No New Washington Union Station, Please

A new presentation dropped for Amtrak’s plans to rebuild Union Station. It is mostly pictorial, but even the pictures suggest that this is a very low-value project, one with little to no transportation value and limited development value. The price tag is now $10 billion (it was $7 billion 10 years ago; the increase is somewhat more than cumulative inflation), but even if two zeros are cut from the budget it’s not necessarily worth it.

What are the features of good train stations?

A train station is interface between passengers and trains. Everything about their construction must serve this purpose. This includes the following features:

  • Platforms that can effectively connect to the trains (Union Station has a mix of high and low platforms; all platforms used by Northeast Corridor trains must be raised).
  • Minimum distance from platform to street or to urban transit.
  • Some concessions and seats for travelers, all in an open area.
  • Ticketing machines.
  • An information booth with maps of the area and station facilities.
  • Nothing more.

In particular, lavish waiting halls not only waste of money but also often have negative transport value, as they either force passenger to walk longer between street and platform or steer them to take an option that involves a longer walk; the new Moynihan Train Hall in New York is an example of the latter failure. Berlin Hauptbahnhof, a rare example of a major urban station built recently in a rich country, has extensive shopping, but it’s all designed around fast street-platform and S-Bahn-intercity connections.

What are the features of Washington Union Station expansion?

The presentation highlights the following features:

  • A new concourse beneath the platforms.
  • A new concourse on H Street with a prominent headhouse, with bus and streetcar connections.
  • An enclosed bus facility.
  • Underground parking.
  • Future air rights development.

All of the above are wasteful. Connections to H Street can be handled through direct egress points from the platforms to the street, and passengers can get between H Street and the main historic station via those egress points and the platforms themselves. The platforms are key circulation spaces at a train station and using them for passenger movements is normal; I can see an argument against that if the platforms are unusually narrow or crowded, as is the case in New York, but in Washington there is no such excuse.

Nor is Union Station a major node for city buses. Washington’s surface transit network serves the station, but it’s not a major bus node – only a handful of buses terminate there and they don’t run frequently – and even if it were, a surface bus loop akin to what Ostbahnhof has in Berlin would have sufficed. Thus, the bus infrastructure should be descoped, and buses should keep using the streets.

So, none of the transit connections have any value. Parking, moreover, has negative value, as it encourages access to the area by car, displacing transit trips. Union Station already has a Metro connection as well as some surface transit. Better rail operations would also improve commuter rail access for intercity rail riders. Unfortunately, the plan does not improve those operations, nor is there any plan for much needed capital investment to go alongside better mainline rail operations, such as Virginia electrification and high platforms.

What about the air rights?

They are a poor use of money. Building towers on top of active railyards is more difficult and more expensive than building them on firma. Hudson Yards projects in New York came in at around $12,000 per square meter in hard costs, twice the cost of Manhattan skyscrapers on firma except those associated with the World Trade Center, which were unusually costly.

Nor is the location just north of the historic Union Station so desirable that developers would voluntarily pay the railyard premium to be there. The commercial center of Washington is well to the west of the site, comprising Metro Center and Farragut. More office towers around Union Station would be nice for rebalancing and for generating demand for future mainline rail improvements, but the place for them is on firma around the existing station and not on top of the approach tracks.

What should be done?

The plan should be rejected in its entirety and no further funding should be committed to it. Good transit activists should demand that spending on public transportation and intercity rail go to those purposes and not toward building unnecessary train halls. Moreover, it is unlikely the managers at Amtrak who pushed for it and who still are the client for the project understand modern rail operations, nor is it likely that they will ever learn. With neither need nor use for the project, it should be canceled and the people involved in its management and supervision laid off.

How Many Tracks Do Train Stations Need?

A brief discussion on Reddit about my post criticizing Penn Station expansion plans led me to write a very long comment, which I’d like to hoist to a full post explaining how big an urban train station needs to be to serve regional and intercity rail traffic. The main principles are,

  • Good operations can substitute for station size, and it’s always cheaper to get the system to be more reliable than to build more tracks in city center.
  • Through-running reduces the required station footprint, and this is one of the reasons it is popular for urban commuter rail systems.
  • The simpler and more local the system is, the fewer tracks are needed: an urban commuter rail system running on captive tracks with no sharing tracks with other traffic and with limited branching an get away with smaller stations than an intercity rail station featuring trains from hundreds of kilometers away in any direction.

The formula for minimum headways

On subways, where usually the rush hour crunches are the worst, trains in large cities run extremely frequently, brushing up against the physical limitation of the tracks. The limit is dictated by the brick wall rule, which states that the signal system must at any point assume that the train ahead can turn into a brick wall and stop moving and the current train must be able to brake in time before it reaches it. Cars, for that matter, follow the same rule, but their emergency braking rate is much faster, so on a freeway they can follow two seconds apart. A metro train in theory could do the same with headways of 15 seconds, but in practice there are stations on the tracks and dealing with them requires a different formula.

With metro-style stations, without extra tracks, the governing formula is,

\mbox{headway } = \mbox{stopping time } + \mbox{dwell time } + \mbox{platform clearing time }

Platform clearing time is how long it takes the train to clear its own length; the idea of the formula is that per the brick wall rule, the train we’re on needs to begin braking to enter the next station only after the train ahead of ours has cleared the station.

But all of this is in theory. In practice, there are uncertainties. The uncertainties are almost never in the stopping or platform clearing time, and even the dwell time is controllable. Rather, the schedule itself is uncertain: our train can be a minute late, which for our purpose as passengers may be unimportant, but for the scheduler and dispatcher on a congested line means that all the trains behind ours have to also be delayed by a minute.

What this means that more space is required between train slots to make schedules recoverable. Moreover, the more complex the line’s operations are, the more space is needed. On a metro train running on captive tracks, if all trains are delayed by a minute, it’s really not a big deal even to the control tower; all the trains substitute for one another, so the recovery can be done at the terminal. On a mainline train running on a national network in which our segment can host trains to Budapest, Vienna, Prague, Leipzig, Munich, Zurich, Stuttgart, Frankfurt, and Paris, trains cannot substitute for one another – and, moreover, a train can be easily delayed 15 minutes and need a later slot. Empty-looking space in the track timetable is unavoidable – if the schedule can’t survive contact with the passengers, it’s not a schedule but crayon.

How to improve operations

In one word: reliability.

In two words: more reliability.

Because the main limit to rail frequency on congested track comes from the variation in the schedule, the best way to increase capacity is to reduce the variation in the schedule. This, in turn, has two aspects: reducing the likelihood of a delay, and reducing the ability of a delay to propagate.

