I Gave a Talk About Canadian Construction Costs

There was a conference I got invited to, consisting of three talks, two about state capacity by me and by Tyler Cowen, and one by a Canadian extramural Conservative politician named Ginny Roth (she’s a columnist but her talk was about how Conservatives could use the insights of state capacity to win elections, hence my appellation). It was run by entrepreneurs named Chris and Matt Spoke, doing a series of online meetings trying to introduce fresh ideas to what they hope will be the next crop of Tory leaders; there’s going to be one on housing in the future, and the YIMBY comments I made seemed popular with the crowd.

Here is a link to my slides. They shouldn’t be too surprising given my usual talk on construction costs and what I said before about the growth in Canadian costs. But I made sure to put the increase in costs in Canada all together in two slides, one about Toronto, sourced to Stephen Wickens, and one about the rest of Canada, sourced to both our database and to a comparison of Calgary’s costs through the 2000s with Calgary’s West LRT costs.

The organizers are in Toronto, so I didn’t talk too much about the situation in Vancouver. I said a few sentences about how I can see there was a real increase in costs from a difference between the half-elevated Canada Line and the 87% underground Broadway subway under construction, but I didn’t go into the history of the Canada Line’s cut-and-cover method or the cost estimates from the early 2010s, which had the Broadway subway costing C$250 million/km. I talked more about Toronto, where the increase in costs is larger; Vancouver, even with the cost increases, remains North America’s lowest-construction-cost city, since the other cities have had even bigger increases, including Toronto, Los Angeles, and Seattle.

I want to highlight, as I brought up 1.5 years ago, that while Canada has American (i.e. bad) mainline rail, and Americanizing construction costs, it is YIMBYer than both the US and Europe. I worry it won’t last for long, because the style of Canadian redevelopment is at fairly small radius from an arterial or a subway station and those will eventually run out, forcing upzoning of large swaths of single-family land for the benefit of everyone except the handful of aggrieved homeowners who dominate municipal politics. (There was not enough time to talk about the importance of high-level decisionmaking, that is at the provincial level and not the municipal one.)


  1. Ks

    I listened. I am not a Tory.

    I actually liked that Cowen was there. He served as a good intro to you on state capacity varying by domain. I also liked your answers to his questions. Very “panellist”. Disagreeing without being disagreeable.

    I had a question. Cowen suggests that state capacity is immutable, so just focus on what you are best at. Do you agree? Do you know of a country becoming good at something it was previously bad at (eg operations)?

    • Alon Levy

      Yeah, the order of him, then me worked out well. I had no idea what he was going to say when I made my presentation, it was by coincidence.

      Re acquiring state capacity: it’s very much not immutable. Spain, for example, learned how to build high-speed rail in the 1980 by learning from France and Germany. Italy made genuine improvements in anti-corruption laws regarding infrastructure – in general, if a country is stereotyped negatively regarding some problem, then it’s often the case that nowadays it’s good at dealing with that specific problem thanks to the political agitation coming from opposition to it. So the US has pretty good anti-racism laws as a legacy of its apartheid era, and at this point my perception of American racism is that it’s middle of the pack by first-world standards; Italy has pretty good anti-corruption laws, and I know that overall Italy has more corruption than Northern Europe but in infrastructure specifically, which was ground zero for the tangentopoli and mani pulite, it’s way better than (say) Germany and on a par with Scandinavia.

      More in general, one sanity check on the “state capacity is immutable” line is that it predicts that economic development is impossible, and we know a bunch of counterexamples. For example, Finland today is tagged as a Nordic country, but its history until the 1920s was more intertwined with that of Russia, and it even had a Russian-style civil war, it’s just that the whites beat the reds, and afterward it made a choice to become Nordic and align its legal and political system with Swedish norms.

      Likewise, it’s all fine and dandy to speak of East Asian state capacity now, but until around the 1980s, everyone in the West treated Confucianism as a cultural drag on development. There’s a school of economic historians of modernization who look to the history of Japanese reopening and find that even before Commodore Perry arrived, Japan was the most developed non-Western country (for example, it was around 15% urban in 1850, which is high by pre-industrial standards). In contrast, no such examples of early development are found for either Korea or Taiwan. Korea was 3% urban in the early 20th century, and evidence of its development only appears after the war, aided by NGOs that treated it as a charity case and invested in literacy (as would happen a generation later in Bangladesh); and just as Finland chose to be Nordic, South Korea and Taiwan chose to be institutionally Japanese. The other half of Korea chose to be Soviet, with Soviet results.

