TY - JOUR
AU - Hanlon, W, Walker
AB - Abstract Can temporary input cost advantages have a long-run impact on production patterns? I study this question in the context of shipbuilding from 1850 to 1911. Although North America was the dominant wood shipbuilding region in the mid-19th century, the introduction of metal shipbuilding shifted the industry to Britain, where metal inputs were less expensive. After 1890, Britain’s input price advantages largely disappeared but its dominant position in the industry persisted. I show that American shipbuilders exposed to British competition struggled to transition to metal shipbuilding and present evidence that the mechanism behind Britain’s persistent lead was the development of pools of skilled workers. 1. Introduction Can initial input cost advantages have a persistent influence on patterns of trade and production, even after those advantages disappear? This is a classic question in international trade, with implications for our understanding of the origins of current trade patterns as well as the impact of certain forms of industrial policy. The answer to this question is particularly relevant today, given ongoing debates over the use of government intervention to protect domestic industries. An ideal empirical setting for studying these issues would be characterized by a set of similar locations, some of which enjoy an initial input cost advantage that eventually disappears, such that all locations face similar cost and demand conditions in the long-run. Identifying settings fitting this description has proven difficult. As a result, our understanding of the extent to which temporary advantages can have long-run effects on trade patterns remains extremely limited, particularly given the importance of the issues at stake, which are central to current trade policy debates. This paper studies the international shipbuilding industry from 1850 until just before World War I, a setting that provides the features needed in order to look at the long-run effects of temporary input cost advantages.1 In the mid-19th century, North American shipbuilders were the dominant producers in this industry. However, Britain had a more advanced iron industry in the mid-19th century that resulted in lower iron input prices. This initial input cost advantage, together with the rise of metal shipbuilding after 1850, allowed British shipyards to attain a dominant position in the industry by 1880. However, during the 1880s and 1890s Britain’s initial input costs advantage largely disappeared due to the discovery of new iron reserves in North America and the development of successful American iron and steel producers. The main analysis in this study thus focuses on the decade after 1900, when initial differences has essentially disappeared and locations in Britain and Eastern North America faced similar cost and demand conditions. I show that, despite losing their advantage in metal input costs, British producers maintained a dominant position in the shipbuilding industry after 1900, whereas North American shipbuilders, despite their earlier dominance in the industry, struggled to adapt the new metal shipbuilding technology. This pattern is particularly striking given the broad success of American manufacturing in general, and metal goods manufacturing in particular, during this period. The goal, then, is to understand the role that Britain’s temporary initial advantage played in this process, and specifically, why the dominant position that British shipyards established by 1880 persisted after Britain’s initial cost advantages had disappeared. To make progress here, I take a somewhat novel approach. Rather than focusing on British shipyards and trying to understand why they were successful, I instead turn the question around and ask: why were North American producers largely unable to catch up to the British after 1890, despite having similar raw materials costs as well as a long history of shipbuilding? Focusing on North American producers is useful because it allows me to take advantage of two novel sources of variation in exposure to British competition. Studying the effect of exposure to British competition can reveal whether Britain’s initial advantage persisted in part because British competition retarded the development of the industry in other locations. Having exogenous variation in exposure to this competition is necessary in order to rule out the possibility that North American producers failed to successfully transition to metal shipbuilding because of other factors. The first source of variation in exposure to British competition that I exploit is generated by the fact that shipbuilders in the Great Lakes were protected from foreign competition because of the difficulty of moving large ships through the locks and canals connecting the Lakes with the Atlantic. This geographic barrier created a market that remained largely isolated from foreign competition until the construction of the St. Lawrence Seaway in the 1950s. Other than selling into separate output markets, I show that shipbuilders did not face systematic differences in input costs or demand conditions on the Great Lakes relative to the Atlantic Coast. I also exploit a second source of exogenous variation in exposure to foreign competition driven by access to government protection. In particular, although the United States used a range of protective policies to aid domestic shipbuilders, Canada was unable to offer similar protections to domestic producers because it was part of the British Empire. These sources of variation allow me to develop a counterfactual for the development of North American shipbuilding in the absence of competition from initially advantaged British producers. Comparing this counterfactual to the development of the industry in Atlantic Canada, which was fully exposed to British competition, identifies the impact of exposure to initially advantaged British producers on the development of the North American industry. Moreover, focusing the analysis on a comparison between wood and metal shipbuilding helps me to deal with a variety of factors, such as unskilled wage levels, access to finance, or the availability of shipyard space, that affected both types of shipbuilding. To track the development of the shipbuilding industry in each location, I draw on detailed new data covering ship output by location from 1850 to 1911. These unique data are available because in order to obtain insurance ships need to be inspected and listed on a register, such as Lloyd’s Register. These registers provide a catalog of the majority of large merchant ships constructed during the study period, including information on their size, construction material, location and year of construction, and so forth. The register data used in this paper were digitized from two sources, Lloyd’s and the American Bureau of Shipping. The data come from thousands of pages of raw documents and cover tens of thousands of individual ships, providing a fairly comprehensive view of the development of the shipbuilding industry in North American and Britain. My main findings show that North American shipbuilders that were protected from British competition, either because they were in the Great Lakes or the protected U.S. Coastal market, rapidly adopted metal ship production once the cost of metal inputs in North America converged to British levels. In contrast, in those areas fully exposed to British competition, such as the Canadian coast, the industry failed to make the transition and was effectively eliminated as an important industrial sector. These results indicate that British shipyards remained more productive than the North American yards that competed with them, even after their initial input cost advantages had disappeared, and that the failure of North American producers to successfully transition to metal shipbuilding can be linked directly to exposure to competition from more productive British yards. This evidence raises questions about the specific mechanisms through which Britain’s initial advantage was translated into persistently higher productivity. In the second half of the paper I try to shed light on these mechanisms. A natural starting point for thinking about how a temporary initial cost advantage might have had long-lasting effects is the presence of learning-by-doing. In fact, previous work by Thompson (2001) has already documented the presence of important dynamic learning in shipyards. However, the learning documented by that study occurred within shipyards.2 This type of learning is unlikely to explain the persistent advantage of British producers because individual shipyards remained small relative to industry output and concentration in ship production was never high. A more likely explanation is that, in addition to within-firm learning, there were also localized learning spillovers. Such external learning spillovers could explain the persistent dominance of British shipbuilding firms as well as why these firms were spatially concentrated in the areas around Glasgow, Newcastle-upon-Tyne, Sunderland, and Belfast. However, direct empirical evidence is needed in order to verify that external learning effects were in fact a feature of the shipbuilding industry. In order to provide evidence that this industry was characterized by localized learning effects, I exploit the locations of Navy Shipyards in the United States. These shipyards were established around 1800, long before the introduction of metal ship production, so their locations were unlikely to have been chosen to advantage metal shipbuilding. When the U.S. Navy began large-scale metal ship construction in the 1880s, these shipyards began producing and repairing metal ships. Thus, the Navy Yard locations provide plausibly exogenous variation in local experience in metal ship production. Using this, I study whether this experience had benefits for nearby private shipbuilding firms, consistent with localized learning effects. Indeed, my analysis shows that private shipyards located near Navy shipyards were much more likely to make the transition from wood to metal ship production. These effects disappear for locations more than 50 km from Navy yards. This result continues to hold when controlling for other ways that firms might have benefited from proximity to Navy yards, such as winning government contracts. Thus, I conclude that the shipbuilding industry was characterized by dynamic localized learning effects. The presence of these dynamic effects can explain why Britain’s initial advantage resulted in a persistent lead. In the last part of the paper I consider the specific nature of the localized learning effects I have documented. In principal these could take a variety of forms, including direct knowledge spillovers, specialized input producers, or the development of pools of local workers with specialized skills. An extensive review of contemporary sources and historical studies of the shipbuilding industry leads me to conclude that the most important factor translating initial input cost advantages into persistent trade patterns was likely the development of large pools of skilled craft workers. Metal shipbuilding required a variety of skills that were acquired through experience. These skills were crucial, and differed in important ways from the skills needed in either wood shipbuilding or other metalworking industries. Contemporary reports describe how Britain’s initial advantage in metal shipbuilding led to the development of pools of skilled workers that substantially improved the productivity of British yards. Because these skills were embodied in a large number of workers, and because production required a wide variety of skills, coordination problems made the relocation of shipyards difficult, locking in a source of local advantage. North American shipbuilders lacked easy access to these pools of skilled workers, resulting in higher wages and costs. Although they compensated by substituting toward unskilled labor and capital, the high cost of skilled work that could not be eliminated left them less productive than their British competitors. Although this evidence is historical, rather than statistical, and should be evaluated as such, it offers a compelling and coherent explanation for why initial input cost advantages allowed British shipbuilders to gain, and then maintain, a dominant position in the industry in the decades before World War I, as well as why these advantages remained localized. The role of temporary initial advantages in influencing long-run trade patterns and welfare outcomes is the subject of a substantial theoretical literature in international trade (e.g., Krugman 1987; Lucas 1988; Grossman and Helpman 1991; Young 1991; Matsuyama 1992; Lucas 1993). However, generating empirical evidence in this area has proven to be challenging. This study contributes to a limited set of empirical research in this area, including Krueger and Tuncer (1982), Baldwin and Krugman (1988), Head (1994), Irwin (2000), Juhasz (2018), Lane (2016), and Mitrunen (2019), as well as work on persistence in urban economies such as Bleakley and Lin (2012). Among this set, the most related recent paper is Juhasz (2018), which exploits the Napoleonic blockade to show that temporary protection from foreign competition in output markets can have persistent effects. An important difference in my study, relative to previous work, is that I focus on the impact of temporary input cost advantages, rather than output market protection. This is an important distinction, particularly given that input subsidies have been one of the main tools used in some prominent industrial policy cases, such as in Korea (Lane 2016). One difference between this study and previous work such as Juhasz (2018) is that I attempt to delve deeper into the underlying mechanisms behind the persistent effects that I document. Although external learning is thought to be important for both the rate of growth and the spatial distribution of economic activity, it is often difficult to study empirically. This study provides new statistical evidence of learning effects as well as historical evidence indicating that the likely source of these effects is the development of pools of skilled local workers, a channel often overlooked in existing work. The importance of skilled workers helps explain a number of features of the shipbuilding industry. For example, the role of experience in generating worker skills provides a potential explanation for the dynamic learning effects documented in existing studies (Searle 1945; Rapping 1965; Argote, Beckman, and Epple 1990; Thompson 2001).3 The existence of locked-in sector-specific skills can also help explain the continuance of wood shipbuilding in Eastern North America long after the technology was clearly inferior to metal and wood supplies had dwindled (Harley 1970, 1973). Finally, the importance of skilled worker pools can also help explain the geographic concentration of the industry despite the relatively small size of individual firms. 