Paris, World Capital of Expensive Regional Trains

I have found something a European city does worse than the United States in public transit. Paris has just announced its new bilevel design for the RER B, currently the only line running single-deck trains due to restricted clearances. The new double-deckers, dubbed MI 20, are expected to cost 2.56b€ for 146 trainsets, each 104 meters long, for a total of 168,600€ per meter of train length.

I’ve criticized Paris’s use of double-deckers in the past. The cost premium for a double-decker, usually around 25-50%, at best matches the gain in seated capacity, and leads to other capacity problems with access and egress, which are of especial importance on urban rail like the RER. Not for nothing, bilevel trains are not used in Tokyo except for the occasional first-class car (“green car”), which is less crowded by design than the legendarily crowded subway and regional rail cars.

However, this is a lot worse than the usual premium. The only comparably expensive bilevel I can find is the Stadler KISS order for Caltrain, which at $230,000/m for the base order (and only $160,000/m for an equal-size option) comes at a large premium over usual KISSes (both around 130,000€/m) due to client interference and micromanagement coming from low competence by American railroaders.

But the KISS is a high-performance train, at the expensive end in Europe, too. Moreover, it is fully bilevel, whereas the MI 20 has a mix of single- and double-deck cars, with high-platform boarding. Comparable split-level trains go well below 130,000€/m. Canalblog has a compendium of recent Coradias: the single-level example for Milan is 6.25m€ per 84-meter train, or 74,400€/m, and the mixed single- and double-deck examples are 96,700€/m in Luxembourg and 117,600€/m in Germany. The mixed-deck Siemens Desiro HC has a range of costs: its RRX order is 1.7b€ for 82 150-meter trainsets, which is 138,200€/m, but a smaller order for the Berlin RegionalBahn is 300m€ for 21 six-car and 2 four-car trainsets, or 89,600€/m, which is a high but not unheard of cost for a single-decker, let alone a double-decker. The Desiro HC is being delivered to Israel as well, at a cost of 900m€ for 60 trainsets totaling 330 cars, or 109,100€/m.

There’s nothing special about Paris that justifies such a cost – the highest in the world so far, even beating the Americans. Rather, the problem is most likely that Paris thinks it’s special and won’t buy a standard platform. Canalblog points out that the Coradia Duplex formed the basis of the X’Trapolis, currently delivered for the RER D and E – and the X’Trapolis’s first tranche spent 29% of its budget on design and engineering, driving the cost up to a stratospheric 21.83m€ per train of length 112 or 130 meters. Even averaged over the entire order of 255 trainsets, at which point economies of scale kick in and the bespoke design is less harmful, the cost is 121,400€/m, which is 25% more than the more standard Luxembourg design.

Update: Clem Tillier asked me about Madrid’s recent Cercanías order. This is a mix of Stadler trains and Coradias, both mixed single- and double-deck; the Coradias, using the same platform as the Paris X’Trapolis trains and built in Spain rather than in France, cost 1.447b€/152 trainsets, or 95,200€/m, and the Stadlers, mixing KISS and FLIRT technology, cost 998m€ for 24 100-meter trains and 35 200-meter trains, or 106,200€/m.

In a megacity like Paris, it’s tempting to think one is special and must have special equipment. But the resulting high costs are particularly damaging in such a city. The RER B runs every 3 minutes at rush hour, which means that high rolling stock costs are proportionally a bigger problem than on a less frequent system. The cost premium of the order over standard single-deck trains is a factor of around 2; half the cost is 1.3b€, which would be enough to build some necessary tunnel extensions, like quad-tracking the combined two-track tunnel for the RER B and D between Châtelet-Les Halles and Gare du Nord, or if RER investment is not desired then around 6 km of tunnel for Grand Paris Express after the latest cost overruns.

France needs to let go of its pride and recognize that Paris is merely the largest city in the Union, with the same standards and regulations as the other 440 million of us who do not live in Ile-de-France. Vanilla Coradias and Desiros that work elsewhere should also work for the RER, with minor tweaks to take into account high platforms and the loading gauge, both of which the vendors are experienced in dealing with due to common intra-European variation. The people who sign extravagant contracts may feel special about the train design, but the passengers who end up not getting the investment the cost premium would have gone to are going to keep feeling packed on rush hour RERs. The region ought to do better and hire managers who are better than this.

The Other People’s Money Problem

I did a poll on Patreon about cost issues to write about. This is the winning option, with 12 votes; project- vs. budget-driven plans came second with 11 and I will blog about it soon, whereas neighborhood empowerment got 8.

OPM, or other people’s money, is a big impediment to cost reform. In this context, OPM refers to any external infusion of money, typically from a higher-level government from that controlling an agency. Any municipal or otherwise local agency, not able or willing to raise local taxes to fund itself, will look for external grants, for example in a federal budget. The situation then is that the federal grantor gives money but isn’t involved in the design of where the money goes to, leading to high costs.

OPM at ground level

Local and regional advocates love OPM. Whenever they want something, OPM lets them have it without thinking in terms of tradeoffs. Want a new piece of infrastructure, including everything the local community groups want, with labor-intensive methods that also pay the wages the unions hop for? OPM is for you.