      The issue is that it’s hard for Americans to choose to be something else, because they’re so enamored with their superpower status they don’t notice that the rest of the world bypasses them. Americans mock Europe over corona (e.g. because we lag on vaccination) while the US has a higher death rate, Americans are unaware of European biotech (the number of times I’ve seen Americans speak of the vaccine as if BioNTech doesn’t exist is huge), Americans make the silliest excuses for why they don’t need high-speed rail, etc.

      Thankfully, Canada doesn’t have to be this way. It wasn’t until 15 years ago, not in infrastructure. The Canadian elites just chose to be American, since the US is big and has high inequality (and thus, to the political and economic elites, the US is better than Canada). But they can stop this and choose to be something else. The domestic expertise from 20 years ago is still around – it takes longer than that for expertise to die out. There are plenty of European examples to learn from as well (though few British or French ones), and Canada doesn’t have the US’s superiority complex toward Europe. There are also some Asian examples, especially with integration of transportation and development – it’s just that in Vancouver, the political elite is white and anti-Asian, so it tries to keep as much of the region single-family zoned as it can.

      • Gok

        > More in general, one sanity check on the “state capacity is immutable” line is that it predicts that economic development is impossible

        Sorry, you’re saying economic development is impossible without increasing state capacity?

        • Luke

          I think that in a lot of cases, they have to move together or else they get in each others’ way. The alternative seems to be to me that the people who most benefit from economic development usurp what state capacity exists for their own interests lest that growing state capacity end up being utilized for other purposes, or those most interested in state capacity end up trapping a country behind the technological/bureaucratic frontier, as their desire to increase state capacity exceeds the ability of the current level of economic development’s to support that increase. Of course, this is assuming that the two parties don’t simply have opposed ideas of what “state capacity” and “economic development” mean.

      • Sarapen

        Uh, if state capacity is immutable doesn’t that mean the state is immutable? Which it clearly isn’t. It’s a strange belief to hold.

      • Brendan

        ” it’s just that in Vancouver, the political elite is white and anti-Asian, so it tries to keep as much of the region single-family zoned as it can.”

        That’s a statement with a lot of explanatory power in, say, 1929, but that argument seems somewhat strained in the Vancouver today

        • Alon Levy

          You know this better than I do, but my impression from seeing what goes on in Shaughnessy and how the province taxes immigrants is that the political elite remains white and Sinophobic and uses explicitly racist rhetoric in arguing for single-family zoning.

          • Brendan

            Every subcomponent of that statement is true individually I just don’t think thing the whole thing strings together in that particular way. After all, in a world where the villainous foreigners are stereotyped as rich, single family zoning is no protection from them and their villainous foreign ways, as opposed to 1928 when it was.

          • Alon Levy

            Does BC not have the California gambit in which condos are automatically held to be for the rich whereas single-family houses are for salt of the Earth working-class millionaires?

          • Brendan

            That gambit has traction in say, East Vancouver, where the stereotypical salt-of-the-Earth-working class millionaire is an aging immigrant, but you may recall that the villainous-foreigners panic that crested in 2016-2018 or so found a lot of it’s legs in Andy Yan’s study of West Side home buyers, for which the instrumental variable to determine foreign-ness was that people who had non-Anglicized Chinese names were presumed to be foreign and therefore concluded that 2/3 of the West Side houses sold were going to Mainland China buyers.

    • borners

      Investing in state capacity to turn Calgary-Edmonton into a dual-core urban corridor with a high-speed line is good policy if you can get the costs right, TOD them, and keep throwing immigrants at the cities Maximum Canada style. The problem is the oil-sands economy is terminal decline and lots of people have an economic and emotional stake in trying to revive it even at the cost of giving Alberta a future (and the planet). Add to that Prairie cultural conservatism is real, still powerful locally but now isolated from the Canadian mainstream so the Federals Liberals use gay marriage/abortion etc as a wedge issue against them. So Alberta’s historic conservative supermajority is in a really bad way and the staunchly conservative but institutionally PC who oversaw the urbanism that Alon has praised are gone. Reform Party low-tax-high-carbon-anti-state populism has taken over with Oil gimmicks and Trudeau hysteria pipeline/equalization brinkmanship taking the place of anything useful. Heck they can’t even squeeze out upgrading oil sand freight rail line properly which would give them something useful for the future (unlike the pipelines).

      Its interesting that the conservatives involved are Ontario rather than Prairie types. I’ve often found it interesting that Preston Manning hasn’t taken up YIMBYism as a cause given how much Canada’s progressives are responsible for the housing crisis (looking at you Ontario Liberals). Must go to show that conservatives are gonna abandon free-market magic the moment it moves a direction they don’t like.

