Category: Construction Costs

I Gave a Talk About Construction Costs

Two years ago, I gave a talk at NYU about regional rail, and as promised, uploaded slides the next day for discussion. Yesterday I gave another such talk, about construction costs.

But here there are two things to upload: the slides, and the data table. I’ve been intermittently adding cities to a spreadsheet of various urban rapid transit lines and their construction costs, and by now there is a total of 207 distinct items, ranging from 1 km extensions to 3-figure packages like 200-km GPX and 160-km Delhi Metro phases. The total length of the lines in this database right now is 3610 km, of which 2090 are underground. These are almost exclusively new lines – most of them aren’t even open, and most of the rest opened this decade, so be cautious since much of the cost estimation is ex ante and a number of the soon-to-open line on the list have had serious cost overruns.

I hope people make use of this dataset and the preliminary analysis contained in the slides, and I ask that people look at both, since the slides do have some interpretive notes about confounding variables. One note that I did not include in the slides and explained verbally is what source means in the table: media means I’m drawing costs from popular media, trade means trade media like Railway Gazette, plan means official plans (either ex ante or ex post), wiki means Wikipedia (as always, a reliable source for line length and station count, never cost), measured means I measured line length on Google Earth lacking any alternative. One item, Crossrail, has its tunnel cost coming from a freedom of information request submitted by an alert reader who I will credit upon request; the headline budget is somewhat higher as it includes surface improvements, a common confounder for regional rail projects (the RER E extension, for example, splits its budget about 50/50 between the tunnel and above-ground works).

More detailed analysis is forthcoming, either here or in print.

What is the Anglosphere, Anyway?

As I’m putting more and more urban rail lines and their construction costs into one table, I have to notice trends. One that I’ve talked about for many years is that construction costs in the Anglosphere are higher than in the rest of the developed world, not just in world leader New York but also in other American cities as well as in Britain, Canada, Singapore, and so on. For years I identified this with common law, which I no longer do. Instead, I want to expand on this by asking what exactly the Anglosphere even means.

The features of the Anglosphere

Within the developed world, a subset of countries consists of the Anglosphere. The core is Britain, the US, Canada, Australia, and New Zealand, but Ireland has to be on the list too, as should Singapore and to varying extents Israel and Hong Kong. Which features separate them from the remainder of the first world:

  • For the most part, they use English as their usual language – but Israel, Hong Kong, and Quebec do not, and Singapore only does as a public language while maintaining Chinese, Malay, and Tamil as home languages.
  • They use English common law – but Quebec uses a French-derived code for civil law.
  • They have extensive right to trial by jury – but Israel and Singapore have no juries.
  • They use single-member districts in elections – but Singapore and Hong Kong are undemocratic, Israel and New Zealand use proportional representation, Ireland uses single transferable vote, and Australia’s single-member districts use instant runoff (cf. France’s single-member districts with runoffs).
  • They have higher economic inequality than other developed countries, lower taxes and government spending, and weaker unions – but there are some exceptions (e.g. Canada and Australia are less unequal than Italy, and South Korea and Japan have lower taxes than most of the Anglosphere), and moreover the ranges within both the Anglosphere and the rest of the developed world are quite wide.
  • They make extensive use of privatization and public-private partnerships for infrastructure and services – but Stockholm contracts out its urban rail whereas no major American city does, and France built one of its recent high-speed lines, the one to Bordeaux, as a PPP.
  • The smaller countries see the US, the UK, or both as inspirations for what modern prosperity looks like – but Israel compares itself with both the US and Western Europe (especially Germany), Singapore’s cultural cringe extends toward both the US/UK and bigger East Asian countries, and Hong Kong is torn between Western and Chinese models.

Every distinguishing feature of the Anglosphere can be made to correlate with high construction costs, but that tells us little, because it could be that this is just a spurious relationship, the real cause being something else about the Anglosphere. When making a claim about what makes the US, UK, and Canada so expensive to build in, it’s useful to test it against special cases – that is, countries that are part of the Anglosphere in general but fail that specific criterion.

The legal system

With respect to common law, Quebec is the ideal testing ground. Montreal and Toronto share more social and economic features than do other pairs of major cities with their respective languages. A large Toronto premium over Montreal would suggest that remaining differences, such as the legal code or maybe the peculiarities of Quebec politics, matter to construction costs.

But what we see is the opposite. In the 2000s, Toronto and Montreal both built subway extensions at pretty reasonable costs. Since then, costs have risen in both cities in tandem, placing the planned Blue Line extension in Montreal and the planned Ontario Line and Scarborough replacement in Toronto among the most expensive non-New York subways. So it’s likely that common vs. civil law makes no great difference to costs.

Electoral politics

By the same token as with the use of common versus civil law, we can look at the electoral system. Israel and New Zealand use fully proportional elections, and Israel has national lists, without any local empowerment. Both countries have cheap recent electrification projects, but when it comes to tunneling, both Tel Aviv and Auckland are on the expensive side.

Conversely, France has single-member districts with runoffs; the lack of a spoiler effect weakens political parties, but they’re still stronger than in the US, and in practice independent candidates mostly run explicitly as left or right. Any reasonable mechanism for why single-member districts should raise construction costs should apply regardless of whether these districts are elected by plurality or with runoffs (and besides which, Melbourne has extreme costs and Sydney fairly high ones). And yet, French costs are decidedly average: Grand Paris Express is the median world subway by construction costs, and other Metro extensions in Paris and other French cities are somewhat cheaper.

Unions and inequality

The political factor – the Anglosphere’s socioeconomic policy is generally to the right of that of Continental European countries – has its own special cases too. The American left and center-left has in particular seized upon the importance of health care to construction costs, since the US has high health care costs and employers, especially in the public sector, are expected to pay most of the costs of workers’ health insurance. But the UK and Canada both have largely public systems that the American left uses as inspiration for its single-payer health care plans, and the UK also has very good cost control; and yet both countries have very high infrastructure construction costs. Singapore, whose health care system is private and unequal but also low-cost, has very expensive subway construction as well.

We can similarly look at inequality in general, or at union power. The correlation between inequality and national construction costs should be fairly high, if only because the Anglosphere has high inequality as well as high construction costs. However, per Branko Milanovic’s data for after-tax-and-transfers inequality, Canada, Britain, and Australia all have slightly lower inequality than Spain, and are comparable to Greece and Italy.

Unions can affect construction costs in either direction. The American center-right and right complain that the power of public-sector unions warps public incentives and forces high construction and operating costs, citing union hostility to productivity improvements that include layoffs, or such regulations as prevailing wage laws. However, the most unionized countries in the developed world are in Scandinavia, where costs are low. The OECD has union density figures by country, and the big cleave is Scandinavia versus the rest. The Anglosphere is on the weaker side.

