Amtrak Releases Bad Scranton Rail Study
There’s hot news from Amtrak – no, not that it just announced that it hired Andy Byford to head its high-speed rail program, but that it just released a study recommending New York-Scranton intercity rail. I read the study with very low expectations and it met them. Everything about it is bad: the operating model is bad, the proposed equipment is bad and expensive, the proposed service would be laughed at in peripheral semi-rural parts of France and Italy and simply wouldn’t exist anywhere with good operations.
This topic is best analyzed using the triangle of infrastructure, rolling stock, and schedule, used in Switzerland to maximize the productivity of legacy intercity line, since Swiss cities, like Scranton, are too small to justify a dedicated high-speed rail network as found in France or Japan. Unfortunately, Amtrak’s report falls short on all three. There are glimpses there of trying and failing, which I personally find frustrating; I hope that American transportation planners who wish to imitate European success don’t just read me but also read what I’ve read and proactively reach out to national railways and planners on this side of the Atlantic.
What’s in the study?
The study looks at options for running passenger trains between New York and Scranton. The key piece of infrastructure to be used is the Lackawanna Cutoff, an early-20th century line built to very high standards for the era, where steam trains ran at 160 km/h on the straighter sections and 110 km/h on the curvier ones. The cutoff was subsequently closed, but a project to restore it for commuter service is under construction, to reach outer suburbs near it and eventually go as far as the city’s outermost suburbs around the Delaware Water Gap area.
Amtrak’s plan is to use the cutoff not just for commuter service but also intercity service. The cutoff only goes as far as the Delaware and the New Jersey/Pennsylvania state line, but the historic Lackawanna continued west to Scranton and beyond, albeit on an older, far worse-built alignment. Thus, the speed between the Water Gap and Scranton would be low; with no electrification planned, the projected trip time between New York and Scranton is about three hours.
I harp on the issue of speed because it’s a genuine problem. Google Maps gives me an outbound driving time of 2:06 right now, shortly before 9 pm New York time. The old line, which the cutoff partly bypassed, is curvy, which doesn’t just reduce average speed but also means a greater distance must be traversed on rail: the study quotes the on-rail length as 134 miles, or 216 km, whereas driving is just 195 km. New York is large and congested and has little parking, so the train can afford to be a little slower, but it’s worth it to look for speedups, through electrification and good enough operations so that timetable padding can be minimized (in Switzerland, it’s 7% on top of the technical travel time).
The operations and timetabling in the study are just plain bad. There are two options, both of which include just three trains a day in each direction. There are small French, Italian, and Spanish towns that get service this poor, but I don’t think any of them is as big as Scranton. Clermont-Ferrand, a metro area of the same approximate size as Scranton, gets seven direct trains a day to Paris via intermediate cities similar in size to the Delaware Water Gap region, and these are low-speed intercities, as the area is too far from the high-speed network for even low-speed through-service on TGVs. In Germany and Switzerland, much smaller towns than this can rely on hourly service. I can see a world in which a three-hour train can come every two hours and still succeed, even if hourly service is preferable, but three roundtrips a day is laughable.
Then there is how these three daily trains are timetabled. They take just less than three hours one-way, and are spaced six hours apart, but the timetable is written to require two trainsets rather than just one. Thus, each of the two trainsets is scheduled to make three one-way trips a day, with two turnarounds, one of about an hour and one of about five hours.
Worse, there are still schedule conflicts. The study’s two options differ slightly in arrival times, and are presented as follows:
Based on the results of simulation, Options B and D were carried forward for financial evaluation. Option B has earlier arrival times to both New York and Scranton but may have a commuter train conflict that remains unresolved. Option D has later departure times from New York and Scranton and has no commuter train conflicts identified.
All this work, and all these compromises on speed and equipment utilization, and they’re still programming a schedule conflict in one of the two options. This is inexcusable. And yet, it’s a common problem in American railroading – some of the proposed schedules for Caltrain and high-speed rail operations into Transbay Terminal in San Francisco proposed the same.
Equipment and capital planning
The study does not look at the possibility of extending electrification from its current end in Dover to Scranton. Instead, it proposes a recent American favorite, the dual-mode locomotive. New Jersey Transit has a growing pool of them, the ALP-45DP, bought most recently for $8.8 million each in 2020. Contemporary European medium-speed self-propelled electric trains cost around $2.5 million per US-length car; high-speed trains cost about double – an ongoing ICE 3 Neo procurement is 34 million euros per eight-car set, maybe $6 million per car in mid-2020s prices or $5 million in 2020 prices.
And yet somehow, the six-car dual-mode trains Amtrak is seeking are to cost $70-90 million between the two of them, or $35-45 million per set. Somehow, Amtrak’s rolling stock procurement is so bad that a low-speed train costs more per car than a 320 km/h German train. This interacts poorly with the issue of turnaround times: the timetable as written is almost good enough for operation with a single trainset, and yet Amtrak wants to buy two.
There are so many things that could be done to speed up service for the $266 million in capital costs between the recommended infrastructure program and the rolling stock. This budget by itself should be enough to electrify the 147 km between Dover and Scranton, since the route is single-track and would carry light traffic allowing savings on substations; then the speed improvement should allow easy operations between New York and Scranton every six hours with one trainset costing $15 million and not $35-45 million, or, better yet, every two hours with three sets. Unfortunately, American mainline rail operators are irrationally averse to wiring their lines; the excuses I’ve seen in Boston are unbelievable.
The right project, done wrong
There’s an issue I’d like to revisit at some point, distinguishing planning that chooses the wrong projects to pursue from planning that does the right projects wrong. For example, Second Avenue Subway is the right project – its benefits to passengers are immense – but it has been built poorly in every conceivable way, setting world records for high construction costs. This contrasts with projects that just aren’t good enough and should not have been priorities, like the 7 extension in New York, or many suburban light rail extensions throughout the United States.
The intercity rail proposal to Scranton belongs in the category of right projects done wrong, not in that of wrong projects. Its benefits are significant: putting Scranton three hours away from New York is interesting, and putting it 2.5 hours away with the faster speeds of high-reliability, high-performance electric trains especially so.
As a note of caution, this project is not a slam dunk in the sense of Second Avenue Subway or high-speed rail on the Northeast Corridor, since the trip time by train would remain slower than by car. If service is too compromised, it might fail even ignoring construction and equipment costs – and we should not ignore construction or equipment costs. But New York is a large city with difficult car access. There’s a range of different trips that the line to Scranton could unlock, including intercity trips, commuter trips for people who work from home most of the week but need to occasionally show up at the office, and leisure trips to the Delaware Water Gap area.
Unfortunately, the project as proposed manages to be both too expensive and too compromised to succeed. It’s not possible for any public transportation service to succeed when the gap between departures is twice as long as the one-way trip time; people can drive, or, if they’re car-free New Yorkers, avoid the trip and go vacation in more accessible areas. And the sort of planning that assumes the schedule has conflicts and the dispatchers can figure it out on the fly is unacceptable.
There’s a reason planning in Northern Europe has converged on the hourly, or at worst two-hourly, frequency as the basis of regional and intercity timetabling: passengers who can afford cars need the flexibility of frequency to be enticed to take the train. With this base frequency and all associated planning tools, this region, led by Switzerland, has the highest ridership in the world that I know of on trains that are not high-speed and do not connect pairs of large cities, and its success is slowly exported elsewhere in Europe, if not as fast or completely as it should be. It’s possible to get away without doing the work if one builds a TGV-style network, where the frequency is high because Paris and Lyon are large cities and therefore frequency is naturally high even without trying hard. It’s not possible to succeed on a city pair like New York-Scranton without this work, and until Amtrak does it, the correct alternative for this study is not to build the line at all.
No, the Anglosphere isn’t Especially NIMBY
There’s an article going around social media on Financial Times, by John Burn-Murdoch, making the case that slow housing growth, with consequent rises in rents, is a pan-Anglosphere phenomenon. A non-paywalled summary can be found on New York Magazine by Eric Levitz, reproducing the FT graphs showing changes in the number of housing units per capita in various developed countries, and making some general comments about Anglo culture. The problem with this analysis is that it’s completely false. As someone who did once err in an analysis of the Anglo problem of high construction costs – a problem that Britain did not have until the 1990s and Canada and Australia until the 2000s or even 2010s – let me throw some cold water on this Anglo NIMBY theory.
