about European rail travel, which is almost entirely domestic, based on domestic networks, and planned and debated in the local language and not in English.

The Europe of the tourists

To estimate how foreign tourists may view Europe, we need some information on tourist travel within the bloc. The best I have is lists of the most visited cities in the world, and unfortunately, the only lists I have that go beyond the global top 10 are from before corona. But 2019 should not be too different to first order from the present. Here are international arrivals, from the global top 50:

CityMillions of arrivals (2019)
London19.55
Paris19.08
Istanbul14.71
Rome10.31
Prague9.15
Amsterdam8.83
Barcelona7.01
Vienna6.63
Milan6.6
Athens6.3
Berlin6.19
Moscow5.96
Venice5.59
Madrid5.59
Dublin5.46

Notably, there’s almost no intersection with any of the busiest intercity rail links in Europe. The top two are the trunk from Paris on the LGV Sud-Est to the bifurcation between Dijon and Lyon, and the Frankfurt-Mannheim trunk line. Paris is a huge international tourist draw, but nothing on the LGV Sud-Est and its extensions is; the top department outside Ile-de-France in tourism overnight stays is Alpes-Maritimes, a 5.5-6 hour trip from Paris by TGV. Germany has little tourism for its size, especially not in Mannheim – foreigners come to Berlin or Munich, or maybe Frankfurt for business trips. Only two city pairs in Europe with solid high-speed rail links appear in the table above, Milan-Rome and Madrid-Barcelona.

The upshot is that the American tourist who comes here and marvels at the fact that even in Germany the trains are faster and more reliable than in the United States isn’t really experiencing the system as most users do. If they take the TGV, it’s much likelier that they’re taking Eurostar and dealing with its premium prices and probably also with its security theater if they’re going to London rather than Brussels or Amsterdam. They have nothing to do in Lyon or Bordeaux or Strasbourg or Lille, so it’s unlikely they see the workhorse domestic lines. It’s even more unlikely they take the train to the smaller cities with direct TGVs, such as Saint-Etienne, Chambéry, and others that beef up the ridership of the LGV Sud-Est without serving Lyon itself; there were considerable errors made by American analysts in the Obama era about high-speed rail coming from looking only at the million-plus metro areas and not at these secondary ones.

By the same token, the American tourist in question is much likelier to be riding Spanish trains with their brand and price differentiation by speed than to be riding the workhorse regional and intercity trains anywhere in Northern Europe. ICEs charitably average 160 km/h on a handful of lines when they’re on time, which isn’t often, and on key corridors like Berlin-Cologne or Berlin-Frankfurt are closer to 120 km/h. The reason Germany is close to even with France on ridership per capita and well ahead of Italy and Spain is that these trains have decent connections with one another and with slower regional trains, so that people can connect to those secondary cities better. Trips from Berlin to Augsburg with a connection in Munich are not hard to plan, or trips to city cores in the Rhine-Ruhr and other polycentric regions. These are largely invisible to the foreign tourist, who doesn’t have anything to do in a city like Münster.

This also applies to the European tourist, not just the American or Asian or Middle Eastern one. A German who visits France is interested in trains from Germany to Paris, and those are not that good, but will probably not be taking TGVs between Paris and Rennes or Lille. From that, they’ll conclude the TGVs aren’t that useful in general.

The Europe of the typical intercity rail traveler

In contrast with the tourists’ picture of the countries of Europe, the typical intercity rail traveler uses the system in a way that the table above doesn’t really capture. All of the following characteristics are likely:

  • They are traveling domestically since cross-border rail within Europe is practically never good.
  • They are traveling based on domestic business, leisure, and social networks: if French, they can be going between Paris and anywhere else in France, and very occasionally even between two places outside Ile-de-France; if German, they are likely going between two major cities or maybe between a major and a midsize city.
  • They are a regular traveler, which implies good knowledge of the system and its quirks, experience with large complex stations allowing getting between the train and the street within minutes, and probably also some kind of discounted fare card such as the BahnCard 25 in Germany or the half-fare card of Switzerland.
  • They have the disposable income to drive, and choose to take the train because of a combination of speed, fares, and convenience rather than because they truly can’t afford a car or because they are ideologically opposed to travel modes with high greenhouse gas emissions.

The upshot is that finicky systems like the TGV and ICE are useful to their current travelers, even if foreigners and people who move in pan-European networks find them unreliable for various reasons. Any kind of EU-wide policy on rail has to acknowledge that SNCF and DB may have problems but are the main providers of solid intercity rail within Europe and are not the enemy, they just focus on city pairs that reflect their domestic travel needs.