Perhaps the correlation must then go the other way? That is, weak unions increase costs, for example by creating a siege mentality among those workers who do have stable union jobs (including rail workers, as the industry’s economic and political situation is friendly to unionization)? But the data does not support that, either. Spain’s union density is barely higher than the US’s and much lower than Britain’s, and Greece’s is comparable to Britain’s. The available data strongly suggests that union power has no effect on construction costs, positive or negative.

Could it be privatization?

Privatization and the reliance on PPPs is the least clean of the Anglosphere’s special features – that is, it is not always used throughout the countries I identify with the Anglosphere, and conversely it may be used elsewhere, even in countries with generally left-wing economic policy like Sweden. Nonetheless, among the political, legal, social, and economic factors, it is the only one I cannot rule out.

The issue is not precisely contracting out something, as Stockholm is doing with urban rail. Rather, it is more specifically privatizing the planning aspects of the state, such as engineering. Spain relies heavily on in-house engineering and design, while the US and UK, and by imitation the rest of the Anglosphere, prefer private consultants. To the extent I have cost comparisons within the same city or country with different levels of privatization, they’re suggestive that it matters: the publicly-funded LGV Est Phase 2 cost €19 million per kilometer (with a tunnel covering 4% of the route), the PPP LGV Sud-Europe-Atlantique cost €23 million per kilometer (with no tunnels), the two lines opening within a year of each other. This is not an enormous cost difference, but accounting for the tunnel makes the cost noticeable, perhaps a factor of 1.5.


Overrelying on a single case is not particularly robust. In light of the similarities between costs of different lines in the same city, and even those of different cities in the same country, the N for a quantitative comparison is not large – my data table currently has 38 unique countries, and even accounting for a few misses for which I haven’t included data yet, like Israel, the number is not much larger than 40. It is not responsible to use multivariable regressions or other advanced statistical techniques in such a situation.

In that case, looking at one or two cases provides a powerful sanity check. As far as I can tell, the Anglosphere’s tendency toward privatization and using consultants, often reinforced by different English-speaking countries learning one another’s practices, could be a serious cost raiser. However, the other special features of the Anglosphere – common law, winner-take-all elections leading to two-party systems, and weak unions and welfare states – are unlikely to have a significant effect.

Construction Costs are National

I’m about to send a thinktank a draft of a table of subway construction costs, and I’d like to preview one of the most important findings from the data. This is based on 125 distinct items, totaling 2,297 kilometers – some complete, some under construction, a handful proposed. I’ve alluded to this here before, for example when writing about national traditions (US, Soviet, UK) or about Russian and Nordic costs. But the basic observation is that construction costs are not really a feature of an individual metro line, but of a city, and usually an entire country.

What this means is that if one line in Madrid is cheap, then we can expect other lines in Madrid to be cheap, as well as in the rest of Spain; if one line in London is expensive, then we can expect other lines in London to be expensive, as well as in the rest of the UK. In fact, in both countries the construction costs of metro systems in the capitals also accord with the construction costs of intercity high-speed rail: cheap in Spain, expensive in Britain, with Germany somewhere between Spain and Britain and France somewhere between Spain and Germany.

Some examples

The examples in this section are somewhat cherrypicked to be the ones with narrower ranges, but there are very few examples with truly large ranges over a similar period of time (i.e. not secular increases as in Canada). I am specifically excluding regional rail, as it tends to be more expensive per kilometer than subways.

Panama: Line 1 cost around PPP$260 million per kilometer for 53% underground construction, and Line 2 is cheaper, around $150 million, but is entirely above-ground. This is consistent with a factor-of-2.5 underground premium over elevated lines, well in line with the literature.

Greece: Athens Line 4 is €104 million per km, with construction having started recently. Thessaloniki has two lines in the database, the main line due to open next year and an extension to Kalamata due to open in 2021, and Athens is also about to wrap up an extension of Line 3 to Piraeus. All cost figures may be found here on PDF-p. 9. The two Thessaloniki projects are respectively €135 million/km and €118 million/km, the former at least including rolling stock and I believe the latter too; the Athens Line 3 extension, without rolling stock and with somewhat wider stop spacing, is much cheaper, €61 million/km, but this rises to €82 million/km with rolling stock.

Sweden: the Stockholm Metro extensions under construction all cost pretty much the same per kilometer. Three extensions are under construction at once, in three different directions; per this source, the costs per kilometer (in kronor) are 1 billion, 1.25 billion, and 1.15 billion, with the most expensive of the three involving brief underwater tunneling.

Russia: I asserted in an old post that Russian construction is expensive, with only a handful of projects. Since then I’ve found a source asserting that the entire 2011-20 program is 1.3 trillion rubles, for what appears to be 150 km, 57% underground. This is in PPP terms $364 million per km. Other costs are vaguely in that range – Railway Gazette claims the cost of boring in Moscow is (again in PPP terms) $400-600 million/km, Line 11 is around $310 million/km for underground suburban construction, one line mentioned on Railway Gazette in St. Petersburg is $310 million/km underground, another St. Petersburg line is maybe $360 million/km.

What does this mean?

That there’s correlation between different cities’ construction costs within the same country suggests the differences in costs are predominantly institutional or socio-political, rather than geological. This is further reinforced by looking at countries with very similar socio-political regimes, namely the Nordic countries: all of them are cheap, and even though Stockholm and Helsinki both have similar gneiss geology, the Oslo line I use for comparison does not (and neither does somewhat more expensive Copenhagen).

To further reinforce the institutional point, the costs of high-speed rail in different countries seem to follow the same order as the costs of metros. Spain is cheap: Ferropedia quotes construction costs below €20 million per kilometer. The UK, in contrast, just announced a cost overrun on HS2, a 540 kilometer network, to £88 billion, and even allowing for future inflation, this is maybe 7 or 8 times as expensive as in Spain. France and Germany are in between, in the same order as their metro costs. China, as far as I can tell comparable to France in its metro construction costs, has a high-speed rail construction cost range somewhat higher than France’s, mostly explainable by using more (generally avoidable) viaducts.

New York Rolling Stock Costs are Skyrocketing

The Metropolitan Transportation Authority has just released its capital plan for 2020-4. The cost is very high and the benefits substantial but limited, and I urge people to look over criticism by Henry Grabar at Slate about elevators and Ben Kabak’s overview at Second Avenue Sagas. Here I am going to focus on one worrying element: the cost of the trains themselves, on both the subway and commuter rail.

I started comparing subway construction costs nearly ten years ago. Here’s an early post on Second Avenue Sagas, hoisting something I wrote in comments. Over here I started writing about this in 2011. Early on, I was asked about the costs of the trains themselves rather than the tunnels, and said that no, there’s no New York premium there. At the time the most recent rolling stock order for the subway was the R160, for which the base order cost was $1.25 billion for 620 cars (source, PDF-p. 34), or about $110,000 per meter of length. Commuter rail was similar, about $2 million per 25-meter-long M7 in the early 2000s and $760 million for 300 M8s of the same length in the mid-2000s. London’s then-current order, the S Stock, cost £1.5 billion for 191 trains and 1,395 cars, around $90,000 per meter of length for narrower trains; Paris’s MP 05, a driverless rubber-tired train, cost €474 million for 49 trainsets, around $140,000 per meter.

