Category: Urban Transit

When Transit Serves the Poor Better Than the Rich

In major transit cities, rich areas have better access to public transportation than poor areas – in fact, what makes them valuable is precisely the easy access to high-paying jobs. Even in cities with bad transit, this is often the case: the transit systems of cities with mode shares in the 10-15% area, like Boston and Chicago, tend to be good at serving city center and little else, and city center workers tend to be richer because professional work tends to cluster whereas low-skill work tends to disperse.

However, there are exceptions to this rule. One, the French Riviera, occurs in a city region with a transit mode share of 13%, comparable to that of American city regions where transit commuters outearn solo drivers. Two more cities are would-be exceptions, for opposite reasons: Providence has no public transit to speak of, but if it invested in creating a transit network, the natural corridors would serve the poor better than the rich; and Vancouver currently has better SkyTrain service in working-class areas than in richer ones, but its current investment is in middle-class areas, and moreover its extensive transit-oriented development has been middle-class as well.

Moreover, all three cities have patterns that generalize. The situation in the Riviera arises because of the classed nature of work there, and generalizes to other places with extensive tourism. That in Providence arises because of the city’s industrial history, and may generalize to other deindustrialized small cities with underutilized legacy rail networks. In Vancouver, part of this situation is because easy rail corridors were more readily available in poorer areas for an essentially random reason, but another part is extensive transit-oriented development concentrating working-class jobs near train stations.

The Riviera: the casinos are walkable, the tech jobs aren’t

Before I go any further, I’d like to stress something important: my observation of the Riviera is largely based on qualitative observations. I don’t know of INSEE data comparable to the Census Bureau’s Means of Transportation to Work by Selected Characteristics table, which could allow me to test the theory that transit ridership in the Riviera skews poor. All I am going by is what I have seen riding trains and occasionally buses as well as what I know of the distribution of jobs.

What I’ve seen is that transit use in the Riviera skews working class. Middle-class Parisians sometimes drive and sometimes take the trains. In contrast, the rich people who I’ve met in the Riviera have as far as I can tell never set foot on the TER. This is despite the fact that the TER is competitive with driving on the area’s main arterial road, the Moyenne Corniche, and is even competitive with the A8 freeway over short distances because the A8 has difficult access time to the relevant exits. Not for nothing, train stations in rich areas have very little ridership: per SNCF’s ridership data, stations in rich areas like Cap d’Ail and Cap Martin-Roquebrune have around 60,000 boardings plus alightings per year, so around 100 weekday boardings, whereas in working- and lower-middle-class Menton the annual total is 1.4 million, or around 2,300 weekday boardings.

The train stations, too, signal poverty. They’re not neglected, but what I’ve seen of them reminded me of working-class suburbs of Paris like Boissy much more than middle-class ones like Bures-sur-Yvette. I was even warned off of spending too much time near Nice’s train station by people echoing local middle-class prejudices. The buses look even poorer: the main east-west bus on the Moyenne Corniche is full of migrant workers.

A key clue for what is happening can be found when selecting a destination station at the fare machines in Menton. As far as I remember, the first option given is not Nice, but Monaco. SNCF’s data table doesn’t include ridership for Monaco, but Wikipedia claims 5.5 million a year without citation, and SNCF’s own blurb claims more than 6 million. Either figure is narrowly behind Nice’s 6.9 million for second in the Riviera and well ahead of third-place Cannes’s 3.2 million – and Nice also has some intercity traffic.

While Monaco’s residents are rich, its commuters are not. There are no corporate jobs in Monaco, because its tax haven status does not extend to corporations with substantial sales outside the city-state, only to local businesses like restaurants and stores. The commuters work low-pay service jobs at hotels and casinos, which they access by train, or perhaps on foot if they live in Beausoleil, as many a domestic service worker in Monaco does.

In contrast, the mass of middle-class jobs cluster in a purpose-built edge city in Antibes, called Sophia-Antipolis. While Antibes itself has a decent transit mode share for residents (10.5%, cf. Menton’s 14.8% and Nice’s 25.4%), and its train station gets 1.6 million annual boardings and alightings, the edge city is unwalkable and far from the train. There’s some traffic in the Riviera, but not enough that middle-class people, who can afford cars, clamor for transit alternatives to their suburban jobs.

The main lesson here is that while the jobs most likely to cluster are usually middle-class city center jobs, working-class tourism jobs cluster as well in regions that have plenty of them. Tourism in the Riviera is the most intense in Monaco specifically and in other coastal cities generally, which encourages travel along the linear corridor, where rail shines. It’s usually hard to see, because for the most part the top tourist destinations are enormous like London, Paris, and New York, but in specialized tourist regions the separation is clearer.

Already we see some evidence of this in Las Vegas, where working-class jobs cluster along the Strip. The city has a monorail, serving the hotels and casinos rather than city center. Were it interested in improving public transportation, it could build an elevated railroad on the Strip itself for better service.

Orlando is another potential example. I named it as a specific example of a region that would be difficult to retrofit for public transit earlier this year, but Disney World remains a major clustering of working-class jobs as well as some middle-class leisure travel. The problem there is that Disney World is far from the train and, unlike the Riviera, does not lie on any line with other potential ridership draws; nonetheless, a train connecting the Orlando CBD, the airport, and Disney World could get some traffic.

Finally, picturesque mountain resorts that happen to lie near rail could see working-class travel on the train to their tourism jobs. Many of these resorts are where they are specifically because a legacy rail trunk happened to be there and the railroad developed the area to generate demand for its services; this is the case for Jasper, Lake Louise, and Banff, all on the Alberta side of the Continental Divide. Aspen is not on a railroad, but is on a road where buses carry working-class commuters displaced by the town’s high housing costs.

Providence: once upon a time, there were factories near the railroad

When I lived in Providence seven years ago, I discussed transit improvements with local urbanists who I met through Greater City: Providence. We talked about improvements to both bus and rail; we had little appetite for the proposed city center streetcar, which has since been downgraded to a proposed frequent bus, and instead talked about improvements to the busiest buses as well as rail service along the main spine of the Northeast Corridor.

The improvements to the busiest buses were already under discussion by the state, including signal priority on key routes and investment in queue jump lanes and shelter amenities. The two routes that were by far the state’s busiest, the 99 on North Main and 11 on Broad, were permanently combined to a single through-running service branded as the R bus, for rapid, with limited-stop service. These routes serve very poor parts of the built-up area, including Pawtucket on the 99 and South Providence on the 11. This is a consequence of the fact that transit in Rhode Island is so bad that only the poor use it, and thus the preexisting busy routes serve poor areas; the best physical bus infrastructure is a bus tunnel to College Hill, the richest neighborhood in the city, but ridership there is weak and therefore the routes were never high priorities for further investment.

The improvements to rail never went beyond blogging; we didn’t have the pull of Boston’s TransitMatters, which itself is better at proposing small improvements than big ones that go up against political obstruction. What we called for was frequent local rail within the urban area: Peter Brassard wrote up the initial proposal, and I added some refinements. The Northeast Corridor, where the service would run, is primarily an intercity rail corridor, but there is room for four tracks in the right-of-way, and while there is freight traffic, it runs at the same approximate speed of a local passenger train.

As we discussed this proposal, Greater City’s Jef Nickerson noted something: what the train would do if implemented is produce better transit service in working-class areas than in more comfortable ones. Unlike the situation with the buses, this was not an intentional process. We would like Rhode Island to improve rail service using an existing right-of-way, which happens to serve Central Falls, Pawtucket, Olneyville, Hartford, Cranston, and Warwick, and miss the East Side and the middle-class suburbs. We realized that the city and inner-suburbs like Pawtucket are poorer than the proper suburbs, but that the train would serve Olneyville but not the East Side seemed like a coincidence.

But is it really a coincidence? Providence developed from east to west. The city was initially founded on the western side of what is now the East Side, sloping down to the river. What is now Downcity was only the second part of the city to develop. It became the center of the city because, as the Northeast Corridor was constructed, it was not possible to provide through-service via the hilly historic core of the city, only via the flatter areas that are now Downcity. A tunnel across College Hill opened in 1908, but by then the city’s basic urban geography was set: the university and port jobs on the East Side, industrial jobs to the west near the rail mainline.

The industrial jobs are long gone now. New England was the first part of America to industrialize and the first to deindustrialize, the mills moving to lower-wage Southern states already in the middle of the 20th century. In very large cities, declining industrial jobs can be replaced with urban renewal serving the middle class: the West India Docks became Canary Wharf, the freight railyards of Gare de Lyon became Bercy, the industrial Manhattan and Brooklyn waterfronts became sites for condos with nice views. In Providence-size cities, no such urban renewal is possible: there is no large mass of middle-class people clamoring to live or work in Olneyville, so the neighborhood became impoverished.

While factories may seem like attractive targets for transit commuting, they’re so clustered, in reality they have not been walkable ever since electrification made open-plan single-story factories viable. Factories are land-intensive and have been since around the 1910s. Moreover, whereas hotels and retail have a reason to locate in walkable areas for their consumption amenities – tourists like walking around the city – factories do not, and if anything depress an area’s desirability through noise and pollution. Working industrial districts are not attractive for transit, but post-industrial ones are, even if they are not gentrified the way so much of London, Paris, and New York have.

A large number of cities share Providence’s history as a medium-size post-industrial city. Nearly every English city except London qualifies, as do the cities of the American Northeast and Midwest below the size class of Boston and Philadelphia. Moreover, all of these cities have undergone extensive middle-class flight, with the racial dimension of white flight in the US but even without it in Britain; thus, the relatively dense neighborhoods, where transit service is more viable, are disproportionately poor. However, the feasibility of mainline rail service to post-industrial neighborhoods is uneven, and depends on local idiosyncrasies.

One positive example I’m more familiar with that’s a lot like Providence is in New Haven. Its best potential local rail route, the Farmington Canal Trail, serves lower middle-class areas like Hamden, and fortunately parallels the busiest bus route, the D-Dixwell. While Hamden is not poor, such service would still lead to the inversion we discussed for Providence, since the rich live in thoroughly auto-oriented suburbs or within walking distance of Yale. The main drawbacks are that it would require replacing an active trail with rail service, and that either street running or brief tunneling would be needed in the final few hundred meters in Downtown New Haven.

Vancouver: easy corridors and TOD for the working class

With a modal share of 21%, Vancouver is in a somewhat higher class of transit quality than the Riviera, Boston, or Chicago. However, it remains a far cry from the numbers beginning with a 3, 4, and 5 seen in New York and in European and Asian transit cities. As with the Riviera, I am somewhat speculating from my own observations, lacking a table that clearly states transit usage by socioeconomic class. However, two factors make me believe that transit in Vancouver serves the working class better than it does the middle class.

The first factor is the corridors served by SkyTrain. The first to be built, the Expo Line, runs in a preexisting interurban right-of-way, with minor greenfield elevated and underground construction; even the downtown tunnel is repurposed from a disused mainline rail branch. It passes through a mixture of working-class and lower middle-class neighborhoods on its way to Surrey, which is working-class and very negatively stereotyped. The second, the Millennium Line, branches east, to lower middle-class suburbs, running on a greenfield el. The third, the Canada Line, is a partially tunneled, partially elevated route through the middle-class West Side to working-class Richmond. Only the fourth line to be built, the Evergreen extension of the Millennium Line, finally serves a comfortable area, as will the next line, the Broadway extension of the Millennium Line deeper into the West Side.

The second factor is the job distribution within Metro Vancouver. Usually, we see concentration of professional jobs in city centers and dispersal of working-class jobs among many stores. In the Riviera this relationship between job concentration and income is only inverted because the working-class jobs are disproportionately in tourism while the professional ones are in an edge city. In Vancouver I don’t believe there is any such inversion, but there is leveling: jobs of either type are concentrated in transit-rich areas. This leveling is the result of extensive commercial transit-oriented development, most notably Metrotown, which has many office jobs on top of Canada’s third largest shopping mall.