And any attempt to learn from Europe and adapt our intercity rail successes has to look beyond what a tourist visiting for a few days would notice. It’s not just the wow effect of speed; Eurostar has that too and its ridership is an embarrassment, with fewer London-Paris trips per day than Paris-Lyon even though metro London is around six times the size of metro Lyon. It’s other details of the network, including how far it reaches into the longer tail of secondary markets.

The secondary markets require especial concern, first because they form a large fraction (likely a majority) of European high-speed rail travel, second because they’re invisible to tourists, and third because they require careful optimization.

One issue is that secondary markets are great for cars, decent for trains, and awful for planes. The TGV owns them at distances where cars take too many hours longer than the train, which helps extend the trains well past the three- to four-hour limit that rail executives quote as the upper bound for competitive train trip time. At shorter range, high-speed rail competes with cars more than with planes, and so the secondary markets lose value.

Another issue is that it’s easy to overdo secondary markets at the expense of compromising speed on the primary ones. This is usually not because of tourists, who almost never ride them, but rather because of domestic travelers who are atypically familiar with and dependent on the system and will use it not just on city pairs like Berlin-Augsburg or Berlin-Münster but also things like Wismar-Jena, on which most people will just drive. In the United States, groups of users of Amtrak trains outside the Northeast Corridor like the Rail Passengers’ Association (RPA, distinct from the New York-area planning organization) routinely make this mistake and overrate the viability of slow night trains. I bring this up here because it is possible to overcorrect from the principle of “don’t rely on tourist reports too much, and do pay attention to the secondary markets” and instead pay too much attention to the secondary markets.

Why is Janno Lieber Constantly Blaming Other People for Problems?

The Editorial Board posted an interview with MTA head Janno Lieber about sundry public transit-related issues. His answers for the most part aren’t bad until he gets to construction costs (and misgenders me), but alongside other recent news about Penn Station Access, they reveal a pattern: Lieber loves blaming other people for problems – nothing is ever the MTA’s fault, everything is someone else’s fault. Nor is he curious about acquiring expertise, to the point that everything is defensive, and everything is about reducing transparency and accountability. Someone like this should not be heading a public transit agency.

Penn Station Access

Penn Station Access, the project to run Metro-North trains from New Rochelle to Penn Station via the Hell Gate Line currently used only by Amtrak, was announced earlier this month to be delayed by a further two years, from 2028 to 2030. The MTA blames Amtrak, which owns most of the line, for not giving it enough work windows.

And, excuse me, but this is bullshit for two separate reasons. The first is that the opening date was said to be 2027 until this year and then 2028. Other people made plans based on MTA announcements; quite a lot of behind-the-scenes advocacy was designed specifically around this date. The state was among those other people: in March, it decided to buy new battery-powered locomotives, each costing $23.45 million (about the same as an eight-car EMU set), on the grounds that it would take too long to acquire new EMUs that were compatible with the different electrification systems used on the line. It’s not at all hard to get new EMUs compatible with both the 12 kV 60 Hz electrification used on most of the line and the 12 kV 25 Hz system used in the last few km into Penn Station based on current New York lead times if the project opens in 2030. But the state made a decision based on the assumption it would need this well before 2030.

In other words, the MTA only discovered that there would be Amtrak-induced delays around two and a half years before planned opening for a project that had been going on for three years and approved for six – and now it’s blaming it on Amtrak instead of on its own poor project management and lack of transparency.

The second reason it’s bullshit is that the relationship between Amtrak and the MTA is mutually abusive. Amtrak is not giving the MTA enough work windows on the Hell Gate Line; the MTA is slowing down Amtrak trains on the New Haven Line between New Rochelle and New Haven, where it owns the tracks, the only part of the Northeast Corridor that is both owned and dispatched by a commuter railroad and not Amtrak (in Massachusetts the MBTA owns the tracks but Amtrak controls dispatching). The maximum allowed cant deficiency on Metro-North territory is based on unmodernized Metro-North values and not based on the modern values that Amtrak rolling stock has been tested for, and there is no attempt to keep Amtrak and Metro-North trains separate east of Stamford, where there are four tracks and light enough traffic that it’s possible, that the top speeds can have a mismatch.

In other words, the MTA complains about being abused by Amtrak, and is likely correct, but refuses to stop abusing Amtrak where it does have control. It could manage this relationship better, but it doesn’t and Lieber isn’t competent enough to know how to do it better.