This was a big problem for the Green Line Extension’s first iteration. Somerville made ridiculous demands for signature stations and even a bike path (“Somerville Community Path”) thrown in – and all of these weren’t jut extra scope but also especially expensive, since the funding came from elsewhere. The Community Path, a 3 km bike path, was budgeted at $100 million. The common refrain on this is “we don’t care, it’s federally funded.” Once there’s an outside infusion of money, there is no incentive to spend it prudently.

OPM modifying projects

In capital construction, OPM can furthermore lead to worse projects, designed to maximize OPM rather than benefits. Thus, not only are costs high, but also the results are deficient. In my experience talking to New Englanders, this takes the form of trying to vaguely connect to a politician’s set of petty priorities. If a politician wants something, the groups will try pitching a plan that is related to that something as a sales pitch. The system thus encourages advocates and local agencies to invest in buying politicians rather than in providing good service.

This kind of behavior can persist past the petty politician’s shelf life. To argue their cases, advocates sometimes claim that their pet project is a necessary component of the petty politician’s own priority. Then the petty politician leaves and is replaced by another, but by now, the two projects have been wedded in the public discourse, and woe betide any advocate or civil servant who suggests separating them. With a succession of petty politicians, each expressing interest in something else, an entire ecosystem of extras can develop, compromising design at every step while also raising costs.

The issue of efficiency

In the 1960s, the Toronto Transit Commission backed keeping a law requiring it to fund its operations out of fares. The reason was fear of surplus extraction: if it could receive subsidies, workers could use this as an excuse to demand higher wages and employment levels, and thus the subsidy would not go to more service. As it is, by 1971 this was untenable and the TTC started getting subsidies anyway, as rising market wages required it to keep up.

In New York, the outcome of the cycle of more subsidies and less efficiency is clearer. Kyle Kirschling’s thesis points out on PDF-p. 106 that New York City Transit’s predecessors, the IRT and BMT, had higher productivity measured in revenue car-km per employee in the 1930s than the subway has today. The system’s productivity fell from the late 1930s to 1980, and has risen since 1980 but (as of 2010) not yet to the 1930s peak. The city is one of a handful where subway trains have conductors; maintenance productivity is very low as well.

Instead of demanding efficiency, American transit advocates tend to demand even more OPM. Federal funding only goes to capital construction, not operations – but the people who run advocacy organizations today keep calling for federal funding to operations, indifferent to the impact OPM would have on any effort to increase efficiency and make organizations leaner. A well-meaning but harmful bill to break this dam has been proposed in the Senate; it should be withdrawn as soon as possible.

The difference between nudging and planning

I am soon going to go over this in more details, but, in brief, the disconnect between funding and oversight is not a universal feature of state funding of local priorities. In all unitary states we’ve investigated, there is state funding, and in Sweden it’s normal to mix state, county, and municipal funding. In that way, the US is not unique, despite its federal system (which at any case has far more federal involvement in transportation than Canada has).

Where the US is unique is that the Washington political establishment doesn’t really view itself as doing concrete planning. It instead opts for government by nudge. A federal agency makes some metrics, knowing that local and state bodies will game them, creating a competition for who can game the other side better. Active planning is shunned – the idea that the FTA should have engineers who can help design subways for New York is unthinkable. Federal plans for high-speed rail are created by hiring an external consultant to cobble together local demands rather than the publicly-driven top-down planning necessary for rail.

The same political advocates who want more money and care little for technical details also care little for oversight. They say “regulations are needed” or “we’ll come up with standards,” but never point to anything concrete: “money for bus shelter,” “money for subway accessibility,” “money for subway automation,” etc. Instead, in this mentality the role of federal funding is to be an open tab, in which every leakage and every abnormal cost is justified because it employed inherently-moral $80,000/year tradesmen or build something that organized groups of third-generation homeowners in an expensive city want. The politics is the project.

Mixing High- and Low-Speed Trains

I stream on Twitch (almost) every week on Saturdays – the topic starting now is fare systems. Two weeks ago, I streamed about the topic of how to mix high-speed rail and regional rail together, and unfortunately there were technical problems that wrecked the recording and therefore I did not upload the video to YouTube as I usually do. Instead, I’d like to write down how to do this. The most obvious use case for such a blending is the Northeast Corridor, but there are others.

The good news is that good high-speed rail and good legacy rail are complements, rather than competing priorities. They look like competing priorities because, as a matter of national tradition of intercity rail, Japan and France are bad at low-speed rail outside the largest cities (and China is bad even in the largest cities) and Germany is bad at high-speed rail, so it looks like one or the other. But in reality, a strong high-speed rail network means that distinguished nodes with high-speed rail stations become natural points of convergence of the rail network, and those can then be set up as low-speed rail connection nodes.

Where there is more conflict is on two-track lines with demand for both regional and intercity rail. Scheduling trains of different speeds on the same pair of tracks is dicey, but still possible given commitment to integration of schedule, rolling stock, and timetable. The compromises required are smaller than the cost of fully four-tracking a line that does not need so much capacity.