      • Brendan

        RE Ontario vs Prairie Conservatives and YIMBYism – that’s because there isn’t anything like the housing issue in Calgary & Edmonton that there is in Toronto or Vancouver. Urban Alberta home prices peaked in 2007 or 2014, and have been mostly on the way down since 2014-15. The politics are not so averse to greenfield development either, so there’s nowhere near the salience

      • df1982

        In favour of Calgary-Edmonton is the fact that the distance is pretty ideal for intercity rail (300km) and the per km cost of any line should be among be the lowest in the world: there are existing rail alignments for city centre access, and between the cities is perfectly flat, virtually uninhabited farmland (apart from Red Deer, which would get an intermediate station, there are only a handful of tiny, easily avoided settlements). There is also an existing road alignment (Highway 2) as well as the legacy rail line that could be used as corridors if necessary. Added to that is the fact that the airports for both cities are on the near side, and are close to potential rail alignments, so airport rail links for both cities could be incorporated into the whole project. You could have a system with, say, half-hourly trains between Calgary and Edmonton, interleaved by half-hourly trains from each city-centre to the city’s airport.

        So despite the low aggregate population the conditions are nearly perfect for high-speed rail (or even just higher-speed rail) to work there, as long as construction costs are well contained.

        • Eric2

          Alon’s model has Houston-Dallas (385km) as barely profitable, and each of those cities is about 7x higher population than Calgary/Edmonton (so 49x more ridership by a pure gravity model, or 22x if we use a power of 0.8). Which suggest to me that frequent buses, or maybe tilting trains on the existing (very wiggly) track, would be a better investment.

          • borners

            I did say Saunders style 100 million population Canada, so quite a few improbable policy choices. Putting aside implementation, the real problem is that it would be a closed system unless someone figures out cheap maglev to get to Vancouver. Network effects make heavy rail pay off.

            There are middling solutions with adding straighter bits for passenger rail, electrification etc, it takes 3-4 hours by car right now, assuming a high performing conventional regional train at 160km could definitely beat that by enough to steal a lot of mode share.

            But its fun to think about since Alberta is concentrated in two sploges on a linear alignment with huge areas of flat land with nobody to complain about noise. Greenfield crayon fantasy!

            That said we know that Alon’s model underpredicts Japanese megacity exburbs use of the Shinkansen and that model doesn’t consider high quality conventional JR rail and legacy private rail competitors and no through running through Tokyo station. If you don’t have a built out urban heavy rail network, why not try and integrate normal urban and high-speed inter-city rail by having them use high-speed alignments in the empty middle section and then go slower inside the city, especially on the far side of the CBD. Could you do branching by having the train formations split Tohoku shinkansen style in the CBD?

          • df1982

            Why treat that model as gospel? Alon says its level of sophistication is extremely low (viz. the barbarian jokes) and it has a very mixed record when compared to real-world examples. In particular it overstates high-population long-distance journeys over smaller-population shorter-distance journeys. I think it also over-rates the network effect over isolated city-pairs, hence why lines like Atlanta-Orlando or Buffalo-Cleveland are rated over the PNW. Which is not to say the network effect doesn’t exist, but in Alon’s model it produces some obviously counter-intuitive priorities.

            In the real world, Houston-Dallas is profitable enough that a private company is willing to sink something like $15b into a brand new line, which is not happening to the same degree anywhere else in the world (Florida and LA-LV will be leveraging existing or state-funded infrastructure).

            Of course Alberta is dealing with a much smaller population catchment, but the key to making it work would be cost-control of construction. Luckily it has ideal natural conditions for achieving this goal, which is what makes it more viable to my mind (Alon’s model does not take relative construction costs into account when calculating profitability, as far as I’m aware). You could even use a lot of the alignment of the existing railroad with tilt trains if you want, with a few bypasses to straighten out the few circuitous sections.

            The model could be something like Sweden’s Stockholm-Gothenburg line. Since that has a smaller population base than Calgary-Edmonton and more distance to cover (and also very little in between), there’s no reason why intercity rail in Alberta wouldn’t be as successful.

          • Alon Levy

            So, one specific problem with scaling down to Calgary-Edmonton is that you’re starting to get frequency artifacts. Texas Central plans to run 1 short tph off-peak, 2 peak. At that distance, that’s already a frequency artifact, and if there’s a way to run 2 tph all day, it’ll meaningfully increase ridership.

            In the other direction, to scale down to Calgary-Edmonton, you need to either run way less than 1 tph, or run smol trains. Smol trains cost more to operate, for example because the driver’s salary is split across fewer passengers, the required conductor-passenger ratio is a bit higher, and even energy consumption is a bit larger because frontal air resistance is the same, and it’s only the side winds that depend on train length. Less frequent trains repel ridership.

