People are sharing various maps of the high-speed rail network the US could build if it were interested in alternative transportation, and I promised I’d make one myself. I did this on camera on Twitch a week ago but was not finished, so I streamed it again just now – this is going to be a regular occurrence, always at 18:00 my time every Saturday. There’s a recording, but Twitch is being weird about letting me upload it, so it might make it to YouTube instead.
Here is the map:
A full-size image can be found here. Red lines are high-speed rail. Blue lines are marginal lines: New Haven-Springfield and Milwaukee-Green Bay are good legacy lines that may or may not work as full HSR (the former probably better than the latter), while Nashville-Memphis, the Pacific Northwest system, and Phoenix-Tucson are marginal between no service at all and HSR.
Florida High-Speed Rail
I did the calculations for Atlanta-Florida on camera. I was surprised that it turned out to work out well, even with semi-decent return on investment based on my Metcalfe’s lawformulas, around 3%. The rub is that Orlando is pretty big, and even though it is sprawl hell, it is also an unusually strong tourist destination, and the rail line would serve Disney World and Daytona Beach. This makes me more confident in a formula trained on Japanese and European cities with public transit than a connection between two random no-transit medium size cities like Cleveland and Cincinnati.
This itself is an example of Metcalfe’s law in action: the Miami-Orlando-Tampa system by itself only returns 2.2% per the formula, and an extension to Jacksonville 2.6%. I also have more certainty in the figures for the larger system, because the impact of sprawl on mode choice is smaller when distances get longer, because it doesn’t affect the air/rail mode choice as much as the car/rail mode choice.
Even at medium distances, observe that the South Florida urban area is linear, around 20 km wide but more than 100 long, which makes intercity rail service more reasonable. Every county can have a stop, and if the 0.8 exponent in the gravity model formula is applied to counties separately, then the sum rises to 6.1, whereas 7^0.8 = 4.74, which means that this refinement provides a 28% boost to ridership. Orlando is not linear, but its subsidiary metro areas, Lakeside and Daytona Beach, could get stops as well.
Alignment questions
I drew the system in a zoom level 7 on OpenStreetMap, which is too high-altitude to see individual railroads. I tried to approximate existing rail alignments that are worth using, but it’s not perfect, so please do not take the map as any assertion about pixel-level alignment, and even some station decisions can be quibbled with.
However, please do take the map as a definitive assertion about macro-scale alignments. The Northeast Corridor should go via I-95 and not via Hartford. This decision is fairly close and could go either way, though the benefits of HSR in the Northeast are so great that the absolute magnitude of such decisions remains momentous. Elsewhere, the Chicago-Minneapolis line could go along I-94 via Eau Claire or via a more southerly route via Rochester and the Mayo Clinic; I’ve gone back and forth on this, and it’s a second-order question, but I think the Mayo Clinic generates more trips, probably. The Albany-Montreal route could be entirely in the state of New York or take a slight detour through easier terrain in Vermont, which is likely cheaper. Toronto-Ottawa could go via Kingston or Peterborough, but the Peterborough route looks more direct. Chicago-St. Louis is sometimes proposed to detour via Champaign rather than go straight via Bloomington, but the benefit of serving UIUC probably doesn’t justify the extra cost. North Carolina HSR could go via the Triad or direct from Raleigh to Charlotte, but the model says the benefit of serving Greensboro is much greater than that of slightly faster trips coming from bypassing the Triad. Texas is a compromise route extending the under-construction line to Downtown Houston and creating a new leg connecting this system to Austin and San Antonio.
The most contentious questions are in California. HSR there should go via a partially high-speed coastal alignment from San Diego up to Los Angeles, then up the Grapevine and Tejon Pass, then across Altamont Pass and a Dumbarton tunnel. None of these decisions is close, and the official alignment decisions to detour via the Inland Empire and Palmdale and to go via Pacheco are all bad and played a role in the failure of the project. Los Angeles-San Diego is in a way the most frustrating: it was left to a future phase, but a medium-speed rail alignment along the coast could be done relatively quickly with electrification and some strategic investments, speeding up trains to about 1:45.
Frequency
I talked about frequency a little bit in the video, but not in much detail. The biggest problem is that Philadelphia is set up poorly: ideally trains coming from New York should branch to either Washington or Pittsburgh, but instead, 30th Street Station requires New York-Pittsburgh trains to reverse direction. This can be handled through actual reversal, as is done today at Frankfurt, with 4-minute turnarounds (cf. 10 at Philadelphia), or through having New York-Pittsburgh trains skip Philadelphia, as was historically done, with a stop at North Philadelphia instead.
With that in mind, my best guess, based partly on the model and partly on intra-metropolitan fudge factors like New York-New Haven, is as follows:
8 tph New York-Boston, 4 New York-Springfield
8 tph New York-Washington, 4 New York-Pittsburgh-Cleveland, 4 Washington-Philadelphia-Pittsburgh-Cleveland
8 tph New York-Albany, 4 short Boston-Albany, 8 Albany-Buffalo (4 short), 4 Buffalo-Toronto, 4 Albany-Montreal, 2 short Buffalo-Cleveland
The main difference is that public transportation does not have competition the way private industry does. In many travel markets, for example rush hour travel to city center, it is a monopoly. In others, it isn’t, but it remains fundamentally different from the competition, whereas private-sector marketing generally involves competition between fairly similar products, such as different brands of cars or computers or supermarkets. This also means that trying to turn public transit into a competition between similar providers is overrated: it is bad from the perspective of good planning, but it turns the industry into something private-sector marketers are more familiar with, and is therefore at risk of being adopted (for example, with EU competition mandates) despite being counterproductive.
Brand identity
Companies that make products that are very similar to their competition engage in extensive marketing. Coke vs. Pepsi is the most cliché example, but different brands of cookies, fast food, cars, computers, and smartphones do the same. The differences between these brands are never zero: I can generally tell different brands of bottled water by taste, Samsung- and Sony-made Androids have some differences (let alone iPhones), and so on. But it’s not large either.
Objectively, the cost of switching firms is small, so marketers first of all spend enormous amounts of money on advertising, and second of all aim to create identity markers to impose an emotional cost on customers who switch: “I am a Mac.” If the small differences involve differences in price point, then this can include a marker of class identity; even if they don’t, there’s no shortage of ways to tell people what brand of alcohol or food or video game best fits their microidentity. Establishing brand identity also involves loyalty programs, like airline miles and hotel points: why compete when you can lock passengers into your airline alliance?
This can even bleed into product development to some extent. Microsoft’s embrace, extend, exterminate strategy was designed around getting people to switch to Microsoft products from competitors. This was not a marketing gimmick – the people who developed Excel made sure everything that Lotus 1-2-3 users were used to would also feature in Excel in order to reduce the cost of switching to Microsoft, before using Windows’ power to lock people into Office.
Mass transit is not like this
Public transportation competes with cars as a system. It has a monopoly in certain travel markets, namely rush hour travel to city center, but the existence of those markets itself comes from real estate competition, in which it is necessary to entice companies to choose to locate in city center rather than in a suburban office park. Of note, the following features, all unusual for private-sector competition, apply:
Competition is for the most part binary: public transportation versus cars. (Bikes complement transit.)
The public transit side of the competition has economies of scale because of the importance of frequency of arrival, and thus is harmed by any internal competition, whereas the car industry has different automakers and works just fine that way.
The service has very little customization – everyone rides the same trains. Attempts to introduce product differentiation are harmful because of the frequency effect.
The product is completely different from the competition – useful at different times of day, in different neighborhoods, for different destinations. Switching incurs costs of similar magnitude to those of migration.
Much of the competition is not for customers, but for development – city center development is good for public transit, sprawl is good for cars.
There is competition over public resources, which cannot be divorced from the mode even in an environment of privatization – someone still has to build roads and finance subways.
The consequences of mass transit Fordism
Public transportation is and remains a Fordist product – no product differentiation, highly regimented worker timetables, one-size-fits-all construction, vertical integration. The vertical integration aspects go even farther than early-20th century industry, covering infrastructure, timetables, the equipment, and development. User choice is extensive regarding where to go within the system – I have access to far more variety of products as a consumer and jobs as a worker in Berlin (and had even more in Paris) than I would have driving in a sprawl environment, but I can’t choose what brand of train to use.
This is particularly important when preferences are heterogeneous. Different users have different walking speeds, transfer penalties, idiosyncrasies about access to wifi on board, etc. Planning has to use averages, and for the most part this works without too many seams, but it means that the standard way private businesses use product differentiation doesn’t work.
Of note, this Fordism also exists for the road network, if not for the cars themselves. It’s just far less visible. Drivers may have different preferences that translate to different costs and benefits for a cloverleaf versus a four-level interchange, but engineers can’t have two sets of interchanges, they just build one based on criteria of traffic density. However, the experience of driving on the interchange is not visible as part of the system to the drivers, who occasionally grumble about traffic at a particular intersection but don’t see it as clearly as transit riders see specific transfer stations or modal questions like streetcar vs. subway.
