Category: Transportation

I’m Giving a Talk About Construction Costs Tomorrow

By popular demand, I’m giving the talk I gave 2 weeks ago at NYU, again. The database will be revised slightly to include more examples (like Ukraine, which I added between when I gave the talk and when I blogged about it), and I may switch around a few things, but it should be similar to what I already said.

Where? Halyards in Brooklyn at 3rd Avenue and 6th Street, near the 4th Avenue/9th Street subway stop where the F/G and R intersect.

When? Monday December 2nd at 9 pm, for an hour.

Do I need to RSVP? No.

Will there be food? To some extent – the bar has minimal selection, although what it does have on the menu seems better for the price than most American bar food (which, to be fair, is like saying “better public transportation than Los Angeles”).

Little Things That Matter: Jerk

When you ride a subway train, and the train decelerates to its station, you feel your body pulled forward, and your muscles tense to adjust, but then when the train reaches a sudden stop, you are suddenly flung backward, since you are no longer decelerating, but your muscles take time to relax and stop fighting a braking that no longer exist. This effect is called jerk, and is defined to be change in acceleration, just as acceleration is change in speed and speed is change in position. Controlling jerk is crucial for a smooth railway ride. Unfortunately, American mainline rail is not good at this, leading to noticeable jolts by passengers even though speed limits on curves and acceleration rates are very conservative.

This is particularly important for speeding up mainline trains around New York and other legacy cities in the US, like Boston. Speeding up the slowest segments is more important than speeding up the fastest ones; my schedules for New York-New Haven trains, cutting trip times from 2:09 to 1:24, save 4 minutes between Grand Central and 59th Street just through avoiding slowdowns in the interlocking. The interlocking is slow because the switches have very conservative speed limits relative to curve radius (that is, lateral acceleration), which in turn is because they are not designed with good lateral jerk control. The good news is that replacing the necessary infrastructure is not so onerous, provided the railroads know what they need to do and avoid running heavy diesel locomotives on delicate infrastructure.

Spirals and jerk

In practice, the worst jerk is usually not forward or backward, except in the last fraction of a second at the end of acceleration. This is because it takes about a second for train motors to rev up, which controls jerk during acceleration. Rather, the worst is sideways, because it is possible to design curves that transition abruptly from straight track, on which there is no lateral acceleration, to curved track, on which there is, in the form of centrifugal force centripetal force.

To reduce jerk, the transition from straight track to a circular arc is done gradually. There are a number of usable transition curve (see Romain Bosquet’s thesis, PDF-p. 36), but the most common by far is called the clothoid, which has the property of having constant change in curvature per unit of arc length – that is, constant jerk. Different countries have different standards for how long the clothoid should be, that is what the maximum lateral jerk is. Per Martin Lindahl’s thesis, the limit in Sweden is 55 mm/s (PDF-p. 30) and that in Germany is 69.44 mm/s (PDF-p. 38), both measured in units of cant deficiency; in SI units, this is 0.367 m/s^3 and 0.463 m/s^3 respectively. In France, the regular limit is 50 mm/s (Bosquet’s thesis, PDF-p. 35), that is 0.333 m/s^2, but it is specifically waived in turnouts.

Track switches are somehow accepted as sites of very high jerk. A presentation about various technical limits in France notes on p. 106 that in switches (“appareils de voie” or “aiguilles” or “aiguillages,” depending on source, just like “switch” vs. “turnout” in English), the jerk can be increased to 100 and even 125 mm/s. On p. 107 it even asserts that in exceptional circumstances, abrupt change in cant deficiency of up to 50 mm on main track and 100 on the diverging direction on a switch is allowed; see also PDF-pp. 13-15 of a pan-European presentation. Abrupt changes are not good for passengers, but will not derail a train.

Turnout design in the advanced world

Second derivative control, that is acceleration and cant deficiency, can be done using calculus and trigonometry tools. Third derivative control, that is clothoids and jerk, requires numerical calculations, but fortunately they are approximated well by pretending the clothoid is half straight line, half circular arc, with the length determined by the maximum jerk. Working from first principles, it’s possible to figure out that at typical turnout needs – e.g. move a train from one track to a parallel track 4 meters away – the clothoid is far longer than the curve itself, and at 50 mm/s jerk and 150 mm cant deficiency it’s not even possible to hit a curve radius of 250 meters.

Turnouts are inherently compromises. The question is just where to compromise. Here, for example, is a French turnout design, in two forms: 0.11 and 0.085. The numbers denoting the tangent of the angle at the frog, and the radius is proportional to the inverse square of the number, thus the speed is proportional to the inverse of the number. The sharper turnout, the 0.11, has a radius of 281 meters, a maximum speed of 50 km/h, and a total length of 26 meters from point to frog (“lead” in US usage), of which the clothoid curve (“point”) takes up 11, to limit jerk to 125 mm/s at a cant deficiency of 100 mm. The 0.085 turnout has a radius of 485 meters, a maximum speed of 65 km/h, a lead of about 38 meters, and a point of about 14.5 meters.

In Germany, turnouts have somewhat independent numbers and radii – some have shorter leads than others. The numbers are the inverse of those of France, so what France calls 0.11, Germany calls 1:9, but at the end of the day, the curve radius is the important part, with a cant deficiency of 100 mm. A higher cant deficiency may be desirable, but lengthening the point requires almost as much space as just increasing the curve radius, so might as well stick with the more comfortable limits.

Turnout design in the United States

American turnouts look similar to French or German ones, at first glance. I’ve seen a number of different designs; here’s one by CSX, on PDF-pp. 22 (#8) and 24 (#10), the numbers being very roughly comparable to German ones and inverses of French ones. CSX’s #10 has a curve radius of 779.39′, or 238 meters, and a lead of 24 meters, both numbers slightly tighter than the French 0.11. The radius is proportional to the square of the number, and so speed is proportional to the number.