Reducing delays

The central insight about delays is that they may occur anywhere on the line, roughly in proportion to either trip time or ridership. This means that on a branched mainline railway network, delays almost never originate at the city center train station or its approaches, not because that part of the system is uniquely reliable, but because the train might spend five minutes there out of a one-hour trip. The upshot is that to make a congested central segment more reliable, it is necessary to invest in reliability on the entire network, most of which consists of branch segments that by themselves do not have capacity crunches.

The biggest required investments for this are electrification and level boarding. Both have many benefits other than schedule reliability, and are underrated in Europe and even more underrated in the United States.

Electrification is the subject of a TransitMatters report from last year. As far as reliability is concerned, the LIRR and Metro-North’s diesel locomotives average about 20 times the mechanical failure rate of electric multiple units (source, PDF-pp. 36 and 151). It is bad enough that Germany is keeping some outer regional rail branches in the exurbs of Berlin and Munich unwired; that New York has not fully electrified is unconscionable.

Level boarding is comparable in its importance. It not only reduces dwell time, but also reduces variability in dwell time. With about a meter of vertical gap between platform and train floor, Mansfield has four-minute rush hour dwell times; this is the busiest suburban Boston commuter rail station at rush hour, but it’s still just about 2,000 weekday boardings, whereas RER and S-Bahn stations with 10 time the traffic hold to a 30-second standard. This also interacts positively with accessibility: it permits passengers in wheelchairs to board unaided, which both improves accessibility and ensures that a wheelchair user doesn’t delay the entire train by a minute. It is fortunate that the LIRR and (with one peripheral exception) Metro-North are entirely high-platform, and unfortunate that New Jersey Transit is not.

Reducing delay propagation

Even with reliable mechanical and civil engineering, delays are inevitable. The real innovations in Switzerland giving it Europe’s most reliable and highest-use railway network are not about preventing delays from happening (it is fully electrified but a laggard on level boarding). They’re about ensuring delays do not propagate across the network. This is especially notable as the network relies on timed connections and overtakes, both of which require schedule discipline. Achieving such discipline requires the following operations and capital treatments:

  • Uniform timetable padding of about 7%, applied throughout the line roughly on a one minute in 15 basis.
  • Clear, non-discriminatory rules about train priority, including a rule that a train that’s more than 30 minutes loses all priority and may not delay other trains at junctions or on shared tracks.
  • A rigid clockface schedule or Takt, where the problem sections (overtakes, meets, etc.) are predictable and can receive investment. With the Takt system, even urban commuter lines can be left partly single-track, as long as the timetable is such that trains in opposite directions meet away from the bottleneck.
  • Data-oriented planning that focuses on tracing the sources of major delays and feeding the information to capital planning so that problem sections can, again, receive capital investment.
  • Especial concern for railway junctions, which are to be grade-separated or consistently scheduled around. In sensitive cases where traffic is heavy and grade separation is too expensive, Switzerland builds pocket tracks at-grade, so that a late train can wait for a slot without delaying cross-traffic.

So, how big do train stations need to be?

A multi-station urban commuter rail trunk can get away with metro-style operations, with a single station track per approach track. However, the limiting factor to capacity will be station dwell times. In cases with an unusually busy city center station, or on a highly-interlinked regional or intercity network, this may force compromises on capacity.

In contrast, with good operations, a train station with through-running should never need more than two station tracks per approach track. Moreover, the two station tracks that each approach track splits into should serve the same platform, so that if there is an unplanned rescheduling of the train, passengers should be able to use the usual platform at least. Berlin Hauptbahnhof’s deep tracks are organized this way, and so is the under-construction Stuttgart 21.

Why two? First, because it is the maximum number that can serve the same platform; if they serve different platforms, it may require lengthening dwell times during unscheduled diversions to deal with passenger confusion. And second, because every additional platform track permits, in theory, an increase in the dwell time equal to the minimum headway. The minimum headway in practice is going to be about 120 seconds; at rush hour Paris pushes 32 trains per hour on the shared RER B and D trunk, which is not quite mainline but is extensively branched, but the reliability is legendarily poor. With a two-minute headway, the two-platform track system permits a straightforward 2.5-minute dwell time, which is more than any regional railway needs; the Zurich S-Bahn has 60-second dwells at Hauptbahnhof, and the Paris RER’s single-level trains keep to about 60 seconds at rush hour in city center as well.

All of this is more complicated at a terminal. In theory the required number of tracks is the minimum turn time divided by the headway, but in practice the turn time has a variance. Tokyo has been able to push station footprint to a minimum, with two tracks at Tokyo Station on the Chuo Line (with 28 peak trains per hour) and, before the through-line opened, four tracks on the Tokaido Main Line (with 24). But elsewhere the results are less optimistic; Paris is limited to 16-18 trains per hour at the four-track RER E terminal at Saint-Lazare.

At Paris’s levels of efficiency, which are well below global best practices, an unexpanded Penn Station without through-running would still need two permanent tracks for Amtrak, leaving 19 tracks for commuter traffic. With the Gateway tunnel built, there would be four two-track approaches, two from each direction. The approaches that share tracks with Amtrak (North River Tunnels, southern pair of East River Tunnels) would get four tracks each, enough to terminate around 18 trains per hour at rush hour, and the approaches that don’t would get five, enough for maybe 20 or 22. The worst bottleneck in the system, the New Jersey approach, would be improved from today’s 21 trains per hour to 38-40.

A Penn Station with through-running does not have the 38-40 trains per hour limit. Rather, the approach tracks would become the primary bottleneck, and it would take an expansion to eight approach tracks on each side for the station itself to be at all a limit.

Watch Our Webinar on Construction Costs Tomorrow

The Italy case, done by Marco Chitti, is up on the website. I encourage people to read the entire report on how Italy has set things up in the last 20-30 years so as to have one of the lowest-cost urban rail infrastructure programs in the world. The Turkey case, by Elif Ensari, will be up shortly.

This is leading to a webinar, to be done tomorrow at 16:00 my time, 10:00 New York time, in which Marco and Elif will present their cases to the general public. I encourage people to register; you’ll be able to ask us questions and we’ll answer in chat or on video. But if you can’t make it, it will be recorded.

Intercity Rail Frequency and the Perils of Market Segmentation

SNCF loves market segmentation. Run by airline execs, the company loves to create different trains for different classes of people. Not only do individual trains have opaque pricing run on the basis of yield management, in which similar seats on the same train at the same time of day and day of week may have different fares, but also there are separately-branded trains for separate fare classes, the higher-fare InOui and the lower-fare OuiGo. On international trains, SNCF takes it to the limit and thus Eurostar and Thalys charge premium fares (both about twice as high as domestic TGVs per passenger-km) and don’t through-ticket with domestic TGVs. This has gotten so bad that in Belgium, some advocates have proposed a lower-priced service on the legacy Paris-Brussels line, which would have to be subsidized owing to the high cost of low-speed intercity rail service.