2. Empirical Setting The shipbuilding industry was an important industrial sector in the British, U.S., and Canadian economies in the 19th century.4 This industry underwent dramatic changes during the period covered by this study, including the shift from wood ships to vessels made of iron or, later, of steel. In the 1850s, iron shipbuilding was still in its infancy. By the last decade of this study, iron and steel shipbuilding had come to dominate and metal ships accounted for 96.4% of the tonnage produced in the United Kingdom, United States, and Canada. However, in the United States and Canada wood shipbuilding remained important, accounting for 17.5% of the tonnage produced from 1901 to 1910. The transition from wood to iron and steel was driven by two main factors. One was the shift from sail to steam power.5 The share of steamships in total production rose from near zero before 1850, passed 50% of production after 1880, and made up over 95% of production in 1900–1910 (see Online Appendix A.4). This advantaged metal ships, which were better able to handle the increased vibration and hull stress associated with steam power (Harley 1973). One implication of this fact is that, although wood and metal ships were highly substitutable for many purposes, they were not perfect substitutes.6 The second key factor driving the shift to metal hulls was improvement in the quality and reduction in the price of iron and steel inputs, together with the increasing scarcity of timber resources near the main shipbuilding locations. At the beginning of the study period there was a distinct pattern of input cost advantages in the shipbuilding industry that determined production patterns (Pollard and Robertson 1979). In particular, the forests of the Eastern U.S. and Canada gave North American shipbuilders cheap access to wood. As a result, the United States was the world’s leading shipbuilder, whereas Canada was also an important ship producer. Not only were the North American producers larger, they were also more innovative, introducing new designs such as the clipper. However, shipbuilders in Britain had access to cheaper iron inputs thanks to their large domestic iron industry, giving British producers an early lead in iron shipbuilding. These advantages were important; although quantitative estimates of the share of costs accounted for by metal or wood inputs are scarce, and would have depended substantially on the specifics of any given design as well as fluctuating input prices, historical sources make it clear that the price of these inputs played a major part in determining overall costs.7 For example, evidence from Report of the Merchant Marine Commission to Congress (1905) indicates that, in 1900, the cost of metal inputs accounted for between one-quarter and one-third of the price of a standard 5,000-ton freighter. By the late 19th century, however, these initial input price differences had almost completely disappeared. This is shown in Figure 1. For wood prices, shown in the top panel of Figure 1, the rise in (eastern) U.S. prices was due to the increasing scarcity of forests near the shipbuilding areas (Hutchins 1948). As a result, by the late 19th century, shipbuilders on the Atlantic coast of North American often had to import wood from the Great Lakes region (Hutchins 1948). For iron prices, shown in the middle panel of Figure 1, the convergence between North American and British prices was driven by the discovery of new iron ore reserves in the United States, such as the rich reserves in the Mesabi iron ore range in Minnesota.8 These discoveries led to an expansion in U.S. iron and steel production and drove a surge in manufacturing exports starting in the 1890s (Irwin 2003).9 Although Figure 1 describes iron prices, similar patterns appear for steel.10 U.S. iron and steel exports surged from $25.5 million (3% of exports) in 1890 to $121.9 million (9% of exports) in 1900 and reached $304.6 million (12.5% of exports) in 1913 (Irwin 2003). By 1900, U.S. manufacturers were even exporting substantial amounts of iron and steel to Britain.11 Figure 1. Open in new tabDownload slide Input prices and relative prices in the United States and United Kingdom, 1850–1913. U.K. iron prices are from the Abstract of British Historical Statistics. U.K. wood prices are from the Statistical Abstract of the United Kingdom. U.S. prices are from Historical Statistics of the United States, Colonial Times to 1870, Vol. 1. U.K. prices are converted into dollars using the exchange rates from MeasuringWorth.com. Figure 1. Open in new tabDownload slide Input prices and relative prices in the United States and United Kingdom, 1850–1913. U.K. iron prices are from the Abstract of British Historical Statistics. U.K. wood prices are from the Statistical Abstract of the United Kingdom. U.S. prices are from Historical Statistics of the United States, Colonial Times to 1870, Vol. 1. U.K. prices are converted into dollars using the exchange rates from MeasuringWorth.com. In Canada, the development of local coal mining and iron and steel production had similar effects. This occurred both in the Great Lakes and along the Atlantic Coast. Of the Canadian Atlantic Coast, an area that is particularly important for this study, Sager and Panting (1990, p. 15) write that, “It is difficult to show that the Atlantic region as a whole lacked the resources necessary to make the transition to iron steamships, and all the more difficult when Nova Scotia acquired an iron and steel complex. The region possessed coal, iron ore, capital, a labor ‘surplus’, and long experience in ship construction and management”. Supporting this, Online Appendix A.7 shows that Canadian iron and wood price trends were similar to U.S. prices. The dramatic reduction in transport costs that occurred in the second half of the 19th century, together with changes in tariff policy, also contributed to input price convergence, by giving coastal North American shipyards easier access to foreign suppliers.12 As a result of this combination of factors, the strong initial patterns of comparative advantage driven by input prices that defined the shipbuilding industry in the mid-19th century had essentially disappeared by 1900, as shown in the bottom panel of Figure 1. One feature of shipbuilding during the period I study was the highly competitive and fragmented nature of the industry. Hutchins (1948), for example, describes shipbuilding as “naturally one of the most highly competitive of all markets. . .” The main reason for this diffuse market structure appears to be geographic constraints that limited the size of individual shipyards, particularly the older yards located in larger towns. Competition in the industry was also increased by the very low cost of transporting a ship between navigable locations (relative to the cost of production). This meant that shipyards had to compete directly even with very distant competitors in a global market. The Great Lakes represented an important exception to the global ship market. In particular, prior to the opening of the St. Lawrence Seaway in the 1950s it was difficult for large vessels to transit between the Great Lakes and the Atlantic Ocean. This geographic barrier created an effectively isolated Great Lakes market. As evidence of this, my data show that in 1912, 97% of the vessels (by tonnage) homeported on the Great Lakes were also constructed on the Great Lakes, whereas over 94% of the tonnage constructed on the lakes remained there.13 In terms of size, in the decade from 1901–1910 the Great Lakes market accounted for 2.3 million tons of production or 12.5% of total tonnage produced in the United Kingdom, United States, and Canada. The main reason for this isolation was the limitation placed on the size of vessels that could pass through the canals connecting the Great Lakes to the Atlantic, particularly the Welland Canal, which bypassed Niagara Falls to connect Lake Erie and Lake Ontario, and the Lachine Canal on the St. Lawrence River at Montreal. To pass these canals, large vessels had to be cut apart and then later reconstructed. This was a time-consuming and costly process.14 The Annual Report to the Commissioners of the Navy (1901, p. 15) states that, as a result, “Construction on the seaboard and on the lakes up to the present time should be considered as different industries, indirectly related”. Though protected from foreign competition, the other factors driving the transition from wood to metal in the Great Lakes market were similar to conditions on the Atlantic Coast. For example, the data presented in Table 1 show that there were no systematic differences between iron and wood prices on the Great Lakes compared to the Atlantic Coast. Although iron prices were relatively low in some Lakes states, like Illinois, they were high in others, such as Michigan and Ohio. Similarly, there is no evidence that Atlantic coast producers had a relative advantage in wood prices.15 Table 1. Iron and wood prices in some Atlantic and Great Lakes States, 1900. Open in new tab Table 1. Iron and wood prices in some Atlantic and Great Lakes States, 1900. Open in new tab On the demand side, incentives for producing metal rather than wood ships in the Lakes were also similar to on the coast. This is important because one of the identifying assumptions in the main analysis is that there were no factors that systematically increased the demand for metal ships relative to wood ships (I do not need to assume that trends in overall demand for shipping capacity were similar across locations). For example, the transition from sail to steamships that took place in the Lakes was similar to the transition in the Atlantic market as a whole, as described in Online Appendix A.4. The incentives for using metal provided by opportunities to construct larger ships were actually weaker in the Great Lakes than on the Coast, because, as shown in Online Appendix A.6, maximum ship sizes in the Lakes remained smaller than in the Atlantic.16 On the other hand, metal ships did last longer on the Lakes because freshwater was less corrosive, which may have provided some increased incentive for metal ship production there. Although ships on the Lakes did have different designs than those on the coast, such as being longer and skinnier to maximize use of the available locks, there does not seem to have been any important differences in the techniques used to construct lake ships.17 One may wonder if ships on the lakes were more constrained by depth limitations in a way that increased the attractiveness of building with metal. I examine this possibility in Online Appendix A.13 using data on the drafts of 1,000 ships collected from the 1912 ABS registry. These data provide no evidence that Great Lakes ships were built using designs that reduced depth, which suggests that depth limitations did not impose a substantially different constraint on the lakes than on the coast. Of course, this is setting aside the impact of depth on the Welland and Lachine Canals connecting the Lakes to the Atlantic, where depth limitations and other size constraints essentially precluded the passage of large vessels. Large ships constructed on the lakes were never intended to pass through these. It is also natural to wonder whether the structure of ship ownership was different in the Great Lakes than on the Atlantic Coast. An examination of this issue, in Online Appendix A.14, shows that the concentration of ownership was low in all locations and fairly similar for ships homeported on the Atlantic Coast (HHI = 503) and the Great Lakes (HHI = 700). Thus, variation in ownership concentration in different areas is unlikely to be an important factor in the analysis. This study takes advantage of a second form of exogenous variation in protection from foreign competition generated by access to government protection. Specifically, I take advantage of the fact that the United States actively protected domestic shipbuilders whereas Canada could not offer similar protection.18 Support from the U.S. government came in two forms. First, the United States imposed a ban on the use of foreign-built ships for direct trade between American ports (coastal trade). This policy, which existed throughout the study period and continues today, created a protected market for U.S. shipbuilders. Essentially, this policy acts like a prohibitively high tariff on the import of ships for use in the coastal trade. The size of this market in 1901–1910 was equal to about 8.7% of the total tonnage produced in the United States, United Kingdom, and Canada during this period.19 A second important channel of government influence on shipbuilding was through the Navy. Warship construction gave domestic shipyards experience and may have helped generate pools of skilled workers.20 From 1901 to 1910 the U.S. Navy bought vessels totaling 643,441 tons. Although Navy vessels sizes are measured in displacement tons, which is not directly comparable to the tonnage measure for merchant vessels, this is roughly equivalent to 3.3% of total U.S., U.K., and Canadian tonnage. Although the United States had access to the full range of protective policies, Canada, as part of the British Empire, did not have the ability to enact similar policies. Specifically, Canada could not close coastal trade to British-built ships, nor did it have an independent navy during this period to provide orders to domestic yards or to operate government shipyards.21 As a result, data for 1912 show that 46% of the total tonnage homeported in Canada in that year was constructed in the United Kingdom. In contrast, only 7.6% of the tonnage homeported along the U.S. coast was built in the United Kingdom. Thus, comparing the experience of the United States and Canada allows us to observe the evolution of this industry with and without access to government protection. Although my analysis takes advantage of output market segmentation at the regional level (U.S. Great Lakes, U.S. Coast, Canada Great Lakes, Canada Coast), within these regions there was enormous heterogeneity across locations. The length of the Great Lakes stretches over 1000 km West to East, from Minnesota to New York State, and over 700 km from North to South, with over 7,000 km of coastline. Shipbuilding took place in large cities such as Chicago, Toronto and Detroit, but also in many small out-of-the-way locations, such as Thunder Bay, ON and Saugatuck, MI. Coastal shipbuilding in Canada spanned a distance of over 1,600 km, from Montreal to St. John’s, Newfoundland. On the U.S. Coast, shipbuilding locations stretched over 2,000 km from Maine to Florida. As a result, even within a region, individual shipyards faced variation in input prices, availability and quality of shipyard space, labor market conditions, and so forth. This is reflected in the wide variation in state-level input prices within regions shown in Table 1, despite the fact that I do not observe systematic differences in input prices across the regions. This variation motivates my use of individual locations as the unit of analysis. The one factor that tied together the heterogeneous set of shipyard locations within each region was segmentation in the output market, the key source of variation exploited in this study. 