Category: High-Speed Rail
Quick Note: California HSR Could Save $4 Billion on the Grapevine
California HSR’s just-released July progress report, as reported on bakersfield.com, contains the pleasant surprise that switching the alignment from the Tehachapis and Palmdale to the I-5 alignment on the Grapevine could save $4 billion.
Furthermore, the study indicating such cost savings “identified more than one feasible alignment over the mountain pass.” The Grapevine option was rejected in 2005 because the preliminary engineering found only one feasible alignment that crosses known faults at-grade and has a maximum tunnel length of 6 miles and maximum 3.5% grade, compared with hundreds through the Tehachapis. Therefore finding multiple alignments, such that even if further meter-scale geological studies discover new faults then some option will make it through, is likely to tilt the field back toward the Grapevine.
Robert Cruickshank is surprisingly pessimistic about the Grapevine, on the grounds that Palmdale is an important market to serve. In reality, Palmdale is a small commuter market – i.e. it has a strong peak and low revenue per rider – so giving it up is a small deal, probably fully canceled out by the gain of about 10 minutes’ trip time on the shorter Grapevine.
But most importantly, it’s most important to get an initial operable segment ready, and this means connecting the Central Valley to the LA Basin. As I’ve explained before, a major advantage of the Grapevine is that it allows connecting to the legacy Metrolink line at Santa Clarita rather than at Palmdale, avoiding tens of kilometers of sharp curves on the climb between the LA Basin and Antelope Valley.
I’m unable to find the progress report, so I don’t know to what extent “$4 billion in savings” literally means coming in $4 billion under budget. If it does, it means that theoretically, the money available suffices to build from Los Angeles to a point between Bakersfield and Fresno; Obama’s now-moribund $4 billion for HSR, matched 50:50, would be more than enough to build from Los Angeles to Fresno.
Update: here is the progress report. The relevant section is on page 27. It says only that “an alternative via the Grapevine may save between $1B and $4B in capital cost” – still unclear whether it means coming $1-4 billion under budget, or staying within budget while avoiding a $1-4 billion cost overrun on the Tehachapis.
It’s too bad the approximate amount remains unclear. The required budget is on the same order as the amount that may become available in the next two years depending on Congressional machinations, and so it’s important for California to know how much it should be asking for. For example, if it were made clear that an additional $2.5 billion in federal funding were enough to complete LA-Fresno, then Dianne Feinstein might try to include the full amount for high-speed rail in the transportation bill for 2012 rather than just $100 million.
Obama Proposes $4 Billion for HSR
President Obama’s new jobs bill includes $50 billion for infrastructure construction, including $10 billion for an infrastructure bank, $4 billion for high-speed rail, and $2 billion for Amtrak. Assuming it can get past the Republican Congress and that it will not be watered down as it already has been since the beginning of the year, the question arises: where to spend the money?
Fortunately, the separate grant for Amtrak suggests that the Northeast Corridor will be funded from a separate pile of money. This means that it’s more feasible to spend 100% of the HSR money in California. I claim that, in light of California’s present funding situation, this is the best possible use of the money, and, furthermore, the federal government should let California know of this as soon as possible, before it lets contracts out to tender.
Recall that California’s present HSR money is sufficient to build from Bakersfield to a point between Fresno and Merced, at least in principle, as the Environmental Impact Report projects slightly higher costs. Recall further that the $8 billion that could be made available to California – Obama’s $4 billion plus matching funds from Proposition 1A – are more or less enough to build from Bakersfield to Sylmar.
More precisely, the cost estimate for Bakersfield-LA is $12.6 billion, but according to CARRD, which independently of this also thinks the cost is going to be $18.6 billion, Palmdale-Sylmar is half the cost of Palmdale-LA, and as a result adding up Bakersfield-Sylmar using the 2009 Business Plan numbers works out to just under $9 billion. The approximately $1 billion in missing funds could either be obtained from local or private sources, or diverted from the plans to build north of Fresno; the segment that goes through and north of Fresno is expected to cost $1-2 billion, and diverting all north-of-Fresno money to Bakersfield-Sylmar should suffice to build the system from Sylmar to Fresno, with a cheap electrified legacy onward connection to LA.
Alternatively, if it turns out that going from Bakersfield to the LA Basin through Tejon Pass rather than through Palmdale is cheaper, then it’s possible to terminate the line in Santa Clarita and have trains continue further south at lower speed. This is in principle possible even through Palmdale, but then the legacy segment of the line would be both longer and curvier.
In other words, by spending all possible HSR money in California now, the Obama administration can guarantee a useful initial operating segment from LA to Fresno. On the margin the benefit of this is much bigger than its share of the cost, since it makes the difference between an upgraded San Joaquins train and a Phase 0 high-speed line.
If the administration funds California in full, then people will be able to ride a fairly long segment at full speed, connecting at lower speed to a major city. Some people are still going to call this a train to nowhere until it connects to San Francisco, but fewer people will use this epithet on LA-Fresno than on Bakersfield-Merced.
The primary problem with American transportation planning is that there is no transportation policy in the US. There is an industrial policy, a jobs policy, and construction for pizzazz on both sides, as well as the joy of hippie-punching among conservatives. An open HSR segment that is not a complete cost or ridership disaster could at least blunt the hippie-punching, if not develop local expertise that could eventually lead to transportation policy. In countries where HSR is in operation, or something close enough to it, the conservatives do not oppose its construction, even quite right-wing ones such as Berlusconi and Cameron.