            The upshot is that the model’s assumption of a constant profit rate per p-km assumes certain economies of scale that do not exist in Alberta. (It also assumes flat demand all day – if you’re peaky, operating costs rise; I will fully justify this assumption for the Northeast Corridor because of its combination of different kinds of trips with different peaks, but two cities with nothing in between them may not be like that.)

          • Eric2

            We can think of good reasons why Alon’s model is imprecise. But even if you put all those reasons together, it’s hard to imagine them making up for a ridership which is 22 (!) times lower.

            I think immigration to Canada is a good thing. But like in the US, in Canada I think we should attempt to concentrate future population growth in big cities like Toronto, where it will be better both economically and environmentally, rather than in small cities like Calgary/Edmonton.

          • Alon Levy

            Calgary isn’t growing out of push factors. It’s an insanely wealthy city, especially before oil prices fell in 2014-5.

          • borners

            The future of Alberta’s economy is still iffy given ongoing the renewable revolution. It didn’t completely waste the booms, its not Alaska/Scotland they have good universities etc the old comparison with Texas has positive sides. And are more affordable than Toronto/Vancouver. @Eric2 You guys have the space for Alberta to aim for bigger especially since its geography is easier for expansion than the big 3. And given the angst over the oil economy and the dominance of Nimby politics in ON/BC maybe there is an opening.

          • Henry Miller

            Why would you run a new HSR that isn’t fully automated? A little more up front, but you can afford to run smol (I think this means short?) trains all day. If needed just give anyone with a current first-aid/CPR certification a slight discount if they are willing to keep watch over the train while they are on it anyway. You still pay more for wind resistance (a big cost), but the frequency should allow you to make up at least some of that.

            I understand if you have an existing line upgrading everything might not be worth it, but when you are building new anyway the costs are not that much more.

          • Phake Nick

            @Henry Miller I am not aware of any fully automated HSR in existence in the world
            @Alon Levy Can you quantify how long a regular/short/smol trains would be? 8-car trains, 4-car trains, and a 1-car railbus?

          • Phake Nick

            @Eric2 From my memory, the problem with existing track is that years ago they determined the cost of improvement is too much to get to the speed that would make it competitive against buses on highway, and they also have opposition from freight train company, so they determined it’s more economical to build new track

            Speak of which, how Nordic countries manage to serve semi-high speed trains across sparsely populated area?

          • Alon Levy

            Higher legacy construction standards than are typical in the American West. Same way British main lines from the 1830s average 140 km/h nowadays. American railroads in the 19th century were built to lower standards because the cost of capital was higher due to high population growth – the mentality was that it was better to build fast, cheap, and bad, and then rebuild later as population grew. The Northeast Corridor in New Jersey is in large part a cutoff from the 1860s.

            P.S. Sometimes you need to hit refresh to see your comment, WordPress sometimes doesn’t show new comments from the last few minutes.

          • adirondacker12800

            Nobody knew how to build railroads when they built the Camden and Amboy because there weren’t many. Your design specifications are going to be bit different if this is the kind of train you are planning on running.

            By the time Alberta had population the Model A had replaced the Model T and had been discontinued for years.

          • Alon Levy

            Sure, but contemporary British railroads were built double-track from the start, with hilariously shallow grades (the Liverpool and Manchester was 0.03%, with a segment of 1% pulled by cables) and wide curves. Andrew Odlyzko’s monograph on the Railway Mania has citations for some mid-1840s railway boosters promising that they can lower construction costs by building lines single-track, relying on the invention of the telegraph to timetable trains.

            The Eastern US had some railroads built British-style, like the Boston and Providence. The Boston and Lowell was single-track initially – I think it only had one trainset operating at a time – but had wide curves. But then there’s the pain that is Connecticut, or anything inland in Massachusetts, where those early railways were never replaced with cutoffs.

          • Henry Miller

            @Phake Nick

            Perhaps none exist yet. However all the technology is there. Lower speed fully automated rail exists. It should be but a small cost to bring it to HSR at this point. There isn’t a lot that needs to change to do this, it won’t be free, but it should be cheap enough to be worth doing. Assuming we reuse existing technology of course. If we decide to start from scratch we can easily blow the costs out to the point where it isn’t worth doing.

          • df1982

            @Alon: doing some maths here. A 200m ICE3 has a capacity of about 400. At 50% average loading, running half-hourly 16 hours a day you get about 4.7m riders. The 2014 Alberta HSR study predicted ridership of 5m+ with population growth, so it should be able to sustain a half-hourly service, if not from day one then at least within a few years of beginning operations.

            And is it really that much cheaper per passenger to run a 16-car train rather than an 8-car train? The only constant cost would be the driver. Attendants would be based on the number of seats, while fuel, maintenance and rolling stock depreciation would all be based on the number of cars (I’m ignoring the dining car because it would probably be unnecessary on 90-min runs).

