Poor Rich Countries and Isomorphic Mimicry

A curious pattern can be found in subway construction costs around the world, based on GDP per capita. On the one hand, poor countries that have severe cultural cringe, such as former colonies, have high construction costs, and often the worst projects are the ones that most try to imitate richer countries, outsourcing design to Japan or perhaps China. On the other hand, poor-rich countries, by which I mean countries on the periphery of the developed world, have similar cultural cringe and self-hate for their institutions, and yet their imitation of richer countries has been a success; for example, Spain copied a lot of rail development ideas from Germany and France. This can be explained using the development economic theory of isomorphic mimicry; the rub here is that a poor country like India or Ethiopia is profoundly different from the richer countries it tries to imitate, whereas a poor-rich country like Spain is actually pretty similar to Germany by global standards.

What is isomorphic mimicry?

In the economic development literature, the expression isomorphic mimicry refers to when a poor country sets up institutions that aim to imitate those of richer countries in hope that through such institutions the country will become rich too, but the imitation is too shallow to be useful. A common set of examples is well-meaning regulations on safety, labor, environmental protection, and anti-corruption that are not enforced due to insufficient state capacity. Here is a review of the concept by Andrews, Pritchett, and Woolcock, with examples from Mozambique, Uganda, and India, as well as some history from the American private sector. More examples using the theory can be found in Turczynowicz, Gautam, Rénique, Yeap, and Sagues concerning Peru’s one laptop per child program, in Evans’ interpretation of Bangladesh’s domestic violence laws, and in Rajagopalan and Tabarrok on India’s poor state of public services.

While the theory regarding institutions is new, analogs of it for tangible goods are older. Postwar developmental states engaged in extensive isomorphic mimicry, building dams, steel plants, and coal plants hoping that it would transform them into wealthy states like the United States, Western Europe, and Japan; for the most part, they had lower economic growth than did the developed world until the 1980s. The shift within international development away from tangible infrastructure and toward trying to fix institutions came about because big projects like the Aswan Dam failed to create enduring economic growth and often had ill side effects on agriculture, the environment, or human rights.

How does isomorphic mimicry affect public transportation?

The best example of isomorphic mimicry leading to bad transit that I know of is the Addis Ababa light rail system. This is funded by China, whose ideas of global development are similar to those of the postwar first and second worlds, that is providing tangible physical things, like railroads. Unfortunately, usage is low, because of problems that do not exist in middle-income or rich countries but are endemic to Ethiopia. Christina Goldbaum, the New York Times’ transit reporter, who lived in East Africa and reported from Addis Ababa, mentioned four problems:

  1. Electricity is unreliable, so the trains sometimes do not work. In early-20th century America, electric railroads and streetcar companies built their own power supply and were sometimes integrated concerns providing both streetcar and power service; but in more modern countries, there is reliable power for urban rail to tap.
  2. Not many people work in city center rather than in the neighborhood they live in. This, again, has historical analogs – there were turn-of-the-century Brooklynites who never visited Manhattan. Thus, a downtown-centric light rail system won’t get as much ridership as in a more developed city.
  3. The train is expensive relative to local incomes, so many people stick with buses or ride without paying.
  4. The railroad cuts through streets at-grade, to save money, and blocks off pedestrian paths that people use.

The Addis Ababa light rail system at least had reasonable costs. A more typical case for countries that poor is to build urban rail at premium cost, and the poorer the country, the higher the cost. The reason is most likely that such countries tend to build with Chinese or Japanese technical assistance, depending on geopolitics, and therefore import expensive capital for which they pay with weak currencies.

In India, the most functional and richest of the countries in question, there is much internal and external criticism that its economic growth is not labor-intensive, that is the most productive firms are not the ones employing the most people, and this stymies social development and urban growth. I suspect that this also means there is reluctance to use labor-intensive construction methods, that is cut-and-cover with headcounts that would be typical in New York, Paris, and Berlin in the early 20th century, or perhaps mid-20th century Milan and Tokyo. International consultancies are centered on the rich world and recommend capital-intensive methods to avoid hiring too many sandhogs at a fully laden employment cost of perhaps 8,000€ a month; in India, that is the PPP-adjusted gross salary of an experienced construction worker per year, and if capital is imported then multiply its cost by 3 to account for the rupee’s exchange rate value.

Poor-rich countries

Poor-rich countries are those on the margin of the developed world, such as the countries of Eastern and Southern Europe, Turkey, Israel, to a lesser extent South Korea, and the richer countries of Latin America such as Chile. These are clearly poorer than the United States or Germany. Their residents, everywhere I’ve asked, believe that they are poorer and institutionally inferior; convincing a Spaniard or an Italian that their country can do engineering better than Germany is a difficult task. Thus, these countries tend to engage in mimicry of those countries that they consider the economic center, which could be Germany in Southern Europe, Japan in South Korea, or the US or Spain in Spanish America.

However, being a poor-rich country is not the same as being a poor country. Italy is, by American or German standards, poor. Wages there are noticeably lower and living standards are visibly poorer, and not just in the South either. But those wages remain in the same sphere as American and German wages. The labor-capital cost ratios in Southern Europe are sufficiently similar to those of Northern Europe that it’s not difficult to imitate. Spain even mixed and matched, using French TGV technology for early high-speed rail but preferring the more advanced German intercity rail signaling system, LZB, to the French one.

Such imitation leads to learning. Spain imported German and French engineering ideas but not French tolerance for casual rioting or German litigiousness, and therefore can build infrastructure with less NIMBYism. Turkey invited Italian consultants to help design the early lines of the Istanbul Metro, but subsequently refined their ideas domestically in order to build more efficiently, for example shrinking station footprint and tunnel diameter to reduce costs. Seoul has a subway system that looks like Tokyo’s in many ways, but has a cleaner network shape, with far fewer missed connections between lines. As a result, all three countries – Spain, Turkey, Korea – now have innovative domestic programs of rail construction and can even export their expertise elsewhere, as Spain is in Ecuador.

Openness to novelty

Andrews-Pritchett-Woolcock stress the importance of openness to novelty in the public sector, and cite examples of failure in which bureaucrats at various levels refused to implement any change, even one that was proven to be positive, because their goal was not to rock the boat.

Cultural cringe is in a way a check on that. Isomorphic mimicry is an attempt to combine agenda conformity and closeness to novelty with a desire to have what the richest countries have. But in poor-rich countries, isomorphic mimicry is real imitation – there is ample state penetration in a country like Spain or Turkey rather than outsourcing of state capacity to traditional heads of remote villages, and education levels are high enough that many people know how Germany works and interact with Germany regularly. A worker who earns 2,000€ a month net and a worker who earns 3,000€ a month can exchange tips about how to apply for jobs, how to prepare food, what brands of consumer goods to buy, and where to go on vacation. They cannot have this conversation with a worker who earns 10,000€ a month net.

Within the rich world, what matters then is the realization that something is wrong and the solution is to look abroad. It doesn’t matter if it’s a generally poor-rich region like Southern Europe or a region with a poor-rich public sector like the United States – there’s enough private knowledge about how successful places work, but what’s needed is a public acknowledgement and social organization encouraging imitation and lifting voices that are most expert in implementing it.

And for all the jokes about how the United States or Britain is like a third-world country, they really aren’t. Their public-sector dysfunctions are real, but are still firmly within the poor-rich basket; remember, for example, that despite its antediluvian signaling capacity, the New York City Subway manages to run 24 trains per hour per track at the peak, which is better than Shanghai’s 21. Health and education outcomes in the United States are generally better than those of middle-income and poor countries on every measure. This is a public sector that compares poorly with innovation centers in Continental Europe and democratic East Asia, but it still compares; to try to do the same comparison in a country like Nigeria would be nonsensical.

The upshot then is that implementing best practices in developed countries that happen to be bad at one thing, in this case public transportation in the United States, can work smoothly, much like Southern Europe’s successful assimilation of and improvements on Northern European engineering, and unlike the failures in former colonies in Africa and Asia. But people need to understand that they need to do it – that the centers of innovation are abroad and are in particular in countries that speak English non-natively.

Sorry Eno, the US Really Has a Construction Cost Premium

There’s a study by Eno looking at urban rail construction costs, comparing the US to Europe. When it came out last month I was asked to post about it, and after some Patreon polling in which other posts ranked ahead, here it goes. In short: the study has some interesting analysis of the American cost premium, but suffers from some shortcomings, particularly with the comprehensiveness of the non-American data. Moreover, while most of the analysis in the body of the study is solid, the executive summary-level analysis is incorrect. Streetsblog got a quote from Eno saying there is no US premium, and on a panel at Tri-State a week ago T4A’s Beth Osborne cited the same study to say that the US isn’t so bad by European standards, which is false, and does not follow from the analysis. The reality is that the American cost premium is real and large – larger than Eno thinks, and in particular much larger than the senior managers at Eno who have been feeding these false quotes to the press think.

What’s the study?