The worst thing that can be done is spreading the money thin. The not-really-high-speed lines funded elsewhere, or, even worse, funding to Amtrak’s massively overpriced Vision plan, can only lead to small, barely noticeable improvements, ensuring there are plenty of disaffected people to continue the treatment of intercity transportation as a cultural political football. The only place where $4 billion in federal money makes a difference between having a usable system and not having one is California, and this is the basket the administration should put its eggs in.
California High-Speed Rail Alignment Questions
The most contentious technical issue about the California High-Speed Rail project is which alignment to use to get from the Central Valley to the Bay Area. The two options are Altamont Pass, roughly paralleling 580, and Pacheco Pass, much farther to the south. A summary of all alternatives can be found on page 115 of the revised Bay Area-Central Valley EIR. For more detailed examination of the alternatives, see the old EIR: the base Altamont option is on pages 903-4, the base Pacheco option on pp. 969-70. Although Altamont is somewhat longer, the two alignments are about even on travel time from Los Angeles to San Francisco (in fact, Altamont is 2 minutes faster).
The basic tradeoff is that Pacheco is somewhat faster for LA-San Jose and serves San Jose and San Francisco on one line, while Altamont is much faster for Bay Area-Sacramento and requires less construction overall and has separate branches to San Francisco and San Jose. Overall, Altamont is superior because of its advantage for travel from the Bay Area to Sacramento and the Upper Central Valley (except Merced, whose commute ties to the Bay Area are weaker than those of Modesto and Stockton). Transit activists and environmentalists either preferred Altamont or did not have an opinion. However, San Jose didn’t want to be left on what it perceived as merely a branch, and lobbied hard for Pacheco, and as a result Pacheco became the preferred alternative; in addition, unlike the NIMBYs on the Peninsula, the NIMBYs in Pleasanton and Tracy complained about HSR early.
A third option is to go via Altamont but enter San Francisco from Oakland via a second Transbay Tube (old EIR, pp. 957-8). The EIR projected it to have the highest ridership, since it serves both San Francisco and Oakland on one branch and has the shortest LA-SF travel time. It was rejected because a second tube would be very expensive, though in fact the EIR pegs the cost of this option at a few hundred million dollars more than the base Pacheco and Altamont options; urban construction along the Caltrain line is expensive as well. In a crunch trains could continue along an electrified but not otherwise upgraded Caltrain line at lower speed, reducing cost, but by a similar token people could transfer to BART at Livermore under any Altamont option and at West Oakland under a second tube option. However, should a second tube be built anyway to relieve the near-capacity BART tube, such an option would become far and away the best, making all others redundant.
The choice of Pacheco became one of the galvanizing features of the technicals in California, who without exception preferred Altamont. To answer concerns that Bay Area-Sacramento travel has to be served, both the HSR Authority and various politicals have proposed a cure that’s worse than the disease: build a high-speed commuter overlay along Altamont (the official version) or the I-80 corridor used by Capitol Corridor trains (consensus among pro-Pacheco blog commenters, see e.g. this map with a second tube just for SF-Sacramento trains).
Pacheco itself is mildly defensible. It would arguably have been superior if Sacramento did not exist, and I-80 would have been the better alignment for SF-Sacramento service if LA did not exist. But given that LA and Sacramento both exist, Altamont’s ability to serve LA, Sacramento, and SF with just one expensive bit through the pass becomes more valuable. If Altamont is built, there would be no point in a Pacheco overlay, whose primary use would then be a frankly uncompetitive connection to Monterey. But Pacheco leads to demands for an overlay service, one that’s almost certainly too expensive to build just for Bay Area-Sacramento travel.
The Capitol Corridor, the other option for SF-Sacramento service, is too slow. With a bus connection from SF to Emeryville, the fastest service takes 2:08 from downtown SF to Sacramento. Even Pacheco beats that: express trains detouring through Gilroy and Merced will nominally take 1:53; service via Altamont is a little more than an hour . For SJ-Sacramento travel, it’s 3:05 on Amtrak and 1:24 via Pacheco. Substantial upgrades are impossible since Union Pacific owns the track and restricts passenger trains’ performance in order to remove a headache for freight operations. The remaining option is to build passenger-dedicated bypasses, at considerable cost and with little benefit over doing it right the first time.
Construction Costs, Third World Edition
It’s a commonplace that building things is cheap in third-world countries, with low wages, few labor and environmental controls, and lax regulations. The reality is quite different. The difference disappears once one makes sure to do a PPP adjustment; poor countries’ currencies are persistently undervalued relative to their PPP exchange rate, and often also relative to true market value, and this could lead to a distortion in cost structure.
Recall that in Continental Europe, a fully-underground subway line costs anywhere between $110 million and $250 million per km, removing one outlier at each end from my list. Spanish construction costs are generally much lower than the European average, with commuter tunnels coming in well under $100 million/km.
In Delhi, the Metro’s construction costs are very high. The next phase involves 108 km, of which 41 are underground and the rest elevated, and is scheduled to cost 30,000 crores. At current exchange rates this is $6.7 billion, but at the PPP rate it’s $17.6 billion, i.e. $163 million per kilometer. Such a cost is normal by European standards for a fully-underground line; it’s not normal for a line that’s majority-elevated. It is almost as expensive as mostly-above ground extensions of American lines, for example the Silver Line in Washington.
In Beijing, the subway construction costs are also higher than one would expect given low wages, but only about as high as those of Europe. Fully-underground lines are about $150 million per km: these include Line 8 Phase 2 ($2.5 billion/17 km), Line 6 Phase 1 ($4.9 billion/30 km), and Line 14 Phase 1 ($4.5 billion/30 km); the first two are confirmed to be fully underground, and while I can’t find a claim in either direction for the last, all lines it intersects are fully underground. Chinese high-speed rail costs are quite similar to European costs as well: the lines rated at 350 km/h are between $19 and 50 million per km; there’s little tunneling on most lines, but long viaducts, e.g. the $42 million/km Beijing-Shanghai HSR line is 1.2% in tunnel and 86.5% elevated.