Complementarity

Whenever a high-speed line runs separately from a legacy line, they are complements. This occurs on four-track lines, on lines with separate high-speed tracks running parallel to the legacy route, and at junctions where the legacy lines serve different directions or destinations. In all cases, network effects provide complementarity.

As a toy model, let’s look at Providence Station – but not at the issue of shared track on the Northeast Corridor. Providence has a rail link not just along the Northeast Corridor but also to the northwest, to Woonsocket, with light track sharing with the mainline. Providence-Woonsocket is 25 km, which is well within S-Bahn range in a larger city, but Providence is small enough that this needs to be thought of as scheduled regional rail. A Providence-Woonsocket regional link is stronger in the presence of high-speed rail, because then Woonsocket residents can commute to Boston with a change in Providence, and travel to New York in around 2 hours also with a change in Providence.

More New England examples can be found with Northeast Corridor tie-ins – see this post, with map reproduced below:

The map hides the most important complement: New Haven-Hartford-Springfield is a low-speed intercity line, and the initial implementation of high-speed rail on the Northeast Corridor should leave it as such, with high-speed upgrades later. This is likely also the case for Boston-Springfield – the only reason it might be worthwhile going straight from nothing to high-speed rail is if negotiations with freight track owner CSX get too difficult or if for another reason Massachusetts can’t electrify the tracks at reasonable cost and run fast regional trains.

There’s also complementarity with lines that are parallel to the Northeast Corridor, like the current route east of New Haven, which the route depicted in the map bypasses. This route serves Southeast Connecticut communities like Old Saybrook and can efficiently ferry passengers to New Haven for onward connections.

In all of these cases, there is something special: Woonsocket-Boston is a semireasonable commute, New London connects to the Mohegan Sun casino complex, New Haven-Hartford and Boston-Springfield are strong intercity corridors by themselves, Cape Cod is a weekend getaway destination. That’s fine. Passenger rail is not a commodity – something special almost always comes up.

But in all cases, network effects mean that the intercity line makes the regional lines stronger and vice versa. The relative strength of these two effects varies; in the Northeast, the intercity line is dominant because New York is big and off-mainline destinations like Woonsocket and Mohegan are not. But the complementarity is always there. The upshot is that in an environment with a strong regional low-speed network and not much high-speed rail, like Germany, introducing high-speed rail makes the legacy network stronger; in one that is the opposite, like France, introducing a regional takt converging on a city center TGV station would likewise strengthen the network.

Competition for track space

Blending high- and low-speed rail gets more complicated if they need to use the same tracks. Sometimes, only two tracks are available for trains of mixed speeds.

In that case, there are three ways to reduce conflict:

  1. Shorten the mixed segment
  2. Speed up the slow trains
  3. Slow down the fast trains

Shortening the mixed segment means choosing a route that reduces conflict. Sometimes, the conflict comes pre-shortened: if many lines converge on the same city center approach, then there is a short shared segment, which introduces route planning headaches but not big ones. In other cases, there may be a choice:

  • In Boston, the Franklin Line can enter city center via the Northeast Corridor (locally called Southwest Corridor) or via the Fairmount Line; the choice between the two routes is close based on purely regional considerations, but the presence of high-speed rail tilts it toward Fairmount, to clear space for intercity trains.
  • In New York, there are two routes from New Rochelle to Manhattan. Most commuter trains should use the route intercity trains don’t, which is the Grand Central route; the only commuter trains running on Penn Station Access should be local ones providing service in the Bronx.
  • In the Bay Area, high-speed rail can center from the south via Pacheco Pass or from the east via Altamont Pass. The point made by Clem Tillier and Richard Mlynarik is that Pacheco Pass involves 80 km of track sharing compared with only 42 km for Altamont and therefore it requires more four-tracking at higher cost.

Speeding up the slow trains means investing in speed upgrades for them. This includes electrification where it’s absent: Boston-Providence currently takes 1:10 and could take 0:47 with electrification, high platforms, and 21st-century equipment, which compares with a present-day Amtrak schedule of 0:35 without padding and 0:45 with. Today, mixing 1:10 and 0:35 requires holding trains for an overtake at Attleboro, where four tracks are already present, even though the frequency is worse than hourly. In a high-speed rail future, 0:47 and 0:22 can mix with two overtakes every 15 minutes, since the speed difference is reduced even with the increase in intercity rail speed – and I will defend the 10-year-old timetable in the link.

If overtakes are present, then it’s desirable to decrease the speed difference on shared segments but then increase it during the overtake: ideally the speed difference on an overtake is such that the fast train goes from being just behind the slow train to just ahead of it. If the overtake is a single station, this means holding the slow train. But if the overtake is a short bypass of a slow segment, this means adding stops to the slow train to slow it down even further, to facilitate the overtake.