But why is market segmentation on rail so bad? The answer has to do with frequency and cost structures that differ from those of airlines. Both ensure that the deadweight loss from market segmentation exceed any gains that could be made from extracting consumer surplus.

The issue of frequency

A segmented market like that of domestic TGVs reduces frequency on each segment. To maintain segmentation, SNCF has to make the segments as difficult to substitute for each other as possible. OuiGo serves Marne-la-Vallée instead of Gare de Lyon and forcing passengers onto a 20-minute RER connection, or even longer if they’re arriving in Paris and the wave of 1,000 TGV riders creates long lines at the ticketing machines; on other LGVs it serves the traditional Parisian station and thus the segments are more substitutable.

The situation of Eurostar and Thalys reduces frequency as well: the high fares discourage ridership and send much of it to intercity buses or suppress travel. Fewer riders, or fewer riders per segment as in the case of domestic TGVs, lead to fewer trains. What’s the impact of this on ridership?

The literature on high-speed rail ridership elasticities has some frequency estimates. In Couto’s thesis (PDF-p. 225), it is stated that passenger rail ridership has an elasticity of 0.53 with respective to overall service provision. There are also multiple papers estimating the elasticity with respect to travel time: in Cascetta-Coppola the elasticity ranges from -1.6 to -2.2, in Börjesson it is -1.12, and in a Civity report it is stated based on other work that it is -0.8 to -2. The lowest values in Börjesson are associated with the premium-fare AVE, while the range for the original TGV, priced at the same level as the slower trains it replaced, is -1.3 to -1.6. The upshot is that halving frequency through market segmentation reduces ridership by a factor of 2^0.53 = 1.44, which is far more than the benefit yield management is claimed to have, which is a 4% increase in revenue per SNCF’s American proposals from 2009.

Why are trains different?

Planes and buses happily use yield management. High-speed trains do not, except for those run by SNCF or RENFE – and ridership in France isn’t really higher than in fixed-fare Northern Europe or East Asia while ridership in Spain is much lower. Why the difference?

The reason has to do with the ratio of waiting time to trip time. Thalys connects Paris and Brussels in 1.5 hours, every half hour at rush hour and every 2 hours midday. At rush hour, frequency is sort of noticeable; off-peak, it dominates travel time. This is nothing like planes – even short-distance trips involve hours of access, waiting, and egress time, and therefore trips are not usually spontaneous, and day trips are rare except for business travelers.

Buses, finally, are so small that a market like New York-Philadelphia supports multiple competitors each running frequently, and passenger behavior is such that different companies are substitutable, so that the effective frequency is multiple buses per hour.

Cost structure and bad incentives

It’s typical to price high-speed rail higher than legacy rail, even when otherwise there is no yield management. This is bad practice. The operating costs of high-speed rail are lower than those of slow trains. The crew is paid per hour; electricity costs are in theory higher at higher speed but in practice greenfield high-speed lines are constant 300 km/h cruises whereas legacy lines have many acceleration and deceleration cycles; high-speed trainsets cost much more than conventional ones (by a factor of about 2 in Europe) but also depreciate by the hour and not by the km and therefore are somewhat cheaper per seat-km.

This is comparable to the bad practice, common in the United States and in developing and newly-industrialized countries, of pricing urban rail higher than a bus. The metro is nicer for consumers than a bus, but it also has far lower operating costs and therefore a wise transit agency will avoid incentivizing passengers to take buses and instead use integrated fares. The same is true for slow and fast trains: the solution proposed by the Belgian advocates is to incentivize passengers to take a high-cost, low-price train over a low-cost, high-price one, and therefore is no solution at all.

Moreover, the cost structure of trains is different from that of planes. Planes don’t pay much for fixed infrastructure; in effect, every plane trip costs money, and then the challenge is to fill all the seats. High-speed railways instead pay a lot for infrastructure, while their above-the-rails costs are a few cents per passenger-km (€0.06/seat-km on the TGV, including trainset costs and a lot of labor inefficiency). Their challenge is how to fill the tracks with trains, not how to fill the trains with passengers. This is why the fixed clockface frequency common in Germany, Switzerland, Austria, and the Netherlands is so powerful: the off-peak trains are less full, but that’s fine, as the marginal operating cost of an off-peak train is low.

Just lower the fares

Bear in mind that frequency is not exogenous – it is set based on demand. This means that anything that affects ridership has its impact magnified by the frequency-ridership spiral. An exogenous shock, such as improvement in trip time or fare reduction, is magnified through the spiral, by a factor of 1/(1-0.53) = 2.13. In other words, every elasticity estimated in isolation must be multiplied by a factor of about 2.

And once this is understood, suddenly the optima for service look very different from what Thalys has settled on. The optimum is to charge fares to pay infrastructure costs but not much more – especially if you’re SNCF and the railway workers’ union will extract all further profit through strikes, as it did 10 years ago. And this means making sure that except at very busy times, known in advance, Paris-Brussels tickets should be 30€, not 50-100€.

The Northeastern United States Wants to Set Tens of Billions on Fire Again

The prospect of federal funds from the Bipartisan Infrastructure Bill is getting every agency salivating with desires for outside money for both useful and useless priorities. Northeastern mainline rail, unfortunately, tilts heavily toward the useless, per a deep dive into documents by New York-area activists, for example here and here.

Amtrak is already hiring project management for Penn Station redevelopment. This is a project with no transportation value whatsoever: this is not the Gateway tunnels, which stand to double capacity across the Hudson, but rather a rebuild of Penn Station to add more tracks, which are not necessary. Amtrak’s current claim is that the cost just for renovating the existing station is $6.5 billion and that of adding tracks is $10.5 billion; the latter project has ballooned from seven tracks to 9-12 tracks, to be built on two levels.

This is complete overkill. New train stations in big cities are uncommon, but they do exist, and where tracks are tunneled, the standard is two platform tracks per approach tracks. This is how Berlin Hauptbahnhof’s deep section goes: the North-South Main Line is four tracks, and the station has eight, on four platforms. Stuttgart 21 is planned in the same way. In the best case, each of the approach track splits into two tracks and the two tracks serve the same platform. Penn Station has 21 tracks and, with the maximal post-Gateway scenario, six approach tracks on each side; therefore, extra tracks are not needed. What’s more, bundling 12 platform tracks into a project that adds just two approach tracks is pointless.