Like our research group at Marron, Eno is comparing American urban rail construction costs per kilometer with other projects around the world. Three key differences are notable:

  1. Eno looks at light rail and not just rapid transit. We have included a smattering of projects that are called light rail but are predominantly rapid transit, such as Stadtbahns, the Green Line Extension in Boston, and surface portions of some regional rail lines (e.g. in Turkey), but the vast majority of our database is full rapid transit, mostly underground and not elevated. This means that Eno has a mostly complete database for American urban rail, which is by construction length mostly light rail and not subways, whereas we have gaps in the United States.
  2. Eno only compares the United States with other Western countries, on the grounds that they are the most similar. There is a fair amount of Canada in their database, one Australian line, and a lot of Europe, but no high-income Asia at all. Nor do they look at developing countries, or even upper-middle-income ones like Turkey.
  3. Eno’s database in Europe is incomplete. In particular, it looks by country, including lines in Britain, Spain, Italy, Germany, Austria, the Netherlands, and France, but even there it has coverage gaps, and there is no Switzerland, little Scandinavia (in particular, no ongoing Stockholm subway expansion), and no Eastern Europe.

The analysis is similar to ours, i.e. they look at average costs per km controlling for how much of the line is underground. They include one additional unit of analysis that we don’t, which is station spacing; ex ante one expects closer station spacing to correlate with higher costs, since stations are a significant chunk of the cost and this is especially notable for very expensive projects.

The main finding in the Eno study is that the US has a significant cost premium over Europe and Canada. The key here is figure 5 on takeaway 4. All costs are in millions of PPP dollars per kilometer.

Tunnel proportionMedian US costMedian non-US cost

However, the study lowballs the US premium in two distinct ways: poor regression use, and upward bias of non-US data.

Regression and costs

The quotes saying the US has no cost premium over Europe come from takeaways 2 and 3. Those are regression analyses comparing cost per km to the tunnel proportion (takeaway 3) or at-grade proportion (takeaway 2). There are two separate regression lines for each of the two takeaways, one looking at US projects and one at non-US ones. In both cases, the American regression line is well over the European-and-Canadian line for tunneled projects but the lines intersect roughly when the line goes to 0% underground. This leads to the conclusion that the US has no premium over Europe for light rail projects. Moreover, because the US has outliers in New York, the study concludes that there is no US premium outside New York. Unfortunately, these conclusions are both false.

The reason the regression lines intersect is that regression is a linear technique. The best fit line for the US construction cost per km relative to tunnel proportion has a y-intercept that is similar to the best fit line for Europe. However, visual inspection of the scattergram in takeaway 3 shows that at 0% underground, most US projects are somewhat more expensive than most European projects; this is confirmed in takeaway 4. All this means that the US has an unusually large premium for tunneled projects, driven by the fact that the highest-cost part of the US, New York, builds fully-underground subways and not els or light rail. If instead of Second Avenue Subway and the 7 extension New York had built high-cost els, for example the plans for a PATH extension to Newark Airport, then a regression line would show a large US premium for elevated projects but not so much for tunnels.

I tag this post “good/interesting studies” and not just “shoddy studies” because the inclusion of takeaway 4 makes this clear: there is a US premium for light rail, it’s just smaller than for subways, and then regression analysis can falsely make this premium disappear. This is an error, but an interesting one, and I urge people who use statistics and data science to study the difference between takeaways 2 and 3 and takeaway 4 carefully, to avoid making the same error in their own work.

Upward bias

Eno has a link to its dataset, from which one can see which projects are included. It’s notable that Eno is comprehensive within the United States, but not in Europe. Unfortunately, this introduces a bias into the data, because it’s easier to find information about expensive projects than about cheap ones. Big projects are covered in the media, especially if there are cost overruns to report. There is also a big-city premium because it’s more complicated to build line 14 of a metro system than to build line 1, and this likewise biases incomplete data because it’s easier to find what goes on in Paris than to find what goes on in a sleepy provincial town like Besançon. Yonah Freemark thankfully has good coverage of France and includes low-cost Besançon, but Eno does not – its French light rail database is heavy on Paris and has big gaps in the provinces. French Wikipedia in fact has a list, and all of the listed systems, which are provincial, have lower costs than Paris.

There is also no coverage of German tramways; we don’t have such coverage either, since there are many small projects and they’re in small cities like Bielefeld, but my understanding is that they are not very expensive. Traditionally German rail advocates held the cost of a tramway to be €10 million/km, which is clearly too low for the 2010s, but it should lower the median cost compared to the Paris-heavy, Britain-heavy Eno database.

Friends Don’t Let Friends Build PPPs

Three examples of public-private partnerships screwing up urban transit are on my mind. The Canada Line in Vancouver is not new to me – I was poking around a few years ago. But the other two in this post are. The Maryland Purple Line in the suburbs of Washington was supposed to be the smooth PPP offering low-risk orbital light rail connecting suburbs to other suburbs without having to go through Downtown Washington, and now it is in shambles because the contractor walked away. Milan is not a new example either, but it is new to me, as we’ve discovered it during the construction costs project comparing high American (and British) costs to low Southern European ones; even there, the PPP bug bit, leading not so much to high capital costs but to high future operating charges. In no case is such a PPP program good government; the bulk of construction and risk must always lie in the public sector, and if your public sector is too incompetent to build things itself, as in the United States, then it’s equally incompetent at overseeing a PPP, as we’re seeing in Maryland. Don’t do this.

Washington: the Purple Line

Maryland planned on building two major urban rail projects last decade, stretching into the current one: the Red Line and the Purple Line. The Red Line was to be a conventional public project to build a subway in Baltimore, mostly serving low-income West Baltimore neighborhoods. The Purple Line, a light rail project in the DC suburbs acting as an orbital for Metro, was designed as a PPP. Governor Larry Hogan canceled the Red Line, most likely for racist reasons. The physical construction costs per rider were higher on the Red Line, but the overall disbursement including very high operating charges made the Purple Line more expensive, and yet Hogan kept the more expensive system and tossed the cheaper one.

One might expect that the PPP structure of the Maryland Purple Line would allow it to at least resist cost escalation – the risk was put entirely on the private contractor. And yet, the opposite happened. Costs turned out to be higher than expected, so the contractor just quit. Once the contract is signed, no matter what it says, the risk is in practice public, and this is no exception. The contractor stopped all work and left the region with a linear swath of ripped up roads; eventually the concessionaire and the state came into a settlement in which the state would pay $250 million extra and the concessionaire would hire a new contractor. The cost overrun was $800 million and the state said that the deal was going to save taxpayers $500 million, but what it signals is that even with very high public-sector payouts over decades that intend to put the entirety of the risk on the private concession, the public sector shares a high proportion of the risk, and the private bidders know this. This is a lose-lose situation and under no circumstances should countries put themselves in it.


Vancouver provides another good example of PPPs and operating costs. SkyTrain operates driverless equipment throughout the system, which means that operating costs should be low, and, moreover, should not depend on train size much. The Expo and Millennium Lines, built and operated publicly, cost C$3.20 to run per car-km, cheaper than on any system for which I have data (mostly very large ones plus Oslo) and less than half as expensive as the major European systems. But the Canada Line, operated by a concessionaire as part of a PPP scheme, costs $17.90/car-km, which is considerably worse than any system for which I have data except PATH. Even taking into account that the Canada Line cars are somewhat bigger, this is a difference of a factor of more than 3.

This is not a matter of economies of scale. The Canada Line’s trunk runs every 3.5 minutes most of the day, which is better than the vast majority of non-driverless systems I am familiar with off-peak, so the high costs there cannot be ascribed to poor utilization. In fact, before the Evergreen extension of the Millennium Line opened in 2016, the two systems’ total operating costs were almost identical but the operating costs per car-km were about 3.5 times worse on the Canada Line – economies of scale predict that unit costs should be degressive, not almost flat.


Marco Chitti is busy collecting information and conducting interviews regarding subway construction in Italy as part of our construction costs report. Italian costs are low, which makes it feasible to build metros even in very small cities like Brescia, where per Wikipedia the cost of the metro was around €65 million per km and €15,000 per weekday rider. However, the use of PPPs has not been good in the places where it happened, due to fiscal austerity following the Great Recession.

  • What is the impact on the cost of the PPP? The impact on costs of the potential transfer of risk from the Public to the Private is hard to calculate, but it appears to have an impact more on higher gross operational costs (the fee that the Municipality will pay in the 26 years of the concession for the operation and pay back a return to the private operators) than on the actual construction cost. But that is unclear yet. A bit of detail: the municipality will pay to the concessionaire a 1.09 €/passenger as a minimum granted fee up to 84 million passengers/year, 0.45€/passenger for each additional user up to a maximum determined as an increase of the IRR of 2 percentage points more than the “base IRR” of 5.93%. That means that this is basically the rate at which the private investors are de facto borrowing the money to the municipality, with most of the risk from low ridership transferred to the municipality. What makes calculations complicate is that the city is directly a majority stakeholder of the concessionaire Metro M4 S.p.A. and also, indirectly, as the owner of ATM, which will be the “private” operator. It’s very blurred compared to other PPP schemes where the concessionaire is 100% private (like M5).
  • PPP emerges as a stratagem to finance the project without increasing the municipal public debt. The PPP schemes is used to compensate for the lack of local public funds matching the national ones, limited due to the debt cap imposed by the so-called “internal Stability Compact”, an austerity measure implemented after the 2011 debt crisis, which strongly limits the capacity of local governments to borrow money for infrastructure projects. It was suspended in 2016.