A good example of this principle is at the New York/Connecticut border, one of the slowest segments of the Northeast Corridor today. A bypass along I-95 is desirable, even at a speed of 200-230 km/h, because the legacy line is too curvy there. This bypass should also function as an overtake between intercity trains and express commuter trains, on a line that today has four tracks and three speed classes (those two and local commuter trains). To facilitate the overtake, the slow trains (that is, the express commuter trains – the locals run on separate track throughout) should be slowed further by being made to make more stops, and thus all Metro-North trains, even the express trains, should stop at Greenwich and perhaps also Port Chester. The choice of these stops is deliberate: Greenwich is one of the busiest stops on the line, especially for reverse-commuters; Port Chester does not have as many jobs nearby but has a historic town center that could see more traffic.

Slowing down the intercity trains is also a possibility. But it should not be seen as the default, only as one of three options. Speed deterioration coming from such blending in a serious problem, and is one reason why the compromises made for California High-Speed Rail are slowing down the trip time from the originally promised 2:40 for Los Angeles-San Francisco to 3:15 according to one of the planners working on the project who spoke to me about it privately.

What’s Frustrating About Bus Redesigns

The transit vlogger Alexandra Rose said something deeply disturbing and yet true on Twitter:

More and more I think that bus network redesigns are too often just managed decline. I was really into zero cost change redesigns a few years ago, less so recently. Literally shuffling deck chairs.

The issue isn’t that net zero cost redesigns are bad. They’re not. The results in recent years look pretty good; Nova Xarxa really did lead to ridership growth, the American redesigns were for the most part helpful too, and I stand by the claim in our report, on pp. 36-37, that the Brooklyn bus redesign we propose would raise ridership 20%.

And yet, Alexandra is completely right, because 20% of zero is still zero. Even in New York, what we call would only get New York bus ridership back to where it was on the eve of the Great Recession and the ensuing service cuts. This really concerns two separate problems of bus service and systemwide changes.

Bus decline management

Managing the decline of buses is inevitable. Buses are too labor-intensive in a developed country to remain a cost-effective solution in the long run – and the sort of cities in the first world with both the best public transit and the best prospects for growth also tend to have the highest wages. Bus drivers in New York earn $85,000/year, and that’s market-rate – there aren’t hordes of unemployed people clamoring to work as bus drivers; agencies that pay significantly less relative to local wages, like San Francisco’s Muni, find it hard to recruit drivers.

How labor-intensive are buses? Well, in New York there are around 12,000 bus drivers and 4,000 subway drivers. Subway ridership was 2.5 times bus ridership in 2019, and overall vehicle-hours, counting cars rather than trains, were 60% longer, with subway cars still substantially bigger than buses. And there’s a lot more inefficiency in crew scheduling on the subway than on the bus network – and today it’s possible to automate subways but not buses. A subway train today carries as many passengers as 17 buses on average, and is 2.5 times faster, for an overall labor efficiency factor of 44.5, without automation; it’s in practice less than this taking crew inefficiencies and maintenance into account, but remains well over a full order of magnitude.

The upshot is that the sort of service-hours that could be run with the wages of 60 years ago stopped being financially sustainable 40 years ago, and the service-hours of 20 years ago are not financially sustainable today. Net zero redesigns are about the best that is possible – because service-hours are expensive and getting more so over time.

All of these bus reforms – network redesigns, dedicated lanes, bus shelter, real-time information, signal priority – push back the decline, but they do not halt it. Eventually, something other than labor-intensive buses will be required, most likely some kind of light rail and subway combination as with the railstitutions happening here in Berlin or in Paris.

Compounding growth factors

A 20% increase in systemwide ridership is great! But, 20% of zero is still zero, and 20% of a low number is in absolute terms low growth. The question is what comes next.

If a city builds a subway line and gets noticeable ridership growth, it can compound. The one subway line succeeded, so now it can built more to new areas, not served by this line. Large increases in systemwide ridership can come from a project that is not systemwide, and then in a large city it’s easy enough to add more such projects. This is not mere linear growth as new lines open – a city that builds a subway system automatically makes city center an attractive place for business, leading to naturally-occurring transit-oriented development. It is natural for public transport advocates to be optimistic in such a situation.

Systemwide improvements compound with everything else, but are frustrating one-time affairs. Yes, a redesign can raise ridership 20% – and then what? Our Brooklyn proposal is aggressive – more so than Nova Xarxa, which included a pre-agreed number of routes with dedicated lanes, I believe 12, but nothing like the proposal we made that every route except for a handful in low-traffic areas on the edge of the borough get two-way dedicated lanes.

The only big thing our proposal didn’t touch on is bus shelter, because we didn’t realize how important it is. But bus shelter interacts negatively with interventions that increase bus frequency, since its effect is to reduce the disutility of waiting for the bus, and if the wait has already been reduced to 5-6 minutes then shelter is useful but its impact is not the 30% increase in ridership that my bus shelter post posits. Other than shelter, there’s conditional signal priority making buses less likely to bunch, but it too interacts negatively with everything else, and the speed benefits (as opposed to the more speculative reliability benefits) of signal priority are known and small.

Is it worthless?

No! Just frustrating. Bus upgrades are a one-time thing, holding back the long-term decline of the mode as better alternatives emerge. These ridership increases are nothing to sneeze at, but there’s no alternative to transitioning to rapid transit, maybe with trams as a feeder layer (or as the primary one if you’re a sub-million metro), with enough transit-oriented development that people can just walk to the subway. Everything else can b fine in the short run, but in a wealthy city that run is short indeed.