However, the cant deficiency is just 50 mm. The point is not always curved; Amtrak’s low-number switches are not, so the change in cant deficiency is abrupt. Judging by what I experience every time I take a train between New York and New Haven, Metro-North’s switches have abrupt change in cant deficiency even on the mainline. The recommended standards by AREMA involve a curved point, but the point is still much shorter than in France (19.5′, or just under 6 meters, on a #12), so a 125 mm/s jerk only gets one up to about 62 mm cant deficiency.

The reason for this is that European turnouts are curved through the frog, whereas American ones are always straight at the frog. Extremely heavy American freight trains do not interact well with curved frogs and long points.

One might ask, why bother with such turnout design on rail segments that never see a heavy freight locomotive or 130-ton freight car? And on segments that do see the odd freight locomotive, like the approaches to Grand Central and Penn Station with the rare dual-mode locomotive, why not kick out anything that doesn’t interact well with advanced track design? Making a handful of passengers transfer would save around 4 minutes of trip time on the last mile into Grand Central alone for everyone else, not to mention time savings farther up the line.

Fare Regulations

Public transportation companies may have the ability to raise fares arbitrarily based on market demands, for examples British buses outside London and American freight railroads. Or they may be subject to regulations capping the fare, for example Japanese railroads. Mixed systems exist as well, such as British rail fares. In Britain, the privatized, mostly deregulated approach is so commonly accepted that a Conservative recently called Labour dangerous socialists for proposing municipalizing bus systems, as in such socialist states as the US, Japan, Germany, etc. In reality, in the case of rail specifically (and perhaps buses as well), there’s a theoretical case with some empirical backing for why reasonable fare caps as in Japan can lead to more investment and more capacity, whereas wholly unregulated fares lead to hoarding and capacity cuts to create shortages.

The model

I’m stealing the economic model for this post from Paul Krugman, who used it to explain the California blackouts of 2000-1. The demand curve is inelastic: the demand is 1,000 units at $20/unit, decreasing to 900 units at $1,000/unit, at which point the curve goes flat. The supply curve is a constant $20/unit, but the market is oligopolistic (say, there are very high barriers to entry because building your own power plant is hard), and there are 5 producers, each with 200 units. If the price is regulated at $20/unit, each producer will supply 200 units. If the price is unregulated, then each producer alone gets an incentive to hold back production, since 100*1000 > 200*20, and then production will be curtailed to 900 units.

The model is simplified in a number of ways: real supply curves slope up; the part about demand going flat at 900 units is unrealistic and exists purely to avoid dealing with optimizing where at 800-something units each producer has an incentive to go back to producing more; capacity constraints involve escalating production costs rather than a God-given restriction on the number of suppliers and their capacity. But with all these caveats, it fits markets that have the following characteristics:

  • There are steep barriers to entry, for example if large amounts of capital are required to enter (to build a power plant, set up a rail operating company, etc.).
  • Demand is highly inelastic.
  • Adding new capacity is expensive.

The issue of capacity

In rail, we can start plugging real numbers for both demand elasticity and the cost of new capacity.

In the above model the price elasticity is -0.0244 in the 900-1,000 units range, which is ridiculously inelastic, on purpose so as to highlight how the model works. TCRP Report 95 says the elasticity in a number of large cities studied is about -0.18, and a VTPI review in a mixture of cities and circumstances (peak vs. off-peak, bus vs. rail, etc.) asserts a short-term average of about -0.3. Unregulated fares will lead to supply reductions if the elasticity times the number of producers is more than -1 (or less than 1 if you flip signs); if no producer has <18% of the market, there will be supply restrictions under unregulated fares, just as a monopolist will hold back supply and raise fares if demand is inelastic.

The cost of new capacity of course depends on the line and the characteristics of competition between different railroads. It’s higher in Japan, where separate railroads run their own lines and trains, than in Britain, where different companies franchise to run trains on the same tracks. But even in Britain, getting a franchise requires a commitment to running service for many years. The significance of this is that the long-run public transport ridership elasticity with respect to fare is more elastic (VTPI recommends a range of -0.6 to -0.9), with a few estimates even going below -1.

For the purposes of this section, we do not distinguish capital from operating costs. Thus, the cost of new capacity is not given in units of capital costs, but in units of operating costs: if increasing service by 1% raises operating expenses by 2% counting the extra investment required, then we say the supply elasticity is 2. Note that supply curves slope up so the elasticity is always positive, but the elasticity can be below 1, for example if economies of scale are more important than the need to invest in new capacity.

Set the following variables: u is quantity of service, r is total revenue (thus, fare is r/u), c is total costs. The railroad is assumed profitable, so r > c. We are interested in the change in profit based on quantity of service, i.e.

\frac{dr}{du} - \frac{dc}{du}.

The important thing to note is that price controls keep dr/du higher in an oligopoly (but not in a competitive environment, like housing – a single landlord can’t meaningfully create a housing shortage). With price controls, we get

\frac{dr}{du} = \frac{r}{u} = \mbox{constant fare}

whereas without price controls, with elasticity e_{d} < 0, we get

\frac{dr}{du} = \frac{r}{u} + \frac{r}{ue_{d}} = \frac{r}{u}(1 + 1/e_{d}).

And likewise, with supply elasticity e_{s} > 0, we get

\frac{dc}{du} = \frac{c}{ue_{s}}.

Note, moreover, that price controls as construed in Japan let operating companies recover profits, letting them raise prices if they invest in more capacity, so that dr/du is actually higher than r/u.

The real world

I do not know to what extent the lack of fare regulation on many British trains contributes to capacity shortages. However, there is some evidence that the same situation is holding back investment in the United States, on Amtrak. Amtrak is a monopolist facing some fare regulations, for example congressional rules limiting the spread between the lowest and highest fares on a given train, but within its ability to set its own capacity in the medium run, it has relatively free hand, and in fact a strong incentive to maximize fares, in order to subsidize money-losing trains outside the Northeast Corridor.