How private-sector marketing can harm transit
Because mass transit is a single system for everyone, standard private-sector marketing schemes involve changes to service that harm the overall system.
Creating brand identification with a specific subgroup of users, such as when some private buses market themselves to tech workers with wifi and USB chargers and charge higher fares, and still can’t make money. Public transportation has to work on an any vehicle, any place, any time principle. Only a handful of hyper-frequent routes can take multiple brands without losing passengers due to the lower frequency of each brand, but on those routes the only reliable way to timetable service is to run on headway management in which case any vehicle can substitute for any vehicle, which means you can’t brand.
This is especially bad when the brands are different modes: bus, bike, streetcar, subway, commuter train. When some modes are marketed to the rich and others are to the poor, capacity is wasted and frequency within each class is lower. Moreover, infrastructure planning is weaker with such differentiation, because often a region or subclass will be close to the wrong mode, forcing expensive additional construction. The United States fails by running commuter rail just for the rich while subways are for the rest, while India fails by doing the exact opposite; both countries build unnecessary infrastructure and underinvest in intermodal integration as a result.
Finally, because public transportation is a complex system, trading the need for inter-organization and interdepartmental organization for much lower overall provision costs, people who come into it from consumer product markets may miss some of the required connections. This is especially true of development – people who sell consumer products, including cars, don’t need to think how urban design has to look for their product to succeed. Even people who have heard of transit-oriented development may get it wrong; in the United States, it is common to build some apartment buildings next to a train station but neglect retail and local services, and YIMBY as a movement is at best indifferent to city center office towers.
I’m going to stream on Twitch at 18:00 tonight my time (UTC+1). This is usually intended as a platform for streaming playing video games, but can also be used for other things, such as discussions that would benefit from video and screen sharing. This may turn into a regular feature, covering a different topic every time – probably monthly, I don’t see myself having enough time to do this weekly unless it’s extremely impromptu.
Today’s topic is crayons:
How I make maps such as these ones: how I grab the base map, how I draw lines, etc.
Good tips for how to figure out which lines are useful, including very rudimentary cost-per-rider analyses.
An example of making a map, probably a US-wide high-speed rail map based on my Metcalfe’slaw posts but maybe something else if there’s popular demand.
I had an idea that the reaction to my last post would go along the lines it did. It was incredibly nitpicky, sometimes pointing out valid concerns, but often just asserting certain things are impossible or incompatible with Long Island culture that happen thousands of times a day in large European and Japanese cities. And it speaks to different communities in different cities – as I said in the post, I wrote that article in the style of a TransitMatters Regional Rail appendix, like the just-released Newburyport/Rockport Line appendix. This matters, in a number of different ways. The most important is this:
The complexity of regional rail
Naively, Boston should be a harder system to modernize than New York. The system is not electrified, and only around half the stations don’t even have high platforms. Most lines have single-track bottlenecks, especially the world-of-pain Old Colony system. Then there’s the need to tie everything together with the North-South Rail Link, to be built as a four-track rail tunnel underneath the Big Dig.
And yet, it’s still easy mode. Boston has four commuter lines on each side: Fitchburg, Lowell, Haverhill, Eastern; Worcester, Providence, Fairmount, Old Colony. There are some alignment questions in that the Franklin Line can get to Boston via the Providence Line (that is, the Northeast Corridor) or the Fairmount Line and that the outer Haverhill Line can be turned into a branch off Lowell, but otherwise the eight lines are independent. The North Station approach has a six-track pinch point, but a schedule that is optimized for short turnarounds is a schedule that automatically avoids at-grade conflicts in the throat. So each system can be planned independently; the line-by-line appendix structure comes from this technical fact.
Moreover, on the level of ridership and social planning, the lines’ sheds are mostly disjoint too. It makes complete sense to speak of jobs on the Providence Line: Providence, for one, has 30,000 jobs within 1 km of the station, which is a near-tie with Ruggles for #4 in the system after the two Boston terminals and Back Bay; and there is never much of an overlap between different lines’ sheds. This also leads to some socioeconomic targeting – the Eastern Lines appendix linked at the top of this post goes into the economic justice implication of service to working-class suburbs on the line, like Chelsea and Lynn.
New York is not like this, at all. The lines are too dense: the Hempstead Line is closely parallel to the Main Line, and it matters that Garden City and Mineola’s walk sheds overlap, albeit only marginally. The branching structure is such that one must treat each system as a whole from the start, with only a handful of plausible cleaves. My Hempstead Line post winks at that by briefly talking about how to schedule Babylon trains, but the reality is that the technical complexity of timetabling New York-area commuter railroads is genuinely high, especially once through-running comes in. (Our Boston Regional Rail timetables are pre-NSRL, so without through-running.)
None of this should be taken to mean that this is not possible. Nor does this mean it is only possible with the sort of expenditures the MTA and Port Authority are used to. Surface improvements should be budgeted in the billions of dollars, but that is equally true of Boston, with New York’s greater size trading off against the mostly-electrified, mostly-high-platform character of its network. The real difference is that Boston can do this in $1-2 billion chunks and each chunk has meaningful improvements on a few lines, whereas in New York there are more moving parts and planning has to be more integrated.
In a way, this shouldn’t surprise. Big cities trade off greater efficiency for greater needs of social organization. New York’s efficiency means that most of it can have every bus and train run every six minutes, but then someone has to plan these bus networks and train connections. At metro area scale, it has enormous potential transit ridership – I think Metro New York beats every other American metro region for how much extra transit ridership it can get in absolute numbers through better capital construction and operations, even Los Angeles – but this doesn’t easily lend itself to line-by-line planning.
Line by line in New York
New York plausibly has the weakest directionality of any city I’m familiar with. Its directional identities are New Jersey vs. in-state suburbs, and to some extent North vs. South Shore of Long Island but city vs. suburbs is the dominant distinction, so social targeting on one commuter line is completely pointless.
This also has an implication for infrastructure. The Hempstead Line is also really two different things – the Hempstead Line proper and the local tracks within the city, with different needs. The East Garden City branch is a way to fit them together, using the jobs around East Garden City to really look for ways to create more local service within Eastern and Central Queens. The upshot is that in New York, a large fraction of the missing ridership on regional rail – the difference between the region’s 800,000 daily riders and Ile-de-France’s 3.5 million on a smaller population – comes from within the city. This matters; the suburbs have a strong not-the-city identity, manifesting itself in resistance to school integration, to densification, to transit fare integration.
Essentially, the correct structure of a New York regional rail proposal is to start from generalities, as we did at TransitMatters, but then transition not to line-by-line, but to topical aspects: scheduling, extensions of electrification, near-city infrastructure (new interlockings, etc.), transit-oriented development, in-city and inner-suburban service, the Gateway tunnel, further through-running tunnels.
What does it mean to be technical?
The first Boston Regional Rail report came out three years ago. We started working on the appendices immediately; things slowed down because of juggling of priorities. We spent much of 2019 trying to address a certain technical commentary gallery that would ask about South Station capacity concerns, back when there was real risk of Massachusetts funding the completely pointless South Station Expansion project; this led to the proof of concept, released at the same time as the Worcester appendix. But that essentially led to trust in TM’s technical acumen, meaning the remaining appendices have been written to a broader audience. This is why we talk about the destinations one can go into near the stations, which I reproduce in a shorter form in my Hempstead writeup.
New York is different. Its size makes it parochial, and thus there is an enormous base of people who know the minutiae of its various pieces and only them – and thanks to its size, they respond extremely aggressively toward any outside knowledge. There are too many things that people on Subchat and New York City Transit Forums, which communities include a fair number of insiders, treat as obvious and yet are incorrect. This is worse when it comes to mainline rail – the insularity is worse, the not-regular-transit identity means that even things that are unremarkable on the subway are treated with suspicion on commuter rail.
Essentially, this forces a dialogue that we didn’t really need to have in Boston about the items (about costs, yes, but that’s a separate matter). Keolis knows what a clockface timetable is, thankfully. Enough insiders are reasonable that the only real technical need was to rebut the idea of South Station Expansion. New York, in contrast, is full of people who think the city is on top of the world and not just in corona death rates, and has capital plans that spend $40 billion in a cycle to do what other cities get done on more than a full order of magnitude less.
How special is New York?
New York is uniquely large, just like all other large cities. What this means is that there are always little places where it needs more. For example, regional rail planning in New York needs to be able to reliably run 24 trains per hour on a two-track line; this is routine, but this is not something seen in every city, because most cities don’t need this. Tokyo, Paris, and London of course all have this, even on fairly fast lines – the RER A averages 49.5 km/h, a hair less than the LIRR, and the Tokaido Main Line scratches 60 km/h. Munich has this, but is a fairly special case for a city that’s objectively kind of small. But Berlin for example tops at 18 tph on the Stadtbahn and even short-turns some trains that really should be going through.