In Baghdad, the under-construction above-ground metro line, built by Alstom, is costing $1.5 billion for 22–25 km. With a PPP adjustment, this goes up to $83-94 million/km, depending on whose report of the line’s length one believes. It’s better than India, but not especially good.
Turkey is proving itself to be the Spain of the developing world. Its construction costs are often high per kilometer, but only because Istanbul’s geography is such that lines have to cross under major bodies of water, in seismic terrain. Marmaray, a commuter rail tunnel connecting the European and Asian halves of the city, cost $3.5 billion for 13.6 km of tunnel; while the overall cost, $333 million/km after PPP conversion, is high, it must be weighed against the extreme complexity of the project. The extension of the Istanbul Metro’s M2 line going under the Golden Horn rather than the Bosporus, is $148 million/km, again with PPP conversion. In contrast, the fully underground first phase of M4 is, if I understand the reference, and that’s a big if, $40 million per km (add all three cost amounts, then convert to US dollars); when a line goes underground rather than underwater, Istanbul builds it as cheaply as Madrid. Mainline rail construction in Turkey is also inexpensive: Turkey plans to build 14,000 km of rail, with a substantial portion permitting 250 km/h speeds, for $45 billion; that’s $4 million per km.
Iranian construction costs are low as well. Tehran Metro Line 3, as usual after PPP conversion, is $61 million per km; it is two-thirds underground.
Although there are no third-world lines that have breached $500 million per km, as several first-world lines have, this is probably entirely due to the fact that India, with the highest construction costs, builds its subways mostly above ground. A fully underground Delhi Metro line will probably cost as much as one in Tokyo, despite Delhi’s much less densely built existing network.
The pattern we see here is, first, that the one country on the list following the English legal and political tradition also has English construction costs. And, second, third-world countries do not build rail more cheaply than first-world countries, after adjusting for living costs but not wages; in other words, they spend more of their income on building those lines.
While labor costs in China are lower than in Europe, so is the productivity of labor. If everything in China cost across the board less than in the first world, it would be as rich as the first world; the reason it’s not as rich is precisely that labor doesn’t go as far as in more industrialized countries. China’s rapid growth should be thought of as a process of catching up to what the developed world learned over two hundred years of industrialization that has made it so much more efficient now than it was in 1800.
MBTA-HSR Compatibility
There is going to be major investment in the Northeast Corridor, and several possibilities, including Amtrak’s NEC Master Plan, call for running trains at higher frequency and somewhat higher speeds than today on the Providence Line, and assumes electrification of commuter service. Since the line is already being used by the MBTA, which according to Amtrak is limiting the number of intercity train slots for capacity reasons, this calls for a good measure of schedule integration, based on the principle of organization before electronics before concrete.
Amtrak’s Master Plan calls for three-tracking the entire Providence Line south to Attleboro (one viaduct excepted) instead, at a cost of $464 million – $80 million in Phase 1, $384 million in Phase 2 – in addition to money spent on unnecessary expansion at South Station. Such a cost is excessive, suggesting that better MBTA-HSR compatibility is required. Full-fat HSR programs go even further and avoid the Providence Line in favor of a greenfield alignment or an I-90 alignment, instead of making use of the existing high-speed track in Rhode Island and Massachusetts. To reduce costs, a better plan would four-track short segments for passing sidings, and time the overtakes. The principle is similar to that of the blended Peninsula plan in California, in the version proposed by Clem Tillier.
In many ways, for example the metro area populations involved and the current ridership level, the Providence Line is similar to the Caltrain line. The main difference is that the Providence Line has fewer stops and therefore can expect higher average speeds. In addition, the Providence Line is straighter and passes through less developed areas, so that even today Acela trains plow it at 240 km/h, and about 330 km/h is possible with true high-speed trains and higher superelevation.
In Switzerland, trains run as fast as necessary, not as fast as possible. In this context, this means running just fast enough to meet a good clockface schedule. Boston-Providence travel time on the MBTA today is about 1:10; for a good takt, this should be cut to about 55 minutes, allowing hourly service with two trainsets and half-hourly service with four.
For the purposes of schedule symmetry and avoiding switching moves at high speed, passing segments should have four tracks rather than three when possible. Costs should be controlled by making those passing segments much shorter than the three-tracking Amtrak proposes.
Finally, the timetables proposed here are based on the following performance assumptions: regional trains have a top speed of 160 km/h, accelerate like a FLIRT (45 seconds acceleration plus deceleration penalty), have an equivalent cant of 300 mm, and dwell at stations for 30 seconds. Intercity trains accelerate like an idealized N700-I, have an equivalent cant of 375 mm, and dwell for 60 seconds. The equivalent cant is by and large unimportant; the acceleration and dwell times for regional trains are. The approach into and out of South Station has a speed limit of 70 km/h through the 90-degree curve toward Back Bay, and 100 km/h to south of the curve at Back Bay; intercity trains are limited to 200 km/h south to Readville and 250 km/h south to the Canton viaduct, and, at the southern end, 225 km/h west of the curve in Attleboro and, curves permitting, 200 km/h in Rhode Island. Regional trains turn in 5 minutes, or 4 at a minimum, and intercity trains turn in 10 minutes at a minimum. Signaling allows a headway of 2 minutes at a speed of 200 km/h and 3 minutes at higher speed, but if a regional train starts from a siding stop, it can follow a high-speed train more tightly initially, say 1 minute, still far higher than a safe stopping distance, since the spacing rapidly increases over time. Grades are ignored; the Providence Line is flat enough that they’re not an issue. Timetables should be padded 7% from the technical time.