Amtrak generally runs the trains it has on the Northeast Corridor, without explicitly holding back on capacity. However, this is in an environment with very low utilization rates. There are 20 Acela trainsets, but only 16 run in service at a given time, giving them the moniker “hangar queens.” There is no real interest within Amtrak at raising speed just enough to be able to run consistent service intervals, for example hourly with two trainsets coupled to form a 16-car train south of New York. Nor is there any interest in making small investments to permit such long trainsets to run – most Acela stops from New York to the south have platforms long or almost long enough for such trains, but the rest need to be lengthened, within right-of-way so that the cost is positive but low.

In the future, capacity cliffs may prove serious enough to stymie American passenger rail development. Right now the main obstacle are Amtrak itself and obstructive commuter railroads such as Metro-North, but assuming competent, profit-maximizing investment plans, it is not so expensive to invest in capacity and speed so as to permit around 4 long high-speed trains per hour north of New York (or even New Haven) and 6 south of it. But then the next few trains per hour require further bypasses, for example four-tracking most of the Providence Line. High supply elasticity – let’s say around 2 – is plausible. Then eventually a dedicated pathway to intercity trains through New York becomes necessary, raising supply elasticity even higher. In an environment with uncapped, profit-maximizing fares, a rational Amtrak management may well just keep what it has and jack up prices rather than build more capacity.

I Gave a Talk About Transportation in Connecticut

I gave a second talk this week about transportation, this time at Hartford Station, concerning the plans for Connecticut transportation. The starting point is Governor Lamont’s $21 billion plan for investment, including both expansion and repairs (read: the State of Good Repair black hole), of which $14 billion is highways, $6.2 billion is rail, and $450 million is buses. But most of the talk concerns what Connecticut should be doing, rather than the specifics of Lamont’s plan.

Here are my slides. The talk itself took around 40-45 minutes out of a nearly 2-hour meeting, so it was designed around taking many questions, and around further explanations. Something I didn’t put in the slides but explained verbally is how easy the modern track renewal process is. Nowadays, there are machines that use no infrastructure except the tracks themselves, running on the tracks at very low speed (slower than walking) and systematically replacing the rails, ties, and ballast. They can also regrade the tracks’ superelevation angle independently of the drainage angle, changing the tracks’ cant as they go. The upshot is that increasing the cant on tracks is almost cost-free, and would enable large increases in train speed on both regional and intercity trains.

Other technology that has negative cost in the future is getting higher-performance EMUs than the current equipment. The current trains are obsolete technology, built around superseded federal regulations. There’s no point in getting more of the same. They’re okay to run until end of life, but new purchases should involve electrification and modern European EMUs. Whereas infrastructure costs are rising (see here and here), technology costs are falling in real terms. The fall in train costs is not so quick as that of computer costs, but still the rolling stock factories are designed around making products for the 2020s, not the 1990s, and retooling them for older technology costs extra.

Hence my slogan from the talk: better things are possible, on a budget.

One question I was asked at the talk that I didn’t have an answer to was, why is construction in Connecticut so expensive? Plans for infill stations are budgeted extravagantly, ranging between $50 million and $100 million without any special construction difficulties. Boston builds infill stations (counting high-platform upgrades as infill since the preexisting stations have no facilities) for $20-30 million counting various hidden costs (e.g. regular MBTA employees, like project managers, count as operating and not capital costs even if they only work on capital costs); Berlin does for €10-20 million.

After the talk, Roger Senserrich explained to me (and a planner at the MBTA confirmed to me) that in Connecticut there’s no in-house design at all. Massachusetts has a mix of in-house design review, with the team stymied by uncompetitive wages making hiring and retention difficult, and outsourcing work to consultants. CDOT exclusively outsources to consultants, and has no in-house expertise to evaluate whether the contracts are fair or whether it’s being overcharged.

I Gave a Talk About Construction Costs

Two years ago, I gave a talk at NYU about regional rail, and as promised, uploaded slides the next day for discussion. Yesterday I gave another such talk, about construction costs.

But here there are two things to upload: the slides, and the data table. I’ve been intermittently adding cities to a spreadsheet of various urban rapid transit lines and their construction costs, and by now there is a total of 207 distinct items, ranging from 1 km extensions to 3-figure packages like 200-km GPX and 160-km Delhi Metro phases. The total length of the lines in this database right now is 3610 km, of which 2090 are underground. These are almost exclusively new lines – most of them aren’t even open, and most of the rest opened this decade, so be cautious since much of the cost estimation is ex ante and a number of the soon-to-open line on the list have had serious cost overruns.

I hope people make use of this dataset and the preliminary analysis contained in the slides, and I ask that people look at both, since the slides do have some interpretive notes about confounding variables. One note that I did not include in the slides and explained verbally is what source means in the table: media means I’m drawing costs from popular media, trade means trade media like Railway Gazette, plan means official plans (either ex ante or ex post), wiki means Wikipedia (as always, a reliable source for line length and station count, never cost), measured means I measured line length on Google Earth lacking any alternative. One item, Crossrail, has its tunnel cost coming from a freedom of information request submitted by an alert reader who I will credit upon request; the headline budget is somewhat higher as it includes surface improvements, a common confounder for regional rail projects (the RER E extension, for example, splits its budget about 50/50 between the tunnel and above-ground works).

More detailed analysis is forthcoming, either here or in print.

Urban Freeways and Rapid Transit

A ride-hailing trip today reminded me of something about freeway travel in cities – namely, it is untethered from the surface street network. Oddly enough, for a different reason this is equally true of rapid transit. The commonality to these two ways of travel is that they change the geography of the city, rather than just extending the range of walking along the usual paths as surface arterial streets and surface transit do.

Rapid transit compression

Rapid transit networks compress distances along the lines, and by the same token magnify distances in orthogonal directions. Manhattan is a good example of how this works: north of Midtown the subway only runs north-south, not east-west, so there are separate East Side and West Side cultures. Moreover, as middle-class gentrifiers are displaced by rising rents coming from even richer gentrifiers, they tend to move along subway lines, and thus people from the Upper West Side and Columbia end up in Washington Heights and Inwood.