But precisely because the number of examples, while solid, is finite, it’s easy to excuse them away. People who are committed to making New York better can figure out how to modify things as necessary – where the signal blocks need to be shortened, where a track needs to be moved, etc. Learning from Spain should be especially easy given the large number of Spanish-speaking New Yorkers. However, that requires a willingness to learn, and as Sandy points out in his master’s thesis, American commuter rail encourages insularity. New York’s size then means it’s easier to be insular there than elsewhere, to the point of not even noticing advances in other American cities, let alone actually good regional rail networks outside North America. This, in turn, forces every plan to degenerate into either accepting incorrect service assumptions, or getting into too much detail about why those assumptions are wrong.
This is a writeup I prepared for modernization of the Hempstead Branch of the LIRR in the same style as our ongoing Regional Rail line by line appendices for Boston at TransitMatters, see e.g. here for the Worcester Line. This will be followed up in a few days by a discussion of the writing process and what it means for the advocacy sphere.
Regional rail for New York: the Hempstead Line
New York has one of the most expansive commuter rail networks in the world. Unfortunately, its ridership underperforms such peer megacities as London, Paris, Tokyo, Osaka, and Seoul. Even Berlin has almost twice as much ridership on its suburban rail network, called S-Bahn, as the combined total of the Long Island Railroad, Metro-North, and New Jersey Transit. This is a draft proposal of one component of how to modernize New York’s commuter rail network.
The core of modernization is to expand the market for commuter rail beyond its present-day core of 9-to-5 suburban commuters who live in the suburbs and work in Manhattan. This group already commutes by public transportation at high rates, but drives everywhere except to Manhattan. To go beyond this group requires expanding off-peak service to the point of making the commuter railroads like longer-range, higher-speed Queens Boulevard express trains, with supportive fares and local transit connections.
The LIRR Hempstead Line is a good test case for beginning with such a program. It is fortunate that on this line the capital and operating costs of modernization are low, and service would be immediately useful within the city as well as dense inner suburbs. With better service, the line would still remain useful to 9-to-5 commuters – in fact it would become more useful through higher speed and more flexibility for office workers who sometimes stay at the office until late. But in addition, people could take it for ordinary transit trips, including work trips to job centers in Queens or on Long Island, school trips, or social gatherings with friends in the region.
The Hempstead Line
The Hempstead Line consists of the present-day LIRR Hempstead Branch and a branch to be constructed to East Garden City. The Hempstead Branch today is 34 km between Penn Station and Hempstead, of which 24 km lie within New York City and 10 lie within Long Island.
Most trains on the branch today do not serve Penn Station because of the line’s low ridership, but instead divert to the Atlantic Branch to Downtown Brooklyn, and Manhattan-bound passengers change at Jamaica to any of the branches that run through to Midtown. Current frequency is an hourly train off-peak, and a train every 15-20 minutes for a one-hour peak. Peak trains do not all run local, but rather one morning peak train runs express from Bellerose to Penn Station.
Ridership is weak, in fact weaker than on any other line except West Hempstead and the diesel tails of Oyster Bay, Greenport, and Montauk. In the 2014 station counts, the sum of boardings at all stations was 7,000 a weekday, and the busiest stations were Floral Park with 1,500 and Hempstead with 1,200. But commute volumes from the suburbs served by the Hempstead Branch to the city are healthy, about 7,500 to Manhattan and another 10,500 to the rest of the city, many near LIRR stations in Brooklyn and Queens. Moreover, 13,500 city residents work in those suburbs, and they disproportionately live near the LIRR, but very few ride the train. Finally, the majority of the line’s length is within the city, but premium fares and low frequency make it uncompetitive with the subway, and therefore ridership is weak.
Despite the weak ridership, the line is a good early test case for commuter rail modernization in New York. Most of it lies in the city, paralleling the overcrowded Queens Boulevard Line of the subway. As explained below, there is also a healthy suburban job market, which not only attracts many city reverse-commuters today, but is likely to attract more if public transportation options are better.
Destinations
The stations of the Hempstead Line already have destinations that people can walk to, so that if service is improved as in the following outline, people can ride the LIRR there. These include the following:
JFK, accessible via Jamaica Station.
Adelphi University, midway between Garden City and Nassau Boulevard, walkable to both.
York University, fairly close to Jamaica and very close to a proposed Merrick Boulevard infill station.
Primary and secondary schools near stations within the city, where students often have long commutes.
Penn Station as an intercity station – passengers from Queens and Long Island traveling to Boston, Philadelphia, and Washington would benefit from faster and more frequent trains.
Many jobs near stations in Queens and on Long Island as described below.
Jobs
Within a kilometer of all stations except Penn Station, there is a total of 182,000 jobs in Queens and 50,000 on Long Island. The spine of the Main Line through Queens closely parallels the overcrowded Queens Boulevard express tracks, and in the postwar era was proposed for a Queens Super-Express subway line. But on Long Island, too, it serves the edge city cluster of Garden City and the city center of Hempstead. All of those jobs should generate healthy amounts of reverse-peak ridership and ridership terminating short of Manhattan.
Station
Jobs within 1 km
Penn Station
522904
Queensboro Plaza (@ QB)
62266
Sunnyside Jct (@ 43th)
23655 (with QBP: 78219)
Woodside
14409 (with Sunnyside: 36469)
Triboro Jct (@ 51st Ave)
14339 (Elmhurst Hospital)
Forest Hills
21926
Kew Gardens
17855
Jamaica
19794
Merrick Blvd
17020 (with Jamaica: 29260)
Hollis
2918
Queens Village
4758
Bellerose
3014 (with QV: 7735)
Floral Park
5389 (with Bellerose: 6776)
Stewart Manor
3203
Nassau Blvd
859
Garden City
9643
Country Life Press
5404 (with GC: 10865)
Hempstead
10896 (with CLP: 15823)
East Garden City (@Oak)
12461
Nassau Center (@Endo)
6352 (with EGC: 17904)
Required infrastructure investment
The LIRR has fairly high quality of infrastructure. Every single station has high platforms, permitting level boarding to trains with doors optimized for high-throughput stations. Most of the system is electrified with third rail, including the entirety of the Hempstead Branch. High-frequency regional rail can run on this system without any investment. However, to maximize utility and reliability, some small capital projects are required.
Queens Interlocking separation
Queens Interlocking separates the Hempstead Line from the Main Line. Today, the junction is flat: two two-track lines join together to form a four-track line, but trains have to cross opposing traffic at-grade. The LIRR schedules trains around this bottleneck, but it makes the timetable more fragile, especially at rush hour, when trains run so frequently that there are not enough slots for recovering from delays.
The solution is to grade-separate the junction. The project should also be bundled with converting Floral Park to an express station with four tracks and two island platforms; local trains should divert to the shorter Hempstead Line and all express trains should continue on the longer Main Line to Hicksville and points east. Finding cost figures for comparable projects is difficult, but Harold Interlocking was more complex and cost $250 million to grade-separate, even with a large premium for New York City projects.
Turnout modification
Trains switch from one track to another at a junction using a device called a switch or turnout. There are two standards for turnouts: the American standard, dating to the 1890s, in which the switch is simpler to construct but involves an abrupt change in azimuth, called a secant switch; and the German standard from 1925, adopted nearly globally, in which the switch tapers to a thin blade to form what is called a tangential switch.
Passengers on a train that goes on a secant turnout are thrown sideways. To maintain adequate safety, trains are required to traverse such switches very slowly, at a speed comparable to 50 mm of cant deficiency on the curve of the switch. In contrast, German and French turnout standards permit 100 mm on their tangential switches; the double cant deficiency allows a nominal 40% increase in speed on a switch of given number (such as an American #10 vs. a German 1:10 or a French 0.1, all measuring the same frog angle). The real speed increase is usually larger because the train sways less, which creates more space in constrained train station throats.
With modern turnouts, Penn Station’s throat, currently limited to 10 15 mph (16 24 km/h), could be sped up to around 50 km/h, saving every train around 2 minutes just in the last few hundred meters into the station. Installation typically can be done in a few weekends, at a cost of around $200,000 per physical switch, which corresponds to high single-digit millions for a station as large as Penn. Amtrak has even taken to installing tangential switches on some portions of the Northeast Corridor, though not at the stations; unfortunately, instead of building these switches locally at local costs, it pays about $1.5 million per unit, even though in Germany and elsewhere in Europe installation costs are similar to those of American secant switches.
Speed
In addition to modifying the physical switches as outlined above, the LIRR should pursue speedups through better use of the rolling stock and better timetabling. In fact, the trains currently running are capable of 0.9 m/s^2 acceleration, but are derated to 0.45 without justification, which increases the time cost of every stop by about 30 seconds. In addition, LIRR timetables are padded about 20% over the technical running time, even taking into account the slow Penn Station throat and the derating. A more appropriate padding factor is 7%, practiced throughout Europe even on very busy mainlines, such as the Zurich station throat, where traffic is comparable to that of the rush hour LIRR.