          • Alon Levy

            Yes, I think 5m+ even with population growth is insanely high. Berlin-Hamburg for example is 6 million, but that includes other city pairs like Leipzig-Hamburg and Berlin-Kiel, and if you try to net these out then my modeled prediction of 3 million looks about right. The real issue with density is that in a dense place or even a medium-density one (=France), there are always a lot more city pairs than the primary one people think of.

            16 vs. 8 isn’t a big difference, but on realistic Alberta ridership, which is not 5m – try 1m with a lot of growth (600,000 at current pop) – it can’t even fill 8-car trains at any usable frequency.

          • Sassy

            @Henry Miller

            HSR and metros are different though. If a passenger related issue arises on an automated metro, staff can be available at the next station within a couple minutes at most. With HSR, the next station could be an hour or more away. Therefore, you’d have to have staff on trains in driverless HSR.

            Beijing–Zhangjiakou is driverless, but still has staff. The Chuo Shinkansen can’t even be manually driven, and is still planned to have staff when it enters commercial service.

            @Phake Nick

            Since Alon hasn’t replied, here’s my logic on what regular/short/smol translate to.

            Texas Central plans to use 8-car trains, and that was described as “short.”

            This means that the “regular” probably refers to the standard 16-car N700S configuration. This strikes me as a bit long for “regular,” since HSR trains don’t get much longer, but if an 8-car train that seats 546 is considered short, then “regular” has to a big boy.

            “smol” is hardest to pin down. I’d say anything comparable in seating capacity to a 737 or A320 plane could be considered “smol,” since you could probably justify running trains of that size between Edmonton and Calgary at 1tph based on pre-existing travel between the two cities. There’s about one regional jet per hour between the two cities, and probably plenty of people driving, and you can more easily afford not full trains.

          • adirondacker12800

            The real issue with density is that in a dense place or even a medium-density one (=France), there are always a lot more city pairs than the primary one people think of.

            Utica/Pueblo. People in Pueblo can go to Colorado Springs, Denver, Boulder, Fort Collins and a dozen of them a week to Wyoming. Uticans will be able to get to New York. And Boston. Philadelphia. Probably Baltimore and Washington DC. Three hours Buffalo to New York implies an hour and half from Utica. Which makes Washington DC three hours. You can’t get to Ohio from the Rochester or Syracuse airports. Ohio. Toronto.
            From Edmonton or Calgary you can get to Red Deer. There aren’t going to be high speed trains to Vancouver or Winnipeg, they are too far away.
            … Buffalo, Rochester, Syracuse, Utica and Albany have more people than all of Alberta….. Population of Wyoming is Union County New Jersey give or take a few tens of thousand.

          • df1982

            @Alon: at 1m riders per annum, you’re suggesting that the average resident of Calgary/Edmonton would make a return trip between the two cities approximately once every 5-6 years. That obviously seems way off. Even one round trip a year per resident would result in 5-6m trips per year. In fact, it’s likely that you would get a sizeable number of weekly commuters between the two cities, which would turbo-charge ridership numbers.

          • Alon Levy

            If you have two cities of 1 million people and the average resident makes one trip per year, then at 20 million the average resident makes 20^0.6 = 6 trips. The gravitational pull is proportional to population (to the 0.8th power); the model predicts 20^(2*0.8) = 121 times the ridership, yes, but there are 20 times as many people, so the average person takes 6 times as many trips.

            It’s hard to exactly sanity-check this, to be fair, because nobody builds HSR between two cities the size class of Calgary and Edmonton with nothing else in the vicinity. But down to the size class of the cities along the Berlin-Munich HSR line it’s valid.

          • df1982

            But why would the average person in the 20m city make 6 times the number of trips as the person in the 1m city? It’s not like we’re talking about rural hicks here. Ca. 1m population cities include places like Zürich, Oslo, Edinburgh, etc., which are prosperous places with highly mobile citizens, as are both Calgary and Edmonton (the tyranny of distance aside).

            What’s the ridership of Stockholm-Gothenburg? That would be a good sanity check. It’s at least enough to have 1-2 tph without people decrying it as a white elephant (quite the opposite, the plans are to improve speeds on the line). Calgary-Edmonton would probably massively out-perform it because it would be a 90min as opposed to 3hr ride.

          • Henry Miller

            Don’t forget the build it and they will come. Today it is hard to get between cities (time consuming), over a few years people will figure out good service exists and use it. This will at most bring up the ridership to the model though as the model is trained on cities that have had a connection long enough to have accounted for those factors.