Housing construction rates
Housing construction rates per capita show no generic Anglosphere effect. The highest rates are in Austria, the Nordic countries and Canada, New Zealand, and Australia. Here are the numbers as far as I’ve been able to find, all expressed in dwelling completions per 1,000 people in 2021:
Australia (starts): 9
New Zealand: 6.9
France (starts): 4.7
The FT article’s data mostly ends in 2020, whereas the above list is from 2021. But looking at earlier years doesn’t change much. The annual average in 2016-20, relative to 2018 and not 2021 population, was 8.2 in Australia, 5.8 in New Zealand, and 5.2 in Canada – slightly lower per capita than in 2021, and yet higher than in all comparison countries. In those other comparison countries the numbers are usually fairly stable as well going back to the mid-2010s recovery from the Great Recession; the only notable changes are in Spain, Portugal and Denmark, which saw sharp rises in construction from the mid-2010s (in Spain’s case, still a far cry from pre-Great Recession rates).
Some trends can be discerned. Southern Europe has low construction rates, owing to the poor state of its economy – but note that Europe’s top builder, Finland, was hit hard by the Great Recession, when coincidentally the smartphone revolution devastated Nokia, and took until last year to recover to its pre-recession GDP per capita. Germany builds the least in Northern Europe; Austria builds the most, for which difference I have no explanation. However, there is no trend separating the Anglosphere into its own group. The US and UK build less than most countries they’re like to be compared with, but those comparison countries include their Anglo peers.
So why does Burn-Murdoch think there’s an Anglo trend here?
Burn-Murdoch uses a different statistic from construction rates per capita. He instead looks at the rate of change in the overall number of dwellings per capita in the above countries I listed, minus Austria and Switzerland. The Anglo countries have stagnated at 400-450 dwellings per 1,000 people since the 1980s; the non-Anglo European countries have kept developing housing and are now in the 500-550 range.
The problem is that housing per capita is the wrong measure to use. It’s influenced by both housing construction rates and population growth, the latter coming from birthrates and immigration. Canada, Australia, and New Zealand are all notable for their high immigration rates, and therefore Canada and Australia have seen slow rises in dwellings per capita and New Zealand has even seen decreases. The same is true of Sweden and Norway, which build a fair amount of housing but are not seeing a large increase in the dwelling stock per capita, because people keep coming in to fill these new apartments.
Instead, on FT’s graphic of growth in housing per capita in the last 10 years, the standouts are France, Portugal, Italy, and Finland. Finland indeed builds a lot of housing, but its issue is that its weak economy in the last 15 years has not been able to attract as many immigrants as Sweden and Norway. Italy and Portugal are literally the two lowest per capita builders on this list, and have negative population growth thanks to weak economies and very low birthrates, so their per capita housing stock looks like it’s doing well.
Where is the housing built?
A real distinction, motivating YIMBY movements even in fast builders like Canada, is where the housing is built. This is an important question at both the national level and the regional level. At the national level, one should expect housing to be built where there is the most demand, typically in the richest city regions. At the regional level, one should likewise expect housing to be built in the areas with the best access to work, which can be infill near city center, or new areas opened by the construction of urban rail lines.
The links on the list above often include subnational breakdowns that one can peruse. Thus, for example, in Norway, we find that Oslo built less housing per capita than the rest of the country in 2021, only 3.7/1,000 people, but Viken, a gerrymandered county collecting Oslo’s suburbs, built more, 7.5/1,000, averaging to 6.2 regionwide. France is less certain, since my regional data is approvals and not starts or completions. In Ile-de-France in 2021, the approval rate for new dwellings was 5.9/1,000 people, with Paris itself at a pitiful 1.2, and same source gives the national rate as 7/1,000. But going a few years back, the French rate is still around 5/1,000, whereas the Francilien one is about 7/1,000 (still with little construction in the city).
A uniquely American misfeature is that while the overall rate of housing construction is below average for a growing country rather than terrible, the interregional pattern of where housing is built is awful. The richest regions of the United States don’t build very much, with the exception of Seattle. New York, the largest by far of these regions, builds well below the national average. Thus, while in stagnant Italy, Spain, and Portugal (or for that matter Japan) the rich main cities are still growing, in the United States the richest city regions have below-average population growth, which is seen at every congressional reapportionment once per decade.
But even this is not an Anglo feature: there’s a detailed local breakdown for England, and while London does build less than the rest of the country, it’s not by a large margin, about 2.5/1,000 people averaged over the last few years versus 3 overall. And in Canada, there’s a detailed local breakdown by metro area and within each such region, and there we see 2021 completion rates of 7.3/1,000 in Toronto, 4.8/1,000 in Toronto’s suburbs, 7/1,000 in Calgary, 9.1/1,000 in Edmonton, and 9.5/1,000 in Metro Vancouver (of which 9.9/1,000 were in Vancouver proper – this isn’t sprawl).
To temper my praise for Vancouver and its high growth rates, I should specify that while Canada is building housing in decent if not eye-popping quantities, in the regions where it’s most needed, it’s not building housing in the neighborhoods where it’s most needed. Metro Vancouver builds transit-oriented development on SkyTrain but not in its richest places: the West Side of the city remains strongly NIMBY, despite its excellent location between city center and UBC, forcing students into hour-long commutes; an indigenous West Side housing project built without needing to consult local NIMBYs is deeply controversial among those same NIMBYs.
That said, “housing is not built in rich urban neighborhoods” is not a national-scale statistic, nor a particularly Anglo one (very little housing is built in Paris proper). So why is it so appealing to posit NIMBYism as a uniquely Anglo problem?
The false appeal of deep roots
Middlebrow writers love talking about deep roots – that is, processes that are said to be part of a shared cultural heritage that stretches a long way back, and is therefore by implication hard to impossible to change through policy. An American bestselling book argued that the South’s political institutions come from its unique history of Scottish rather than English settlement (and not from, say, slavery) – institutions that are nowhere in sight in modern Scotland. Often (but not always!), it’s a thin veneer for racism, normalizing the idea that non-Westerners could never perform on a par; until the growth of the Asian Tigers was impossible to ignore, there was a common belief in the West that Confucianism was a deeply-rooted obstacle to growth, which now has flipped to an argument that it’s a deeply-rooted accelerator of growth.
In the case of housing, it’s therefore important to note that even in the US and UK, there’s no longstanding pattern of NIMBYism beyond what’s found in every non-city-state. The US had rapid urban growth around the turn of the century, which romantics found offensive – but that’s little different from the concurrent urbanization of Germany. Romantic and nationalistic interests fought against this urban growth throughout this era, from the 1870s to World War Two. Japan and South Korea today are famous in YIMBY circles for their high capital-region housing growth rates, but neither country is happy with its capital-centricity, and South Korea is even relocating capital functions to a new city in the far suburbs of Seoul.
There’s a real longstanding difference between London and comparable Continental cities like Paris and Berlin, in that London’s housing typology, the rowhouse, is much less dense than the mid-rise apartment blocks of the Continent. This goes back to early industrialization, when Paris, Berlin, and other Continental cities were walled for tax purposes and British cities were not. Thus, Britain evolved a culture of “gentlemen don’t live on shelves” whereas the French and German urban middle classes were happy with mid-rise apartments.
However, New York behaves in exactly the same way as Continental cities: there were historic impediments to urban sprawl coming from the width of the Hudson and East Rivers, leading to a mid-rise urban form and the now-familiar pattern in which middle-class city residents live in a single-story apartment in a multistory building (British dwellings were multistory even for the working class). And New York’s elite hated the city, fleeing to segregated suburbs more than a 100 years ago far away from Jewish and Catholic immigrants, and inventing modern zoning to keep Jews out of Fifth Avenue department stores. The city is fiercely NIMBY today, building little housing by the standards of Berlin or of Paris with its inner suburbs.