The first factor is idiosyncratic. The easy corridors happened to serve poorer areas, on a line from East Vancouver to Surrey. The rich live in North Vancouver, which has a ferry and doesn’t have enough population density for a SkyTrain tunnel; on the West Side, which is separated from downtown by False Creek and was thus late to get a rail connection; and in Port Moody and Coquitlam, which were only connected to SkyTrain recently via the Evergreen extension.

The second factor is more systemic. While American and European cities rarely have big urban shopping malls, Canadian cities are full of them. The Metropolis at Metrotown has 27 million annual visitors, not far behind the 37 million of the Forum des Halles, at the center of a metro area five times the size of Metro Vancouver – and the Metropolis has more than twice the total commercial floor area. In this, Canada is similar to Israel and Singapore, which like Canada have harsh climates, only hot instead of cold. Moreover, Vancouver has encouraged this centralization through TOD: Burnaby built Metrotown from scratch in the 1980s, simultaneously with the Expo Line.

It is difficult to engage in concerted residential TOD for the working class, since it requires extensive housing subsidies. Vancouver’s residential TOD near SkyTrain stations is thoroughly middle-class. However, concerted commercial TOD is easier: hospitals, universities, and shopping centers all employ armies of unskilled workers (the first two also employing many professional ones), the first two while satisfying general social goals for health care and education provision and the last while making the owners a profit on the open market.

Moreover, Vancouver’s TOD within downtown, too, has made it easier to provide transit service for the working and lower middle classes. Where constraints on office towers lead to high office rents, only the most critical jobs are in city centers, and those are typically high-end ones; in the US, it’s common for big corporations to site their top jobs in the center of New York or Chicago or another large city but outsource lower-end office jobs to cheaper cities. In Vancouver, as elsewhere in Canada, extensive downtown commercialization means that even semi-skilled office jobs like tech support can stay in the center rather than at suburban office parks.


Based on my own observations, I believe the Riviera provides better public transportation for the working class than for the middle class, and to some extent so does Vancouver. Providence provides uniformly poor transit service, but its lowest-hanging fruit are in working-class urban neighborhoods.

The reasons vary, but the unifying theme is that, in the Riviera and Vancouver, there is none of the typical big-city pattern in which the rich work in walkable city centers more than the poor (e.g. in New York). In Vancouver it’s the result of commercial TOD as well as a Canadian culture of urban shopping centers; in the Riviera it’s the result of unique dependence on tourism. In Providence the situation is not about job concentration but about residential concentration: lower-income neighborhoods are likelier to arise near rail because historically that’s where industry arose, and all that remains is for Providence to actually run local passenger trains on the mainline.

It is not possible to replicate culture. If your city does not have the tourism dependence of Monaco, or the shopping mall culture of Vancouver, or the post-industrial history of Providence, there’s little it can do to encourage better urban geography for working-class transit use. At best, can build up more office space in the center, as Vancouver did, and hope that this encourages firms to locate their entire operations there rather than splitting them between a high-end head office and lower-end outlying ones. Fortunately, there exist many cities that do have the special factors of the Riviera, Vancouver, or Providence. In such cities, transit planners should make note of how they can use existing urban geography to help improve transit service for the population that most depends on it.

The Red-Blue Connector and the Importance of Connectivity

The Boston rapid transit network has the shape of the hex symbol, #. In Downtown Boston, the two north-south legs are the Green Line on the west and the Orange Line on the east, and the two east-west legs are the Red Line on the south and the Blue Line on the north. The Orange and Green Lines meet farther north, but the Red and Blue Lines do not. The main impact of this gap in systemwide connectivity is that it’s really hard to get between areas only served by the Blue Line, i.e. East Boston, and ones only served by the Red Line, i.e. Cambridge, Dorchester, and Quincy. However, there is a second impact: people who do transfer between the Red and Blue Lines overload one central transfer point at Park Street, where the Red and Green Lines meet. This way, the weak connectivity of the Boston rapid transit network creates crowding at the center even though none of the individual lines is particularly crowded in the center. The topic of this post is then how crowding at transfer points can result from poor systemwide connectivity.

The current situation in Boston

Connecting between the Blue and Red Lines requires a three-seat ride, with a single-stop leg on either the Orange or Green Line. In practice, passengers mostly use the Green Line, because the Orange Line has longer transfer corridors.

Travel volumes between East Boston and Cambridge are small. Only 1,800 people commute from East Boston, Winthrop, and the parts of Revere near the Blue Line to Cambridge, and only 500 commute in the other direction. I don’t have data on non-work travel, but anecdotally, none of the scores of Cantabrigians I know travels to the Blue Line’s service area except the airport, and to the airport they drive or take the Silver Line, and moreover, only two people moved from Cambridge or Somerville to the area, a couple that subsequently left the region for Bellingham. Travel volumes between East Boston and the southern legs of the Red Line are barely larger: 1,200 from East Boston to Dorchester, Mattapan, and Quincy, 1,600 in the other direction, most likely not taking public transit since cars are a good option using the Big Dig.

Nonetheless, this small travel volume, together with connections between East Boston and South Station or Dorchester, is funneled through Park Street. According to the 2014 Blue Book, which relies in 2012 data, transfer volumes at Park Street are 29,000 in each direction (PDF-p. 16), ahead of the Red/Orange connection at Downtown Crossing, where 25,000 people transfer in each direction every weekday. Riders connecting between the Blue and Red Lines are a noticeable proportion of this volume – the East Boston-Cambridge connection, where I believe the transit mode share is high, is around 8% of the total, and then the East Boston-Dorchester connection would add a few more percentage points.

Why Soviet triangles exist

In a number of metro networks, especially ones built in the communist bloc, there are three lines meeting in a triangle, without a central transfer point. This is almost true of the first three subway lines in Boston, omitting the Red Line: they meet in a triangle, but the Green and Orange Lines do not cross, whereas in true Soviet triangles lines meet and cross.

The reason for this typology rather than for the less common one in which all three lines meet at one station, as in Stockholm, is that it spreads transfer loads. Stockholm’s transfer point, T-Centralen, has 184,000 daily boardings (source, PDF-p. 13), almost as many as Times Square, which is served by 14 inbound tracks to T-Centralen’s 5 and is in a city with 5.6 million weekday trips to Stockholm’s 1.1 million. Urban transit networks should avoid such situations, which lead to central crowding that is very difficult to alleviate. Adding pedestrian circulation is always possible, but is more expensive at a multilevel central station than at a simple two-line crossing.

The triangle is just a convenient way of building three lines. As the number of lines grows beyond three, more connectivity is needed. Moscow’s fourth line, Line 5, is a circle, constructed explicitly to decongest the central transfer station between the first three lines. More commonly, additional lines are radials, especially in cities with water constraints that make circles difficult, like Boston and New York; but those should meet all the older radii, ideally away from existing transfer stations in order to reduce congestion. When they miss connections, either by crossing without interchange or by not crossing at all, they instead funnel more cross-city traffic through the existing transfer points, increasing ridership without increasing the capacity required to absorb it.

The way out

The situation is usually hard to fix. It’s much harder to fix missed connections, or parallel lines that diverge in both directions, than to connect two parallel lines when one of these lines terminates in city center, which Boston’s Blue Line does. The one saving grace is that cities with many missed connections, led by New York and Tokyo, also have very expansive networks with so many transfer points that individual interchanges do not become overloaded.

In large cities that do have problems with overcrowded transfer points, including London and Paris, the solution is to keep building out the network with many connections. London tries to weaken the network by reducing transfer opportunities: thus, Crossrail has no connection to Oxford Circus, the single busiest non-mainline Underground station, in order to prevent it from becoming any more crowded, and the Battersea extension of the Northern line deliberately misses a connection to the Victoria line. Paris has a better solution – it invests in circumferential transit, in the form of Metro Line 15 ringing the city at close distance, as well as extensions to Tramway Line 3, just inside city limits.

While the solution always involves investing more in the transit network, its precise nature depends on the city’s peculiar geography. In Paris, a compact city on a narrow river, adding more circles is an option, as is adding more RER lines so that people would be able to avoid difficult Metro-to-RER transfers. In London, the population density is too low and the construction costs are too high for a greenfield circle; the existing circle, the Overground, is cobbled together from freight bypasses and is replete with missed radial Underground connections. Thus, the solution in London has to come from radials that offer alternatives to the congestion of the Victoria line.

In Boston, a much smaller city, the Red-Blue Connector is easier since the Red and Blue Lines almost touch. It only takes about 600 meters of cut-and-cover tunnel under a wide road to continue the Blue Line beyond its current terminus in Downtown Boston and meet the Red Line at Charles-MGH; to first order, it should cost not much more than $100 million. The transfer would not be easy, since the Red Line is elevated there and the Blue Line would be underground, but it would still be better than the three-seat ride involving the Green Line. A competent state government with interest in improving transportation connectivity for its residents – that is, a government that is nothing like the one Massachusetts has – would fix this problem within a few years. Boston is fortunate in not needing painful deep tunneling under a medieval city center like London or hundreds of kilometers of inner suburban tunneling like Paris – it only needs to kick out the political bums, unfortunately a much harder task.

What Does “On Demand” Mean, Anyway?

One of the tech industry’s buzzwords for transportation is “on demand” – that is, available to the passenger immediately, without fixed schedules. When I said something about schedules at my Hyperloop One interview, the interviewers gently told me that actually, they intend their system to be on-demand; I forget what I said afterward, but I do remember I didn’t press the point. More commonly, people who insist on using ride-hailing apps rather than public transit talk about how great it is that they don’t have to follow fixed schedules.

But what does this really mean? Calling a cab, or hailing one via a TNC app, does not mean it comes immediately. There’s a wait time of several minutes. How many minutes depends on time of day and which city one is in. A Dallas air travel blog describes wait times of around 10 minutes. In and around Boston, Patreon supporter Alexander Rapp says “2-7 minutes is the typical range” with New York waits slightly longer. In Los Angeles, a dissertation studying racial bias finds that the average predicted wait times are 6-7 minutes for black people and 5-6 minutes for others (PDF-p. 147); both the absolute numbers and the difference are much higher for street-hailed cabs. In 2014, the median wait time in New York was 3 minutes.

On an urban transit line, an average wait of 10 minutes is equivalent to a 20-minute frequency, and an average wait of 5 minutes is equivalent to a 10-minute frequency. At the lower end, the fixed schedule is actually better – a well-run transit system with 20-minute frequencies publicly posts clockface schedules and sticks to them, so people know in advance how to time themselves to the bus or train’s arrival time.

Even the wait times of 2014’s New York, not since achieved in the city or elsewhere, are only equivalent to a subway train that comes every 6 minutes, which is decidedly mediocre. The outer subway branches in New York get a train every 8-10 minutes off-peak, but they are not what TNCs compete with. Bruce Schaller’s report alleging that TNCs are responsible for the decline in subway ridership uses data from mid-2016, when 56% of TNC trips in New York were in Manhattan south of West 110th and East 96th Streets and another 22% were in inner-ring neighborhoods, mostly before the subway branch points. And subway frequency in New York is not good for how busy the system is; during the daytime, longer headways than 5 minutes are rare on the Paris Metro and on the trunks in London and Stockholm. Milan, hardly a transit city, runs its driverless metro line every 4 minutes off-peak.

All of the comparable waits get longer outside the city. Stockholm’s highly-branched metro system runs every 10 minutes on some branches off-peak, and Tube waits go up to 10 minutes on some branches as well. Commuter rail waits start from 10 minutes on the busiest branches, like those of the RER A, and go up from there, sometimes even to a train every half hour off-peak in suburbs of respectable European transit cities. But the branches are not where people ride TNCs. Just as in New York the vast majority of TNC trips are downstream of the branch points, in London as of 2013, 74% of taxi trips were within Inner London; if New York’s subsequent evolution is any indication, TNC traffic is somewhat less dominated by the center, but has only differed from street-hailed cab traffic patterns in degree rather than kind.