Note that contra the plan to build the system without public debt, the PPP does in fact include borrowing. It’s opaque, but the payment per rider is a form of borrowing. Driverless metro operating costs are lower than €1.09 per unlinked trip. The Expo and Millennium Lines cost C$1.55, which in PPP terms is about €0.90, and feature much longer trips, as the Expo Line is 36 km long and one-tailed, which means many people ride end-to-end, whereas Milan M4 is to be 15 km and two-tailed, which means few trips are longer than half the total. In effect, this is high-interest borrowing, kept off the books in an atmosphere of strict budgetary austerity

Don’t do this

PPP-built lines do not have to have high construction costs. The Canada Line was cheap to build – it was Canada’s last reasonable-cost subway, and since then costs have exploded around the country. M4 in Milan is inexpensive as well, around €110 million per kilometer at current estimates even while going underneath older subways in city center. The current annual ridership projection of M4, 87 million, means that the current projected cost per weekday trip is €6,000, which represents an enormous social surplus in a region that builds up to around €30,000-40,000 before even pro-transit activists demand cancellation.

But in those cases, the structure of the contract keeps the operating costs artificially high, privatizing what should be public-sector profit from building a very inexpensive-to-operate system. This is especially bad if it is bundled into construction costs as an up-front payment, as in Maryland. In Maryland, the extra operating costs raised the construction cost well above the maximum level that is acceptable to the public transportation community over here, and in the United States too, such lines tend to be under threat of cancellation from fiscally conservative governors if they are not portrayed as pro-market PPPs. But those PPPs then have higher costs and, through poor risk allocation, lead to the worst of both worlds: the private concessionaire increases costs in order to deal with the risk of escalation, but if the risk exceeds prior estimates, then the state remains on the hook.

Don’t do this. One can to some extent understand why Italy was forced into this position at the bottom of the financial crisis. This isn’t such a situation – all countries in Europe are engaging in large discretionary deficit spending nowadays, as the market appears to believe that not only will corona pass, but also the new vaccines developed will help prevent the common cold and the flu in the near future, increasing future health outcomes and improving productivity through less lost sick time. In the United States, a $2 trillion stimulus is sold as just the first of two steps, because there’s fiscal room. You, even as a state or local government, can find money in the budget for more spending – raise taxes or sell bonds, and do so transparently. Don’t take opaque high-interest loans just to tell the public that you haven’t borrowed on the open market. It’s not worth it.

Electronics Before Concrete, not Instead of Concrete

The Swiss slogan electronics before concrete, and related slogans like run trains as fast as necessary, not as fast as possible, is a reminder not to waste money. However, I worry that it can be read as an argument against spending money in general. For many years now, Cap’n Transit has complained that this slogan is used to oppose bad transit like the Gateway Tunnel and if the money is not spent on public transportation then it may be spent on other things. But in reality, the Swiss slogans, all emphasizing cost minimization, must be reconciled with the fact that Switzerland builds a lot of concrete, including extensive regional rail tunneling in Zurich and intercity rail tunneling. Electronics precedes concrete, but does not always substitute for it; it’s better to think of these planning maxims as a way to do more with a fixed amount of money, and not as a way to do the same amount of project with less money.

The extent of tunneling in Switzerland

Here is a list of tunnels built in Switzerland since the 1980s, when its modern program of integrated timetable-infrastructure-rolling stock investment began:

This is not a small program. Zurich and Geneva are not large cities, and yet they’ve build regional rail trunk tunnels – and Zurich has built two, the most of any German-speaking country, since Berlin and Hamburg only have one of their trunk lines each in tunnel, the rest running above ground. The Mattstetten-Rothrist line likewise does not run at high speed, topping at 200 km/h, because doing so would raise the cost of rolling stock acquisition without benefiting the national integrated timetable – but it was an extensive undertaking for how small Switzerland is. Per capita, Switzerland has built far more intercity rail tunnels by length than France, and may even be ahead of Germany and Italy – and that’s without taking into account the freight base tunnels.

The issue of passenger experience

It’s best to think of organization-before-electronics-before-concrete as a maxim for optimizing user experience more than anything. The system’s passengers would prefer to avoid having to loiter 20 minutes at every connection; this is why one designs timed transfers, and not any attempt to keep the budget down. The Bahn 2000 investments were made in an environment of limited money, but money is always limited – there’s plenty of austerity at the local level in the US too, it just ends up canceling or curtailing useful projects while bad ones keep going on.

In Europe, Switzerland has the highest modal split for rail measured in passenger-km, 19.3%, as of 2018; in 2019, this amounted to 2,338 km per person. The importance of rail is more than this – commuters who use trains tend to travel by train shorter than commuters who use cars drive, since they make routine errand trips on foot at short distance, so the passenger-km modal split is best viewed as an approximation of the importance of intercity rail. Europe’s #2 and #3 are Austria (12.9%) and the Netherlands (11.2%), and both countries have their own integrated intercity rail networks. One does not get to scratch 20% with a design paradigm that is solely about minimizing costs. Switzerland also has low construction costs, but Spain has even lower construction costs and it wishes it had Switzerland’s intensity of rail usage.

Optimizing organization and electronics…

A country or region whose network is a mesh of lines, like Switzerland or the Netherlands, had better adopt the integrated timed transfer concept, to ensure people can get from anywhere to anywhere without undue waiting for a connecting train and without waiting for many hours for a direct train. This includes organizational reforms in the likely case there are overlapping jurisdictions with separate bus, urban rail, and intercity rail networks. Fares should be integrated so as to be mode-neutral and offer free transfers throughout the system, and schedules should be designed to maximize connectivity.

This should include targeted investments in systems and reliability. Some of these should be systemwide, like electrification and level boarding, but sometimes this means building something at a particular delay-prone location, such as a long single-track segment or a railway junction. In all cases, it should be in the context of relentlessly optimizing operations and systems in order to minimize costs, ensure trains spend the maximum amount of time running in revenue service and the minimum amount of time sitting at a yard collecting dust, reduce the required schedule padding, etc.

…leads to concrete

Systemwide optimization invariably shows seams in the system. When Switzerland designed the Bahn 2000 network, there was extensive optimization of everything, but at the end of the day, Zurich-Bern was going to be more than an hour, which would not fit any hourly clockface schedule. Thus the Mattstetten-Rohrist line was born, not out of desire to run trains as fast as possible, but because it was necessary for the trains to run at 200 km/h most of the way between Olten and Bern to fit in an hourly takt.

The same is true of speed and capacity improvements. A faster, more reliable system attracts more passengers, and soon enough, a line designed around a train every 15 minutes fills up and requires a train every 10 minutes, 7.5 minutes, 6 minutes, 5 minutes, 4 minutes. An optimized system that minimizes the need for urban tunneling soon generates so much ridership that the tunnels it aimed to avoid become valuable additions to the network.

The Munich S-Bahn, for example, was built around this kind of optimization, inventing many of the principles of coordinated planning in the process. It had a clockface schedule early, and was (I believe) the first system in the world designed around a regionwide takt. It was built to share tracks with intercity and freight trains on outer branches rather than on purely dedicated tracks as in the older Berlin and Hamburg systems, and some of its outermost portions are on single-track. It uses very short signaling blocks to fit 30 trains per hour through the central tunnel in each direction. And now it is so popular it needs a second tunnel, which it is building at very high cost; area activists invoked the organization before electronics before concrete principle to argue against it and in favor of a cheaper solution avoiding city center, but at the end of the day, Munich already optimized organization and electronics, and now is the time for concrete, and even if costs are higher than they should be by a factor of 2-3, the line is worth it.

Electronics before concrete, not instead of concrete

Switzerland is not going to build a French-style national high-speed rail network anytime soon. It has no reason to – at the distances typical of such a small country, the benefits of running at 300 km/h are not large. But this does not mean its rail network only uses legacy lines – on the contrary, it actively builds bypasses and new tunnels. Right now there are plans for an S-Bahn tunnel in Basel, and for an express tunnel from Zurich to Winterthur that was removed from Bahn 2000. The same is true of other European countries that are at or near the frontier of passenger rail technology. Even the Deutschlandtakt plan, compromised as it is by fiscal austerity, by high construction costs, by a pro-car transport minister, and by NIMBYs, includes a fair amount of new high-speed rail, including for example a mostly fast path from Berlin to Frankfurt.

When you plan your rail network well, you encourage more people to use it. When you optimize the schedules, fare integration, transfer experience, and equipment, you end up producing a system that will, in nearly every case, attract considerable numbers of riders. Concrete is the next step: build those S-Bahn tunnels, those express bypasses, those grade separations, those high-speed lines. Work on organization first, and when that is good enough, build electronics, and once you have both, build concrete to make maximum use of what you have.

Streets Before Trust

There’s an emerging mentality among left-wing urban planners in the US called “trust before streets.” It’s a terrible idea that should disappear, a culmination of about 50 or 60 years of learned helplessness in the American public sector. Too many people who I otherwise respect adhere to this idea, so I’m dedicating a post to meme-weeding it. The correct way forward is to think in terms of state capacity first, and in particular about using the state to enact tangible change, which includes providing better public transportation and remaking streets to be safer to people who are not driving. Trust follows – in fact, among low-trust people, seeing the state provide meaningful tangible change is what can create trust, and not endless public meetings in which an untrusted state professes its commitment to social justice.

What is trust before streets?

The trust before streets mentality, as currently used, means that the state has to first of all establish buy-in before doing anything. Concretely, if the goal is to make the streets safer for pedestrians, the state must not just build a pop-up bike lane or a pedestrian plaza overnight, the way Janette Sadik-Khan did in New York, because that is insensitive to area residents. Instead, it must conduct extensive public outreach to meet people where they’re at, which involves selling the idea to intermediaries first.