Very little of the problem of NIMBYism in either Britain or the US – or for that matter Germany – is especially deeply rooted. The US has an unusual problem with democratic deficit at the local level, which YIMBYs seek to resolve through disempowering local actors and creating national networks that push for more pro-development policy; they are starting to see some success in California. New Zealand, without federalism, imitated some of the California YIMBYs’ proposals and is seeing a wave of new construction and falling rents in parts of the country. Germany is the NIMBYest place in Northern Europe, but high rents are understood as a problem and so SPD has, in its usual slow pace, sought to embrace YIMBYism, Olaf Scholz pledging to increase the housing construction rate here from 250,000 units a year (3/1,000) to 400,000 (4.8/1,000) and the party’s next generation within Jusos openly calling themselves YIMBYs. The UK has a parliamentary casework system that lets petty actors constrain the otherwise unitary state, but not when the state makes something a priority, and so Labour runs on increasing housing production.
In fact, in the US, UK, and Germany, we’re even seeing the same political pattern emerge: in response to slow housing production and high rents, national and nationally-looking center-left forces are politicizing the issue in order to flush out urban NIMBYs, who vote center-left as well but are locally rather than nationally rooted and so have opinions out of touch with those of the median voter or party supporter. Even there, we see a difference: the UK also has center-right thinktanks pushing for the same on neoliberal grounds, and this is also seen in Canada, whereas CSU is proudly NIMBY and the Republicans are, from their origin of embracing housing construction in Texas, slowly trending that way too.
None of this is deeply-rooted or Anglo. Sometimes, social trend evolve in parallel in multiple countries. It’s easy to pattern-match this to Anglo or not; I do this for infrastructure construction costs and have to constantly remind people that until the 1990s, London built urban rail tunnels for the same per-km cost as Milan and Rome, and Canadian cities only lost their ability to build efficiently 10-20 years ago. The same is true of housing: first of all, there’s no Anglo-wide pattern at all, the UK and US differing profoundly from Canada, Australia, and New Zealand, and second of all, their shared characteristics are also shared with Germany.
Push and Pull Factors and Measuring Modal Shift
There’s a longstanding debate among activists and academics about what the best way of effecting modal shift from cars to public transport is. Pull factors concern making public transport better through building more rail lines, running them more frequently, improving service convenience, or reducing fares. Push factors concern making driving harder through speed limits, fuel taxes, congestion pricing, and reallocation of street space from cars to public and non-motorized transport. There’s a tendency on the New Left to favor push factors (but the East Asian developmental states are best characterized as push-before-pull and not pure pull).
This has been refined by researchers at the climate research institute, the Ariadne Project, who published a paper in late 2021 rating various push and pull policies on effectiveness for reducing transport emissions. They conclude that push factors dominate, and pull factors are small, with construction of new public transit almost insignificant, only worth a reduction of around 300,000 tons of CO2 a year Germany-wide, 0.039% of national emissions as of 2021; instituting a 120 km/h speed limit on the Autobahn is said to have about 10 times that effect, while the biggest effects yet would come from carbon taxes. The study laments that pull factors are so much more popular than push factors, which they admit suppress society-wide consumption.
The research suffers from the same problem as other work in this direction, in that it is bad at estimating the impact of public transport on mode shift. It briefly argues that construction of public transport increases overall consumption and therefore doesn’t do much to reduce emissions. This way, it’s like 2020’s carbon critique of U-Bahn expansion, which I criticized two months ago; the carbon critique argues that each kilometer of U-Bahn built only reduces CO2 emissions by 714 tons a year through mode shift, under the assumption that only 20% of public transport riders are diverted from cars.
This doesn’t pass a sanity check. 300,000 divided by 714 is 420 km, which is about comparable to the total route length of the four grade-separated U-Bahn systems in Germany plus the Wuppertal Schwebebahn; I think the two figures, 300,000 and 714/km, come from different sources, and judging by the other elements in the study, I suspect 300,000 assumes less construction than a full doubling of Germany’s rapid transit network length. Nonetheless, even under a more generous assumption, this is far too low compared with macro trends in public transport usage.
The best way to use macro trends as a sanity check is to look at some cases with much more and much less public transport than the present. Do they look like it’s a total difference of 0.039%? No, and that’s even taking into account that transit cities tend to be wealthier, stimulating more consumption and more production. As I pointed out in my post two months ago, while Germany averages 9.15 t-CO2/capita, Berlin only does 5.38, and while Germany averages 580 cars per 1,000 people, Berlin only does 327. The difference is largely about Berlin’s pull factors. Push factors in the city are not extensive, and what exists is implemented only in areas that already have very low car use.
Even lower household emissions in Berlin must be viewed as downstream of the density that is enabled by the presence of a large urban rail network. Cars are a low-capacity mode of transport, so an auto-oriented region, like American metro regions, has to spread out its homes and destinations to limit congestion, and this increases household emissions (single-family houses emit more than apartment buildings) and also encourages people to travel longer distances for their commute and routine non-commute trips.
This is not easy to measure. Public transport projects have gotten fairly good in the last generation at estimating ridership, but estimating the responsibility of one particular project to modal shift is hard. It interacts with the entire city region. For example, building one rail line can be measured to shift modes in the neighborhoods it serves, but it also encourages destinations to locate in city center since people from the neighborhoods the line serves can now access it, and the increase in office, retail, and community development then leads to a small modal shift citywide. Worse, trying to tease out the effect of the rail line on modal shift sufficiently carefully may lead researchers to count this citywide effect negatively, since one econometric technique is to compare the neighborhoods near the line with neighborhoods in the same city not on the line.
In practice, the construction of rail lines tends to co-occur with other policies that improve public transport, which may be pull or push factors. This means that it’s very easy to misattribute the effect of urban rail expansion to those other factors. I am convinced that this is what is happening here; the proper comparison must be at the level of an entire region, looking at the emissions of different regions with different levels of public transport usage.
The upshot is that if it is hard to measure the effect of public transport construction on modal shift and emissions, then the uncertain factors should not be set to zero. Rather, they should be set to sanity-check levels. For example, one can compare New York with the rest of the United States, since it’s a starker difference between a transit city and an auto-oriented country than anywhere in Europe, and correct for non-transport effects like climate and electricity mix, both of which are easy to measure.
Within Germany, Berlin has 42% lower emissions than the rest of the country per capita. Berlin achieves this with an urban rail network that, in 2019, got 1,289 million rail trips, nearly all within the city of 3.7 million, a minority in the suburban region of perhaps 1.3 million. This is around 250 trips/person regionwide, and 320/person citywide assigning around 20% of S-Bahn ridership to suburbs like Potsdam and Oranienburg. What’s more, Germany doesn’t start from zero; this is not the United States, with multiple large cities with around 10 annual rail trips per capita. Netting out buses from VDV’s data (p. 25) gets around 6.3 billion rail trips in Germany in 2019 including trams, or 75 per capita.
The difference between 320 and 75 is around 250 – I know it’s actually 245 but at this point I’m deliberately reducing precision because those are sanity-check estimates and I don’t want people thinking they’re correct to three significant figures (try 1.5). If we attribute the entire Berlin-Germany difference of about 3.8 t-CO2/capita to public transport and downstream changes to the urban layout, then we get 0.015 t saved per annual trip generated. To get from there to 300,000 tons saved, we just need 20 million annual rail riders, or around 65,000 daily ones, which is easy to generate on a single line; the approximately 2 km extension of U8 to Märkisches Viertel that Berlin keeps postponing is estimated to generate 25,000-30,000.
Now, to sanity-check the sanity check, the estimate here is that every trip on urban rail saves 15 kg-CO2. This is an aggressive figure; new cars nowadays average 100 g/km and averaged 180 g/km in 2001 (source, PDF-p. 15), and the average displaced car trip is not 150 km or even 80 km – Americans average around 45 km/day, or somewhat more when only adults are considered. Rather, the issue is a combination of factors:
- Because the limiting factor to car transport is capacity, in practice what happens in an auto-oriented region is that it fills from the inside outward, and any modal shift ends up displacing the outermost and longest car trips. I proposed a model for that in a blog post from four years ago.