This is a combined $17 billion that Amtrak wants to spend with no benefit whatsoever; this budget by itself could build high-speed rail from Boston to Washington.

Or at least it could if any of the railroads on the Northeast Corridor were both interested and expert in high-speed rail construction. Connecticut is planning on $8-10 billion just to do track repairs aiming at cutting 25-30 minutes from the New York-New Haven trip times; as I wrote last year when these plans were first released, the reconstruction required to cut around 40 minutes and also upgrade the branches is similar in scope to ongoing renovations of Germany’s oldest and longest high-speed line, which cost 640M€ as a once in a generation project.

In addition to spending about an order of magnitude too much on a smaller project, Connecticut also thinks the New Haven Line needs a dedicated freight track. The extent of freight traffic on the line is unclear, since the consultant report‘s stated numbers are self-contradictory and look like a typo, but it looks like there are 11 trains on the line every day. With some constraints, this traffic fits in the evening off-peak without the need for nighttime operations. With no constraints, it fits on a single track at night, and because the corridor has four tracks, it’s possible to isolate one local track for freight while maintenance is done (with a track renewal machine, which US passenger railroads do not use) on the two tracks not adjacent to it. The cost of the extra freight track and the other order-of-magnitude-too-costly state of good repair elements, including about 100% extra for procurement extras (force account, contingency, etc.), is $300 million for 5.4 km.

I would counsel the federal government not to fund any of this. The costs are too high, the benefits are at best minimal and at worst worse than nothing, and the agencies in question have shown time and time again that they are incurious of best practices. There is no path forward with those agencies and their leadership staying in place; removal of senior management at the state DOTs, agencies, and Amtrak and their replacement with people with experience of executing successful mainline rail projects is necessary. Those people, moreover, are mid-level European and Asian engineers working as civil servants, and not consultants or political appointees. The role of the top political layer is to insulate those engineers from pressure by anti-modern interest groups such as petty local politicians and traditional railroaders who for whatever reasons could not just be removed.

If federal agencies are interested in building something useful with the tens of billions of BIL money, they should instead demand the same results seen in countries where the main language is not English, and staff up permanent civil service run by people with experience in those countries. Following best industry practices, $17 billion is enough to renovate the parts of the Northeast Corridor that require renovation and bypass those that require greenfield bypasses; even without Gateway, Amtrak can squeeze a 16-car train every 15 minutes, providing 4,400 seats into Penn Station in an hour, compared with around 1,700 today – and Gateway itself is doable for low single-digit billions given better planning and engineering.

German Rail Traffic Surges

DB announced today that it had 500,000 riders across the two days of last weekend. This is a record weekend traffic; May is so far 5% above 2019 levels, representing full recovery from corona. This is especially notable because of Germany’s upcoming 9-euro ticket: as a measure to curb high fuel price from the Russian war in Ukraine, during the months of June, July, and August, Germany is both slashing fuel taxes by 0.30€/liter and instituting a national 9€/month public transport ticket valid not just in one’s city of domicile but everywhere. In practice, rail riders respond by planning domestic rail trips for the upcoming three months; intercity trains are not covered by the 9€ monthly pass, but city transit in destination cities is, so Berliners I know are planning to travel to other parts of Germany during the window when local and regional transit is free, displacing trips that might be undertaken in May.

This is excellent news, with just one problem: Germany has not invested in its rail network enough to deal with the surge in traffic. Current traffic is already reaching projections made in the 2010s for 2030, when most of the Deutschlandtakt is supposed to go into effect, with higher speed and higher capacity than the network has today. Travel websites are already warning of capacity crunches in the upcoming three months of effectively free regional travel (chaining regional trains between cities is possible and those are covered by the 9€ monthly pass). Investment in capacity is urgent.

Sadly, such investment is still lagging. Germany’s intercity rail network rarely builds complete high-speed lines between major cities. The longest all-high-speed connection is between Cologne and Frankfurt, 180 km apart. Longer connections always have significant slow sections: Hamburg-Hanover remains slow due to local NIMBY opposition to a high-speed line, Munich’s lines to both Ingolstadt and Augsburg are slow, Berlin’s line toward Leipzig is upgraded to 200 km/h but not to full high-speed standards.

Moreover, plans to build high-speed rail in Germany remain compromised in two ways. First, they still avoid building completely high-speed lines between major cities. For example, the line from Hanover to the Rhine-Ruhr is slow, leading to plans for a high-speed line between Hanover and Bielefeld, and potentially also from Bielefeld to Hamm; but Hamm is a city of 180,000 people at the eastern margin of the Ruhr, 30 km from Dortmund and 60 from Essen. And second, the design standards are often too slow as well – Hanover-Bielefeld, a distance that the newest Velaro Novo trains could cover in about 28 minutes, is planned to be 31, compromising the half-hourly and hourly connections in the D-Takt. Both of these compromises create a network that 15 years from now is planned to have substantially lower average speeds than those achieved by France 20 years ago and by Spain 10 years ago.

But this isn’t just speed, but also capacity. An incomplete high-speed rail network overloads the remaining shared sections. A complete one removes fast trains from the legacy network except in legacy rail terminals where there are many tracks and average speeds are never high anyway; Berlin, for example, has four north-south tracks feeding Hauptbahnhof with just six trains per hour per direction. In China, very high throughput of both passenger rail (more p-km per route-km than anywhere in Europe) and freight rail (more ton-km per route-km than the United States) through the removal of intercity trains from the legacy network to the high-speed one, whose lines are called passenger-dedicated lines.

So to deal with the traffic surge, Germany needs to make sure it invests in intercity rail capacity immediately. This means all of the following items:

  • Building all the currently discussed high-speed lines, like Frankfurt-Mannheim, Ulm-Augsburg (Stuttgart-Ulm is already under construction), and Hanover-Bielefeld.
  • Completing the network by building high-speed lines even where average speeds today are respectable, like Berlin-Halle/Leipzig and Munich-Ingolstadt, and making sure they are built as close to city center as possible, that is to Dortmund and not just Hamm, to Frankfurt and not just Hanau, etc.
  • Purchasing 300 km/h trains and not just 250 km/h ones; the trains cost more but the travel time reduction is noticeable and certain key connections work out for a higher-speed D-Takt only at 300, not 250.
  • Designing high-speed lines for the exclusive use of passenger trains, rather than mixed lines with gentler freight-friendly grades and more tunnels. Germany has far more high-speed tunneling than France, not because its geography is more rugged, but because it builds mixed lines.
  • Accelerating construction and reducing costs through removal of NIMBY veto points. Groups should have only two months to object, as in Spain; current practice is that groups have two months to say that they will object but do not need to say what the grounds for those objections are, and subsequently they have all the time they need to come up with excuses.