3. Data The main analysis relies on a unique new data set derived from individual ship listings on two registers, one produced by Lloyd’s and the other by the American Bureau of Shipping (ABS, sometimes called “American Lloyd’s”). The primary purpose of these registers was to provide insurers and merchants with a rating of the quality of each ship. This provided shipowners with a strong incentive to have their ship included on at least one major register, and often more than one. As a result, the registration societies claimed that the vast majority of major merchant ships (e.g., over 100 tons) were included on one of the lists.22 The data cover only merchant ships; warships are not included in the analysis. The vast majority of these were cargo carriers, though the data also include passenger liners, some fishing and whaling vessels, and other miscellaneous types (tugs, large barges, etc.). The registers were published annually and included a variety of information about each ship. Figure A.1 in the Online Appendix provides an example of the data from the first page of the Lloyd’s Register for 1871–1872. From each register, I have digitized the ship name, type (sail vs. steam), construction material (wood vs. metal), tonnage, the location and year of construction, and in some cases the shipyard and current home port.23 This study uses data from registers for three years, 1871, 1889, and 1912.24 Because the registers include all active ships in these years, and because ships generally last many years after construction, these snapshots provide coverage for most ships built between 1850 and 1911.25 Specifically, I use the 1871 register to track ships built before 1871, the 1889 register to track ships built between 1871 and 1887, and the 1912 register to track ships from 1888 to 1911.26 For each snapshot year I digitized both the Lloyd’s Register and the ABS Register. Table A.2 in the Online Appendix describes the number of vessels included in the data from each of the registers used in this study. The full data set includes just over 69,000 ships. Most of the analysis focuses on the subset of these built in the United States or Canada from 1851 to 1910. The data required extensive processing to clean and standardize location names, eliminate duplicate entries that appeared in both registers, identify the construction material for each ship, and so forth. After eliminating duplicates, the main analysis relies on observations for 18,700 ships built in the Great Lakes or Atlantic Coast of U.S. or Canada between 1851 and 1910. Within the regions that I study, it is possible to identify the exact location of construction for the vast majority of ships.27 Some summary statistics for the data on production by location used in the main analysis are reported in Online Appendix Table A.1. Maps of the data are available in Online Appendix A.5. In 1901–1910, the most important period for my analysis, my data cover 160 active shipbuilding locations. Although this may not appear to be a large sample size, it is comparable or larger than the sample size available in most previous studies in this literature.28 These locations, which form the main unit of observation, are towns or cities. Most of these contained one active shipyard, though some may have contained several.29 Of the locations in my data in 1901–1910, 74 were located on the U.S. Atlantic Coast, 60 on the Canadian Atlantic Coast, and 26 on the Great Lakes. Metal ships were being produced in 43 locations, some of which also produced wood ships, with 21 on the U.S. Atlantic Coast, 16 in the Great Lakes, and just 6 metal shipbuilding locations surviving in Coastal Canada. Thus, the majority of North American shipbuilding locations continued to focus on wood ship production. In addition to the main data, I have also constructed several controls using Census data. I control for nearby employment in metal-working or wood-working industries using county-level Census data from the United States and Canada in 1880 (see Online Appendix A.8 for details). These county data are not available for Newfoundland so some observations are lost when these controls are included. Controls for iron and lumber prices at the state level, available only for the United States, come from the Census of 1900 (see Online Appendix A.9 for details). 4. Main Analysis The aim of the analysis presented in this section is to establish two main results. First, that British shipbuilders developed a leading position in metal shipbuilding in the mid-19th century, a time when they enjoyed advantageous cost conditions, and that this advantage was largely maintained after the initial cost advantages disappeared in the 1890s. Second, that the failure of North American producers in certain regions to successfully transition to metal shipbuilding after their metal input costs converged to British levels can be directly linked to exposure to competition from initially-advantaged British producers. The first of these results is readily apparent from the data, so most of this section focuses on the second. A useful starting point is Figure 2, which describes overall ship output in North America (United States and Canada) and Britain from 1845 to 1911. We can see that North America was initially the largest shipbuilding area, but it was soon surpassed by Britain. By the 1880s Britain dominated the market and this continued up to World War I (WWI) despite the fact that Britain’s input price advantages essentially disappeared in the 1890s. Figure 2. Open in new tabDownload slide Merchant ship production in the Unitd States, United Kingdom, and Canada 1845–1911. Data based on both the Lloyd’s and ABS Registers. The United Kingdom includes all of Ireland. The United States and Canada data cover only the Atlantic and Great Lakes regions. The two vertical lines in this figure mark the points at which the registers providing the data switch. The fact that I do not observe sharp drops at these points suggests that I am not losing too many observations by digitizing registries every 20 years. Figure 2. Open in new tabDownload slide Merchant ship production in the Unitd States, United Kingdom, and Canada 1845–1911. Data based on both the Lloyd’s and ABS Registers. The United Kingdom includes all of Ireland. The United States and Canada data cover only the Atlantic and Great Lakes regions. The two vertical lines in this figure mark the points at which the registers providing the data switch. The fact that I do not observe sharp drops at these points suggests that I am not losing too many observations by digitizing registries every 20 years. The next set of charts, in Figure 3, can help us make sense of the overall production patterns. These graphs present output for each country divided into wood or metal ships. Here we can see that the United Kingdom transitioned to metal ship production early, in the 1860s and 1870s, with wood ship production essentially disappearing by the 1870s. In the United States, the transition to metal ship production happened much later, mainly in the 1890s, when U.S. iron and steel prices were falling to U.K. levels. A third pattern is offered by Canada, where we see no evidence of a substantial move into metal ship production. Instead, most Canadian producers remained tied to the declining wood shipbuilding industry. Figure 3. Open in new tabDownload slide Shipbuilding tonnage by construction material. Data based on both the Lloyd’s and ABS Registers. Figure 3. Open in new tabDownload slide Shipbuilding tonnage by construction material. Data based on both the Lloyd’s and ABS Registers. The key question posed by Figures 2 and 3 is why, after the price of metal inputs fell in the 1890s, were North American shipbuilders unable to catch up to British production levels? One possible answer to this question is that the early lead enjoyed by British producers made them more productive and that exposure to these more productive foreign competitors made it difficult for North American producers to adopt the new metal shipbuilding technology. An alternative possibility, however, is that other factors made North America generally unsuitable for metal ship production. One way to evaluate these alternatives is to exploit plausibly exogenous variation in exposure to British competition across locations within North America that faced similar environments in other respects. Next, I provide graphical evidence describing the two key dimensions of variation that I will use to isolate the impact of exposure to foreign competition on the development of the North American industry. Figure 4 looks at the share of output (by tonnage) of metal ships in the United Kingdom, on the Atlantic coast of the United States and Canada, and in the Great Lakes. The key feature to note in this graph is the production pattern observed on the Atlantic Coast of North American and the pattern observed in the Great Lakes. Although the share of metal ship production was similar in these two regions until 1880, after 1880 we can see that there was a dramatic shift. Shipbuilders in the Great Lakes rapidly converged to the pattern of production observed in the overall Atlantic market (United Kingdom, United States, and Canada) as the price of metal inputs in North America fell, whereas this convergence process was much slower among North American Atlantic Coast producers. Thus, this graph reveals the impact of exposure to foreign competition on Atlantic Coast producers. Figure 4. Open in new tabDownload slide Evolution of production patterns by region. Data based on both the Lloyd’s and ABS Registers. The United Kingdom includes Ireland. The “All Atlantic” category includes production in the United Kingdom, United States, and Canada. Figure 4. Open in new tabDownload slide Evolution of production patterns by region. Data based on both the Lloyd’s and ABS Registers. The United Kingdom includes Ireland. The “All Atlantic” category includes production in the United Kingdom, United States, and Canada. Figure 5 compares shipbuilding in the United States and Canada to highlight the role that exposure to British competition played in the transition from wood to metal shipbuilding. The left-hand panel shows that, on the Atlantic Coast, U.S. shipbuilders transitioned to metal more rapidly than Canadian builders. In contrast, on the Lakes, where shipbuilders were more protected from foreign competition, the United States and Canada show similar patterns. These patterns reflect the fact that the protection offered to U.S. shipbuilders was important on the coast, whereas U.S. government support had less effect on the Great Lakes, where producers were already protected from foreign competition by geographic barriers. Figure 5. Open in new tabDownload slide Evolution of metal share on the Coast versus the Lakes. Data based on both the Lloyd’s and ABS Registers. Figure 5. Open in new tabDownload slide Evolution of metal share on the Coast versus the Lakes. Data based on both the Lloyd’s and ABS Registers. Overall Canadian ship production in the Great Lakes was low, even accounting for the smaller Canadian population in the lakes region. Although the population of Ontario, the Canadian province bordering the Great Lakes, was equal to 7% of the population of U.S. states bordering the lakes in 1900 (or 12% if New York and Pennsylvania are excluded), Canada built only 3.8% of the tonnage produced on the Great Lakes in 1900–1910. One important factor behind this pattern was the prohibition on using foreign-built ships to serve routes connecting any two U.S. ports. Since most ports on the Great Lakes were American, a ship built in Canada would face a permanent limitation on the routes on which it could be used. On the other hand, a ship built in the United States still had the ability, if needed, to service routes between Canadian ports. This provided an incentive for production on the Great Lakes to take place in the United States rather than Canada. However, Figure 5 shows that Canada and the United States exhibited roughly similar ratios of metal to wood ship production in the Great Lakes region.30 The key to reconciling the protection available to U.S. producers, and their greater share in overall production, with the similar patterns shown in Figure 5 is to note that, although protection from the U.S. government benefited U.S. lakes producers, it did not specifically benefit metal ship production relative to wood ship production.31 Thus, it should not affect relative production in these two sectors in the United States relative to Canada. In contrast, on the Atlantic Coast protection from British competition was particularly important for metal ship producers relative to wood because it was in metal ships that British competition mattered. Next, I turn to the econometric analysis. I begin by looking at the extensive margin, that is, whether locations were active in a particular sector (wood or metal). I then turn to the intensive margin, that is, the amount of tonnage produced conditional on being active. The first set of results are obtained from cross-sectional regressions focused on the 1901–1910 period, after the input price differences had largely disappeared. Later, I also consider the timing of when protection mattered using the full panel of data. To study the extensive margin, I apply multinomial logit (ML) regressions. The specification is $$\begin{align} A_{ls} &= 1[a^*_{ls} > 0], \\ \nonumber a^*_{ls} & = \alpha _1 \mathit{LAKES}_{l} + \alpha _2 {\mathit U\!Scoast}_{l} + X_{js} \Gamma + e_{ls},\end{align}$$ (1) where Als is an indicator variable for whether location l was active in shipbuilding sector s ∈ {wood, metal, both} in the 1901–1910 decade (with inactive as the reference category), and |$a^*_{ls}$| is an unobserved latent variable that depends on the set of explanatory variables. LAKESl is an indicator variable for whether the location is in the Great Lakes region whereas UScoastl is an indicator for whether the location is on the Atlantic coast of the United States. I treat these two areas separately because they experienced varying levels of protection from British competition. The reference region is Atlantic Canada, which was fully exposed to foreign competition. The error term els follows a logistic distribution. Among the control variables that I consider is whether a location had been active in shipbuilding in some past decade (typically 1871–1880, which avoids the decade of the U.S. Civil War but predates the input price convergence) at all, or in sector s specifically, and if so, the tonnage produced in that past decade in the location overall or in sector s specifically.32 These controls help capture a location’s physical assets for ship production such as a deep harbor or easier access to inputs. In some specifications I also control for shipbuilding in other nearby locations, county-level employment in other metal industries and lumber mills, county population, and state-level iron and wood input prices.33 To help reduce potential endogeneity concerns with the input price controls, I use prices for products in a relatively raw state (e.g., pig iron and generic lumber) that are used as inputs in a wide variety of goods as well as shipbuilding, rather than inputs more directly related to shipbuilding (e.g., steel plates).34 The fact that shipbuilding was only one of many uses for these raw materials should limit endogeneity concerns. One potential identification concern in this study, as well as other studies using a similar identification strategy, is that there could be some other time-varying regional shock to treated locations, such as those in the Great Lakes, that is not captured by the available control variables. In this study, the availability of two sources of plausibly exogenous variation—comparing the Great Lakes to the Atlantic and the United States to Canada—provides some protection against such concerns. One may also worry about spatial correlation in my regressions. To examine this possibility, I include in all of the regression results (in square brackets) standard errors that are clustered by the 31 major shipbuilding areas during this period.35 These clustered standard errors are presented in square brackets in the results tables in addition to the robust standard errors in parenthesis. Clustered standard errors are often similar to or slightly smaller than the robust standard errors, particularly in my preferred specifications, so I focus on the robust standard errors when indicating statistical significance (*). Table 2 presents ML regression results based on equation (1). These regressions are run on the full set of U.S. and Canadian shipbuilding locations on the East Coast or Great Lakes that were active at some point in the 1850–1910 period.36 Column (1) presents results without any additional controls while columns (2) and (3) add in additional controls for activity in the location in the 1871–1880 decade, county level population and industry composition, and past production in nearby locations.37 The results in columns (1)–(3) suggest that locations in the Great Lakes were more likely to be active in the production of metal ships, either alone or in combination with wood shipbuilding, relative to exiting the market. There is also some evidence that coastal locations in the United States were more likely to remain active, but this result does not remain significant as controls are added. Table 2. Multinomial logit regression results. . (1) . (2) . (3) . (4) . (5) . A = 1: Location active in wood shipbuilding only in 1901–1910 U.S. Coastal Locations −0.082 0.009 0.268 (0.209) (0.228) (0.412) [0.379] [0.402] [0.445] Great Lakes Locations 0.324 0.603 0.542 0.282 0.241 (0.382) (0.418) (0.460) (0.610) (0.649) [0.371] [0.378] [0.429] [0.487] [0.560] A = 2: Location active in metal shipbuilding only in 1901–1910 U.S. Coastal Locations 0.630 1.049 0.316 (0.712) (0.902) (1.064) [0.834] [0.818] [0.899] Great Lakes Locations 2.991*** 1.671* 1.351 1.479* 1.790* (0.697) (0.848) (0.941) (0.737) (0.786) [0.756] [0.745] [0.746] [0.830] [0.866] A = 3: Location active in both wood and metal shipbuilding in 1901–1910 U.S. Coastal Locations 1.546* 2.554** 0.771 (0.637) (0.842) (1.151) [0.536] [1.016] [1.383] Great Lakes Locations 2.991*** 4.655*** 3.265** 3.423*** 2.873** (0.697) (0.941) (1.139) (0.815) (0.915) [0.597] [1.084] [1.418] [0.622] [0.764] Observations 833 833 779 274 274 Testing effect on A = 2 different from A = 1 (p-values, robust SEs) United States 0.3315 0.2599 0.9801 Great Lakes 0.0004 0.2420 0.4264 0.1832 0.1074 Testing effect on A = 3 different from A = 1 (p-values, robust SEs) United States 0.0138 0.0030 0.6867 Great Lakes 0.0004 0.0000 0.0224 0.0009 0.0107 . (1) . (2) . (3) . (4) . (5) . A = 1: Location active in wood shipbuilding only in 1901–1910 U.S. Coastal Locations −0.082 0.009 0.268 (0.209) (0.228) (0.412) [0.379] [0.402] [0.445] Great Lakes Locations 0.324 0.603 0.542 0.282 0.241 (0.382) (0.418) (0.460) (0.610) (0.649) [0.371] [0.378] [0.429] [0.487] [0.560] A = 2: Location active in metal shipbuilding only in 1901–1910 U.S. Coastal Locations 0.630 1.049 0.316 (0.712) (0.902) (1.064) [0.834] [0.818] [0.899] Great Lakes Locations 2.991*** 1.671* 1.351 1.479* 1.790* (0.697) (0.848) (0.941) (0.737) (0.786) [0.756] [0.745] [0.746] [0.830] [0.866] A = 3: Location active in both wood and metal shipbuilding in 1901–1910 U.S. Coastal Locations 1.546* 2.554** 0.771 (0.637) (0.842) (1.151) [0.536] [1.016] [1.383] Great Lakes Locations 2.991*** 4.655*** 3.265** 3.423*** 2.873** (0.697) (0.941) (1.139) (0.815) (0.915) [0.597] [1.084] [1.418] [0.622] [0.764] Observations 833 833 779 274 274 Testing effect on A = 2 different from A = 1 (p-values, robust SEs) United States 0.3315 0.2599 0.9801 Great Lakes 0.0004 0.2420 0.4264 0.1832 0.1074 Testing effect on A = 3 different from A = 1 (p-values, robust SEs) United States 0.0138 0.0030 0.6867 Great Lakes 0.0004 0.0000 0.0224 0.0009 0.0107 Notes: Standard errors in brackets are clustered on shipbuilding area. The analysis covers all locations active in shipbuilding from 1850 to 1910 in the Atlantic Coast or Great Lakes regions of the United States and Canada. Column (2) includes controls for whether a location is active in metal or wood shipbuilding in 1870 as well as separate variables for tonnage produced in metal or wood in 1870. Column (3) adds additional controls for metal or wood shipbuilding at other locations within 100 km, county log population, the county employment share in metalworking industries, and the employment share in lumber. Note that the county data are not available for some locations. Column (4) includes the controls in column (2) together with the log price of pig iron and log lumber index price in the state. These are only available for a subset of U.S. states, so the number of observations drops substantially. Column (5) includes the controls in column (4) together with controls for county log population, the county employment share in metalworking industries, and the employment share in lumber. Tests of coefficient differences use robust SEs. Based on robust SEs, shown in parentheses: *p < 0.1; **p < 0.05; ***p < 0.01. Open in new tab Table 2. Multinomial logit regression results. . (1) . (2) . (3) . (4) . (5) . A = 1: Location active in wood shipbuilding only in 1901–1910 U.S. Coastal Locations −0.082 0.009 0.268 (0.209) (0.228) (0.412) [0.379] [0.402] [0.445] Great Lakes Locations 0.324 0.603 0.542 0.282 0.241 (0.382) (0.418) (0.460) (0.610) (0.649) [0.371] [0.378] [0.429] [0.487] [0.560] A = 2: Location active in metal shipbuilding only in 1901–1910 U.S. Coastal Locations 0.630 1.049 0.316 (0.712) (0.902) (1.064) [0.834] [0.818] [0.899] Great Lakes Locations 2.991*** 1.671* 1.351 1.479* 1.790* (0.697) (0.848) (0.941) (0.737) (0.786) [0.756] [0.745] [0.746] [0.830] [0.866] A = 3: Location active in both wood and metal shipbuilding in 1901–1910 U.S. Coastal Locations 1.546* 2.554** 0.771 (0.637) (0.842) (1.151) [0.536] [1.016] [1.383] Great Lakes Locations 2.991*** 4.655*** 3.265** 3.423*** 2.873** (0.697) (0.941) (1.139) (0.815) (0.915) [0.597] [1.084] [1.418] [0.622] [0.764] Observations 833 833 779 274 274 Testing effect on A = 2 different from A = 1 (p-values, robust SEs) United States 0.3315 0.2599 0.9801 Great Lakes 0.0004 0.2420 0.4264 0.1832 0.1074 Testing effect on A = 3 different from A = 1 (p-values, robust SEs) United States 0.0138 0.0030 0.6867 Great Lakes 0.0004 0.0000 0.0224 0.0009 0.0107 . (1) . (2) . (3) . (4) . (5) . A = 1: Location active in wood shipbuilding only in 1901–1910 U.S. Coastal Locations −0.082 0.009 0.268 (0.209) (0.228) (0.412) [0.379] [0.402] [0.445] Great Lakes Locations 0.324 0.603 0.542 0.282 0.241 (0.382) (0.418) (0.460) (0.610) (0.649) [0.371] [0.378] [0.429] [0.487] [0.560] A = 2: Location active in metal shipbuilding only in 1901–1910 U.S. Coastal Locations 0.630 1.049 0.316 (0.712) (0.902) (1.064) [0.834] [0.818] [0.899] Great Lakes Locations 2.991*** 1.671* 1.351 1.479* 1.790* (0.697) (0.848) (0.941) (0.737) (0.786) [0.756] [0.745] [0.746] [0.830] [0.866] A = 3: Location active in both wood and metal shipbuilding in 1901–1910 U.S. Coastal Locations 1.546* 2.554** 0.771 (0.637) (0.842) (1.151) [0.536] [1.016] [1.383] Great Lakes Locations 2.991*** 4.655*** 3.265** 3.423*** 2.873** (0.697) (0.941) (1.139) (0.815) (0.915) [0.597] [1.084] [1.418] [0.622] [0.764] Observations 833 833 779 274 274 Testing effect on A = 2 different from A = 1 (p-values, robust SEs) United States 0.3315 0.2599 0.9801 Great Lakes 0.0004 0.2420 0.4264 0.1832 0.1074 Testing effect on A = 3 different from A = 1 (p-values, robust SEs) United States 0.0138 0.0030 0.6867 Great Lakes 0.0004 0.0000 0.0224 0.0009 0.0107 Notes: Standard errors in brackets are clustered on shipbuilding area. The analysis covers all locations active in shipbuilding from 1850 to 1910 in the Atlantic Coast or Great Lakes regions of the United States and Canada. Column (2) includes controls for whether a location is active in metal or wood shipbuilding in 1870 as well as separate variables for tonnage produced in metal or wood in 1870. Column (3) adds additional controls for metal or wood shipbuilding at other locations within 100 km, county log population, the county employment share in metalworking industries, and the employment share in lumber. Note that the county data are not available for some locations. Column (4) includes the controls in column (2) together with the log price of pig iron and log lumber index price in the state. These are only available for a subset of U.S. states, so the number of observations drops substantially. Column (5) includes the controls in column (4) together with controls for county log population, the county employment share in metalworking industries, and the employment share in lumber. Tests of coefficient differences use robust SEs. Based on robust SEs, shown in parentheses: *p < 0.1; **p < 0.05; ***p < 0.01. Open in new tab It is worth noting that adding in controls for previous production in columns (2) and (3) affects the interpretation of the results. Without controlling for past production patterns, the estimates in column (1) should capture the impact of both current protection from foreign competition as well as the effect of protection in the past operating through learning effects. Adding in past production patterns helps control for locational advantages in a particular type of shipbuilding, but these controls will also soak up some of the effect of past protection operating through learning. Since I am primarily interested in the impact of protection in the period after which the gap between British and North American input prices had narrowed, my preferred results are those that include controls for production patterns in the 1870s. In columns (4) and (5) I include additional controls for state-level iron and lumber prices. Note that these data are available only for the United States, which means that fewer observations are available for these regressions and I cannot compare the U.S. coast to Canada. Despite the smaller sample size, I still tend to find evidence that locations in the Great Lakes were more likely to be active in metal shipbuilding than those on the coast. It is worth noting that these results are identifying the effect of the additional protection provided by being in the Great Lakes (and in the United States) compared to being in the United States but on the coast. At the bottom of the table I include additional tests comparing the probability of being active in metal shipbuilding or in both sectors to the probability of being active in wood shipbuilding alone. These tests are important because comparing metal to wood ship production in the Great Lakes helps me deal with concerns that the results are just reflecting more rapid growth in shipbuilding in the Great Lakes overall. In general, the effect of the Great Lakes on whether a location is active in metal (in combination with wood) is statistically different from the impact of the Great Lakes on activity in wood only. The results in Table 2 are consistent with the idea that North American shipbuilders that were not exposed to British competition were able to rapidly switch to metal shipbuilding once metal input prices fell. This suggests that it was exposure to initially advantaged British producers, rather than other factors, that were likely behind the inability of Coastal North America shipbuilders to catch up to their British competitors after 1900. Two additional sets of ML results are available in Online Appendix A.15. The first considers both the ship’s construction material and power source (sail vs. steam). These results show that differences in metal ship production between the Lakes and the Coast were not driven by differences in demand for sailing versus steamships. The second set of results treats the U.S. and Canadian areas of the Great Lakes regions separately and shows that both areas exhibit fairly similar patterns, though the relatively small number of Canadian Great Lakes shipyards means results for that group are imprecise. Next, I study the intensive margin of production, that is, how much tonnage a shipyard produced from 1901 to 1910 in a sector conditional on being active in that sector. I use $$\begin{eqnarray} \ln (Y_{ls}) & =& \beta _0 {\mathit METAL}_{s} {+} \beta _1 \mathit{LAKES}_{l} {+} \beta _3 {U\!Scoast}_{l} {+} \beta _4 ( {\mathit METAL}_{s} {\times} \mathit{LAKES}_{l}) \nonumber\\ && +\, \beta _5 ({\mathit METAL}_{s} {\times} {UScoast}_{l}) {+} X_{js} \Gamma {+} \epsilon _{js},\end{eqnarray}$$ (2) where Yls is ship tonnage of type s produced in location l, METALs is an indicator for the metal ship sector, and the remaining variables are defined as before. The main coefficients of interest in this regression are β4 and β5 that reflect the impact of being in the Great Lakes market or in the United States, respectively, on metal ship output relative to wood. I use log tonnage as the dependent variable in these regressions, but similar results are obtained if instead I use the level of tonnage (Online Appendix Table A.11). This tells us that the results are not being driven by the fact that the log specification places more weight on smaller observations. Table 3 presents the results of regressions based on equation (2). Column (1) presents baseline results while columns (2) and (3) add in additional controls following the same format as in Table 2. Columns (4) and (5) present results including state-level price controls and using only observations from the United States.38 These results suggest that, conditional on a location being active in a particular sector, tonnage of metal ship production was higher in locations in the Great Lakes region and in the Coastal United States compared to Coastal Canada. The magnitudes of these effects are large; being in the Great Lakes is associated with an increase in tonnage of 4-5 log points relative to Coastal Canada and about 2 log points relative to the Coastal United States (columns (4) and (5)). Being in the Coastal United States is associated with a tonnage increase of about 2 log points relative to Coastal Canada.39 Additional results, in Online Appendix A.16, show that these patterns are being driven entirely by steamships. Moreover, the impact of the Great Lakes and the U.S. markets continues to hold when we look only within steamships, so these effects are not being driven by a different mix of steamships versus sailing ships in different markets. Table 3. Tonnage regression results. DV: Log of tons in 1901–1910 by location and material . . (1) . (2) . (3) . (4) . (5) . Great Lakes Locations × Metal 5.174*** 4.802*** 4.703*** 2.522*** 2.547*** (0.731) (0.798) (0.811) (0.895) (0.887) [0.775] [0.752] [0.738] [0.691] [0.729] U.S. Coastal Locations × Metal 2.467*** 2.204*** 2.396*** (0.700) (0.714) (0.782) [0.942] [0.694] [0.705] Metal indicator Yes Yes Yes Yes Yes U.S. Coast ind. Yes Yes Yes Great Lakes ind. Yes Yes Yes Yes Yes Activity in 1871 Yes Yes Yes Yes Tonnage in 1871 Yes Yes Yes Yes Nearby tons in 1871 Yes County controls Yes Yes Input prices Yes Yes Observations 186 186 182 82 82 R-squared 0.427 0.516 0.551 0.620 0.640 DV: Log of tons in 1901–1910 by location and material . . (1) . (2) . (3) . (4) . (5) . Great Lakes Locations × Metal 5.174*** 4.802*** 4.703*** 2.522*** 2.547*** (0.731) (0.798) (0.811) (0.895) (0.887) [0.775] [0.752] [0.738] [0.691] [0.729] U.S. Coastal Locations × Metal 2.467*** 2.204*** 2.396*** (0.700) (0.714) (0.782) [0.942] [0.694] [0.705] Metal indicator Yes Yes Yes Yes Yes U.S. Coast ind. Yes Yes Yes Great Lakes ind. Yes Yes Yes Yes Yes Activity in 1871 Yes Yes Yes Yes Tonnage in 1871 Yes Yes Yes Yes Nearby tons in 1871 Yes County controls Yes Yes Input prices Yes Yes Observations 186 186 182 82 82 R-squared 0.427 0.516 0.551 0.620 0.640 Notes: Standard errors in brackets are clustered on shipbuilding area. Regressions are run only on sector locations that were active in 1901–1910. Column (2) includes controls for whether a location is active in metal or wood shipbuilding in 1870 as well as separate variables for tonnage produced in metal or wood in 1870. Column (3) adds additional controls for metal or wood shipbuilding at other locations within 100 km, county log population, the county employment share in metalworking industries, and the employment share in lumber. Note that the county data are not available for some locations. Column (4) includes the controls in column (2) together with the log price of pig iron and log lumber index price in the state. These are only available for a subset of U.S. states, so the number of observations drops substantially. Column (5) includes the controls in column (4) together with controls for county log population, the county employment share in metalworking and the share in lumber. Based on SEs clustered by location, shown in parentheses: ***p < 0.01. Open in new tab Table 3. Tonnage regression results. DV: Log of tons in 1901–1910 by location and material . . (1) . (2) . (3) . (4) . (5) . Great Lakes Locations × Metal 5.174*** 4.802*** 4.703*** 2.522*** 2.547*** (0.731) (0.798) (0.811) (0.895) (0.887) [0.775] [0.752] [0.738] [0.691] [0.729] U.S. Coastal Locations × Metal 2.467*** 2.204*** 2.396*** (0.700) (0.714) (0.782) [0.942] [0.694] [0.705] Metal indicator Yes Yes Yes Yes Yes U.S. Coast ind. Yes Yes Yes Great Lakes ind. Yes Yes Yes Yes Yes Activity in 1871 Yes Yes Yes Yes Tonnage in 1871 Yes Yes Yes Yes Nearby tons in 1871 Yes County controls Yes Yes Input prices Yes Yes Observations 