Fares

The conversation in The Editorial Board heavily features talking about fares, in context of fare evasion and mayoral frontrunner Zohran Mamdani’s proposal for free buses. Lieber is suggesting that instead of free buses, buses can have all-door boarding without free fares, unlocking the speed benefits without forgoing the revenue. He’s right and I want to sympathize with his critique of free buses. But it was Lieber who scuttled plans by Andy Byford to install back-door OMNY card readers and enable all-door boarding without free fares. He calls for all-door boarding as an alternative to free buses now, but when all-door boarding was available as an internally developed plan, he killed it.

He speaks about Europe this and Europe that in the interview, but he’s too ignorant and incurious to understand how things go here and how we make all-door boarding work with proof-of-payment. And the best way to see that is his abominable line, “had a kid who did a semester abroad in Stockholm, and you see them all over in Europe.” That’s his only reference – his kid did a semester abroad. He didn’t ring up any transit agency to ask how to do it. It’s all superficial, almost tourist-level understanding of better-run systems.

This is especially bad in context of what he says about construction costs at the end. He says,

I don’t accept the Alon Levy theory, which, you know, you’re articulating — that somehow, if we just had like this massive in-house force, we would be building everything way, way cheaper. That’s like, hiring— you cannot compete with private-sector engineering. And we don’t have one project after another, like he loves, like Madrid, which built all these subways in a row.

Setting aside the fact that calling me “he” in New York, a city with better access to gender-neutral bathrooms than my own, is obnoxious, we didn’t do a report on Madrid, but did do one on Stockholm. He’s aware of the report (and of the points it makes about ridership per station, the excuse he uses farther down the line for bigger stations). And he still reduces Stockholm to where his kid did a one-semester study abroad to give a little anecdote on fare evasion, which boils down to Americans being so detached from internal national discourses in Europe (except maybe the UK) that they don’t know that we’ve had to deal with the same questions they did, we just have public agencies run by competent people who sometimes make the right decisions and not by people like Janno Lieber.

Reverse-Branching on Commuter Rail

Koji asked me 3.5 days ago about why my proposal for New York commuter rail through-tunnels has so much reverse-branching. I promised I’d post in some more detail, because in truth, reverse-branching is practically inevitable on every commuter rail system with multiple trunk lines, even systems that are rather metro-like like the RER or the S-Bahns here and in Hamburg.

Berlin S-Bahn schematic. Source: Wikipedia.

This doesn’t mean that reverse-branches, in this case the split from the Görlitzer Bahn trunk toward the Stadtbahn via S9 and the Ring in two different directions via S45/46/47 and S8/85, are good. It would be better if Berlin invested in turning this trunk into a single trunk into city center, provided it were ready to build a third through-city line (in fact, it is, but this project, S21, essentially twins the North-South Tunnel). However, given the infrastructure or small changes to it, the current situation is unavoidable.

Moreover, the current situation is not the end of the world. The reasons such reverse-branches are not good for the health of the system are as follows:

  • They often end up creating more frequency outside city center than toward it.
  • If there is too much interlining, then delays on one branch cascade to the others, making the system more fragile.
  • If there is too much interlining, then it’s harder to write timetables that satisfy every constraint of a merge point, even before we take delays into account.

All of these issues are more pressing on a metro system than on a commuter rail system. The extent of branching on commuter rail is such that running each line as a separate system is unrealistic; tight timetabling is required no matter what, and in that case, the lines could reverse-branch if there’s no alternative without much loss of capacity. The S-Bahn here is notoriously unreliable, but that’s the case even without cascading delays on reverse-branches – the system just assumes more weekend shutdowns, less reliable systems (28,000 annual elevator outages compared with 1,800 on the similar-size U-Bahn), and worse maintenance practices.

So, on the one hand, the loss from reverse-branching is reduced. On the other hand, it’s harder to avoid reverse-branching on commuter rail. The reason is that, unlike a metro (including a suburban metro), the point of the system is to use old commuter lines and connect them to form a usable urban and suburban service. Because the system relies on old lines more, it’s less likely that they’re at the right places for good connections. In the case of Berlin, it’s that there’s an east-west imbalance that forces some east-center-east lines via S8, which was reinforced by the context of the Cold War and the Wall.

In the case of New York, consider this map:

The issue is that too much traffic wants to use the Northeast Corridor lines in both New Jersey and Connecticut. Therefore, it’s not possible to segregate everything, with lines using the preexisting North River Tunnels and the new Gateway tunnels having to share tracks. It’s not optimal, but it’s what’s possible.