We Ran a Conference About Rail Modernization (Again)

Modernizing Rail 2021 just happened. Here’s a recording of the Q&A portion (i.e. most) of the keynote, uploaded to YouTube.

As more people send in materials, I’ll upload more. For now, here are the slides I’ve gotten:

A bunch of us tweeted the talks using the hashtag #ModernRail2021, including some that were not recorded.

Modernizing Rail, and a Note on Gender

Modernizing Rail starts in 15 hours! Please register here, it’s free. The schedule can be viewed here (and the Zoom rooms all have a password that will be given to registered attendees); note that the construction costs talk is not given by me but by Elif Ensari, who for the first time is going to present the Turkish case, the second in our overall project, to the general public. But do not feel obligated to attend, not given what else it’s running against.

I made a video going into the various breakout talks that are happening, in which I devoted a lot of time to the issue of gender. This is because Grecia White didn’t have enough time at last year’s equity session to talk about it, so this time she’s getting a full session, which I have every intention of attending. The video mentions something that fizzled out because of difficulties dealing with US census UI, which is a lot harder to use than the old Factfinder: the issue of gender by commuting. So I’d like to give this more time, since I know Grecia is going to talk about something adjacent but not the same.

The crux is this: public transit ridership skews female, to some extent. US-wide, 55% of public transit trips are by women; an LA-specific report finds that there, women are 54% of bus riders and 51% of rail riders. The American Community Survey’s Means of Transportation to Work by Selected Characteristics Table has men at 53% of the overall workforce but transit commuters splitting 50-50; the difference, pointed out in both links, is that women ride more for non-work trips, often chaining trips for shopping and child care purposes rather than just commuting to work.

In the video, I tried to look at the gender skew in parts of the US where transit riders mostly use commuter rail, like Long Island, and there, the skew goes the other way, around 58-42 for men. In Westchester, it is 54-46. In New York, which I struggled to find data for in the video, the split is 52-48 female – more women than men get to work on public transit.

But an even better source is the Sex of Workers by Means of Transportation to Work table, which (unlike Selected Characteristics) details commute mode choice not just as car vs. transit but specifies which mode of public transit is taken. There are, as of the 2019 ACS, 3,898,132 male transit commuters and 3,880,312 female ones – that’s the 50-50 split above. But among commuter rail riders, the split is 533,556 men, 387,835 women, which is 58-42. Subways split about 50-50, buses skew 52-48 female.

In the video, I explain this referencing Mad Men. Commuter rail is stuck in that era, having shed all other potential riders with derogatory references to the subway; it’s for 9-to-5 suburban workers commuting to the city, and this is a lifestyle that is specifically gendered, with the man commuting to the city and the woman staying in the suburb.

This impacts advocacy as well. In planning meetings for the conference, we were looking for more diverse presenters, but ran into the problem that in the US, women and minorities abound in public transit advocacy but not really in mainline rail, which remains more white and male. I believe that this is true of the workforce as well, but the only statistics I remember are about race (New York subway and bus drivers are by a large majority nonwhite, commuter rail drivers and conductors are the opposite), and not gender. Of course there are women in the field – Adina Levin (who presented last year) and Elizabeth Alexis are two must-read names for understanding what goes on both in general and in the Bay Area in particular – but it’s unfortunately not as deep a bench as for non-mainline transit, from which it is siloed, which has too many activists to list them all.

Do Costs Run Over or Are They Underestimated?

The literature on cost overruns for infrastructure projects is rich, much more so than that for absolute costs. The best-known name in this literature is Bent Flyvbjerg, who in the early 2000s collated a number of datasets from the 1980s and 90s to produce a large enough N for analysis, demonstrating consistent, large cost overruns, especially for urban rail. Subsequently, he’s written papers on the topic, focusing on underestimation and on how agencies can prospectively estimate costs better and give accurate numbers to the public for approval. This parallels an internal trend in the US, where Don Pickrell identified cost overruns in 1990 already, using 1980s data; Pickrell’s dataset was among those analyzed by Flyvbjerg, and subsequent to Pickrell’s paper, American cost overruns decreased to an average of zero for light rail lines.

But a fundamental question remains: are cost overruns really a matter of underestimation, or a true overrun? In other words, if a project, say Grand Paris Express, is estimated to cost 22.6 billion € in 2012 (p. 7) and is up to 35.6 billion € today (p. 13), does it mean the cost was 35.6b€ all along and the 2012 analysis just failed to estimate it right? Or dos it mean the cost was 22.6b€ then, and then the budget ran over due to failures of planning that could have been avoided?

Transit agencies that just want to avoid the embarrassment of media headlines saying “they said it costs X but it costs 2X” care mostly about underestimation. This is also true of both generic project managers and political appointees, two groups that do not care about the details of how to build a subway, and think of everything in abstract terms in which a subway might as well be a box of shampoo bottles.

However, the concrete examples that I have seen or heard of for cost overruns look like overruns rather than underestimation. That is, those projects could have been done at the original cost, but planning mistakes drove the budget up, or otherwise created conditions that would enable other forces to drive the budget up.