          • Phake Nick

            Regarding city size and pull of population, I just saw a post on a Japanese forum from 20 years ago on the traffic between the three Northern Kanto prefectures/cities. https://uub.jp/arc/arc443.html
            The three neighbouring prefectures, with Ibaraki’s capital city at Mito, Tochigi’s capital Utsunomiya, and Gunma’s capital Maebashi with the neighboring Takasaki where Shinkansen passes through, all have ~300k-500k population inside the city and each lead their prefecture with population in the number of million, but they have very limited connection and traffic between them
            After the opening of expressway in the area about 10 years ago, bus companies have tried to run express bus services between these cities pair, and those buses are faster than railway so they should be able to attract major amount of transit users in-between (Although Northern Kanto is a automobile-centered area and could thus affect bus ridership), but still the Mito-Takasaki/Maebashi terminated in year 2011, Maebashi/Takasaki-Utsunomiya terminated in year 2014, and Mito-Utsunomiya was only maintaining 6 bus a day until the start of coronavirus pandemic and the service is now suspended due to the pandemic.
            From the linked discussion thread, each of these cities have their own needed necessities and facilities, and when residents there find there are something not sufficient and would want to travel to other cities for more options, they usually visit Tokyo which is about 1 hours by fast train or Shinkansen, or 2 hours by local trains, instead of spending the 2 hours to visit another not-quite-big city similar to the city they themselves are in, and thus the traffic and link between them have been quite limited.
            I guess it can also explain why, in the previous checks on Shinkansen traffic, cities pairs like Hiroshima – Kumamoto massively underperformed the formula estimation, since there aren’t much reasons for travellers to travel from one small city in one area to another small city in another area while there are more attractive alternative with either same or shorter travel time. The same can also be observed from world air travel network where even with the advent of LCC opening up more point-to-point routes, most of these points to points routes are still linking different sort of hub cities on one end, very few routes are really connecting one smaller cities to another smaller cities directly, especially when those smaller city pairs doesn’t have some particular linkage or when they don’t have some specific attraction like beaches or such for people to visit

      • df1982

        The fact that Alon’s model produces a differential factor of 22 between Dallas-Houston and Calgary-Edmonton shows what’s wrong with this model. The aggregate population of the two city-pairs is 15m vs 3m, so a factor of five. They are a similar distance from each other, and they are both pretty isolated from any network effects since the amount of through running on Dallas-Houston would be nugatory. And in both cases the two cities are roughly equal in size to each other, so there is no real hierarchy between them.

        Simply put, given the same transport conditions (in this case, a 300km HSR ride taking roughly 90min), a resident of Dallas/Houston is not more than four times more likely per capita to travel than a resident of Calgary/Edmonton. In fact Clagary/Edmonton residents would probably be more likely because average access/egress times to the station would be better.

        That’s a key problem with the model.

        • Eric2

          “The aggregate population of the two city-pairs is 15m vs 3m, so a factor of five.”

          It’s not based on the aggregate (sum) population though. It’s based on the product of the two populations.

          As an example: If you have two cities of 1M, they will have a steady stream of travel between them. If you have one city of 1,999,950 people and a village of 50 people, they will have very little travel between them. In each case the sum of populations is the same, but the product is very different.

          This makes a lot of physical sense. Let’s say you’re in city A (population 1M) and considering visiting cities B (population 2M) and C (population 1M). City B has twice as many houses and twice as many offices, compared to C. So you are twice as likely to have family and friends to visit there, and twice as likely to need to visit for business. So travel volume will be twice as high. This corresponds to the product of populations: A*B is 2 trillion, while A*C is 1 trillion. It does not correspond to the sum of populations: A+B is 3M, while A+C is 2M – only a 50% difference.

          (There is one extra factor that HSR is less attractive in big cities because it’s hard to get to the station, therefore Alon sets the exponent to be 0.8 not 1, so the ratio of Houston/Dallas to Calgary/Edmonton is (7^0.8)*(7^0.8)=22 not 7*9=49)

          • Eric2

            (By the way, this is the reason cities exist in the first place. A city that’s 7 times larger will have 49 times as many possible connections between its residents, so it is much easier to find someone with a particular skill set for your business, and businesses are more successful. This more than offsets the difficulties with cities, like getting food/water in and waste out.)

          • df1982

            Using the product applies well to city-pairs within an integrated network, hence why it is useful for something like SNCF, because someone in Marseille is obviously much more likely to travel to Paris than, say, Bordeaux.

            But it doesn’t work well for comparing two independent sets of city-pairs, particularly when the cities in both pairs are both roughly equally sized, which is the case here. There is simply no reason to think that someone in Dallas has >4 times the likelihood to travel to Houston than someone in Calgary has to travel to Edmonton. Rather, the sum of the populations is a better predictor here, which is probably why it’s hard to produce an all-purpose model predicting HSR ridership, and also why you shouldn’t just point to such an all-purpose model to argue why HSR shouldn’t be built somewhere..