The contrast here is with surface transit. Bus networks are far too dense to have the same effect. A citywide bus grid would offer 15 km/h transit in all directions in New York, and a tramway grid like what parts of Berlin have (and what big Eastern European cities like Prague and Budapest have) offers 15-20 km/h transit in all directions. It extends walking, in the sense that the most important throughfares probably get their own routes, or if they don’t they are closely parallel with roads with surface transit.

This is not how rapid transit works. A handful of very strong orthogonal routes can and should get rapid transit, hence the Ringbahn, M2/M6 in Paris, and the under-construction M15 – and by the same token, 125th Street in New York should get a subway extension off of Second Avenue. But that still leaves the city with a wealth of major routes that have no reason to get rapid transit, ever. Most of these are crosstown routes, for example the east-west streets of Manhattan, but in less gridded cities they can just be major streets that don’t quite fit into a regionwide radial metro network.

Rapid transit spikiness

I get a lot of pushback when I talk about this, but rapid transit encourages spiky density. This does not mean that every transit city is spiky and every spiky city is a transit city. Density in Paris within the city is fairly uniform, aided by zoning rules that prohibit high-rises even though many could succeed commercially on top of Métro transfer points or RER stations. In the other direction, some American auto-oriented cities have spiky density near transit, like San Diego’s Mission Valley or Atlanta’s Buckhead, but it’s not big enough a development to permit people to comfortably walk and take transit to all destinations.

Nonetheless, for the most part, rapid transit tends to be associated with spiky development forms, especially if it’s been built more recently and if the interstation is long (as in Vancouver, Singapore, Hong Kong, or Stockholm). This isn’t really how a pedestrian city works: pedestrians have no need for spikiness because they don’t have particular distinguished stations – at most, the corner nodes are distinguished, but that includes all corners, which are placed at far shorter intervals than subway stops.

Freeways as street bypasses

Surface transit promotes urban forms that look like an extended pedestrian city. This is equally true of surface roads designed around car access. The car was originally not supposed to take over the entire city, but merely provide convenient intra-urban transportation at a faster speed than walking. It was originally just a faster, more private, more segregated streetcar. The effect on urbanism was to reduce overall density (as did the streetcars and rapid transit in New York, which used to have inhuman overcrowding levels on the Lower East Side), but not to change the urban form beyond that.

Freeways, like rapid transit, are completely different. This does not mean that they change the city in the same way as rapid transit, just that both operate independently of the usual street grid. Freeways, like rapid transit, compress travel distances along the freeway, and simultaneously lengthen them in all other directions because of the effect of traffic congestion.

Moreover, freeways are different from rapid transit in typical alignments. They are far more land-intensive, which is why they tend to be placed in formerly marginal parts of the city. This can include the waterfront if it is originally industrial and low-value, as it was in midcentury America, rather than a place of high-end residential consumption because of the views.

Interface with the street

How does a surface street transit network interface with either rapid transit or freeways?

With rapid transit, the answer is that surface transit is slow, so it should feed rapid transit using transfers, which may be timed if the trains are not so frequent (say, 15 minutes or worse, as is common on suburban rail branches). Rapid transit should then be constructed to connect with surface transit this way, that is the stations should be at intersections with arterial corridors for bus connections.

With freeways, the answer is that often interface is impossible. San Diego provides a convenient example: there is an arterial road that’s great for buses running northward from city center toward the beachfront neighborhood of Pacific Beach. But there’s also a parallel freeway inland, so drivers mostly use the freeway, and commerce on the north-south arterial is neglected. In contrast, the main east-west arterials feeding the freeway are bustling, and one of them has of the city’s strongest buses. Buses can make stops on these arterials and then express to city center on the freeway, but on the freeway itself the buses are not very efficient since there’s minimal turnover, and chaining a few neighborhoods together on one frequent route is usually not possible.

Fare Evasion

There’s a moralistic discourse in the United States about fare evasion on public transport that makes it about every issue other than public transport or fares. It’s a proxy for lawlessness, for police racism, for public safety, for poverty. In lieu of treating it as a big intra-urban culture war, I am going to talk about best practices from the perspective of limiting revenue loss to a minimum.

This is an issue where my main methodology for making recommendations for Americans – looking at peer developed countries – is especially useful. The reason is that Americans practically never look at other countries on hot-button culture war issues, even less than (say) the lip service the center-left pays to foreign universal health care systems. Americans who support immigration liberalization practically never listen when I try bringing up the liberal work visa, asylum, and naturalization policies of Germany or Sweden. Knowing stuff about the rest of the world is a type of competence, and competence is not a factor in a culture war. The upshot is that successful policies regarding fare collection in (for example) Germany are obscure in the United States even more than policies regarding wonkier transportation issues like train frequency.

The current situation in New York

In the summer, Governor Cuomo announced a new initiative to hire 500 cops to patrol the subway. The justification for this scheme has varied depending on who was asking, but the primary goal appears to be to defeat fare evasion. Per Cuomo’s office, fare evasion costs $240 million a year on the subway and buses, about 5% of total revenue. The MTA has also mentioned a higher figure, $300 million; I do not know if the higher figure includes just urban transit or also commuter rail, where conductors routinely miss inspections, giving people free rides.

But New York fare evasion is mostly a bus problem: the rate on buses is 22%. On the subway the rate is only 4%, and there is somewhat more revenue loss on buses than on subways. This, in turn, is because bus fares are enforced by drivers, who for years have complained that fare disputes lead to assaults on them and proposed off-board fare collection as an alternative. On many buses, drivers just let it go and let passengers board without paying, especially if nearly all passengers are connecting from the subway and therefore have already paid, as on the B1 between the Brighton Beach subway station and Kingsborough Community College or on the buses to LaGuardia.