To get to 7%, it is necessary to design the infrastructure so that delays do not propagate. Grade-separating Queens Interlocking is one key component, but another is better timetabling. Complex timetables require more schedule padding, because each train has a unique identity, and so if it is late, other trains on the line cannot easily substitute for it. In contrast, subway-style service with little branching is the easiest to schedule, because passengers do not distinguish different trains; not for nothing, the 7 and L trains, which run without sharing tracks with other lines, tend to be the most punctual and were the first two to implement CBTC signaling.
In the case of the LIRR, achieving this schedule requires setting things up so that all Hempstead Line trains run local on the Main Line to Penn Station, and all trains from Hicksville and points east run express to Grand Central. Atlantic and Babylon Branch trains can run to Atlantic Terminal, or to the local tracks to Penn, depending on capacity; Babylon can presumably run to Penn while the Far Rockaway and Long Beach Lines, already separated from the rest of the system, can run to Downtown Brooklyn.
Infill stations
Within the city, commuter rail station spacing is sparse. The reason is that the frequency and fares are uncompetitive. Historically, the LIRR had tight spacing in the city, with nine more stations on the Main Line within city limits, but it closed most of them in the 1920s and 30s as the subway opened to Queens. The subway offered very high frequency for a 5-cent fare compared with the LIRR’s 20-to-30-cent fares. Today, the fares remain unequal, but this can be changed, as can the off-peak frequency. In that case, it becomes useful to open some additional infill stops.
The cost of an infill station is unclear. There is a wide range; Boston and Philadelphia both open infill stations with high platforms for about $15-25 million each, and the European range is lower. Urban infill stations in constrained locations like Sunnyside can be more expensive, but not by more than a factor of 2. In the past, LIRR and Metro-North infill stops, such as those for Penn Station Access, have gone up to the three figures, and it is critical to prevent such costs from recurring.
Queensboro Plaza
This station is already part of the Sunnyside Yards master plan, by the name Sunnyside, and is supposed to begin construction immediately after the completion of the East Side Access project. This proposal gives it a different name only because there is another station called Sunnyside (see below).
Located at the intersection of the Main Line with Queens Boulevard, this would be a local station for trains heading toward Penn Station. It is close to the Queensboro Plaza development, which has the tallest building in the city outside Manhattan and more jobs than anywhere in the Outer Boroughs save perhaps Downtown Brooklyn. Within a kilometer of the station there are more than 60,000 jobs already, and this is before planned redevelopment of Sunnyside Yards.
Sunnyside Junction
The opening of East Side Access and Penn Station Access will create a zone through Sunnyside Yards where trains will run in parallel. LIRR trains will run toward either Penn Station or Grand Central, and Metro-North trains will run toward Penn Station.
It is valuable to build an express station to permit passengers to transfer. This way, passengers from the Penn Station Access stations in the Bronx could connect to Grand Central, and passengers from farther out on the New Haven Line who wish to go to Penn Station Grand Central could board a train to either destination, improving the effective frequency. Likewise, LIRR passengers could change to a different destination across the platform at Sunnyside, improving their effective frequency.
The area is good for a train station by itself as well. It has 24,000 jobs within a kilometer, more than any other on the line except Penn Station and Queensboro Plaza. There is extensive overlap with the 1 km radius of Queensboro Plaza, but even without the overlap, there are 16,000 jobs, almost as many as within 1 km of Jamaica, and this number will rise with planned redevelopment of the Yards.
Triboro Junction
This station is at 51st Avenue, for future transfers to the planned Triboro RX orbital. Population and job density here are not high by city standards: the 14,000 jobs include 5,000 at Elmhurst Hospital on Broadway, which is at the periphery of the 1 km radius and is poorly connected to the railroads on the street network. The value of the station is largely as a transfer for passengers from Astoria and Brooklyn.
Merrick Blvd
About 1.5 km east of Jamaica, Merrick Boulevard catches the eastern end of the Jamaica business district. It also connects to one of Eastern Queens’ primary bus corridors, and passengers connecting from the buses to Manhattan would benefit from being able to transfer outside the road traffic congestion around Jamaica Station.
The East Garden City extension
The Hempstead Branch was historically part of the Central Railroad of Long Island. To the west, it continued to Flushing, which segment was abandoned in 1879 as the LIRR consolidated its lines. To the east, it continued through Garden City and what is now Levittown and ran to Babylon on a segment the LIRR still uses sporadically as the Central Branch. The right-of-way between Garden City and Bethpage remains intact, and it is recommended that it be reactivated at least as far as East Garden City, with an East Garden City station at Oak Street and a Nassau Center station at Endo Boulevard. This is for two reasons.
Jobs
Long Island is unusually job-poor for a mature American suburb. This comes partly from the lack of historic town centers like Stamford or Bridgeport on the New Haven Line or White Plains and Sleepy Hollow in Westchester. More recently, it is also a legacy of Robert Moses, who believed in strict separation of urban jobs from suburban residences and constructed the parkway system to feed city jobs. As a result of both trends, Long Island has limited job sprawl.
However, East Garden City specifically is one of two exceptions, together with Mineola: it has a cluster with 18,000 jobs within 1 km of either of the two recommended stations. Reopening the branch to East Garden City would encourage reverse-commuting by train.
Demand balance
Opening a second branch on the Hempstead Line helps balance demand in two separate ways. First, the population and job densities in Queens are a multiple of those of Long Island and always will be, and therefore the frequency of trains that Queens would need, perhaps a local train every 5 minutes all day, would grossly overserve Hempstead. At the distance of Hempstead or East Garden City, only a train every 10-15 minutes (in a pinch, even every 20) is needed, and so having two branches merging for city service is desirable.
And second, having frequent Hempstead Line local service forces all of the trains on the outer tracks of the Main Line in Queens to run local, just as the subway has consistent local and express tracks. The LIRR gets away with mixing different patterns on the same track because local frequency is very low; at high frequency, it would need to run like the subway. Because passengers from outer suburbs should get express trains, it is valuable to build as much infrastructure as possible to help feed the local tracks, which would be the less busy line at rush hour.
Train access and integration
Today, the LIRR primarily interfaces with cars. LIRR capital spending goes to park-and-rides, and it is expected that riders should drive to the most convenient park-and-ride, even on a different branch from the one nearest to their home. This paradigm only fills trains at rush hour to Manhattan, and is not compatible with integrated public transportation. In working-class suburbs like Hempstead, many take cheaper, slower buses. Instead, the system should aim for total integration at all levels, to extend the city and its relative convenience of travel without the car into suburbia.
Fare integration
Fares must be mode-neutral. This means that, just as within the city the fares on the buses and subways are the same, everywhere else in the region a ticket should be valid on all modes within a specified zone. Within the city, all trains and buses should charge the same fares, with free intermodal transfers.
Such a change would entice city residents to switch from the overcrowded E and F trains to the LIRR, which is by subway standards empty: the average Manhattan-bound morning rush hour LIRR train has only 85% of its seats occupied. In fact, if every E or F rider switches to the LIRR, which of course will not happen as they don’t serve exactly the same areas, then the LIRR’s crowding level, measured in standees per m^2 of train area, will be lower than that of the E and F today.
In the suburbs, the fares can be higher than in the city, in line with the higher operating costs over longer distances. But the fares must likewise be mode-neutral, with free transfers. For example, within western Nassau County, fares could be set at 1.5 times subway fare, which means that all public transit access between the city and Hempstead would cost $190 monthly or $4.00 one-way, by any mode: NICE bus, the LIRR, or a bus-train combo.
This would be a change from today’s situation, where premium-price trains only attract middle-class riders, while the working class rides buses. In fact, the class segregation today is such that in the morning rush hour, trains run full to Manhattan and empty outbound and NICE buses, which carry working-class reverse-commuters, are the opposite. Thus, half of each class’s capacity is wasted.
Bus redesign and bus access
Instead of competing with the trains, buses should complement them, just as they do within the city with the subway. This means that the NICE system should be designed along the following lines:
More service perpendicular to the LIRR, less parallel to it.
Bus nodes at LIRR stations, enabling passengers to connect.
Timed transfers: at each node the buses should arrive and depart on the same schedule, for example on the hour every 20 minutes, to allow passengers to change with minimal hassle. This includes timed transfers with the trains if they run every 15 minutes or worse, but if they run more frequently, passengers can make untimed connections as they do in the city.
Bike access
Urban and suburban rail stations should include bike parking. Bikes take far less space than cars, and thus bike park-and-ride stations in the Netherlands can go up to thousands of stalls while still maintaining a walkable urban characteristic.
In many countries, including the United States on the West Coast, systems encourage riders to bring their bikes with them on the train. However, in New York it’s preferably to adopt the Dutch system, in which bikes are not allowed on trains, and instead stations offer ample bike parking. This is for two reasons. First, New York is so large and has such a rush hour capacity crunch that conserving capacity on board each train is important. And second, cultures that bring bikes on trains, such as Northern California, arise where people take trains to destinations that are not walkable from the station; but in New York, passengers already connect to the subway for the last mile from Penn Station to their workplaces, and thus bikes are not necessary.
