The Local Impact of Containerization

Finance and Economics Discussion Series Divisions of Research & Statistics and Monetary Affairs Federal Reserve Board, Washington, D.C. The Local Impact of Containerization Leah Brooks, Nicolas Gendron-Carrier, and Gisela Rua 2018-045 Please cite this paper as: Brooks, Leah, Nicolas Gendron-Carrier, and Gisela Rua (2018). “The Local Impact of Containerization,” Finance and Economics Discussion Series 2018-045. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/FEDS.2018.045. NOTE: Staff working papers in the Finance and Economics Discussion Series (FEDS) are preliminary materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the research staff or the Board of Governors. References in publications to the Finance and Economics Discussion Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers.

Working Paper The Local Impact of Containerization Leah Brooks Trachtenberg School of Public Policy and Public Administration George Washington University Nicolas Gendron-Carrier Department of Economics University of Toronto Gisela Rua Division of Research and Statistics Board of Governors of the Federal Reserve System Friday 1st June, 2018 Thanks to Elliot Anenberg, Nathaniel Baum-Snow, Paul Carrillo, Jonathan Dingel, Jessie Handbury, Thomas Holmes, Peter Morrow, Luu Nguyen, Justin Pierce, Lukas Püttmann, JordanRappaport,andMatthewTurnerformanyhelpfulcommentsanddiscussions. We arealsoveryappreciativetoresearchassistantsJamesCalello, AdrianHamins-Puertolas, Arthi Rabbane, Alex Severn, and Daniel Walker who helped with data entry. We are 1956 grateful to Matthew Turner and Gilles Duranton for sharing the County Business Patterns data. Finally, we thank Nora Brooks, retired reference librarian and Leah’s 1948 mother, whose combination of digital and pre-digital skills uncovered the portlevel trade data. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by the Board of Governors of the Federal Reserve System or its staff. i

The Local Impact of Containerization We investigate how containerization impacts local economic activity. Containerization is premised on a simple insight: packaging goods for waterborne trade into a standardized container makes them dramatically cheaper to move. We use a novel costshifter instrument – port depth pre-containerization – to contend with the non-random adoption of containerization by ports. Container ships sit much deeper in the water than their predecessors, making initially deep ports cheaper to containerize. Consistent with New Economic Geography models, we find that counties near container ports grow an 70 1950 2010 additional percent from to . Gains predominate in counties with initially low population density and manufacturing. Leah Brooks Nicolas Gendron-Carrier Trachtenberg School of Public Policy Department of Economics George Washington University University of Toronto 805 21 601 150 st St. NW, Room F St. George Street 20006 5 3 7 Washington, DC Toronto, ON, Canada, M S G lfbrooks@gwu.edu nicolas.gendron.carrier@mail.utoronto.ca Gisela Rua Division of Research and Statistics Board of Governors of the Federal Reserve System 20 82 th and C Streets, Mail Stop 20551 Washington, DC gisela.rua@frb.gov ii

Underlying the second wave of globalization following World War II is a vast improvement in the ability to transport goods. New York City’s Herald Square Macy’s now finds it cheaper to source a dress from Malaysia than from the city’s own rapidly 2008 3 disappearing garment district (Levinson, , p. ). This decline in the importance of physicaldistanceowesmuchtothedevelopmentandriseofcontainerization. Container- 1960 ization, which took off in the early s, is premised on a simple insight: packaging goods for waterborne trade into a standardized container makes them cheaper to move. Containerization simplifies and speeds packing, transit, pricing, and the transfer from ship to train to truck. It also limits previously routine and lucrative pilferage. These cost declines have yielded sea changes in trade. From the advent of container- 1956 1981 ization in to , containerization caused international trade to grow by more than 1000 2016 , percent (Bernhofen et al., ). Containerized cargo now accounts for over half of global non-commodity trade (United Nations Conference on Trade and Development, 2013 ). In this paper, we use novel data and a new identification strategy to understand how the drastic decline in trade cost brought by containerization impacts local economic activity. We address the non-random adoption of the new shipping technology by ports with a novel cost-shifter instrument: port depth pre-containerization. Such a instrument isolates exogenous, cost-driven port containerization from adoption due to local demand. Because container ships sit much deeper in the water than their predecessors, they require deeper ports in which to dock. Dredging a harbor to increase depth is possible, but it is extremely costly. Wefindthatthecostadvantageconferredbyadeepharborinthepre-containerization era makes a port more likely to containerize. To ensure that the instrument works through the cost of supplying a container port and not through a port’s initial competitive advantage, we limit the instrument to only ports that are “very deep” pre- 1

containerization. Intuitively, these ports had a depth beyond any pre-containerization advantage. To undertake this analysis, we combine a large variety of data sources for the period 1910 2010 to . WeuseUScountiesasourunitofanalysis,withcounty-leveldemographics 1910 2010 and income from the Decennial Census ( to ) and employment and payroll data 1956 1971 2011 from the County Business Patterns ( and to ). We supplement these data 1953 2014 with information on ports from and , containerization adoption, and port-level foreign trade in the pre-containerization era. To measure contemporaneous alternative 1950 transportation, we use newly digitized highway and rail routes circa . We assess the economic impact of containerization’s most important feature—the reduction in trade costs—through the lens of a New Economic Geography (NEG) model 1995 2008 2017 (Helpman, ; Redding and Sturm, ; Redding and Rossi-Hansberg, ). In these models, agglomeration and dispersion forces account for the spatial distribution of economic activity, and population moves in response to changes in real wages. A NEG model predicts that containerization’s reduction in trade costs causes an increase population near containerized ports, and that this effect decays as distance from the port increases. Containerization has an ambiguous impact on nominal wages, depending on the balance between productivity gains from access to a larger market and greater competition from lower-priced distant firms. 1950 2010 Our findings are consistent with these theoretical predictions. From to , our 100 instrumental variable estimates report that counties within km of a container port 53 experience population gains of an additional percentage points, which corresponds 70 60 to an increase in population of about percent over the year period. We find smaller, but still economically meaningful, gains in employment, and no substantive change in nominalwages. Thesegainsinpopulationandemploymentdissipatewithdistancefrom 300 the port, and are indistinguishable from zero beyond roughly km. 2

We find that measures of initial land values mediate gains to containerization. Containerization requires large extensions of land, as port activity shifts from water-based fingerpierstogiantcranesandvastmarshallingyards. NewEconomicGeographymodels also predict that initially smaller locations experience a proportionally larger increase in access to new markets. Consistent with both the NEG intuition and the fact that container ports are cheaper and easier to develop in initially low land value areas, we find that gains from containerization, in percentage terms, are concentrated in places where we expect low pre-containerization land values: counties with initially lower population density and manufacturing employment. Our paper adds to several literatures. First, our findings contribute to the debate on 1999 the impact of globalization on economic activity. Following Romer and Frankel ( ), a large literature has emerged to understand how improved access to international mar- 2009 2017 1 kets affects country level outcomes such as GDP (e.g. Feyrer, a,b; Pascali, ). Our paper contributes to this literature by looking at how the reduction in trade costs brought by containerization affects the spatial distribution of economic activity within countries. In doing so, our results shed light on the potential uneven impacts of globalization. To the best of our knowledge, only one other paper isolates the causal effects of 2017 globalization on local economic activity: Campante and Yanagizawa-Drott ( ). These authors exploit constraints on the capacity of airplanes to fly long distances to obtain a source of exogenous variation in access to international markets at the city level. As do we, they find large positive effects of access to international markets on local economic 2 activity. 1 Most papers in this literature find that improved access to international markets has large positive effectsonGDP,withtheexceptionofPascali(2017)whodocumentsmainlynegativeeffects. Pascali(2017) isparticularlyrelatedtoourpaperinthatheexploitsamajorimprovementintheshippingtechnology— the advent of the steamship—to examine how a decline in international transportation costs impacts economicactivity. 2 Our paper also complements a growing literature in international trade that looks at the impact of trade shocks on local labour markets (e.g. Topalova, 2010; Autor et al., 2013; Kovak, 2013). These papers 3