So realistically the subway fare evasion level is closer to $110 million a year. The total cost of the new patrol program is $56 million in the first year, escalating by 8% annually thanks to a pre-agreed pay hike scale. Whereas today the program is a net revenue generator if it halves subway fare evasion, a level that already seems strained, within ten years, assuming normal fare escalation, it will need to cut fare evasion by about 90%, which is a complete fantasy. A sizable proportion of riders who do not pay would just stop riding altogether, for one. The governor is proposing to spend more on fare enforcement than the MTA can ever hope to extract.

The American moral panic about fare evasion regrettably goes far beyond New York. Two years ago, BART announced that it would supplement its fare barriers with proof-of-payment inspections, done by armed cops, and lied to the public about the prevalence of such a belts-and-suspenders system. More recently, it trialed a new turnstile design that would hit passengers in the face, but thankfully scrapped it after public outcry. Boston, too, has its moral panic about fare evasion, in the form of campaigns like the Keolis Ring of Steel on commuter rail or Fare is Fair.

There is another way

In talking to Americans about fare evasion, I have found that they are generally receptive to the idea of minimizing revenue loss net of collection costs. However, what I’ve encountered more resistance about is the idea that people should just be able to walk onto a bus or train.

In the urban German-speaking world, everyone with a valid fare can walk onto a bus, tram, or train without crossing fare barriers or having to pay a driver. This system has been copied to American light rail networks, but implementation on buses and subways lags (except on San Francisco buses). In New York, the SBS system uses proof of payment (POP), but passengers still have to validate fares at bus stops, even if they already have paid, for example if they have a valid monthly pass.

In the vast majority of cities, no excuse exists to have any kind of overt fare control. Tear down these faregates. They are hostile to passengers with disabilities, they cost money to maintain, they constrain passenger flow at busy times, and they don’t really save money – evidently, New York’s subway fare evasion rate is within the range of Berlin, Munich, and Zurich. Fare enforcement should be done with POP alone, by unarmed civilian inspectors, as in Berlin. Some people will learn to dodge the inspectors, as is the case in Berlin, and that’s fine; the point is not to get fare evasion to 0%, but to the minimum level net of enforcement costs.

New York itself may have an excuse to keep the faregates: its trains are very crowded, so peak-hour inspections may not be feasible. The question boils down to how New York crowding levels compare with those on the busiest urban POP line, the Munich S-Bahn trunk. But no other American city has that excuse. Tear down these faregates.

What’s more, the fare inspection should be a low-key affair. The fine in Berlin is €60. In Paris on the RER I can’t tell – I believe it’s three figures of which the first is a 1. Inspectors who can’t make a citation without using physical violence should not work as inspectors.

Make it easy to follow the law

The most important maxim when addressing a low-level crime is to make it easy to follow the law. Mistakes happen; I’ve accidentally fare-dodged in Berlin twice, only realizing the error at the end of the trip. This is much more like parking violations or routine mistakes in tax filing.

The turnstile acts as a reminder to everyone to pay their fare, since it’s not possible to fare-dodge without actively jumping it. (I did turnstile-jump in Paris once, with a valid transfer ticket that the turnstile rejected, I think because Paris’s turnstile and magnetic ticket technology is antediluvian.) However, turnstiles are not necessary for this. A better method is to ensure most passengers have prepaid already, by offering generous monthly discounts. My fare dodges in Berlin happened once before I got monthlies and once on my way to the airport on my current trip, in a month when I didn’t get a monthly since I was only in Berlin 6 days.

New York does poorly on the metric of encouraging monthlies. Passengers need to swipe 46 times in a 30-day period to justify getting a monthly pass rather than a pay-per-ride. This is bad practice, especially for passengers who prefer to refill at a ticketing machine rather than at home or on their phone with an app, since it means passengers visit the ticketing machines more often, requiring the agency to buy more to avoid long lines. In Berlin, the breakeven point is 36 trips. In Zurich, it’s 20 trips; ZVV does whatever it can to discourage people from buying single tickets. In both cities, there are further discounts for annual tickets.

Unfortunately, the problem of indifference to monthlies on urban rail is common around the Anglosphere. Singapore has no season passes at all. In Vancouver, Cubic lobbying and a New Right campaign about fare evasion forced TransLink to install faregates on SkyTrain, and when the faregate project had predictable cost overruns, the campaigners took that as evidence the agency shouldn’t get further funding. London’s fare capping system is weekly rather than monthly – there are no monthly passes, and all fares are set at very high levels. Britain generally overuses faregates, for example on the commuter trains in London. London generally gives off an impression of treating everyone who is not a Daily Mail manager as a criminal. Paris is better, but not by much. The German-speaking world, as irrational as Britain and France about urban crime rates that are far lower than they were a generation ago, still treats the train and bus rider as a law-abiding customer unless proven otherwise.

Social fares

American transit agencies and activists resist calls for large monthly discounts, on a variety of excuses. The most common excuse is revenue loss, which is weird since realistically New York would transition to a large discount through holding the monthly fare constant and hiking the single-ride fare. It’s the second most common excuse that I wish to deal with here: social fares, namely the fact that many low-income riders don’t have the savings to prepay for an entire month.

On social fares, as on many other socioeconomic issues, it is useful for Americans to see how things work in countries with high income compression and low inequality under the aegis of center-left governments. In Paris, various classes of low-income riders, such as the unemployed, benefit from a solidarity fare discount of 50-75%. In both Paris and Stockholm, the monthly pass is flat regionwide, an intentional program of subsidizing regular riders in the suburbs, which are on average poorer than the city.

The flat fare is not really applicable to American cities, except possibly the Bay Area on BART. However, the large fare reductions to qualifying low-income riders are: a number of cities have used the same definition, namely Medicaid eligibility, and give steep discounts for bikeshare systems. On the same principle, cities and states can discount fares on buses and trains.

The right way to view fares

Fares are an important component of public transport revenue; the taxes required to eliminate fares are significant enough that there are probably better uses for the money. By the same token, the issue of fare evasion should be viewed from the lens of revenue loss, rather than that of crime and disorder. The transit agency is not an individual who is broken by being mugged of $100; it should think in terms of its own finances, not in terms of deterrence.