This calculation does not mean that transit is better than TNCs on out-of-vehicle times. It is not. Walk times to stations are considerable. Trips that require transfers have extra wait time. In New York, there appear to be about 1.6 unlinked trips per linked trip, but most likely multiple-seat rides have shorter waits on average, because they include local-express transfers, which passengers make preferentially if the waits are short. In London, judging by the origin-destination matrix, 61% of trips do not require any interchange, and another 34% require just one. So even with transfers, frequent subways are still a little bit ahead, but then the walk time to the station makes a big difference in favor of TNCs.

But here’s the thing: tech workers who talk about the greatness of on-demand transportation do not talk about station access time. Evidently, Hyperloop One, which has to use stations, talks about on-demand service. The company did try to think about how to branch in the cities in order to reduce station access time, but reduce does not mean eliminate. Moreover, the same kind of branching is already available to trains and even more so to intercity buses, and yet they rarely make use of it: intercity buses do not make milk runs within cities, leading to awkward situations in which a person in Upper Manhattan traveling to Boston has to take the subway to Midtown and then get on a bus that slogs through Manhattan streets going toward Upper Manhattan on its way to the freeways to Boston.

So what’s going on here? There’s a legitimate advantage to cars over transit in that they don’t require you to travel to a subway station or transfer, but that’s not the argument that opponents of transit who talk about TNCs and app-hailed services and on-demand travel make. They talk about wait times, never mind that well-run urban transit offers shorter wait times than app-hailed TNCs.

My suspicion is that this involves business culture. Urban transit is extremely Fordist: it has interchangeable vehicles and workers, relentlessly regular schedules, and central allocation of resources based on network effects. The tech industry has corners that work like this as well, like Amazon’s monitoring its warehouse workers’ bathroom breaks, but for the most part the industry comes from a post-Fordist world. The idea that there should be people writing down precise schedules for service is alien, as is any coordinated planning; order should be emergent, and if it doesn’t work at the scale of a startup, then it’s not worth pursuing.

There have been positive examples of using better software technology to improve public transportation. The Internet itself has been amazing at improving access to information; the single most important technology for transit reform in lagging regions like North America is Google search, followed by Wikipedia, and even in places with healthy transit these tools are valuable. Within schedule planning, new software tools make it easier to track delays. Tech is a tool, and as such it has been very useful for transit, as for many other industries.

However, all of this occurs within the usual culture of transportation planning. In contrast with this culture, most companies that produce software use a culture of startups, which have to work at a small scale to get anywhere. Where network effects are required, as with social media, it’s necessary to find a small, high-prestige network of early adopters, e.g. Harvard students for Facebook. Anything that requires more initial capital than a VC is willing to risk on a single firm is out; thus Hyperloop One views itself as a consultancy developing a technology rather than as a railroad actually building its own Hyperloop infrastructure.

A corollary of this is that people within the tech industry dismiss schedules out of hand. Thus they insist that transportation be on-demand, even when in practice the wait is longer than on a competing mode of travel that is scheduled. The idea of on-demand travel is reassuring, and because Swiss scheduling precision is alien to the American tech entrepreneur, it’s not a big deal if on-demand means a promised 5-minute wait and an actual 10-minute wait.

But what reassures the tech entrepreneur does not reassure the average rider. By overwhelming numbers, people who have a choice between even mediocre public transportation and TNCs slog through 9-minute bus and train frequencies; people who have access to good public transportation keep taking it where available. In New York, where transit isn’t even that great by the standard of large European cities, there is an ongoing panic about a 2% decline in annual subway ridership, which Schaller wrongly attributes to TNCs rather than to internal decline in subway service quality. Ultimately, the experience of waiting a few minutes for a train is annoying and passengers try to avoid it, but over time they don’t find it any more annoying than the experience of waiting for the app-hailed car driver to show up. Rhetoric about on-demand service aside, passengers do notice how long they’re actually waiting.

Difficult Urban Geography Part 1: Narrow Streets

I did a complex Patreon poll about series to write about. People voted for general transit network design, and more posts about national traditions of transit in the mold of the one about the US. Then I polled options for transit network design. There were six options, and people could vote up or down on any. Difficult urban geography was by far the most wanted, and three more alternatives hover at the 50% mark. To give the winning option its due course, I’m making it a mini-series of its own.

There are cities that, due to their street layout, make it easy to run transit on them. Maybe they are flat and have rivers that are easy to bridge or tunnel under. Maybe they have a wealth of wide arterials serving the center, with major cross streets at exactly the right places for stations and an underlying bus grid. Maybe they spread out evenly from the center so that it’s easy to run symmetric lines. Maybe their legacy mainline rail network is such that it’s easy to run interpolating buses and urban rail lines.

And then there are cities that are the exact opposite. In this post I’m going to focus on narrow or winding streets and what they mean for both surface and rapid transit. The good fortune for transit planners is that the city that invented urban rapid transit, London, is a prime example of difficult urban geography, so railway engineers have had to deal with this question for about 150 years, inventing some of the necessary technology in the process.

Rapid transit with narrow streets

The easiest ways to build rapid transit are to put it on a viaduct and to bury it using cut-and-cover tunneling. Both have a minimum street width for the right-of-way – an el requires about 10 meters, but will permanently darken the street if it is not much wider, and a tunnel requires about 10 meters for the tracks but closer to 18 for the stations.

Nonetheless, even cities with narrow streets tend to have enough streets of the required width. What they don’t always have is streets of the required width that are straight and form coherent spines. The labyrinth that is Central London does have wide enough streets for cut-and-cover, but they are not continuous and often miss key destinations such as major train stations. The Metropolitan line could tunnel under Euston Road, but the road’s natural continuation into the City is not so wide, forcing the line to carve a trench into Farringdon. Likewise, the District line could tunnel under Brompton Road or King’s Road, but serving Victoria and then Westminster would have required some sharp curves, so the District Railway carved a right-of-way, demolishing expensive Kensington buildings at great expense.

While London is the ur-example, as the city that invented the subway, this situation is common in other cities with large premodern cores, such as Rome, Milan, and Istanbul. Paris only avoided this problem because of Haussmann’s destruction of much of the historic city, carving new boulevards for aesthetics and sewer installation, which bequeathed the Third Republic a capital rich in wide streets for Metro construction.

Dealing with this problem requires one of several solutions, none great:

Deep-bore tunneling

London’s solution was to invent the tunnel boring machine to dig deep Tube lines, avoiding surface street disruption. With electric-powered trains and reliable enough TBMs to bore holes without cave-ins, London opened the Northern line in 1890, crossing the Thames to provide rapid transit service to South London. Subsequently, London has built nearly all Underground lines bored, even in suburban areas where it could have used cut-and-cover.

The main advantage to TBMs is that they avoid surface disruption entirely. Most first-world cities use them to bore tunnels between stations, only building stations cut-and-cover. The problem is that TBMs are more expensive to use than cut-and-cover today. While turn-of-the-century London built Tube lines for about the same cost per km as the Metropolitan line and as the cut-and-cover Paris Metro and New York City Subway, in the last half century or so the cost of boring has risen faster than that of shallow construction.

The worst is when the stations have to be mined as well. Mining stations has led to cost blowouts in New York (where it was gone gratuitously) and on London Crossrail (where it is unavoidable as the tunnel passes under the older Underground network). A city that cannot use cut-and-cover tunnels needs to figure out station locations that are easily accessible for vertical digging.

The alternative is the large-diameter TBM. Barcelona is using this technique for Line 9/10, which passes under the older lines; the city has a grid of wide boulevards, but the line would still have to pass under the older metro network, forcing the most difficult parts to be deep underground. The large-diameter TBM reduces the extent of construction outside the TBM to just an elevator bank, which can be dug in a separate vertical TBM; if higher capacity is desired, it’s harder but still possible to dig slant bores for escalators. The problem is that this raises construction costs, making it a least bad solution rather than a good one; Barcelona L9, cheap by most global standards, is still expensive by Spanish ones.

Carving new streets

Before the 1880s, London could not bore the Underground, because the steam-powered trains would need to be close to surface for ventilation. Both the Metropolitan and District lines required carving new right-of-way when streets did not exist; arguably, the entire District line was built this way, as its inner segment was built simultaneously with the Victoria Embankment, under which it runs. The same issue happened in New York in the 1910s and again in the 1920s: while most of the city is replete with straight, wide throughfares, Greenwich Village is not, which forced the 1/2/3 to carve what is now Seventh Avenue South and later the A/C/E to carve the southern portions of Sixth Avenue.

This solution is useful mostly when there are wide streets with absolutely nothing between them that a subway could use. The reason is that demolishing buildings is expensive, except in very poor or peripheral areas, and usually rapid transit has to run to a CBD to be viable. If the entire route is hard to dig, a TBM is a better solution, but if there are brief narrows, carving new streets New York did could be useful, especially if paired with improvements in surface transit.

Looking for station sites

Milan built its first metro lines cut-and-cover. However, lacking wide streets, it had to modify the method for use in a constrained environment. Instead of digging the entire street at a sloped angle and only then adding retaining walls, Milan had to dig the retaining walls first, allowing it to dig up streets not much wider than two tracks side by side. This method proved inexpensive: if I understand this article right, the cost was 30 billion lire in 1957-1964 prices, which is €423 million in 2018 prices, or €35 million per km. Milan’s subsequent construction costs have remained low, even with the use of a TBM for Metro Line 5.

The problem with this method is that, while it permits digging tunnels under narrow medieval streets, it does not permit digging stations under the same streets. Milan is fortunate that its historical center is rich in piazzas, which offer space for bigger digs. One can check on a satellite map that every station on Lines 1 and 2 in city center is at a piazza or under a wide street segment; lacking the same access to easy station sites, Line 3 had to be built deeper, with tracks stacked one under the other to save space.

I have argued in comments that Paris could have used this trick of looking for less constrained sites for stations when it built Metro Line 1, permitting four tracks as in New York as long as the express stations under Rue de Rivoli stuck to major squares like Chatelet. However, Paris, too, is rich in squares, it just happens to be equally rich in wide streets so that it did not need to use the Milan method. London is not so fortunate – its only equivalent of Milan’s piazzas is small gardens away from major streets. It could never have built the Central line using the Milan method, and even the Piccadilly line, which partly passes under wide streets, would have been doubtful.

Surface transit

Rapid transit benefits from being able to modify the shape of the street network to suit its needs. Surface transit in theory could do the same, running in short tunnels or widening streets as necessary, but the value of surface modes is not enough to justify the capital expense and disruption. Thus, planners must take the street network as it is given. The ideal surface transit route runs in the street median on two dedicated lanes, with boarding islands at stops; creating a parking lane, a moving lane, and a transit lane in each direction on a street plus some allowance for sidewalks requires about 30 meters of street width or not much less. Below 25, compromises are unavoidable.

Cutting car lanes

A lane is about 3 meters wide, so removing the parking lanes reduces the minimum required street width by about 6 meters. Contraflow lanes instead allow the street to have the same four lanes, but with a moving lane and a parking lane in one direction only. In extreme cases it’s possible to get rid of the cars entirely; a transit mall is viable down to maybe 12-15 meters of street width. The problem is that deliveries get complicated if the city doesn’t have alleyways or good side street access, and this may force compromises on hours of service (perhaps transit doesn’t get dedicated lanes all day) or at least one parking lane in one direction.

Single-track streetcars

Some city cores with very narrow streets don’t have double-track streetcars. A few have one-way pairs, but more common is single-track segments, or segments with two overlapping tracks so that no switching is needed but trams still can’t pass each other. Needless to say, single-tracking is only viable over short narrows between wider streets, and only when the network is punctual enough that trams can be scheduled not to conflict.

