Key Finding Number One: Modest Tendency For Rising Real Costs Over Time

As noted above, the key dependent variable in our study is the cost of building a kilometre of subway line in 2021 USD.  We can observe considerable discrepancies in unit construction costs in different decades. Decades are, of course, arbitrary units, so we need to look for specific years that seem to be turning points, so we can do a rough-and-ready breakpoint analysis. When we do so,  the First World War appearing to be a key turning point in which construction costs across the western world began to creep upwards. As can be observed in Figure 1, these costs have tended to rise of over time. Prior to the outbreak of the First World War, the unit construction cost of all subway lines on both sides of the Atlantic was below $60m 2021USD per kilometre.

Looking across our whole data set, we see that the average cost per kilometre in 2021 dollars of constructing the lines opened before the outbreak of the First World War was $23,111,542, while for lines built after the war it is $45,593,824.  This increase is inflation-adjusted unit construction costs doesn’t appear to have been driven by lines becoming better, for instance by a worldwide trend towards more stations per mile of track or the stations acquiring longer platforms or becoming noticeably fancier. (Some of the pre-1914 stations were relatively opulent and featured nice toilets (see Image 3), as was the case in some of the New York stations that opened to the public in 1904). Physically, the subway infrastructure being built after the First World War was basically similar in quality to that built before the war. Indeed, in some of the subway projects built in the interwar period, the proportion of the lines that were underground was lower than in some of the pre-war lines, as subway lines were extended out further from the urban core and into the suburbs, where they tended to run on elevated tracks, which is cheaper to build that tunnels, or just the surface, which is the cheapest of all. For instance, more than half of the extension of Berlin U2 line that opened in June 1930, which cost 8,580,000 Reichmarks to build, was elevated and less than 47% was tunnelled. In effect, societies after the First World War were paying more for infrastructure of equivalent quality, which strongly suggests to us that changes in the socio-economic and political systems in which the projects were embedded were driving the increase in costs.[1] 

Prior to the First World War, there does not seem to be a marked difference in the costs of projects on both either side of the Atlantic: we can find many examples of American and European lines being built for well below $50m per kilometre. Construction costs tend to cluster around the level pretty much everywhere.  In the interwar period, we start to see much greater dispersion of costs, with some of the lines built in the United States in the 1930s New Deal Era racking up much higher unit construction costs. In the postwar period, we see even greater dispersion of costs, with some of American projects, such as BART and MARTA in Atlanta, incurring eye-watering high construction costs. The MARTA East-West Line that opened to the paying public in 1979 cost an astonishing $294,444,444 per kilometre to build. The Los Angeles subway line that opened to paying passengers in 1993 and which is now known as the B Line (it was originally called the Red Line), cost $268,181,818 per kilometre to build.

 We also see that well into the late twentieth century, subway lines were still being built in some Western cities at low costs, which shows that the problem of rising costs was not a universal trend but is something that can be beaten. London extended the Piccadilly line out to Heathrow Airport in 1977 at a far lower cost ($44,116,035 in 2021USD), even though 100% of this line was underground. The MARTA line opened in 1979 was 10.8km in length, of which just 1.6km were underground. The discrepancy between subway construction costs in Europe and the United States that many researchers have observed appear to open up during and after the New Deal. Below, in the Analysis section, we offer some theoretically-informed explanations for why that would be the case.

Figure 1. Unit Construction Costs

Key Finding Number Two: Major Differences in Construction Costs are a Long-Standing Rather Than New Phenomenon

Other patterns are observable in our data. One of them is that average inflation-adjusted constructions costs vary considerably between Western countries and have varied for about a century. Researchers who compare subway constructions costs in the twenty-first century frequently out that it costs more to build subway lines in the US. Our research shows that this isn’t a new phenomenon at all.  The average cost of all of the US lines in our historical data set is $65,216,703 per kilometre. In the UK (basically London plus the Glasgow subway system), the average cost is lower $40,134,774 per kilometre. The French unit cost of $8,792,660 per kilometre is very low, although it would be risky to put too much credence in this particular figure as we were unable to find cost data for some of the major post-war Parisian metro projects. However, we have a tolerably complete run of data for the German projects in Hamburg and Berlin and find that the costs in Germany are consistently lower than in the United States, at $18,454,887 per kilometre. (Many of these projects were lines that were 100% underground).

Table 1. Comparison of Average Unit Construction Costs in Different Countries

	Cost Per KM in 2021 USD
UK lines, Average Unit Costs	$40,134,774
US lines, Average Unit Costs	$65,216,703
French lines, Average Unit Costs	$8,792,660
German lines, Average Unit Costs	$18,454,887
Canadian lines, Average Unit Costs	$24,492,538
All projects, average unit cost	$35,125,151

It is tempting to look at the data and  just conclude that Europe as a whole is simply better at building good infrastructure cheaply than in North America as a whole and then perhaps conjecture that this difference is driven by certain unchangeable geographical facts of life in North America, such as lower population density, the greater severity of the winters in northern cities like New York and Boston, or the use of different construction equipment and techniques. However, there would be a number of serious problems with such arguments. First, pre-1914 unit construction costs on both sides of the Atlantic were basically similar, as we have noted above. The major differential in costs between the US and Europe emerges later. Second, the Canadian subways in Toronto and Montreal have significantly lower construction costs than the US lines, even though Canadian cities are very similar to those in the northern United States in terms of spatial organization, population density, and construction techniques. Toronto is, famously, often the location in which films set in New York are filmed because the buildings and the roads look basically similar. Yet for some reason, the costs of building subways in Canada was lower.   To us, this pattern is further evidence to support the view that differences in the socio-economic and political systems in which infrastructure projects are embedded explain the differences in construction costs that observers who used present-day data have noted.[2] 

Key Finding Number Three: Ratio of Stations to Line Length Doesn’t Matter

Subway lines vary in terms in how far apart the stations are spaced. In some cities, there are long distances between stations, which force people to walk further to catch a train. In other places, the stations are closer together, which means that a line of a particular length will have more costs associated with building stations. In the historical period covered by our data, the ratio of stations to line length had very little impact on the cost of projects. That’s true for the period before 1914, the interwar period, and indeed the whole period covered by our project as a whole. Similarly, the percentage of a line that is tunnelled (as opposed to running on the surface or on elevated tracks doesn’t appear to matter that much. 

	All years	Pre 1914	After 1918	1918-1939
Correlation for tunnel percentage	0.036336847	0.185848182	0.028900823	0.281355593
Correlation for year	0.45065061	-0.171072239	0.572856991	-0.096231849
Correlation for stations per km	-0.184645765	0.110480413	-0.231721074	-0.149887707
Correlation for length	0.029732141	-0.065515911	0.153251994	0.417972841
Correlation for tunneled KM	0.119579747	0.010539811	0.239836273	0.25382516
Correlation for country	0.377442239	0.449378199	0.468030026	0.464752557
Correlation for Anglo Saxon	0.352638634	0.576314001	0.327688749	0.413676508

---