Train scheduling
Trains should run intensively, with as little distinction between the peak and off-peak as is practical. At most, the ratio between peak and off-peak service should be 2:1. Already, the LIRR’s high ratio, 4:1 on the Hempstead Branch, means that trains accumulate at West Side Yard at the end of the morning peak. The costs of raising off-peak service to match peak service are fairly low to begin with, but they are especially low when the alternative is to expand a yard in Midtown Manhattan, paying Midtown Manhattan real estate prices.
For an early timetable in which the Babylon Branch provides extra frequency in the city, the following frequencies are possible:
Segment
Peak
Off-peak
Penn Station-Garden City
5 minutes
10 minutes
Garden City-Hempstead
10 minutes
20 minutes
Garden City-Nassau Center
10 minutes
20 minutes
A more extensive service, with all LIRR South Side diverting to a separate line from the Main Line, perhaps the Atlantic Branch to Downtown Brooklyn, requires an increase in off-peak urban service:
Segment
Peak
Off-peak
Penn Station-Garden City
5 minutes
5 minutes
Garden City-Hempstead
10 minutes
10 minutes
Garden City-Nassau Center
10 minutes
10 minutes
Further increases in peak service may be warranted for capacity reasons if there is more redevelopment than currently planned or legal by city and suburban zoning codes.
Travel times
With rerating the LIRR equipment to its full acceleration rate, a fix to the Penn Station throat, and standard European schedule padding, the following timetable is feasible:
Station
Time (current)
Time (future, M7)
Time (Euro-EMU)
Penn Station
00:00
00:00
00:00
Queensboro Plaza
—
00:04
00:04
Sunnyside Jct
—
00:06
00:06
Woodside
00:10
00:09
00:09
Triboro Jct
—
00:12
00:11
Forest Hills
00:15
00:15
00:13
Kew Gardens
00:17
00:17
00:15
Jamaica
00:22
00:19
00:17
Merrick Blvd
—
00:21
00:19
Hollis
00:29
00:24
00:21
Queens Village
00:31
00:26
00:23
Bellerose
00:35
00:28
00:25
Floral Park
00:38
00:30
00:27
Stewart Manor
00:41
00:32
00:29
Nassau Blvd
00:44
00:34
00:31
Garden City
00:46
00:36
00:33
Country Life Press
00:49
00:38
00:35
Hempstead
00:52
00:40
00:37
East Garden City
—
00:38
00:35
Nassau Center
—
00:40
00:37
Providing peak service every 10 minutes to each of Hempstead and Nassau Center requires 20 trainsets, regardless of whether they are existing LIRR equipment or faster, lighter European trainsets.
The background to the Boston regional rail schedule is that corona destroyed ridership. In December of 2020, the counts showed ridership was down by about an order of magnitude over pre-crisis levels. American commuter rail is largely a vehicle for suburban white-collar commuters who work in city center 9 to 5; the busiest line in the Boston area, the Providence Line, ran 4 trains per hour at rush hour in the peak direction but had 2- and 2.5-hour service gaps in the reverse-peak and in midday and on weekends. Right now, the system is on a reduced emergency timetable, generally with 2-hour intervals, and the trains are empty.
But as Americans get vaccinated there are plans to restore some service. How much service is to run is up in the air, as is how it’s to be structured. Those plans may include flattening the peak and going to a clockface schedule, aiming to start moving the system away from traditional peak-focused timetables toward all-day service, albeit not at amazing frequency due to budget limits.
The plan I’ve been involved with is to figure out how to give most lines hourly service; a few low-ridership lines may be pruned, and the innermost lines, like Fairmount, get extra service, getting more frequency than they had before. The reasoning is that the frequency that counts as freedom is inversely proportional to trip length – shorter trips need more frequency and shorter headways, so even in an environment of austerity, the Fairmount Line should get a train every 15 or 20 minutes.
Optimization
In an environment of austerity, every resource counts. We were discussing individual trains, trying to figure out what the best use for the 30th, the 35th, the 40th trainset to run in regular service is. In all cases, the point is to maximize the time a train spends moving and minimize the time it spends collecting dust at a terminal. However, this leads to conflict among the following competing constraints:
At outer terminals like Worcester and Lowell, it is desirable that the train should have a timed transfer with the local buses.
At the inner terminals, that is South and North Stations, it is desirable that all trains arrive and depart around the same time (“pulse“), to facilitate diagonal transfers, such as from Fitchburg to Salem or from Worcester to Brockton.
Some lines have long single-track segments; the most frustrating is the Worcester Line, which is in theory double-track the entire way but in practice single-track through Newton, where only the nominally-westbound track has platforms.
The lines should run hourly, so ideally the one-way trip time should be 50 minutes or possibly 80 minutes, with a 10-minute turnaround.
Unfortunately, it is not possible to satisfy all constraints at once. In an environment with some avenues for investment, it’s possible to double-track single-track bottlenecks, as the MBTA is already planning to do for Newton in the medium run. It’s also possible to speed up lines on the “run as fast as necessary” principle to ensure the trips between knots take an integer or half-integer multiple of the headway; in our higher-investment regional rail plan for Worcester, this is the case, and all transfers and overtakes are tight. However, in a no-investment environment, something has to give. The Worcester Line is 90 minutes end-to-end all-local, and the single-track section is between around 15 and 30 minutes out of South Station, which means it is not possible to conveniently pulse either at South Station with the other commuter lines or at Worcester with the buses. But thankfully, the length of the single-track segment between the crossovers is just barely enough to allow bidirectional local service every 30 minutes.
Discussion
No-investment and low-investment plans are great for highlighting what the most pressing investment needs are. In general Boston needs electrification and high platforms everywhere, as do all other North American commuter lines; it is unfortunate that not a single system has both everywhere, as SEPTA is the only all-electric system and the LIRR (and sort of Metro-North) is the only all-high-platform system. However, more specifically, there are valuable targets for early investment, based on where the seams in the system are.
In the case of integrated timetabling, it’s really useful to be able to make strategic investments, including sometimes in concrete. They should always be based on a publicly-communicated target timetable, in which all the operational constraints are optimized and resolved for the maximum benefit of passengers. For example, in the TransitMatters Regional Rail plan, the timed transfers at the Boston end are dealt with by increasing frequency on the trunk lines to every 15 minutes, at which point the average untimed transfer is about as good as a timed hourly transfer in a 10-minute turnaround; this is based on expected ridership growth as higher frequency and the increase in speed from electrification and high platforms both reduce door-to-door trip times.
The upshot is that austerity is not good for efficiency. Cutting to grow is difficult, because there are always little seams that require money to fix, even at agencies where overall spending is too high rather than too low. Sometimes the timetables are such that a speedup really is needed: Switzerland’s maxim on speed is to run as fast as necessary, not as fast as trains ran 50 years ago with no further improvement. This in turn requires investment – investment that regularly happens when public transportation is run well enough to command public trust.
Subways can be built in two ways: cut-and-cover, and bored tunnel. Cut-and-cover means opening up the street top-down, building the system, and roofing it to restore surface traffic; bored tunnel means opening up one portal and digging horizontally, with less surface disturbance. In the last generation or two there has been a shift toward bored tunnel even in places that used to build cut-and-cover, despite the fact that bored tunnel is the more expensive technique in most cases. Regrettably, people don’t seem to even recognize it as a tradeoff, in which they spend more money to avoid surface disruption – some of our sources have told us that avoiding top-down cut-and-cover is an unalloyed good, a kind of modernity. Even more regrettably, this same thinking is common in much of the developing world, where subways tend to be bored.
What are cut-and-cover and bored tunnel?
Cut-and-cover refers to a family of construction techniques all of which involve top-down tunneling. In New York, one of the sources cited on NYCSubway.org refers to the subway as “a covered trench” rather than a real tunnel. The oldest cut-and-cover subways were dug by hand, but in the last 100 years there have been technological innovations to mechanize some of the work as well as to reduce surface disruption, which is considerable and lasts for a few years. These innovations include the cover-and-cut system invented in 1950s Milan (“Milan method”) and the caisson system used to build T-Centralen in Stockholm. The Milan method sinks piles into the street early and builds retaining walls to allow for truly vertical construction, whereas traditional cut-and-cover must be sloped, which requires a wider street than the tunnel, like the Manhattan avenues or Parisian boulevards but not Milan’s Renaissance streets. The caisson method builds a concrete structure and then lowers it into the ground, which facilitates multistory cut-and-cover structures at transfer stations.
Bored tunnel involves digging just one portal, or sometimes a few to speed up work, and then drilling horizontally. This used to be called a tunneling shield, but the shield has been automated to the point that a small crew, only 8-12 people, are required to supervise it nowadays, and now it is called a tunnel-boring machine, or TBM. This method was first invented in London for the construction of the Thames Tunnel, and has been used for all of the London Underground lines since the first two, as London lacks for wide streets for cut-and-cover work. Most American, European, and East Asian cities have switched to this method in the last generation; thus for example New York started to build Second Avenue Subway in the 1970s cut-and-cover, but the program since the 1990s has always been bored.