On longer stretches without enough room for two tracks or two lanes, one-way pairs are unavoidable; these complicate the network, and unless the streets the two directions of the bus or tram run on are very close to each other they also complicate interchanges between routes. New York has many one-way pairs on its bus network, even on wide and medium-width streets in order to improve the flow of car traffic, and as a result, some crosstown routes, such as the B35 on Church, are forced to stop every 250 meters even when running limited-stop. While New York’s network complexity is the result of bad priorities and can be reversed, cities with premodern street networks may not even have consistent one-way pairs with two parallel streets on a grid; New York itself has such a network in Lower Manhattan.

Bus network redesign

The best way to avoid the pain associated with running buses on streets that are not designed for fast all-mode travel is not to run buses on such streets. Boston has very little surface transit in city center, making passengers transfer to the subway. In Barcelona, part of the impetus for Nova Xarxa was removing buses from the historic core with its narrow streets and traffic congestion and instead running them on the grid of the Eixample, where they would not only provide a frequent system with easy transfers but also run faster than the old radial network.

However, this runs into two snags. First, there must be some radial rapid transit network to make people connect to. Boston and Barcelona both have such networks, but not all cities do; Jerusalem doesn’t (it has light rail but it runs on the surface). And second, while most cities with a mixture of wide and narrow streets confine their narrow streets to premodern historic cores, some cities have streets too narrow for comfortable bus lanes even far out, for example Los Angeles, whose north-south arterials through the Westside are on the narrow side.

What not to do: shared lanes

It’s tempting for a transit agency to compromise on dedicated lanes whenever the street is too narrow to feature them while maintaining sufficient auto access. This is never a good idea, except in outlying areas with little traffic. The reason is that narrow streets fed by wide streets are precisely where there is the most congestion, and thus where the value of dedicated transit lanes is the highest.

In New York, the dedicated bus lanes installed for select bus service have sped up bus traffic by around 30 seconds per kilometer on all routes Eric Goldwyn and I have checked for our Brooklyn bus redesign project, but all of these figures are averaged over long streets. Within a given corridor, the short narrows that the transit agency decides to compromise on may well feature greater time savings from dedicated lanes than the long arterial stretch where it does set up dedicated lanes. This is almost certainly the case for the Silver Line in Boston, which has unenforced dedicated lanes most of the way on Washington Streets but then uses shared lanes through Downtown Boston, where streets are too narrow for dedicated lanes without reducing auto access.

Urban Transit Vs. Commuter Transit

The Geary corridor in San Francisco is a neat model for transit ridership. The Golden Gate Park separates the Richmond District from the Sunset District, so the four east-west buses serving the Richmond – the 38 on Geary itself and the closely parallel 1 California, 31 Balboa, and 5 Fulton – are easy to analyze, without confounding factors coming from polycentric traffic. Altogether, the four routes in all their variations have 114,000 riders per weekday. The 38 and 1 both run frequently – the 1 runs every 5-6 minutes in the weekday off-peak, and the 38 runs every 5 minutes on the rapid and every 8 on the inner local.

I was curious about the connection between development and travel demand, so I went to OnTheMap to check commute volumes. I drew a greater SF CBD outline east of Van Ness and north of the freeway onramp and creek; it has 420,000 jobs (in contrast, a smaller definition of the CBD has only 220,000). Then I looked at how many people commute to that area from due west, defined as the box bounded by Van Ness, Pacific, the parks, and Fell. The answer is 28,000. Another 3,000 commute in the opposite direction.

Put another way: the urban transit system of San Francisco carries about twice as many passengers on the lines connecting the Richmond and Japantown with city center as actually make that commute: 114,000/2*(28,000 + 3,000) is 1.84. This represents an implausible 184% mode share, in a part of the city where a good number of people own and drive cars, and where some in the innermost areas could walk to work. What’s happening is that when the transit system is usable, people take it for more than just their commute trips.

The obvious contrast is with peak-only commuter rail. In trying to estimate the potential ridership of future Boston regional rail, I’ve heavily relied on commute volumes. They’re easier to estimate than overall trip volumes, and I couldn’t fully get out of the mindset of using commuter rail to serve commuters, just in a wider variety of times of day and to a wider variety of destinations.

In Boston, I drew a greater CBD that goes as far south as Ruggles and as far west as Kendall; it has a total of 370,000 jobs. Of those, about 190,000 come from areas served by commuter rail and not the subway or bus trunks, including the southernmost city neighborhoods like Mattapan and Hyde Park, the commuter rail-adjacent parts of Newton, and outer suburbs far from the urban transit system. But MBTA commuter rail ridership is only about 120,000 per weekday. This corresponds to a mode share of 32%.

I tried to calculate mode shares for the MBTA seven years ago, but that post only looked at the town level and excluded commuter rail-served city neighborhoods and the commuter rail-adjacent parts of Newton, which contribute a significant fraction of the total commute volume. Moreover, the post included suburban transit serving the same zones, such as ferries and some express buses; combined, the mode share of these as well as commuter rail ranged from 36% to 50% depending on which suburban wedge we are talking about (36% is the Lowell Line’s shed, 50% is the Providence Line’s shed). Overall, I believe 32% is consistent with that post.

Part of the difference between 32% and 184% is about the tightness of economic integration within a city versus a wider region. The VA Hospital in San Francisco is located in the Outer Richmond; people traveling there for their health care needs use the bus for this non-commuter trip. On a regional level, this never happens – people drive to suburban hospitals or maybe take a suburban bus if they are really poor.

That said, hospital trips alone cannot make such a large difference. There are errand trips that could occur on a wider scale if suburban transit were better. Cities are full of specialty stores that people may travel to over long distances.

For example, take gaming. In Vancouver I happened to live within walking distance of the area gaming store, but during game nights people would come over from Richmond; moreover, the gaming bar was in East Vancouver, and I’d go there for some social events. In Providence I’d go to Pawtucket to the regional gaming store. In the Bay Area, the store I know about is in Berkeley, right on top of the Downtown Berkeley BART station, and I imagine some people take BART there from the rest of the region.

None of this can happen if the region is set up in a way that transit is only useful for commute trips. If the trains only come every hour off-peak, they’re unlikely to get this ridership except in extreme cases. If the station placement is designed around car travel, as is the case for all American commuter lines and some suburban rapid transit (including the tails of BART), then people will just drive all the way unless there’s peak congestion. Only very good urban transit can get this non-work ridership.

In-Motion Charging

While electric cars remain a niche technology, electric buses are surging. Some are battery-electric (this is popular in China, and some North American agencies are also buying into this technology), but in Europe what’s growing is in-motion charging, or IMC. This is a hybrid of a trolleybus and a battery-electric bus (BEB): the bus runs under wire, but has enough battery to operate off-wire for a little while, and in addition has some mechanism to let the bus recharge during the portion of its trip that is electrified.

One vendor, Kiepe, lists recent orders. Esslingen is listed as having 10 km of off-wire capability and Geneva (from 2012) as having 7. Luzern recently bought double-articulated Kiepe buses with 5 km of off-wire range, and Linz bought buses with no range specified but of the same size and battery capacity as Luzern’s. Iveco does not specify what its range is, but says its buses can run on a route that’s 25-40% unwired.

Transit planning should be sensitive to new technology in order to best integrate equipment, infrastructure, and schedule. Usually this triangle is used for rail planning, but there’s every reason to also apply it to buses as appropriate. This has a particular implication to cities that already have large trolleybus networks, like Vancouver, but also to cities that do not. IMC works better in some geographies than others; where it works, it is beneficial for cities to add wire as appropriate for the deployment of IMC buses.

Vancouver: what to do when you’re already wired

Alert reader and blog supporter Alexander Rapp made a map of all trolleybus routes in North America. They run in eight cities: Boston, Philadelphia, Dayton, San Francisco, Seattle, Vancouver, Mexico City, Guadalajara.

Vancouver’s case is the most instructive, because, like other cities in North America, it runs both local and rapid buses on its trunk routes. The locals stop every about 200 meters, the rapids every kilometer. Because conventional trolleybuses cannot overtake other trolleybuses, the rapids run on diesel even on wired routes, including Broadway (99), 4th Avenue (44, 84), and Hastings (95, 160), which are in order the three strongest bus corridors in the area. Broadway has so much ridership that TransLink is beginning to dig a subway under its eastern half; however, the opening of the Broadway subway will not obviate the need for rapid buses, as it will create extreme demand for nonstop buses from the western end of the subway at Arbutus to the western end of the corridor at UBC.

IMC is a promising technology for Vancouver, then, because TransLink can buy such buses and then use their off-wire capability to overtake locals. Moreover, on 4th Avenue the locals and rapids take slightly different routes from the western margin of the city proper to campus center, so IMC can be used to let the 44 and 84 reach UBC on their current route off-wire. UBC has two separate bus loops, one for trolleys and one for diesel buses, and depending on capacity IMC buses could use either.

On Hastings the situation is more delicate. The 95 is not 25-40% unwired, but about 60% unwired – and, moreover, the unwired segment includes a steep mountain climb toward SFU campus. The climb is an attractive target for electrification because of the heavy energy consumption involved in going uphill: at 4 km, not electrifying it would brush up against the limit of Kiepe’s off-wire range, and may well exceed it given the terrain. In contrast, the 5 km in between the existing wire and the hill are mostly flat, affording the bus a good opportunity to use its battery.

Where to add wire

In a city without wires, IMC is the most useful when relatively small electrification projects can impact a large swath of bus routes. This, in turn, is most useful when one trunk splits into many branches. Iveco’s requirement that 60-75% of the route run under wire throws a snag, since it’s much more common to find trunks consisting of a short proportion of each bus route than ones consisting of a majority of route-length. Nonetheless, several instructive examples exist.

In Boston, the buses serving Dorchester, Mattapan, and Roxbury have the opportunity to converge to a single trunk on Washington Street, currently hosting the Silver Line. Some of these buses furthermore run on Warren Street farther south, including the 14, 19, 23, and 28, the latter two ranking among the MBTA’s top bus routes. The area has poor air quality and high rates of asthma, making electrification especially attractive.

Setting up wire on Washington and Warren Streets and running the Silver Live as open BRT, branching to the south, would create a perfect opportunity for IMC. On the 28 the off-wire length would be about 4.5 km each way, at the limit of Kiepe’s capability, and on the 19 and 23 it would be shorter; the 14 would be too long, but is a weaker, less frequent route. If the present-day service pattern is desired, the MBTA could still electrify to the northern terminus of these routes at Ruggles, but it would miss an opportunity to run smoother bus service.

In New York, there are examples of trunk-and-branch bus routes in Brooklyn and Queens. The present-day Brooklyn bus network has a long interlined segment on lower Fulton, carrying not just the B25 on Fulton but also the B26 on Halsey and B52 on Gates, and while Eric Goldwyn’s and my plan eliminates the B25, it keeps the other two. The snag is that the proportion of the system under wire is too short, and the B26 has too long of a tail (but the B52 and B25 don’t). The B26 could get wire near its outer terminal, purposely extended to the bus depot; as bus depots tend to be polluted, wire there is especially useful.

More New York examples are in Queens. Main Street and the Kissena-Parsons corridor, both connecting Flushing with Jamaica, are extremely strong, interlining multiple buses. Electrifying these two routes and letting buses run off-wire on tails to the north, reaching College Point and perhaps the Bronx on the Q44 with additional wiring, would improve service connecting two of Queens’ job centers. Moreover, beyond Jamaica, we see another strong trunk on Brewer Boulevard, and perhaps another on Merrick (interlining with Long Island’s NICE bus).

Finally, Providence has an example of extensive interlining to the north, on North Main and Charles, including various 5x routes (the map is hard to read, but there are several routes just west of the Rapid to the north).