But since then, costs have rapidly risen. The gap is still far smaller than that for infrastructure, which New York builds for an order of magnitude higher cost than the rest-of-world median. But it’s no longer a rounding error. Subway rolling stock costs are rising, and commuter rail rolling stock are rising even faster. The latest subway order, the R211, costs $1.45 billion for 535 cars, or $150,000 per meter, for the base order, and $3.69 billion for 1,612 cars, or $130,000 per meter, including options. Commuter rail equipment costs, once about $100,000 per meter of train length, inched up to $2.7 million per car in 2013, or $110,000 per meter, and then rose to $150,000 per meter for the M9 order.

Construction costs: subway trains

The 2020-4 capital plan has showcased even further rolling stock cost escalation. Go to the link for the MTA capital plan again. On PDF-p. 23 there’s a breakdown of different items on the subway, and rolling stock is $6.057 billion for a total of 1,977 cars, of which 900 are 15 meters long and the rest (I believe) 18, for a total of $185,000 per linear meter.

I’ve blogged before about comparative costs of light rail and regional rail rolling stock. In Europe, both still cluster around $100,000 per linear meter for single-level, non-high-speed equipment. There is no apparent premium over early- and mid-2000s cost even without adjusting for inflation, which is not surprising, as the real prices of manufactured goods tend to fall over time. But what about metros? Here, too, we can look at first-world world comparisons.

In London, a recent Piccadilly line order is, in exchange rate terms, $190,000/meter (the trains are 103 m long) – but it includes 40 years of maintenance and spare parts. In Singapore, a recent order is S$2.1 million per car, which is about $70,000 per meter in exchange rate terms. Grand Paris Express’s first tranche of orders costs €1.3 billion for 183 trains totaling 948 cars, each (I believe) 15 meters long, around $120,000 per meter. Metro Report states Busan’s recent order as 55.6 billion for 48 trainsets (replacing 140-meter long trains), which is almost certainly an error; assuming the actual cost is 556 billion, this is $70,000/meter in exchange rate terms and $90,000/meter in PPP terms (PPP is relevant as this is an entirely domestic order).

In Berlin, the situation is the diciest, with the highest costs outside New York (not counting London’s maintenance-heavy contracts). An emergency order of 20 52-meter trains, tendered because cracks were discovered in the existing trains, cost €120 million, around $150,000 per linear meter. A longer-term contract to supply 1,500 cars (some 13 meters long, most 16.5 meters long) for €3 billion by 2035 is on hold due to litigation: Siemens had already sued over the emergency order of Stadler cars, but now Alstom made its own challenge. But even here, costs are well below the levels of New York, even before we adjust for inflation since Berlin’s future contract is in 2020-35 prices and New York’s is in in 2020-24 prices.

Construction costs: New York-area commuter rail

Commuter rail is faring even worse. On PDF-p. 27 the LIRR is listed as spending $242 million on 17 coaches and 12 locomotives, and on PDF-p. 29 Metro-North is listed as spending $853 million on 80 EMU cars and 30 locomotives.

Figuring out exact comparisons is not easy, because locomotives do cost more than multiple-units and unpowered coaches, and there is a range of locomotive costs, with uncertainty due to currency conversions, as most information I can find about European locomotives is in Eastern Europe with its weak currencies, since Western Europe mostly uses multiple-units. Railway Gazette’s pages on the world rolling stock market suggest that a European locomotive is around €5 million (e.g. the PKP Vectron order), or $6.5 million; PKP’s domestic order (including some dual-modes) is around $4.2 million per unit measured in exchange rate terms, but twice as much in PPP terms; Bombardier has a sale to an undisclosed customer for about $4.8 million. Siemens claims the Vectron costs €2.5 million per unit, although all the contracts for which I can find prices are substantially more expensive.

For what it’s worth, in the US dual-mode locomotives for New Jersey Transit cost around $9.5 million apiece, which is still evidently lower than what the LIRR and Metro-North plan on spending. 242 – 9.5*12 = 128, and 128/17 = 7.5, or $300,000 per linear meter of unpowered coach; similarly, 853 – 9.5*30 = 568, and 568/80 = 7.1, or $280,000 per linear meter of new Metro-North EMU. If we take the normal-world cost of a locomotive at $6 million and that of an EMU or coach at $2.5 million per US-length car, then the LIRR has a factor-of-2.1 cost premium and Metro-North a factor-of-2.2 premium.

The equipment is conservative

The FRA recently realigned its regulations to permit lightly-modified European mainline trains to run on American tracks. Nonetheless, no American commuter rail operator has taken advantage of the new rules – the only ones buying European equipment had plans to do so even before the revision, going through costly waiver process that increased costs. At a public meeting last month, Metro-North’s vice president of engineering did not even know FRA rules had changed. The LIRR and Metro-North are buying the same equipment, to the same standards, as they have for decades.

The subway, likewise, is conservative. It is a laggard in adopting open gangways: the R211 order is the first one to include any, but that is just two test trainsets, the rest having doors between cars like all other older New York trainsets. It is not buying any of the modular products of the global vendors, like Bombardier’s Movia platform or the Alstom Metropolis. It is buying largely the same kind of equipment it has bought since the 1990s.

Despite this conservatism, costs are very high, consistent with a factor somewhat higher than 2 on commuter rail and somewhat lower than 2 on the subway.

But perhaps the conservatism is what increases costs in the first place? Perhaps the reason costs are high is that the world market has moved on and the MTA and some other American operators have not noticed. In Chicago, Metra found itself trying to order a type of gallery car that nobody makes any longer, using parts that are no longer available. Perhaps the same kind of outmoded thinking is present at the MTA, and this is why costs have exploded in the last 10 years.

A secular increase in costs of infrastructure construction is nearly universal. No such trend can be seen in rolling stock: nominal costs in Paris are 15% lower than they were 15 years ago, and real costs are about 30% lower, whereas in New York nominal costs are 70% higher than 10 years ago and real costs about 40% higher. Paris keeps innovating – M1 and M14 have the highest frequency of any metro system in the world, a train every 85 seconds at the peak, and M1 is the first driverless line converted from earlier manual operations rather than built from scratch. In contrast, New York is stuck in the 1990s, but far from keeping a lid on costs, it has seen rolling stock cost explosion.

Update 9/24: I just saw a new commuter rail coach order in Boston. These are bilevels so some cost premium is to be expected, but $345 million for 80 unpowered coaches, or $170,000 per meter, is excessive, and TransitMatters tried hard to fight against this order, arguing in favor of EMUs on the already-electrified Providence Line.