With the above assumptions, the technical time for regional trains is 38 minutes with the present stopping pattern, which yields 41 minutes with padding; this compares with 46 minutes for the fastest Acela. Clearly, if Acela service levels remain similar to what they are today – which includes the Master Plan, which calls for a 10% reduction in Boston-New York travel time (see page 40 on the PDF linked above) – there’s no need for passing segments. To raise travel time to 55 minutes, trains should make more frequent stops, and/or run to T. F. Green Airport always. Although the speed profile of regional and intercity trains would be different, the average speed would be the same, and given that the corridor has a small number of trains per hour of each type, this mismatch is no cause for concern. The $464 million Amtrak is proposing would then be a complete waste, and the federal government should spend any money toward this goal on electrifying more MBTA lines and funding EMUs.
However, in a scenario involving a significantly improved intercity service, the best technical time for nonstop Boston-Providence service with a top speed of 300 km/h decreases to about 19 minutes (20.5 with pad), and this makes overtakes necessary. A slowdown to 250 km/h only adds about one minute of travel time, so the operating pattern is almost identical.
If 15-minute service, both regional and high-speed, is desired, then regional trains can be about 11 minutes slower between successive passing segments, since 11 = 15-3-1 or 15-2-2. A single mid-line overtake is theoretically possible: 41-20.5 = 20.5 < 2*11. However, such an overtake would have to be exactly at the midline, and, in addition, there could be merge conflicts at Providence, whose station tracks include two on the mainline and two on one side of the mainline as opposed to one on each side.
It’s still possible, but tight, to have a single overtake at Sharon. The immediate station vicinity would be four-tracked; this is no trouble, since the area around the station is undeveloped and reasonably flat. In addition, there’s more than enough time in the Providence area, making the merge conflict a lesser problem. However, this is very tight near Boston South, beyond signaling capability unless four-tracking extends a few kilometers further north. One way to counter this problem is to slow high-speed trains by making them all stop at Back Bay and/or Route 128, adding precious minutes to the schedule but reducing the speed difference. Conversely, the current weekday pattern of Providence Line trains skipping Ruggles could be made permanent. There is no room for infill stops; the overtake would only add 4 minutes to regional train travel time, so there’s time to run further to the airport at 160 km/h, and even make an extra stop at Cranston.
Another possibility is to have two overtakes, taking advantage of existing four-tracking around Attleboro. The capital costs are similar; it would require four-tracking around Route 128, possibly extending north to Readville if an on-the-fly overtake is desired. The operating complexity is much higher, since there’s one more opportunity for a late train to mess up the entire schedule. However, there is plenty of slack south of Attleboro and north of Route 128 allowing for additional stops. Under this option, the train loses 4 minutes waiting at Attleboro and about 2.5 at Readville, since the overtake is not completely on-the-fly, raising travel time to 47.5 minutes. There’s no time for airport trains, not on the same takt. However, there’s space in the schedule for 5-6 infill stops in addition to Readville; Forest Hills, Pawtucket, Central Falls, and perhaps one more in each of Boston and Providence closer to city center.
In principle, it’s possible to extend this analysis to 10-minute service, with three overtake segments, at Route 128, Sharon, and Attleboro. In practice, this is operationally cumbersome, and the operating profits coming from filling six full-length high-speed trains from New York to Boston ought to be able to pay for four-tracking the entire line, even the viaduct.
Not included in this analysis are the branches. Those are not a worry since north of Readville there are three tracks, and frequencies on the other lines are low. The Stoughton Line is a bigger problem; however, with the three tracks through Boston, it could still be shoehorned. Electrifying it should not be difficult due to its short length, though the proposed Taunton extension would make it harder.
Every Time You Justify Infrastructure on Competitiveness Grounds, A Kitten Dies
You’ve heard it before: the US is falling behind China and Europe, and has to build more infrastructure to stay competitive in the 21st century. It’s unavoidable in almost any Thomas Friedman article. Boosters, construction industry interests, and even ordinary high-speed rail supports keep asking, how can a country grow without matching other countries’ HSR investment? Never once do they stop to ask why HSR should do anything to help increase competitiveness, beyond vague promises about reducing oil dependence and carbon emissions, issues for which HSR is roughly priority #20.
Countries do not in fact compete with one another. This is made clear in Paul Krugman’s 1994 article in Foreign Affairs, Competitiveness: A Dangerous Obsession. If China builds HSR and becomes richer as a result, the US does not suffer. It’s not competing with Chinese productivity in any meaningful way. In principle, the effect on US wages could be negative if production moves to China or positive if the larger Chinese market buys more American goods; in practice, the effect of other countries’ growth on the US is negligible.
But let’s zoom in and discuss how exactly HSR, or other large infrastructure projects, could lead to more competitiveness. They could boost productivity, but that is mostly an issue for freight transportation. Passenger transportation is mainly a consumer product, not a producer product. In fact, during its own spurt of fast growth from the 1960s to 1997, South Korea lagged in building passenger transportation, explicitly because it prioritized capital investments in industry over such consumer products as highways.