IMC and grids

The examples in New York, Providence, and Boston are, not coincidentally, ungridded. This is because IMC interacts poorly with grids, and it is perhaps not a coincidence that the part of the world where it’s being adopted the most has ungridded street networks. A bus grid involves little to no interlining: there are north-south and east-west arterials, each carrying a bus. The bus networks of Toronto, Chicago, and Los Angeles have too little interlining for IMC to be as cost-effective as in New York or Boston.

In gridded cities, IMC is a solution mainly if there are problematic segments, in either direction. If there’s a historic core where wires would have adverse visual impact, it can be left unwired. If there’s a steep segment with high electricity consumption, it should be wired preferentially, since the cost of electrification does not depend on the street’s gradient.

Overall, this technology can be incorporated into cities’ bus design. Grids are still solid when appropriate, but in ungridded cities, trunks with branches are especially attractive, since a small amount of wire can convert an entire swath of the city into pollution-free bus operation.

New York City Subway Expansion Proposal

I wrote a post proposing disentangling the subway in New York a few months ago. On the same basis, I’ve drawn some extra lines that I think should be built in the event the region can get its construction costs under control:

A higher-resolution image (warning: 52 MB) can be found here. The background image is taken from OpenStreetMap. Python 2.7 code for automatically downloading tiles and pasting them into a single image can be found here. Make sure you get PIL or else the file won’t run; first run, and choose whichever tiles you’d like (the boundaries I used for this image are given in the code as x1, x2, y1, y2), and then run, changing the x1, x2, y1, y2 variables in the code as needed. As a warning, pasting images together makes them much bigger – the sum of the individual tiles I used is 15 MB but pasted together they became 46 MB.


Local stations are denoted by black circles, express stations by bigger circles with white filling. On four-track lines and three-track lines with peak-direction express trains (that is, the 2, 6, and D in the Bronx and the 7 in Queens), the local/express designation is straightforward. Two-track tails are denoted as all local; for the most part the trains continue as express on the three- or four-track lines, but on the Brighton Line the expresses keep turning at Brighton Beach while the locals are the trains that go into Coney Island. On a few two-track segments stations are denotes as express and not local, for example the 2 in Harlem or the A in Lower Manhattan and Downtown Brooklyn: this occurs when a two-track line turns into a three- or four-track line farther out, so that people don’t get the impression that these are local-only stations that the express trains skip.

The local and express patterns are barely changed from today. On Eastern Parkway trains run local east of Franklin Avenue, without skipping Nostrand and Kingston-Throop as the 4 does today. Skip-stop on the J train is eliminated, as is express-running between Myrtle and Marcy Avenues. On Queens Boulevard and Central Park West, the trains serving Sixth Avenue (i.e. the orange ones) run express and the ones serving Eighth (i.e. the blue ones) run local, but I’m willing to change my mind on at least one of these two designations; on Queens Boulevard, 36th Street is also turned into an express station, so that passengers can transfer to 63rd or 53rd Street.

As far as possible, I’ve tried to be clear about which stations are connected and which aren’t. The rule is that circles that touch or are connected by a black line denote transfer stations. However, in the lower-resolution version it may hinge on a single pixel’s worth of separation in Downtown Manhattan. The only new interchanges in Downtown Manhattan connect the 1 with PATH in the Village and at World Trade Center (and the latter connection also connects to the R, E, and 2/3).

No existing subway station is slated for closure. If an existing subway station is missing a circle, it’s an error on my part. Edit: I found one mistaken deletion – the 9th Street PATH station (which should be connected with West 4th, but the West 4th circle doesn’t touch PATH).

New lines

Most of this map should be familiar to people who have followed discussions among railfans in New York (and not just myself) about the next priorities after Second Avenue Subway. Utica and Nostrand are there, with stops that match nearly all of the east-west buses. Northern Boulevard, which Yonah Freemark pointed is a denser corridor than Utica, is also there. Triboro RX is there: the route through the Bronx includes a little more tunneling to connect with the 2 train better, forced by incursions onto the right-of-way farther north. LaGuardia gets an elevated extension of the N, which I’ve periodically argued is superior to other alignments and sound in its own right. Second Avenue Subway continues west under 125th Street, providing crosstown service on a street where buses are very busy despite being slower than walking.

In New Jersey, a hefty proportion of the lines already exist, as part of PATH or the Hudson-Bergen Light Rail. PATH is completely dismembered in this proposal: the line from Newark to World Trade Center is connected with the 6 train, an idea that I don’t think is a top priority but that some area advocates (such as IRUM) have proposed; most of the rest is turned into a 7 extension and connected with the two southern HBLR branches, both of which are extended, one to Staten Island and one to Newark; what remains is reduced to a shuttle from Hoboken to Sixth Avenue. Note that the 6-PATH train also gets an infill stop at Manhattan Transfer for regional rail connections.

The other extensions come from a number of different places:

  • The 6 is extended to Co-op City, the 7 is extended to College Point, and the 1 to the edge of the city. The first two are big ridership generators, and all three also extend lines beyond their bumper tracks, increasing turnback capacity.
  • The Queens Boulevard express trains branch in Jamaica, as they do today, and both branches are extended to near city limits. The southern extension also increases turnback capacity (some E trains run to Jamaica-179th and not Jamaica Center today for this reason), but the primary purpose is to improve coverage to areas of the city that are already at worst missing middle density and redevelopable as mid-rise apartment blocks, and have very long commutes today.
  • The 1 is extended to Red Hook. This was proposed by AECOM a few years ago; my alignment differs somewhat in that it doesn’t connect Red Hook with the subway within Brooklyn, but does connect it directly with South Brooklyn, where in the event of such a subway extension a high-frequency bus (the B71) could run onward.
  • Instead of the periodically mooted 7 extension to Secaucus, the L is extended there, with a four-track tunnel under the Hudson providing for easy 7/L transfers.
  • There’s a preexisting bellmouth for connecting the C train to New Jersey across the George Washington Bridge; it is activated in this plan, with an extension to Paterson elevated over Route 4, with tunneling within Paterson itself. Route 4 is a freeway, but it’s flanked by shopping centers in Paramus, has good regional rail connections and good potential connections if the Northern Branch and West Shore Line are reactivated, and terminates in a dense working-class city.
  • The old Erie Main Line gets converted to subway operations, running elevated through the built-up area of Secaucus.
  • To connect some of the new lines to one another, two new Manhattan trunk lines, both two-track, are built: under 50th Street, and under Third Avenue, the latter substituting for phases 3 and 4 of Second Avenue Subway in order to avoid reverse-branching. Third then connects to the northern reaches of Eighth Avenue Line via a super-express line, with new stations at 110th and 125th; the alignment through Central Park is designed to allow cheap cut-and-cover construction.
  • Bergenline Avenue, where traffic fills a bus every 2 minutes, gets a subway. One station is designed for a commuter rail transfer to new Hudson tunnels with a Bergenline stop. The segment south of Journal Square is weaker and can be removed from scope, but as it can be done in an existing above-ground right-of-way, it’s also cheaper than the rest.
  • The D train gets a two-stop extension to the north to connect to Metro-North at Williams Bridge and the 2 train at Gun Hill Road.

Conspicuous absences

There is no subway connection to JFK or Newark Airport on this map. The JFK AirTrain is adequate with better regional rail and fare integration; so is a Newark connection at the current commuter rail station. A direct JFK regional rail connection may be included in a 9-line regional rail map (for reference, the map I usually peddle has 5 or 6 trunk lines, not 9). A Newark rapid transit connection may be included in a much more expansive version, but even then it’s unlikely – the only reason to build such a connection is for extra capacity, and it’s better to resolve mainline rail capacity crunches by building more mainline rail.

There is no R train to Staten Island, an extension that some railfans (including myself many years ago) periodically call for; this could be added, but is a low priority, as regional rail could provide faster service to Downtown Brooklyn with a transfer than the R train ever could.

But the biggest absence is Second Avenue Subway phases 3 and 4. Phase 3 is replaced with a subway under Third Avenue, and phase 4 is omitted entirely. The reason for this omission is, as mentioned above, to avoid reverse-branching, and permit the new system to consist of separate lines without track-sharing, which is more reliable than today’s heavily interlined system.

Phase 4 is also difficult and not all that useful. Lower Manhattan construction is sometimes necessary but should be avoided when it isn’t, as the area has narrow rights-of-way, complex underground station footprints, and archeology going back to the 17th century. There is no capacity crunch heading to Lower Manhattan – southbound trains unload in Midtown in the morning peak – and the area is so small and has so many subways that there is no coverage gap that Second Avenue Subway would fill. Even phase 3 mostly duplicates the Lexington Avenue Line, but serves a large and growing business district in East Midtown where trains do have a capacity crunch, hence the Third Avenue subway.

Scope and costs

The map has around 110 km of new subway and 100 km of new els and other open-air lines (such as the Triboro and Erie rights-of-way). Some of the subways can be built cut-and-cover given sufficient political cajoling, including Nostrand, most of Bergenline, parts of Third and Utica, Northern, and the outer Queens extension. But many cannot: there are 6 new river crossings (50th*2, 7, L, Utica, 1), a kilometer of pure pain in connecting the 6 with PATH, another PATH pain involving a new Exchange Place dig for platforms for the 7, and some new stations that have to be mined (e.g. 50th Street).

At what I consider a normal first-world cost, the tunnels would be around $25 billion in last decade’s money, so maybe $30 billion in today’s money, and the els would add around $10 billion. To put things in perspective, the current five-year MTA capital program is spending $33 billion, nearly all of which is routine maintenance. It’s affordable within a decade if the region gets its construction costs under control.

The American Way of Building Rapid Transit

I’ve sporadically discussed how some countries or regions have traditions of how to build rapid transit. For example, in a City Metric article last year I made an off-hand comment about how communist bloc metros, from Europe to North Korea, have widely-spaced stops just like Moscow, while French metros and French-influenced Montreal Metro have short stop spacing just like Paris. I intend to write some posts covering different traditions, starting from one I’ve barely discussed as such: the American one. There are commonalities to how different American cities that build subways choose to do so, usually with notable New York influences, and these in turn affect how American transit activists think about trains.

For the most part, the American tradition of rapid transit should be viewed as one more set of standards, with some aspects that are worth emulating and others that are not. Most of the problems I’ve harped on are a matter of implementation more than a matter of standards. That said, that something is the local tradition does not immediately mean it works, even if on the whole the tradition is not bad. Some of the traditions discussed below definitely increase construction costs or reduce system effectiveness.

The situation in New York

A large majority of American rapid transit ridership, about two thirds, is in New York. The city’s shadow is so long that the systems built in the postwar era, like the Washington Metro and BART, were designed with New York as a reference, whether consciously or not. Only the Boston subway and Chicago L are old enough to avoid its influence – but then their elevated system design still has strong parallels in New York, whether due to direct influence or a common zeitgeist at the end of the 19th century. Thus, the first stop on the train of thought of the American rapid transit tradition must be New York practice.

New York has nine subway main lines. Five are north-south through Manhattan and four-track, three are east-west and two-track, and one avoids Manhattan entirely. Nearly all construction was done cut-and-cover between 1900 and 1940, forcing lines to hew to the street network. As New York has wide, straight streets, a trait shared with practically all American cities, this was not a problem, unlike in London, where carving right-of-way for the Underground was so difficult that every line from the third onward was built deep-bore.

With four tracks on most of the Manhattan trunks, there is local and express service. This allows trains to go around obstacles more easily, increasing redundancy. It’s in this context that New York’s 24/7 service makes sense: there is no absolute need for nighttime maintenance windows in which no train runs. This approach works less well on the two-track lines, and the L, the only one that’s two-track the entire way, has occasional work orders with very low train frequency because of single-tracking.

Outside the core of the city as it was understood during construction, lines run elevated. The standard New York el is an all-steel structure, which reduces construction costs – the First Subway’s subway : el cost ratio was 4:1, whereas today the average is about 2.5:1 even though tunneling uses the more expensive boring technique – at the cost of creating a boombox so noisy that it’s impossible to have a conversation under the tracks while a train is passing. Moreover, splitting the difference between two and four tracks, the standard el has three tracks, which allows peak-direction express service (on the 2/5, 6, and 7) or more space for trains to get around obstacles (on the 1, 4, and N/W).

