What Should Our Counterfactuals Be?

Railroads and American Economic Growth

How valuable were the railroads to American economic growth? Intuition tells us that they were extremely valuable. As Robert Fogel writes, describing the prevailing view in the middle of the 20th century, “it was the sine qua non of American economic growth, the prime force behind the western movement of agriculture, the rise of the corporation, the rapid growth of modern manufacturing industry, the regional location of trade, the pattern of urbanization, and the structure of interregional trade.” Roads would not be capable of moving large quantities of goods at remotely competitive rates until the interstate, and the canal, the initial capital intensive transportation technology in the US, was almost entirely replaced by railroads. It was responsible for a very large percentage of GDP, though how much could hardly be estimated; it may have also had a transformative impact on the nature of economic growth. Walter Rostow argued that it was a leading sector of the economy, which induced much of industrialization by itself through backwards linkages to other sectors.

It may then be surprising to you that Robert Fogel, one of the most prominent early economic historians, wrote to attack this conventional wisdom. These arguments can be found in the paper summary of his book, “Railroads and American Economic Growth”. In Fogel’s telling, the railroad was not, in aggregate, a substantial contributor to economic growth. Certainly it shaped where economic growth occurred, but we cannot assume that people would settle in the same way as the world in which there were railroads. Canals are quite capable of shipping bulk goods, and people would settle in places where canals were most suitable. Moreover, we cannot tell the degree by which railroads were superior to canals, only that it was better. As he writes, “one cannot … leap from data demonstrating the victory of the railroad over waterways … to the conclusion that the development of the railroad network … was a prerequisite for the rapid, continuous growth of the internal market. The only inference that one can safely draw is that railroads were producing the same (or a similar) service at a lower cost to the buyer.” (p. 166) Fogel wants to get some sense of the “social savings” involved in the change from canals and waterways to railroads. How much did we save? Some key assumptions are that the costs and benefits of railroads are constant. There are no railroads with much larger benefits than others, and the same with canals. It is also assumed that canals would have constant marginal costs, even as they increased greatly in number and size to replace the railroads. The unobserved parts of the supply curve are complete conjecture, but one may well question the realism of it. (As Jeffrey Williamson does in a review of the book). 

First, he deals with long-haul transportation. In this, railroads have no advantage, as indeed any historian would tell you. L.H. Jenks, author of the earliest economic history works on American railroads, wrote in 1944, “There is no convincing evidence … that railroads have ever carried freight at lower costs either to shippers or to society than canals and waterways. The advantages that early railways showed over canals…are analogous to those of modern highway transport over the railroad.” Fogel’s initial estimate was that the railroad reduced gross domestic product; allowing for additional costs reflecting the increased friction transporting to and from waterways mean that railroads add, at most, 6/10th of one percent in 1890. 

Next, he considers intraregional agricultural hauling. Here is where previous exponents of the railroad hung their caps, and indeed, he does find that most of the social savings of the railroad was here. His procedure is two-fold. First, he finds what the boundaries of feasible agricultural land would be in the absence of railroads. He assumes that people would settle differently without the railroad to sustain them. Then, he finds the cost of shipping by wagon within the new region of feasibly arable land, and comes up with a loss of 2.1% of GDP. However, allowing for the construction of 5,000 miles of new canals reduces the loss to 1.6%, and allowing for changes in road quality reduces it to 1.2%. Note that, while GDP per capita is not much affected, the population of the United States would be much lower than otherwise. There wouldn’t be enough arable, accessible land to sustain them. There is no question, by the way, that the road might have replaced canals and waterways. He estimates the social savings as compared to roads as being a whole third of GDP.

Finally, he looks at backward linkages to other sectors of economy. Railroads required enormous quantities of steel and iron. Did the demand for it induce learning by doing, and a takeoff into modern development? Fogel can find no sign of discontinuities in growth rates. This part is necessarily speculative, in a way that the other sections aren’t. He is not able to apply much hard data to the question. Thus, it is the biggest weakness of his thesis — if railroads did have an impact on productivity, it could be a very large part of GDP.

The maximum estimate that Fogel can reach regarding social saving is that it might be responsible for 7.1% of GDP. He believes that, since everything which could be bent toward that was, estimates below 5% of GDP are far more reasonable. This is simply not a very large proportion. Railroads, then, were not especially relevant for American economic growth. To put in perspective, present day GDP per capita (according to the World Bank) is 81,000 dollars per person, and a four percent reduction would be around 78,000 dollars. We would go from the US to Iceland. 

The primary reaction was that it was an extremely important work, but may not quite be true. It is a result which so deviated from common sense that there must surely be a methodological error, or perhaps counterfactual analysis is ill-suited for large questions. Marc Nerlove, in a review of the book, notes that the implied rate of return on capital invested in railroads was extremely low, and that any other investment would get a return at least three times higher. This suggests that, rather than America being richer if it didn’t invest in railroads and putting it all in wagons and canals, that capital markets then were incredibly imperfect, and that we cannot assume perfect competition. Marc Fishlow published, soon after Fogel, another work assessing the contribution of the railroad, in which he estimated a 15% decline in GDP per capita were railroads to be unavailable in 1890. Even Fogel’s estimates which do not allow for additional construction of canals come nowhere near that. 

Preston McAfee published a satirical article (in which I detect an underlying warmth and good humor toward Fogel) nevertheless savaging the proposition that railroads were immaterial. In “American Economic Growth and the Voyage of Columbus”, he asks what would happen if, instead of discovering North America, Christopher Columbus had in fact fallen off the edge of the earth. The consequences are naturally absurd — Australia is unhitched by Welsh divers and attached to Britain, with this becoming known as “America”, the European powers fight great balloon wars, and so on. 