Second, our paper contributes to a growing academic literature investigating the consequences of improvements in transportation infrastructure on local economic activity 2007 2008 2012 (Baum-Snow, ; Michaels, ; Duranton and Turner, ; Donaldson and Horn- 2016 beck, ). These studies examine how investments in highways and railways have shaped the spatial distribution of economic activity within countries. Our paper is the first to study how large investments in maritime transportation infrastructure, specifically new container terminals, affect the economic conditions of target areas. Methodologically, our paper contributes a new instrumental variable strategy to contend with the non-random allocation of transportation infrastructure. Specifically, we introduce a cost-shifter instrument to obtain a source of quasi-random variation in the observed 2015 infrastructure. See Redding and Turner ( ) for a recent survey of the literature. Finally, our work enhances the growing literature on containerization by expanding its focus beyond the shipping and trade industries. In this burgeoning literature, Rua 2014 2016 ( ) investigates the global adoption of containerization, and Bernhofen et al. ( ) 3 2007 2014 estimate its impact on world trade. Hummels ( ), Bridgman ( ) and Cos¸ar and 2018 Demir ( ) all analyze containerization’s impact on shipping costs. The remainder of this paper is organized as follows. The following section provides 3 background on containerization, Section outlines the theoretical motivation, and Sec- 4 5 tion discusses the data. We present empirical methods in Section , and results in 6 7 Section . We conclude with Section . compare locations within a country that have similar access to international markets but that, because of initial differences in industry composition, are differentially affected by changes in a trading partner’s economic activity (e.g. China). In contrast, we control for initial differences in industry composition and comparelocationsthatexperiencedifferentialgainsinaccesstointernationalmarkets. 3 TheseminalbookonthistopicisLevinson(2008). 4

2 Containerization Beforegoodswentintothebox,shippingwasexpensiveandslow. Vesselsspentweeksat portswhilegangsofdockworkershandledcargopiecebypiece. Portcostsaccountedfor a sizeable share of the total cost of the movement of goods. The American Association of Port Authorities estimated that in-port costs, primarily labor, accounted for half the cost 1960 of moving a truckload of medicine from Chicago to Nancy, France in (Levinson, 2008 9 , p. ). In response to these high costs, producers searched for alternatives. Trucker and entrepreneur Malcolm McLean is generally credited with being the first to match vision 58 1956 with reality when he moved truck trailers on a ship from Newark to Houston in on the maiden container voyage. Containerized trade relies on two key innovations. The first is the mechanization of container movements. Rather than workers with carts, specialized container cranes lift containers in and out of ships, around the port, and onto rail cars and trucks. This mechanization substantially decreased per unit labor costs, cut time in ports and made 17 ever-larger ships viable. Today’s Post-Panamax ship is more than times larger than 1956 1 the first ship to carry container goods in (see ship sizes in Appendix Figure ). The second key innovation of containerization is the development of common standards for container size, stacking techniques, and grip mechanisms. These standards allow a container to be used across modes of transportation—ships, trucks, rail—and 1960 across countries. The U.S. standard for containers was adopted in the early s, and 1960 the international standard followed in the late s. To achieve economies of scale, containerization requires physical changes to ports. In breakbulk ports, as cargo ports were known before the rise of containerization, ships pulled into finger piers and workers on- and off-loaded items by hand and cart. Ports 5

were centrally located within cities and used a large amount of labor and a moderate amount of land for warehousing and storage. In contrast, containerized ports require substantially less labor per unit of weight and a much larger amount of land. Land is used both for the large cranes that move containers and for the marshalling of containers 2014 and trucks (Rua, ). Despite containerization’s small-scale start, it diffused extremely rapidly across the 1960 United States. The bulk of domestic containerization adoption occurred in the s, as 1 shown in Figure a, which reports the total number of US containerized ports by year. In 1960 the early s, the cost decreases from containerization were perceived as primarily a domesticbenefit, orfollowingBenjaminChinitz, “atrendfarmoreadvancedindomestic 1960 85 waterhaulsthaninforeigntrade”(Chinitz, , p. ). Containerizationadoptioninthe 1970 1980 United States continued at a slower pace throughout the s and s and plateaued thereafter. Post-containerization, the distribution of dominant ports has shifted. Of the ten 1955 largest ports before containerization (in , measured in terms of international trade), two never containerized: New York (Manhattan), NY and Newport News, VA. In fact, 1956 the Port of Manhattan, the largest in the world in , no longer exists as a freight port. 25 25 Of today’s largest ports, four did not rank in the pre-containerization top . Only two of the modern ten largest ports were in the pre-containerization top ten: Norfolk, 4 VA and Los Angeles, CA. Adoption of containerization in the rest of the world followed a similar pattern, 1 roughly one decade delayed. Figure b shows that the majority of containerization out- 1970 2014 side the US occurred in the s (see also Rua ( )). The pace of adoption in the US and across the world is consistent with the initial pattern of containerized trade. Until at 1960 least the mid s, containerized trade was primarily domestic. The first international 4 SeeKubyandReid(1992)onportconcentration. 6

1966 container service did not begin until , nearly a decade after the first US shipment. 2016 Containerized trade is now central to the global economy. Bernhofen et al. ( ) 1000 estimate that containerization caused international trade to grow by more than , 15 1966 2013 percent over the years following . As of , containerized trade accounted for over half of global non-commodity trade (United Nations Conference on Trade and 2013 5 Development, ). The literature credits containerization with substantially decreasing the cost of wa- 2014 2007 terborne trade. While Bridgman ( ) and Hummels ( ) note only a small decline in shipping rates, traditional measures of shipping costs understate the true cost advantage yielded by containerization. Containerization cuts the time ships spend at port and thus 2013 the total time in transit. Hummels and Schaur ( ) estimate that each day in transit is 06 21 worth between . to . percent of the value of the good, showing that the time benefits of containerized shipping are non-negligible. In addition, losses to pilferage plummeted 1982 25 with containerization. Wilson ( ) estimates loses to pilferage at roughly percent in 6 the breakbulk era, and near zero in the container era. Finally, containers ease logistics costs by protecting goods from unintentional damage and allowing different kinds of 2017 goods, with different destinations, to be shipped together (Holmes and Singer, ). 2013 2018 Using export transaction data for Turkey, Cos¸ar and Demir ( ) find that con- 16 22 tainerization decreases variable shipping costs between to percent. 5 While containers are appropriate for carrying many goods, as diverse as toys and frozen meat, some goodsarenotyetcontainerizable. Both“non-drycargo”and“dry-bulkcommodities”suchasoil,fertilizers,ore,andgraincannotbeshippedinside“thebox.” 6 It is therefore no surprise that Scottish whiskey bound for US markets was on the first international containertrip(Levinson,2008,p. 165). 7

3 Theoretical Motivation We now turn to the theoretical literature to frame our empirical work and understand containerization’s potential impact. Containerization’s most important feature is the reduction in waterborne transit costs it generates. Because almost all goods transported by water require additional land-based movement, reductions in trade costs due to containerization are largest in percentage terms at the port and decay as distance to the port increases. We assess the impact of this reduction in trade costs through the lens of a standard 1995 2008 New Economic Geography model (e.g. Helpman, ; Redding and Sturm, ; Red- 2017 ding and Rossi-Hansberg, ). In this class of models, agglomeration and dispersion forces explain the uneven distribution of economic activity across space, resulting in particular from people moving in response to changes in real wages. Variation in real 1 2 3 wages typically results from changes in ( ) nominal wages, ( ) the cost of living, and ( ) land prices. Containerization’s reduction in trade costs has three main short run effects. First, when firms produce differentiated products and consumers love variety, locations with lower trade costs become more attractive to consumers. These locations offer a greater variety of goods at lower prices, reducing the cost of living and increasing real wages. Second, if there are increasing returns to scale in production, a reduction in trade costs also increases the profitability of firms because firms can access a larger market for their products. This “home market effect” yields an increase in nominal wages and, therefore, an increase in real wages. Third, due to increased trade, firms encounter more lower-priced competitors. This heightened competition, known as the “market crowding effect,” acts as a dispersion force and causes both nominal and real wages to decline. 8