- Public transport displaces car trips on a more than one-to-one basis (and certainly more than 20% as in the carbon critique of the U-Bahn). This is because public transport users also walk and bike, and transit cities have high modal splits for active transport by the standards of auto-oriented cities, if not by the standards of Dutch cities. Berlin’s all-trip modal split in 2018 was 26% car, 27% public transport, 18% bike, 30% walking – and the high active transport modal split exists not because of road diets, which are few and far between, but because of the presence of a large core fed by the U- and S-Bahn.
- Public transport reduces household energy usage by encouraging people to live in apartment buildings with shared walls rather than in single-family houses, which have much greater heating requirements; this is also the mechanism through which transit cities have relatively high usage of active transport even without trying very hard.
I don’t think these factors fully explain away the gap between 45 km/day and 150 km per trip (so around 300/day), but they explain a large enough fraction of it that the installation of a system like what Berlin has – or, better, what Tokyo has – should be a climate priority. If your model says it doesn’t, it needs a lot more work than to just talk about the consumption effects of more public transport (if you’re bothered by how Berlin is poor for its size, compare New York with the rest of the United States).
In fact, if estimating modal shift is hard, then it’s best to approximate it by ridership. It’s imperfect because there is the effect of walking and biking; some lines really do just compete with walking, like city-center streetcars, but usually, to first order, it’s a good enough estimate. If it’s hard to estimate the benefits then they should not be set to zero, but rather set proportionally to something easier to measure, in this case ridership. Investment should follow ridership-maximizing strategies, and only deviate from them in corner cases.
TransitCenter’s Commuter Rail Proposal
Last week, TransitCenter released a proposal for how to use commuter rail more effectively within New York. The centerpiece of the proposal is to modify service so that the LIRR and Metro-North can run more frequently to stations within the city, where today they serve the suburbs almost exclusively; at the few places near the outer end of the city where they run near the subway, they have far less ridership, often by a full order of magnitude, which pattern repeats itself around North America. There is much to like about what the proposal centers; unfortunately, it falls short by proposing half-hourly frequencies, which, while better than current off-peak service, are far short of what is needed within the city.
Commuter rail and urban ridership
TransitCenter’s proposal centers urban riders. This is a welcome addition to city discourse on commuter rail improvement. The highest-ridership, highest-traffic form of mainline rail is the fundamentally urban S-Bahn or RER concept. Truly regional trains, connecting distinct centers, coexist with them but always get a fraction of the traffic, because public transit ridership is driven by riders in dense urban and inner-suburban neighborhoods.
A lot of transit and environmental activists are uncomfortable with the idea of urban service. I can’t tell why, but too many proposals by people who should know better keep centering the suburbs. But in reality, any improvement in commuter rail service that does not explicitly forgo good practices in order to discourage urban ridership creates new urban ridership more than anything else. There just aren’t enough people in the suburbs who work in the city (even in the entire city, not just city center) for it to be any other way.
TransitCenter gets it. The proposal doesn’t even talk about inner-suburban anchors of local lines just outside the city, like Yonkers, New Rochelle, and Hempstead (and a future update of this program perhaps should). No: it focuses on the people near LIRR and Metro-North stations within the city, highlighting how they face the choice between paying extra for infrequent but fast trains to Midtown and riding very slow buses to the edge of the subway system. As these neighborhoods are for the most part on the spectrum from poor to lower middle-class, nearly everyone chooses the slow option, and ridership at the city stations is weak, except in higher-income Northeast Queens near the Port Washington Branch (see 2012-4 data here, PDF-pp. 183-207), and even there, Flushing has very little ridership since the subway is available as an alternative.
To that effect, TransitCenter proposes gradually integrating the fares between commuter rail and urban transit. This includes fare equalization and free transfers: if a bus-subway-bus trip between the Bronx and Southern Brooklyn is covered by the $127 monthly pass then so should a shorter bus-commuter rail trip between Eastern Queens or the North Bronx and Manhattan.
Interestingly, the report also shows that regionwide, poorer people have better job access by transit than richer people, even when a fare budget is imposed that excludes commuter rail. The reason is that in New York, suburbanization is a largely middle-class phenomenon, and in the suburbs, the only jobs accessible by mass transit within an hour are in Midtown Manhattan, whereas city residents have access to a greater variety of jobs by the bus and subway system. But this does not mean that the present system is equitable – rich suburbanites have cars and can use them to get to edge city jobs such as those of White Plains and Stamford, and can access the entire transit network without the fare budget whereas poorer people do have a fare budget.
The issue of frequency
Unfortunately, TransitCenter’s proposal on frequency leaves a lot to be desired. Perhaps it’s out of incrementalism, of the same kind that shows up in its intermediate steps toward fare integration. The report suggests to increase frequency to the urban stations to a train every half an hour, which it phrases in the traditional commuter rail way of trains per day: 12 roundtrips in a six-hour midday period.
And this is where the otherwise great study loses me. Forest Hills, Kew Gardens, and Flushing are all right next to subway stations. The LIRR charges higher fares there, but these are fairly middle-class areas – richer than Rosedale in Southeast Queens on the Far Rockaway Branch, which still gets more ridership than all three. No: the problem in these inner areas is frequency, and a train every half hour just doesn’t cut it when the subway is right there and comes every 2-3 minutes at rush hour and every 4-6 off-peak.
In this case, incremental increases from hourly to half-hourly frequency don’t cut it. The in-vehicle trip is so short that a train every half hour might as well not exist, just as nobody runs subway trains every half hour (even late at night, New York runs the subway every 20 minutes). At outer-urban locations like Bayside, Wakefield, and Rosedale, the absolute worst that should be considered is a train every 15 minutes, and even that is suspect and 10 minutes is more secure. Next to the subway, the absolute minimum is a train every 10 minutes.
All three mainlines currently radiating out of Manhattan in regular service – the Harlem Line, the LIRR Main Line, and the Port Washington Branch – closely parallel very busy subway trunk lines. One of the purposes of commuter rail modernization in New York must be decongestion of the subway, moving passengers from overcrowded 4, 5, 7, E, and F trains to underfull commuter trains. The LIRR and Metro-North are considered at capacity when passengers start having to use the middle seats, corresponding to 80% of seated capacity; the subway is considered at capacity when there are so many standees they don’t meet the standard of 3 square feet per person (3.59 people/m^2).
To do this, it’s necessary to not just compete with buses, but also directly compete with the subway. This is fine: Metro-North and the LIRR can act as additional express capacity, filling trains every 5 minutes using a combination of urban ridership and additional ridership at inner suburbs. TransitCenter has an excellent proposal for how to improve service quality at the urban stations but then inexplicably doesn’t go all the way and proposes a frequency that’s too low.
Quick Note on Los Angeles and Chicago Density and Modal Split
A long-running conundrum in American urbanism is that the urban area with the highest population density is Los Angeles, rather than New York. Los Angeles is extremely auto-oriented, with a commute modal split that’s only 5% public transit, same as the US average, and doesn’t feel dense the way New York or even Washington or Chicago or Boston is. In the last 15 years there have been some attempts to get around this, chiefly the notion of weighted or perceived density, which divides the region into small cells (such as census tracts) and averaged their density weighted by population and not area. However, even then, Los Angeles near-ties San Francisco for second densest in the US, New York being by far the densest; curiously, already in 2008, Chris Bradford pointed out that for American metro areas, the transit modal split was more strongly correlated with the ratio of weighted to standard density than with absolute weighted density.
DW Rowlands at Brookings steps into this debate by talking more explicitly about where the density is. She uses slightly different definitions of density, so that by the standard measure Los Angeles is second to New York, but this doesn’t change the independent variable enough to matter: Los Angeles’s non-car commute modal split still underperforms any measure of density. Instead of looking at population density, she looks at the question of activity centers. Those centers are a way to formalize what I tried to do informally by trying to define central business districts, or perhaps my attempts to draw 100 km^2 city centers and count the job share there (100 km^2 is because my French data is so coarse it’s the most convenient for comparisons to Paris and La Défense).
By Rowlands’ more formal definition, Los Angeles is notably weaker-centered than comparanda like Boston and Washington. Conversely, while I think of Los Angeles as not having any mass transit because I compare it with other large cities, even just large American cities, Brookings compares the region with all American metropolitan areas, and there, Los Angeles overperforms the median – the US-wide 5% modal split includes New York in the average so right off the bat the non-New York average is around 3%, and this falls further when one throws away secondary transit cities like Washington as well. So Los Angeles performs fairly close to what one would expect from activity center density.