It is silly, but it is asking us an important question about counterfactuals. In order to say that something caused another thing, we require the assumption of a counterfactual world in which, were it not for that cause, the thing would never have happened. Heck, even something as simple and a supply and demand curve implies counterfactual thinking — there are combinations of prices and quantities that might have happened, but didn’t. To this must be appended a sense of which causes are most likely to be modified under different circumstances. If one thinks of why they were late to school in the morning, it is true that “but for” the traffic, you would have made it on time; but also true that if you had woken earlier, or you lived closer to school, or school was held online, or that it didn’t snow, or that one thousand years ago someone died of plague setting off a chain of consequences such that we did not invent teleportation as soon as will are all equally causes. 

As changes get bigger, they imply causes of their own, which in turn imply massive deviations from how things actually happen. What is a world in which the railroad is not invented? The railroad was not a historical contingency — it was a natural and obvious result of the development of the steam engine, and of modern steel manufacture. To not invent the railroad implies enormous changes. What can we say, then, is the counterfactual with no railroad? It would have to be a world so alien from our own as to have little resemblance. We cannot say that it was the railroad which caused things, so much as it was the complex of associated technological developments which did, and if we remove those, we have no capacity for predicting how things might have been in the slightest. All counterfactuals imply some degrees of imagination, but too many degrees leads us into mathematical make-believe. 

The degree of plausibility is what separates out finding causality from alternate history. In economics, we might want to know what would happen if a group of children receives one more or less year of education. People receive different amounts of education all the time. Meanwhile, a historian might want what would happen if Horatio Nelson were unexpectedly shot at the beginning of the Battle of Trafalgar. (Oh, he was shot? I see the time-traveling French have been at it again). There is no obvious alternative — it implies a delicate chain of contingent events which cannot be forecast from data on very similar events elsewhere. 

I have had a (entirely one-sided, and indeed unnoticed, as the elephant is unbothered by a dung beetle) disagreement with professors Heblich, Redding and Voth over their paper “Slavery and the Industrial Revolution”. They seek to answer what the contribution of slavery was to the Industrial Revolution in Britain (as you may well have divined from the title), and use the length of slave voyages across the Atlantic as a plausible instrument for slave wealth, which they connect to specific regions back in England. They compare the difference in industrialization, and find the implied impact of additional capital on growth. So far they have only found the local effects. They have found why some regions might have grown relatively faster. To find the impact on England as a whole, they construct a simple model of economy, and simulate what would happen if access to slavery investments were cut off, holding everything else equal.

It is the holding everything else equal which is questionable. The sugar colonies, which were the primary source of revenue from slaveholding, were government protected and subsidized. English consumers had to pay a substantial tariff for imported sugar, which raised prices well above those of Nantes, France. The sugar islands were guarded by a garrison of troops and a small fleet, at considerable expense. One cannot “hold all else equal” — would you say that Britain would invest in exactly the same troops to guard nothing at all? One is caught between wild speculation, and patent unreality, and I cannot tell you which is better.

There has been later reassessment of the Fogel thesis. Richard Hornbeck has two papers, one with Dave Donaldson, and the other with Martin Rotemberg, on more accurately assessing the impact of railroads on American economic growth. The first, with Donaldson, is “Railroads and American Economic Growth: A ‘Market Access’ Approach”. This uses new methods, which were not available at the time. (Also computers. A lot of the paper is simply doing calculations which could not be done under any circumstances in 1964). Infrastructure expands into new counties discontinuously — either you have access to railroads, or you don’t. Economic historians have carefully compiled maps of the expansion of railroads over time, tracking when a railroad arrived in each county. They use this to account for spillovers from trade. Because trade networks are interlinked, and improvement in one place affects relative prices and trade quantities everywhere. 

Their headline results track closely with Fogel’s. It would be a massive blow to agriculture. Without railroads, agricultural land not already near a waterway would lose 60% of its value, which implies a loss of 3.22% of GDP. It could be at most 5.35% of GDP, if land not by a waterway lost all of its value. Additional canals could only mitigate 30% of this. Their estimates are not sensitive to population reallocation. Population in the US would substantially decline, and world output and utility would decline.

Neither Donaldson-Hornbeck nor Fogel were able to seriously consider the effects of the railroad on productivity in manufacturing. For a long time, the data didn’t exist. There is a project underway to digitalize the Census of Manufactures, and Hornbeck and Rotemberg have a recent paper using it to answer some really big questions. They drop the assumption of perfect competition, or price equaling marginal cost, because they have records of all of firms’ inputs and all of their outputs and the price they charged for each. If there is a substantial markup of a firm’s output price over their input cost, the firm needs to be producing more! They are at an inefficiently low level of production. Because they are concerned only with manufacturing, their estimates supplement prior ones.

Hornbeck and Rotemberg find that not only are there substantial markups, and that these vary by county, but that resource use is changed when railroads come to a place. It is not that they reduce distortions, but that they lead to a greater expansion of things which should expand. Exposure to a railroad led to a 20% increase in county productivity. Their low end estimate, keeping the same population in the US, is 8.2% of GDP lost. Their high end, allowing for immigration out (or immigration never happening) is 27% of GDP. They estimate an annual social return on capital of 48%, with railroads only capturing 7% of that. 

So how’s that for a plot twist? The railroad likely was responsible for an enormous part of America’s economic growth, once you have more and better data.