If there are gains from trade, as New Economic Geography models assume, the cost of living effect and the home market effect should dominate the market crowding effect. 7 Thus, we expect a short run increase in real wages in locations near container ports. In the long run, however, higher real wages should attract people to locations near container ports. As population increases, land prices rise, in turn lowering real wages. Migration ceases when real wages equalize across space. Since the containerization-induced reduction in trade costs declines with distance from the port, we anticipate that the impact of containerization on population is greatest in places near container ports and declines as distance to the port increases. The simplest New Economic Geography framework, outlined above, assumes that places are all ex-ante homogeneous. However, an extension to the basic framework can allow the same shock to impact cities unevenly, as a function of the city’s initial characteristics. In the empirical section, we consider variation in both initial population and land values. Firms in initially less populous cities rely more heavily on the demand from non-local consumers. We therefore expect containerization to have a larger impact, in percentage terms, in initially smaller cities relative to initially larger cities. Wealsoexpectcontainerizationtohaveanuneveneffectbasedonpre-containerization land prices. Because containerports requirelarge swaths ofland forgiant cranes andextensive marshalling yards, rather than the water-based finger piers of the breakbulk era, container ports may be more viable in locations with initially low land value. However, as local productivity shocks are ultimately capitalized into the value of land (Moretti, 2011 ), low land value cities tend also to be small cities, all else equal. Thus, empirically, the distinction between being initially low population and initially low land value may not be empirically visible. 7 Note that even if there are gains from trade, the net effect on nominal wages is ambiguous because thehome-marketeffectandthemarket-crowdingeffectgoinoppositedirections. 9

In sum, New Economic Geography models predict that containerization’s reduction in trade costs causes population to increase near container ports. This effect diminishes as distance to the container port increases. Containerization’s net effect on nominal wages remains theoretically ambiguous because the productivity gains associated with access to a larger market may be offset by the intensified competition from distant firms. Finally, for a given distance to a container port we anticipate greater population growth in initially smaller cities. These smaller cities receive a proportionately larger increase in access to new markets and have relatively cheap land, which is key to container port development. 4 Data Tostudytheimpactofcontainerizationonlocaleconomicactivity,weconstructacountylevel panel dataset that includes population and employment information, as well as proximity to port and port characteristics. This section gives an overview of the data, and we present full details in the data appendix. 1910 20108 OursampleframeistheDecennialCensus,fortheyears to . Weassemblea 1950 2010 time invariant panel of counties by aggregating counties to their counterparts 1910 2010 andbydroppingaveryfewcountieswithlargelandareachanges. From to we 1950 2010 observe population; and from to income and demographic characteristics. We also observe total employment, total payroll, and employment and payroll by industry 1956 1971 20119 from the County Business Patterns from and then annually to . We omit 1950 Alaska from our analysis because its administrative districts in do not correspond 8 Forthe2010sample,weusetheDecennialCensusforpopulationfiguresandtheAmericanCommunitySurvey(years2008–2012)forotherdemographiccovariates. 9 We are very appreciative of digitized 1956 County Business Patterns from Matt Turner and Gilles Duranton. Seethedataappendixformoreinformationaboutthesedata. 10

3023 10 to modern counties. This yields , counties with complete data. To this sample frame, we add port attribute data. Our universe of ports is all ports 1953 2015 1953 2015 that existed in either or , as defined by the and World Port Index. For each port, we observe its location (latitude and longitude), size (in four discrete categories), and depth (in eight discrete categories). We gather the year of first con- 1968 1970 tainerization from the Containerisation International Yearbook, volumes and to 201011 1948 1955 . We also observe and international trade in dollars by port from the CensusBureau’sForeignTradeStatistics. Weassociateeachcountywithavectorofports and port characteristics, which include the distance from each county to each port, the 1953 1953 number of nearby ports, the maximal depth of nearby ports in , and the total 1948 195512 value of international trade at nearby ports in and . We also include variables that characterize the state of the transportation network 1957 1960 now and at the advent of containerization (c. for highway and c. for rail). We measure total rail kilometers, highway kilometers, and waterway kilometers in each county, per square kilometer of each county’s area. In addition to these detailed US data, we construct a less detailed panel dataset of 2014 world cities. The sample frame for world cities is the United Nation’s Revision 1692 of World Urbanization Prospects. This dataset contains all , urban agglomerations 300000 1950 2014 with populations exceeding , at any time between and . By construction, 1950 this sample over-represents fast growing cities that were small in but grew rapidly in the second half of the twentieth century. To mitigate this sampling issue, we restrict 50000 1950 1051 the sample to cities with population over , in , yielding a world panel of , 10 Estimations using County Business Patterns data use a slightly smaller sample because the provider suppressesdataforcountiesundercertainconditions;seedataappendixforcompletedetails. 11 For the purposes of this paper, and consistent with the industry definition, we call a port “containerized” when it has special infrastructure and equipment to handle containers. Specifically, the port has invested in equipment to handle shipping containers which enables their movement in and out of ship andontoatrainoratruck. 12 Wecalculatealldistancesfromthecountycentroid. 11

cities. 5 Empirical Methods We now turn to our empirical strategy for estimating the causal effect of containerization on local economic activity. We first present a difference-in-differences framework to analyze the impact of proximity to a containerized port on economic activity and illustrate its strengths. We then discuss remaining concerns with causality, followed by a motivation for and details about our instrumental variable strategy. 5.1 Difference-in-Differences Our goal is to understand how local economic activity responds to the advent of containerization. Specifically, we test the theoretical predictions that population and employment increase in locations close to container ports and that these gains attenuate with distance from the port. We also test whether percentage gains are larger in locations with initially low land values, all else equal. Empirically, we ask whether county proximity to a containerized port is associated with changes in key economic outcomes, conditional on a host of covariates. We estimate ∆ ln(y ) = β +β ∆ C +β X +∆ (cid:101) , ( 1 ) i,t 0 1 i,t 2 i i,t where i ∈ I indexes counties and t ∈ T indexes years. Our primary dependent variable, y , is population. We also investigate the impact that containerization has on nominal i,t ∆ wages,industrialcomposition,andincome. Theoperator denoteslongrundifferences, so that ∆ ln(y ) = ln(y )−ln(y ). 13 Capital letters denote vectors. i,t i,t i,1950 13 WhenweuseCountyBusinessPatternsdata,theinitialyearis1956. 12

Our key explanatory variable is an indicator for proximity to a containerized port ∆ 1950 at time t, C , which is equivalent to C , as no containerized ports existed in i,t i,t (C = 0 ∀i ∈ I). We allow for potential non-linear impacts of proximity to a coni,1950 tainerized port by using indicator variables for port proximity by distance bin. Figure 3 100 a shows this parameterization. Counties in the darkest blue are located within km 100 200 of a containerized port, counties in mid-blue are between and km from a con- 200 300 tainerized port, counties in light blue are between and km from a containerized 300 port, and counties in light pink are more than km away from a containerized port. Mathematically, we parameterize proximity to a containerized port as β 1 ∆ C i,t ≡ ∑ β 1,d 1{Closest containerized port is between d 1 and d 2 km} i,t , ( 2 ) d∈D where d ∈ D are a set of distance bins of {0−100,100−200,200−300} kilometers. We interpret β as the percentage change in the dependent variable for counties 1,{0−100} 100 300 within km of a containerized port relative to counties more than km away from a containerized port, conditional on covariates. Coefficients β and β 1,{100−200} 1,{200−300} refer to the remaining distance bins. Theory suggests that population increases in counties proximate to container ports (β > 0). Inaddition,standardNewEconomicGeographymodelspredictthatcontainer- 1 ization’simpactattenuateswithdistancefromtheport,sothat β > β > 1,{0−100} 1,{100−200} 14 β . However,theorydoesnotclearlypredictwheretheimpactofcontaineriza- 1,{200−300} 300 tion stops, so this bound of km comes from the data (see a more detailed discussion 62 24 on this in Section . , footnote ). To establish the causal effects of containerization on local economic activity, we must contend with the non-random assignment of containerized ports to counties. The 14 Thisframeworkdoesnotallowustodistinguishbetweengrowthandreallocation. Seefootnote21for adiscussionofthemagnitudeofreallocationrequiredforgrowthtobenegligible. 13