But curiously, Chicago registers as weaker-centered than Los Angeles. I suspect this is an issue of different definitions of activity centers. Chicago’s urban layout is such that a majority of Loop-bound commutes are done by rail and a supermajority of all other commutes are done by car; the overall activity center density matters less than the raw share of jobs that are in a narrow city center. Normally, the two measures – activity center density and central business district share of jobs – correlate: Los Angeles has by all accounts a weak center – the central 100 km^2, which include decidedly residential Westside areas, have around 700,000 jobs, and this weakness exists at all levels. Chicago is different: its 100 km^2 blob is uninspiring, but at the scale of the Loop, the job density is very high – it’s just that outside the Loop, there’s very little centralization.
Berlin Greens Know the Price of Everything and Value of Nothing
While trying to hunt down some numbers on the costs of the three new U5 stations, I found media discourse in Berlin about the U-Bahn expansion plan that was, in effect, greenwashing austerity. This is related to the general hostility of German urbanists and much of the Green Party (including the Berlin branch) to infrastructure at any scale larger than that of a bike lane. But the specific mechanism they use – trying to estimate the carbon budget – is a generally interesting case of knowing the costs more certainly than the benefits, which leads to austerity. The underlying issue is that mode shift is hard to estimate accurately at the level of the single piece of infrastructure, and therefore benefit-cost analyses that downplay ridership as a benefit and only look at mode shift lead to underbuilding of public transport infrastructure.
The current program in Berlin
In the last generation, Berlin has barely expanded its rapid transit network. The priority in the 1990s was to restore sections that had been cut by the Berlin Wall, such as the Ringbahn, which was finally restored with circular service in 2006. U-Bahn expansion, not including restoration of pre-Wall services, included two extensions of U8, one north to Wittenau that had begun in the 1980s and a one-stop southward extension of U8 to Hermannstrasse, which project had begun in the 1920s but been halted during the Depression. Since then, the only fully new extension have been a one-stop extension of U2 to Pankow, and the six-stop extension of U5 west from Alexanderplatz to Hauptbahnhof.
However, plans for much more expansive construction continue. Berlin was one of the world’s largest and richest cities before the war, and had big plans for further growth, which were not realized due to the war and division; in that sense, I believe it is globally only second to New York in the size of its historic unrealized expansion program. The city will never regain its relative wealth or size, not in a world of multiple hypercities, but it is growing, and as a result, it’s dusting off some of these plans.
Most of the lines depicted in red on the map are not at all on the city’s list of projects to be built by the 2030s. Unfortunately, the most important line measured by projected cost per rider, the two-stop extension of U8 north (due east) to Märkisches Viertel, is constantly deprioritized. The likeliest lines to be built per current politicking are the extensions of U7 in both directions, southeast ti the airport (beyond the edge of the map) and west from Spandau to Staaken, and the one-stop extension of U3 southwest to Mexikoplatz to connect with the S-Bahn. An extension to the former grounds of Tegel is also considered, most likely a U6 branch depicted as a lower-priority dashed yellow line on the map rather than the U5 extension the map depicts in red.
The carbon critique
Two days after the U5 extension opened two years ago, a report dropped that accused the proposed program of climate catastrophe. The argument: the embedded concrete emissions of subway construction are high, and the payback time on those from mode shift is more than 100 years.
The numbers in the study are, as follows: each kilometer of construction emits 98,800 tons of CO2, which is 0.5% of city emissions (that is, 5.38 t/person, cf. the German average of about 9.15 in 2021). It’s expected that through mode shift, each subway kilometer saves 714 t-CO2 in annual emissions through mode shift, which is assumed to be 20% of ridership, for a payback time of 139 years.
And this argument is, frankly, garbage. The scale of the difference in emissions between cities with and without extensive subway systems is too large for this to be possibly true. The U-Bahn is 155 km long; if the 714 t/km number holds, then in a no U-Bahn counterfactual, Berlin’s annual greenhouse gas emissions grow by 0.56%, which is just ridiculous. We know what cities with no or minimal rapid transit systems look like, and they’re not 0.56% worse than comparanda with extensive rapid transit – compare any American city to New York, for one. Or look again at the comparison of Berlin to the German average: Berlin has 327 cars per 1,000 people, whereas Germany-wide it’s 580 and that’s with extensive rapid transit systems in most major cities bringing down the average from the subway-free counterfactual of the US or even Poland.
The actual long-term effect of additional public transport ridership on mode shift and demotorization has to be much more than 20%, then. It may well be more than 100%: the population density that the transit city supports also increases the walking commute modal split as some people move near work, and even drivers drive shorter distances due to the higher density. This, again, is not hard to see at the level of sanity checks: Europeans drive considerably less than Americans not just per capita but also per car, and in the United States, people in New York State drive somewhat shorter distance per car than Americans elsewhere (I can’t find city data).
The measurement problem
It’s easy to measure the embedded concrete of infrastructure construction: there are standardized itemized numbers for each element and those can be added up. It’s much harder to measure the carbon savings from the existence of a better urban rail system. Ridership can be estimated fairly accurately, but long-term mode shift can’t. This is where rules of thumb like 20% can look truthy, even if they fail any sanity check.
But it’s not correct to take any difficult to estimate number and set it to zero. In fact, there are visible mode shift effects from a large mass transit system. The difficulty is with attributing specific shifts to specific capital investments. Much of the effect of mode shift comes from the ability of an urban rail system to contribute to the rise of a strong city center, which can be high-rise (as in New York), mid-rise (as in Munich or Paris), or a mix (as in Berlin). Once the city center anchored by the system exists, jobs are less likely to suburbanize to auto-oriented office parks, and people are likelier to work in city center and take the train. Social events will likewise tend to pick central locations to be convenient for everyone, and denser neighborhoods make it easier to walk or bike to such events, and this way, car-free travel is possible even for non-work trips.
This, again, can be readily verified by looking at car ownership rates, modal splits (for example, here is Berlin’s), transit-oriented development, and so on, but it’s difficult to causally attribute it to a specific piece of infrastructure. Nonetheless, ignoring this effect is irresponsible: it means the carbon benefit-cost analysis, and perhaps the economic case as well, knows the cost of everything and the value of little, which makes investment look worse than it is.
I suspect that this is what’s behind the low willingness to invest in urban rail here. The benefit-cost analyses can leave too much value on the table, contributing to public transport austerity. When writing the Sweden report, I was stricken by how the benefit-cost analyses for both Citybanan and Nya Tunnelbanan were negative, when the ridership projections were good relative to costs. Actual ridership growth on the Stockholm commuter trains from before the opening of Citybanan to 2019 was enough to bring cot per new daily trip down to about $29,000 in 2021 PPP dollars, and Nya Tunnelbanan’s daily ridership projection of 170,000 means around $23,000/rider. The original construction of the T-bana cost $2,700/rider in 2021 dollars, in a Sweden that was only about 40% as rich as it is today, and has a retrospective benefit-cost ratio of between 6 and 8.5, depending on whether broader agglomeration benefit are included – and these benefits are economic (for example, time savings, or economic productivity from agglomeration) scale linearly with income.
At least Sweden did agree to build both lines, recognizing the benefit-cost analysis missed some benefits. Berlin instead remains in austerity mode. The lines under discussion right now are projected between 13,160€ and 27,200€ per weekday trip (and Märkisches Viertel is, again, the cheapest). The higher end, represented by the U6 branch to Tegel, is close to the frontier of what a country as rich as Germany should build; M18 in Paris is projected to be more than this, but area public transport advocates dislike it and treat it as a giveaway to rich suburbs. And yet, the U6 branch looks unlikely to be built right now. When the cost per rider of what is left is this low, what this means is that the city needs to build more infrastructure, or else it’s leaving value on the table.
One- and Two-Seat Rides
All large urban rail networks rely on transfers – there are too many lines for direct service between any pair of stations. However, transfers are still usually undesirable; there is a transfer penalty, which can be mitigated but not eliminated. This forces the planners who design urban and suburban rail systems to optimize: too many transfers and the trips are too inconvenient, too few and the compromises required to avoid transfers are also too inconvenient. How do they do it? And why?