186 186 182 82 82 R-squared 0.427 0.516 0.551 0.620 0.640 DV: Log of tons in 1901–1910 by location and material . . (1) . (2) . (3) . (4) . (5) . Great Lakes Locations × Metal 5.174*** 4.802*** 4.703*** 2.522*** 2.547*** (0.731) (0.798) (0.811) (0.895) (0.887) [0.775] [0.752] [0.738] [0.691] [0.729] U.S. Coastal Locations × Metal 2.467*** 2.204*** 2.396*** (0.700) (0.714) (0.782) [0.942] [0.694] [0.705] Metal indicator Yes Yes Yes Yes Yes U.S. Coast ind. Yes Yes Yes Great Lakes ind. Yes Yes Yes Yes Yes Activity in 1871 Yes Yes Yes Yes Tonnage in 1871 Yes Yes Yes Yes Nearby tons in 1871 Yes County controls Yes Yes Input prices Yes Yes Observations 186 186 182 82 82 R-squared 0.427 0.516 0.551 0.620 0.640 Notes: Standard errors in brackets are clustered on shipbuilding area. Regressions are run only on sector locations that were active in 1901–1910. Column (2) includes controls for whether a location is active in metal or wood shipbuilding in 1870 as well as separate variables for tonnage produced in metal or wood in 1870. Column (3) adds additional controls for metal or wood shipbuilding at other locations within 100 km, county log population, the county employment share in metalworking industries, and the employment share in lumber. Note that the county data are not available for some locations. Column (4) includes the controls in column (2) together with the log price of pig iron and log lumber index price in the state. These are only available for a subset of U.S. states, so the number of observations drops substantially. Column (5) includes the controls in column (4) together with controls for county log population, the county employment share in metalworking and the share in lumber. Based on SEs clustered by location, shown in parentheses: ***p < 0.01. Open in new tab In Online Appendix A.16, I look at whether similar results to those shown in Tables 2 and 3 are obtained if we look at the impact of being in the Great Lakes within only the United States or only Canada, or the impact of being in the United States in only the Lakes or only the Atlantic. I find that locations in the Great Lakes produce more metal ship tonnage in 1901–1910 in both the United States and Canada, but the protection afforded by the Lakes is more important for Canadian shipbuilders. Focusing only on the Atlantic Coast, I find evidence that shipyards in the U.nited States produced more metal ship tonnage, whereas I find no strong evidence that being in the United States mattered in the protected Lakes market (though the small number of Canadian shipyards on the Great Lakes means that these results should be interpreted with some caution). Next, I look at the timing of the effects using the full panel of data, focusing on the intensive margin of production. The specification is $$\begin{eqnarray} Y_{lst} & =& \sum _t \beta _{0t} ({\mathit METAL}_{s}\times D_t) + \sum _t \beta _{1t} (\mathit{LAKES}_{l}\times {\mathit METAL}_s \times D_t) \nonumber\\ &&+ \sum _t \beta _{2t} (\mathit{LAKES}_{l}\times {\mathit WOOD}_s \times D_t) + \sum _t \beta _{3t} (US_{l}\times {\mathit METAL}_S \times D_t)\nonumber\\ &&+ \sum _t \beta _{4t} (US_{l}\times {\mathit METAL}_S \times D_t) + X_{jst} \Gamma + \sum _t \eta _t D_t + \phi _{ls} + \epsilon _{js},\end{eqnarray}$$ (3) where Ylst is ship tonnage (in 10,000s), |${\mathit WOOD}_s$| is an indicator variable for the wood shipbuilding sector, Dt is a set of indicator variables for each decade, and φls is a set of fixed effects for each sector-location. These regressions allow me to look at the impact of being in the Great Lakes or in the United States on iron ship output while controlling for changes in output over time as well as differences in regional production patterns over time. Because of concerns about serial correlation in these regressions, standard errors are clustered by location. I focus on tonnage rather than log tons in this specification to avoid dropping observations for locations that were inactive (produced zero tons) in at least some decades. The coefficients of interest in equation (3) are the vectors β1t − β4t, which reflect the impact of being in the Great Lakes or being in the United States in each decade within each ship type. These estimates, together with 95% confidence intervals, are described in Figure 6. Figure 6. Open in new tabDownload slide Panel data regression results. Estimates based on decadal data from 1850–1910. Figures show coefficients for the interaction of an indicator for metal shipbuilding with an indicator for the Great Lakes (top panel) or the United States (bottom panel) on tons produced (in 10,000s) and similar coefficients for interactions using an indicator for wood shipbuilding. Regressions include decade effects and a full set of location-by-sector fixed effects. Ninety-five percent confidence intervals based on standard errors clustered by location. Figure 6. Open in new tabDownload slide Panel data regression results. Estimates based on decadal data from 1850–1910. Figures show coefficients for the interaction of an indicator for metal shipbuilding with an indicator for the Great Lakes (top panel) or the United States (bottom panel) on tons produced (in 10,000s) and similar coefficients for interactions using an indicator for wood shipbuilding. Regressions include decade effects and a full set of location-by-sector fixed effects. Ninety-five percent confidence intervals based on standard errors clustered by location. The top panel of Figure 6 shows the coefficients estimated for each decade on the interaction between the Great Lakes and either metal or wood shipbuilding. These results suggest that being located in the Great Lakes was, if anything, associated with lower production tonnage prior to the 1880s. Then, starting in the 1890s, there was a relative increase in tonnage produced on the Great Lakes that was concentrated in metal shipbuilding.40 This timing corresponds with the fall in U.S. iron and steel prices as well as an increase in demand for Great Lakes shipping. The bottom panel shows that starting in the 1890s metal shipbuilding experiencing a relative increase in the Coastal U.S. compared to Coastal Canada. This timing corresponds to the fall in metal prices and the expansion of the U.S. Navy. One potential concern with this analysis is that there may have been other initial differences that gave some locations an advantage in metal ship production relative to others. One way to provide some additional evidence on this point is to focus on a set of very similar locations that differed in their exposure to competition. A natural candidate is to compare shipyards in Maine to those in Nova Scotia and New Brunswick. These three areas were all major wood shipbuilding centers in the late 19th century and none of these locations had meaningful metal ship production prior to the 1880s. Their economies were also similar in a number of other respects (see details in Online Appendix A.17). However, only shipyards in Maine enjoyed protection from British competition, whereas those in nearby Nova Scotia and New Brunswick were not protected. Figure 7 compares the evolution of wood and metal shipbuilding in these similar locations. Both areas show similar patterns of wood ship production, with output peaking in the 1860s and then declining starting in the 1870s, though the decline in wood ship output was not as steep in Maine due to the protection afforded by U.S. policies. The key feature in this graph is the pattern of metal ship production. Despite the initial similarity of these areas, only in Maine do we see the emergence of any substantial metal ship production. The fact that the patterns identified in the broader statistical analysis also emerge when focusing only on these initially very similar shipbuilding areas provides confidence that the broader results are not being driven by initial differences. Instead, a location's chances of successfully switching from wood to metal ship production appears to be closely linked to the extent to which the location was exposed to foreign competition. Figure 7. Open in new tabDownload slide Evolution of shipbuilding in Maine versus Nova Scotia/New Brunswick. Figure compares wood tonnage (left axis) and metal tonnage (right axis) produced in Maine (USA) to tonnage produced in Nova Scotia and New Brunswick (Canada). Figure 7. Open in new tabDownload slide Evolution of shipbuilding in Maine versus Nova Scotia/New Brunswick. Figure compares wood tonnage (left axis) and metal tonnage (right axis) produced in Maine (USA) to tonnage produced in Nova Scotia and New Brunswick (Canada). As a final piece of this analysis, it is useful to consider the key margins for competition between British, U.S., and Canadian shipbuilders. The data make it clear that competition between British and U.S. shipbuilders was not mainly over sales of vessels to shipping companies based in those countries. Most (91%) the vessels homeported in the United States were built there (7.6% of tonnage homeported on the U.S. Atlantic Coast was built in Britain), and essentially all of the tonnage homeported in Britain in 1912 was also produced there.41 Instead, the key margin of competition was the market for vessels operated by firms in other countries that served the U.S. market. We can study this group by looking at vessels registered in the U.S.—presumably because they were serving U.S. ports—but homeported outside of the United States, Canada, and Britain. This was a large market because there were many countries where the needs of shipping firms substantially exceeded the capacity of local shipbuilding (e.g., Greece, Italy, Brazil, the Netherlands).42 Table 4 uses data on homeport location digitized from the 1912 ABS registry to identify the construction location of ships homeported outside of the United States, United Kingdom, and Canada. Since these data come from the ABS—a U.S. registry—it is likely that at some point these ships were active in serving North American ports, so if anything we would expect U.S. shipbuilders to have some advantage relative to British producers. Despite this, the data show that British producers were dominant in supplying ships based outside of the United States, United Kingdom, and Canada. Moreover, there is little change in this pattern when British colonies are excluded. Nor does the pattern appear to be driven by ships that were re-sold by their original owners; if I focus only on ships built recently, which are likely to still be held by their original owner, there is little change in the results. Thus, British dominance was not linked mainly to the size of their own merchant marine relative to American merchant marine, but rather to their ability to dominate sales to shipping firms based in other countries that served the U.S. market. Table 4. Source of ships in the U.S. registry homeported outside of the United Kingdom, United States, and Canada. . All ships in 1912 ABS . Newer ships in 1912 ABS . . Ships homeported . Ships homeported . Country of . All foreign . Foreign locations . All foreign . Foreign locations . construction . locations (%) . except UK colonies (%) . locations (%) . except UK colonies (%) . United Kingdom 55.65 54.95 48.57 47.54 United States 0.51 0.52 0.09 0.09 Canada 0.09 0.07 0.02 0.00 All others 43.74 44.46 51.32 52.37 . All ships in 1912 ABS . Newer ships in 1912 ABS . . Ships homeported . Ships homeported . Country of . All foreign . Foreign locations . All foreign . Foreign locations . construction . locations (%) . except UK colonies (%) . locations (%) . except UK colonies (%) . United Kingdom 55.65 54.95 48.57 47.54 United States 0.51 0.52 0.09 0.09 Canada 0.09 0.07 0.02 0.00 All others 43.74 44.46 51.32 52.37 Notes: Data from the 1912 ABS registry. Newer ships are those built after 1904. Note that the increase in the “All others” share when focusing only on newer ships suggest that other countries were gaining market share during this period. These gains were driven primarily by Germany, where the government was subsidizing shipbuilding during this period, as well as by increases in Japan and Italy. Open in new tab Table 4. Source of ships in the U.S. registry homeported outside of the United Kingdom, United States, and Canada. . All ships in 1912 ABS . Newer ships in 1912 ABS . . Ships homeported . Ships homeported . Country of . All foreign . Foreign locations . All foreign . Foreign locations . construction . locations (%) . except UK colonies (%) . locations (%) . except UK colonies (%) . United Kingdom 55.65 54.95 48.57 47.54 United States 0.51 0.52 0.09 0.09 Canada 0.09 0.07 0.02 0.00 All others 43.74 44.46 51.32 52.37 . All ships in 1912 ABS . Newer ships in 1912 ABS . . Ships homeported . Ships homeported . Country of . All foreign . Foreign locations . All foreign . Foreign locations . construction . locations (%) . except UK colonies (%) . locations (%) . except UK colonies (%) . United Kingdom 55.65 54.95 48.57 47.54 United States 0.51 0.52 0.09 0.09 Canada 0.09 0.07 0.02 0.00 All others 43.74 44.46 51.32 52.37 Notes: Data from the 1912 ABS registry. Newer ships are those built after 1904. Note that the increase in the “All others” share when focusing only on newer ships suggest that other countries were gaining market share during this period. These gains were driven primarily by Germany, where the government was subsidizing shipbuilding during this period, as well as by increases in Japan and Italy. Open in new tab The same pattern does not hold for Canada, where there was no restriction on Canadian shipping firms using British-built ships. There, the data show that even a large fraction of vessels homeported in Canada had been built in Britain (see Online Appendix Table A.4). Thus, the key margin for competition between Canadian and British shipbuilders was not the third party market, but instead sales to Canadian shipping firms. This reflects the key difference between the protection afforded U.S. shipbuilders and the lack of protection available to Canadian shipyards. The main conclusion to draw from this section is that exposure to competition from initially advantaged British producers substantially retarded the ability of North American shipbuilders to transition to metal ship production. These econometric results are reinforced by historical evidence. In 1897, for example, the Baltimore Journal of Commerce wrote of the U.S. shipbuilding industry, “on the lakes, where it receives the most effective protection, the ship building industry enjoys its highest prosperity and reaches its most splendid proportions; whereas on the ocean, where it has no protection at all, it is gradually falling into decay under the aggressive competition of more enterprising nations”.43 Whether the transition from wood to metal was made determined the ultimate success of the industry in each location as wood shipbuilding disappeared in the early 20th century. 