The typical method used in the world is really a mix – the tunnels are bored, the stations are cut-and-cover. This is because, while the TBM is capable of building tunnels easily, it cannot build stations. Mining or blasting a station is expensive, and many modern examples run up to $500 million or more, not just in high-cost New York but also in otherwise low-cost Rome. This mixed method involves opening up the street at station sites for 1.5-2 years in Paris, intermediate costs, and disruption only at sites that would benefit from the opening of a station.
How much do these techniques cost?
The cost of a mined station starts at $500 million and goes up. But very few cities mine stations – New York and London do, and very rarely other cities do in constrained historic centers like Rome’s. The typical cost of bored tunnel is much less; the lines for which we have seen a breakdown in costs between tunneling and stations, which are a small fraction of our database, have tunneling costs ranging from around $50 million per km to somewhat more than $100 million per km, not counting systems, overheads, or stations. With everything included, this should be viewed as about $200 million per km; the actual median for subways in our database is about $250 million/km, but it includes expensive lines with mined stations, city center tunnels that can’t easily build cut-and-cover stations, and projects that are unusually bad.
Cut-and-cover is generally cheaper. The only cut-and-cover example in our database from Paris, the Line 13 extension to Courtilles, cost 83M€/km, which is around $130 million/km in today’s money; other Paris Métro extensions from the last 15 years are 50-100% more expensive, and the next tranche is even costlier, as Parisian costs are regrettably increasing. Low-cost cities in Southern Europe bore the majority of their subways, but their suburban subway extensions are often a mix of TBMs and cut-and-cover, which is one of many reasons they have low construction costs and Paris does not.
Bear in mind that the superiority of cut-and-cover to bored tunnel depends on the presence of an at least moderately wide straight street for it to go under. London ran out of such streets after it built the Metropolitan line; the District line was, per Wikipedia, three times as expensive, about $110 million/km in today’s money, because it needed to demolish property in Kensington, already then an expensive neighborhood. New York used bored tunnel to cross under rivers and under the hills of Washington Heights, switching to cut-and-cover elsewhere; readers who have gone to the New York Subway Museum will remember the exhibits about the dangerous work of the sandhogs underwater. However, that bored tunnel was no more expensive in turn-of-the-century London than cut-and-cover was in contemporary Paris and New York does not mean these relative costs persist today. Today, on the sort of streets most cities build subways under, cut-and-cover is cheaper, by a factor that appears to be 1.5-2.
The situation in developing countries
In developing countries, I am not aware of any cut-and-cover, which does not mean there isn’t any, just that in the places I’ve looked most closely, namely India and Thailand, the tunnels seem bored. Of note, both India and Thailand build extensive elevated networks, so their subways are to some extent built where elevated construction is infeasible or undesirable. However, to some extent is doing a lot of work here. The Bangkok MRT goes under Rama IV Road, which is about 35 meters wide, and under Asok, which is 30 meters wide. This is comparable to the Sukhumvit, a 35-meter-wide road that hosts the BTS el. Deep-level construction is not necessary on the main roads of Bangkok.
What of other developing-world cities? Bangkok may be unusual, in that it’s a solidly middle-income city, the dominant capital of a middle-income country with comparable GDP per capita to China. What of genuinely poor cities? At least in the bigger ones, wide boulevards for cut-and-cover are not in shortage. Nairobi has vast roads hosting matatu routes. Lagos has such wide main roads that when I crayoned it I proposed that the main radials be elevated, as the under-construction Blue Line is, to avoid having to tunnel underwater from the mainland to Lagos Island. In most cases, short bored segments may be needed, or else short segments that involve the purchase and demolition of private property, as happened in New York when the city carved Seventh Avenue South and Sixth Avenue through the Village.
I suspect the reason this is not done is that planners believe that TBMs are more modern. The physical TBM is an engineering marvel, and looks like advanced technology, even if what it produces is comparable in quality to what cut-and-cover could do when there are wide roads to tunnel under. Planners in the United States have treated it as a given that it’s better to avoid top-down construction. This isn’t even isomorphic mimicry, in which poor countries improperly imitate rich ones; this is proper imitation of a technique whose use in rich countries too is often in error.
Cut-and-cover is underrated
Instead of tunneling wherever possible, I would urge urban subway planners to look to cut-and-cover more. In poor countries, it can be done with the same labor-intensive techniques that produced $40 million/km subways (in today’s money) in New York and Paris. In rich ones, it can be done with more advanced technology to save labor and keep costs under control. This involves more surface disruption, but this disruption can be mitigated by using the Milan method on roads that are wider than those of the center of Milan, and the ultimate benefit is that a lot more subway can be built.
Robert Jackel asked me an excellent question in comments: what is a pulse? I’ve talked about timed transfers a lot in the last almost 10 years of this blog, but I never wrote a precise definition. This is a critical tool for every public transportation operation with more than one line, making sure that trains and buses connect with as short a transfer window as possible given other constraints. Moreover, pulse-oriented thinking is to plan capital investment and operations to avoid constraints that make transfers inconvenient.
When are pulses needed?
Passengers perceive the disutility of a minute spent transferring to be more than that of a minute spent on a moving vehicle. This is called the transfer penalty and is usually expressed as a factor, which varies greatly within the literature. In a post from 2011 I quoted a since-linkrotted thesis with pointers to Boston and Houston’s numbers, and in a more recent post I found some additional literature in a larger variety of places, mostly in the US but also the Netherlands. The number 2 is somewhere in the middle, so let’s go with this.
Observe that the transfer penalty measured in minutes and not in a factor is, naturally, larger when service runs less frequently. With a factor of 2, it is on average equal to the headway, which is why it is likely the number is 2 – it represents actual time in the worst case scenario. The upshot is that the value of an untimed transfer is higher the higher the frequency is.
I used the principle of untimed transfers and frequency to explain why small subway networks do not look like small bus networks – they have fewer, more frequent lines. Subway lines that run every 3-4 minutes do not need transfer timing, because the time cost of an untimed transfer is small compared to the likely overall trip time, which is typically in the 15-30 minute range. But the lower the frequency, the more important it is to time transfers. Thus, for example, Berlin times the U6/U7 transfer at Mehringdamm in the evening, when trains run every 10 minutes, but does not do so consistently in the daytime, when they run every 5.
But note: while the value of an untimed transfer is higher at higher frequency, the value of a timed transfer is the same – it is zero-penalty or close to it no matter what. So really, the relative value of timing the transfer decreases as frequency increases. But at the same time, if frequency is higher, then more passengers are riding your service, which justifies more investment to try to time the transfer. The German-speaking planning tradition is the most concerned with transfer timing, and here, it is done commonly at 10 minutes, occasionally at 5 minutes, and never that I know of at higher frequency.
Easy mode: one central station
If all your buses and trains serve one transit center, then a pulse means that they all run at the same frequency, and all meet at the center at the same time. This doesn’t usually happen on urban rail networks – a multi-line urban rail system exists in a high-ridership, high-frequency context, in which the value of serving a mesh of city center lines is high, and the cost of bringing every subway tunnel to one location is high. Instead, this happens on buses and on legacy regional rail networks.
The pulse can be done at any frequency, but probably the most common is hourly. This is routine in small American towns with last-resort bus networks serving people too poor or disabled to drive. Two and a half years ago a few of us on Transit Twitter did a redesign-by-Twitter of the Sioux City bus network, which has ten bus routes running hourly, all pulsing in city center with timed connections. A similar network often underlies the night buses of a larger city that, in the daytime, has a more complete public transport network, such as Vancouver.
Even here, planners should keep two delicate points in mind. First, on buses in mixed traffic, there is an upper limit to the frequency that can be timetabled reliably. The limit depends on details of the street network – Jarrett Walker is skeptical that timetabling buses that run every 15 minutes is feasible in a typical American city, but Vancouver, with no freeways within a city and a rich arterial grid, manages to do so every 12 minutes on 4th Avenue. A half-hourly pulse is definitely possible, and even Jarrett writes those into his bus redesigns sometimes; a 20-minute pulse is probably feasible as well even in a typical American city. The current practice of hourly service is not good, and, as I point out in the Sioux City post, involves slow, meandering bus routes.
The second point is that once the takt is chosen, say half an hour, the length of each roundtrip had better be an integer multiple of the takt, including a minimal turnaround time. If a train needs 5 minutes to turn, and runs half-hourly, then good times for a one-way trip from city center are 10, 25, 40, 55 minutes; if there is no turnaround at city center, for example if there is through-running, then half as many turnarounds are needed. This means that short- and long-term planning should emphasize creating routes with good trip times. On a bus, this means straightening meanders as needed, and either extending the outer end or cutting it short. On a train, this means speedup treatments to run as fast as necessary, or, if the train has a lot of spare time, opening additional infill stops.
The issue of branching
Branches and pulses don’t mix well. The ideal way to run a system with a trunk and branches is to space the branches evenly. The Berlin S-Bahn runs every 3-4 minute on the Stadtbahn trunk and on the North-South Tunnel, mixing services that run every 10 and 20 minutes at roughly even intervals. In such an environment, timed transfers in city center are impossible. This is of course not a problem given Stadtbahn headways, but becomes serious if frequency is sparser. A one-trunk, two-branch regional rail system’s planners may be tempted to run each branch every half hour and interpolate the schedules to create a 15-minute headway on the trunk, but if there’s a half-hourly pulse, then only one branch can participate in it.