Of note, the strategies detailed below are valid for both urban rail and suburban commuter rail systems. Multi-line commuter rail networks like the RER and the Berlin S-Bahn tend to resemble urban rail in their core and work in conjunction with the rest of the urban rail network, and therefore strategies for reducing the onerousness of transferring work in much the same way for both kinds of systems. Suburban strategies such as timing half-hourly trains to meet connecting buses are distinct and outside the scope of this post.
Passengers universally prefer to avoid transfers between vehicles, keeping everything else constant. The transportation studies literature has enough studies on this pattern that it has a name: transfer penalty. The transfer penalty consists of three elements:
- Walking time between platforms or bus curbs
- Waiting time for the connecting train or bus
- An independent inconvenience factor in addition to the extra time
One meta-study of this topic is by Iseki-Taylor-Miller of the Institute for Transportation Studies. There’s a bewildering array of different assumptions and even in the same city the estimates may differ. The usual way this is planned in elasticity estimates is to bundle the inconvenience factor into walking and waiting times; passengers perceive these to be more onerous than in-vehicle time, by a factor that depends on the study. Iseki-Taylor-Miller quote a factor as low as 1.4-1.7 and Lago-Mayworm-McEnroe’s classic paper, sourced to a Swedish study, go up to 3; Teulings-Ossokina-de Groot suggest it is 2, which is the figure I usually use, because of the convenience of assuming worst-case scenario for waiting time (on average, the wait is half the headway).
The penalty differs based on the quality of station facilities, and Fan-Guthrie-Levinson investigate this for bus shelter. However, urban rail estimates including those in the above meta-studies are less dependent on station facilities, which are good in all cases.
Mitigating the transfer penalty
Reducing the transfer penalty for riders can be done in three ways, if one believes the model with a constant penalty factor (say 2):
- Reducing the number of transfers
- Reducing walking time between platforms
- Reducing waiting time for trains
All three are useful strategies for good urban rail network planning, and yet all three are useful only up to a point, beyond which they create more problems than they solve.
The most coherent network planning principle for reducing passengers’ need to transfer is to build radial rail networks. Such networks ideally ensure each pair of lines intersects once in or near city center, with a transfer, and thus there is at most one transfer between any pair of stations. A circumferential line may be added, creating some situations in which a three-legged trip is superior in case it saves a lot of time compared with the two-legged option; in Moscow, the explicit purpose of the Circle Line is to take pressure off the congested passageway of the central transfer connecting the first three lines.
In general, the most coherent radial networks are those inherited from the Soviet tradition of metro building; the London Underground, which influenced this tradition in the 1920s, is fairly radial itself, but has some seams. It’s important in all cases to plan forward and ensure that every pair of lines that meets has a transfer. New York has tens of missed connections on the subway, and Tokyo has many as well, some due to haphazard planning, some due to an explicit desire to build the newer lines as express relief lines to the oversubscribed older lines.
On a regional rail network, the planning is more constrained by the need to build short tunnels connecting existing lines. In that case, it’s best to produce something as close to a coherent radial network with transfers at all junctions as possible. Through-running is valuable here, even if most pairs of origins and destinations on a branched commuter line trunk still require a transfer, for two reasons. First, if there is through-running, then passengers can transfer at multiple points along the line, and not just at the congested city center terminus. And second, while through-running doesn’t always cut the transfer for suburb-to-suburb trips, it does reliably cut the transfer for neighborhood-to-suburb trips involving a connection to the metro: a diameter can be guaranteed to connect with all radial metro lines, whereas a radius (terminating at city center) will necessarily miss some of them, forcing an extra transfer on many riders.
Reducing walking time
The ideal transfer is cross-platform, without any walking time save that necessary to cross a platform no more than 10-15 meters wide. Some metro building traditions aim for this from the outset: London has spent considerable effort on ensuring the key Victoria line transfers are cross-platform and this has influenced Singapore and Hong Kong, and Berlin has accreted several such transfers, including between the U- and S-Bahn at Wuhletal.
However, this is not always viable. The place where transfers are most valuable – city center – is also where construction is the most constrained. If two lines running under wide streets cross, it’s usually too costly to tilt them in such a way that the platforms are parallel and a cross-platform transfer is possible. But even in that case, it’s best to make the passageways between the platforms as short as possible. A cruciform configuration with stairs and an elevator in the middle is the optimum; the labyrinthine passageways of Parisian Métro stations are to be avoided.
Reducing waiting time
The simplest way to reduce waiting time is to run frequently. Passengers’ willingness to make untimed transfers is the highest when frequency is the highest, because the 2-minute wait found on such systems barely lengthens one’s trip even in the worst case, when one has frustratingly just missed the train.
Radial metro networks based on two- rather than one-seat rides pair well with high frequency. Blog supporter and frequent commenter Threestationsquare went viral last month when he visited Kyiv, a Soviet-style three-line radial system, and noted that due to wartime cuts the trains only run every 6-7 minutes off-peak; Americans amplified this and laughed at the idea that base frequency could be so high that a train every 7 minutes takes the appellation “only.”
When frequency is lower, for example on a branch or at night, cross-platform transfers can be timed, as is the case in Berlin. But these are usually accidental transfers, since the core city center transfers are on frequent trunks, and thus the system is only valuable at night. Moreover, timed transfers almost never work outside cross-platform transfers, which as noted above are not always possible; the only example I’m aware of is in Vienna, where a four-way transfer with stacked parallel platforms is timed.
This is naturally harder on a branched commuter rail system. In that case, it’s possible to set up the timetable to make the likeliest origin-destination pairs have short transfer windows, or even one-seat rides. However, in general transfers may require a wait as long as the system’s base clockface intervals, which is unlikely to be better than 20 minutes except on the busiest trunks in the largest cities; even Paris mixes 10-, 15-, and occasionally 20- and 30-minute intervals on RER branches.
More on Six-Minute Service in New York
Two years ago I wrote about how New York should aim to run every bus and subway service every six minutes off-peak. Buses would require a combination of aggressive bus redesign and speedup treatments for this to be viable. The subway already has very low variable operating costs off-peak and such a boost in frequency would naturally increase efficiency; New York City Transit gets around 550 service-hours annually per train driver, whereas the Berlin U-Bahn with its flat all-day schedule gets around 900. But now, the more mainstream New York-area transit advocacy group Riders’ Alliance has its own proposal for six-minute service, which it has aggressive marketed using the hashtag #6minuteservice.
This is a good campaign and I hope more people in the region take notice and push for it until the state implements it in full. The impact on passenger convenience is massive, not just in the form of shorter waits but also higher reliability coming from better timetabling, and hopefully also slightly more speed coming from said higher reliability. The proposal says that it would take $250 million a year in extra spending to effect this system, and it’s unknown but plausible that it would increase ridership by enough to defray this cost entirely, even without any efficiency treatments to reduce unit costs.
What’s in the Riders’ Alliance proposal?
Between 5 am and 9 pm on weekdays, and between 8 am and 10 pm on weekends, all subway routes and the top 100 bus routes in the city should run at worst every six minutes. This echoes a report by the comptroller’s office from last year, recommending this as an alternative to rush hour-focused service by bringing up corona-related ridership decreases.
It’s not stated but I think the subway routes in question are reckoned by letter or number, which means the A train runs every six minutes but each of its two branches runs every 12. This is fine – the two branches of the A are exceptionally far out, which is why a single service splits to them, where elsewhere in New York each branch gets its own number or letter.
The implications for timetabling
Timetabling a consistent all-day service is much easier than timetabling bespoke service patterns. The Riders’ Alliance proposal aims to face the general public rather than planners and therefore omits this benefit, but this benefit reaches passengers as well, in non-obvious ways.
First, if all trains and buses run every six minutes, then it’s possible to set up clockface timetables. These don’t matter very much if they run every six minutes, but they do if they run every 12, as I expect the two A branches to. The same is true of buses that branch: some outer ends may run every 12 minutes, in which case they can and should run on repeating clockface timetables that passengers can memorize. Passengers who can remember “my bus leaves at :01, :13, :25, :37, and :49” without having to consult timetables or trip planners all the time are likelier to take the trip; this was my commute for a year in Vancouver.