Systemic Investments in the New York City Subway

Subway investments can include expansion of the map of lines, for example Second Avenue Subway; proposals for such extensions are affectionately called crayon, a term from London Reconnections that hopped the Pond. But they can also include improvements that are not visible as lines on a map, and yet are visible to passengers in the form of better service: faster, more reliable, more accessible, and more frequent.

Yesterday I asked on Twitter what subway investments people think New York should get, and people mostly gave their crayons. Most people gave the same list of core lines – Second Avenue Subway Phase 2, an extension of the 2 and 5 on Nostrand, an extension of the 4 on Utica, an extension of the N and W to LaGuardia, the ongoing Interborough Express proposal, and an extension of Second Avenue Subway along 125th – but beyond that there’s wide divergence and a lot of people argue over the merits of various extensions. But then an anonymous account that began last year and has 21 followers and yet has proven extremely fluent in the New York transit advocacy conversation, named N_LaGuardia, asked a more interesting question: what non-crayon systemic investments do people think the subway needs?

On the latter question, there seems to be wide agreement among area technical advocates, and as far as I can tell the main advocacy organizations agree on most points. To the extent people gave differing answers in N_LaGuardia’s thread, it was about not thinking of everything at once, or running into the Twitter character limit.

It is unfortunate that many of these features requiring capital construction run into the usual New York problem of excessive construction costs. The same institutional mechanisms that make the region incapable of building much additional extension of the system also frustrate systemwide upgrades to station infrastructure and signaling.

Accessibility

New York has one of the world’s least accessible major metro systems, alongside London and (even worse) Paris. In contrast, Berlin, of similar age, is two-thirds accessible and planned to reach 100% soon, and the same is true of Madrid; Seoul is newer but was not built accessible and retrofits are nearly complete, with the few remaining gaps generating much outrage by people with disabilities.

Unfortunately, like most other forms of capital construction in New York, accessibility retrofits are unusually costly. The elevator retrofits from the last capital plan were $40 million per station, and the next batch is in theory $50 million, with the public-facing estimates saying $70 million with contingency; the range in the European cities with extensive accessibility (that is, not London or Paris) is entirely single-digit million. Nonetheless, this is understood to be a priority in New York and must be accelerated to improve the quality of universal design in the system.

Platform screen doors

The issue of platform screen doors (PSDs) or platform edge doors (PEDs) became salient earlier this year due to a much-publicized homicide by pushing a passenger onto a train, and the MTA eventually agreed to pilot PSDs at three stations. The benefits of PSDs are numerous, including,

  • Safety – there are tens of accident and suicide deaths every year from falling onto tracks, in addition to the aforementioned homicide.
  • Greater accessibility – people with balance problems have less to worry about from falling onto the track.
  • Capacity – PSDs take up platform space but they permit passengers to stand right next to them, and the overall effect is to reduce platform overcrowding at busy times.
  • Air cooling – at subway stations with full-height PSDs (which are rare in retrofits but I’m told exist in Seoul), it’s easier to install air conditioning for summer cooling.

The main difficulty is that PSDs require trains to stop at precise locations, to within about a meter, which requires signaling improvements (see below). Moreover, in New York, trains do not yet have consistent door placement, and the lettered lines even have different numbers of doors sometimes (4 per car but the cars can be 60′ or 75′ long) – and the heavily interlined system is such that it’s hard to segregate lines into captive fleets.

But the biggest difficulty, as with accessibility, is again the costs. In the wake of public agitation for PSDs earlier this year, the MTA released as 2019 study saying only 128 stations could be retrofitted with PSDs, at a cost of $7 billion each, or $55 million per station; in Paris, PSDs are installed on Métro lines as they are being automated, at a cost of (per Wikipedia) 4M€ per station of about half the platform length as in New York.

Signaling improvements

New York relies on ancient signaling for the subway. This leads to multiple problems: maintenance is difficult as the international suppliers no longer make the required spare parts; the signals are designed around the performance specs of generations-old trains and reduce capacity on more modern trains; the signals are confusing to drivers and therefore trains run slower than they can.

To modernize them, New York is going straight to the most advanced system available: CBTC, or communications-based train control, also known as moving-block signaling. This is already done on the L and 7 trains and is under installation on other lines, which are not isolated from the rest of the system. CBTC permits much higher peak capacity in London; in New York, unfortunately, this effect has been weaker because of other constraints, including weak electrical substation capacity and bumper tracks at the terminals of both the L and the 7.

Moreover, in New York, the L train’s performance was derated when CBTC was installed, to reduce brake wear. The effect of such computer control should be the opposite, as computers drive more precisely than humans: in Paris, the automation of Line 1 led to a speed increase of 15-20%, and CBTC even without automation has the same precision level as full automation.

As before, costs form a major barrier. I can’t give the most recent analogs, because such projects tend to bundle a lot of extras, such as new trainsets and PSDs in Paris. In Nuremberg, the first city in the world to permanently convert a preexisting metro system to driverless operations, the cost of just the driverless system is said to have been 110M€ in the late 2000s, for what I believe is 13 km of U2 (U3 was built with driverless operations in mind, and then U2, from which it branches, was converted). It is said that automating U1 should cost 100M€ for 19.5 km, but this project is not happening due to stiff competition for federal funds and therefore its real cost is uncertain. In contrast, Reinvent Albany quotes $636 million for the 7 train in New York, of which $202 million must be excluded as rolling stock conversion; the Flushing Line is 16 km long, so this is still $27 million/km and not the $7-12 million/km of Nuremberg.

Maintenance regime

The maintenance regime in New York involves heavy slowdowns and capacity restrictions. Trains run 24/7 without any breaks for regular maintenance. Instead, maintenance is done one track at a time during off-peak periods, with flagging rules that slow down trains on adjacent tracks and have gotten more onerous over the last 10-20 years; only recently have planners begun to use temporary barriers to reduce the burden of flagging.

The result of this system is threefold. First, track maintenance productivity is extremely low – the train on an adjacent track slows down as it passes but the work stops as it passes as well. Second, speeds are unreliable off-peak and the timetable is in perpetual firefighting mode. And third, parts of the system are claimed to be incapable of running more than about 16 trains per hour off-peak, which means that if there is any branching, the branches are limited to 8, which is not enough frequency on a major urban metro system.