This is always sold as a racial justice or social justice measure, and thus the idea of trust centers low-income areas and majority-minority neighborhoods (and in big American cities they’re usually the same – usually). Thus, the idea of trust before streets is that it is racist to just build a pedestrian plaza or bus lanes – it may not be an improvement, and if it is, it may induce gentrification. I’ve seen people in Boston say trust before streets to caution against the electrification of the Fairmount Line just because of one article asserting there are complaints about gentrification in Dorchester, the low-income diverse neighborhood the line passes through (in reality, the white population share of Dorchester is flat, which is not the case in genuinely gentrifying American neighborhoods like Bushwick).

I’ve equally seen people use the expression generational trauma. In this way, the trust before streets mentality is oppositional to investments in state capacity. The state in a white-majority nation is itself white-majority, and people who think in terms of neighborhood autonomy find it unsettling.

Low trust and tangible results

The reality of low-trust politics is about the opposite of what educated Americans think it is. It is incredibly concrete. Abstract ideas like social justice, rights, democracy, and free speech do not exist in that reality, to the point that authoritarian populists have exploited low-trust societies like those of Eastern Europe to produce democratic backsliding. A Swede or a German may care about the value of their institutions and punish parties that run against them, but an Israeli or a Hungarian or a Pole does not.

In Israel, this is visible in the corona crisis: Netanyahu’s popularity, as expressed in election polls, has recently risen and fallen based on how Israel compares with the Western world when it comes to handling corona. In March, there was a rally-around-the-flag effect in Israel as elsewhere, giving Netanyahu cover to refuse to concede even though parties that pledged to replace him as prime minister with Benny Gantz got 62 out of 120 seats, and giving Gantz cover not to respond to hardball with hardball and instead join as a minister in Netanyahu’s government. Then in April and May, as Israel suppressed the first wave and had far better outcomes than nearly every European country, let alone the US, Netanyahu’s popularity surged while that of Gantz and the opposition cratered. The means did not matter – the entire package including voluntary quarantine hotels, Shin Bet surveillance for contact tracing, and a tight lockdown that Netanyahu, President Rivlin, and several ministers violated nonchalantly, was seen to produce results.

In the summer, this went in reverse. The second wave hit Israel earlier than elsewhere, and by late summer, its infection rate per capita was in the global top ten, and Israel had the largest population among those top ten countries. In late September it reached around 6,000 cases a day, around 650 per million people. The popularity of Netanyahu’s coalition was accordingly shot; Gantz himself is being nearly wiped out in the polls, but the opposition was holding steady, and Yamina, a party to the right of Likud led by Naftali Bennett that is not currently in the coalition and is perceived as more competent, Bennett himself having done a lot to moderate the party’s line, surged from its tradition 5-6 seats to 16.

Today the situation is unclear – Israelis have seen the state fight the second wave but it was not nearly as successful as in the spring, and right now there is a lot of chaos with vaccination. On the other hand, Israel is also the world’s vaccination capital, and eventually people will notice that by March Israel is (most likely) fully vaccinated while Germany is less than 10% vaccinated. Low-trust people notice results. If they’re disaffected with Netanyahu’s conduct, which most people are, they can then vote for a right-wing-light satellite party like New Hope, just as many voted Kulanu in 2015, which advertised itself as center, became kingmaker after the results were announced, and immediately joined under Netanyahu without trying to seriously negotiate.

Streets lead to trust

The story of corona in Israel does not exist in isolation. Low trust in many cases exists because people perceive the state to be hostile to their interests, which happens when it does not provide tangible goods. Many years ago, talking about his own history immigrating from the Soviet Union in the 1970s, Shalom Boguslavsky credited the welfare state for his integration, saying that if he’d immigrated in the 1990s he’d probably have ended up in a housing project in Ashdod and voted for Avigdor Lieberman, who at the time was running on Russian resentment more than anything.

In Northern Europe, perhaps trust is high precisely because the state provides things. My total mistrust of the German state in general and Berlin in particular is tempered by the fact that, at queer meetups, people remind me that Berlin’s center-left coalition has passed universal daycare, on a sliding scale ranging from 0 for poor parents to about €100/month for wealthy ones. This more than anything reminds me and others that the state is good for things other than dithering on corona and negatively stereotyping immigrant neighborhoods.

Such provisions of tangible goods cannot happen in a trust before streets environment. This works when the state takes action, and endless public meetings in which every objection must be taken seriously are the death of the state. It says a lot that in contrast with Northern Europe, in the United States even in wealthy left-wing cities it is unthinkable that the municipality can just raise taxes to pay teachers and social workers better. Low trust is downstream of low state capacity. Build the streets and trust will follow.

Is Remote Work Viable?

No, not in the long run.

This has big implications for cities in the future, because it means firms will want to cluster more near production amenities – that is, other high-productivity firms. A city like New York manifestly has very weak consumption amenities, because in the spring it proved that its government is dangerously incompetent in a crisis – but its production amenities are likely to grow, because more firms will want to locate there and in other big, rich cities.

Remote work and the tech industry

The tech industry has long been familiar with remote work. The big multinationals have offices worldwide and some teams are remote, and some small firms are even all-remote. Much of this is an adaptation to the industry’s inability to bring everyone to San Francisco and Silicon Valley, where housing is too expensive and work visas are scarce. This has led to a big internal debate about the future of work; for decades now there have been predictions that the Internet would facilitate remote work and therefore reduce the need for cities to exist as office work centers.

The industry also reacted to corona slightly faster than the rest of the Western world. I’m not sure why – usually the American tech industry sneers at anything that comes out of Asia. But for whatever reason, Google sent its workers home in early March, and has been on work-from-home since, as have the other tech employers.

However, this was always intended to be a temporary arrangement. Workers were told to go back to the office when the crisis ended, at a date that keeps being pushed back and is now September 2021. Moreover, it appears that the industry wants to consolidate rather than disperse: Google, Amazon, Facebook, and Apple are all buying up office space in Manhattan, planning to add 22,000 jobs there. This is not San Francisco, but it’s the closest thing: New York is the United States’ second richest metropolitan region, and (I believe) the second biggest tech job center, with New York hosting the largest non-Bay Area Google office.

The problems with remote work

I have asked a number of people to talk to me about their experience with working from home. All are American professionals; this is far and away the easiest socioeconomic class to do an ethnography of. At no point did anyone ever tell me that everyone in their office is as productive working from home as they had been working as a team at the office. The work from home productivity loss is real; it does not affect everyone, but it affects enough people to be noticeable.

Specific problems I was told include,

  • Corona specifically is a very stressful event, so everyone is on edge and less productive than the usual.
  • Without continuous office work, it’s harder to onboard junior workers, even when senior workers are fine at home. Junior workers also lose the benefits of close mentoring.
  • Parents with children have to take on additional care duties, and without a stay-at-home parent this is difficult.
  • I believe in one case I was told the opposite of the above – that given that children are at home, it’s easier for parents than for non-parents.
  • At least per the CEO of United, who is obviously biased on this, firms perceive in-person sales to be more successful than virtual ones. In general, I’ve been told that work facing clients is less productive when it’s virtual and law firms can work remotely in the short run with their existing client base but in the long run they need the office.

The standard production theory, articulated for example by Alain Bertaud, is that working from home is less productive because there are no spontaneous interactions, and this seems true although I don’t recall anyone telling me this exact thing literally, but very similar problems are apparent.

What does this mean for cities?

Before corona, it was not always clear whether advances in telecommunications would make remote work viable. It increasingly looks like the answer is no, and therefore the most productive firms are likely to center around their usual clusters, just as the tech firms are buying up Manhattan office space. The upshot, then, is that high-cost, high-productivity city centers are likely to see more commercial demand in the medium and long runs.

One model that I’ve heard from multiple sources is mixed, for example 2-4 days a week at the office, 1-3 days remote. If this happens, then it will mean that people commute fewer days. This has opposite effects on office and residential geography: fewer commutes mean it’s more acceptable to live farther out and have longer work trips on work-at-office days, which encourages either suburbanization or hopping over to the next city over; for the exact same reason, it’s also more acceptable to site offices in areas with more traffic congestion, that is city center.

What does this mean for public transportation?

More urban job concentration universally requires better public transportation, since rapid transit is far and away the most efficient mode of transportation measured in capacity provided per unit of right-of-way width. However, the details are subtle. Most importantly, the American upper middle class mostly does not work 9 to 5 at the most productive firms. The tech industry tends toward shifted hours, especially on the East Coast in order to overlap Silicon Valley better, and even for the same reason in Israel. So the impact of more tech employment in Midtown is not that New York desperately needs more subway capacity, but rather that it needs to broaden the peak to last until 10 in the morning rather than 9. This conclusion does not depend much on whether workers show up at the office every day or only 3-4 days a week, because 60-80% of rush hour traffic still requires peak or near-peak train throughput.

There were many Americans who, back when corona seemed to be first and foremost a New York problem, predicted the end of cities, or the conversion of cities to spaces of consumption. Joel Kotkin even blamed New York’s density for corona and praised Los Angeles’s sprawl; now that Los Angeles is running out of hospital beds, nobody in the US blames density anymore. (One could also point out Seoul and Tokyo’s density, but not even 460,000 deaths and counting will make Americans say “our country needs to be more like other countries.”)

But this is not looking to happen. The most productive firms in the US are urbanizing – and those are the most productive firms in the world; it averages out with horrific American public-sector inefficiency to about the same GDP per hour as in Germany. And this means that going forward, the richest, most productive, and most expensive cities will remain spaces of high-end production, and will need to build sufficient numbers of office towers and residences and improve public transportation infrastructure to accommodate.