Why I Write About Rail Costs, not the Cost of Other Things

I’ve been asked from time to time, Alon, you write about comparative rail costs all the time, but what about roads? Sometimes the question expresses curiosity about whether roads display the same American construction cost premium as urban rail does; sometimes it expresses frustration that The Discourse doesn’t complain about road costs. Regardless of why people ask, I’d like to explain my reasoning in depth, especially now that serious people are asking why this is the focus of my comparative research.

There’s an easy answer and a hard answer. The easy answer is that I’m a railfan. I got into this because I was living in Morningside Heights and taking the subway to social events in Brooklyn and Queens, which involved 3- and sometimes 4-seat rides. It got me interested in coverage gaps and subway extensions, which got me interested in the construction costs of such extensions.

But that’s not really it. From my original purpose of comparing a few urban infill subways in large global cities I got into operating costs, and high-speed rail, and light rail, and electrification, and even road tunnels (here is my comparison of urban road tunnel projects). What’s more, other people have looked at comparative costs, and even without sharing my not-knowing-how-to-drive origin story, they don’t compare individual road projects much. The Brookings study about the Interstates looked at the entire cost of the US Interstate program rather than teasing it out project by project.

What’s really going on is that subways are megaprojects. Megaprojects are visible, and I don’t just mean physically – they’re widely discussed in the media and politics, and cost overruns invite intense criticism by the opposition and by investigative reporters. Everybody in New York knows about Second Avenue Subway, and everybody in New Jersey knows about the Gateway tunnel, and everybody in London knows about Crossrail.

The upshot is that megaproject cost estimates are just more reliable than those of anything else. What I mean is not that cost overruns are unlikely. Rather, what I mean is that cost overruns are difficult to hide, unless the agency goes the Canadian route of fluffing the budget with very high contingencies. The current budget for Grand Paris Express is around €35 billion, up from €25 billion when it was first announced. If it actually ends up at €36 billion and not €35 billion then it may be possible to scrounge extra funds from a few sources sub rosa, but not if it ends up at €45 billion.

The largest source of wasteful spending in the world is the American military. It has a budget of $700 billion a year, debated largely behind the scenes, with boisterous generals and their lackeys ready to publicly defend every $600 toilet seat and every procurement item in the district of any member of Congress who dares object. There is a shroud of secrecy around everything that can be justified as national security. There is no exit threat – the military can’t be shut down the way an underperforming state railroad can be privatized. Hidden costs are rampant, and as far as I understand, they are on the order of a few billion dollars at a time.

I bring up American military waste not to justify civilian waste on infrastructure, but to compare which costs can be plausibly hidden. If the US military can miss a few billion dollars, the transport planners of Ile-de-France can miss tens to hundreds of millions of euros on a 15-year, 200-kilometer project. Those of Madrid can probably miss an amount of money on the same order of magnitude as those of Paris. The low construction costs in Madrid have been plugged into additional construction, giving Madrid Europe’s third longest metro network after London and Moscow; those hundreds of kilometers built in the last 25 years could not have cost the same as in France, let alone the US, because this would have been too big of a difference, and the media would have noticed.

The same situation equally occurs for road megaprojects, such as tunnels or big urban reconstruction projects, such as the lane additions in Los Angeles. But it does not occur for run-of-the-mill road widening outside urban areas or for small projects to increase the capacity of a junction from a cloverleaf to a four-level interchange. These are not sufficiently visible for me to be able to trust that there is full cost accounting in the trade and popular press.

I’m happy to compare the costs of road tunnels between different cities; the few examples I have found paint the same picture as the subway cost comparison. But above-ground road construction is harder, just because “above-ground” can mean anything from a complex viaduct-over-viaduct to simple at-grade construction. Even then, ancillary costs like unnecessary street reconstruction may be bundled into the overall budget, and since above-ground construction isn’t so expensive, these extras may be a sizable fraction of the cost.

For a similar reason, I don’t look at airports so much: they’re just harder to compare. I do not know how big the Berlin-Brandenburg disaster is compared with other airports under construction, so I do not know how much it should cost; I don’t even know what the equivalent metric of cost per km or cost per new station excavated is. In contrast, to take another well-known German infrastructure disaster, Stuttgart21 has a definite tunnel length – 30 kilometers, as well as another 25 above ground – so I can compare with other regional rail projects and say that actually the cost of Stuttgart21 (€6.5 billion) is not so high relative to how much urban mainline rail tunneling costs elsewhere in the world.

For the exact same reason, when I look at above-ground urban rail I try to separate out truly at-grade light rail from elevated lines. The only times I try to do a deep dive are when these projects encroach on the cost range of subways, like the Boston Green Line Extension. Elsewhere, ancillary costs can be substantial, as with the Nice tramway: 70% of the budget was the tramway itself and 30% was stormwater drainage, rebuilding a public plaza, tree planting, and other extras. Extras introduce an error term into comparisons that are harder to ignore when the cost is $50 million per kilometer than when it is $300 million per kilometer.

Road costs remain a powerful sanity check. All of the reasons I (and others) believe are behind the American construction cost premium are equally applicable to roads and urban rail. So far, looking at road tunnels confirms the subway pattern, but there just aren’t a lot of road tunnels built around the world – they’re expensive for the capacity they provide. And if it’s possible to carefully tease out above-ground road megaproject costs then a comparison is welcome as well. But they are unlikely to form the backbone of any comparison.

Metro tunnels, for all the handwringing about special circumstances, are pretty consistent. Some places have easier rock and some have harder rock, but usually this will be noted in the trade and popular press; the most fundamental quantities, length and the number of stations, are if anything easier to find than the headline costs; ancillary extra costs are usually not significant, and when they are, they tend to be bundled into quantifiable metrics like station size and depth. The only big difference in reporting regimes is that some places (like Spain) bundle together infrastructure and rolling stock costs whereas most don’t.

The main approach to project-level comparison of infrastructure costs across countries has to be about urban rail, because that’s by far what’s most common across the world. The error bars around ex post costs are small enough that even a relatively restricted sample is suggestive of the real global effect as I’m learning when adding more and more projects to my database (currently about 130 projects totaling 2,000 km). This is the most comparable list of public infrastructure projects, and what we may learn about why various American urban rail lines cost so much and why Spanish and Korean and Nordic ones cost so little is likely to generalize.

New Report on Construction Costs Misses the Mark

In the last few years, ever more serious and powerful actors have begun investigating the fact of high American infrastructure construction costs. First it was Brian Rosenthal’s excellent New York Times exposé, and then it was the Regional Plan Association’s flop of a study. At the same time, I was aware that the congressional Government Accountability Office, or GAO, was investigating the same question, planning to talk to sources in the academic world as well as industry in order to make recommendations.

The GAO report is out now, and unfortunately it is a total miss, for essentially the same reason the RPA’s report was a miss: it did not go outside the American (and to some extent rest-of-Anglosphere) comfort zone. Its literature review is if anything weaker than the RPA’s. Its interviews with experts are telling: out of nine mentioned on PDF-p. 47, eight live in English-speaking countries. Even when more detailed information about non-English-speaking countries is readily available, even in English, the GAO report makes little use of it. It is a lazy study, and people who ideologically believe the American federal government does not work should feel confident citing this as an example.