1 difference-in-differences specification in Equation ( ) goes some way to this end by netting out any time-invariant county-specific characteristics correlated with the location of containerized ports. Such characteristics include geography, proximity to population centers, climate, and historical antecedents for the location of particular industries. This 1950 method also nets out any national changes that impact all counties equally from to 2010 . In the event that county proximity to a containerized port is also a function of timevarying county attributes, we also include a vector of baseline covariates, X . Including i initial covariates in the difference-in-differences model is akin to allowing for differential trends in the dependent variable by the initial covariates. We list these in greater detail 6 in Section , but X includes regional fixed effects, distance to the ocean, measures of i 1953 geographic proximity to ports in , the extent of the initial transportation network, 1950 initialdemographiccharacteristics,initialindustrymix,andpre- countypopulation. 2010 Weclusterstandarderrorsthroughoutatthe commutingzonetoaccountforspatial dependence in the error. A commuting zone is a grouping of counties that approximate 44 15 a local labor market. The average commuting zone includes . counties. This empirical strategy yields a causal estimate of the effect of proximity to a containerized port on local economic activity when proximity to a containerized port is uncorrelated with the error term. This is equivalent to saying that β can be interpreted 1 asacausalestimatewhenproximitytoacontainerizedportisrandomlyassigned,conditional on time-invariant county-level factors and the included initial covariates. Because we include a host of initial period covariates, these estimates cannot be driven by, for ex- 15 We have also made standard error estimates with the spatial HAC method, using radii of 100, 200 and300km. Becausethesestandarderrorsareingeneralsmallerthanthoseusingcommutingzones,and because these spatial standard errors are not (to the best of our knowledge) yet available for the instrumental variable case, we use commuting zone clustering throughout. Even in principle, commuting zone clustering may be preferred, as commuting zone counties are linked by economic activity and therefore likelytobespatiallycorrelated. Incontrast,countieswithinafixedradiusmaybelesslikelytoberelated inaneconomicallymeaningfulway. 14

ample, regional trends in population growth, or differential population growth related to proximity to the coast. To test the predictions that gains vary by initial conditions, we introduce an interaction term that allows β to vary below the median of a given covariate. Call this 1 covariate h and let H = 1 when h < median(h ) and 0 otherwise. 16 We therefore i i i i 1 modify Equation ( ): ∆ ln(y ) = γ +γ ∆ C ++γ ∆ C ∗ H +γ X +γ H +∆ (cid:101) . ( 3 ) i,t 0 1 i,t 2 i,t i 3 i 4 i i,t Now γ reports the average impact of proximity to a container port on population 1 growth, and γ reports whether there is any incremental population gain or loss in 2 counties when h is below the median. We expect containerization induced population i growth to be larger, in percentage terms, in locations with low initial population and low initial land values. We therefore anticipate γ > 0 when h is a measure of initial 2 i land values or population. 1 3 While both equations ( ) and ( ) net out county-specific time-invariant factors as well as trends by initial conditions – including distance to the ocean and initial share of employment in manufacturing – it may still be the case that an element in the error ∆ (cid:101) remains correlated with both containerization and the outcome variable of interi,t est. For example, if counties near container ports were more likely to specialize in an 1950 agricultural commodity that became tradeable since the s, we could conflate local economic growth due to the increase in the trade of the agricultural commodity with local economic growth related to containerization. 16 H relative to the overall distribution and H relative to the treated distribution are both of interest. i i Weconsiderbothempirically;inpracticethedifferenceinestimatesisquitesmall. 15

5.2 Instrumental Variables To address this type of concern – and any other remaining non-randomness in the assignment of containerized ports to counties – we use proximity to a very deep port in 1953 ∆ , Z , as an instrument for proximity to a containerized port, C . Specifically, we i i,t instrument proximity to a containerized port with proximity to initially very deep ports as ∆ C = α +α Z +α X +∆ η , ( 4 ) i,t 0 1 i 2 i i,t where α Z is 1 i α 1 Z i ≡ ∑ α 1,d 1{Closest very deep port in 1953 is between d 1 and d 2 km} i . ( 5 ) d∈D Thus, we have three potentially endogenous variables and three instruments. For the 3 interaction specification in Equation ( ), we use both proximity to a very deep port, Z , i and that proximity interacted with being below the median of a given covariate, Z ∗ H , i i as instruments—so, six instruments overall. There are two requirements for the instrument to yield a causal estimate of proximity to a containerized port on local economic activity. The first is a strong relationship 1953 between proximity to a containerized port and proximity to a very deep port in . The second requirement is that, conditional on covariates, proximity to a very deep 1953 port in is uncorrelated with unobserved determinants of changes in local economic 1950 1953 activity from to period t. In other words, proximity to a very deep port in impacts changes in local economic activity only through its impact on proximity to a containerized port, conditional on covariates (Cov(Z , ∆ (cid:101) ) = 0). We discuss each of i i,t these requirements in turn. First, we anticipate that proximity to a containerized port should be strongly related 16

1953 to proximity to a very deep port in because container ships require deeper ports 1 than their predecessors. As Appendix Figure illustrates, container ships are much larger than their predecessors and larger ships sit deeper in the water and thus require greater depth to navigate and dock. It is possible, but quite expensive, to drill, blast or dredge an initially shallow port sufficiently deep to accept container ships. Given enough money and sufficiently lax environmental regulation, a harbor can arguably be made arbitrarily deep. However, port depth is only malleable at great cost. Therefore, initially deep ports have a competitive advantage when technology changes to favor very deep ports. This inability of ports to adjust equally is confirmed by Broeze, who notes that while “ship designers [keep] turning out larger and larger vessels,” and “the engineering limits of port construction and channel deepening have by no means been reached[, t]his, however, may not be said 2002 of the capacity of all port authorities to carry the cost of such ventures” Broeze ( , 175 177 pp. – ). Thus, initial port depth is a key component of the cost of converting a breakbulk port into a containerized port. Our instrument is therefore analogous to a cost shifter instrument often used in the industrial organization literature. Port depth should affect the supply of ports after the advent of containerization, but have no effect on the demand for ports. The intuition that port depth is a key driver of containerization is borne out in prac- 2 tice by containerization’s pattern of adoption. Figure a shows the likelihood that a 300 county becomes proximate to (within km of) a containerized port over time by the 300 195317 maximal depth of ports within km of the county in . Thick lines indicate depths we consider “very deep.” 1953 It is immediately clear that proximity to deep ports in is a strong predictor of 17 We use depth of the wharf in 1953 as our measure of pre-containerization port depth. Results are robusttousinganchorageandchanneldepth,whichtheWorldPortIndexalsoreports. 17