Quick Note: Consumption and Production Theories of Berlin

I’ve periodically written about consumption and production theories of cities – that is, whether people mostly move to cities based on consumption or production amenities. The production theory is that what matters is mostly production amenities, that is, jobs, and this underlies YIMBYism. Consumption theory is that people move for consumption amenities, and, moreover, these amenities are not exactly consumption in the city, for example good health outcomes, but consuming the city itself, that is neighborhood-level amenities in which who lives in the city matters. The latter theory, for example promulgated by Richard Florida, is that jobs follow consumption amenities like gay bars, and not the other way around. It is wrong and production theory is right, and I’d like to give some personal examples from Berlin, because I feel like Berliners all believe in consumption theory.

The situation in Berlin

Berlin is an increasingly desirable city. After decades in which it was economically behind, the city is growing. Unemployment, which stood at 19% in 2005, was down to 7.8% last year. With higher incomes come higher rents, and because Berlin for years built little housing as there was little demand, rents rose, and it took time for housing growth to catch up; on the eve of corona, the city was permitting about 6 annual dwellings per 1,000 people, up from about 1 in the early 2000s.

This is generally attributed to tech industry growth. There are a lot of tech startups in the city. I don’t want to exaggerate this too much – Google’s biggest Germany office is by far Munich’s, and the Berlin office is mostly a sales office with a handful of engineers who are here because of a two-body problem. But the smaller firms are here and the accelerator spaces are very visible, in a way that simply didn’t exist in Paris, or even in Stockholm.

Berlin’s production amenities

I might not have thought that Berlin should attract so much tech investment. My vulgar guess would be that tech would go to cities with many preexisting engineers, like Munich and Stuttgart, or maybe to Frankfurt for the international flight connections. But Berlin does make sense in a number of ways.


The city is mostly fluent in English. Jakub Marian’s map has France 39% Anglophone and Germany 56%, which doesn’t seem too outlandish to me. But Paris seems in line with the rest of France, whereas in Berlin, service workers seem mostly Anglophone, which is not the case in (say) Mainz or Munich.

The global tech industry is Anglophone, and good command of English is a huge production amenity. Other English-dependent industries seem to favor Anglophone European cities as well, for example various firms fleeing Brexit moved their European headquarters not to Paris but to Amsterdam or maybe Dublin.

The capital

The federal government is here. This is not relevant to tech – the startups here don’t seem to be looking for lobbying opportunities, and at any case German lobbying works differently from American lobbying and firm-level proximity to the capital is unimportant. However, the government stimulates local spending, which has increased employment. The government’s move here has been gradual, with institutions that during division were spread all over West Germany slowly migrating to Berlin.

Good infrastructure

The quality of infrastructure in Berlin is very good. The urban rail network was built when Berlin was Western Europe’s third largest city, after London and Paris, and has even grown after the war because the West built U7 and U9 to bypass Mitte. This means that commute pain here is not serious, especially on any even vaguely middle-class income. Moreover, Berlin has benefited from post-reunification investment, including Hauptbahnhof and two high-speed rail lines.

Consumption theory and the counterculture

The queer counterculture that I am involved with in Berlin tells a different story. To hear them tell it, Berlin has a quirky, individualistic, nonconforming culture, unlike the stifling normality of Munich. Artists moved here, and then other people moved here to be near the artists, paying higher rents until the artists could no longer afford the city. This story is told at every scale, from Berlin as a city to individual neighborhoods like Prenzlauer Berg and Neukölln. A lot of the discourse about Berlin repeats this uncritically, for example Feargus O’Sullivan at CityLab/Bloomberg Cities writes about the cool factor and about gentrification of old buildings.

It is also a completely wrong story. This is really important to understand: nobody that I know in the sort of spaces that are being blamed for gentrification, that is the tech industry and its penumbra, has any interest in the counterculture. I go to board games meetups full of tech workers who are fluent in English and often don’t know any German, and they have no connections at all to the local counterculture. They interact with immigrant culture spaces, not with the 95%+ white counterculture as defined by queer spaces in Neukölln that complain about gentrification in a neighborhood undergoing white flight at the rate of postwar New York (compare 2019 data, PDF-pp. 25 and 28, with 2016, PDF-pp. 28 and 31). Occasionally there are crossovers, as when an American comedian hosted live standup in February and then there were tech workers and said American also interacts with the counterculture, but a standup comic is not why Berliners complain.

Nor do I find foreign tech workers especially interested in German minutiae comparing Berlin with Munich. By my non-German standards, Berliners already jaywalk at indescribably lower rates, and I gather that Munich is stuffier but that’s not why I’m here and not there, the rents and the language are.

We’re not even particularly oppositional to the counterculture. I personally am because seeing queer space after queer space host indoor events during corona without masks was a horrifying experience; I went to a queer leftist meetup in late October in which people huddled together maskless and I was the only one with a mask on, except for one trans Australian physicist who drank a beer and then masked after finished. But the rest? They don’t care, nor should they. The counterculture is not the protagonist or the antagonist of Berlin’s story; it’s barely a bystander. Consumption theory is just what it promotes in order to convince itself that it’s important, that it spreads ideas and not viruses.

Costs Matter: Some Examples

A bunch of Americans who should know better tell me that nobody really cares about construction costs – what matters is getting projects built. This post is dedicated to them; if you already believe that efficiency and social return on investment matter then you may find these examples interesting but you probably are not looking for the main argument.

Exhibit 1: North America


I wrote a post focusing on some North American West Coast examples 5 years ago, but costs have since run over and this matters from the point of view of building more in the future. In the 2000s and 10s, Vancouver had the lowest construction costs in North America. The cost estimate for the Broadway subway in the 2010s was C$250 million per kilometer, which is below world median; subsequently, after I wrote the original post, an overrun by a factor of about two was announced, in line with real increases in costs throughout Canada in the same period.

Metro Vancouver has always had to contend with small, finite amounts of money, especially with obligatory political waste. The Broadway subway serves the two largest non-CBD job centers in the region, the City Hall/Central Broadway area and the UBC, but in regional politics it is viewed as a Vancouver project that must be balanced with a suburban project, namely the lower-performing Surrey light rail. Thus, the amount of money that was ever made available was about in line with the original budget, which is currently only enough to build half the line. Owing to the geography of the West Side, half a line is a lot less than half as good as the full line, so Vancouver’s inability to control costs has led to worse public transportation investment.


Like Vancouver, Toronto has gone from having pretty good cost control 20 years ago to having terrible cost control today. Toronto’s situation is in fact worse – its urban rail program today is a contender for the second most expensive per kilometer in the world, next to New York. The question of whether it beats Singapore, Hong Kong, London, Melbourne, Manila, Qatar, and Los Angeles depends on project details, essentially on scoring which of these is geologically and geographically the hardest to build in assuming competent leadership, which is in short supply in all of these cities. I am even tempted to specifically blame the most recent political interference for the rising costs, just as the adoption of design-build in the 2000s as an in-vogue reform must be blamed for the beginning of the cost blowouts.

The result is that Toronto is building less stuff. It’s been planning a U-shaped Downtown Relief Line for decades, since only the Yonge-University-Spadina (“YUS”) line serves downtown proper and is therefore overcrowded. However, it’s not really able to afford the full line, and hence it keeps downgrading it with various iterations, right now to an inverted L for the Ontario Line project.

Los Angeles

Los Angeles’s costs, uniquely in the United States, seemed reasonable 15 years ago, and no longer are. This, as in Canada, can be seen in building less stuff. High-ranking officials at Los Angeles Metro explained to me and Eric that the money for capital expansion is bound by formulas decided by referendum; there is a schedule for how to spend the money as far as 2060, which means that anything that is not in the current plan is not planned to be built in the next 40 years. Shifting priorities is not really possible, not with how Metro has to buy off every regional interest group to ensure the tax increases win referendums by the required 2/3 supermajority. And even then, the taxes imposed are rising to become a noticeable fraction of consumer spending – even if California went to majority vote, its tax capacity would remain very finite.

New York

The history of Second Avenue Subway screams “we would have built more had costs been lower.” People with deeper historic grounding than I do have written at length about the problems of the Independent Subway System (“IND”) built in the 1920s and 30s; in short, construction costs were in today’s terms around $140 million per km, which at the time was a lot (London and Paris were building subways for $30-35 million/km), and this doomed the Second System. But the same impact of high costs, scaled to the modern economy, is seen for the current SAS project.

The history of SAS is that it was planned as a single system from 125th Street to Hanover Square. The politician most responsible for funding it, Sheldon Silver, represented the Lower East Side. But spending capacity was limited, and in particular Silver had to trade that horse for East Side Access serving Long Island, which was Governor George Pataki’s base. The package was such that SAS could only get a few billion dollars, whereas at the time the cost estimate for the entire 13-km line was $17 billion. That’s why SAS was chopped into four phases, starting on the Upper East Side. Silver himself signed off on this in the early 2000s even though his district would only be served in phase four: he and the MTA assumed that there would be further statewide infrastructure packages and the entire line would be complete by 2020.

Exhibit 2: Israel

Israel is discussing extending the Tel Aviv Metro. It sounds weird to speak of extensions when the first line is yet to open, but that line, the Red Line, is under construction and close enough to the end that people are believing it will happen; Israelis’ faith that there would ever be a subway in Tel Aviv was until recently comparable to New Yorkers’ faith until the early 2010s that Second Avenue Subway would ever open. The Red Line is a subway-surface Stadtbahn, as is the under-construction Green Line and the planned Purple Line. But metropolitan Tel Aviv keeps growing and is at this point an economic conurbation of about 3-4 million people, with a contiguous urban core of 1.5 million. It needs more. Hence, people keep discussing additions. The Ministry of Finance, having soured on the Stadtbahn idea, bypassed the Ministry of Transport and introduced a complementary three-line underground driverless metro system.