It takes a small amount of capital spending to increase efficiency of maintenance, through procuring more advanced machinery, installing barriers between tracks, and installing crossovers at appropriate locations. But it takes a large degree of operations and management reform to get there, which is necessary for reducing the high operating costs of the subway.

Deinterlining

New York has the most complicated interlining of any global metro network. Only four lines – the 1, 6, 7, and L – run by themselves without any track sharing with other lines. The 2, 3, 4, and 5 share tracks with one another. Then the lettered trains other than the L all share tracks on various segments, without any further segregation. Only some commuter rail networks are more complex than this – and even Tokyo has greater degree of segregation between different trunk lines, despite extensive through-service to commuter rail. The New York way guarantees more direct service on more origin-destination pairs, but at low frequency and with poor speed and reliability.

London, the second most interlined system, has long wanted to reduce interlining to increase capacity. The Northern line traditionally had just one southern segment reverse-branching to two central trunks, combining and splitting into two northern branches. When CBTC opened, the busier of the central trunks got 26 peak trains per hour; the more recent Battersea extension removed the interlining to the south, permitting boosting capacity up to 32 tph, and full deinterlining to the north would boost it to 36 tph, as on the most captive Underground lines.

In New York, it is desirable to remove all reverse-branching. At DeKalb Avenue in Downtown Brooklyn, the interlocking switches the four express (bridge) tracks from an arrangement of the B and D on one track pair and the N and Q on the other to the B and Q on one track pair and the D and N on the other; the process is so complex that every train is delayed two minutes just from the operation of the switches. Everywhere within the system, interlining creates too much dependency between the different trains, so that delays on one line propagate to the others, reducing reliability, speed, and capacity.

Some of the problem is, as usual, about high costs. Rogers Avenue Junction controls the branching of the 2, 3, 4, and 5 trains in Brooklyn, transitioning from the 2 and 3 sharing one track pair and the 4 and 5 sharing another to the 3 and 4 running on dedicated tracks and the 2 and 5 sharing tracks. For a brief segment, the 2, 3 and 5 trains all share tracks. This devastates capacity on both trunk lines, which rank first and third citywide in peak crowding as of the eve of the opening of Second Avenue Subway. There are already internal designs for rebuilding the junction to avoid this problem – at a cost of $300 million.

But some of the problem is also about operating paradigms. New York must move away from the scheduling ideas of the 1920s and 30s and understand that independently-operated lines with dedicated fleets and timetables, with passengers making transfers as appropriate, are more robust and overall better for most riders. DeKalb can be deinterlined with no capital spending at all, and so can Columbus Circle. It’s Rogers and Queens Plaza where spending is ideal (but even then, not strictly required if some operational compromises are made), and the 142nd Street Junction in Harlem where an extensive rebuild is obligatory in order to permit splitting the 2 from the 5 in the Bronx permanently.

Labor changes

Staffing levels in New York are very high. Trains have conductors and not just drivers; this is not globally unheard of (Toronto and some lines in Tokyo still have conductors) but it’s rare. With good enough signaling, a retrofit even for full automation is possible, as in Nuremberg, Paris, and Singapore. Maintenance work is likewise unproductive, not because people don’t work hard, but because they work inefficiently.

Improving this situation involves changes on both sides of the ledger – staffing and service. Conductors have to be cut for efficiency and not all of them can be absorbed by other roles, and the same is true of some station facilities and maintenance functions. In contrast, the low productivity of drivers in New York – they spend around 550 hours a year driving a revenue train whereas Berlin’s drivers, who get 6 weeks of annual paid vacation, scratch 900 – is the result of poor off-peak frequency, and must be resolved through increases in off-peak service that increase efficiency without layoffs.

Ultimate goal: six-minute service

I wrote two years ago about what it would take to ensure every public transit service in New York runs every six minutes off-peak, calling it a six-minute city.

Riders Alliance argues for the same goal, with the hashtag #6minuteservice; I do not know if they were basing this on what I’d written or if it’s convergent evolution. But it’s a good design goal for timetabling, with implications for labor efficiency, maintenance efficiency, the schedule paradigm, and the bus system.

No tradeoffs

It is fortunate that the agenda of systemwide improvements does not exhibit significant tradeoffs in investment. Other parts of the transit agenda do not need to suffer to implement those improvements. On the contrary, they tend to interact positively: accessibility and PSDs can be combined (and federal law is written in such a way that PSDs void the grandfather clause permitting the subway to keep most of its stations inaccessible), faster and more reliable trains can be run more frequently off-peak, better service means higher ridership and therefore higher demand for extensions. Only the issue of labor exhibits a clear set of losers from the changes, and those can be compensated in a one-time deal.

Moreover, the budget for such an agenda is reasonable, if New York can keep its construction costs under control. At the per-elevator costs of Berlin or Madrid, New York could make its entire network wheelchair-accessible for around $3.5-4 billion. Parisian PSDs, pro-rated to the greater size of New York trains, would be around $10 million a station, or $5 billion systemwide. Full automation at German costs would be maybe $6 billion with triple- and quad-track lines pro-rated. The entire slate of changes required for full deinterlining, including a pocket track for the 3 train at 135th Street, a rebuild of the 36th Street station in Queens, and a connection between Queensboro Plaza and Queens Plaza, should be measured in the hundreds of millions, not billions.

The overall program still goes into double-digit billions; it requires a big push. But this big push is worth two to three years’ worth of current New York City Transit capital spending. A New York that can do this can also add 50-100 km to its subway network and vice versa, all while holding down operating costs to typical first-world levels. For the most part, the planners already know what needs to be done; the hard part is getting construction costs to reasonable levels so that they can do it on the current budget.

Why We Adjust Costs for PPP

The Transit Costs Project adjusts all construction costs for purchasing power parities. This means that, for example, a Chinese subway is converted into dollars not at the exchange rate of $1 = 6.7¥, but at the PPP rate of $1 = 4.2¥; this means that present-day Chinese subways look 1.5 times more expensive in our analysis than in analyses that use exchange rate values, and projects from 10 years ago look twice as expensive. I believe our choice is correct, and would like to explain why, since it has gotten some criticism from serious people, who’s prefer exchange rates.

Local costs

I started this comparing mature developed countries. The US and Europe have largely separate markets for construction, and so American work is almost entirely done in dollars and European work in euros (or pounds, or kronor, etc.). Japan is likewise very local and so is China. In that case, local costs matter far more than international ones.

But what’s interesting is that even in countries that use imported technology and international consultants and contractors and have low wages, costs are almost entirely local. I wrote about this last year, referencing an article out of India about the small cost impact of indigenization and an interview I made with a Philippine planner who told me 90% of the value of civil works is local. Rolling stock is internationally traded, but we exclude it from our cost estimates whenever possible.