300 proximity to a containerized port at time t. Counties within km of a port with depth 40 300 greater than feet are always within km of a containerized port by the end of 300 35 40 the sample period, as are almost all counties with km of a port to feet deep. 20 300 25 35 Roughly percent of counties within km of a port with depth between and feet are not near a containerized port by the end of the sample period. For counties 300 within km of less deep ports, however, containerization is decidedly not a certainty. 20 Indeed, counties near initially shallow ports—those less than feet deep—are never 300 within km of a containerized port. An alternative way to view the strength of our instrument is to compare Figures 3 3 a and b. The top panel is the map of US counties, where treated counties are blue and deeper blue indicates greater proximity to a containerized port. The bottom panel repeats this map, but re-colors treated counties in green when the instrument predicts treatment. “Predicting treatment” means that a county is both between d and d km 1 2 2010 from the nearest containerized port in and between d and d km from the nearest 1 2 1953 very deep port in . This picture demonstrates that while the instrument frequently fails to predict treatment in the Midwest, it predicts treatment quite accurately on the 18 ocean coasts. Given this evidence of a strong relationship between the endogenous variables and theinstruments,wenowturntothesecondconditionforinstrumentvalidity—thatprox- 1953 imity to a very deep port in affects local economic activity only through its impact on proximity to a containerized port. A key concern with the instrument is that proximity to a deep port may explain changes in county economic activity even before containerization. This is surely true, as ports have long been engines of growth. For this reason, rather than rely on the full distribution of port depth, we use an indicator variable for a county being proximate to 18 WeaddressthiscasewheretheinstrumentfailstopredicttreatmentinSection6.2. 18

a very deep port pre-containerization. Specifically, we call a port “very deep” when it is 30 1953 feet or more deep in . We choose this depth cut-off because the historical record 30 indicatesnoperceivedadvantagetodepthgreaterthan feetinthepre-containerization era. Before containerization, while port depth conveyed some advantage, it was not particularly useful for a port to be very deep given the draft of breakbulk ships. This is clear 1953 even from how data on port depth was collected. The World Port Index’s deepest 40 2015 category is “ feet and above,” while the deepest category in the World Port Index 76 is “ feet and over.” Thus, intuitively, our instrument measures how much more likely a county is to become proximate to a container port if it is proximate to a very deep 1953 port in , conditional on initial covariates. Our specification includes covariates that allow for differential growth trends in the dependent variable by the number of ports in 1953 300 100 within km in km bins and the values of international trade at these ports 1955 100 in , also measured in km bins. Therefore, the instrument captures the impact of proximity to an initially very deep port above and beyond proximity to many ports in 1953 1955 and to high value ports in . 30 Our claim that depths beyond feet were not particularly advantageous to port 1938 success is supported by a number of contemporary commentators. A monograph 30 notes the critical -foot cut-off, arguing that “For the ports with which we are dealing, 30 the -foot channel at low-water will be taken as the minimum standard in relation to 1938 19 the needs of modern ships” (Sargent, ). However, he notes that the cost of making a channel deeper is no small endeavor: “It is a question how far the rest of the world, Europe in particular, is prepared, except in special circumstances, to face the very heavy 1938 21 cost of providing for the needs of the ocean mammoth” (Sargent, , p. ). This author’s focus on the irrelevance of extreme depth is not unique. Even as late as 19 HegoesontowritethatintheU.S.,a35-footdraughtisbecomingstandard(p. 21). 19

1952 , F. W. Morgan argues in Ports and Harbours that beyond a certain level, depth is not a particularly useful feature of a port: The importance for a few ports of maintaining a ruling depth sufficient to ad- 40 mit the largest liners [a draft of feet] emphasizes unduly their importance to the port world. A super-liner which comes into a port every few weeks will, it is true, amplify that port’s tonnage figures by half a million tons or so annually. ...The greater part of world trade by sea and the greater part of the traffic of many ports is concerned with ships of more modest size. Itwouldcertainlybepossibletodeviseaclassificationofportsbythedraught of ship which can be berthed in them. Halifax and Wellington would appear in the first class, and their ability to berth the largest ships is a great asset in wartime. It tells, however, only a little about their normal significance as 15 1952 ports. (p. , Morgan ( )) Thus, pre-containerization, being very deep was not a particularly valuable port attribute. This instrumental variables strategy implies multiple tests for validity. First, if our claimsabouttheroleof“verydeep”portsaretrue,weshouldseenoimpactofproximity to very deep ports on population growth in the pre-containerization era. In addition, in anysub-samplewhereourinstrumentdoesnotpredicttreatment, theinstrumentshould have no direct impact on population growth. Finally, the instruments should not be correlated with potential confounders that might be in the error term. We turn to these tests in the instrumental variables results section. 6 Results With this empirical framework in hand, we now turn to estimation. The first subsection reports summary statistics and the difference-in-differences results. The second subsection presents tests of instrument validity, discusses our main instrumental variable results, and assesses whether the results are robust to alternative specifications. The 20

third subsection tests whether containerization’s impact is larger, in percentage terms, in places with low initial land values. 6.1 Difference-in-Differences We begin with the difference-in-differences specification to test the theoretical prediction that containerization increases local economic activity. The summary statistics in 1 Table illustrate the comparison at hand and preview the main results. The three left- 100 most columns report county means by distance to the nearest containerized port by 300 km bins; the fourth column shows means for all observations within km of a containerized port, and the final column reports means for all other counties, which we call “never containerized.” A county may appear in only one distance bin. The number of observations in the “ever” and “never” columns sum to the total sample size (final row). On average, counties near container ports have experienced about forty years of containerization. The figures on log population in the first rows of this table clearly show that counties near containerized ports were larger pre-containerization and that counties closest to 1910 1950 containerized ports were largest. From to —the pre-containerization years— log population in counties near future containerized ports is larger and increases at a faster rate than in counties farther from future containerized ports. These differences between counties generate a possible bias in the OLS estimation that we address in the IV section. The summary statistics also show some additional differences between counties by proximity to a containerized port. Across census regions, counties near containerized ports are over represented in the Northeast, under represented in the Midwest and West, and about proportionately represented in the South. Counties near containerized ports 1956 had a substantially larger share of workers in manufacturing in , on average. 21

In addition, these summary statistics illustrate our main finding that counties near containerized ports grow at a faster pace after the advent containerization than the average untreated county. This relative increase is visible not only in the population data, but also in the employment and payroll per employee data from the County Business Patterns. 2 Moving to a regression framework, Table presents difference-in-differences results, testing the prediction that proximity to a containerized port is associated with greater 1 population growth after the advent of containerization. Column presents estimates 68 including only regional fixed effects and shows a percentage point increase in pop- 100 ulation growth for counties within km of a containerized port relative to counties 300 35 more than km away from a containerized port. This coefficient declines to per- 100 200 24 centage points for counties between and km from a containerized port and to 200 300 20 percentage points for counties between and km from a containerized port. The remaining columns in this table add additional covariates. To address the concern that counties of different size may grow at different rates—especially since counties 2 near containerized ports are uniformly initially larger—Column controls for log of 1920 1930 1940 population in years , and . We also add controls for the share of population with a college degree and share African American by county, both measured as of 1950 . To isolate the impact of containerization from proximity to the coast, initial port 3 intensity and pre-containerization port prominence, Column adds additional controls. 1953 These are distance to the ocean, three variables for the number ports in within 300 100 1955 km, measured in bins of km, and three variables for the total value of 300 100 international trade at ports within km, again measured in bins of km. Results 20 In this and all estimates in this paper, we cluster standard errors by the 2010 commuting zone to accountforspatialdependenceacrosscounties. Seefootnote15formoredetails. 22