This is visible when one compares S-Bahn and RegionalBahn systems. High-frequency S-Bahn systems don’t use timed transfers in city center, because there is no need. I can get from Jannowitzbrücke to Ostkreuz without consulting a schedule, and I would get to the Ring without consulting a schedule either, so there is no need to time the crossing at Ostkreuz. There may be sporadic transfer timing for individual branches, such as between the S9 branch of the Stadtbahn, which diverts southeast without serving Ostkreuz, and the Ring, but S9 runs every 20 minutes, and this is not a pulse, only a single-direction timed connection.
In contrast, RegionalBahn systems, running at longer ranges and lower frequencies, often tend toward timed transfers throughout. The tradeoff is that they don’t overlie to create high-frequency trunks. In some cases, trains on a shared trunk may even platoon, so that all can make the same timed transfer, if high trunk frequency is not desired; this is how intercity trains are run on the Olten-Bern line, with four trains to a platoon every 30 minutes.
Medium mode: dendritic networks
A harder case than the single pulse is the dendritic network. This means that there is a central pulse point, and also secondary pulse points each acting as a local center. All cases I am aware of involve a mainline rail network, which could be S-Bahn rather than RegionalBahn, and then bus connections at suburban stations.
Already, this involves more complex planning. The reason is that the bus pulse at a suburban station must be timed with trains in both directions. Even if planners only care about connections between the suburban buses and trains toward city center, the pulse has to time with inbound trains for passengers riding from the suburban buses to the city and with outbound trains for passengers riding from the city to the buses. This, in turn, means that the trains in both directions must arrive at the station at approximately the same time. A few minutes of leeway are acceptable, since the buses turn at city center so the connection always has a few minutes of slack, but only a few minutes out of what is often a half-hourly takt.
Trains that run on a takt only meet every interval equal to half the takt. Thus, if trains run half-hourly, they can only have suburban pulses every 15 minutes of travel. This requires planners to set up suburban pulses at the correct interval, and speed up or sometimes slow down the trains if the time between suburban nodes. Here is an example I’ve worked on for a Boston-Worcester commuter train, with pulses in both Framingham and Worcester.
Hard mode: meshes
The next step beyond the dendritic network is the multi-node network whose graph is not simply connected. In such a network, every node must have a timed transfer, which imposes considerable planning constraints. Optimizing such a network is an active topic of research in operations and transportation in European academia.
Positive examples for such networks come from Switzerland. Large capital investments are unavoidable, because there’s always going to be some line that’s slower than it needs to be. The key here is that, as with dendritic networks, nodes must be located at consistent intervals, equal to multiples of half the headway, and usually the entire headway. To make multiple timed transfers, trains must usually be sped up. This is why pulse-based integrated timed transfer networks require considerable planning resources: planning for rolling stock, infrastructure, and the timetable must be integrated (“the magic triangle”) to provide maximum convenience for passengers connecting from anywhere to anywhere.
I wrote a long thread about regional rail and population density, and I’d like to explain more and give more context. The upshot is that higher population density makes it easier to run a rail network, but the effects are most visible for regional rail, rather than either urban rail or high-speed intercity rail. This is visible in Europe when one compares the networks in high-density Germany and low-density Sweden, and has implications elsewhere, for example in North America. I stress that high-speed rail is not primarily affected by background density, but only by the populations of cities within a certain range, and thus France, which has one of Western Europe’s lowest densities, manages to have high per-capita ridership on the TGV. However, the density of a regional mesh comes from background density, which is absent in such countries as France, Sweden, and Spain.
What is density?
Population density is population divided by area. This post is concerned with overall density at the level of an entire country or region, rather than the more granular level of the built-up urban area of a single city. What this means is that density is in large part a measurement of how close cities are to one another. In a high-density area like western Germany, Northern Italy south of the Alps, England, or the Low Countries, cities are spaced very close together, and thus people live at densities surpassing 300/km^2. In contrast, low-density areas have isolated cities, like Sweden, Australia, Canada, or the Western United States.
For example, take Stockholm. The region has about 2.5 million people, and has a strong urban and suburban rail network. However, there just aren’t a lot of cities near Stockholm. The nearest million-plus metro areas are Oslo, Gothenburg, and Helsinki, all about 400 km away, none much bigger than 1 million; the nearest 2 million-plus metro area is Copenhagen, 520 km away. The region I use as an example of German polycentrism, Rhine-Neckar, is about the same size as Stockholm, and has a good deal more suburban sprawl and car usage. The nearest million-plus region to Mannheim is Karlsruhe, 55 km away; it is a separate metropolitan area even though the Rhine-Neckar S-Bahn does have an hourly train to Karlsruhe. Frankfurt is 70 km away. A 400 km radius from Mannheim covers nearly the entirety of Germany, Switzerland, and the Low Countries; it reaches into Ile-de-France and into suburbs that share a border with Amsterdam. A 520 km radius covers Paris, Berlin, Hamburg, Milan, and Prague, and reaches close to Vienna.
Density and regional rail
Kaiserslautern is a town of 100,000 people, served by the Rhine-Neckar S-Bahn every half hour even though it is not normally seen as part of the Rhine-Neckar region. It has, in addition to the east-west S-Bahn, independent regional lines reaching north and south. When I visited two years ago, I saw these lines pulse while waiting for my delayed TGV back home to Paris.
This is viable because there are towns ringing Kaiserslautern, close enough that a low-speed regional train could connect them, with their own town centers such that there is a structure of density around their train stations. This in turn exists because the overall population density in Germany is high, even in Rhineland-Pfalz, which at 206/km^2 is slightly below the German average. The alternative structure to that of Germany would have fewer, larger cities – but that structure lends itself well to regional rail too, just with fewer, thicker lines running more frequently. If those smaller towns around Kaiserslautern did not exist but people instead lived in and right around Kaiserslautern, then it would be a city of about 400,000, and likewise Mainz might have 500,000 and the built-up area of Mannheim would have more people in Mannheim itself and in Ludwigshafen, and then there would be enough demand for a regional train every 10-20 minutes and not just every half hour.
I bring up Sweden as a low-density contrast, precisely because Sweden has generally well-run public transport. Stockholm County’s per capita rail ridership is higher than that of any metropolitan area of Germany except maybe Berlin and Munich. Regional rail ridership in and around Stockholm is rising thanks to the opening of Citybanan. Moreover, peripheral regions follow good practices like integrated intermodal ticketing and timed transfers. And yet, the accretion of a mesh of regional lines doesn’t really exist in Sweden. When I visited Växjö, which is not on the main intercity line out of Stockholm, I had a timed connection at Alvesta, but the timetable there and at Växjö looked sporadic. Växjö itself is on a spur for the network, but poking around the Krösatågen system it doesn’t look like an integrated timed transfer system, or if it is then Alvesta is not a knot. I was told in the replies on Twitter that Norrbotten/Västerbotten has an integrated network, but it runs every 2 hours and one doesn’t really string regional rail lines together to form longer lines the way one does in Germany.
Integrated regional networks
The integrated timed transfer concept, perfected in Switzerland, is ideal for regional and intercity networks that form meshes, and those in turn require high population density. With these meshes, regional rail networks overlap, underlaying an intercity network: already one can get between Frankfurt and Stuttgart purely on lines that are branded as S-Bahn, S-Bahn-like, or Stadtbahn, and if one includes RegionalBahn lines without such branding, the network is nationally connected. Even in Bavaria, a state with lower density than the German average, nearly all lines have at least hourly service, and those form a connected network.
It’s perhaps not surprising that Italy, which has high density especially when one excludes unpopulated alpine areas, is adopting German norms for its regional rail. As in Germany, this originates in urban networks, in Italy’s case that of Milan, but Trenord operates trains throughout Lombardy, most of whose population is not the built-up area of Milan, and even lines that don’t touch Milan run hourly, like Brescia-Parma. Italy is not unusual within Southern Europe in looking up to Germany; it’s only unusual in having enough population density for such a network..
Once the network is in place, it is obligatory to run it as an integrated timed transfer system. Otherwise, the connections take too long, and people choose to drive. This in turn means setting up knots at regular intervals, every 30 minutes for a mixed hourly and half-hourly system, and investing in infrastructure to shorten trip times so that major cities can be knots.
The concept of the knot is not just about regional service – high-speed rail can make use of knots as well. Germany has some low-hanging fruit from better operations and under-construction lines that would enable regularly spaced knots such as Frankfurt, then Mannheim, then Stuttgart, and far to the north Hanover and then Bielefeld. The difference is that Germany’s ideal high-speed rail network has around 20 knots and its existing regional rail network has about as many in Hesse alone. Nor can regional rail networks expect to get away with just building strong lines and spamming frequency on those, as the Shinkansen does – regional rail uses legacy alignments to work, generating value even out of lines that can only support an hourly train, whereas high-speed lines need more than that to be profitable.