The impact of currency changes

Using PPPs, if a country undergoes a bout of inflation, this should be reflected in changes in construction costs. This is intentional. The example given to me in the critique linked in the lede is that if Bangladeshi food prices rise, then this makes the PPP exchange rate look less favorable (a taka in Bangladesh can then buy less relative to a dollar in the US). But that’s fine – if Bangladeshi food prices rise then this forces Dhaka to pay higher wages to MRT construction workers, so overall it’s just domestic inflation. It’s no different from how, today, we’re seeing nominal construction cost growth in the United States and Europe because of high inflation.

At least the inflation today is moderate by any developing-country standard. Core inflation in the United States is 6%; in Germany it’s 3%. This may introduce third-order errors into the database as we deflate costs to the midpoint of construction. In contrast, 50-60% annual inflation is sustained over years in some middle-income countries like Iran, and then the choice of year for prices has significant impact, to the point that Iranian costs have a significant error bar. But that’s regardless of whether one adjusts for PPP or not, since usually inflation leads to deteriorating terms of trade.

In contrast, if prices are compared in exchange rate terms, then international fluctuations create fictitious changes in construction costs. When China permitted the renminbi to appreciate in the mid-2000s, this would have looked like an increase in costs of about 20% – but the costs of local inputs did not change, so in reality there was no increase in costs. The euro:dollar rate peaked around 1€ = $1.58 in 2008, before tumbling to 1€ = $1.28 in the financial crisis – but nothing material happened that would reduce European construction costs by 19% relative to American ones; right now it’s trading at 1€ = $1.05, but this again does not mean that construction in Europe is suddenly a third cheaper compared with in the US relative to 15 years ago.

Unusual currency values

Some patterns are systemic – richer countries have stronger currencies relative to PPP value than poor countries. But others are not, and it’s important to control for them. A currency can be weak due to the risk of war or disaster; the Taiwanese dollar is unusually weak for how rich Taiwan is, and this should not mean that Taiwanese construction costs are half what they really are. Or it can be strong or weak based on long-term investment proposition: investors will bid up the value of a currency in a country they expect to profit in in the long term, perhaps due to population growth coming from high birthrates or immigration, and this does not mean that today, it builds infrastructure more expensively.

In any of those cases, the unusual value of the currency really reflects capital availability. Capital for investment in Australia is plentiful, but this by itself does not raise its construction costs; capital for investment in Taiwan is scarce, but this certainly does not make it a cheap place to build infrastructure.

Foreign-denominated construction

In some peripheral countries with unstable currencies, costs are quoted in foreign currency – dollars or euros. Some Turkish contracts are so quoted, and this is also common in Latin America and sometimes Southeast Asia. But ultimately, the vast majority of the contract’s value is paid out in the local currency, not just labor but also locally-made materials like concrete. This creates a weird-looking statistical artifact in which we convert dollars or euros to local currency in exchange rate terms and then back in PPP terms.

This, we do because the quotation of the contract (in dollars or euros) is not the real value. Rather, it comes out of one of two artifacts. The first is data reporting: we rely on international trade media, and those often quote prices in exchange rate dollars or euros, even if the contract is in local currency (and in all cases where we’ve seen both, they match in exchange rate value).

The second is that an international consultancy may demand actual payment in foreign currency as a hedge against currency depreciation; in that case its rate of profit should be dollar- or euro-denominated. However, this again is a small minority of overall contract value. Moreover, if a country’s institutions can’t produce enough capital stability to do business in their own currency, it’s a problem that should be reflected in global indices; ultimately, if costs are higher in PPP terms as a result, this means that the country really does have greater problem affording infrastructure.

A posteriori justification

The above reasoning is all a priori. When I started comparing costs in the early 2010s, I was comparing developed countries and the euro:dollar rate was in flux in the early financial crisis, so I just went with one long-term PPP rate.

However, a posteriori, there is another positive feature of PPP adjustment: it levels the differences in construction costs by income. There is positive correlation between metro cost per km and the GDP per capita of the country the metro is built in, about 0.22, but it comes entirely out of the fact that poorer countries (especially India) build more elevated and fewer subway lines; correcting for this factor, the correlation vanishes. This is as it should be: PPP is a way of averaging out costs in different countries, first because it levels short-term fluctuations such as between different developed countries, and second because exchange rate value is dominated by internationally tradable goods, which are relatively more expensive in poor countries than non-tradable goods like food and housing.

What this says is that infrastructure should be viewed as an average-tradable good, at least a posteriori: its variation in costs across the world is such that there is no correlation with GDP per capita, whereas food prices display positive correlation even after PPP adjustment, and tradables like smartphones display negative correlation (because they cost largely the same in exchange rate terms).

Tails on Commuter Rail

An interesting discussion on Twitter came out of an alternatives analysis for Philadelphia commuter rail improvements. I don’t want to discuss the issue at hand for now (namely, forced transfers), but the discussion of Philadelphia leads to a broader question about tails. Commuter rail systems sometimes have low-frequency tails with through-service to the core system and sometimes don’t, and it’s useful to understand both approaches.

What is a tail?

For the purposes of this post, a tail is whenever there is a frequent line with trains infrequently continuing farther out. Frequency here is relative, so a subway line running every 2.5 minutes to a destination with every fourth train continuing onward is a tail even though the tail still has 10-minute frequency, and a commuter line running every 20 minutes with every third train continuing onward also has a tail, even though in the latter case the core frequency is lower than the tail frequency in the former case.

The key here is that the line serves two markets, one high-intensity and frequent and one lower-intensity warranting less service, with the outer travel market running through to the inner one. Usually the implication is that the inner segment can survive on its own and the contribution of the outer segment to ridership is not significant by itself. In contrast, it’s common enough on S-Bahn systems to have a very frequent trunk (as in Berlin, or Munich, or Paris) that fundamentally depends on through-service from many suburban segments farther out combining to support high frequency in the core; if ridership farther out is significant enough that without it frequency in the core would suffer, I would not call this a tail.

When are tails useful?

Tails are useful whenever there is a core line that happens to be along the same route as a lower-intensity suburban line. In that case, the suburban line behind can benefit from the strong service in the core by having direct through-service to it at a frequency that’s probably higher than it could support by itself. This is especially valuable as the ridership of the tail grows in proportion to that of the core segment – in the limiting case, it’s not even a tail, just outer branches that combine to support strong core frequency.