The Netherlands: early commitment

Bert van Wee is among the world’s top researchers on cost overruns, even if he’s less well-known to the public than Flyvbjerg. He spoke to me about the problems of early commitment in Dutch planning, in which politicians commit to a project before design is finalized. Once the political decision has been made, it is easy for actors to extract surplus, because the state or city cannot walk away easily, while a 20% cost overrun is much easier to explain to the public. This problem plagued 2000s investments like HSL Zuid. To deter this, after 2009 the Netherlands passed reforms that attempt to tackle this problem, aiming to defer the formal political decision to later in the process.

This factor seems to correlate with absolute costs, if not with overruns. American planning is extremely politicized; Canadian planning is fairly politicized too, with individual subway projects identifiable as the brainchildren of specific politicians or parties; Southern European and Nordic planning is highly bureaucratized, with design driven by the civil service and politicians making yes or no decisions late in the process.

Sweden: changes in rules

According to a senior planner at Nya Tunnelbanan, the project has run over from 22.506 billion kronor in 2013 to 31.813 today, both in 2016 price levels; in US dollars, this is $2.551b/19.6 km to $3.606b/19.6 km, all underground. The reasons for the escalation come largely from tighter regulations as well as litigation:

  • Safety requirements have been tightened midway through the project, requiring a service tunnel in addition to the two track tunnels, raising excavation volume almost 50%
  • An environmental court ruling slowed down excavation further
  • Consensus with stakeholders took longer than expected
  • Excavated rock was reclassified midway through the project from useful building material to waste that must be disposed of

Focusing on underestimation is not really germane to what’s happened in Stockholm. The problem isn’t that early 2010s engineers failed to anticipate regulations that were not in force at the time. It’s that regulations were changed later. The rock removal process today actually increases greenhouse gas emissions, just because of the need to freight it away, let alone the systemwide effects on climate of making it harder to build subways.

California: scope creep and change orders

California High-Speed Rail is such a big project that its cost overruns, in multiple stages, were amply discussed in the media. The original announcements in the early 2010s, for example here, were largely about scope creep. At-grade segments turned into viaducts; above-ground segments, particularly in the Bay Area, were turned into tunnels. The reasons were mostly about agency turf battles.

Only in one case was the problem more about underestimation than overrun: the Central Valley segment had originally been planned to follow railroad rights-of-way, but had to be redesigned to have more viaducts and swerve around unserved small towns. This was bad planning, at two points: first, the original designs assumed trains could go at 350 km/h through unserved towns, which they don’t anywhere; and second, once the redesign happened, it was so rushed that land acquisition was time-consuming and acrimonious. Even then, much of the overdesign as identified by a Deutsche Bahn postmortem could have been prevented.

The second stage is more recent: the Central Valley construction contracts have long busted their budgets due to change orders. Change orders are a common problem in California, and in this case, it involved not only the change order king Tutor-Perini, but also the usually reasonable Dragados. The situation here must be ascribed to overrun rather than underestimation: a transparent process for handling changes, based on itemized costs, is an emerging best practice, known since the early 2000s to people who cared to know, and more recently seen in the economics literature for general infrastructure. That California failed to follow this practice – which, again, was available already in the late 2000s – is the source of malpractice. The original bids could have held if the process were better.

Absolute costs and cost overruns

Cost overruns are not the same as absolute costs. They are not even obviously correlated: witness the way the US eliminated most overruns on surface light rail projects in the 1990s and 2000s, to the point that projects with large overruns like the Green Line Extension are exceptional, while absolute costs have skyrocketed. But if we understand the problem to be about cost overruns from an ambitious but achievable budget rather than about underestimating a final cost that could not be improved on, then the study of the two topics is inherently intertwined.

Problems that recur in postmortems of cost overruns are not just about estimation. They’re about building better and cheaper. A bureaucratized planning process in which politicians retain the right to make yes-or-no decisions on complete design reduces cost overruns by reducing leakage and surplus extraction; the overruns such a process prevents are preventable extra costs, rather than higher initial estimates. The same is true of avoiding overbuilding, of not introducing extraneous regulations, of treating environmental questions as systemic and quantitative rather than as local under a do-no-harm principle. Even the question of change orders is more transparently about reducing absolute costs in the literature, since the overruns prevented tend to be seen in higher risk to the contractor leading to higher profit margin demands.

The upshot is that this makes the study of absolute costs easier, because we can reuse some of the literature for the related problem of cost overruns. But conceptually, it means that agencies need to be more proactive and treat early budgets as standards to be adhered to, rather than just blow up the budgets preemptively so that it’s easier to stick to them.

What Berlin Should Be Building

A week and a half ago, I crayoned Berlin U- and S-Bahn expansion on video. With some tweaks, here is the final product:

Solid lines exist, dashed ones are under construction or (as far as I can tell) done deals, dotted ones are things that I am proposing or that are officially maybes; a few station names change to dissimilate different U- and S-Bahn stations and coname colocated stations

Here is the full-size version. (I know I’ve been asked to provide lighter JPGs, but my attempt at JPG compression turned 86 MB to 37 MB, hardly a coup de grâce.)