          • Eric2

            “There is simply no reason to think that someone in Dallas has >4 times the likelihood”

            Why not? I gave a reason – what do you think is wrong with it?

          • df1982

            Because if you extrapolate it mathematically you get ridiculous results. If you have two cities of 1m people and the average resident makes one trip a year between, then if you have two comparable cities of 20m people, you’re suggesting the average person would make 400 trips a year between them. That is patently absurd.

          • df1982

            Correction: I got carried away: 20 trips per year per resident. But the logic is still there.

          • Eric2

            Is that unrealistic? Ask residents of Binghamton, I would bet they travel to the NY area 20 times as often as they travel to Buffalo. (The population ratios of NY and Buffalo areas are about 20:1, and the distance similar)

          • df1982

            Yep, and as I said that calculation model works well in integrated regions where the largest city obviously exerts the biggest gravitational pull. It doesn’t work so well when you’re comparing isolated city pairs. If you’re in Edmonton there’s no other nearby cities exerting that kind of pull apart from Calgary. For everywhere else you have to get on a plane.

          • adirondacker12800

            Pshaw. there’s a train and you could always drive.

            There might even be bus.

          • Eric2

            Can you give any concrete reason why isolated city pairs would be different? At first glance the logic of more people=more family=more business would apply equally there.

            In the particular case of Edmonton/Calgary, maybe there is a factor that both cities have a lot of fossil fuel industry so there is a disproportionate amount of business travel between the two, compared to random cities of the same size and distance. But Dallas/Houston has a bit of this factor as well.

          • df1982

            It obviously would scale with population, but not exponentially so (not even with a 0.8 factor). If you have an isolated city-pair, then one city would tend to monopolise the possible journeys made by the other. If you have a brother who moves from Edmonton, but wants to stay nearby, he would overwhelmingly choose Calgary. If you have a company doing business in Alberta, it would likely have branches in both cities, and that’s it. If you’re itching for a weekend away but without the hassle and expense of flying, Calgary is your only destination. So the random person in Edmonton is going to have just as many reasons to travel to Calgary as the random person in Dallas does to Houston.

            This is different to small cities within the orbit of a bigger city. Say you have cities A: 20m, B: 2m and C: 1m. A-B would then probably have double the travel market than A-C (let alone B-C), so here the product rather than the sum does apply. I know rom experience when I lived in New Haven I would go to New York every couple of weeks, Boston maybe a few times a year, and Hartford/Providence precisely never (of course the MetroNorth tipped the scales here).

            This is why Alon’s model works well when comparing city-pairs within a single network, but not when comparing self-contained lines that exist in isolation from each other.

          • Alon Levy

            Okay, but the model trained on Japanese data does okay in Korea, Germany, and Spain. In France I’m less sure because the system is so open and there’s such a long tail of small towns with TGV through-running, and the one place where I have city pair data, Paris-Nice, overperforms massively presumably because of Nice’s unusually strong tourism industry. I can even sort of predict NY-New Haven and Boston-Providence rail ridership with this model – both city pairs overperform somewhat (this is really Boston-RI and NY-NH County); of course at such short distance lots and lots of people drive off-peak, but the model accounts for that with the 500 km minimum distance.

          • df1982

            But what Japan, Korea, Germany, Spain and France all have in common is that they all have integrated networks where the bulk of the country is within HSR distance of the dominant city (although in Germany this is a bit different due to its polycentric nature), and that’s precisely where your model works well.

            The broader NE network is similar to this, as would be a Mid-West network with a Chicago hub and even a greater California system. But what makes North America different is it also has other potential lines that are isolated from any conceivable broader HSR network: Texas, Florida, PNW, Alberta. And this is where the model falls down in underestimating their potential.

            Put it like this: your model would correctly judge that Hartford-Providence does not get much ridership since Hartford gravitates towards NY and Providence to Boston (and also NY to a lesser extent). But if you put the exact same cities at the exact same distance on a remote island cut off from any nearby major cities, ridership between them would conceivably be a lot higher, because there’s nowhere else to go.

            That’s why I would suggest looking at ridership for something like Stockholm-Gothenburg, which is not quite as self-contained as Calgary-Edmonton (since there is Malmö, Copenhagen, etc.), but is broadly similar.

            I also tend to think the 500km floor is not justified, as a city pair at 300km distance probably does a lot better than the same one at 500km, but that is a different issue.

          • Alon Levy

            FWIW, my ridership estimate for Texas Central and the company’s own are very close.

            The “nowhere else to go” factor assumes that people take intercity trips at a constant rate, net of long-distance flying. Do they? I know that there are models that assume this, but the one application I saw of such a model was working with commute data (and even that data has certain GIGO problems).