Globally, the lowest-hanging fruit for such a system is in the Northeastern United States, followed by China and India. Population density in the Northeast is high, and cities have intact cores near their historic train stations. There is no excuse not to have a network of regional lines running at a minimum every 30 minutes from Portland down to Northern Virginia and inland to Albany and Harrisburg.
A few modifications to the basic Swiss system are needed to take into account the fact that the Northeast Corridor, run at high speeds, would fill a train every 5 minutes all day, and the core regional lines through New York could as well. But regional rail is not a country bumpkin mode of transportation; it works fine within 100 km of Frankfurt or Milan, and should work equally well near New York. If anything, a giant city nearby makes it easier to support high frequency – in addition to internal travel within the regional system, there are people interested in traveling to the metropole helping fill trains.
What about low-density places?
Low-density places absolutely can support good rail transport. But it doesn’t look like the German mesh. Two important features differ:
It is not possible to cobble together a passable intercity rail network from regional express lines and upgrade it incrementally. Intercity lines run almost exclusively intercity traffic. This tilts countries toward the use of high-speed rail, including not just France but also Spain and now Sweden. This does not mean high-density countries can’t or shouldn’t build high-speed rail – they do successfully in Asia, Italy has a decent network, Britain has high-speed rail plans, and Germany is slowly building a good network. It just means that high-density countries can get away with avoiding building high-speed rail for longer.
The connections between regional and intercity lines are simpler. Different regions’ suburban networks do not connect, and can be planned separately, for example by state-level authorities in Australia or provincial ones in Canada. These networks are dendritic: intercity lines connect to regional lines, and regional lines branch as they leave city center. Lines that do not enter the primary city center are usually weaker, since it’s unlikely that there are enough strong secondary centers at the right places that a line could serve them well without passing through the primary center.
In extreme cases, no long-distance rail is viable at all. Australia is a borderline case for Brisbane-Sydney-Melbourne high-speed rail – I think it’s viable but only based on projections of future population and economic growth. But Perth and Adelaide are lost causes. In the United States, railfans draw nationally-connected proposals, but in the Interior West the cities are simply too far apart, and there is no chance for a train to usefully serve Denver or Salt Lake City unless cars are banned. Connecting California and the Pacific Northwest would be on the edge of viable if the topography were flat, but it isn’t and therefore such a connection, too, is a waste of money in the economic conditions of the early 21st century.
Note that even then, cities can have suburban rail networks – Perth and Adelaide both have these, and their modal splits are about on a par with those of secondary French cities like Nice and Bordeaux or secondary American transit cities like Boston and Chicago. Denver is building up a light rail and a commuter rail network and one day these networks may even get ridership. The difference between the case of Perth or Denver and that of a German city is that Perth and Denver can rest assured their regional rail alignments will never be needed for intercity rail.
In less extreme cases, intercity trains are viable, and can still run together with regional trains on the same tracks. California is one such example. Its population density and topography is such that planning regional rail around the Bay Area and in Los Angeles can be kept separate, and the only place where intercity and regional trains could work together as in Germany is the Los Angeles-San Diego corridor. Blended planning with timed overtakes is still recommended on the Peninsula, but it’s telling that at no point have Bay Area-based reformers proposed a knot system for the region.
Those less extreme low-density cases are the norm, in a way. They include the Midwestern and Southern US, the Quebec-Ontario corridor, the Nordic countries, France, nearly all of Eastern Europe, and Southern Europe apart from Italy; this is most of the developed world already. In all of those places, regional rail is viable, as is intercity rail, but they connect in a dendritic and not meshlike way. Many of the innovations of Germany and its penumbra, such as the takt and the integrated intermodal plan, remain viable, and are used successfully in Sweden. But the exact form of regional rail one sees in Germany would not port.
A country or region that is good at manufacturing cars can export them globally and earn hard cash. But what about public transportation? How can a city that has the ability to build good, low-cost public transport get rich off of it? There is an answer, but it is more complicated than “export this,” mirroring the fact that public transport itself is a more complex system to run than cars. This in turn relates to housing growth rates and urban economies of scale, making this the most useful in a large city with high housing production rates, of which the best example is Seoul. The good news is that the world’s largest and richest cities could gain tremendously if they had better public transport as well as high housing growth rates.
Infrastructure is not exportable
I wrote more than two years ago about the difference between dirty and clean infrastructure. Cars, car parts, and oil are exportable, so the majority of the cost of cars as a system are exportable, making dedicated regions like Bavaria, Texas, and the Gulf states rich. Green tech is not like that – the bulk of the cost is local labor. A large majority of the operating costs of a subway system are local wages and benefits; in New York, depreciation on rolling stock is less than 10% of overall operating costs. Construction costs are likewise almost entirely local labor and management, which is why they are determined by where the project takes place, rather than by which engineering firm builds the project.
The upshot is that Madrid and other low-cost cities can’t just get rich by building other cities’ infrastructure for them. They can’t build turnkey systems for New York and London at Spanish prices – the problems with New York and London come from local standards, management, and regulations, and while a Spanish engineering firm could give valuable advice on what high-cost cities need to change, it’s not going to reap more than a fraction of the construction cost saving in consulting fees.
Good transit as an amenity
What a city can do with low-cost construction is build a large subway network like Madrid, and use that as infrastructure to help local economic production. This works as both a consumption amenity and a production amenity. As a consumption amenity, it enables people to commute without needing to own a car, which reduces living costs and lets employers get away with paying less in nominal terms; this is a bigger influence on local firms, because international ones tend to use cost of living adjustments that make profligate lifestyle assumptions and factor in car costs even in cities where car ownership is low, like Singapore or New York.
As a production amenity, public transit also enables work concentration in city centers. This is separate from the observation that it allows workers to commute more cheaply – if a large city produces in a concentrated center, then without rapid transit, workers can’t get in at all. About 23% of people entering the Manhattan core on a weekday do so by car per the Hub Bound Report, but at the peak hour, 8-9 am, this falls to 9%, because the road capacity is capped around 55,000 cars an hour and a maximum number of parking spots for them. Auto-centric cities of New York’s approximate size exist, not by building massive road capacity to support comparable city centers, but by not having strong city centers to begin with. Los Angeles has maybe 400,000 people in the widest definition of its central business district, where in the same area New York has more than 2 million – and Los Angeles’s secondary centers, like Century City, top in the mid-5 figures before they get completely choked with traffic.
So what a city can do with cheap infrastructure is build a large subway network and support a large high-rise central business district and then use that to produce more efficiently. This is possible, but more complex than just exporting cars or oil, because to export cars one just needs to be good at making cars, and to export oil one just needs to have oil underground, whereas to produce out of public transit one also needs a solid economy in other sectors that can make use of the better infrastructure. I suspect that this is why Southern Europe keeps not growing economically despite building high-quality public transport – the Madrid Metro is great but there isn’t enough of a private economy to make use of it.
The connection with development
To maximize the use of a subway for its economy, a city needs to make sure development can follow it. This means that city center needs high job density, which includes high-rise office towers at the busiest intersections, and many mid-rise office buildings in a radius of a few kilometers. Neither the typical European pattern in which there are few skyscrapers nor the American pattern in which there are skyscrapers for a few blocks and then the rest of the city is subject to strict residential zoning is ideal for this. It’s better to have a city whose central few square kilometers look like Midtown and whose surrounding few tens of square kilometers look like Paris, with the occasional secondary cluster of skyscrapers at high-demand nodes; let’s call this city “Tokyo.”
Residential development has to keep up as well. A city region that has a strong private economy but doesn’t build enough housing for it will end up with capped production. Normally it’s the lowest-end jobs that get exported. However, two problems make it more than a marginal reduction in production. First, expensive cities have political pressure to allocate apartments by non-market processes like rent control, keeping less productive but politically favored people; a large gap between market rent and construction costs creates plenty of surplus to extract, and a mass exodus of firms from cities like San Francisco in such a situation starts from thee least profitable ones, and by the time it affects the most profitable on, the system is entrenched. And second, breaking a firm’s chain between high-end headquarters jobs in a rich city center and lower-end subsidiary jobs elsewhere reduces firmwide productivity, since many connections have to be remote; Google has problems with all-remote teams and tries to center teams in the Bay Area when it gets too unwieldy.
For one example of a city that does everything right, look at Seoul. It has low construction costs, around $150 million per kilometer for urban subways. Thanks to its low costs and huge size, it keeps building up its system even though it already has one of the largest systems in the world, probably third in ridership after Tokyo and Osaka when one includes all commuter lines. It also has high density, high-rise CBDs, and fast housing construction; in 2019 the Seoul region built around 10 units per 1,000 people, representing a decline since the mid-2010s, and the state has plans to accelerate construction, especially in the city, to curb rising prices. This is till a better situation than the weak economy and flagging construction in much of Europe, or the NIMBY growth rates of both much of the rest of Europe and the richest American cities.
Pedestrian Observations 