Because the els are so noisy, the city stopped building them in the 1920s. The lines built in the 1930s were all underground, with the exception of one viaduct over an industrial shipping channel.

Moreover, from the 1930s onward, stations got bigger, with full-length mezzanines (the older stations had no or short mezzanines). Track standards increased, leading to an impressive and expensive array of flying junctions, contrasting with the flat junctions that characterize some older construction like the Chicago L or some foreign examples like much of the London Underground.

Finally, while New York has nine separate subway colors, its number of named lines is far greater. The system comprises several tens of segments called lines, and each route combines different lines, with complex branching and recombination. The infrastructure was never built for discrete lines with transfers between them, but rather for everywhere-to-everywhere one-seat rides, and service choices today reinforce this, with several outer lines reverse-branching to an East Side and a West Side Manhattan trunk.

The desire for 24/7 service

I know of five urban rail networks with 24/7 service. One is the Copenhagen Metro, which is driverless and built with twin bores, making it easy for service to single-track at night for maintenance. The other four are American: the New York City Subway, PATH, PATCO, and the Chicago L. Moreover, the LIRR runs 24/7, which no other commuter rail system I know of does, even ones where an individual outlying station has comparable ridership to the entire LIRR.

The other systems have somewhat of a 24/7 envy. I’ve heard lay users and activists in Washington and the Bay Area complain that the Washington and BART shut down overnight; BART itself feels it has to justify itself to the users on this question. Right now, BART’s decision to temporarily add an hour to the nighttime shutdown window to speed up maintenance is controversial. People are complaining that service is being cut despite increases in funding. In Washington, the more professional activists understand why 24/7 service is unviable, but like BART feel like they have to explain themselves.

Local and express trains

New York is full of four-track mainlines, running both local and express trains. Chicago and Philadelphia have them as well on one line each. The other rapid transit networks in the US don’t, but like 24/7 service desire it. Washington has enough complaints about it that regular reader and Patreon supporter DW Rowlands had to write an article for Greater Greater Washington explaining why it would not be all that useful.

BART is the more interesting case. In any discussion of BART extensions, people bring up the fact that BART can’t skip stops – never mind that its stop spacing is extremely wide owing to its function as suburban rail. The average speed on BART is 57 km/h per the National Transit Database; the RER A, which is the express service here, averages around 50. At BART’s speed, the single longest express segment in New York not crossing water, the A/D between 125th and 59th Streets, would take 7 minutes; in fact it takes about 9. If anything, BART errs in having too few stations in Oakland and San Francisco.

On new-build systems, four tracks are understandable and desirable, provided the construction method is cut-and-cover, as it was in early-20th century America. The earliest subway lines built in New York had little cost premium over London and Paris even though the tunnels were twice as wide for twice as many tracks. However, cut-and-cover is no longer used in developed countries owing to its heavy impact on merchants and residents along the way; already during WW2, Chicago dug the tunnels for the Red and Blue Lines of the L using deep boring. A city that bores tunnels will find that four-track tunnels cost twice as much as two-track tunnels, so it might as well built two separate lines for better coverage.

The shadow of steel els

New York, Boston, Philadelphia, and Chicago all built all-steel els. While cheaper, these structures are so noisy that by the 1930s they became untenable even in far-out neighborhoods, like on the Queens Boulevard Line. New lines in New York were underground; existing els were removed, quickly in New York and more slowly in Boston.

The newer systems built in the US avoided els entirely. BART planned to build one in Berkeley, but community opposition led to a change to an underground alignment; unlike subsequent examples of NIMBYism, Berkeley was willing to pay the cost difference. When tunnels are infeasible due to cost, American rail networks prefer at-grade rights-of-way, especially freeway medians. Rail rights-of-way are popular where available, such as on the realigned Orange Line in Boston, but freeway medians are common where rail alignments don’t exist.

The next generation of American urban rail systems, unable to tunnel in city center, turned to light rail in order to keep things at-grade. Across the border, in Canada, Vancouver built els to cover gaps in the right-of-way that turned into the Expo Line, and then built concrete els on the Millennium Line and outer Canada Line to reinforce the system. These brutalist structures are imposing, but I’ve had conversations under the viaducts in Richmond, just as I have in Paris under the mixed concrete and steel structures or in Sunnyside next to New York’s one concrete el.


New York did not invent reverse-branching. London has had it since the 1860s, when most South London railways ran separate trains to the City (at Cannon Street, London Bridge, or Blackfriars) or the West End (at Victoria or Charing Cross), and multiple North London railways ran trains to their traditional terminals or to the North London Railway for service to Broad Street. Paris has had it since even earlier: the railways operating out of Gare Saint-Lazare and Gare Montparnasse merged in 1851 and treated the two stations as reverse-branches allowing cities farther west to access both the Right Bank and the Left Bank. In both cities, this situation makes it harder to run coherent regional rail – in London the railways are spending considerable resources on disentangling the lines to increase frequency to South London’s many branches, and in Paris the fact that Montparnasse and Saint-Lazare serve similar destinations frustrated plans to connect the two stations with an RER tunnel.

Where New York innovated is in copying this practice on rapid transit, starting with the Dual Contracts era. In Brooklyn, existing as well as new outlying lines could be routed to any number of new crossings to Manhattan; in the Bronx and Eastern Brooklyn, a desire to give branches service to both the West Side and East Side led to reverse-branching even on the numbered lines, which were built from scratch and did not involve older suburban railroads.

Reverse-branching spread across the United States. Boston had it until it removed the Atlantic Avenue El, and even today, railfans occasionally talk about reverse-branching the Red Line along Massachusetts Avenue to Back Bay and Roxbury. Chicago occasionally has it depending on the arrangement of trains on the North Side; today, the Purple and Brown Lines share tracks at rush hour but then go in opposite directions on the Loop. The Broad Street Line in Philadelphia reverse-branches to Chinatown. The Washington Metro has reverse-branches in Virginia, limiting train frequency due to asymmetry at the merge points. BART designed itself to force a three-way wye in Oakland pointing toward San Francisco, Berkeley and Downtown Oakland, and East Oakland on which every pair of destinations has a direct train, or else East Oakland residents would have to change trains to access their own city center – and current plans for a second trans-Bay tube add further reverse-branches instead of using the extra capacity as an opportunity to fix the Oakland junction.

Outside the United States, I know of four reverse-branches on rapid transit that is not historically regional rail: the Delhi Green Line, the Namboku and Mita Lines in Tokyo, the Yurakucho and Fukutoshin Lines also in Tokyo, and the Northern line’s two trunks in London. Of those, the last one is slowly being disentangled: its southern end will be two separate lines once the Battersea extension opens, and its northern end will, severing the line in two, once upgrades to pedestrian circulation are completed at the branch point. Historically Toronto had a three-way wye on the subway, like BART, but it caused so many problems it was discontinued in favor of running two separate lines.

Regional rail

The most prominent feature of American rail networks is not what they do, but what they lack. American (and Canadian, and Chinese) regional rail networks remain unmodernized, run for the exclusive benefit of upper middle-class suburban office workers at the primary CBD. Details differ between cities, but even when management is theoretically part of the same agency as the rapid transit network, as in Boston, New York, and Philadelphia, in practice the commuter railroads are autonomous. There is no hint of fare integration or schedule integration.

This fact influences network design more than anything else, even the low quality of steel els. Service to any destination beyond the dense urban core, which is small outside a handful of relatively dense cities, requires building new rail from scratch. This favors low-cost, low-capacity light rail, often in freeway medians. Smaller cities, unable to afford enough light rail to convince entire counties to tax themselves to build transit, downgrade service one step further and build bus rapid transit, typically treated as a weird hybrid of Latin American busways and European bus lanes.

Does any of this work?

In one word, no. The American tradition of rapid transit clearly doesn’t work – just look at the weak ridership even in old cities like Boston and Philadelphia, whose mode shares compare with medium-size urban regions in the French sunbelt like the Riviera or Toulouse.

Or, more precisely, it doesn’t work in early-21st century America. In the rare occasion an American city manages to round up funding to build a new subway line, I would recommend looking abroad for models of both construction methods and network design. For example, as BART keeps working on designing the second tube, I would strongly advise against new branches on the East Bay – instead, one of the two tubes (old and new) should permanently serve East Oakland, with a new Downtown Oakland transfer station, and the other should serve Berkeley and Concord.

Moreover, the United States owes it to itself to aggressively modernize its mainline passenger rail network. It’s too important to let Amtrak, the LIRR, Metro-North, Metra, and other dinosaurs do what they’ve always done. Toronto’s modernization of GO Transit, named the Toronto RER after the Western world’s premier regional rail network, had wide support among transit planners, but the engineers at GO itself were against it, and Metrolinx had to drag them into the 21st century.

Where the American tradition does work is in contexts that the United States has long left behind. Booming third-world cities direly need rapid transit, and while American construction costs are not to be emulated, the concept of opening up major throughfares, laying four tracks, and covering the system is sound. The mix of underground construction in city center and elevated construction farther out (using concrete structure, not louder steel ones) is sound as well, and is already seeing use in China and India. This is especially useful in cities that have little to no legacy regional rail, in which category India and China do not qualify, but most of the rest of the third world does.

Globalization makes for grand shuffles like this one. Experts in the United States should go to Nigeria, Bangladesh, Pakistan, Colombia, Kenya, Tanzania, Angola, and the Philippines and advise people in these countries’ major cities about how to emulate rapid transit designs from early-20th century America. But in their home country these same experts should instead step aside and let people with experience in the traditions of Japan, South Korea, and the various distinct countries of Western and Central Europe make decisions.

Meme Weeding: Los Angeles Density

If you’re the kind of total nerd that looks up tables on Wikipedia for fun, you may notice a peculiarity: the American built-up area with the highest population density is Los Angeles, followed by the Bay Area and New York. This is not what anyone experiences from even a slight familiarity with the two cities. Some people leave it at that and begin to make “well, actually Los Angeles is dense” arguments; this is especially common among supporters of cars and suburbs, like Randall O’Toole, perhaps because they advocate for positions the urbanist mainstream opposes and enjoy the ability to bring up an unintuitive fact. Others instead try to be more analytic about it and understand how Los Angeles’ higher headline density than New York coexists with its actual auto-centric form.

The answer that the urbanist Internet (blogs, then the Census Bureau, then Twitter) standardized on is that the built-up area of New York has some really low-density outer margins, where auto use is high, but the dense core is larger than that of Los Angeles. Here’s a log graph made by longtime Twitter follower Neil Patel:

New York’s 70th percentile of density (shown as 30 on the graph’s y-axis) is far denser than that of the comparison cities. The term the urbanist blogosphere defaulted to is “weighted density,” which is the average density of census tracts weighted by their population rather than area; see original post by the Austin Contrarian, in 2008.

But one problem remains: Los Angeles is by any metric still dense. Neil’s chart above shows its density curve Lorenz-dominating those of Chicago and Washington, both of which have far higher transit usage. Unfortunately, I haven’t seen too much analysis of why. Jarrett Walker talks about Los Angeles’s polycentrism, comparing it with Paris, and boosting it as a positive for public transit. The reality is the opposite, and it’s worth delving more into it to understand why whatever density Los Angeles has fails to make it have even rudimentary public transit.

Yes, Los Angeles is auto-oriented

The “well, actually Los Angeles is not autopia” line faces a sobering fact: Los Angeles has practically no transit ridership. In this section, I’m going to make some comparisons among American metropolitan statistical areas (MSAs); these exclude many suburbs, including the Inland Empire for Los Angeles and Silicon Valley for San Francisco, but Neil’s graph above excludes them as well, because of how the US defines urbanized areas. In the following table, income refers to median income among people driving alone or taking public transit, and all data is from the 2017 American Community Survey (ACS).