Tokyo makes extensive use of tails. The JR East commuter lines all have putative natural ends within the urban area. For example, most Chuo Rapid Line trains turn at Takao, at the western end of the built-up area of Tokyo – but some continue onward to the west, running as regional trains to Otsuki or as interregional or as intercity trains farther west to Shiojiri.

Munich and Zurich both use tails as well on their S-Bahns. In Munich, the base frequency of each of the seven main services is every 20 minutes, but some have tails running hourly, and all have tails running two trains per hour with awkward alternation of 20- and 40-minute gaps. In Zurich, the system is more complex, and some lines have tails (for example, S4) and some do not (for example, S3); S4 is not a portion of an intercity line the way the Chuo Line is, and yet its terminus only gets hourly trains, while most of the line gets a train every 20 minutes.

What are the drawbacks of tails?

A tail is a commitment to running similar service as in the core, just at lower frequency. In Philadelphia, the proposal to avoid tails and instead force what would be tails into off-peak shuttle trains with timed transfers to the core system is bundled into separate brands for inner and outer service and a desire to keep the outer stations underbuilt, without accessibility or high platforms. Branding is an exercise in futility in this context, but there are, in other places than Philadelphia, legitimate reasons to avoid tails, as in Paris and Berlin:

  • Different construction standards – perhaps the core is electrified and an outer segment is not; historically, this was the reason Philadelphia ended commuter rail service past the limit of electrification, becoming the only all-electrified American commuter rail network. In Berlin, the electrification standards on the mainline and on the S-Bahn differ as the S-Bahn was electrified decades earlier and is run as an almost entirely self-contained system.
  • Train size difference – sometimes the gap in demand is such that the tail needs not just lower frequency than the core but also shorter trains. In the United States, Trenton is a good example of this – New York-Trenton is a much higher-demand line than Trenton-Philadelphia and runs longer trains, which is one reason commuter trains do not run through.
  • Extra tracks – if there are express tracks on the core segment, then it may be desirable to run a tail express, if it is part of an intercity line like the Chuo Line rather than an isolated regional line like S4 in Zurich, and not have it interface with the core commuter line at all to avoid timetabling complications. If there are no extra tracks, then the tail would have to terminate at the connection point with the core line, as is proposed in Philadelphia, and the forced transfer is a drawback that generally justifies running the tail.

Do the drawbacks justify curtailment?

Not really. On two-track lines, it’s useful to provide service into city center from the entire line, just maybe not at high frequency on outer segments. This can create situations in which intercity-scale lines run as commuter rail lines that keep going farther than typical, and this is fine – the JR East lines do this on their rapid track pairs and within the built-up area of Tokyo people use those longer-range trains in the same way they would an ordinary rapid commuter train.

This is especially important to understand in the United States, which is poor in four-track approaches of the kind that the largest European cities have. I think both Paris and Berlin should be incorporating their regional lines into the core RER and S-Bahn as tails, but they make it work without this by running those trains on dedicated tracks shared with intercity service but not commuter rail. Boston, New York, and Philadelphia do not have this ability, because they lack the ability to segregate S-Bahn and RegionalBahn services. This means Boston should be running trains to Cape Cod, Manchester, and Springfield as tails of the core system, and New York should electrify its entire system and run trains to the Hamptons as LIRR tails, and Philadelphia should run tail trains to the entire reach of its commuter rail system.

Quick Note: How to Incentivize Transit-Oriented Development

The Biden administration recently put out a statement saying that it would work to increase national housing production. It talks about the need to close the housing shortfall, estimated at 1.5 million dwellings, and proposes to use the Bipartisan Infrastructure Law (BIL) to dole out transport funding based on housing production. This is a welcome development, and I’d like to offer some guidelines for how this can be done most effectively.

Incentives mean mistrust

You do not need to give incentives to trustworthy people. The notion of incentives already assumes that the people who are so governed would behave poorly by themselves, and that the governing body, in this case the federal government, surveils them loosely so as to judge them by visible metrics set in advance. Once this fundamental fact is accepted – the use of BIL funding to encourage housing production implies mistrust of all local government to build housing – every other detail should be set up in support of it.

Demand conflict with community

Federal funding should, in all cases, require state and local governments to discipline community groups that fight housing and extract surplus from infrastructure. Regions that cannot or do not do so should receive less funding; the feds should communicate this in advance, stating both the principle and the rules by which it will be judged. For example, a history of surrender to local NIMBYs to avoid lawsuits, or else an unwillingness to fight said lawsuits, should make a region less favored for funds, since it’s showing that they will be wasted. In contrast, a history of steamrolling community should be rewarded, showing that the government is in control and prioritizes explicit promises to the feds and the voters over implicit promises to the local notables who form the base of NIMBYism.

Spend money in growth regions

In cities without much housing demand, like Detroit and Cleveland, the problem of housing affordability is one of poverty; infrastructure spending wouldn’t fix anything. This means that the housing grant should prioritize places with growth demand, where current prices greatly exceed construction costs. These include constrained expensive cities like New York and San Francisco, but increasingly also other wealthy cities like Denver and Nashville, whose economic booms translate to population increase as well as income growth, but unfortunately housing growth lags demand. Even poorer interior cities are seeing rent increases as people flee the high prices of richer places, and encouraging housing growth in their centers is welcome (but not in their suburbs, where housing is abundant and not as desirable).

Look at residential, not commercial development

In the United States, YIMBY groups have focused exclusively on residential development. This is partly for political reasons: it’s easier to portray housing as more moral, benefiting residents who need affordable housing even if the building in question is market-rate, than to portray an office building as needing political support. In some cases it’s due to perceived economic reasons – the two cities driving the American YIMBY discourse, New York and San Francisco, have unusually low levels of job sprawl for the United States, and in both cities YIMBY groups are based near city center, where jobs look especially plentiful. At the local and state level, this indifference to commercial YIMBY is bad, because it’s necessary to build taller in city center and commercialize near-center neighborhoods like the West Village to fight off job sprawl.

However, at the federal level, a focus on residential development is good. This is a consequence of the inherent mistrust assumed in the incentive system. While economically, American cities need city centers to grow beyond the few downtown blocks they currently occupy, politically it’s too easy for local actors to bundle a city center expansion with an outrageously expensive urban renewal infrastructure plan. In New York, this is Penn Station redevelopment, including some office towers in the area that are pretty useful and yet have no reason to be attached to the ill-advised Penn Station South project digging up an entire block to build new tracks. Residential development is done at smaller scale and is harder to bundle with such unnecessary signature projects; the sort of projects that are bundled with it are extensions of urban rail to new neighborhoods to be redeveloped, and those are easier to judge on the usual transport metrics.