International corporations looking for a place to site a new factory will not look at the general infrastructure situation; they’ll look at what’s useful to their needs. Nissan chose Smyrna, Tennessee for its plant because it had good freight rail and Interstate access and was in a low-wage, anti-union state. The closest thing to passenger-oriented infrastructure that we could look at in such cases is international airports, and the Nashville area only has a small one; Nissan, and the other Japanese and European companies locating plants in the South, would have clustered in Atlanta, Dallas, and Houston if they’d cared.
Let’s zoom in even more, specifically on Nissan and what it’s done to Smyrna. Smyrna is a company town; Nissan even told it to zone the area around the plant as industrial-only, on the theory that commercial development would distract the workers too much. In any other context, the proponents of competitiveness and high-value-added industrial policy would decry such cases as a race to the bottom; and yet, those are among the few situations in which there’s actual competition among regions. The local drivers of a productive economy, rather than one that’s simply a passive recipient of other companies’ transplant factories, have nothing to do with infrastructure megaprojects. Silicon Valley exists because of Stanford, not because of the Peninsula Line or US 101.
At least, there’s competition among regions looking for foreign investment. In other contexts, it’s not as clear. The effects of HSR on national economic growth are too small to be visible, which means that it’s impossible to conduct a study that reliably tells if they exist. But the effects on regional development, a related trope, are decidedly mixed. It’s clear that HSR promotes development near the station; it’s unclear whether it actually develops the surrounding areas, rather than merely concentrates development near the station. Evidence from the Shinkansen as well as other high-speed systems is decidedly mixed – see for example this review.
Building public infrastructure is not a race. Other countries’ experience is a good teacher of what works and what doesn’t, and, provided adjustments for different circumstances are made, can help gauge whether HSR will be successful in the US. However, there is a very big difference between saying that HSR succeeded on a route similar to an American proposal and saying that the US must build because other countries are building as well.
As Krugman notes, the mentality of treating things as if they were races oversimplifies, and leads to bad projects. In the case of transportation, it means focusing on visibility, prestige, and spectacle rather than on cost-effectiveness, usability, and mode share. This is where development-oriented transit comes in: one of the causes of airport transit boondoggles is the insistence of cities and airport authorities that their airport access be world-class, which means a no-expense-spared people mover or, worse, premium rail link to downtown. Those projects, too, often come with promises of competitiveness, as if an airline is going to choose its hub based on the existence of a rail link with a 10% mode share rather than low landing fees or proximity to many travelers and destinations.
At least, development-oriented transit is transit. Paul Barter’s thesis explains how in the postwar period, Asian cities often started building freeways simply because that was what the US was doing and they wanted to be modern. I’m most reminded by the line from the Onion, attributed to the Chinese government: “this year, a million people in China will die from cancer – cancer is a very modern disease.” HSR exists largely because Japan National Railways President Shinji Sogo refused to accept a railway decline and instead built the Tokaido Shinkansen. Although HSR is not freeways, some of the rhetoric coming from various boosters glorifying China’s lack of environmental and community protection has the same basic problem of placing a national race over quality of life.
(Some) HSR projects are good economic and transportation development; they should be sold as good economic and transportation development. Read this summary on Reason & Rail and note how nowhere does Paulus Magnus mention competitiveness. Japan didn’t build the Shinkansen in order to compete with anyone, and France and Germany didn’t build the LGVs and ICE system in order to compete with Japan. If what they’ve done has succeeded then it’s likely that similar American lines could also succeed and should be built, but it’s not a race and the concept of being behind or of needing to imitate what others have done promotes boondoggles, not good transit.
Cost Overruns: How I Learned to Stop Worrying and Hate Bent Flyvbjerg
Let me preface this post by saying I have nothing against Bent Flyvbjerg or his research. My problem is purely with how it’s used in the public media, and frequently even in other academic studies, which assume overruns take place even when they do not.
Stephen Smith sent me a link to an article in The Economist complaining about cost overruns on the California HSR Central Valley segment. The article gets its numbers wrong – for one, the original cost estimate for Merced-Bakersfield was never $6.8 billion, but instead was $7.2 billion in 2006 dollars and $8 billion in YOE dollars, according to CARRD, and as a result it portrays a 25% overrun as a 100% overrun. But the interest is not the wrong numbers, but the invocation of Flyvbjerg again.
Nowhere does the article say anything about actual construction costs – it talks about overruns, but doesn’t compare base costs. It’s too bad; Flyvbjerg himself did a cost comparison for rapid transit, on the idea that the only way to reliably estimate costs ex ante is to look at similar projects’ ex post costs. His paper has some flaws – namely, the American projects he considers are older than the European projects, and there’s no systematic attempt at controlling for percentage of the line that’s underground, both resulting in underestimating the US-Europe cost difference – but the method is sound. Unfortunately, this paper is obscure, whereas his work on cost overruns is famous.
In the case of high-speed rail, it seems to me, from pure eyeballing, that there is a difference between countries in how much costs run over, and that this correlates strongly with high construction costs. German train projects, including the one example cited by the Economist, run over a lot. French and Spanish high-speed lines do not, and also cost much less.
Of course, this by itself doesn’t mean this correlation should keep holding: up until Barcelona Line 9, originally budgeted at €1.9 billion but now up to €6.5 billion, Spanish subway lines were built within budget. France has not yet had a factor-of-3 overrun on a major project, but it might in the future, and I’m not going to bet my life that it won’t. But what this does suggest is that looking at German overruns as if they’re typical rather than extremal cases is deeply misleading.