The cost of the system is estimated at 130-150 billion shekels, which is around $39 billion. This is not a sum Israelis are used to seeing for a government project. It’s about two years’ worth of IDF spending, and Israeli is a militarized society. It’s about 10% of annual GDP, which in American or EU-wide terms would be $2 trillion. The state has many competing budget priorities, and there are so many other valid claims on the state coffers. It is therefore likely that the metro project’s construction will stretch over many years, not out of planning latency but out of real resource limits. People in Israel understand that Gush Dan has severe traffic congestion and needs better transportation – this is not a point of political controversy in a society that has many. But this means the public is willing to spend this amount of money over 15-20 years at the shortest. Were costs to double, in line with the costs in most of th Anglosphere, it would take twice as long; were they to fall in half, in line with Mediterranean Europe, it would take half as long.

Exhibit 3: Spain

As the country with the world’s lowest construction costs for infrastructure, Spain builds a lot of it, everywhere. This includes places where nobody else would think to build a metro tunnel or an airport or a high-speed rail line; Spain has the world’s second longest high-speed rail network, behind China. Many of these lines probably don’t even make sense within a Spanish context – RENFE at best operationally breaks even, and the airports were often white elephants built at the peak of the Spanish bubble before the 2008 financial crisis.

One can see this in urban rail length just as in high-speed rail. Madrid Metro is 293 km long, the third longest in Europe behind London and Moscow. This is the result of aggressive expansion in the 1990s and 2000s; new readers are invited to read Manuel Melis Maynar’s writeup of how when he was Madrid Metro’s CEO he built tunnels so cheaply. Expansion slowed down dramatically after the financial crisis, but is starting up again; the Spanish economy is not good, but when one can build subways for €100 million per kilometer, one can build subways that other cities would not. In addition to regular metros, Madrid also has regional rail tunnels – two of them in operation, going north-south, with a third under construction going east-west and a separate mainline rail tunnel for cross-city high-speed rail.

Exhibit 4: Japan

Japan practices economic austerity. It wants to privatize Tokyo Metro, and to get the best price, it needs to keep debt service low. When the Fukutoshin Line opened in 2008, Tokyo Metro said it would be the system’s last line, to limit depreciation and interest costs. The line amounted to around $280 million/km in today’s money, but Tokyo Metro warned that the next line would have to cost $500 million/km, which was too high. The rule in Japan has recently been that the state will fund a subway if it is profitable enough to pay back construction costs within 30 years.

Now, as a matter of politics, on can and should point out that a 30-year payback, or 3.3% annual interest, is ridiculously high. For one, Japan’s natural interest rate is far lower, and corporations borrow at a fraction of that interest; JR Central is expecting to be paying down Chuo Shinkansen debt until the 2090s, for a project that is slated to open in full in the 2040s. However, if the state changes its rule to something else, say 1% interest, all that will change is the frontier of what it will fund; lines will continue to be built up to a budgetary limit, so that the lower the construction costs, the more stuff can be built.

Conclusion: the frontier of construction

In a functioning state, infrastructure is built as it becomes cost-effective based on economic growth, demographic projections, public need, and advances in technology. There can be political or cultural influences on the decisionmaking process, but they don’t lead to huge swings. What this means is that as time goes by, more infrastructure becomes viable – and infrastructure is generally built shortly after it becomes economically beneficial, so that it looks right on the edge of viability.

This is why megaprojects are so controversial. Taiwan High-Speed Rail and Korea Train Express are both very strong systems nowadays. Total KTX ridership stood at 89 million in 2019 and was rising on the eve of corona, thanks to Korea’s ability to build more and more lines, for example the $69 million/km, 82% underground SRT reverse-branch. THSR, which has financial data on Wikipedia, has 67 million annual riders and is financially profitable, returning about 4% on capital after depreciation, before interest. But when KTX and THSR opened, they both came far below ridership projections, which were made in the 1990s when they had much faster economic convergence before the 1997 crisis. They were viewed as white elephants, and THSR could not pay interest and had to refinance at a lower rate. Taiwan and South Korea could have waited 15 years and only opened HSR now that they have almost fully converged to first-world Western incomes. But why would they? In the 2000s, HSR in both countries was a positive value proposition; why skip on 15 years of good infrastructure just because it was controversially good then and only uncontroversially good now?

In a functioning state, there is always a frontier of technology. The more cost-effective construction is, the further away the frontier is and the more infrastructure can be built. It’s likely that a Japan that can build subways for Korean costs is a Japan that keeps expanding the Tokyo rail network, because Japan is not incompetent, just austerian and somewhat high-cost. The way one gets more stuff built is by ensuring costs look like those of Spain and Korea and not like those of Japan and Israel, let alone those of the United States and Canada.

Metcalfe’s Law for High-Speed Rail, Redux

Americans are in big infrastructure spending mood, and my post from February using Metcalfe’s law to argue in favor of expansive high-speed rail in the eastern half of the United States has been attracting renewed attention. That post looked at how Metcalfe’s law that the value of a network rises in proportion to the square of the number of nodes implied that once a strong HSR corridor existed, for example the Northeast Corridor, extensions would be strong as well even if they connected much smaller cities. People have been asking me to extend that analysis to more lines that do not touch the Northeast Corridor, so here goes.

As a reminder, I’m using a simple gravity model, of the following level of sophistication:

He doesn’t put his axe down when standing on a crowded train, either. Credit: Anaterate.

The model is that the annual ridership in millions between two metropolitan areas A and B, with populations in the millions, is,

75,000\cdot\mbox{Pop}_{A}^{0.8}\cdot\mbox{Pop}_{B}^{0.8}/\max\{500 \mbox{ km}, d\}^{2}.

The theoretical reason for the 0.8 exponents is diseconomies of scale: the average person in Tokyo is farther from Tokyo Station than the average person in a small city is from their respective intercity rail station. Empirically, the best fit exponent for observed data in Japan and Europe is 0.8 – see sources in my previous post and in this post (sourced to since-rotted links) for France. The 500 km minimum is an artifact of the impact of station access time and the option of driving instead of taking the train.

Fares are set at typical Continental European levels rather than Japanese ones. As in the previous post, this means $0.135 per passenger-km, which breaks down as $0.07/p-km in operating expenses including rolling stock but excluding infrastructure and $0.065/p-km in profit, up to a total profit of $50/passenger. Beyond $50 in profit, which normally occurs at 770 km, fares only rise with operating expenses, to be more competitive with airlines. The goal is to find lines that have annual profits of more than 2-3% of construction costs.

A note of caution on the model

There are arguments to be made to refine the gravity model above in either direction. Ridership estimates in Britain are well above what the model predicts. High Speed 2 projects 3 trains per hour between London and Birmingham, running nonstop between the two cities so that no other city pairs can be added. The model gives an annual ridership equal to,

75,000\cdot 14^{0.8}\cdot 3^{0.8}/500^{2} = 5.97,

which fills around 1 train per hour in each direction to 50% of seated capacity. It’s possible the model does give higher ridership figures for very close-by cities – London and Birmingham are only 180 km apart – or it’s possible some unknown factor exists. Or HS2’s traffic estimates could be completely off.

In the case of the US, it’s likely any HSR will run faster than legacy 1960s Shinkansen. However, there’s a serious malus coming from higher car ownership, lower car traffic levels, and much weaker city centers. This is unlikely to be a problem for traffic to New York, but the last post dealt with that, whereas today we’re looking mostly at lines that aren’t about New York. Even Chicago is extremely auto-oriented by the standards of London or Paris, let alone Tokyo.

Metcalfe’s law for HSR in the Midwest: the initial line

The Midwest benefits from two things: it is flat, which reduces construction costs to $20-25 million per km if European norms are followed, and it has near-megacity Chicago in the middle. Unfortunately, Chicago is big but not big enough, and while the secondary cities are pretty big, there aren’t additional medium-size cities nearby the way Lyon has Saint-Etienne, Marseille has Toulon, etc. HSR can succeed, but the return on investment is for the most part marginal. The one exception is lines that can leverage the Northeastern network, including eventually not just the Northeast Corridor but also tie-ins to Pittsburgh and Cleveland, both of which are at reasonable HSR distance from New York.

By itself, the core Midwestern network would connect Chicago (10 million people) with Toledo (0.8, a distance of 370 km) and thence split toward Detroit (5 million, 100 km from Toledo) and Cleveland (3 million, 180 km). This leads to the following O&D ridership matrix, in millions:

City W\City EToledoDetroitCleveland

And in annual operating profits, in millions of dollars:

City W\City EToledoDetroitCleveland

This is not a lot of ROI. It’s $443 million a year, for a 650 km system, which should cost maybe $15 billion. It’s 3% by itself, which isn’t horrible, but compares poorly with Northeastern lines even though it connects the Midwest’s numbers 1, 2, and 4 metro regions.