The A train today runs every 15 minutes on each branch but it’s not on a consistent clockface schedule, which depresses ridership. In effect, current practice is little different from what Swiss planners warn of: they say the best way to reduce ridership is to run service every 11, 13, or 17 minutes, rather than every 12 or 15 on a clockface pattern.
Second, if all trains run on the same frequency, then service planning on a complexly interlined system like New York’s becomes more tractable. Today, every train runs on a separate frequency, often different from the services it shares track with. The 2 and 3 trains share track most of the way, from Franklin Avenue to 135th Street, but the 2 is just a little more frequent, resulting in the following northbound timetable at Franklin:
This is irregular both on the trunk and on each individual service – the 2 on average runs every eight minutes but has a 12-minute gap, and the 3 runs on average every nine but also has a 12-minute gap. It’s an unavoidable consequence of the combination of extensive reverse-branching and subway frequency guidelines that run different services at different headways. The six-minute service proposal straightens this by aligning the trains to a single frequency, with regular alternation between successive trains on trunks.
And third, another benefit of a regular frequency to planning is that schedule planners can reliably avoid merge conflicts. This, in turn, speeds up service, which is full of planned delays and schedule padding at pain points. It’s not a full substitute for deinterlining, which would eliminate the merge conflicts at the worst junctions, but it makes it viable to no longer write impossible schedules with the planning department that New York City Transit has.
Service quality and demographics
Both Riders’ Alliance and the comptroller report it uses as its source point out demographic differences between peak and off-peak riders: rush hour subway commuters have a median income of $50,783 a year, even higher (slightly) than drivers, but off-peak subway commuters have a median income of $37,048 and bus commuters have a median income of $30,374.
In both reports this is taken to be indicative that off-peak service is mostly for poorer people, but it’s not the right analysis. The picture that emerges from the data is not that in general rush hour commuters outearn off-peak commuters; for one, most off-peak commutes are done by car, not by public transportation. Rather, what’s going on is that off-peak public transit quality is bad and this suppresses ridership among those who can afford a car.
By the same token, we can look at the incomes of commuters in regions of the United States that have no public transit to speak of – maybe some buses or even a few trains but with rounding-error ridership and low single-digit modal split. In metro New York, public transit and car commuters have about the same median income, and in some secondary transit cities like Chicago public transit commuters actually outearn drivers, since service to non-CBD destinations is so bad it suppresses ridership below median income more than above it. But in places like Los Angeles, the median income of transit commuters is not much more than half that of car commuters, because service quality is so bad that anyone who can afford to drive does.
The upshot of this is that better off-peak transit service is going to increase the average income of off-peak transit users, by attracting people who currently drive. This is also going to lead to higher-socioeconomic status shifts: higher levels of degree attainment, a larger proportion of white riders, a larger proportion of native-born riders.
I bring this up because a rise in the relative average income of users as service quality improves means the improvement is working as intended. It doesn’t mean the subway is gentrifying or turns away poorer riders, it just means it no longer repels riders who can afford to drive. This is important, because too much American transit planning is based on market segmentation in which service is supposed to be for a specific class of rider, and if the demographics are changing it means it’s being revamped for a different class. In reality, there’s just one transit system for one city and income differences are indicative of quality differences and not of inherent differences in the travel market.
How much does this cost? What is the ridership impact?
The Riders’ Alliance proposal says the additional cost of the program is $250 million a year in operating expenses. In 2019, NYCT spent $8.8 billion on operations and got $4.6 billion in fares, so this is in theory a 6% increase in subsidy, and in practice a little less as better service attracts more fare-paying riders. This is without any concurrent attempts to use the increase in service to increase efficiency (read: reduce unit staffing levels) and, I think, without bus speedups that permit much higher frequency for the same cost.
It’s unclear what the revenue impact should be; the ridership impact can be estimated from longstanding results in the literature about ridership-frequency elasticity, which in the case of NYCT should be about 0.4. The proposal increases off-peak service on the subway by around 50% in principle and a bit more in practice because of the reduced variability in frequency, say two-thirds: most lines are to go from 10- to six-minute headways and the rest, which are mostly more frequent than this, get a smaller increase that we round up to two-thirds by taking the impact of higher reliability into account. This means an increase in off-peak ridership of around 23%. The bus impact is even larger – in Brooklyn the median bus headway is right between 12 and 15 minutes, and even taking into account that the busiest buses do much better, this is close to a doubling of the effective frequency.
In turn, most ridership is off-peak. In 2019, peak (7-10 am) ridership into the Manhattan core was 923,000 per weekday, amounting to 44% of ridership entering the Manhattan core on a weekday, or around 33% of all inbound weekday ridership and 27% of all ridership. Even adding a bit to account for peak ridership that doesn’t enter Manhattan, only about a third of subway ridership in New York was at the peak before corona; the peak share has fallen since, but is slowly creeping back up as workers slowly return to the office. Raising two-thirds of ridership by 23% is massive – it’s a 15% systemwide increase for a much smaller increase in operating costs, and a somewhat larger increase in bus ridership to boot.
Unfortunately, I can’t turn this into a revenue impact estimate. While the demographics in the section above specify off-peak commuters, the studies that my ridership estimate is based on measure riders, including peak commuters who ride more often for non-work trips. Such riders already have monthly passes, so making it easier for them to ride is excellent for the city’s long-term health but doesn’t defray the added cost. Converted riders who are not already on the system as well as the odd peak rider who doesn’t already have a pass do generate more revenue, but I don’t know how many there are; these need to be a little more than a third of the overall increase in ridership to fully defray costs, which sounds plausible to me.
Eno’s Project Delivery Webinar
Eno has a new report out about mass transit project delivery, which I encourage everyone to read. It compares the American situation with 10 other countries: Canada, Mexico, Chile, Norway, Germany, Italy, South Africa, Japan, South Korea, and Australia. Project head Paul Lewis just gave a webinar about this, alongside Phil Plotch. Eno looks at high-level governance issues, trying to figure out if there’s some correlation with factors like federalism, the electoral system, and the legal system; there aren’t any. Instead of those, they try teasing out project delivery questions like the role of consultants, the contracting structure, and the concept of learning from other people.
This is an insightful report, especially on the matter of contract sizing, which they’ve learned from Chile. But it has a few other gems worth noting, regarding in-house planning capacity and, at meta level, learning from other people.
How Eno differs from us
The Transit Costs Project is a deep dive into five case studies: Boston, New York, Stockholm (and to a lesser extent other Nordic examples), Istanbul (and to a lesser extent other Turkish examples), and the cities of Italy. This does not mean we know everything there is to know about these cases; for example, I can’t speak to the issues of environmental review in the Nordic countries, since they never came up in interviews or in correspondence with people discussing the issue of the cost escalation of Nya Tunnelbanan. But it does mean knowing a lot about the particular history of particular projects.
Eno instead studies more cases in less detail. This leads to insights about places that we’ve overlooked – see below about Chile and South Korea. But it also leads to some misinterpretations of the data.
The most significant is the situation in Germany. Eno notes that Germany has very high subway construction costs but fairly low light rail costs. The explanation for the latter is that German light rail is at-grade trams, the easiest form of what counts as light rail in their database to build. American light rail construction costs are much higher partly because American costs are generally very high but also partly because US light rail tends to be more metro-like, for example the Green Line Extension in Boston.
However, in the video they were asked about why German subway costs were high and couldn’t answer. This is something that I can answer: it’s an artifact of which subway projects Germany builds. Germany tunnels so little, due to a combination of austerity (money here goes to gas subsidies, not metro investments) and urbanist preference for trams over metros, that the tunnels that are built are disproportionately the most difficult ones, where the capacity issues are the worst. The subways under discussion mostly include the U5 extension in Berlin, U4 in Hamburg, the Kombilösung in Karlsruhe, and the slow expansion of the tunneled part of the Cologne Stadtbahn. These are all city center subways, and even some of the outer extensions, like the ongoing extension of U3 in Nuremberg, are relatively close-in. The cost estimates for proposed outer extensions like U7 at both ends in Berlin or the perennially delayed U8 to Märkisches Viertel are lower, and not too different per kilometer from French levels.