Cost comparisons

Brian himself already notes one of the reasons the report is so weak: Congress mandated a comparative study, but the report made no international comparisons at all. Instead, the report offered this excuse (PDF-p. 27):

The complexity of rail transit construction projects and data limitations, among other things, limits the ability to compare the costs of these projects, according to the stakeholders we interviewed. As highlighted above, each project has a unique collection of specific factors that drive its costs. According to FTA officials, each proposed transit project has its own unique characteristics, physical operating environment, and challenges. Some stakeholders said that the wide disparity in the relative effect of different cost factors renders cost comparisons between projects difficult. For example, representatives of an international transit organization said that because of the large number of elements that can affect a project’s costs and the differences in what costs are included in different projects’ data, projects should be compared only at a very granular level and that aggregate cost comparisons, such as between the costs per mile or costs per kilometer of different projects, are likely flawed. Some stakeholders also said that project costs should not be compared without considering the projects’ contexts, such as their complexity. For example, one academic expert contended that project costs cannot be compared without considering the context of each project, and that analysis of projects should focus on leading practices and lessons learned instead.

There is a big problem with the above statement: disaggregated costs for many aspects of urban rail construction do exist. The Manhattan Institute’s Connor Harris has done a lot of legwork comparing tunnel boring machine staffing levels and wages in New York and in Germany, and found that New York pays much higher wages but also has much higher staffing levels, 25-26 workers compared with 12. I have done some work looking at station costs specifically, and at the cost of installing elevators for wheelchair accessibility.

There is a lot of detailed comparative research about the costs of high-speed rail; the report even references one such meta-study undertaken within Europe, but omits the study’s analysis of causes of cost differences and instead asserts that it shows that comparing different projects is hard. In the interim, California contracted Deutsche Bahn to do a post-mortem of its elevated high-speed rail costs, which found that California needlessly built larger structures than necessary, explaining its cost premium over Germany.

Instead of probing these disaggregated estimates, the GAO preferred to say that they are too hard and move on.

Sanity checks

Even without disaggregation, there are some good sanity checks one can make about construction costs. The most important is that big projects – major subway expansions, regional rail tunnels, high-speed rail – cost an appreciable amount of the government’s budget. The budget for the 200-kilometer Grand Paris Express project is €35 billion, plus another €3 billion in contributions for related suburban rail extensions such as that of the RER E. There may be future cost overruns, but they will be reported in the media, just as the current overrun has been; it is extremely difficult to hide cost overruns measured in tens of billions in a Paris-size city, and even in a China-size country it may not be easy.

Is it plausible that GPX is inherently easier to build than New York’s $1+ billion/km subway tunnels? Yes. It’s equally plausible that it is inherently harder. Second Avenue Subway runs under a wide, straight throughfare, a situation that simplifies construction. In Israel, the ministry of transportation has long mentioned the ease of tunneling under wide, straight boulevards in connection with plans to extend the second line of the Tel Aviv subway to North Tel Aviv under Ibn Gabirol, and admitted this even when it opposed the extension on land use grounds.

The most important sanity check is that in a world with several dozens of cities with a wide variety of wealth levels, land use patterns, geologies, and topographies, no city has managed to match or even come close to New York’s construction costs. New York is not special enough to be an edge case in all or even most relevant geographic variables – it is dense but no denser than Seoul or Paris, it is wealthy but no wealthier than London or Paris or Munich and barely wealthier than Stockholm, it has hard rock but less hard than Stockholm (and in Stockholm the gneiss is cited as a cost saver – bored tunnels do not require concrete lining), etc.

Moreover, the cities that have the highest construction costs outside New York are almost without exception in the same set of countries: the US, Canada, Britain, Singapore, Australia. What’s likelier – that there is some special geographic feature common to the entire Anglosphere (including Quebec) but absent from all other developed countries, or that there is a shared set of legal and political traditions that developed in the last 50 years that impede cost-effective construction? Instead of probing this pattern, the GAO preferred to wash its hands and refuse to compare projects across countries.

Internal comparisons

In lieu of making international comparisons, the GAO has engaged in extensive internal comparison. It cites aspects that have raised the costs of Second Avenue Subway above other American subway projects, such as overdesign for stations. Apparently, it’s completely legit to compare two different cities’ construction if they’re in the same country.

Over and over again, it references its own domestic standards. The GAO has 12 design standards, e.g. on PDF-pp. 51-52 and 56-60; the report mentions that existing cost estimation methodologies by the Federal Transit Administration, or FTA, meet 7 of them; thus, it exhorts transit agencies to meet the other 5 standards.

The only problem is that there is no evidence supplied that those design standards are really useful. After all, the United States has very high costs, so why should anyone trust its standards? Even domestically, the report makes no effort to bring up successful examples of low overall costs coming from following prescribed standards. Seattle recently opened a light rail tunnel built for around $400 million per kilometer, a cost that would get most European project managers fired but that is still the lowest for an American urban rail tunnel built in this century. But the report never brings up Seattle at all, never mind that New York would salivate over the prospect of tunneling at Seattle’s cost.

The real internal comparisons then are not between different cities in the United States. Rather, they’re between different stages of cost estimation for the same project. There is published literature on cost overruns, most famously by Bent Flyvbjerg and his research group. The report cites Flyvbjerg. Moreover, one of the nine academic experts it consulted is Don Pickrell, who published a seminal paper on American cost overruns and ridership shortfalls in 1990. Pickrell was influential enough that a 2009 review found that not only had cost and ridership projections improved greatly in the intervening two decades, but also there was an improvement in ridership estimate quality attributable to Pickrell’s paper.

The GAO report is not the best source on cost overruns, but it is not completely useless there. Unfortunately, it remains useless when it comes to discussing absolute costs, a different topic from relative increases. Flyvbjerg’s original paper found that the US did not have higher cost overruns than Europe; but absolute costs in the US are several times as high. Flyvbjerg’s paper found that urban rail has higher cost overruns than road projects; but when a rail tunnel and a road tunnel are built in the same city, the road tunnel is more expensive by a factor of 1.5-2.5, at least in the four-city pilot I reported in 2017, owing to the need to build bigger bores with ventilation to carry heavy car traffic.

Lazy analysis, lazy synthesis

Americans who think of themselves as reformers like to point out real problems to solve, but then propose solutions that they made up without any connection with their analysis. The RPA study is one such example: even though one of its sources (namely, former Madrid Metro CEO Manuel Melis Maynar’s writeup about low Spanish costs) explicitly calls for separation of design and construction, its recommendations include a greater reliance on design-build. The same design-build recommendation appeared in a 2008 report in Toronto comparing the costs of the Sheppard subway, opened in 2002, with those of subways in Madrid; construction costs in Toronto have since tripled, while those of Madrid have barely risen.