In contrast, suppose a Northeastern system preexists, or perhaps is built at the same time, including a Pittsburgh-Cleveland connection. What then? Well, the question is really what the ROI is on connections from west of Cleveland to east of Cleveland. There are four metro areas east of Cleveland on the way to New York: Pittsburgh (2.5 million, 200 km from Cleveland), Harrisburg (0.7, 280 km from Pittsburgh), Philadelphia (7 million, 170 km from Harrisburg), New York (22 million, 140 km from Philadelphia). Washington has 10 million people and is 220 km from Philadelphia, but because a Washington-Philadelphia-Harrisburg route is circuitous, trains can only charge for 220 km, which is $29.70, and then earn the usual rate of $0.135/km farther west up to a maximum of $50 in profit, which is reached 730 km west of Harrisburg, or somewhat west of Toledo. With this in mind, we use the same pair of tables as above for the new city pairs, first ridership and then operating income:

City W\City EPittsburghHarrisburgPhiladelphiaNew YorkWashington
City W\City EPittsburghHarrisburgPhiladelphiaNew YorkWashington

The total operating income is $875 million a year, which combines with our internal $443 million to produce an 8.8% ROI. This relies on estimating HSR ridership at hefty distances – New York-Chicago is 1,340 km and around 5 hours, New York-Detroit is 1,070 km and around 4 hours. But we do have ridership estimates for city pairs of that magnitude in both Europe and Japan and they’re fine, except for airline-dominated Tokyo-Fukuoka. If anything, this is more robust than making assumptions on how many people are willing to travel by train between two cities without public transportation like Cleveland and Detroit.

Metcalfe’s law for the Midwest: further lines

Past plans for a Chicago-centered Midwestern HSR network called for four spokes: east toward Cleveland and Detroit, northwest toward Milwaukee and Minneapolis, southeast toward Indianapolis and Cincinnati, and southwest toward St. Louis and perhaps Kansas City. These spokes do pan out financially, but the ROI is not great. Even a line that doesn’t touch Chicago can work, namely HSR between Cleveland, Columbus, and Cincinnati – those three cities are too small and weak-centered to produce internal ridership, but New York-Columbus is in similar shape to New York-Detroit.

Milwaukee (2 million, 140 km from Chicago)

Milwaukee’s metro area touches Chicago’s. HSR between the two cities alone is not worth it, since at this distance, top speed isn’t as relevant as station access time. However, the addition of other cities makes this worthwhile. Since Milwaukee is just on city, we put ridership and operating income in the same table:

CityRidershipOperating income
New York0.7135.34

The ROI within the Midwest alone on what should be about $3 billion in construction is around 4% – higher than the bare Chicago-Cleveland/Detroit system. With Northeastern tie-ins, this rises to 7%, if one is confident in second-order but noticeable extra revenue from trains from New York, which would necessarily be a two-seat ride and take almost 6 hours with transfer time.

St. Louis (3 million, 460 km from Chicago) and Kansas City (2.5 million, 400 km from St. Louis)

The Chicago-St. Louis line has received some investments in the last 10 years that the state of Illinois pretends are high-speed rail. Those are expensive – there’s extensive surplus extraction by actors including politicians and the freight railroads – and perform exactly as one should expect trains that are slower than the legacy trains that the TGV replaced 40 years ago. However, this says nothing about whether trains that Europeans and East Asians would recognize as fast could succeed on that corridor. Could they?

A reasonable estimate for the Chicago-St. Louis construction cost is $10 billion; St. Louis-Kansas City would be another $10 billion, perhaps slightly costlier per km because Missouri’s terrain isn’t quite so flat as Illinois’s. Ignoring transfer penalties, we get the following ridership and operating income out of it:

City N\City SWSt. LouisKansas City
St. Louis1.5
City N\City SWSt. LouisKansas City
St. Louis39.1

St. Louis generates 6.84 million riders and $240 million in operating profit, which is above our 2% minimum but not by much. Moreover, 6.84 million riders means a train every hour, at which point there are real frequency artifacts for a service that shouldn’t take much longer than an hour and a half to Chicago. So it’s marginal, though still plausible. But if this is plausible, Kansas City isn’t: it generates $150 million. There are small intermediate stop locations like Springfield and Columbia, but they’re too small to make a difference.

The Ohio Hub

The four largest metro areas of Ohio are roughly collinear. Going southwest of Cleveland, one has Columbus (2.5 million, 220 km), then Dayton (1 million, 110 km), and finally Cincinnati (2.3 million, 90 km). Construction costs are likely to be low because of the terrain – only around Cincinnati are there significant hills. 420 km for $10 billion is plausible. What is the ridership, and what is the revenue?

City E\City WColumbusDaytonCincinnati
New York1.810.711.18
City E\City WColumbusDaytonCincinnati
New York90.7135.4459.13

The total operating income is $563 million a year, which is 5.6% ROI. The biggest cells – New York-Columbus, New York-Cincinnati, Philadelphia-Columbus, Washington-Columbus – are reasonably certain. The internal Midwestern numbers are more suspect, as are the numbers involving Pittsburgh – these are cities where car ownership approaches 100% and the remainder are carless out of poverty, and the destinations are fairly decentralized.

Indianapolis and points south

Indianapolis (2.5 million, 280 km from Chicago) is an attractive-looking target. By itself it’s not much, just like slightly-bigger St. Louis, but unlike St. Louis, it has cities behind it in Cincinnati (170 km) and Louisville (1.5 million, 180 km) that are not as far from everything as Kansas City is. Moreover, Indianapolis-Cincinnati also unlocks travel to Columbus, probably with a transfer because the bluffs around Cincinnati force trains from both Indianapolis and Columbus to enter from the north, without through-service.

South of Louisville, it’s attractive to go south to Nashville (2 million, 270 km from Louisville), Chattanooga (1 million, 200 km), and finally Atlanta (7 million, 180 km). But unlike the New York-Atlanta and the New York-Chicago legs of the triangle, the Chicago-Atlanta leg is decent but not amazing, since it omits the largest city.

City S\City NMilwaukeeChicagoIndianapolisLouisvilleNashvilleChattanooga
City S\City NMilwaukeeChicagoIndianapolisLouisvilleNashvilleChattanooga

Reasonable construction costs are $6 billion to Indianapolis, $4 billion to each of Cincinnati and Louisville, $7 billion to Nashville, $6 billion to Chattanooga, and $5 billion to Atlanta. Indianapolis itself doesn’t generate sufficient ROI, but with the addition of Cincinnati it is pretty strong, the combined system generating $483 million, or 4.8% ROI. Then Louisville generates $108 million, or 2.7%; Nashville generates $133 million, or 1.9%; and Chattanooga and Atlanta together generate $360 million, or 3.3%. Note that the last segment generates the highest ROI, and moreover it is not really possible to start from Atlanta and move north, since Chattanooga alone doesn’t generate significant ridership to cities northeast of Atlanta, as those cities are either small (Greenville, Charlotte) or far (Washington).

Update 12-21: Madison (0.9 million, 120 km from Milwaukee) and Minneapolis (4 million, 400 km from Madison)

The above calculations are for expansions from the first Midwestern core line connecting metro regions #1, 2, and 4 to on another. But what about the #3 region, Minneapolis? Minneapolis has a metro area of 4 million, and is by far the largest Midwestern region with population growth, having grown 9% between 2010 and 2019, whereas Chicago, Detroit, and St. Louis were flat and Cleveland declined.

It should not surprise that Chicago-Minneapolis traffic alone is insufficient to justify HSR, given that Chicago-Detroit alone is not and that line requires service to Cleveland as well as points east. Fortunately, Minneapolis’s location is such that through-service from much of the rest of the Midwest is plausible. Distances are long – this isn’t the Northeast or Western Europe – but trips between Minneapolis and secondary cities like Cleveland, Detroit, Cincinnati, and Indianapolis become much faster by rail than by car. Even St. Louis-Minneapolis is feasible, even though nowadays there’s a mostly direct all-freeway route that’s 900 km long vs. 1,120 by HSR. Midwestern travel today is dominated by the car and not the plane, since car ownership is universal and flying between two secondary cities is not necessarily convenient or cheap.

We get the following matrix of ridership:

City S\City NMadisonMinneapolis
St. Louis0.320.44

And here is the matrix of operating income:

City S\City NMadisonMinneapolis
St. Louis14.9921.82

A reasonable construction cost for Milwaukee-Minneapolis is around $13 billion. Overall operating income is $514 million a year, so 4% ROI; one can even scratch a few fractions of 1% by including extra ridership from connections from points east of Cleveland, but I’m comfortable rounding New York-Minneapolis ridership over 2,000 km and a probably-untimed transfer in Chicago from 0.67 million to zero. At most, including East Coast-Minneapolis rail ridership provides cushion against unresolved questions such as whether people would take a 4.5-hour train between Detroit and Minneapolis or continue driving for 10 hours plus rest stops.

Metcalfe’s law in California

In California, the definition of a metro area is dicey. The combined statistical area for the Bay Area has 9.7 million people, but that includes Merced, Modesto, and Stockton, all of which are geographically in the Central Valley and would get dedicated HSR stations, some on a different branch from that going toward the Bay proper. In fact, we have 9.7^0.8 = 6.16, but if we sum each individual MSA component and raise its population to the 0.8th power, even omitting ones without planned HSR stations like Santa Cruz and Napa, we get a total of about 7. So we should use the higher figure. Likewise, in Los Angeles, taking the CSA population yields 18.7^0.8 = 10.41 whereas summing the constituent metro areas separately yields 11.3, and summing the counties, all of which are supposed to have stations, yields 12.8. We use the higher figure, 12.