46 decline by about one-third to one quarter, so that the gradient by distance bin is now , 26 16 , and percentage points, respectively. 1956 Finally, we address the higher rates of manufacturing activity near future con- 1 tainerized ports, as seen in Table ). The fourth column includes this variable and measures of the extent of pre-existing transportation networks as controls. Measures of the 1950 s-era transportation are the length of highways, navigable waterways, and railways per square kilometer. These controls have little additional impact on the size of the co- 45 25 13 efficients. We now estimate , and percentage point increases in population with distance to the closest containerized port. These results are consistent with the theoretical predictions of a standard New Economic Geography model: population increases near containerized ports and gains dissi- 21 pate with distance. Population increases are large and decline monotonically, but not linearly, with distance from the containerized port. We defer a detailed discussion of the 300 magnitude of the estimates and the choice of the km border until the presentation of the instrumental variables results. 6.2 Instrumental Variables Although the difference-in-differences specification addresses many confounding factors potentially correlated with both proximity to a containerized port and population growth—such as past population and initial industrial mix—it is possible that some part of the error term remains correlated with the treatment. We now turn to our instrumental variables estimates. We start with the graphical reduced form intuition, proceed to instrumentstrengthandvalidity,followwithinstrumentalvariableresults,andconclude 21 Our estimation does not discriminate between growth and reallocation. In the period between 1950 and 2010, the US population roughly doubled, from about 150 to roughly 300 million. Thus, our results seem very unlikely to be driven exclusively by reallocation, as they would require approximately half of the1950populationtorelocateduetocontainerization. 23

with measures of robustness. Reduced Form: Relating Proximity to Very Deep Ports and Population Growth 2 To give intuition for the instrument variable analysis, Figure b presents a graphical illustration of the reduced form regression (a regression of change in the log of population on the instrument). This figure presents the average log of population over time 300 by initial depth category. Thick lines indicate counties within km of ports that we 1953 300 30 classify as very deep in ; thin lines are counties within km of ports less than 1953 300 feet deep in . We also include a line for counties not within km of a container port. In essence, the estimation asks whether the thicker lines trend upward more after 1956 (the vertical red line) than do the thin lines. This picture shows that the thick lines of counties near very deep ports do, and that the gains are driven primarily by initially smaller counties—the beige and purple lines. Instrument is Strong and Unrelated to Pre-containerization Population Growth 2 5 We already saw from Figure a (discussed in Section ) that the instrument is strong. 1 Appendix Table validates this intuition, reporting coefficients for the three equations that estimate the full first-stage (one equation per distance bin). The table shows the pattern we expect if the instrument is working as we hypothesize: counties that are 1953 betweend tod kmfromtheclosestverydeepportin aremorelikelytobebetween 1 2 2010 d tod kmfromtheclosestcontainerizedportin . Thesecoefficientsonthediagonal 1 2 05 06 are large—in the . to . range—and strongly significant. Thus, even conditional on 1953 the many covariates we use, proximity to a very deep port in remains an important 2010 predictor of proximity to a containerized port in . The lowest F statistic on the 22 59 instruments in any of these three equations is ; the highest is . Our two-stage least squares estimates tables always report the Kleinberg-Paap F statistic, which summarizes 24

2016 the overall strength of the first-stage, as suggested by Sanderson and Windmeijer ( ). 2122 In our main instrumental variable estimates, this F statistic is never smaller than . Given that the instrument is strong, we now turn to three tests for validity. First, we examine whether proximity to a very deep port is related to pre-containerization 4 population changes; given what we have argued, it should not be. Figure shows the 1910 1950 distribution of population change to , conditional on regional fixed effects and 300 distance to the coast. The red line shows the distribution for counties near (within km) of very deep ports, and the blue line the distribution for counties far from very 95 deep ports. These distributions are virtually indistinguishable. The percent confidence interval on a dummy from a regression distinguishing between these two types of 011004 counties is small relative to the first-stage coefficients and covers zero: [- . , . ]. Thus, we find little evidence that proximity to a very deep port impacts pre-containerization population growth, adding confidence in the validity of the instrument. An additional implication of the IV framework is that, in cases where the instrument failstopredicttreatment,theinstrumentshouldalsobeuncorrelatedwiththedependent variable – since the assumption underlying the instrumental variable specification is that the instrument impacts the dependent variable only through the endogenous variable. In our data, proximity to port depth fails to predict proximity to containerization in the 1953 Great Lakes region. Ports in this area were not very deep in , yet regional ports did adopt containerization. If the proximity to deep ports impacts population and other outcomes only through proximity to containerization, then in cases where port depth is unrelated to containerization, it should also be unrelated to population changes (see 2010 798 Angrist et al. ( ), page ). 300 Limiting our analysis to the roughly seven hundred counties within km of the 22 These first-stage results are also qualitatively robust to defining “very deep” as one category above (greater than 35 feet deep) or one category below (greater than 25 feet deep). The F statistics are larger, andtheestimatesmoreprecise,whenweusethelowerdepthcut-off. 25

Great Lakes, we find a very weak relationship between proximity to port depth and proximitytocontainerization. Further,weseenorelationshipbetweenproximitytodeep ports and population growth. The coefficients on the instrument in the reduced form specification are an order of magnitude smaller than the main estimates (coefficients by 0040 0078 0050 distancebinare- . , . , and . )andareneverdifferentfromzero. SeeAppendix 2 Table for complete results. Our third test of instrument validity evaluates whether the instruments are correlated with county-level characteristics that might plausibly be in the error term. While we cannot do this for all potential confounders, we can observe whether the identifying variation—the residual from a regression of an instrumental variable on the full set of 2 covariates from Table —is correlated with specific pre-treatment covariates, also conditional on covariates. 1920 1930 Recall that our regression specification controls for log of population in , , 1940 and . Were the identifying variation in the instrument to be related to the log of 1910 population (conditional on covariates), this would suggest that the pre-treatment controls were not adequately capturing the historical pattern of population growth. We do not find this to be the case. We do a similar analysis for international trade at ports. 1955 Recall that the regression controls for the value of international trade flows in each of the three distance-to-containerized-port bins. If this covariate did not sufficiently controlfortheimpactofpre-containerizationportstrengthonpopulationgrowth, wewould 1948 expect that the identifying variation would be related to the value of international 23 trade flows by distance-to-containerized-port bins, conditional on covariates. 2 Appendix Figure displays the full matrix of scatterplots showing the correlation 1910 1948 between population and trade and the identifying variation. There are no 23 An alternative method is to include these controls directly in the regression, and results are robust to doing so. We believe that this test, however, highlights the econometric implication of this lack of importance: thattheidentifyingvariationisnotcorrelatedwithlikelyconfounders. 26

significant relationships, and the largest t value for any of these relationships is 2e−8. Instrumental Variable Results Consistent with Difference-in-Differences Findings Given these tests of validity, we report instrumental variable results in the right half of 2 Table . The columns repeat the pattern of covariates from the OLS half of the table. The coefficients are generally quite similar, though slightly larger than the OLS in the 4 8 complete specification (columns and ). Why might IV results be larger? As discussed 3 in section , we expect containerization to have a larger impact on population growth in initially smaller counties. When we use the instrument to correct for endogeneity in the proximity to a containerized port, we are in principle giving more weight to initially smaller counties where the depth is the main driver of the containerization decision. As a result, coefficients in the IV regression increase. 8 53 The most complete model in column shows a percentage point increase in pop- 60 1950 2010 100 ulation growth over the years from to for counties within km of a 300 containerized port relative to counties more than km away from a containerized port. Consistent with the expected relationship between the gains to containerization 29 20 and distance from the port, this coefficient declines to and percentage points for 24 counties slightly farther from containerized ports. 2012 To interpret the magnitude of these results, we turn to Duranton and Turner ( ), 100 13 who find that a % increase of a city’s initial stock of highways yields a percent 20 increase in population over a year period. This corresponds to an annualized increase 24 BothhereandintheOLSestimates,wecomparecountieswithin300kmofacontainerizedporttoall othercounties. Astheorydoesnotprovideguidanceonthephysicaldistanceoverwhichcontainerization might have a measurable impact, we turn to the data as a guide. Appendix Figure 3 shows regression coefficients from a version of Equation (1) where distance to containerized port is measured in 50 km bins. Gray bands are confidence intervals. These results show that the association between proximity to a containerized port and population growth is indistinguishable from zero at 300 km. In our main specification, we use bins of 100 km, rather than the smaller 50 km ones, to increase the power in the estimates. This is particularly important when we examine whether containerization’s impacts differ by initialconditionsinsubsection6.3. 27