5. Evidence of Learning The results in the previous section suggest that North American shipyards were unable to compete with British producers even after Britain’s initial advantage in input prices had disappeared. This tells us that British producers enjoyed some persistent productivity advantage. One explanation for this pattern is that the shipbuilding industry may be characterized by dynamic learning effects, so that current productivity is increasing in previous production experience. Such effects would explain why Britain’s initial lead meant that, later on, North American shipyards exposed to British competition had trouble entering metal shipbuilding. Learning can take many forms, but these forms can be roughly divided into those types that are internal to firms and those that are external. Organizational improvements resulting from experience, for example, represents an internal form of learning, whereas the acquisition of skills by workers is external to firms, since workers can switch jobs. Distinguishing between these forms can therefore shed light on the types of mechanisms likely to be behind the broad patterns documented in the previous section. There is already existing evidence suggesting that the shipbuilding industry was characterized by learning effects (e.g., Thompson 2001; Thornton and Thompson 2001). However, this existing evidence does not distinguish whether the learning was internal or external to an individual yard, or whether external effects were localized.44 Also, because these results come from wartime shipyards, which sought to rapidly produce many ships with a common design, it is unclear the extent to which these results carry over to peacetime yards, which rarely produced more than a couple ships of a given type. Thus, more evidence is needed to understand the nature of learning in this industry. The section uses the location of U.S. Navy shipyards to provide evidence that learning was important in the shipbuilding industry, and that at least part of this learning was external to individual shipyards. Proximity to U.S. Navy Shipyards could benefit nearby private-sector shipyards through technology spillovers or by providing access to pools of skilled metal shipbuilders.45 To identify these effects, I take advantage of the fact that the location of Navy shipyards was plausibly unrelated to a location’s specific advantage in metal, relative to wood, shipbuilding. This is a plausible assumption because the locations of the Navy shipyards in operation during the period that I study (shown in the map in Online Appendix Figure A.15) were all determined around 1800, well before the introduction of metal ships.46 Thus, although Naval shipyards were situated in locations with advantages for shipbuilding overall, there is little reason to believe that they were sited in locations that were particularly advantageous for metal shipbuilding after 1880. Results looking at the impact of proximity to U.S. Navy shipyards are presented in Table 5. These are based on the set of U.S. Atlantic Coast shipyards only.47 The regressions are run using the log tonnage regression specification from equation (2). Columns (1)–(3) present results using all U.S. Atlantic coast locations. All of the results suggest that close proximity to a Navy shipyard—within 50 km—has a positive relationship to the tonnage of metal ships produced. The impact of proximity to Navy shipyards on wood shipbuilding tends to be negative, suggesting that private shipyards near the Navy yards were more likely to switch from wood to metal ship construction, or that metal shipbuilding pushed wooden shipbuilding out of these locations. Column (4) shows that these effects were localized and disappear beyond 50km. Table 5. Results looking at the impact of proximity to U.S. Navy Shipyards. DV: Log of tons in 1901–1910 by location and material . . (1) . (2) . (3) . (4) . Navy yard within 50 km × Metal 2.761** 2.381** 2.390** 3.093** (1.044) (1.006) (1.051) (1.378) [0.955] [0.790] [0.790] [1.385] Navy yard within 50 km −1.178** −1.280*** −1.344*** −0.737 (0.473) (0.370) (0.527) (0.711) [0.504] [0.286] [0.286] [0.358] Navy yard within 100 km × Metal −0.806 (1.252) [1.449] Navy yard within 100 km −1.065 (0.717) [0.693] Metal ind. Yes Yes Yes Yes Controls for Yes Yes Yes active in 1870s County controls Yes Yes Observations 89 89 89 89 R-squared 0.281 0.463 0.475 0.504 DV: Log of tons in 1901–1910 by location and material . . (1) . (2) . (3) . (4) . Navy yard within 50 km × Metal 2.761** 2.381** 2.390** 3.093** (1.044) (1.006) (1.051) (1.378) [0.955] [0.790] [0.790] [1.385] Navy yard within 50 km −1.178** −1.280*** −1.344*** −0.737 (0.473) (0.370) (0.527) (0.711) [0.504] [0.286] [0.286] [0.358] Navy yard within 100 km × Metal −0.806 (1.252) [1.449] Navy yard within 100 km −1.065 (0.717) [0.693] Metal ind. Yes Yes Yes Yes Controls for Yes Yes Yes active in 1870s County controls Yes Yes Observations 89 89 89 89 R-squared 0.281 0.463 0.475 0.504 Notes: Standard errors in brackets are clustered on shipbuilding area. Regressions are run on data from U.S. Atlantic Coast locations only. All regressions include controls for whether the sector was metal. The regressions in columns (2)–(4) also include controls for whether a location was active in 1871–1880, whether it was active in the same sector in 1871–1880, total tonnage produced in the location in 1871–1880, and tonnage produced in the same location and sector in 1871–1880. Regressions in columns (3) and (4) include county-level controls for log population, metalworking employment share and lumber milling employment share. Based on SEs clustered by location, shown in parentheses: **p < 0.05; ***p < 0.01. Open in new tab Table 5. Results looking at the impact of proximity to U.S. Navy Shipyards. DV: Log of tons in 1901–1910 by location and material . . (1) . (2) . (3) . (4) . Navy yard within 50 km × Metal 2.761** 2.381** 2.390** 3.093** (1.044) (1.006) (1.051) (1.378) [0.955] [0.790] [0.790] [1.385] Navy yard within 50 km −1.178** −1.280*** −1.344*** −0.737 (0.473) (0.370) (0.527) (0.711) [0.504] [0.286] [0.286] [0.358] Navy yard within 100 km × Metal −0.806 (1.252) [1.449] Navy yard within 100 km −1.065 (0.717) [0.693] Metal ind. Yes Yes Yes Yes Controls for Yes Yes Yes active in 1870s County controls Yes Yes Observations 89 89 89 89 R-squared 0.281 0.463 0.475 0.504 DV: Log of tons in 1901–1910 by location and material . . (1) . (2) . (3) . (4) . Navy yard within 50 km × Metal 2.761** 2.381** 2.390** 3.093** (1.044) (1.006) (1.051) (1.378) [0.955] [0.790] [0.790] [1.385] Navy yard within 50 km −1.178** −1.280*** −1.344*** −0.737 (0.473) (0.370) (0.527) (0.711) [0.504] [0.286] [0.286] [0.358] Navy yard within 100 km × Metal −0.806 (1.252) [1.449] Navy yard within 100 km −1.065 (0.717) [0.693] Metal ind. Yes Yes Yes Yes Controls for Yes Yes Yes active in 1870s County controls Yes Yes Observations 89 89 89 89 R-squared 0.281 0.463 0.475 0.504 Notes: Standard errors in brackets are clustered on shipbuilding area. Regressions are run on data from U.S. Atlantic Coast locations only. All regressions include controls for whether the sector was metal. The regressions in columns (2)–(4) also include controls for whether a location was active in 1871–1880, whether it was active in the same sector in 1871–1880, total tonnage produced in the location in 1871–1880, and tonnage produced in the same location and sector in 1871–1880. Regressions in columns (3) and (4) include county-level controls for log population, metalworking employment share and lumber milling employment share. Based on SEs clustered by location, shown in parentheses: **p < 0.05; ***p < 0.01. Open in new tab In Online Appendix A.18 I show that these basic results are robust to using a variety of alternative samples or control variables. For example, similar results are obtained if I focus only on locations that were active in wood shipbuilding in 1870s. This is important because it reduces the possibility that the results could be do merely to selection, that is, that more productive metal shipyards may have chosen to locate closer to Navy Shipyards. One may also worry that locations near Navy shipyards had better infrastructure connections. Naturally, all shipyards had good access to water transport, but railroad connections may have varied. I have examined the railroad connections of the U.S. shipyard locations using the data produced by Jeremy Atack (2016). These data show that all of the shipyard locations had railroad connections before 1890, with the exception of a few small island locations in Maine or Massachusetts (see Online Appendix Figure A.8). Since my results are not sensitive to dropping these states from the analysis (see Oline Appendix A.18), differences in infrastructure connections are unlikely to be behind the effects of proximity to Navy shipyards that I estimate. In the Online Appendix, I also present ML results that show that locations near Navy yards were somewhat more likely to be active in the 1901–1910 period but that this effect was not differential for wood versus metal shipbuilding. This provides additional evidence that the main impact of Navy shipyards was on the intensive margin and that selection is not likely to be a driving force behind my results. I also show that the effect of Navy shipyards survives even when controlling for whether a yard received Navy contracts, though the magnitude falls. This suggests that at least a substantial portion of the effect of proximity to Navy shipyards was due to factors other than access to Navy contracts. Finally, I show that my results are not being driven by any one shipbuilding region. To summarize, the results in this section suggest that learning that was external to firms was a salient feature of the shipbuilding industry. This provides some indication of the mechanisms behind how Britain’s initial advantage generated a persistent lead in the shipbuilding industry. In the next section I draw on historical evidence to shed further light on the nature of this learning. 6. A Discussion of the Channels The previous section provides evidence that the shipbuilding industry was characterized by dynamic learning effects that were external to firms. In this section I draw on historical sources to shed some additional (suggestive) light on the specific mechanisms likely to be at work in this context. A good starting point for this exploration is to consider the types of mechanisms suggested by existing theories. These include productivity advantages gained through learning-by-doing in general (Krugman 1987; Young 1991) or learning that was specifically embodied in worker skills (Lucas 1988, 1993; Stokey 1991), through an R&D lead (Grossman and Helpman 1991), or through achieving internal economies of scale. Two of these channels can be discarded at the outset. An explanation based on internal economies of scale is inconsistent with the highly competitive and fragmented nature of the shipbuilding industry. Also, historical sources indicate that shipyards typically did not invest in R&D, most likely because the highly competitive nature of the industry left them with little surplus to invest, whereas the ease of copying give them little incentive.48 In addition, sources such as Pollard and Robertson (1979) indicate that U.S. shipyards were actually developing and using more advanced technologies, such as hydraulic riveters and larger cranes, than British yards by the early 20th century. Of the remaining channels, available historical evidence points to the development of pools of skilled metal shipbuilders as the key factor that translated Britain’s initial advantages into a persistent lead. As Pollard and Robertson (1979) write in their authoritative history of the British shipbuilding industry (p. 129), “While foreign builders were able to choose better sites and design more efficient yards and shops, they were unable to overcome completely the greater efficiency of British labor, an efficiency that in part derived from Britain’s longer tradition as a producer of iron and steel steamships”. One aspect of this channel was the vital importance played by labor costs, and skilled labor costs in particular, in the industry. In 1877 Scientific American reported that for metal ships, “The greatest [cost] item, however, is labor, the cost of which constitutes fully 60% of that of a steamer, and at least 50% of that of a sailing vessel; or starting with the pig iron and sawn lumber, it is estimated to amount to 80% per cent. . .”.49 Similarly, in 1893 Hichborn (1893) indicates that labor accounted for 69% of the cost of constructing the Navy cruiser Charleston. Most of this labor cost came from skilled craft workers. In British yards these workers made up 70%–80% of the workforce.50 A wide variety of specialized skills were required for the production of large metal ships, including riveters, tinsmiths, boilermakers, carpenters, plumbers, riggers, fitters and draftsmen.51 Although some of these skilled were also applicable in sectors other than shipbuilding (so called “amphibians”) and others were used in both wood and metal ship production (e.g., carpenters and riggers), many important skills were unique to metal shipbuilding. For example, the skills involved in bending and shaping large metal plates into curved and irregular shapes were unique, and vital, to metal shipbuilding. One factor that increased the importance of skills is the fact that the vast majority of ships were bespoke products produced to designs supplied by the buyer.52 This increased the need for skilled workers who could move flexibly between different ship types.53 Skills were acquired primarily through experience. In Britain, this typically meant formal apprenticeships lasting 5–7 years. Only a very small subset of the most skilled workers, such as marine engineers and naval architects, had any formal education.54 In contrast to British yards, evidence suggests that North American producers wishing to begin metal shipbuilding in the late 19th century faced a scarcity of experienced metal shipbuilders. Pollard and Robertson (1979) describe how, to compensate for the lack of skilled workers, North American shipyards used more capital in order to substitute toward unskilled workers. Unfortunately for these yards, “expensive equipment could not compensate for the lower level of skills and more irregular output...Thus, despite Britain’s inferior capital equipment, the output per man hour was still highest in Britain at the end of the [19th] century”. Although, “In the United States, vast overheads crippled builders in all but the best years. British yard owners were able to take advantage of their more highly skilled workforces by investing only in equipment that was absolutely necessary. . .and by refusing to purchase as many labor-saving machines as German and American builders did.” Consistent with this, Hutchins (1948) found that (p. 50), “American shipyard work which could be effectively mechanized cost no more than that in Britain, but handicraft work, of which there was a large amount, was much more expensive”. Thus, despite using more capital and advanced technology, evidence suggests that the lack of skilled workers meant that the cost of producing most merchant ship types in North American yards was much higher than in Britain.55 One illustration of the challenges faced by North American shipyards is provided by the 1905 Report of the Merchant Marine Commission to Congress. This report provides the following example. Convincing proof on this point was offered in 1900, when steel plates and beams, because of labor troubles abroad, were selling at $40.86 in England, and $28 in the United States. Boston shipowners at that time invited bids from an American and a British builder for a cargo steamship of about 5,000 tons capacity. With both yards figuring for a small competitive profit, the American estimate was $275,000 and the English $214,000. The material of the American ship would have cost $63,000; of the English ship, $80,000. But this difference was more than offset by the higher wages paid to the American shipyard mechanics. The higher wages in U.S. yards do not appear to be a result of union activity. Unions were strong in British shipyards, but largely absent from American shipyards before WWI. As one American shipyard superintendent wrote after visiting British yards in 1897, “in my judgment, the worst feature today of the British yards is the tremendous power of the labor unions. . .the exactations and obstructions of all kinds that are thrown about the work by these unions is almost inconceivable” in contrast with U.S. yards where, “we manage our own business. . .the union and the walking delegate are not all powerful”.56 This suggests that, if anything, union activity should have given U.S. yards and advantage. It also does not appear that the United States experienced a shortage of those very high-skilled workers in jobs, such as naval architect or marine engineer, that required formal education. Pollard and Robertson (1979) report that by the early 20th century the United States was turning out more university-trained ship designers than Britain, from places like Cornell, MIT, and the Webb Institute. Instead, the key constraint appears to have been the much larger body of craft workers that relied on skills gained through experience, rather than in the classroom. There is also evidence that a scarcity of skilled labor was the key constraint in Atlantic Canada. For example, focusing on the Maritime Provinces, Sager and Panting (1990) write that (p. 12), “The best contemporary estimates were that Nova Scotia possessed all the necessary advantages for steel shipbuilding except skilled labor”. This is a telling statement, particularly given that Nova Scotia had been one of the foremost (wood) shipbuilding areas in the mid-19th century. Additional evidence on the scarcity of skilled metal shipbuilders in the United States in the late 19th century is offered by Hanlon (2019), which uses census data to study the composition of the workforces in two U.S. Atlantic Coast shipyards that successfully transitioned into metal shipbuilding, Newport News Shipyard in Virginia and Bath Iron Works in Maine, as well as one Great Lakes shipyard in Loraine, OH. That study shows that, early in their life, these shipyards substituted away from skilled workers toward unskilled workers and capital and relied on immigrants from Britain to fill key skilled positions that could not be eliminated. Once established, these yards began to train native-born workers to fill skilled positions. 