To the GAO report’s credit, it does not recommend design-build. It even mentions the biggest drawback of design-build: it shifts cost risk to the private contractor, who compensates by demanding more public money up front. Nonetheless, it does not follow through and does not make the correct recommendation on this subject – namely, that cities and states should cease using this approach. It buries a recommendation for in-house expertise alongside a fad for peer review of projects.

Instead of lazily proposing design-build, the GAO lazily proposes two barely relevant tweaks (PDF-p. 43):

  • The FTA administrator should ensure that FTA’s cost estimating information for project sponsors is consistent with all 12 steps found in GAO’s Cost Estimating and Assessment Guide and needed for developing reliable cost estimates.
  • The FTA Administrator should provide a central, easily accessible source with all of FTA’s cost estimating information to help project sponsors improve the reliability of their cost estimates.

In other words, the report makes no recommendation about how to reduce costs, only about how to tell the public in advance that costs will be unaffordably high.

Why are these reports so bad?

This is not the first time a serious group releases an incurious study of American construction costs. What gives?

I suspect the answer has to be a combination of the following problems:

  • Reform factions often have a lot of internal ideas about how to improve things based on what they already know. They will cite new information if they feel like they must do so to save face, but they will not let new evidence change their conclusions. A little knowledge can be dangerous.
  • Finding information from outside the US, especially outside the English-speaking world, puts Americans (and Canadians) at a disadvantage. They know few to no foreigners, have little experience with cities abroad except as tourists, and do not speak foreign languages. Even when machine translation is decently accurate, which it is in the engineering literature in European languages, they are intimidated by the idea of dealing with non-English material. The process of learning is humbling, and some people prefer to remain proud and ignorant.
  • Open-ended analysis does not always lend itself to easy explanations or easy solutions. Even when solutions do present themselves, they may not flatter the people in power. Ten years ago I did not think senior management at American transit organs should be fired; today I think mass layoffs of the top brass, especially the political appointees, are somewhere between very useful and essential.

All three problems interact. For example, senior management is even less likely to be multilingual than junior staffers, who may be second-generation immigrant heritage speakers of a foreign language; thus, anything relying on foreign material disempowers the high-ups in favor of up-and-comers. The quick-and-easy-and-wrong solutions reformists seize upon if they find a little bit of knowledge let outfits like the GAO feel more powerful without actually challenging any obstructive politician or interest group, and if those solutions fail, they can always keep churning reports about implementation.

Last year, I did not know whether the GAO was capable of providing a blueprint for improving American infrastructure at lower cost. I assumed good faith because I had no reason not to. With this report, it is clear to me as well as to other observers of American public transit that the GAO is not so capable. Instead of doing what was in the country’s best interest, the people who commissioned and wrote the report delivered the minimal product that would get them kudos from superiors who do not know any better. They could have learned, or made a serious effort to learn, but that might challenge their assumptions or those of the high political echelons, and thus they preferred to say nothing and propose to do nothing.

Costs are Rising, US Highway Edition

There’s a preliminary paper circulating at Brookings, looking at American infrastructure construction costs. Authors Leah Brooks and Zachary Liscow have tabulated the real costs of the American Interstate program over time, from the 1950s to the 1990s, and find that they increased from $5.3 million per km ($8.5 million/mile) in 1958-63 to $21.3 million/km ($34.25 million/mile) in 1988-93.

Moreover, they have some controls for road difficulty, expressed in slope (though not, I believe, in tunnel quantity), urbanization, and river and wetland crossings, and those barely change the overall picture. They go over several different explanations for high American infrastructure costs, and find most of them either directly contradicted by their results or at best not affirmed by them.

I urge readers to read the entire paper. It is long, but very readable, and it is easy to skip the statistical model and go over the narrative, including favored and disfavored explanations, and then poke at the graphs and tables. I’m going to summarize some of their explanations, but add some important context from cross-national comparisons.

Why costs (probably) aren’t rising

The authors identify four hypotheses they rule out using their research, in pp. 19-23 (they say five but only list four):

Difficult segments postponed and built later – they have some controls for that, as mentioned above. The controls are imperfect, but the maps depicted on pp. 59-61 for the Interstate network’s buildout by decade don’t scream “the segments built after 1970 were harder than those built before.”

Time-invariant features – these include cross-national comparisons, since the United States has always been the United States. I will discuss this in a subsequent section, because two separate refinements of what I’ve seen from cross-national comparisons deal with this issue specifically.

Input prices – this is by far the longest explanation the authors deal with. Anecdotally, it’s the one I hear most often: “labor costs are rising.” What the authors show is that labor and materials costs did not rise much over the period in question. Construction worker wages actually peaked in real terms in 1973 and fell thereafter; materials costs jumped in the aftermath of the oil crisis, but came down later, and were back at pre-crisis levels by the 1990s (p. 48). Land costs did rise and have kept rising, but over the entire period, only 17.7% of total costs were preliminary engineering and land acquisition, and the rest were in construction.

Higher standards – the authors looked and did not find changes in standards leading to more extensive construction.

There are several more incorrect explanations that jump from the data. I was surprised to learn that throughout the 1970s and 80s, completion time remained mostly steady at 3-3.5 years of construction; thus, delays in construction cannot be the explanation, though delays in planning and engineering can be.

The authors themselves list additional explanations that have limited evidence but are not ruled out completely from their data, on pp. 32-35. Construction industry market concentration may be an explanation, but so far data is lacking. Government fragmentation, measured in total number of governments per capita, has no effect on the result (for example, California has high costs and not much municipal fragmentation); I’ll add that Europe’s most municipally fragmented country, France, has middle-of-the-road subway construction costs. State government quality, as measured by corruption convictions, has little explanatory power – and as with fragmentation, I’ll add that in Europe we do not see higher costs in states with well-known problems of clientelism and corruption, like Italy and Greece. Work rules requiring the addition of more workers may be relevant, but unionization and left-right politics are not explanatory variables (and this also holds for rail costs).

Economies of scale look irrelevant as well: there is negative correlation between costs and construction, but the causality could well go the other way. Finally, soft budget constraints are unlikely, as the federal government can punish states that mismanage projects and take more money; it’s possible that as the Interstate program ended states felt less constrained because there wouldn’t be money in the future either way (“end of repeated game”), but the fact that costs keep rising in subway construction suggests this is not relevant.

Favored explanations

Two explanations stand out to the authors. The first is that nearly the entire increase in construction costs over time can be attributed to a mix of higher real incomes and higher house prices. While the construction workers themselves did not see their wages rise in the late 1970s and 80s, a richer population may demand more highways, no matter the cost.

Higher real estate costs could have an impact disproportionate to the share of land acquisition in overall costs by forcing various mitigations that the paper does not control for, such as sound walls and tunnels, or by sending roads over higher-cost alignments.