Nor is making it easier to follow the law going to encourage more crime – to the contrary. Transit agencies should aim at a fare system, including enforcement, that allows passengers to get on and off trains quickly, with minimum friction. Turnstiles do not belong in any city smaller than about 10 million people. The fare structure should then encourage long-term season passes, including annual passes, so that nearly all residents who take public transport have already paid. Random inspections with moderate fines are the layer of enforcement, but the point is to make enforcement largely unneeded.

And tear down the faregates.

What is the Anglosphere, Anyway?

As I’m putting more and more urban rail lines and their construction costs into one table, I have to notice trends. One that I’ve talked about for many years is that construction costs in the Anglosphere are higher than in the rest of the developed world, not just in world leader New York but also in other American cities as well as in Britain, Canada, Singapore, and so on. For years I identified this with common law, which I no longer do. Instead, I want to expand on this by asking what exactly the Anglosphere even means.

The features of the Anglosphere

Within the developed world, a subset of countries consists of the Anglosphere. The core is Britain, the US, Canada, Australia, and New Zealand, but Ireland has to be on the list too, as should Singapore and to varying extents Israel and Hong Kong. Which features separate them from the remainder of the first world:

  • For the most part, they use English as their usual language – but Israel, Hong Kong, and Quebec do not, and Singapore only does as a public language while maintaining Chinese, Malay, and Tamil as home languages.
  • They use English common law – but Quebec uses a French-derived code for civil law.
  • They have extensive right to trial by jury – but Israel and Singapore have no juries.
  • They use single-member districts in elections – but Singapore and Hong Kong are undemocratic, Israel and New Zealand use proportional representation, Ireland uses single transferable vote, and Australia’s single-member districts use instant runoff (cf. France’s single-member districts with runoffs).
  • They have higher economic inequality than other developed countries, lower taxes and government spending, and weaker unions – but there are some exceptions (e.g. Canada and Australia are less unequal than Italy, and South Korea and Japan have lower taxes than most of the Anglosphere), and moreover the ranges within both the Anglosphere and the rest of the developed world are quite wide.
  • They make extensive use of privatization and public-private partnerships for infrastructure and services – but Stockholm contracts out its urban rail whereas no major American city does, and France built one of its recent high-speed lines, the one to Bordeaux, as a PPP.
  • The smaller countries see the US, the UK, or both as inspirations for what modern prosperity looks like – but Israel compares itself with both the US and Western Europe (especially Germany), Singapore’s cultural cringe extends toward both the US/UK and bigger East Asian countries, and Hong Kong is torn between Western and Chinese models.

Every distinguishing feature of the Anglosphere can be made to correlate with high construction costs, but that tells us little, because it could be that this is just a spurious relationship, the real cause being something else about the Anglosphere. When making a claim about what makes the US, UK, and Canada so expensive to build in, it’s useful to test it against special cases – that is, countries that are part of the Anglosphere in general but fail that specific criterion.

The legal system

With respect to common law, Quebec is the ideal testing ground. Montreal and Toronto share more social and economic features than do other pairs of major cities with their respective languages. A large Toronto premium over Montreal would suggest that remaining differences, such as the legal code or maybe the peculiarities of Quebec politics, matter to construction costs.

But what we see is the opposite. In the 2000s, Toronto and Montreal both built subway extensions at pretty reasonable costs. Since then, costs have risen in both cities in tandem, placing the planned Blue Line extension in Montreal and the planned Ontario Line and Scarborough replacement in Toronto among the most expensive non-New York subways. So it’s likely that common vs. civil law makes no great difference to costs.

Electoral politics

By the same token as with the use of common versus civil law, we can look at the electoral system. Israel and New Zealand use fully proportional elections, and Israel has national lists, without any local empowerment. Both countries have cheap recent electrification projects, but when it comes to tunneling, both Tel Aviv and Auckland are on the expensive side.

Conversely, France has single-member districts with runoffs; the lack of a spoiler effect weakens political parties, but they’re still stronger than in the US, and in practice independent candidates mostly run explicitly as left or right. Any reasonable mechanism for why single-member districts should raise construction costs should apply regardless of whether these districts are elected by plurality or with runoffs (and besides which, Melbourne has extreme costs and Sydney fairly high ones). And yet, French costs are decidedly average: Grand Paris Express is the median world subway by construction costs, and other Metro extensions in Paris and other French cities are somewhat cheaper.

Unions and inequality

The political factor – the Anglosphere’s socioeconomic policy is generally to the right of that of Continental European countries – has its own special cases too. The American left and center-left has in particular seized upon the importance of health care to construction costs, since the US has high health care costs and employers, especially in the public sector, are expected to pay most of the costs of workers’ health insurance. But the UK and Canada both have largely public systems that the American left uses as inspiration for its single-payer health care plans, and the UK also has very good cost control; and yet both countries have very high infrastructure construction costs. Singapore, whose health care system is private and unequal but also low-cost, has very expensive subway construction as well.

We can similarly look at inequality in general, or at union power. The correlation between inequality and national construction costs should be fairly high, if only because the Anglosphere has high inequality as well as high construction costs. However, per Branko Milanovic’s data for after-tax-and-transfers inequality, Canada, Britain, and Australia all have slightly lower inequality than Spain, and are comparable to Greece and Italy.

Unions can affect construction costs in either direction. The American center-right and right complain that the power of public-sector unions warps public incentives and forces high construction and operating costs, citing union hostility to productivity improvements that include layoffs, or such regulations as prevailing wage laws. However, the most unionized countries in the developed world are in Scandinavia, where costs are low. The OECD has union density figures by country, and the big cleave is Scandinavia versus the rest. The Anglosphere is on the weaker side.

Perhaps the correlation must then go the other way? That is, weak unions increase costs, for example by creating a siege mentality among those workers who do have stable union jobs (including rail workers, as the industry’s economic and political situation is friendly to unionization)? But the data does not support that, either. Spain’s union density is barely higher than the US’s and much lower than Britain’s, and Greece’s is comparable to Britain’s. The available data strongly suggests that union power has no effect on construction costs, positive or negative.