This is based on ongoing U-Bahn expansion plans plus the 2030 S-Bahn plan.

S-Bahn

The most significant variation is that the dashed S-Bahn line from Gesundbrunnen to Hauptbahnhof and Potsdamer Platz, dubbed S21, is turned into a northwest-southeast trunk line in my plan, following a proposal by Felix Thoma in Zukunft Mobilität. The plan for S21 today is to stay north-south and link with Südkreuz and Schöneberg, beefing up frequency on the north-south S-Bahn.

I believe my routing to be superior, due to traffic on the Görlitzer Bahn, seen below (source, p. 6):

Currently, peak traffic on both the Stadtbahn and the North-South Tunnel is 18 trains per hour in each direction. This is low; Munich achieves 30 tph with very short signal blocks and more branching than Berlin has, splitting into seven branches on each direction rather than three or four. 30 is a limit value, but 24 is more common, and would substantially simplify operations.

The North-South Tunnel splits into a western branch, currently carrying S1 via Schöneberg to Wannsee every 10 minutes, and an eastern, carrying S2/S25/S26 via Südkreuz every 10/20/20 minutes; since the two branches have roughly equal ridership, each should run every 5 minutes, unlike today, where only Südkreuz gets such service. To the north, each of the two main branches can run every 5 minutes as well.

The Stadtbahn is asymmetric. Only 12 out of 18 tph continue west of Westkreuz: Spandau and Potsdam get 10-minute service, and in addition S5, turning at Westkreuz, runs every 10 minutes. As such, all growth in traffic on the western branches should be encouraged. This is thankfully already done, with expansion plans west of Spandau. To the east, traffic is the most overloaded, and will remain so even with the opening of the U5 extension last year. Going up from 18 to 24 maximum tph means 10-minute service on each of the four branches – S3 to Erkner, S5 to Strausberg-Nord, S7 to Ahrensfelde, S75 to Wartenberg (proposed to be extended into a loop going northwest). Today, S3 runs every 20 minutes, and S75 doesn’t run through but rather only runs from Warschauer Strasse east, and conversely, S9 curves from the Stadtbahn to the Görlitzer Bahn to the airport.

Rerouting S21 to connect to the Görlitzer Bahn means that trunk, currently carrying 18 trains per hour, can all run through to city center, and then either go to the Siemensbahn or loop from Hauptbahnhof to Gesundbrunnen. Such service also removes reverse-branching from the rest of the system, allowing all services to run more regularly and reliably since each of the four trunks, including the Ring, would run independently of the others, and delays wouldn’t propagate.

U-Bahn

U-Bahn expansion in Berlin is mostly mothballed. The city prefers trams, even where they are inappropriate due to low speed over long stretches or forced transfers. Plans for U-Bahn expansion to Märkisches Viertel are uncertain, unfortunately. Plans for expansion to Tegel along a branch of U6 look dead, hence my resurrection of an older unbranched U5 extension; the current plan is to connect the Urban Tech Republic complex with the rest of the city via tram. Trams are cheaper but you get what you pay for; the ideal use of a tram is for cross-city routes, not primary routes to the center.

Hence various extensions that I think should be built. U7 to the airport looks like a done deal, and U7 to Staaken is favorable too, as is the low-cost, low-ridership one-stop extension of U3 to Mexikoplatz. U9 to Pankow and U2 to Pankow-Kirche are much-discussed, as is U8 to Märkisches Viertel, whose current cost/rider projection is favorable by international standards.

My additions are U1 extensions at both ends, the U5 extension to Tegel and then looping to intersect U6 and U8 in Reinickendorf, and the resurrection of the U10 plan as a U3 link (and not as a line to Steglitz, which gets extra S-Bahn service either way). The U1 extension to the west is forced to use cut-and-cover since the U1 tunnel under Kurfürstendamm is 1900s cut-and-cover, which is disruptive but cheaper than bored tunnel. The other two lines are long-term desires of the city and have been safeguarded for decades, with intersecting stations built to accommodate them.

Whether lines run in this configuration or another is up for debate. At Wittenbergplatz it’s easiest to link the new U10 system to U1 to Uhlandstrasse and then connect U3 to Krumme Linke with the existing Warschauer Strasse terminus. This would be an awkward system of U1, U2, and U3 in which the line going farthest north going east also goes farthest north going west and the line going farthest south to the east goes farthest south also to the west. If there’s a way to flip the situation, pairing U10 with present-day U1-west, U2-east with U3-west, and U3-east with U2-west, it should be done; this system in general has undergone many such changes over the generations.

Quick Note: Do Costs Ever Go Down?

Bad agencies have a ratchet process in costs: they can go up, but not down. If there’s a cost saving, it does not reduce the budget, but only cancels out with unspecified cost increases. Agency heads and politicians trumpet their value engineering while costs never go down, leading to premium-cost, substandard quality projects.