The second explanation is what the authors call citizen voice. Regulatory changes in the 1960s and early 70s gave organized local groups greater ability to raise objections to planning and force changes, reducing community impact at the cost of higher monetary expenditures. The authors give an example from suburban Detroit, where a highway segment that disrupted a Jewish community center took 25 years to be built as a result of litigation.

The authors don’t say this explicitly, but the two explanations interact well together. The citizen voice is very locally NIMBY but is also pro-road outside a handful of rich urban neighborhoods. Higher incomes may have led to public acceptance of higher costs, but local empowerment through citizen voice is the mechanism through which people can express their preference for higher costs over construction inconvenience.

How time-invariant are national features, anyway?

The authors contrast two proposed explanations – higher incomes and property values, and stronger NIMBY empowerment – with what they call time-invariant features, which could not explain an increase in costs. But can’t they?

I spent years plugging the theory that common law correlates with high subway construction costs, and it does in the developed world, but upon looking at more data from developing countries as well as from before the last 25 years, I stopped believing in that theory. It started when I saw a datapoint for Indonesia, a civil-law country, but even then it took me a few more years to look systematically enough, not to mention to wait for more civil-law third-world countries to build subways, like Vietnam. By last year I was giving counterexamples, including Montreal, low rail electrification costs in some common law countries, and the lack of a London cost premium over Paris until the late 20th century.

In lieu of common law, what I use to explain high costs in the US relative to the rest of the world, and to some extent also in most first-world common law countries as well as third-world former colonies, is weak civil service. In the developed world, the theory behind this is called adversarial legalism, as analyzed by Robert Kagan. Adversarial legalism enforces the law through litigation, leading to a web of consent decrees. Some are naked power grabs: for example, in Los Angeles, a union sued a rolling stock vendor for environmental remediation and agreed to drop the lawsuit in exchange for a pledge that its factory be unionized, which may play a role in why the trains cost around 50% more than equivalent European products.

American litigiousness developed specifically in the 1970s – it’s exactly how what the authors of the paper call citizen voice is enforced. In contrast, on this side of the Channel, and to some extent even generally on this side of the Pond, laws are enforced by regulators, tripartite labor-business-government meetings, ombudsmen, or street protests. French riotousness is legendary, but its ability to systematically change infrastructure is limited, since rioting imposes a real cost on the activist, namely the risk of arrest and backlash; in contrast, it is impossible to retaliate against people who launch frivolous lawsuits.

I bring up the fact that I said most of this last year, and the rest at the beginning of this year, whereas I was not aware of the paper under discussion until it was released a few hours ago, to make it clear that I’m not overfitting. This is something that I’ve been talking about for around a year now, and a jump in American construction costs in the 1970s and 80s – something that also looks to be the case in subway construction – is fully compatible with this theory.

Costs are Rising

This is a partial data dump from an in-progress database I’m compiling for subway construction costs around the world. The key point is that costs are rising: in cities with enough historical data points we can see a secular increase in construction costs. The difference between expensive cities like New York and London is that their costs have been high for a while, whereas cheap ones like Madrid and Seoul are seeing construction cost growth from a very low basis.

As a note of caution, while growth in costs seems universal, the rate of growth is not the same everywhere. Some cities, most notably Singapore and Toronto, have seen a cost explosion in the last 15-20 years; others, most notably Seoul, have seen only a moderate increase in costs.

I also urge readers to look at some 20th century historical costs here, as in this post I am going to focus on the very end of the 20th century and the 21st century.

Paris: the original Metro Line 14 cost €1.174 billion for 9.2 km (link), built between the 1990s and 2007; deflated to 2012 euros, the baseline year used for ongoing extensions, this is around €160 million per km. More recent projects in Paris cost around the same, including the Line 1 extension and Grand Paris Express – but those are mostly suburban extensions, whereas M14 had to go underneath central Paris.

Toronto: Jonathan English, who has been working with me on Canadian construction costs issues, notes that the Sheppard subway, opened in 2002, cost C$1 billion for 5.5 km, but ongoing projects are far more expensive. The one-stop 6.2 km Scarborough subway is projected to cost around C$3.5 billion not including extra items for interfacing with the existing rapid transit line along the same alignment, which is to be dismantled. Not only is the nominal cost 3 times higher – and the real cost is still around twice as high – but also the Scarborough subway has just one station to be constructed, which makes it a simpler project.

Montreal: Jonathan equally looks at the cost explosion in Montreal. The Laval extension was built in the 2000s and opened in 2007, costing C$742 million for 5.2 km and 3 stations, or C$143 million per km, crossing under the Rivière des Prairies. In contrast, an extension of the Blue Line planned for next decade is to cost C$3.9 billion for 5.8 km and 5 stations, or C$672 million per km – and even adjusting for inflation only reduces the cost differential to a factor of about 3.

(In case Canadian readers wonder why I’m not covering Vancouver, even though the escalations on the Broadway subway have pushed its per-km cost well beyond that of the Canada Line, the reason is that the Canada Line was built cut-and-cover whereas the Broadway subway will be bored.)

Singapore: Singapore’s cost overrun history in the 21st century has been unusually severe. Built mainly in the 2000s, the original Circle Line cost S$10 billion for 33.3 km, or S$300m/km, an overrun of 50% over the original budget. Subsequently, the Downtown MRT Line, built from 2008 to 2017, cost S$21 billion for 42 km, and the Thomson Line S$24 billion for 43 km. The Thomson Line has a complex interchange at Orchard, but also long segments in easy suburban areas – Upper Thomson Road, after which it is named, is very wide and borders modernist housing projects on one side and a forest on the other. Moreover, the last stage of the Circle Line, completing the circle, is to cost S$4.85 billion for 4 km and 3 stations – depending on PPP rates, it may be the first line outside New York to cross the US$1 billion/km line.

Seoul: South Korean costs are fairly stable. JRTR has data for the Seoul Metro going back to its start in the 1970s. After adjusting for inflation, costs were initially about $70 million per km, and rose gently to $80-90 million. The cost increases are continuing, albeit at a slow pace. As best as I can tell, the 2020s’ expansion program is budgeted at about $110 million per km in PPP terms.

Madrid: in the 1995-2003 period the city built tunnels for very low costs. The 1995-8 program cost $55 million per km, all underground, and the 1999-2003 program cost €3.147 billion for 74.7 km, 77% underground, around $52 million per km based on the era’s PPP conversion rate. In the conditions of 2010 this would be roughly $65-70 million per km – but the Line 2 extension, built 2008-11, cost €315 million for 4.6 km and 4 stations, and the Line 9 extension, built 2009-15, cost €191 million for 3 km and 2 stations, about $80-90 million per km.