7. Conclusions The experience of the international shipbuilding industry documented in this study offers a window into understanding how temporary initial advantages can influence long-run patterns of production and trade. My main results show that initial input price advantages can have a long-run impact on the spatial distribution of production and trade patterns. Due to lower input costs, British shipbuilders were able to take an early lead in metal ship construction, overcoming the dominant position that North American producers held in the shipbuilding industry in the first half of the 19th century. Despite losing this advantage in the 1890s, British producers were able to maintain their dominant position in metal shipbuilding into the 20th century, whereas North American firms that were exposed to competition from British producers struggled to make the transition from wood to metal and went into decline. A natural explanation for this pattern is that the industry was characterized by some sort of dynamic learning effects. My analysis of Navy shipyards indicates that in fact shipbuilding was characterized by learning spillovers, and that these were highly localized in nature. Such learning spillovers can explain how a temporary input cost advantage generated a persistent lead, as well as why successful metal shipyards tended to be clustered in just a few locations. Finally, a review of historical evidence suggests that the development of pools of skilled shipyard workers was likely to have been a key channel through which these localized learning effects operated. One implication of this study is that interventions that allow one location to gain an early lead in an industry characterized by learning effects may have long-term benefits. Although this suggests that industrial policy may be successful in certain cases, my findings also highlight the limits of such policies. Despite having access to a completely protected coastal market and a variety of other supporting policies there is no evidence that U.S. Atlantic Coast shipbuilders were able to eventually compete with the British on the international market. The most likely explanation here is that the timing of intervention is crucial. An initial advantage may help a country establish a dominant position in a new industry, but industrial policy may be less useful once producers in another country are already established. It is interesting to note that after WWII, Japan, Korea and China all became internationally competitive shipbuilders with the help of temporary government aid (see, e.g., Lane (2016) on Korea). In future work it will be interesting to consider why shipbuilders in these countries were successful in markets where, even after decades of protection, U.S. shipbuilders were not. One potential explanation for this difference is that the changing nature of the shipping industry, including the changes in ship size and design induced by containerization, fundamentally altered the ship production process in ways that reduced the importance of craft skills. Footnotes 1. I end the study period just before World War I to avoid the massive disruption in the shipbuilding industry caused by this conflict. 2. Thornton and Thompson (2001) looks at learning occurring across shipyards, but it is important to note that those shipyards are typically not located near to one another. Thus, that study may miss learning effects that are localized, exactly the sort that I find evidence for. 3. These papers have been primarily focused on estimating the magnitude of learning effects rather than identifying the mechanisms that drive them. Thornton and Thompson (2001) extend this analysis to a variety of ship types during the World War II (WWII) period. 4. In Britain, Pollard and Robertson (1979) estimate that aggregate wages in shipbuilding made up roughly 1%–2% of total British wages from employment in the period from 1871 to 1911 (p. 36). The importance of the industry in the United States is harder to estimate, but likely to be similar. 5. The shift from sail to steam was due in large part to improvements in engine efficiency (Pascali 2017). 6. Another dimension in which these were not perfect substitutes had to do with ship size. As shown in Online Appendix A.6, the largest ships could only be built of metal. 7. See, for example, Culliton (1948) or Pollard and Robertson (1979). 8. I focus on pig iron prices here and in later discussions despite the fact that this would have to go through several other production steps before being used by shipbuilders. One reason is that pig iron was more standardized than products further down the production chain, so prices are easier to compare across locations. A second reason is that pig iron was a key input into more specialized products used by shipbuilders. A third important reason is that products made from pig iron were used in a wide set of industries, so production is less likely to be endogenously affected by the local shipbuilding than products more specialized for use in ships. 9. In addition to providing a ready supply of ore, the chemical composition of Mesabi ore improved productivity (Allen 1977, 1979). 10. Allen (1981) reports that, “Before the 1890s American [steel] prices substantially exceeded British prices, and the American industry achieved a large size only because of high tariffs. During the 1890s American prices dropped to British levels or below, and America emerged as a major exporter of iron and steel”. Focusing on steel rails in particular, Allen found that, “Between 1881 and 1890 the average price of steel rails at Pennsylvania mills was $37.01 whereas the average British price was $23.62. During the period 1906–1913 the American price had fallen to $28.00 whereas the British price had risen to $29.46”. 11. It is worth noting that U.S. steel producers with market power in the United States may have been dumping steel in Britain in some years. 12. Jacks and Pendakur (2010) and Jacks, Meissner, and Novy (2008) provide evidence that international trade costs fell substantially during this period. For shipbuilding, the Dingley Tariff of 1897 helped reduce the cost of inputs by specifically exempting from duty steel used in the construction of vessels for the foreign trade (Dunmore 1907). This gave shipbuilders the option to buy from European steelmakers and increased the foreign competition faced by U.S. steel producers, particularly on the coast. 13. In contrast, only 82% of the vessels (by tonnage) homeported on the Atlantic Coast of the United States and Canada in 1912 were also constructed there and only 83.5% of the tonnage constructed on the Atlantic Coast between 1890 and 1912 remained there in 1912. Of course, this understates the openness of the coast market because the coastal ports of North American were also served by a large number of vessels homeported in other countries that operated on international routes, whereas Great Lakes ports were served only by vessels homeported on the Lakes. In Online Appendix A.12, I review additional evidence comparing the openness of the Great Lakes and Atlantic ship markets. 14. Thompson (1991) writes (p. 45), “The larger foreign-built ships, those too long to negotiate the locks in the Welland or St. Lawrence. . .had their midbodies removed, and the remaining bow and stern sections were welded together. With the midbody sections stowed in their cargo holds, the downsized ships made their way through the locks. . .Once above the Welland, the vessels would again be cut in half and the midbody sections reinstalled before the ships were put into service”. The Annual Report to the Commissioners of the Navy (p. 15) says of this method, “The experiment of building large vessels, cutting them in two to pass the locks, and then reuniting the parts has been made successfully in a few instances, but at the present time it does not appear that this method. . .will become general”. There are also reports of ships that moved into the Great Lakes by going up the Mississippi river and through the Illinois and Michigan Canal, but this required that the ships have their entire superstructure removed in order to pass under the river bridges along the route. In addition, there were small metal vessels called canallers because they were built to be able to pass through the small St. Lawrence and Welland Canals. Some of these made their way into the Great Lakes in the 1890s, but these smaller ships were usually under 250 ft long. 15. The 1900 Census, the source of these data, mentions that the very low price observed for Illinois is likely to be understated because in that location most pig iron was used internally by firms to produce steel. 16. The smaller size of ships on the Great Lakes was due to the limitations imposed by locks and canals, particularly the lock between Lake Superior and the lower Great Lakes. 17. One sign of the similarity of techniques used on the Lakes and the Coast is provided by the Annual Report of the Commissioners of the Navy from 1901, which suggests that coastal shipbuilders may be able to learn from the more successful yards on the Great Lakes (p. 15): “. . .through the training of shipbuilders, the invention and improvement of shipbuilding tools, machinery, and materials, and through experience gained in the financial and industrial organization of shipyards, the establishments on the Great Lakes are promoting the chance for seaboard growth”. 18. It is worth noting that Britain also had some policies that benefited British shipbuilders during this period, despite the country’s general laissez-faire economic philosophy. Among these was Naval shipbuilding as well as support for fast mail steamers. Although these policies certainly aided British shipbuilders to some extent, as I discuss at the end of Section 4 the key margin of competition between British and North American shipbuilders was in sales to third-party buyers, a market segment that would not have been directly impacted by these British policies. 19. This estimate is based on the size of U.S. coastal production. Given that only U.S.-built ships could be used in the U.S. coastal trade, and given evidence suggesting that U.S. shipbuilders were largely unsuccessful in selling ships outside of that market, I use the size of U.S. coastal production as a measure of the size of the U.S. coastal market. 20. Hutchins (1948) suggests that the substantial expansion of the U.S. Navy in the late 1880s and 1890s, often described as the “New Navy” because the new ships were metal rather than wood, played an important role in the development of U.S. shipbuilding. Online Appendix A.10 describes the increases in U.S. Navy shipbuilding during the study period. Another type of industrial policy was the subsidization of passenger liners on mail-carrying routes that had to be served with domestically built ships. This form of protection was particularly important during the inter-war period. 21. Canada’s status as part of the British Dominion made enacting protection against the mother country “scarcely thinkable” (Sager and Panting 1990, p. 171). Although Canada did impose tariffs on some British goods, ships were different from other products for a number of reasons. Among these was the fact that Canadian ships were protected by British maritime power, since Canada did not possess its own Navy until the Royal Canadian Navy was founded in 1910 (and initially it was equipped with surplus Royal Navy vessels). Canadian shipping companies were also dependent on access to British ports as well as access to international ports granted through treaty agreements between the British Empire and other nations allowing the free entry of ships, which Canada was bound by. These geopolitical concerns made it practically impossible for Canada to protect shipbuilders from British competition. Practical difficulties also made it hard to exclude British vessels. Sager and Panting (1990) explain that because Canada used the British registration system for vessels, it was “virtually impossible to distinguish between British and Canadian ships, and hence a customs duty on British ships [in the Canadian foreign trade] would be impossible to enforce”. Finally, the relatively weak political position of the Maritime Provinces in the Canadian Confederation also limited support for shipbuilders. 22. To be included on a register, a ship had to be inspected. This often occurred multiple times during the construction process and at periodic intervals after construction was complete. To complete these inspections, the registration societies employed a set of local inspectors in the majors shipbuilding areas of the world. 23. The register also included additional information about the current owner, home port and master of each ship. These data were not entered for cost reasons. The home port of each ship was entered for the 1912 ABS Register only. 24. The use of these snapshots is driven primarily by cost concerns. Digitizing each register requires entering data from thousands of pages of documents by hand, so even with outsourcing this to low-cost providers the cost is substantial. 25. The patterns over time described in my data are similar to those found in available aggregate statistics (see Online Appendix A.3), which provides some confidence that the values derived from the registers are reasonable. 26. The registers often did not have complete coverage for ships in the year in which they were published. 27. For ships built in the United States and Canada, I am able to identify the construction location for over 99% of ship tonnage in data from the 1912 register, over 96% of tonnage in the 1889 registers. In data from the 1871 registers, the share of tonnage linked to a location within the United States and Canada, respectively, is 97.1% and 88.3%. The larger share of tonnage with missing locations in the Canadian data is due to the fact that only the province of construction was provided for many Canadian ships registered in the 1871 Lloyd’s. 28. For example, the analysis in Juhasz (2018) looks across 88 French Departments, whereas the main firm-level analysis in Irwin (2000) covers 45 firms. 29. I use towns as the unit of observation because individual shipyards are difficult to consistently identify in the data, particularly over time. 30. There were no Navy shipyards on the Great Lakes and no substantial Naval vessels were produced there. 