This sounds like a criticism, because it mostly is. But as we’ll see below, even if they missed the ongoing changes in Nordic project delivery, what they’ve found from elsewhere points to the exact same conclusions regarding the problems of what our Sweden report calls the globalized system, and it’s interesting to see it from another perspective; it deepens our understanding of what good cost-effective practices for infrastructure are.
The issue of contract sizing in the Transit Costs Project
Part of what we call the globalized system is a preference for fewer, larger contracts over more, smaller ones. Trafikverket’s procurement strategy backs this as a way of attracting international bidders, and thus the Västlänken in Gothenburg, budgeted at 20,000 kronor in 2009 prices or around $2.8 billion in 2022 prices, comprises just six contracts. A planner in Manila, which extensively uses international contractors from all over Asia to build its metro system (which has reasonable elevated and extremely high underground costs), likewise told us that the preference for larger contracts is good, and suggested that Singapore may have high costs because it uses smaller contracts.
While our work on Sweden suggests that the globalized system is not good, the worst of it appeared to us to be about risk allocation. The aspects of the globalized system that center private-sector innovation and offload the risk to the contractor are where we see defensive design and high costs, while the state reacts by making up new regulations that raise costs and achieve little. But nothing that we saw suggested contract sizing was a problem.
And in comes Eno and brings up why smaller contracts are preferable. In Chile, where Eno appears to have done the most fieldwork, metro projects are chopped into many small contracts, and no contractor is allowed to get two adjacent segments. The economic logic for this is the opposite of Sweden’s: Santiago wishes to make its procurement open to smaller domestic firms, which are not capable of handling contracts as large as those of Västlänken.
And with this system, Santiago has lower costs than any Nordic capital. Project 63, building Metro Lines 3 and 6 at the same time, cost in 2022 PPP dollars $170 million/km; Nya Tunnelbanan is $230 million/km if costs don’t run over further, and the other Nordic subways are somewhat more expensive.
Other issues of state capacity
Eno doesn’t use the broader political term state capacity, but constantly alludes to it. The report stresses that project delivery must maintain large in-house planning capacity. Even if consultants are used, there must be in-house capacity to supervise them and make reasonable requests; clients that lack the ability to do anything themselves end up mismanaging consultants and making ridiculous demands, which point comes out repeatedly and spontaneously for our sources as well as those of Eno. While Trafikverket aims to privatize the state on the British model, it tries to retain some in-house capacity, for example picking some rail segments to maintain in-house to benchmark private contractors against; at least so far, construction costs in Stockholm are around two-fifths those of the Battersea extension in London, and one tenth those of Second Avenue Subway Phase 1.
With their broader outlook, Eno constantly stresses the need to devolve planning decisions to expert civil servants; Santiago Metro is run by a career engineer, in line with the norms in the Spanish- and Portuguese-language world that engineering is a difficult and prestigious career. American- and Canadian-style politicization of planning turns infrastructure into a black hole of money – once the purpose of a project is spending money, it’s easy to waste any budget.
Finally, Eno stresses the need to learn from others. The example it gives is from Korea, which learned the Japanese way of building subways, and has perfected it; this is something that I’ve noticed for years in my long-delayed series on how various countries build, but just at the level of a diachronic metro map it’s possible to see how Tokyo influenced Seoul. They don’t say so, but Ecuador, another low-cost Latin American country, used Madrid Metro as consultant for the Quito Metro.
The Nine-Euro Ticket
A three-month experiment has just ended: the 9€ monthly, valid on all local and regional public transport in Germany. The results are sufficiently inconclusive that nobody is certain whether they want it extended or not. September monthlies are reverting to normal fares, but some states (including Berlin and Brandenburg) are talking about restoring something like it starting October, and Finance and Transport Ministers Christian Lindner and Volker Wissing (both FDP) are discussing a higher-price version on the same principle of one monthly valid nationwide.
The intent of the nine-euro ticket
The 9€ ticket was a public subsidy designed to reduce the burden of high fuel prices – along with a large three-month cut in the fuel tax, which is replaced by a more permanent cut in the VAT on fuel from 19% to 7%. Germany has 2.9% unemployment as of July and 7.9% inflation as of August, with core inflation (excluding energy and food) at 3.4%, lower but still well above the long-term target. It does not need to stimulate demand.
Moreover, with Russia living off of energy exports, Germany does not need to be subsidizing energy consumption. It needs to suppress consumption, and a few places like Hanover are already restricting heating this winter to 19 degrees and no higher. The 9€ ticket has had multiple effects: higher use of rail, more domestic tourism, and mode shift – but because Germany does not need fiscal stimulus right now and does need to suppress fuel consumption, the policy needs to be evaluated purely on the basis of mode shift. Has it done so?
The impact of the nine-euro ticket on modal split
The excellent transport blog Zukunft Mobilität aggregated some studies in late July. Not all reported results of changes in behavior. One that did comes from Munich, where, during the June-early July period, car traffic fell 3%. This is not the effect of the 9€ ticket net of the reduction in fuel taxes – market prices for fuel rose through this period, so the reduction in fuel taxes was little felt by the consumer. This is just the effect of more-or-less free mass transit. Is it worth it?
Farebox recovery and some elasticities
In 2017 and 2018, public transport in Germany had a combined annual expenditure of about 14 billion €, of which a little more than half came from fare revenue (source, table 45 on p. 36). In the long run, maintaining the 9€ ticket would thus involve spending around 7 billion € in additional annual subsidy, rising over time as ridership grows due to induced demand and not just modal shift. The question is what the alternative is – that is, what else the federal government and the Länder can spent 7 billion € on when it comes to better public transport operations.
Well, one thing they can do is increase service. That requires us to figure out how much service growth can be had for a given increase in subsidy, and what it would do to the system. This in turn requires looking at service elasticity estimates. As a note of caution, the apparent increase in public transport ridership over the three months of more or less free service has been a lot less than what one would predict from past elasticity estimates, which suggests that at least fare elasticity is capped – demand is not actually infinite at zero fares. Service elasticities are uncertain for another reason: they mostly measure frequency, and frequency too has a capped impact – ridership is not infinite if service arrives every zero minutes. Best we can do is look at different elasticity estimates for different regimes of preexisting frequency; in the highest-frequency bucket (every 10 minutes or better), which category includes most urban rail in Germany, it is around 0.4 per the review of Totten-Levinson and their own work in Minneapolis. If it’s purely proportional, then doubling the subsidy means increasing service by 60% and ridership by 20%.
The situation is more complicated than a purely proportional story, though, and this can work in favor of expanding service. Just increasing service does not mean doubling Berlin U-Bahn frequency from every 5 to every 2.5 minutes; that would achieve very little. Instead, it would bump up midday service on the few German rail services with less midday than peak frequency, upgrade hourly regional lines to half-hourly (in which case the elasticity is not 0.4 but about 1), add minor capital work to improve speed and reliability, and add minor capital work to save long-term operating costs (for example, by replacing busy buses with streetcars and automating U-Bahns).
The other issue is that short- and long-term elasticities differ – and long-term elasticities are higher for both fares (more negative) and service (more positive). In general, ridership grows more from service increase than from fare cutting in the short and long run, but it grows more in the long run in both cases.
The issue of investment
The bigger reason to end the 9€ ticket experiment and instead improve service is the interaction with investment. Higher investment levels call for more service – there’s no point in building new S-Bahn tunnels if there’s no service through them. The same effect with fares is more muted. All urban public transport agencies project ridership growth, and population growth is largely urban and transit-oriented suburban.
An extra 7 billion € a year in investment would go a long way, even if divided out with direct operating costs for service increase. It’s around 250 km of tramway, or 50 km of U-Bahn – and at least the Berlin U-Bahn (I think also the others) operationally breaks even so once built it’s free money. In Berlin a pro-rated share – 300 million €/year – would be a noticeable addition to the city’s 2035 rail plan. Investment also has the habit to stick in the long term once built, which is especially good if the point is not to suppress short-term car traffic or to provide short-term fiscal stimulus to a 3% unemployment economy but to engage in long-term economic investment.