Update: since people have asked for high-speed rail data, it confirms the same story. Ferropedia has costs in Spain, which have risen from €4.88 million per kilometer in 2001 terms for Madrid-Seville, which opened in 1992, to about €15-20 million per kilometer in 2006-7 terms for subsequent lines from the 2000s and 2010s. France displays the same history of escalation: built in the early 1980s, the LGV Sud-Est cost €5.5 million per km, much less than the late 1980s and early 1990s’ LGVs Nord and Atlantique (which cost €10 million), let alone this decade’s LGV Est (which cost €16 million for Phase 1 and €19 million for Phase 2); all of these lines are through comparable terrain, with very little to no tunneling.

Assume Nordic Costs: London Edition

A month ago I made maps proposing some subway and regional rail extensions in New York and noting what they would cost if New York could build as cheaply as the Scandinavian capitals. Here is the same concept, but with London rather than New York. Here is everything in a single large map:

A full-size (74 MB) map can be viewed here.

Solid lines are existing or under construction, that is Crossrail and the Battersea extension; proposed lines are dashed. Commuter rail lines, that is Thameslink, the soon-to-open Crossrail, and four additional Crossrail tunnels labeled 2 through 5, are always depicted as having separate stations from the other modes, to avoid confusion where one Crossrail station has connections to two adjacent Tube stations (such as Farringdon-Barbican and Moorgate-Liverpool Street). It has many additional interchanges between lines and branches, including some that were left out on purpose, like a Crossrail 1 connection to Oxford Circus, omitted from the under-construction line to discourage riders from using the oversubscribed Victoria line; with four more cross-city lines, the capacity problems would be lessened substantially.

The overall picture is sparser than my New York map. The total projected cost of all of these projects, including some allocated for redoing stations on commuter branches to be given to Tube lines, is £6.8 billion, compared with $37 billion for the New York maps. The reason is that unlike New York, London already has excellent coverage thanks to extensive branching – what it needs is core capacity, which consists of city center tunnels that have high cost per kilometer but need not be long.

There is considerable overbuilding planned in London. Crossrail 2 as depicted on my map is a 6.5 km tunnel between the approach to Victoria Station and the approach to Kings Cross. But as planned, Crossrail 2 extends to a long tunnel parallel to the South West Main Line, a four-track line in a right-of-way that could if truly necessary accommodate six, as well as a long tunnel going north to take over the Lea Valley Lines, which on my map go into Crossrail 5. With gratuitous suburban tunnels and extremely high British construction costs, the budget for Crossrail 2 is around £30 billion, about 20 times what Scandinavia might spend on such a project. Even allowing for the possibility that crossing under three lines at once at Bank is more complex than crossing under two at T-Centralen, this is a difference of a full order of magnitude, counting both total required tunnel length and cost per km.

In addition, there is network simplification. On the Tube this consists of segregating the Northern line’s Bank and Charing Cross branches (already in planning pending the Battersea extension and reconstruction of Camden Town) and through breaking the Circle line into separate Metropolitan and District lines. The latter was estimated by a British blogger to cost £5 billion, based on a rubric in which the Met/District transfer at Aldgate (or Tower Hill) should by itself cost £1 billion; Crossrail and Second Avenue Subway stations cost around half that much, and the more complex T-Centralen and Odenplan stations on Citybanan cost less.

On mainline rail, the service plan is supposed to be deinterlined, as is Transport for London’s long-term goal. The slow tracks of the various mainlines feeding into Central London turn into Crossrail branches, or occasionally Underground extensions, such as Hayes and the Hounslow Loop. The fast tracks stay on the surface to avoid interfering with high-frequency regional metro service. For historic reasons Thameslink mostly stays as-is, with a combination of fast and stopping services, but the curve toward London Bridge should not be used – instead, passengers should have access to Crossrail 3 plus interchanges to the City at London Bridge and a new infill station at Southwark.

London owes it to itself to understand why its construction costs are so high that instead of solving its transport capacity problems with multiple cross-city tunnels in a decade, it’s taking multiple generations to build out such a system. There’s a lot of ongoing discussion about the last-minute delays and cost overruns on Crossrail, but the absolute costs even before the overrun were very high, the highest in the world outside New York City – and Crossrail 2 is set to break that record by a margin.

Assume Nordic Costs

I wrote a post last year proposing some more subway lines for New York, provided the region could bring down construction costs. The year before, I talked about regional rail. Here are touched-up maps, with costs based on Nordic levels. To avoid cluttering the map in Manhattan, I’m showing subway and regional rail lines separately.

A full-size 52 MB version of the subway map can be found here and a 52 MB version of the regional rail map can be found here.

Subways are set at $110 million per km underground, outside the Manhattan core; in more difficult areas, including underwater they go up to $200-300 million per km, in line with Stockholm Citybanan. Lacking data for els, I set them at $50 million per km, in line with normal subway : el cost ratios. The within-right-of-way parts of Triboro are still set at $20 million per km (errata 5/30: 32 out of 35 km are in a right-of-way and 3 are in a new subway, despite what the map text says, but the costs are still correct).

Overall, the subway map costs $22 billion, and the regional rail one $15 billion, about half as high as the figure I usually quote when asked, which is based on global averages. This excludes the $2 billion for separated intercity rail tracks, which benefit from having no stations save Penn (by the same token, putting the express rather than local lines in the tunnel is a potential cost saving for Crossrail 2). It also excludes small surface projects, such as double-tracking the Northern Branch and West Shore Line, a total of 25 and 30 km respectively, which should be $300-550 million in total, and some junction fixes. There may also be additional infill stations on commuter rail, e.g. at intersection points with new subway extensions; I do not have Nordic costs for them, but in Madrid they cost €9 million each.

The low cost led me to include some lines I would not include elsewhere, and decide marginal cases in favor of subways rather than els. There is probably no need for the tunnel connecting the local tracks of Eighth Avenue and Fulton Street Lines, but at just $1.2 billion, it may be worth it. The line on Northern Boulevard and the Erie Main Line should probably be elevated or in a private right of way the entire way between the Palisades and Paterson, but at an incremental cost of $60 million per km, putting the Secaucus and East Rutherford segments underground can be justified.

In fact, the low cost may justify even further lines into lower-density areas. One or two additional regional rail tunnels may be cost-effective at $300 million per kilometer, separating out branches like Port Washington and Raritan Valley and heading to the airports via new connections. A subway line taking over lanes from the Long Island Expressway may be useful, as might another north-south Manhattan trunk feeding University Avenue (or possibly Third Avenue) in the Bronx and separating out two of the Brighton Line tracks. Even at average costs these lines are absurd unless cars are banned or zoning is abolished, but at low costs they become more interesting.

The Nordic capitals all have extensive urban rail networks for their sizes. So does Madrid: Madrid and Berlin are similar in size and density, but Berlin has 151 km of U-Bahn whereas Madrid has 293 km of metro, and Madrid opened a second Cercanías tunnel in 2008 for around $100 million per km and is planning a third tunnel for next decade (source, PDF-pp. 104-108). Things that are completely ridiculous at American costs – say, any future subway expansion – become more reasonable at average costs; things that are completely ridiculous at average costs likewise become more reasonable at Nordic or Spanish costs.