06 100 of about . percent. Our findings are similar. Being within km of a containerized portcausesa 70 percentincreaseinpopulationovera 60 yearperiod(exp(.53)−1 = .70), implying a comparable annual growth rate. Our containerization effect is thus roughly 25 equivalent to a doubling in the initial stock of highways in a county. Containerization’s Impact Increases Over Time To test for changes in the impact of containerization over time, we re-estimate Equation 1 1970 ( ) using different final years, starting in . We report coefficients from these esti- 4 mations in Appendix Figure , which displays results decade-by-decade. Full circles are significant coefficients and hollow circles are insignificant coefficients (at the five percent 100 level). The red line at the top reports the coefficients for counties within km of a con- 100 200 200 300 tainerized port; the orange line to , and the yellow line to km. Apart from 1980 ablipin ,countiesnearcontainerizedportshavelargepopulationgainsthatincrease 1970 15 over time. For example, in , only years after the advent of containerization, coun- 40 ties closest to containerized ports had grown by almost additional percentage points 300 2010 relative to counties more than km away from a containerized port. By , this fig- 55 ure was . While estimates for counties farther from ports are smaller, they also follow this general pattern of increase. This increasing impact decade-by-decade may reflect 1 the increasing size of the containerized port network, as shown in Figure . Results Robust to Additional Considerations 2003 We now turn to threats to identification. Rappaport and Sachs ( ) argue that coastal locations have long been associated with greater economic growth, crediting both in- 25 Containerization required substantial investments. In the years of peak outlays from 1968 to 1973, the U.S. spent about $2015 8 billion of public and private funds on the required port infrastructure (Kendall,1986). Thisisabout$20151.6billionperyear,onefourthoftheannualizedcostoftheInterstate HighwaySystemfrom1956through1991(https://www.fhwa.dot.gov/interstate/faq.cfm,assessedon 08/21/2017). 28

creased productivity and, more recently, better amenities. We can interpret containerization as a productivity-enhancing mechanism that generates part of the Rappaport and Sachs result. However, our estimates show that containerization is more than just coastal proximity: our main results are little changed by the inclusion of a Rappaport and Sachs 3 2 26 coastal indicator (Table column ). 3 To further isolate the impact of containerization from proximity to the coast, Table ’s 3 400 1953 column restricts the sample to counties within km of a port in . The sample 3023 1767 size drops from , to , observations, but the coefficients decline only slightly 1 (compare estimates to column , which repeats the most complete specification from 2 Table ). This suggests that population growth in counties near a containerized port is not driven by a comparison with slower-growing centrally located counties. 1 Furthermore, we know from the summary statistics in Table that counties near containerized ports experience more rapid population growth pre-containerization, and 1956 this trend may have continued after irrespective of containerization. We account 1920 1930 1940 for this in the main estimates by including log population in , , and . Table 3 4 ’s column additionally includes squares of those measures of past population, in the event that previous population impacts population growth non-linearly. Again, the estimates are little changed. As we discussed, our instrument does not predict containerization in the Great Lakes region, which does have container ports. In addition, this region experiences the slowest population growth over our period of analysis. To allay fears that the results are driven 5 by this potentially anomalous treatment of the Midwest, column omits the Midwest 1975 region entirely, leaving , observations. Results in this column are smaller than the originalspecification,butthepatternofdeclinewithdistancetotheclosestcontainerized port remains. Indeed, we should expect smaller coefficients in this estimation because 26 RappaportandSachsmeasurecoastaslocationswithin80kmoftheGreatLakesandoceancoasts. 29

the control group—non-Midwest, non-containerized ports—now has a higher average 0373 population growth. Note the increase in the mean of the dependent variable from . 0508 44 to . (final row of the table). Still, we observe a relative population increase of percentage points near containerized ports, an increase of almost three-quarters of the mean. Research in urban economics strongly suggests that growth is associated with an 2004 5 area’s education and demographic characteristics (Moretti, ). Column includes 25 additional controls for the share of people or older with a high school degree, the 65 share foreign born, the number of government workers per capita, and the share age and older by county. The addition of these covariates decreases the coefficients slightly, with greater impact for the category closest to containerized ports. The coefficients remain sizeable, and retain the pattern of decline with distance to containerized ports. 1956 We conclude this discussion of robustness by considering two additional preinfrastructure investments plausibly correlated with port depth. The first such infrastructure is naval bases. In the US, large military installations may promote local economic activity. If growth-yielding federal investments were concentrated near very deep ports, this could bias the coefficient on proximity to containerization upward. When we 1 300 re-estimate Equation ( ) using instrumental variables, omitting counties within km of any naval base, coefficients are slightly larger and statistically indistinguishable from 27 the main specification. Similarly, if very deep ports were crucial for oil importation, and oil importation caused population growth, our estimate of β would be biased upward. A number 1 27 As of the 1950s, the US had four domestic naval bases, at least 10 naval stations, and over 250 total facilities,whichincludeshospitals,teststations,airstations,andalargevarietyofotherinstallations(U.S. Department of the Navy, 1952, 1959). Naval bases were Pearl Harbor, HI; San Diego, CA; Norfolk, VA and New London, CT. New London was actually taken out of “base” status between 1952 and 1959, but we include it for completeness. Relative to naval bases, naval stations are smaller, serve more limited purposes, and receive less investment (Coletta, 1985). Naval stations are so numerous that 300 km bands aroundthemareindistinguishablefromcoastallocations;seeourcontrolforcoastallocationsinTable3. 30

1948 90 of factors argue against this interpretation. First, as of , percent of US oil was 62 produced domestically and the US accounted for percent of the world oil market 1950 4 1970 (Mendershausen, , p. ). It was not until the s, almost two decades after the advent of containerization, that the US was no longer able to fulfill oil demand with domestic oil. Furthermore, port depth is not a key determinant for suitability as an oil port, allaying concerns about the validity of the instrument. During the period of domestic oil hegemony, most oil moved by pipeline, rather than by ship. Even when oil importation grew, port depth was not as crucial, because oil ships connect to offload via a pipeline, which can be quite long. Therefore, ships need not dock directly at the harbor to offload 1960 oil. Further, until the Suez Canal was dredged in the mid- s, it did not allow vessels 37 2010 43 with a draft deeper than feet (Horn et al., , p. ). Our analysis of robustness concludes by turning to a dataset of world ports and world cities to assess containerization’s global impact. We focus primarily on the United States in this paper because of the rich data available at a relatively small geographic scale. However, containerization is clearly a global phenomenon, and one that may have hadanevenlargerimpactoneconomicactivityincountriesotherthantheUnitedStates. We use world population and port data to estimate regressions that parallel our main 4 4 US regressions. We report results in Table . Columns reports OLS results controlling 1953 300 for country fixed effects, the number of ports in within km of each city (in 100 1950 km distance bins), distance to the ocean, and log population in . We find that 100 9 cities within km of a containerized port experience a percentage point increase in 1950 2010 300 population growth between and relative to cities more than km away from a containerized port. JustasintheUSsample, weareconcernedthattheassignmentofcontainerizedports to cities is not random, generating bias. Using the same instrumenting technique as in 31

1953 the US sample, we find that, similar to the US, proximity to a very deep port in is 2010 3 strongly related to proximity to a containerized port in (Appendix Table presents 4 summary statistics and Appendix Table shows a strong first stage). The instrumental variable coefficients have the same signs as the OLS results, but are substantially larger. 8 200 In the most complete specification in column , we find that cities within km of a 30 20 containerized port grow by an additional percentage points, or about percent of 200 300 the mean. For cities between and km of a containerized port, we estimate a 11 statistically insignificant increase in population growth of about percentage points. These results are smaller in absolute terms than for the US, likely because we consider a sample of international cities that are relatively larger than the majority of US cities. Containerization’s Impact on Other Economic Outcomes Having shown that proximity to a containerized port causes population growth, we test whether proximity to a containerized port also causes an increase in employment, nominal wages, industrial composition, and income. Using instrumental variables estimation 2 1 5 1956 with the full set of covariates from Table , column in Table confirms that, from 2011 28 to , employment increases more in counties near containerized ports. While only the coefficient for counties closest to a container port is statistically significant, the magnitude and pattern of employment increases is strikingly similar to what we find using Decennial Census population data. However, in comparison to the mean, these figures 113 are substantially smaller. The mean change in log employment over the period is . 037 (see final row), compared to a mean increase in log population of . (see final row in 3 Table ). 2 The dependent variable in Column is nominal first quarter payroll per employee. Proximity to a containerized port is virtually unrelated to nominal payroll per employee. 28 EmploymentdataisfromCountyBusinessPatterns(seedetailsinSection4). 32