Place Workers Drive share Drive income Transit share Transit income
US 152,802,672 76.4% $38,689 5% $37,530
New York 9,821,147 50% $48,812 31% $44,978
San Francisco 2,371,803 57% $54,923 17.4% $62,500
Washington 3,320,895 66.4% $53,390 12.8% $60,420
Chicago 4,653,591 70% $41,817 12.2% $46,796
Los Angeles 6,434,177 75.4% $39,627 4.8% $21,153

The income numbers are not typos. In San Francisco, Washington, and Chicago, transit users outearn drivers. In New York the incomes are close, and US-wide they are almost even. But in Los Angeles, drivers outearn the few transit users almost 2:1. It’s not because Los Angeles has better transit in poor neighborhoods than in rich ones: this may have been true for a long while, but with the Expo Line open to Santa Monica, the Westside has bare bones coverage just like the rest of the city. Even with the coverage that exists, public transit in Los Angeles is so bad that people only use it if they are desperately poor.

When public transportation is a backstop service for the indigent, ridership doesn’t follow the same trends seen elsewhere. Transit ridership in Los Angeles rises and falls based on fares; new rail extensions, which have led to big gains in ridership in Seattle and Vancouver, are swamped by the impact of fare changes in Los Angeles. Gentrification, which in New York has steadily raised subway usage in hotspots like Williamsburg and which does the same in San Francisco, has instead (slightly) contributed to falling transit usage in Los Angeles (p. 53).

Job density and CBD job share

Los Angeles has high residential density by American standards – lower than in New York counted properly, but comparable to San Francisco, and higher than Chicago and Washington. However, job density tells a completely different story. New York, Chicago, San Francisco, and Washington all have prominent central business districts. Without a consistent definition of the CBD, I am drawing what look like the peak employment density sites from OnTheMap, all as of 2015:

Place CBD boundaries Area Jobs MSA share Density
New York 33rd, 3rd, 60th, 9th 3.85 825,476 8.4% 214,336
San Francisco Washington, Powell, 5th, Howard, Embarcadero 1.81 224,010 9.4% 123,558
Washington Rock Creek, P, Mass., 7th, Cons., 14th, H 3.26 240,505 7.2% 73,775
Chicago River, Congress, Michigan, Randolph, Columbus 1.61 368,910 7.9% 228,998
Los Angeles US 110, US 101, Alameda, 1st, Main, 7th 2.11 189,767 2.9% 89,937

The two main indicators to look for are the rightmost two columns: the percentage of jobs that are in the CBD, and the job density within the CBD. These indicators are highly not robust to changing the CBD’s definition, but expanding the definition moves them in opposite direction. Washington and San Francisco can be boosted to about 400,000 jobs each if the CBD is expanded to include near-CBD job centers such as Gallery Place, L’Enfant Plaza, SoMa, and Civic Center. Manhattan south of 60th has 1.9 million jobs in 22.2 km^2. Even in Chicago, where job density craters outside the Loop, the 9 km^2 bounded by Chicago, Halsted, and Roosevelt have 567,000 jobs. In making the tradeoff between job density and MSA share, I tried to use smaller CBD definitions, maximizing density at the expense of MSA share.

But even with this choice, the unusually low CBD share in Los Angeles is visible. This is what Jarrett and others mean when they say Los Angeles is polycentric: it is less dominated by its central business district than New York, Chicago, Washington, and San Francisco.

However, the comparisons between Los Angeles and Paris are wildly off-base. I am not including Paris in my above table, because INSEE only reports job numbers at the arrondissement level, and the city’s CBD straddles portions of the 1st, 2nd, 8th, and 9th arrondissements. Those four arrondissements total 405,189 jobs in 8.88 km^2, but in practice few of these jobs are in the outer quartiers, so a large majority of these jobs are in the central 4.64 km^2. The overall job density is then comparable to that of the Los Angeles CBD, but the similarity stops there: CBD employment is 7.1% of the total for Ile-de-France. If there is a US city that’s similar to Paris on the two CBD metrics of density and employment share, it’s Washington, not coincidentally the only big American city with a height-limited city center.

Secondary centers

In all of the American cities I’m comparing in this post except New York, the share of the population using public transit to get to work is not much higher than the share working in the CBD, especially if we add in near-CBD job centers served by public transit like Civic Center and L’Enfant Plaza (and all of the Manhattan core outside Midtown). This is not a coincidence. Outside a few distinguished locations with high job density, it’s easy enough to drive, and hard to take the train (if it even exists) except from one or two directions.

American cities are distinguished from European ones in that their non-CBD employment is likely to be in sprawling office parks and not in dense secondary centers. Paris is polycentric in the sense of having multiple actual centers: La Defense is the most conspicuous outside the CBD, but the city is full of smaller, lower-rise clusters: the Latin Quarter, Bercy, the Asian Quarter, Gare du Nord, the Marais. The 3rd, 4th, 5th, 6th, 7th, 10th, and 12th all have around 20,000-25,000 jobs per square kilometer, not much less than the Upper East Side (which has about 120,000 jobs between 60th and 96th Streets).

A polycentric city needs to have multiple actual centers. Does Los Angeles have such centers? Not really. Century City has 33,000 jobs in about 1.1 km^2. Here is the city’s second downtown, with a job density that only matches that of central Parisian neighborhoods that nobody would mistake with the CBD. The UCLA campus has around 15,000 jobs. Downtown Santa Monica has 24,000 in 2 km^2. El Segundo, which Let’s Go LA plugs as a good site for CBD formation, has 52,000 jobs in 5.2 km^2. Downtown Burbank has about 13,000 in 0.6 km^2. The dropoff in commercial development intensity from the primary CBD is steep in Los Angeles.

What Los Angeles has is not polycentric development. Paris is polycentric. New York is fairly polycentric, with the growth of near-CBD clusters like Long Island City, in addition to older ones like Downtown Brooklyn and Downtown Newark. Los Angeles is just weak-centered.

The structure of density

In his original posts about weighted density from 2008, Chris noted not just the overall weighted density of an American urban area but also the ratio of the weighted to standard density. This ratio is highest in New York, but after New York the highest ratios are in other old industrial cities like Boston and Chicago. This ratio is in stronger correlation with the public transit modal share than weighted density. Much of this fact is driven by the fact that Boston, Chicago, and Philadelphia have high-for-America transit usage and Los Angeles doesn’t, but it still suggests that there is something there regarding the structure of density.

In Chicago and Washington, the population density is low, but it follows a certain structure, with higher density in central areas and in distinguished zones near train stations. These structures are not identical. Chicago has fairly uniform density within each city neighborhood, and only sees this structure in the suburbs, which are oriented around commuter rail stations, where people take Metra to the city at rush hour (and drive for all other purposes). In contrast, in Washington commuter rail is barely a footnote, whereas Metro drives transit-oriented development in clusters like Arlington, Alexandria, Silver Spring, and Bethesda. In these islands of density, the transit-oriented lifestyle is at least semi-plausible.

Paris has fairly uniform density within the city, but it has strong TOD structure in the suburbs: high density within walking distance of RER stations, lower density elsewhere. Some RER stations are also surrounded by job clusters oriented toward the train station: La Defense is by far the biggest and best-known, but Cergy, Val d’Europe and Marne la Vallee, Issy, Noisy, and Saint-Denis are all walkable to job centers and not just housing. Within the city there is no obvious structure, but the density is so high and the Metro so ubiquitous that transit serves the secondary nodes well.

In Los Angeles, there is no structure to density. There are some missing middle and mid-rise neighborhoods, but few form contiguous blobs of high density that can be served by a rapid transit line. Koreatown is in a near-tie with Little Osaka for highest population density in the United States outside New York, but immediately to its west, on the Purple Line Extension, lie kilometers of single-family sprawl, and only farther west on Wilshire can one see any density (in contrast, behind Little Osaka on Geary lies continuous density all the way to the Richmond). With the exception of Century City, UCLA, and Santa Monica, the secondary centers don’t lie on any obvious existing or current transit line.

With no coherent structure, Los Angeles is stuck. Its dense areas are too far away from one another and from job centers to be a plausible urban zone where driving is not necessary for a respectable middle-class lifestyle. Buses are far too slow, and trains don’t exist except in a handful of neighborhoods. Worse, because the density is so haphazard, the rail extensions can’t get any ridership. The ridership projection for the Purple Line Extension is an embarrassing 78,000 per weekday for nearly 15 km and $8.2 billion. The construction cost is bad, but in a large, dense city should be offset by high ridership (as it is in London); but it isn’t, so the projected ridership per kilometer is on a par with some New York City Transit buses and the projected cost per rider is so high that it is usually reserved for airport connectors.

The way out

In a smaller, cheaper auto-centric city, like Nashville or Orlando, I would be entirely pessimistic. In Los Angeles there is exactly one way out: fix the urban design, and reinforce it with a strong rail network.

The fact that this solution exists does not mean it is politically easy. In particular, the region needs to get over two hangups, each of which is separately nearly insurmountable. The first is NIMBYism. Los Angeles is so expensive that if it abolishes its zoning code, or passes a TOD ordinance that comes close to it, it could see explosive growth in population, which would be concentrated on the Westside, creating a large zone of high density in which people could ride the trains. However, the Westside is rich and very NIMBY. Metro isn’t even trying to upzone there: the Purple Line Extension has a 3.2-km nonstop segment from Western to La Brea, since the single-family houses in between are too hard to replace with density. Redeveloping the golf courses that hem Century City so that it could grow to a real second downtown is attractive as well, but even the YIMBYs think it’s unrealistic.

The second obstacle is the hesitation about spending large amounts of money all at once. American politicians are risk-averse and treat all spending as risk, and this is true even of politicians who boldly proclaim themselves forward-thinking and progressive. Even when large amounts of money are at stake, their instincts are to spread them across so many competing goals that nothing gets funded properly. The amount of money Los Angeles voters have approved to spend on transportation would build many rapid transit lines, even without big decreases in construction costs, but instead the money is wasted on showcasing bus lanes (this is Metro’s official blog’s excuse for putting bus lanes on Vermont and not rapid transit) or fixing roads or the black hole of Metro operating costs.

But the fact that Los Angeles could be a transit city with drastic changes to its outlook on development and transportation investment priorities does not mean that it is a transit city now. Nor does it mean that the ongoing program of wasting money on low-ridership subway lines is likely to increase transit usage by the required amount. Los Angeles does not have public transportation today in the sense that the term is understood here or in New York or even in Chicago. It should consider itself lucky that it can have transit in the future if it implements politically painful changes, but until it does, it will remain the autopia everyone outside urbanism thinks it is.

Sometimes, Bus Stop Consolidation is Inappropriate

For the most part, the optimal average spacing between bus stops is 400-500 meters. North American transit agencies have standardized on a bus stop every 200-250 meters, so stop consolidation is usually a very good idea. But this is based on a model with specific inputs regarding travel behavior. In some circumstances, travel behavior is different, leading to different inputs, and then the model’s output will suggest a different optimum. In contrast with my and Eric’s proposal for harsh stop consolidation in Brooklyn, I would not recommend stop consolidation on the crosstown buses in Manhattan, and am skeptical of the utility of stop consolidation in Paris. In Vancouver I would recommend stop consolidation, but not on every route, not do we recommend equally sweeping changes on every single Brooklyn route.