Case in point: the Baltimore and Potomac Tunnel replacement project. The project used to be $750 million, in the 2000s, as a two-track passenger rail tunnel. Over the next decade, this turned into a four-track system with mechanical ventilation for diesel freight trains and enough clearance for double-stacked freight; costs ran over to $4 billion. Well, two months ago Amtrak announced a scope reduction back to two tracks, which it claims would save a billion dollars, cutting cost to… $4 billion.

This is not the first time this happens. Value engineering in California has had the same effect: every attempt to reduce scope – the blended plan for Northern California, plus various design compromises in both the Bay Area and the Central Valley – has failed to reduce costs. At most, they’ve prevented further cost overruns.

And in New York, the removal of the cavern underneath Penn Station in the planning process between the canceled ARC tunnel and the Gateway tunnel did not reduce costs at all. The cost estimate was $10 billion, much of which was the cavern; the cost estimate now is $10 billion for the bare tunnel with less scope than before. ARC was canceled on the grounds of potential cost overruns, and yet as soon as it took over the project, even while descoping the cavern, Amtrak presided over further increases in costs due to extras (Penn South, etc.).

It’s as if once there’s a number circulating out there, it will be spent, no matter what. If there’s a surplus, it will be blown on unspecified extras or on sheer inefficiency. Why spend $3 billion when the political system has already indicated that $4 billion is okay? Thus, 4-1 = 4, and, no doubt, if further value engineering is identified, the cost will stay $4 billion.

At no point does anyone say, okay, if there’s a cost saving, here’s the next slate of projects that the money can be spent on. Nor is there any proactive value engineering. Costs are only a problem insofar as they prevent the political system from saying yes, but even then, if there’s a number out there, even an outlandish one that nobody will say yes to (such as $117 billion for medium-speed rail on the Northeast Corridor), then it is the number. Any cuts from that are against inherently moral workers, communities, etc., in the service of inherently immoral outsiders and experts.

Quick Note: Waste and Missed Projects

If the state spends money on a bad infrastructure project, or too much money beyond what was necessary for the project, then this is waste of money, and should be avoided. But the opposite situation can occur too: some worthwhile projects are not pursued, and that too is a waste, because society forgoes the benefits coming from such projects. This situation should be avoided equally. Moreover, there is no priority between those two types of error. Planning should treat them symmetrically and aim on balance to avoid both equally.

The reason is that just as infrastructure projects are generally not critical, the money that is spent on them is not critical. The US is spending around $1.5 trillion over the lifetime of the program on the F-35 plane, and the money is buried deep in a defense budget that by the standards I grew up with isn’t even that large – and that program consists of documented waste and suffers from poor planning, including serious cost overruns and delays. None of this is an existential threat; the problems the F-35 is intended to solve are not existential but neither are its costs, and likewise, neither infrastructure problems such as delays, capacity limitations, and congestion nor the costs of the projects that intend to fix them are existential.

And if none of this is existential, then the decision of whether to build is about comparing two finite, bounded quantities: costs and benefits. This is why one does a benefit-cost analysis and respect its conclusions, without spiking. But this is also why the state should not systematically aim to err in one direction. If a project with a BCR of less than 1 is built then there is waste, but if a project with a high BCR is not built then there is waste as well.

Note that this principle of not biasing one’s error in one direction (typically the bias is toward inaction) is separate from the question of what the best estimates for costs and benefits are. There is a real tendency to underestimate costs, which is why the minimum BCR that should be funded is not 1 but slightly more, the typical range in Europe being 1.2-1.4. But subject to that limit, decisions should still be symmetric, i.e. if the limit is 1.4, then building 0.7 is symmetrically bad with failing to build 2.8. Alternatively, some projects, like high-speed rail, have upfront costs and long-term benefits, and so it’s better to think of them in terms of financial and social returns on investment, as is done in France (source, pp. 11-12), rather than a BCR in which the discount rate is hidden in a box. But ROI analysis should still be symmetric around one’s chosen limit.

This becomes relevant especially for projects that can expect benefits to rise over time due to economic growth. It is tempting to have a bias toward inaction and only build something once its benefits are unimpeachable, a large multiple of the cost. But this means that in the interim, society has forgone the smaller-but-still-real benefits. Worse, when the BCR grows too large, surplus extraction might pull it back down through an increase in costs, and thus building later can be very risky.

In essence, what this means is that if there’s infrastructure out there with a very high BCR or ROI – and if you ask me, preliminarily, Northeast Corridor high-speed rail done right has a purely financial ROI of maybe 13% – then something is deeply wrong. There shouldn’t be 13% returns out of anything. If there is one, the first question to ask is “why was this not built 50 years ago?”.

In the opposite direction, what looks like building infrastructure prematurely is in fact the prudent decision. South Korea and Taiwan both opened high-speed rail in the 2000s, both underperforming initial expectations. But both have seen steady growth in ridership; at this point, Taiwan HSR returns 4% without social benefits, which is decently healthy, and KTX has somewhat higher ridership than THSR on only slightly higher total construction costs. In the mid-2000s the projects looked like white elephants, that is they were doing just better than minimum. But the 15 years of benefits since then have been considerable. The 20% of society least interested in paying for things should not have veto power; economics exists on the margin and politics on the median.