Together we get 25 million intercity riders, before applying the distance penalty. The distance depends on which pair of stations we look at, since we’re summing over many different stations; it also depends on alignment choices, which don’t all have the same average speed, which means that trip time, whence the distance malus, is not perfectly congruent to distance. To simplify, we assume that LA-SF is 2:45, which at Shinkansen speed is 650 km; this is shorter than the actual LA-SF distance under most alignments, though not by much, and it’s longer than actual distances to subsidiary Northern California destinations.

With this in mind, our formula spits out 14.79 million intercity rail trips. This is a lot lower than California HSR estimates. Those estimates also include San Diego (3 million, 190 km from LA), Bakersfield (0.9 million, 180 km), Fresno (1.3 million, 170 km from Bakersfield), and Sacramento (2.6 million, 270 km from Fresno, 230 km from SF). None of these adds a lot, though. The reasons for the discrepancy include,

  • California HSR assumed heavy HSR commuter traffic – Palmdale-Los Angeles was one of the top city pairs.
  • California HSR assumed somewhat lower fares than the European norm, standing at $79 for LA-SF.
  • California was projecting population into the future, and may have assumed less NIMBYism than the state presently has.
  • The California HSR model may have had flaws; one such flaw was overestimating the impact of frequency at the LA-SF range, to the point that pruning branches such as to San Jose was said to increase ridership by improving frequency to the remaining destinations.

Not that the numbers coming out of my model are bad. The LA-SF numbers alone are worth $625 million in operating profits a year, and with Bakersfield and Fresno this grows to $875 million. The cost of the project without San Diego and Sacramento tie-ins should be on the order of $25-30 billion, in today’s money. Sacramento is maybe 90 extra km and $2 billion depending on alignment, and generates another $260 million or so; Metcalfe’s law is practically a free gift when you have a 90 km spur in flat geography. San Diego is probably something like $6 billion, the higher cost coming from the constrained urban environment and the need for some viaducts and one short tunnel, and adds around $240 million in operating profits.

I am of course aware that at no point was the cost of California HSR $25 billion in 2020 terms. In 2008 the state promised $33 billion in 2008 dollars. The discrepancy comes from some catastrophically bad decisions regarding scope at every stage of the planning phase and bad procurement. But if one looks at what the project needed rather than what has been built in the Central Valley and plugs in standardized costs, the answer is around $25 billion.

What Suburban Poverty? Redux

Earlier this week, I wrote about the incomes of commuters, looking at the incomes of people who commute to the central business districts of six American cities by distance from the center. Contrary to the story of drive-until-you-qualify, in which incomes drop as one moves farther out, in fact incomes tend to rise with commute distance. I was asked by many people in comments and on Twitter, what about the general public, and not just commuters?

The answer is that the answer changes but not by much. The model remains that of the poverty donut, in which people within a certain distance from city center, between 5 and 20 km depending on city size, are poorer than people in both the innermost radius and people who live in the suburbs farther out.

As before, we use data from OnTheMap, which slices jobs by income brackets, of which the highest is $40,000 or more per year in wages. This does not take unemployment or non-wage income into account, but usually these amplify existing inequalities in wages.

Here’s the same table as in the last post, with counts of employed residents and the share of $40,000+ workers within the same radii from the same point as before, without the restriction that people work in the CBD:

CityNew YorkLos AngelesChicagoWashingtonSan FranciscoBoston
PointGrand Central7th/Metro CenterMadison/StateFarragutMarket/2ndDTX
0-5 km680,133203,820176,979177,312222,134219,045
40k+ %67.6%33%68.6%66.4%69.2%58.7%
5-10 km1,123,426506,084342,255297,723239,994379,292
40k+ %50.1%34.7%48%53%57.6%53.5%
10-15 km1,335,294627,797468,107342,649261,568274,212
40k+ %41.5%40.7%39.3%50.6%57.3%58.8%
15-20 km1,114,743736,561368,022306,101248,715223,600
40k+ %45.5%44.3%44.4%51.8%55.8%57.9%
20-30 km1,289,3641,220,414539,332485,355367,591384,671
40k+ %51.8%44.2%47.6%56.6%58.4%57.9%
30-40 km905,2541,020,080630,250551,093469,556387,372
40k+ %57.1%44.9%49.3%53.7%60.3%54.8%
40-50 km753,040754,717633,381478,307377,010377,544
40k+ %55.8%45.9%51.1%56.3%63.7%53.9%
50-60 km623,786632,476554,520435,857405,328421,248
40k+ %55.7%49.6%47.4%47.2%62.5%51.8%
60-70 km535,991405,516399,769410,554498,840434,048
40k+ %55.6%50.2%49.8%48.8%58.8%45.8%
70-80 km484,356518,270189,540232,985410,047402,805
40k+ %54.9%46.5%48.7%51.2%54.4%47.9%


A few valuable footnotes to the table above:

  1. In Boston, the innermost 3 km radius, comprising such neighborhoods as Back Bay, there are 98,691 residents of whom 66.4% earn at least $40,000 a year, but the 5 km level of granularity doesn’t quite see that because the city is smaller than the others. So the swoosh model seen in New York and Washington still holds.
  2. The 50-60 km and 60-70 km annuli around Washington include most of Baltimore, so they are poor once we strip the requirement that people work in the District. They do not show suburban poverty, but urban poverty in a city that, far from getting the transportation investment Massachusetts is putting into the Gateway Cities, had a subway line canceled by a popular moderate Republican governor for what’s almost certainly racist reasons.
  3. The situation in the Bay Area is the reverse of that of Washington. The 40-50 km and 50-60km annuli are wealthy because they happen to include wealthy communities on the Peninsula whose suburban status is awkward, having been both wealthy commuter suburbs of San Francisco and more recently Silicon Valley edge cities with many tech jobs.

What’s going on in Los Angeles?

All other cities on the table have poverty donuts, poorer than both the city core and the suburbs. But in Los Angeles, the $40,000+ share grows nearly monotonically as distance from the CBD increases. The 5 km radius from the center, which in New York comprises the Upper East and West Sides and in Chicago comprises the North Side and the gentrifying parts of the Near South Side, is the poorest group in Los Angeles. It consists of neighborhoods that are not particularly wealthy, like Boyle Heights, Filipinotown, and Koreatown.

The broader question is, how come those neighborhoods have not gentrified the way their counterparts in other American cities did?

The answer to this question has to be that Los Angeles is very weak-centered. The other five cities all have strong CBDs, which means the middle class is willing to pay extra to live near their centers. In Los Angeles, employment in the CBD is weaker, so fewer people of means try to concentrate there.

Place-based policy for commuters

Despite the fact that people who live 50 km from city center are noticeably poorer than people who live 50 km and work at city center, there’s an impulse to focus on rush-hour commuter transportation at this range. This can include highway widening, or commuter rail that is so peak-focused it’s essentially a highway widening, interfacing with the suburban road and parking network but not with any urban public transportation.

Even though the people such policy helps are better-off than most, governments still sell it as a social justice measure that would promote equity. The error here is that while people in (say) New Bedford are poorer than average, the local notables who decide what the New Bedford agenda is are richer than average, and they want to be able to say that they steered spending to the area in order to feel more important.

It’s an awkward situation in which money is wasted on grounds of both efficiency and equality. The local notables are on the wealthy side, like the CBD-bound commuters, but they’re a distinct group with mostly local ties, so they understand the needs of regional rail even worse than 9-to-5 commuters as as class do. So the money is wasted, and it’s wasted in a way that increases inequality rather than decreasing it.

Job access for the working class

The best place for job access for the working class remains city center. In Los Angeles, this is direct from the data: for all the talk about drive-until-you-qualify exurbs in the Inland Empire, incomes there remain higher than in East LA or South Central. But this is true even in the other cities, for two distinct reasons.

In some cities, like Chicago, there is notable directionality – that is, there is a favored quarter (the North Side) and an ill-favored one (the South Side). Job suburbanization generally goes in the direction of the favored quarter because that is where corporate management lives. In Washington, Amazon decided to build HQ2 in the direction of the favored quarter, in Virginia, and offered the ill-favored quarter, the lower middle-class Prince George’s County, a lower-end warehousing job center. This situation seems universal or nearly so: in Paris most job suburbanization goes to the western favored quarter, in Tel Aviv it goes to the northern favored quarter, and so on.

But not all cities have much directionality. New York doesn’t – go in any direction outside the Manhattan core and you’ll find poverty, whether it’s in Harlem, Corona, Bed-Stuy, Jersey City, or Bergen Hill, and go further and you’ll find reasonable levels of comfort.

That said, in New York, off-center jobs are awfully inaccessible. Creating more jobs in Harlem would be great for working-class black and Hispanic job seekers in the area, but would not be very accessible from Brooklyn or Hudson County. Even access from the Bronx may be compromised by East Side vs. West Side divisions: how much access does the South Bronx get to Uptown Manhattan’s biggest job center, Columbia?

What’s more, plans for decentralizing jobs in the New York region don’t focus on Harlem or Jersey City, just as plans in Washington go to Fairfax County and not PG County. The PennDesign plan for high-speed rail, dubbed North Atlantic Rail, calls for a job center on Long Island called Nassau Center, in a homogeneously comfortable part of the region.

So in all cases, keeping jobs as concentrated in city center as possible, and allowing the CBD to organically expand into nearby areas, ensures the best job access for everyone, but this is disproportionately helpful for lower-income workers. There just isn’t enough suburban poverty writ large to justify any deurbanization of job geography on equity grounds.