The model for the optimal stop spacing

If demand along a line is isotropic, and the benefits of running buses more frequently due to higher in-vehicle speed are negligible, then the following formula holds:

\mbox{Optimum spacing} = \sqrt{2\cdot\frac{\mbox{walk speed}}{\mbox{walk penalty}}\cdot\mbox{stop penalty}\cdot\mbox{average trip distance}}

The most important complicating assumption is that if demand is not isotropic, but instead every trip begins or ends at a distinguished location where there is certainly a stop, such as a subway connection, then the formula changes to,

\mbox{Optimum spacing} = \sqrt{4\cdot\frac{\mbox{walk speed}}{\mbox{walk penalty}}\cdot\mbox{stop penalty}\cdot\mbox{average trip distance}}

The choice of which factor to use, 2 or 4, is not exogenous to the bus network. If the network encourages transferring, then connection points will become more prominent, making the higher factor more appropriate. Whether the network encourages interchanges depends on separate policies such as fare integration but also on the shape of the network, including bus frequency. Higher average bus speed permits higher frequency, which makes transferring easier. The model does not take the granularity of transfer ease into account, which would require a factor somewhere between 2 and 4 (and, really, additional factors for the impact of higher bus speed on frequency).

After the choice of factor, the most contentious variable is the walk speed and penalty. Models vary on both, and often they vary in directions that reinforce each other rather than canceling out (for example, certain disabilities reduce both walk speed and willingness to walk a minute longer to save a minute on a bus). In Carlos Daganzo’s textbook, the walk speed net of penalty is 1 m/s. For an able-bodied adult, the walk speed can exceed 1.5 m/s; penalties in models range from 1.75 (MTA) to 2 (a Dutch study) to 2.25 (MBTA). The lower end is probably more appropriate, since the penalty includes a wait penalty, and stop consolidation reduces waits even as it lengthens walk time.

Update 10/31: alert reader Colin Parker notes on social media that you can shoehorn the impact of walk time into the model relatively easily. The formula remains the same with one modification: the average trip distance is replaced with

\mbox{average trip distance} + \frac{\mbox{average distance between buses}\cdot\mbox{wait penalty}}{2}.

The factor of 2 in the formula comes from computing average wait time; for worst-case wait time, remove the 2 (but then the wait penalty would need to be adjusted, since the wait penalty is partly an uncertainty penalty).

The average distance between buses is proportional to the number of service-hours, or fleet size: it obeys the formula

\mbox{revenue service-hours per hour} = \frac{2\cdot\mbox{route-length}}{\mbox{average distance between buses}}.

The factor of 2 in the formula comes from the fact that route-length is measured one-way whereas revenue hours are for a roundtrip.

If we incorporate wait time into the model this way, then the walk and wait penalties used should be higher, since we’re taking them into account; the Dutch study’s factor of 2 is more reasonable. The conclusions below are not really changed – the optima barely increase, and are unchanged even in the cases where stop consolidation is not recommended.

The situation in New York

The average unlinked New York City Transit bus trip is 3.35 km: compare passenger-miles and passenger trips as of 2016. In theory this number is endogenous to the transit network – longer interstations encourage passengers to take the bus more for long trips than for short trips – but in practice the SBS routes, denoted as bus rapid transit in the link, actually have slightly shorter average trip length than the rest. For all intents and purposes, this figure can be regarded as exogenous to stop spacing.

The stop penalty, judging by the difference between local and limited routes, is different for different routes. The range among the routes I have checked looks like 20-40 seconds. However, Eric tells me that in practice the B41, which on paper has a fairly large stop penalty, has little difference in trip times between the local and limited versions. The local-SBS schedule difference is consistent with a stop penalty of about 25 seconds, at least on the B44 and B46.

As a sanity check, in Vancouver the scheduled stop penalty on 4th Avenue is 22 seconds – the 84 makes 19 fewer stops than the 4 between Burrard and UBC and is 7 minutes faster – and the buses generally run on schedule. The actual penalty is a little higher, since the 4 has a lot of pro forma stops on the University Endowment Lands that almost never get any riders (and thus the bus doesn’t stop there). This is consistent with 25 seconds at a stop that the bus actually makes, or even a little more.

Plugging the numbers into the formula yields

\mbox{Optimum spacing} = \sqrt{4\cdot\frac{1.5}{1.75}\cdot 25\cdot 3350} = 536 \mbox{ meters}

if we assume everyone connects to the subway (or otherwise takes the bus to a distinguished stop), or

\mbox{Optimum spacing} = \sqrt{2\cdot\frac{1.5}{1.75}\cdot 25\cdot 3350} = 379 \mbox{ meters}

if we assume perfectly isotropic travel demand. In reality, a large share of bus riders are connecting to the subway, which can be seen in fare revenue, just $1.16 per unlinked bus trip compared with $1.91 per subway trip (linked or unlinked, only one swipe is needed). In Brooklyn, it appears that passengers not connecting to the subway disproportionately go to specific distinguished destinations, such as the hospitals, universities, and shopping centers, or Downtown Brooklyn, making the higher figure more justified. Thus, our proposed stop spacing, excluding the long nonstop segments across the Brooklyn-Battery Tunnel and between borough line and JFK, is 490 meters.

Update 10/31: if we incorporate wait time, then we need to figure out the average distance between buses. This, in turn, depends on network shape. Brooklyn today has 550 km of bus route in each direction, which we propose to cut to 350. With around 600 service hours per hour – more at the peak, less off-peak – we get an average distance between buses of 1,830 meters today or 1,180 under our proposal. Using our proposed network, and a wait and walk penalty of 2, we get

\mbox{Optimum spacing} = \sqrt{4\cdot\frac{1.5}{2}\cdot 25\cdot (3350 + \frac{2\cdot 1180}{2}} = 583 \mbox{ meters}


\mbox{Optimum spacing} = \sqrt{2\cdot\frac{1.5}{2}\cdot 25\cdot (3350 + \frac{2\cdot 1180}{2})} = 412 \mbox{ meters}.

Short bus routes imply short stop spacing

Our analysis recommending 490 meter interstations in Brooklyn depends on the average features of New York’s bus network. The same analysis ports to most of the city. But in Manhattan, the situation is different in a key way: the crosstown buses are so short that the average trip length cannot possibly match city average.

Manhattan is not much wider than 3 km. Between First and West End Avenues the distance is 2.8 km. The likely average trip length is more than half the maximum, since the typical use case for the crosstown buses is travel between the Upper East Side and Upper West Side, but the dominant destinations are not at the ends of the line, but close to the middle. With Second Avenue Subway offering an attractive two-seat ride, there is less reason to take the crosstown buses to connect to the 1/2/3 (and indeed, the opening of the new line led to prominent drops in ridership on the M66, M72, M79, M86, and M96); the best subway connection point is now at Lexington Avenue, followed by Central Park West. On a long route, the location of the dominant stop is not too relevant, but on a short one, the average trip length is bounded by the distance between the dominant stop and the end of the line.

If we take the average trip length to be 1.6 km and plug it into the formula, we get

\mbox{Optimum spacing} = \sqrt{4\cdot\frac{1.5}{1.75}\cdot 25\cdot 1600} = 370 \mbox{ meters}


\mbox{Optimum spacing} = \sqrt{2\cdot\frac{1.5}{1.75}\cdot 25\cdot 1600} = 262 \mbox{ meters.}

A crosstown bus stopping at First, Second, Third, Lex, Madison, Fifth, Central Park West, Columbus, Amsterdam, Broadway, and West End makes 10 stops in 2.8 km, for an average of 280 meters. There isn’t much room for stop consolidation. If the bus continues to Riverside, lengthening the trip to 3 km at the latitude of 96th Street, then it’s possible to drop West End. If the buses running up Third and down Lex are converted to two-way running, presumably on Lex for the subway connections, then Third could be dropped, but this would still leave the interstation at 330 meters, much tighter than anything we’re proposing in Brooklyn.

The only other places where avenues are too closely spaced are poor locations for stop removal. Amsterdam and Broadway are very close, but Amsterdam carries a northbound bus, and if the Columbus/Amsterdam one-way pair is turned into two two-way avenues, then Amsterdam is a better location for the bus than Columbus because it provides better service to the Far West Side. Fifth and Madison are very close as well, but the buses using them, the M1 through M4, are so busy (a total of 32 buses per hour at the peak) that if the two avenues are converted to two-way running then both should host frequent bus trunks. It’s not possible to skip either.

Within Brooklyn, there is one location in which the same issue of short bus routes applies: Coney Island. The B74 and B36 act as short-hop connectors from Coney Island the neighborhood to Coney Island the subway station. The routes we propose replacing them have 7 stops each from the subway connection west, over distances of 2.5 and 2.7 km respectively, for interstations of 360 and 390 meters.

Vancouver supplies two more examples of routes similar to the B74 and B36: the 5 and 6 buses, both connecting the West End with Downtown. The 6 is only 2 km between its western end and the Yaletown SkyTrain station, and the 5 is 2.3 km from the end to the Burrard station and 2.8 km to city center at Granville Street. The average trip length on these buses is necessarily short, which means that stop consolidation is not beneficial, unlike on the main grid routes outside Downtown.

Update 10/31: incorporating wait time into this calculation leads to the same general conclusion. The short routes in question – the Manhattan crosstowns, the B36 and B74, and the 5 and 6 in Vancouver – have high frequency, or in other words short distance between buses. For example, the M96 runs every 4 minutes peak, 6 off-peak, and takes 22-24 minutes one-way, for a total of 6 circulating buses per direction peak (which is 500 meters), or 4 off-peak (which is 750 meters). This yields

\mbox{Optimum spacing} = \sqrt{2\cdot\frac{1.5}{2}\cdot 25\cdot (1600 + \frac{2\cdot 500}{2})} = 281 \mbox{ meters.}

A network that discourages transferring should have more stops as well

In Paris the average interstation on buses in the city looks like 300 meters; this is not based on a citywide average but on looking at the few buses for which Wikipedia has data plus a few trunks on the map, which range from 250 to 370 meters between stations.

The short stop spacing in Paris is justified. First of all, the average bus trip in Paris is short: 2.33 km as of 2009 (source, PDF-p. 24). Parisian Metro coverage is so complete that the buses are not useful for long trips – Metro station access time is short enough that the trains overtake the buses on total trip time very quickly.

Second, there is little reason to transfer between buses here, or to transfer between buses and the Metro. The completeness of Metro coverage is such that buses are just not competitive unless they offer a one-seat ride where the Metro doesn’t. Another advantage of buses is that they are wheelchair-accessible, whereas the Metro is the single least accessible major urban rail network in the world, with nothing accessible to wheelchair users except Line 14 and the RER A and B. It goes without saying that people in wheelchairs are not transferring between the bus and the Metro (and even if they could, they’d have hefty transfer penalties). The New York City Subway has poor accessibility, but nearly all of the major stations are accessible, including the main bus transfer points, such as Brooklyn College and the Utica Avenue stop on the 3/4.

With little interchange and a mostly isotropic city density, the correct formula for the optimal bus stop spacing within Paris is

\mbox{Optimum spacing} = \sqrt{2\cdot\frac{1.5}{1.75}\cdot 25\cdot 2330} = 316 \mbox{ meters,}

which is close to the midpoint of the range of interstations I have found looking at various routes.


The half-kilometer (or quarter-to-a-third-of-a-mile) rule for bus stop spacing is an empirical guideline. It is meant to describe average behavior in the average city. It is scale-invariant – the density of the city does not matter, only relative density does, and the size of the city only matters insofar as it may affect the average trip length. However, while scale itself does not lead to major changes from the guideline, special circumstances might.

If the geography of the city is such that bus trips are very short, then it’s correct to have closer stop spacing. This is the case for east-west travel in Manhattan. It is also common on buses that offer short-hop connections to the subway from a neighborhood just outside walking range, such as the B36 and B74 in Coney Island and the 5 and 6 in Vancouver’s West End.

Note that even in New York, with its 3.3 km average trip length, stop consolidation is still beneficial and necessary on most routes. North American transit agencies should not use this article as an excuse not to remove extraneous stops. But nor should they stick to a rigid stop spacing come what may; on some routes, encouraging very short trips (often 1.5 km or even less), closely spaced stations are correct, since passengers wouldn’t be riding for long enough for the gains from stop consolidation to accumulate.