There’s an argument to be made that California’s inability to rein in the contractors will in fact lead to German cost overruns. California HSR’s projected costs look downright reasonable, whereas rapid transit projects in the state are unusually expensive. The proposed BART to San Jose tunnel is $4 billion for 8 km – very high by general subway standards, and unheard of for a subway in low-density suburbia. Going by Flyvbjerg’s own attempts to find ex ante cost estimates that are reliable, this could be used as evidence for future cost escalations; general overruns couldn’t, not without being more specific.
EMUs Versus Locomotives
I keep getting pushback from Amtrak defenders about my article about its locomotive order. I think I addressed most points, but one that I didn’t that keeps coming up is whether electric multiple units are really better for train service than locomotives hauling unpowered cars. The answer is in Amtrak’s case an unambiguous yes, but it requires more argument.
Ordinarily, the cost tradeoff between multiple units and locomotives is that unpowered cars are less expensive and lower-maintenance than EMUs while locomotives are much more expensive and higher-maintenance. EMUs have definite advantage in performance; they accelerate faster, and, when the consists are short their energy consumption is much lower, since most modern locomotives are optimized for longer freight trains. Because the advantage is the most pronounced for short consists, Amtrak asked Vermont to buy US Railcar’s FRA-compliant DMUs for the Vermonter train, replacing the current diesel loco-hauled setup; Vermont itself puts the breakeven point between DMUs and locos at 4-5 cars, but the DMUs in question have just one vendor and are extraordinarily expensive by global standards.
Conversely, locomotives require much more track maintenance than EMUs, because of their higher axle load. Road wear is proportional to the fourth power of axle load, so the less even the weight distribution is, the higher the road wear is. Track wear does not satisfy such a neat formula; all old comments of mine stating the contrary should be ignored. However, for freight traffic such a formula does hold, and locomotives have axle loads comparable to those of freight trains. One could also observe that in Japan, railroads make every effort to keep axle load low, and therefore avoid articulated bogies; furthermore, almost all Shinkansen axles are powered to keep weight distribution even, whereas European high-speed EMUs only power about half the axles (Siemens’ Velaro has a maximum axle load of 17 t, and an average load of 14 t).
Generally, the trend in countries with well-run passenger rail systems is away from locomotives and toward EMUs. The exceptions come from three cases:
1. Some technologies, most notably the Talgo tilting wheels, can’t be used with powered bogies. The same is true of the tilting TGV test train.
2. Some railroads ignore track maintenance costs and focus on train maintenance. This includes SNCF, since the tracks are the responsibility of RFF.
3. Cultural inertia may make railroads too used to separate power cars. This again includes SNCF, which needed power cars for the TGV because of the technological limitations of the 1970s and 80s, requiring very large transformers.
In the specific case of Amtrak and the Northeast Corridor, not only are reasons 1-2 not an issue, but also the cost question favors EMUs. Look again at Vermont’s report, which seriously posits unpowered coaches costing up to $5.5 million each, more than a standard off-the-shelf EuroSprinter loco; Amtrak’s recent order is much cheaper, at $2.2 million per car, but still comparable to the FRA-compliant M7 EMU and not much less per meter of car length (and more per car) than the Coradia Nordic EMUs used in Sweden or the FLIRTs used in Finland.
In comments elsewhere, I’ve heard that one reason to keep the locomotives is that they can be detached and replaced with diesels on through-trains to unelectrified territory. This is pure cultural inertia; EMUs, and even power cars that are permanently coupled to unpowered coaches, can be attached to a diesel locomotive, as the TGV did to reach Sables d’Olonne. More cynically, the cost of Amtrak’s locomotives is $466 million, which, at Northeast Corridor electrification cost (about $3 million/km), could electrify 155 km of route, almost all the way from Washington to Richmond. At the cost of electrifying the line to Sables d’Olonne (about $1.2 million/km), it could electrify nearly 400 km. Amtrak’s insistence on locomotives is reducing flexibility here rather than increasing it.
But in general, the move toward EMUs is not about flexibility; railroads around the world deprecate it and have semi-permanently coupled trains. It comes from the fact that, outside Amtrak’s uniquely bad experience with Metroliner EMUs, they work better. I’ve already mentioned higher acceleration. In addition, all else being equal, they’re more flexible, and can be scaled to any length: the M7s are married pairs. I’ve seen commenters that claim the exact opposite, by looking only at EMUs with articulated bogies; those have nothing to do with the question at hand (the TGV has articulated bogies, too), and indicate that the operator cares about other things more than about flexible length, for example a walk-through train or reducing the number of bogies.
Another problem with locomotives, besides inferior performance, is limited capacity. A single-deck 200-meter long AGV has 466-510 seats, compared with about 350 for a single-deck TGV and 545 for a double-deck TGV. SNCF is still eschewing the AGV because its capacity limit is so great it needs double-deck trains, but Alstom is developing a train with standard, unarticulated bogies that it claims can reach 600 seats with one deck.
Although Amtrak does not have the capacity problems of the LGV Sud-Est, it too is capacity-constrained, in another way. The limiting factor to Amtrak’s capacity is the lack of cars; as a result, buying EMUs instead of locomotives and coaches would add more capacity per dollar spent. It’s brutal, but true. Even the slightly more expensive Nordic EMUs would be an improvement; they’re still cheaper than coaches plus a single locomotive for all train lengths up to 14 cars (if the loco is an Amtrak Cities Sprinter) or 9 cars (if it’s a TRAXX or Prima).
In reality, the reason Amtrak uses locomotives is entirely cultural inertia. It was burned with the Metroliners, and thinks that unpowered cars last longer because, well, they have to. The reality that the M7, or the average European EMU, lasts 40 years, the same as Amtrak’s coaches; however, that idea was not invented by Amtrak, and is therefore out. It thinks that unpowered coaches are cheaper, while buying coaches that cost the same as EMUs. And so on. This is yet another bad US rail practice, hindering rail revival by making it too expensive and reducing performance.