Could it be privatization?

Privatization and the reliance on PPPs is the least clean of the Anglosphere’s special features – that is, it is not always used throughout the countries I identify with the Anglosphere, and conversely it may be used elsewhere, even in countries with generally left-wing economic policy like Sweden. Nonetheless, among the political, legal, social, and economic factors, it is the only one I cannot rule out.

The issue is not precisely contracting out something, as Stockholm is doing with urban rail. Rather, it is more specifically privatizing the planning aspects of the state, such as engineering. Spain relies heavily on in-house engineering and design, while the US and UK, and by imitation the rest of the Anglosphere, prefer private consultants. To the extent I have cost comparisons within the same city or country with different levels of privatization, they’re suggestive that it matters: the publicly-funded LGV Est Phase 2 cost €19 million per kilometer (with a tunnel covering 4% of the route), the PPP LGV Sud-Europe-Atlantique cost €23 million per kilometer (with no tunnels), the two lines opening within a year of each other. This is not an enormous cost difference, but accounting for the tunnel makes the cost noticeable, perhaps a factor of 1.5.

Conclusion

Overrelying on a single case is not particularly robust. In light of the similarities between costs of different lines in the same city, and even those of different cities in the same country, the N for a quantitative comparison is not large – my data table currently has 38 unique countries, and even accounting for a few misses for which I haven’t included data yet, like Israel, the number is not much larger than 40. It is not responsible to use multivariable regressions or other advanced statistical techniques in such a situation.

In that case, looking at one or two cases provides a powerful sanity check. As far as I can tell, the Anglosphere’s tendency toward privatization and using consultants, often reinforced by different English-speaking countries learning one another’s practices, could be a serious cost raiser. However, the other special features of the Anglosphere – common law, winner-take-all elections leading to two-party systems, and weak unions and welfare states – are unlikely to have a significant effect.

Construction Costs are National

I’m about to send a thinktank a draft of a table of subway construction costs, and I’d like to preview one of the most important findings from the data. This is based on 125 distinct items, totaling 2,297 kilometers – some complete, some under construction, a handful proposed. I’ve alluded to this here before, for example when writing about national traditions (US, Soviet, UK) or about Russian and Nordic costs. But the basic observation is that construction costs are not really a feature of an individual metro line, but of a city, and usually an entire country.

What this means is that if one line in Madrid is cheap, then we can expect other lines in Madrid to be cheap, as well as in the rest of Spain; if one line in London is expensive, then we can expect other lines in London to be expensive, as well as in the rest of the UK. In fact, in both countries the construction costs of metro systems in the capitals also accord with the construction costs of intercity high-speed rail: cheap in Spain, expensive in Britain, with Germany somewhere between Spain and Britain and France somewhere between Spain and Germany.

Some examples

The examples in this section are somewhat cherrypicked to be the ones with narrower ranges, but there are very few examples with truly large ranges over a similar period of time (i.e. not secular increases as in Canada). I am specifically excluding regional rail, as it tends to be more expensive per kilometer than subways.

Panama: Line 1 cost around PPP$260 million per kilometer for 53% underground construction, and Line 2 is cheaper, around $150 million, but is entirely above-ground. This is consistent with a factor-of-2.5 underground premium over elevated lines, well in line with the literature.

Greece: Athens Line 4 is €104 million per km, with construction having started recently. Thessaloniki has two lines in the database, the main line due to open next year and an extension to Kalamata due to open in 2021, and Athens is also about to wrap up an extension of Line 3 to Piraeus. All cost figures may be found here on PDF-p. 9. The two Thessaloniki projects are respectively €135 million/km and €118 million/km, the former at least including rolling stock and I believe the latter too; the Athens Line 3 extension, without rolling stock and with somewhat wider stop spacing, is much cheaper, €61 million/km, but this rises to €82 million/km with rolling stock.

Sweden: the Stockholm Metro extensions under construction all cost pretty much the same per kilometer. Three extensions are under construction at once, in three different directions; per this source, the costs per kilometer (in kronor) are 1 billion, 1.25 billion, and 1.15 billion, with the most expensive of the three involving brief underwater tunneling.

Russia: I asserted in an old post that Russian construction is expensive, with only a handful of projects. Since then I’ve found a source asserting that the entire 2011-20 program is 1.3 trillion rubles, for what appears to be 150 km, 57% underground. This is in PPP terms $364 million per km. Other costs are vaguely in that range – Railway Gazette claims the cost of boring in Moscow is (again in PPP terms) $400-600 million/km, Line 11 is around $310 million/km for underground suburban construction, one line mentioned on Railway Gazette in St. Petersburg is $310 million/km underground, another St. Petersburg line is maybe $360 million/km.

What does this mean?

That there’s correlation between different cities’ construction costs within the same country suggests the differences in costs are predominantly institutional or socio-political, rather than geological. This is further reinforced by looking at countries with very similar socio-political regimes, namely the Nordic countries: all of them are cheap, and even though Stockholm and Helsinki both have similar gneiss geology, the Oslo line I use for comparison does not (and neither does somewhat more expensive Copenhagen).

To further reinforce the institutional point, the costs of high-speed rail in different countries seem to follow the same order as the costs of metros. Spain is cheap: Ferropedia quotes construction costs below €20 million per kilometer. The UK, in contrast, just announced a cost overrun on HS2, a 540 kilometer network, to £88 billion, and even allowing for future inflation, this is maybe 7 or 8 times as expensive as in Spain. France and Germany are in between, in the same order as their metro costs. China, as far as I can tell comparable to France in its metro construction costs, has a high-speed rail construction cost range somewhat higher than France’s, mostly explainable by using more (generally avoidable) viaducts.

When Reliability Matters Above All Else

This post is about situations in which the most important thing for transportation is reliability, more so than average speed or convenience. It’s inspired by two observations, separated by a number of years: one is my own about flying into or out of Boston, the other is from a New York Times article from yesterday describing a working-class subway rider’s experience.

