Eric and I are in the process of building up our database of construction costs and starting to select case studies for in-depth study. Most of the world was already in my original database from late 2019, but there are big gaps, most notably China, which has built more subways in the last 20 years than in my entire database combined. For this, we work with students; I mentioned Min-Jae Park in a previous post, but we have others. A Chinese master’s student of public administration at NYU named Yinan Yao is working with us on this, and has used Chinese sources, mainly official (what I call “plan” in my dataset), to construct a dataset that so far has 5,700 km, of which around two-thirds is underground.
I’m not putting the database out yet – this is still preliminary and subject to some edits, and we’ll publish a merged database of everything when it’s done (probably in the summer of this year, but don’t yell at me if it takes longer). However, I want to point out some observations that come from the data:
Chinese costs are fairly consistent: most recent subways cluster somewhat below 1 billion yuan per kilometer, or around $250 million per kilometer in PPP terms. This is consistent across the entire PRC. Costs are slightly higher in Beijing and Shenzhen than in the rest of China, and are even higher in Shanghai, where they approach 1.5 billion yuan per km. This is in accordance with what I’ve found in the rest of the world: costs are remarkably consistent within countries, especially within cities, to the point that variations, like New York’s higher-than-US-average costs or the difference between Milan and Rome, require separate explanation.
More difficult lines cost more: this is again not surprising, but it’s useful to check this on the largest national database of costs. Yinan points out that certain lines that cost more are more central, in that sense of passing under older lines with many transfer stations. See for example the Shanghai plan for 2018-23, with a map, a list of lines, and their costs (in hundreds of millions of yuan, not billions) on the last page: the highest cost per kilometer is actually a short elevated extension of Line 1, which has to be done while keeping the line’s current Xinzhuang terminal open for service as it is a critical transfer point to Line 5. The same map also shows the cost difference between the more central Lines 19 and 20 and the more suburban Airport Line, which goes around city center as the center is already connected to Pudong Airport via Line 2.
Why is Shanghai more expensive? Shanghai has a more built-out metro system than any other city in China save Beijing. That could explain its cost premium, but then again, relatively suburban lines like the Airport Line have similar costs to rest-of-China lines, including city center tunneling. Yinan suggests that the reason is geological: Shanghai is in the alluvial plain at the mouth of the Yangtze. This theory would suggest that tunneling in other parts of the world at the mouths of big rivers is expensive as well – and this is in fact true in Europe, as construction costs in the Netherlands are high. It is worth investigating, not just because of the implications for China but also for the implications for Europe: if Dutch costs are high for geological reasons, then there is nothing to explain regarding the quality of Dutch institutions, and thus if certain institutions (such as consensus democracy) occur in low-cost countries like Switzerland and the Nordic countries but also in the Netherlands, then the retort “but the Netherlands has this too and is expensive” loses impact.
There is very little regional rail in China. The definition of regional rail in a Chinese context is dicey – China did not inherit big legacy commuter rail networks, unlike India or most developed countries. Suburban rail lines are greenfield metros, rather like the Tsukuba Express or some of the more speculative parts of Grand Paris Express. In our dataset, regional rail is broken out from other urban rail because the concept of regional rail means only tunneling the hardest parts, and doing the rest on the surface using legacy railroads, which cuts overall costs but raises the costs per km of tunneling. China doesn’t do this, so all lines have the tunnel composition of a metro.
Having a lot of quantitative data makes things easier. Chinese costs are in the context of a consistent set of national institutions, and involve a lot of different subway lines. Even income differences are not so huge as to render analysis impossible – there is a lot of geographic inequality in China, but less than between (say) China and the developed world, and for the most part the bigger cities are on the richer side. This makes it easier to formulate hypotheses, for example regarding what exactly it means for a line to be more or less central. Eric, Yinan, and I are trying to come up with a coherent definition, which we can then try to test on other countries that build a lot of subways, like France, Russia, India, South Korea, and Spain.
All data is valuable. I started looking at costs in 2009-10 in order to figure out how to affordably build more subways in New York, and thus focused on the largest and richest world cities, like London and Paris. But really, all data is valuable. Comparing various developing countries is important because of issues like cultural cringe, and likewise figuring out if Shanghai is more expensive due to geology is important because of the implications regarding Dutch institutions. It is ignorant and harmful when New Yorkers reject knowledge that comes from outside their comfort zone of the city and perhaps the few rich global cities it deigns to compare itself with. On the contrary, Chinese data should be of immense value to both richer countries like the US and poorer ones like India, and likewise data from the rest of the world (for example, some Japanese and Korean best practices) should be of immense value in China.
There is a lot of knowledge out there. The point of comparative research is to access knowledge that people in one reference group (in our case, New York) do not have. Eric and I don’t speak Chinese; our language coverage, plus some non-English Google searches, is pretty useful, but far from panglossian. Yinan is so far tremendously helpful to this project. (The other students are helpful too in what they cover – they’ll get posts too, just this one focuses on China.)