California HSR: Where Now?
California is going ahead with construction of the Central Valley segment, and has just publicly released an email saying it will solicit bids in 3 months, totaling about $6 billion from Bakersfield to just south of Merced, a distance of about 200-210 kilometers. The alignment bypasses some small towns but not all, avoiding some of the scope creep that happened in the years leading up to the Business Plan, which required more elevated segments; however, some towns will still require many grade separations and viaducts, and so will Fresno and Bakersfield.
The HSR Authority has just released environmental impact reports for the Bakersfield-Fresno and Fresno-Merced that point out to higher costs: the sum of the two cheapest alternatives is $10 billion, in 2010 dollars, for 300 km; although the cost per km is not much higher, the Fresno-Bakersfield segment is much more expensive, whereas the extra bits included in the EIR but not the bid request are the cheapest.
There is some additional room for value engineering, especially in Fresno, where the currently preferred alternative calls for viaducts, but the potential for cost saving is not that great, especially relative to the $6 billion estimate; projects run over budget much more frequently than they come under. The main interest here is not the cost overrun: the current stage, the bidding, is the one most prone to overruns, and no matter what, we will know in three months what the projected cost is. The interest is the breakdown of costs, which, as expected, are primarily infrastructure and tracks, including grading and grade separations. The cost overruns come from scope creep, with more elevated segments than originally expected (but, due to value engineering, less than expected in 2009).
At any case, there is money to proceed, at least from Bakersfield to Fresno – there is $6.3 billion available, half from federal spending (which has been spared in the latest austerity plan) and half from Proposition 1A’s matching funds. There is another almost $6 billion locked in Prop 1A, but it has to be matched 50:50. Matching funds will almost certainly materialize, if not from the federal government then from foreign governments anxious to pay California to buy their products (for example, Japan’s ambassador to the US offered half the money, and Japan expects China and Korea to offer funding as well). It should be enough to build an initial operable segment, though probably not to build from Los Angeles to San Francisco.
The question is then how to prioritize. The gold standard here should be building all the way from Sylmar to San Jose and electrifying the legacy lines at the two ends. At the Bay Area end, the Caltrain FRA waiver ensures this wouldn’t cause regulatory problems, and while it would limit initial capacity, it would not increase travel time by more than a few minutes. At the Los Angeles Basin end, it would require Metrolink or HSR to seek a waiver, along the lines Caltrain has already gotten; the speed reduction, while still not very large, would be larger, because the travel time simulations assume higher operating speed in the LA Basin, and there will be fewer speed limits due to curves.
Unfortunately, while cutting the initial segment to San Jose-Sylmar will save a large number of billions of dollars in urban grade separations, it may not save enough, though it’s fairly close if one believes the 2009 Business Plan numbers. If California has half the money from foreign sources, then matched with Prop 1A and existing federal money, it has a total of $24 billion, which is not enough. The question then boils down to where to go first from the Central Valley – south or north. North would involve going over Pacheco Pass to San Jose (or, better yet, over Altamont Pass to Livermore and thence Redwood City). South would involve going south to Sylmar, either through Palmdale or directly through Tejon Pass, which carries I-5; although Palmdale is the preferred alternative, the HSR Authority is looking at Tejon again. For a slide show using the existing preferred options, see here. Either alone should be doable with the money available under such a circumstance, which is about $18 billion.
I claim that the southern option is the better one – in fact, that LA-Bakersfield is more important than Bakersfield-Fresno. The reason is, first, a pure numbers game: LA is much larger than anything else in California. And second, Tejon is where the existing legacy transit options are the worst: Amtrak can’t go between Palmdale and Bakersfield at all because the Tehachapi Loop is at capacity, ensuring that a mixed legacy-high speed operation in the mold of the initial TGV runs is not possible even under reformed FRA regulations.
Northern options suffer from different problems. The Pacheco option’s problem is that it uses Pacheco, and is therefore inadequate at linking the Bay Area to Sacramento. This means nothing further can be done until enough money materializes to connect to the Los Angeles Basin. The Altamont option’s problem is that the Phase 0 option connects to Livermore and requires a transfer; connecting to Redwood City is possible, but requires all of the most expensive elements of Altamont, especially crossing the Bay in the vicinity of the Dumbarton Bridge.
Once the southern option is selected, the question is how far to go. Bakersfield-Sylmar is expensive, and although it’s easily doable given 50% foreign funding, lower levels of funding may not suffice. Bakersfield-Palmdale is much easier, and could be done on existing Prop 1A money if it were not required a 50:50 match; however, Palmdale is not in the LA Basin, and the legacy rail line to LA is curvy and steep. Express Metrolink trains do Palmdale-LA in 1:28, versus 0:27 projected for HSR. Higher cant deficiency and acceleration with electrification could cut the travel time somewhat, but not enough to make HSR competitive for travel from LA to the Central Valley. Travel from LA to the Bay Area is another issue, but a situation in which it’s possible to build all the way to San Jose is one in which there’s money to build to Sylmar.
The alternative is to use Tejon and connect to the legacy line in Santa Clarita. It’s more expensive because Tejon is one big crossing whereas the Palmdale route involves two smaller crossings, one to Bakersfield and one to the LA Basin. It should still be affordable, though I have no detailed segment-by-segment breakdown of the Tejon route’s cost. The advantage is that Santa Clarita is much closer to Los Angeles than Palmdale, and the legacy Metrolink route to Palmdale is fairly straight south of Santa Clarita; even now, express trains travel to LA in 42 minutes, half an hour slower than full HSR buildout rather than an hour as with Palmdale, and there’s more potential for an increase in speed.