          • Eric2

            “If you have a brother who moves from Edmonton, but wants to stay nearby, he would overwhelmingly choose Calgary. If you have a company doing business in Alberta, it would likely have branches in both cities, and that’s it. If you’re itching for a weekend away but without the hassle and expense of flying, Calgary is your only destination.”

            I think there is some truth to this. But at the same time, perhaps the brother doesn’t have to move at all, and similarly for the business. Maybe they will move to Calgary IF they move, but because there are no other good candidates for moving, maybe they’ll just stay in Edmonton in higher proportions.

            “So the random person in Edmonton is going to have just as many reasons to travel to Calgary as the random person in Dallas does to Houston.”

            Let’s assume this is correct – it’s still a 7x ridership difference between the city pairs! (Maybe a 7^0.8=4.7x difference if you want to account for travel time like we did before.) If Dallas-Houston is borderline profitable, a 4.7x factor is still enough to sink Edmonton-Calgary. Even if you factor in Edmonton-Calgary’s shorter distance without putting a 500km limit, it’s a factor of 2.9x.

          • df1982

            @Alon: There undoubtedly is some induced demand from the network effect, I just don’t think it’s exponential the way your model implies. And if your model suggests that with an HSR line the average resident of Calgary-Edmonton would make one trip every six years on the line, then that obviously doesn’t pass a sniff test (that might be true if they were both within HSR distance of a New York-sized city, but they’re not).

            @Eric2: Remember Calgary and Edmonton metro areas are both around 1.5m, not 1m as you seem to assume. Regardless, it would undoubtedly have a lot less ridership than Dallas-Houston, but it should also be a lot cheaper to build and somewhat cheaper to operate. And I don’t see why we should see Dallas-Houston as being only borderline profitable, when the private backers of the line selected it as the most potentially profitable route anywhere in North America.

          • Alon Levy

            I’m not assuming it’s exponential, I’m assuming it’s superlinear with exponent 1.6.

            And why is one roundtrip per 6 years obviously false? Albertans can still fly elsewhere.

          • df1982

            Because, a) the two cities are far more integrated with each other than they are with anywhere else. In addition to the distance factor, they are both in the same state (and this state has a strong regional identity), and they are both centred on the same industry.
            And b) it is evidently far more convenient to catch a sub-90min train than to get on a plane, which would have door-to-door times of a minimum of 3 hours (Vancouver) or even 5 hours (Toronto). Your entire work on HSR is predicated on the superior convenience and hence potential ridership of high-speed trains in comparison to planes, so I’m surprised you even ask this question.

  2. SB

    “After comparing the Sheppard Subway unfavorably with
    Madrid Metro expansion, the TTC decided to adopt
    design-build, which in fact Spain does not use.”

    So Toronto looked at construction costs in city with low construction costs (which is good) and decided to not to do what low construction city was doing (which is bad)?

      • Henry Miller

        I know you are not a fan of design-build, but is it actually bad? (you have published on why public-private-partnerships are bad, but not that I’ve seen on design-build). As I recall you had reservations, but it isn’t clear if design-build is a cause of cost blowup, or just something that is happening at the same time as whatever the problem is.

        That is design-build might be a good thing, but the advantage might not be enough to overcome all the other problems and so we can’t tell.

        • Alon Levy


          Americans keep telling me how it is better, but when I probe, it turns out that they just really suck at design-bid-build. The in-house planning and design review teams are too small, politicians micromanage, the risk allocation is adversarial. Because Americans are unlikely to hear of European or Asian projects unless they use American or British consultants, in which case they suffer from the same Anglo groupthink, they earnestly think that Europe is transitioning toward design-build, which it isn’t except for one overrated country with big recent cost overruns.

          It’s telling that among the low-cost countries we’ve looked at, none uses design-build. Turkey comes closest; I should blog it soon – Elif describes it as design-build and generally if Elif says something it’s correct, but it’s really de-bid-sign-build, i.e. early design is separate from final design + construction, with more in-house engineering than one sees in the US or Canada.

          • Tom M

            Noob question/clarification: design-bid-build = conceptual/pre-liminary/final design by in-house, then bid + build by private sector/out-sourced? Does the US typically only do the conceptual stuff in house e.g. choose the route, decide they need monument level stations?

          • Alon Levy

            No, that’s build. Design-bid-build = contract out the design, then bid the design to separate contractor to build.

            The US typically does a lot of weird petty things in-house. There’s a lot of micromanagement; Stadler complained that this is why Caltrain’s KISS order had such a big cost premium. It’s just never any kind of organized design and engineering bid out to a private builder who can then make modification. The interference is haphazard and adversarial.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.