31. That is, there is no evidence that U.S. Lakes producers were substantially ahead or behind Canadian Lakes producers in either sector. 32. I have explored specifications including market access as a control, following Donaldson and Hornbeck (2016). Market access measures do not seem to be correlated with shipbuilding activity. Moreover, the link between broad measures of (inland) market access and shipbuilding is not intuitively obvious. Thus, I have decided to omit market access controls from the main specifications. 33. Shipbuilding in other nearby locations is based on data from the registers. County level employment data are from the 1880 Census. State level price data are from the 1900 Census. 34. See Online Appendix A.9 for further details. 35. A map describing the areas that I use is available in Online Appendix A.7. Using shipbuilding regions is preferable to political boundaries such as states, since it ensures that nearby shipyards (e.g., New York City and Newark, NJ) are included in the same region and distant yards (e.g., New York City and Buffalo, NY) are not. 36. An alternative approach might be to run the analysis at the county level and include all counties that bordered the lakes or the Atlantic. This approach requires that I take a stand on counties suitable for shipbuilding. This determination is not as straightforward as it seems. For example, many shipbuilders located on rivers, whereas many coastal counties with rugged coastlines or in the north of Canada were unlikely shipbuilding locations. Thus, a sample of coastal counties is likely to include many counties that were unsuitable for shipbuilding, which really should not be in the sample. 37. Of the controls included in the regressions in Table 2, the most explanatory are the indicators for whether a location was active in a particular sector in 1870. The other consistently significant control variable is county population, which is positively related to whether a location was active in both metal and wood ship production (outcome three). 38. A table displaying the estimated coefficients for all of the controls variables is in Online Appendix A.16. 39. Mean production in active metal shipbuilding locations in the data in 1901–1910 was 66,000 tons. 40. In terms of magnitude, the estimated effect of being in the Great Lakes or the United States in Figure 6 are somewhat smaller than those obtained from cross-section tonnage regressions in Online Appendix Table A.11. In particular, the cross-sectional results suggest that Great Lakes locations produced around 120,000 more metal tons and U.S. locations produced around 40,000 additional metal tons. The panel results suggest that Great Lakes locations produced around 30,000 additional metal tons and U.S. Coast locations produced around 3,000 tons after 1900. 41. Statistics are for ships registered in the 1912 ABS registry. See Online Appendix Table A.4 for further details. 42. Of the tonnage registered in the 1912 ABS, 32% was homeported outside of the United Kingdom, United States, or Canada. Of this, 60%, just over 4 million tons, was built in a country other than the homeport country. As a point of comparison, total tonnage homeported in the United States was just over 5 million tons. Thus, the contested third-party market available to U.S. shipbuilders was almost as large as the total domestic market. 43. Quoted from the Marine Review, October 28, 1897. 44. This issue has been studied by Thornton and Thompson (2001), but their analysis uses a relatively small number of geographically dispersed yards, which makes it impossible for them to look for evidence of geographically localized spillovers. 45. Proximity to Navy Yards may have also improved access to Navy contracts, which could have had beneficial effects that spilled over into the construction of merchant ships within yards. In the robustness exercises in Online Appendix A.18, I present results from a specification that includes a control for whether each private shipyard received Navy contracts. 46. The five Naval shipyards in operation during the period I study were in Portsmouth, VA (Norfolk NSY, opened 1767), Boston, MA (opened 1800), New York City (Brooklyn NSY, opened 1800), Philadelphia (opened 1801), and Kittery, ME (Portsmouth NSY, opened 1800). The only other early Atlantic shipyard, in Washington, DC, was opened 1799 but this yard largely ceased ship construction after the War of 1812 because the Anacostia River was too shallow to accommodate larger vessels. A Coast Guard shipyard was opened in Baltimore in 1899, but I do not include that in my analysis because it is likely that the location of that yard was influenced by Baltimore’s potential for metal shipbuilding. 47. There are 74 U.S. Atlantic Coast shipyards in this analysis some of which are active in both wood and metal shipbuilding, yielding a total of 89 location × material observations. Of these locations, 24 were located within 50 km of a Navy shipyard, 39 were located from 50 to 100 km from a Navy shipyard, and 11 were more than 100 km from a Navy shipyard. 48. For example, Pollard and Robertson (1979) write that (p. 148), “Many improvements, if not most, however, were developed outside of the industry, in the steel-making, electrical products, or engineering industries. . .it was only necessary for the shipbuilders to adopt innovations after the basic research had been done elsewhere. Few laboratories were established in the yards, and as the reluctance to use experimental tanks [to test ship designs] demonstrates, builders were not even very interested in investing funds to solve problems peculiar to their industry”. 49. Sci. Am. (1877). 50. Pollard and Robertson (1979, Table 8.1, p. 153) show that in 1892 unskilled workers made up 29% and 22% of the labor force in English and Scottish shipyards, respectively, 18% in Scottish yards in 1911, and 25% in Northeast England in 1913. 51. See Pollard and Robertson (1979, p. 78). 52. Pollard and Robertson (1979) write (p. 152), “. . .the fact that they [shipbuilders] produced for the most part a large, custom-made commodity that was not susceptible to many of the techniques of mass production, ensured that a premium continued to be placed upon skilled labor”. 53. This represents an important difference relative to the Liberty shipbuilders studied in previous work, who focused on producing standardized designs. 54. See Pollard and Robertson (1979) for more details. 55. Hutchins (1948), for example, suggests that (p. 47), “British costs were from 30 to 40 percent less”. The Report of the Merchant Marine Commission found that in 1905 the difference was 30%–50% (p. viii). 56. This quote is from W.I. Babcock of The Chicago Ship Building Company, writing in the Marine Review, September 2, 1897. Acknowledgments I am grateful to the International Economics Section at Princeton for support and advice while writing this paper. I thank Philip Ager, Leah Boustan, Stephen Broadberry, Paula Bustos, Dave Donaldson, Capser Worm Hansen, Richard Hornbeck, Reka Juhasz, Ian Kaey, Joan Monras, Petra Moser, Henry Overman, Ariel Pakes, Jean-Laurent Rosenthal, Martin Rotemberg, and seminar participants at Caltech, CEMFI, Chicago, Copenhagen, Michigan State, Princeton, University of Virginia, Warwick, and NYU Stern for helpful comments. Meng Xu and Anna Sudol provided excellent research assistance. A previous version of this paper was circulated under the title “Evolving Comparative Advantage in International Shipbuilding, 1850–1911”. Funding was provided by a Cole Grant from the Economic History Association, the Hellman Fellowship at UCLA, a research grant from UCLA’s Ziman Center for Real Estate, and National Science Foundation Career Grant No. 1552692. Notes The editor in charge of this paper was Paola Giuliano. References ( 1877 ). “ Our Iron Shipbuilding Industry .” Scientific American , 36 . OpenURL Placeholder Text WorldCat Allen R. C. ( 1981 ). “ Accounting for Price Changes: American Steel Rails, 1879-1910 .” Journal of Political Economy , 89 , 512 – 528 . Google Scholar Crossref Search ADS WorldCat Allen Robert C. ( 1977 ). “ The Peculiar Productivity History of American Blast Furnaces, 1840-1913 .” The Journal of Economic History , 37 , 605 – 633 . Google Scholar Crossref Search ADS WorldCat Allen Robert C. ( 1979 ). “ International Competition in Iron and Steel, 1850-1913 .” The Journal of Economic History , 39 , 911 – 937 . Google Scholar Crossref Search ADS WorldCat Argote Linda , Beckman Sara L. , Epple Dennis ( 1990 ). “ The Persistence and Transfer of Learning in Industrial Settings .” Management Science , 36 , 140 – 154 . Google Scholar Crossref Search ADS WorldCat Atack Jeremy ( 2016 ). “ Historical Geographic Information Systems (GIS) database of U.S. Railroads for Years Before 1890 .” https://my.vanderbilt.edu/jeremyatack/data-downloads/, retrieved November 2018. OpenURL Placeholder Text WorldCat Baldwin Richard , Krugman Paul ( 1988 ). Industrial Policy and International Competition in Wide-Bodied Jet Aircraft . University of Chicago Press , pp. 45 – 78 . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Bleakley Hoyt , Lin Jeffrey ( 2012 ). “ Portage: Path Dependence and Increasing Returns in U.S. History .” Quarterly Journal of Economics , 127 , 587 – 644 . Google Scholar Crossref Search ADS PubMed WorldCat Culliton James W. ( 1948 ). “ Economics and Shipbuilding .” In The Shipbuilding Business in the United States of America , edited by F. G. Fasset. The Society of Naval Architects and Marine Engineers, New York , pp. 1 – 13 . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Donaldson Dave , Hornbeck Richard ( 2016 ). “ Railroads and American Economic Growth: A “Market Access” Approach .” Quarterly Journal of Economics , 131 , 799 – 858 . Google Scholar Crossref Search ADS WorldCat Dunmore Walter T. ( 1907 ). Ship Subsidies . Houghton Mifflin Company, Boston and New York . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Grossman Gene , Helpman Elhanan ( 1991 ). Innovation and Growth in the Global Economy . The MIT Press , Cambridge, MA . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Hanlon W. Walker ( 2019 ). “ Skilled immigration and American Industrialization: Lessons from Newport News Shipyard .” Business History Review , 92 , 605 – 632 . Google Scholar Crossref Search ADS WorldCat Harley C. Knick ( 1970 ). “ British Shipbuilding and Merchant Shipping: 1850-1890 .” The Journal of Economic History , 30 , 262 – 266 . Google Scholar Crossref Search ADS WorldCat Harley C. Knick ( 1973 ). “ On the Persistence of Old Techniques: The Case of North American Wooden Shipbuilding .” The Journal of Economic History , 33 , 372 – 398 . Google Scholar Crossref Search ADS WorldCat Head Keith ( 1994 ). “ Infant Industry Protection in the Steel Rail Industry .” Journal of International Economics , 37 , 141 – 165 . Google Scholar Crossref Search ADS WorldCat Hichborn Philip ( 1893 ). “ Shipbuilding Here and Abroad .” The North American Review , 156 , 398 – 407 . OpenURL Placeholder Text WorldCat Hutchins John G.B. ( 1948 ). “ History and Development of the Shipbuilding .” In The Shipbuilding Business in the United States of America , edited by F. G. Fasset. The Society of Naval Architects and Marine Engineers, New York , pp. 14 – 60 . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Irwin Douglas A. ( 2000 ). “ Did Late-Nineteenth-Century U.S. Tariffs Promote Infant Industries? Evidence from the Tinplate Industry .” The Journal of Economic History , 60 , 335 – 360 . OpenURL Placeholder Text WorldCat Irwin Douglas A. ( 2003 ). “ Explaining America’s Surge in Manufactured Exports, 1880-1913 .” The Review of Economics and Statistics , 85 , 364 – 376 . Google Scholar Crossref Search ADS WorldCat Jacks David S. , Meissner Christopher M. , Novy Dennis ( 2008 ). “ Trade Costs, 1870-2000 .” The American Economic Review, Papers & Proceedings , 98 , 529 – 534 . Google Scholar Crossref Search ADS WorldCat Jacks David S. , Pendakur Krishna ( 2010 ). “ Global Trade and the Maratime Transport Revolution .” Review of Economics and Statistics , 92 , 745 – 755 . Google Scholar Crossref Search ADS WorldCat Juhasz Reka ( 2018 ). “ Temporary Protection and Technology Adoption: Evidence from the Napoleonic Blockade .” American Economic Review , 108 ( 11 ), 3339 – 76 . Google Scholar Crossref Search ADS WorldCat Krueger Alan O. , Tuncer Baran ( 1982 ). “ An Empirical Test of the Infant Industry Argument .” American Economic Review , 72 ( 5 ), 1142 – 1152 . OpenURL Placeholder Text WorldCat Krugman Paul ( 1987 ). “ The Narrow Moving Band, the Dutch Disease, and the Competitive Consequences of Mrs. Thatcher .” Journal of Development Economics , 27 , 41 – 55 . Google Scholar Crossref Search ADS WorldCat Lane Nathan ( 2016 ). “ Manufacturing Revolution: Industrial Policy and Networks in South Korea .” Working paper, Monash University, Melbourne, Australia . OpenURL Placeholder Text WorldCat Lucas Robert E. ( 1988 ). “ On the Mechanics of Economic Development .” Journal of Monetary Economics , 22 , 3 – 42 . Google Scholar Crossref Search ADS WorldCat Lucas Robert E. ( 1993 ). “ Making a Miracle .” Econometrica , 61 , 251 – 272 . Google Scholar Crossref Search ADS WorldCat Matsuyama Kiminori ( 1992 ). “ Agricultural Productivity, Comparative Advantage, and Economic Growth .” Journal of Economic Theory , 58 , 317 – 334 . Google Scholar Crossref Search ADS WorldCat Mitrunen Matti ( 2019 ). “ Structural Change and Intergenerational Mobility: Evidence from the Finnish War Reparations .” Working paper, University of Chicago , https://www.dropbox.com/s/d83g2772ttzzwxq/war_reparations_Mitrunen.pdf?dl=0. OpenURL Placeholder Text WorldCat Pascali Luigi ( 2017 ). “ The Wind of Change: Maratime Technology, Trade and Economic Development .” American Economic Review , 107 ( 9 ), 2821 – 2854 . Google Scholar Crossref Search ADS WorldCat Pollard Sidney , Robertson Paul ( 1979 ). The British Shipbuilding Industry, 1870-1914 . Harvard University Press , Cambridge, MA . Google Scholar Crossref Search ADS Google Scholar Google Preview WorldCat COPAC Rapping Leonard ( 1965 ). “ Learning and World War II Production Functions .” The Review of Economics and Statistics , 47 , 81 – 86 . Google Scholar Crossref Search ADS WorldCat Sager Eric W. , Panting Gerald E. ( 1990 ). Maritime Capital . McGill-Queen’s University Press , Montreal . Searle Allan D. ( 1945 ). “ Productivity and Labor in Industry .” Monthly Labor Review , 61 , 1132 – 1147 . OpenURL Placeholder Text WorldCat Stokey Nancy ( 1991 ). “ Human-Capital, Product Quality, and Growth .” The Quarterly Journal of Economics , 106 , 587 – 616 . Google Scholar Crossref Search ADS WorldCat Thompson Mark L. ( 1991 ). Steamboats and Sailors of the Great Lakes . Wayne State University Press, Detroit, Michigan . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Thompson Peter ( 2001 ). “ How Much Did the Liberty Shipbuilders Learn? New Evidence for an Old Case Study .” Journal of Political Economy , 109 , 103 – 137 . Google Scholar Crossref Search ADS WorldCat Thornton Rebecca A. , Thompson Peter ( 2001 ). “ Learning from Experience and Learning from Others: An Exploration of Learning and Spillovers in Wartime Shipbuilding .” American Economic Review , 91 ( 5 ), 1350 – 1368 . Google Scholar Crossref Search ADS WorldCat Young Alwyn ( 1991 ). “ Learning By Doing and the Dynamic Effects of International Trade .” Quarterly Journal of Economics , 106 , 369 – 405 . Google Scholar Crossref Search ADS WorldCat © The Author(s) 2019. Published by Oxford University Press on behalf of European Economic Association. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model)
TI - The Persistent Effect of Temporary Input Cost Advantages in Shipbuilding, 1850 to 1911
JF - Journal of the European Economic Association
DO - 10.1093/jeea/jvz067
DA - 2020-03-23
UR - https://www.deepdyve.com/lp/oxford-university-press/the-persistent-effect-of-temporary-input-cost-advantages-in-S1w2a34duv
DP - DeepDyve
ER -