That said, the debate is most likely academic – Tejon vs. Palmdale is most likely going to be decided primarily on a revisited look at the costs, with other issues (LA County power brokers prefer Palmdale, Tejon is shorter) not much more than tiebreakers. In addition, a situation in which Prop 1A money could be released for the crossing is one in which matching funds have materialized, making the full Bakersfield-LA route realistic with the available money. The primary lesson is that there should be enough money to build a realistic initial operable segment, not going all the way from LA to San Francisco but still serving a fair number of intercity travelers.
Racism and Accidents
As has been widely reported in the news, China had a major rail accident three days ago, killing 43 people. A positive train control system that was supposed to prevent accidents didn’t; it was reportedly shut down due to severe weather, and as a result, when one train stalled on a bridge, another train rear-ended it and derailed, and two of its cars fell from the bridge. The Chinese government’s response was secretive and authoritarian, as can only be expected of a regime that treats breathing exercises as an act of subversion, and a leaked set of propaganda instructions to reporters contains such gems as “From now on, the Wenzhou train accident should be reported along the theme of ‘major love in the face of major disaster.'”
However, more interesting is the reaction of Western media to the disaster. Bloomberg quotes several financial analysts who raise doubts about China’s ability to export technology. A Financial Times blog analogizes high-speed rail to China’s fast-growing economy and warns of overheating. The general mood is one of treating accidents in China as evidence of a defective culture, which does not care about safety. More abstractly, it’s evidence that Asians don’t care about the individual, only about nationality and prestige. It comes from the same place as the San Francisco transit planner who, Richard Mlynarik reports, answered a question about Japan’s short turnaround times with, “Asians don’t value life the same way we do.”
The biggest HSR accident in history is still Eschede. The cause of the accident turned out to be a series of errors in maintenance and design. And yet, nobody doubts the safety record of Germany. They know that German industry turns out high-quality products. Siemens successfully distanced itself from the accident, claiming that it was only partially responsible to the manufacturing and that it was really DB’s train, and has sold its Velaro train in multiple foreign markets. An accident on its maglev test track that killed 23 hasn’t prevented it from marketing its maglev technology, and Germany’s continued rejection of maglev is on grounds of cost rather than safety. DB too was unfazed, made cosmetic changes, and was more recently hit with a less deadly egg on its face in Berlin; it too gets contracts abroad.
Eschede is emblematic of reactions to accidents in the West; Wenzhou is emblematic of reactions to accidents in Asia. (Amagasaki was as far as I can tell somewhere in the middle.) Individual incidents merely confirm what everyone knows.
The reality, buried at the bottom of few articles and unmentioned elsewhere, is that China’s overall safety record is not that bad. If one believes that Wikipedia’s list of accidents is exhaustive, then China’s record is very good. Even if not, on any reasonable estimate of Chinese HSR traffic (including traveling at lower speed, as the trains in question were), its safety is better than in many of the scoffing Western countries. Assume 150 billion passenger-km a year; this compares with an actual figure of 300 million HSR passengers per year as of 2010 and an average trip length of a little more than 500 km on all lines, not just high-speed (computed from data here). To beat the last twenty years’ American railway safety, China’s HSR division will need to have no additional fatal accidents for a year. To beat Germany, make that three years.
The sort of racism that would lead commentators and investors to think less of China’s safety over Wenzhou but not of Germany’s over Eschede is subtle; it’s nothing like overt discrimination in jobs or immigration or housing. As a result, it’s more or less self-solving in the long run: in the 1960s, Westerners thought Japan made shoddy products, in the 1990s they thought the same of South Korea, and in the last decade they’ve shifted the target to China. In twenty years, when China’s GDP starts approaching that of developed countries, they’ll find another target. They’ll of course not stop thinking that Asians are an undifferentiated mass of insects with no thought or creativity (or that Muslims are terrorists), but they’ll appreciate that they can make and even design manufactured products.
The significance is that it’s a telltale sign of the Not Invented Here syndrome. Convincing Americans to adopt European practices and vice versa is hard enough; but convincing them to adopt practices from Japan, let alone China, is anathema. You might as well try to convince an Orthodox Jew to switch from beef to pork. Attacking the assumption that other countries’ experiences are always part of a grand cultural essence is not just good humanity and antiracism; it’s also good technical planning.
In contrast to both the cultural approach and China’s apparatchik guidelines, I’d propose the following way to report accidents, terrorist attacks, and other major disasters:
1. Put individual events into broader statistical context. An aircraft or train crash should be accompanied by a reminder that those modes are still safer than all others.
2. Report on the causes of the accident, both immediate (as described in the first paragraph of this post) and fundamental, including any political or economic pressure to skimp on safety.
3. Avoid overinterpreting high-impact, low-probability events. Thus, avoid questions such as which train design standard is safer unless either directly relevant to the disaster (the wheel broke, the car crumpled, etc.) or backed up by extensive multi-year evidence.
4. If the official story or the source is not credible, pursue a separate investigation, using your own knowledge, or that of outside expert sources; pressure the institutions involved to be more candid about their own failures.
5. Follow up on the lessons learned, and whether they are helpful or not. As an example, consider the various measures taken to improve air safety since 9/11, and think which have been effective and which have not.
6. Avoid fluff at all costs.
For the most part, this list of items boils down to “Report on disasters involving non-Westerners as if they involved Westerners.” People are people, and societies are societies.
Pedestrian Observations 