My observation is that over the years, I’ve used Logan Airport a number of times, sometimes choosing to connect via public transportation, which always involves a bus as the airport is not on the rail network, and other times via taxi or pickup. My choice was always influenced by idiosyncratic factors – for example, which Boston subway line my destination is on, or whether I was visiting someone with a car and free time. However, over the last eight years, a consistent trend is that I am much more likely to use the bus arriving at the airport to the city than departing. I know my own reasoning for this: the bus between South Station and the airport is less reliable than a cab, so when in a crunch, I would take a train to South Station (often from Providence) and then hail a taxi to the airport.

The New York Times article is about a work commute, leading with the following story:

Maribel Burgos barely has time to change into her uniform before she has to clock in at the McDonald’s in Lower Manhattan where she works, even though she gives herself 90 minutes to commute from her home in East Harlem.

It does not take 90 minutes to get between East Harlem and Lower Manhattan on the subway. The subway takes around half an hour between 125th Street and Bowling Green, and passengers getting on at one of the local stations farther south can expect only a few minutes longer to commute with a cross-platform change at Grand Central. Taking walking and waiting time into account, the worst case is around an hour – on average. But the subway is not particularly reliable, and people who work somewhere where being five minutes late is a firing offense have to take generous margins of error.

When is reliability the most important?

What examples can we think of in which being late even by a little bit is unacceptable? Let us list some, starting with the two motivating examples above:

  • Trips to the airport
  • Work trips for highly regimented shift work
  • Trips to school or to an external exam
  • Work trips for safety-critical work such as surgery
  • Trips to an intercity train station

In some of these cases, typically when the riders are of presumed higher social class, the system itself encourages flexibility by arranging matters so that a short delay is not catastrophic. At the airport, this involves recommendations for very early arrival, which seasoned travelers know how to ignore. At external exams, there are prior instructions of how to fill in test forms, de facto creating a margin of tolerance; schools generally do not do this and do mark down students who show up late. Doctors as far as I understand have shifts that do not begin immediately with a life-critical surgery.

But with that aside, we can come up with the following commonalities to these kinds of trips:

  • They are trips to a destination, not back home from it
  • They are trips to a fairly centralized and often relatively transit-oriented destination, such as a big workplace, with the exception of regimented shift work for retail (the original NY Times example), which pays so little nobody can afford to drive
  • They are disproportionately not peak trips, either because they are not work trips at all, or because they are work trips for work that is explicitly not 9-to-5 office work
  • They are disproportionately not CBD-bound trips

The first point means that it’s easy to miss this effect in mode choice, because people can definitely split choice between taxis and transit or between different transit modes, but usually not between cars and transit. The second means that driving is itself often unreliable, except for people who cannot afford to drive. The third means that these trips occur at a point in time in which frequency may not be very high, and the fourth means that these trips usually require transfers.

What does reliability mean?

Reliability overall means having low variance in door-to-door trip time. But for the purposes of this discussion, I want to stress again that trips to destinations that require unusual punctuality are likely to occur outside rush hour. Alas, “outside rush hour” does not mean low traffic, because midday and evening traffic in big cities is still quite bad – to take one New York example with shared lanes, the B35 steadily slows down in the first half of the day even after the morning peak is over and only speeds up to the 6 am timetable past 7 pm. Thus, there are twin problems: frequency, and traffic.

Traffic means the vagaries of surface traffic. Buses are generally inappropriate for travel that requires any measure of reliability, or else passengers have to use a large cushion. Everything about the mixed traffic bus is unreliable, from surface traffic to wait times, and bunching is endemic. Dedicated lanes improve things, but not by enough, and unreliable frequency remains a problem even on mostly segregated buses like the Silver Line to the airport in Boston.

Frequency is the harsher problem. The worker commuting from Harlem to Lower Manhattan is if anything lucky to have a straight-short one-seat ride on the 4 and 5 trains; most people who need to be on time or else are not traveling to city center and thus have to transfer. The value of an untimed transfer increases with frequency, and if every leg of the trip has routine 10-minute waits due to bunching or just low off-peak frequency guidelines, the trip gets intolerable, fast.

What’s the solution?

Bus redesigns are a big topic in the US right now, often pushed by Jarrett Walker; the latest news from Indianapolis is a resounding success, boasting 30% increase in ridership as a result of a redesign as well as other changes, including a rapid bus line. However, they only affect the issue of reliability on the margins, because they are not about reliability, but about making base frequency slightly better. New York is replete with buses and trains that run every 10-15 minutes all day, but with transfers, this is not enough. Remember that people who absolutely cannot be late need to assume they will just miss every vehicle on the trip, and maybe even wait a few minutes longer than the maximum advertised headway because of bunching.

Thus, improving reliability means a wider toolkit, including all of the following features:

  • No shared lanes in busy areas, ever – keep the mixed traffic to low-traffic extremities of the city, like Manhattan Beach.
  • Traffic signals should be designed to minimize bus travel time variance through conditional signal priority, focusing on speeding up buses that are running slow; in combination with the above point, the idea of giving a late bus with 40 passengers the same priority at an intersection as a single-occupant car should go the way of the dodo and divine rights of kings.
  • Off-peak frequency on buses and trains needs to be in the 5-8 minute range at worst.
  • Cross-platform transfers on the subway need to be timed at key transfer points, as Berlin manages routinely at Mehringdamm when it’s late and trains run every 10 minutes (not so much when they run every 5); in New York it should be a priority to deinterline and schedule a 4-way timed pulse at 53rd/7th.
  • Branch scheduling should be designed around regular gaps, rather than crowding guidelines – variation between 100% and 130% of seats occupied is less important to the worker who will be fired if late than variation between waiting 4 and waiting 8 minutes for a train.
  • Suburban transit should run on regular clockface schedules every 30, 20, or 15 minutes, with all transfers timed, including with fare-integrated commuter trains.