3 As discussed in Section , the net effect on nominal wages is theoretically ambiguous becausethehomemarketeffectandthemarketcrowdingeffectgoinoppositedirections. 5 The middle two columns of Table assess whether containerization changed the 3 industrial composition of counties near containerized ports. Column reports the share of employment in manufacturing, the industry most likely to produce products that travel in shipping containers. On average, across all counties, the share of employment 21 1956 2011 in manufacturing declined by about percentage points from to (last row). The coefficients reveal very little evidence of a smaller decline in manufacturing among treated counties. Nonetheless, a more narrow focus on transportation does show relative growth. In 4 column , the dependent variable is the share of employment in transportation services, which is “services which support transportation,” and which includes “air traffic control 29 services, marine cargo handling, and motor vehicle towing”. Relative to the miniscule one-tenth of one percent of employment in this industry on average, counties within 100 km of a container port see a statistically significant gain of three times this mean. 100 Counties more than km away from a container port see no significant change in this sector. Our finding that employment shifts towards transportation services is reminis- 2008 cent of Michaels ( ) who finds that counties connected with highways experience an increase in trade-related activities, such as trucking and retail sales. Finally, in the last three columns, we look at the impact of containerization on the 10 50 90 income distribution. We look at income for the th, th, and th percentiles and find 100 that counties within km of a containerized port experience larger and significant increases in income across the whole distribution. In addition, as with population and overallemployment,thepatternofdeclinewithdistancetotheclosestcontainerizedport 29 For 1956, we use SIC 47 for “services incidental to transportation,” and for 2011 we use NAICS 488 for“supportactivitiesfortransportation.” 33

remains: counties farther away from a containerized port experience smaller additional 300 increases in income relative to counties more than km away from a containerized port. 6.3 Where Gains to Containerization Are Largest In the previous subsections, we show that, on average, proximity to a containerized port causes increases in population and employment. We hypothesize that gains should be greater in initially low land value areas, and this section reports results from testing this claim. 1956 We use three proxies for land values circa . The first is the share of county 1956 employment in the manufacturing sector in . Manufacturing was the high tech 1950 of the s, and we anticipate that productive places should also be high land value 2011 1950 places (Moretti, ). The second proxy is county population density as of , and 1956 the third is assessed land value from the Census of Governments. While this last measure is the closest to a direct measure of the variable of interest, assessed values are notoriously different from market values. Particularly in this period, it was not unusual for assessment practices to vary substantially – and systematically – across jurisdictions 2010 (Anderson and Pape, ). 6 Table reports coefficients on the measure of proximity to a containerized port and coefficients on the interaction of being below the median of variable h and near a coni tainerized port. Again, the dependent variable is the change in log population. The first column shows that half of the containerization-induced population growth in counties 100 within km of containerized ports occurs in counties with lower than median share of workers in manufacturing. For counties slightly farther from the port, almost all of the containerization-induced population growth occurs in counties with lower than median 1956 share of workers in manufacturing in . While no initial condition explains as much 34

of containerization-induced population growth as an initially small manufacturing sector,containerization-inducedpopulationgrowthisalsolargeininitiallylessdenseplaces 2 1956 (column ). We observe no particular pattern in counties with low assessed land 3 values (column ). Overall, these results paint a picture of containerization exerting the greatest influence not in dominant agglomerations—large, wealthy urban areas—but in second-tier agglomerations. These second-tier agglomerations are initially less dense and less con- 1950 centrated in the vanguard technology of the s (manufacturing). This is consistent withcontainerization’sdemandforlargeareasoflandandsuggeststhatcontainerization is easier to implement where land values are initially low. Theseresultsarealsoconsistentwithacomplementarystoryabouttheroleofmarket 2016 access (e.g. Donaldson and Hornbeck, ). This line of argument says that containerization’s impact will be larger, in percentage terms, in areas with initially low market 4 access. This hypothesis is consistent with the results in column , showing larger gains in counties at the bottom half of highway intensity (highways per square km). We see 5 no preferential pattern, however, with railroads (column ). 7 Conclusion Containerization is a fundamental engine of the global economy. Containerization simplifies and speeds packing, transit, pricing, and every transfer from ship to train to truck. It eliminates previously profitable pilferage and makes shipping more reliable. 1956 Since the advent of containerization in , the cost of moving containerizable goods has plummeted. In this paper, we analyze how local economic activity responds to the dramatic decline in trade costs brought by containerization. We use a novel cost-shifter instrument 35

based on the historical depth of ports to show that, consistent with the predictions of a New Economic Geography model, containerization caused substantial population and employment growth in counties near container ports. These gains follow the pattern of decline with distance predicted by theory: counties closer to a containerized port experience larger increases than counties located farther away. Finally, consistent with containerization’s need for substantial land for large cranes and vast marshalling yards, gains are located predominantly in counties with initially low population density and initially low manufacturing employment. Whether and how containerization impacts the location of population, employment, and wages has implications for both the agglomerative forces that drive innovation, and for political representation that yields democratic outcomes. For policymakers to mitigate the uneven impacts of globalization, it is useful to first understand its causes. 36

References 2010 1970 Anderson, Nathan B. and Pape, Andreas D., . “A Reassessment of the s Property Tax Revolt.” Tech. Rep. Working Paper, University of Illinois at Chicago. 2010 Angrist, Joshua, Lavy, Victor, and Schlosser, Analia, . “Multiple Experiments for the Causal Link between the Quantity and Quality of Children.” Journal of Labor 28 4 773 824 Economics ( ): – . 2013 Autor, David H., Dorn, David, and Hanson, Gordon H., . “The China Syndrome: Local Labor Market Effects of Import Competition in the United States.” American 103 6 2121 68 Economic Review ( ): – . 2007 Baum-Snow, Nathaniel, . “Did Highways Cause Suburbanization?” Quarterly 122 2 775 805 Journal of Economics ( ): – . 2016 Bernhofen, Daniel, El-Sahli, Zouheir, and Kneller, Richard, . “Estimating the Effects 98 of the Container Revolution on World Trade.” Journal of International Economics : 36 50 – . 2014 Bridgman, Benjamin, . “Why Containerization Did Not Reduce Ocean Trade Shipping Costs.” Tech. rep. Working Paper. 2002 1950 Broeze, Frank, . The Globalisation of the Oceans: Containerisation From the s 23 to the Present, vol. of Research in Maritime History. St. John’s, Newfoundland: International Maritime Economic History Association. 2017 Campante, Filipe and Yanagizawa-Drott, David, . “Long-Range Growth: Economic Development in the Global Network of Air Links*.” The Quarterly Journal 050 of Economics page qjx . 1960 Chinitz, Benjamin, . Freight and the Metropolis: The Impact of America’s Transport Revolutions on the New York Region. Havard University Press. 2018 Cos¸ar, Kerem and Demir, Banu, . “Shipping Inside the Box: Containerization and 11750 Trade.” Tech. Rep. Discussion Paper No. DP , CEPR. 1985 Coletta, Paolo E., . “Preface.” In Paolo E. Coletta and Jack K. Bauer, (Eds.) “United States Navy and Marine Corps Bases, Domestic,” Greenwood Press. 2016 Donaldson, Dave and Hornbeck, Richard, . “Railroads and American Economic 131 2 Growth: A Market Access Approach.” The Quarterly Journal of Economics ( ): 799 858 – . 2014 Duranton, Gilles, Morrow, Peter M., and Turner, Matthew A., . “Roads and Trade: 81 681 724 Evidence from the US.” Review of Economic Studies : – . 37

2012 Duranton, Gilles and Turner, Matthew, . “Urban Growth and Transportation.” 79 1407 1440 Review of Economic Studies : – . 2009 1967 1975 Feyrer, James, a. “Distance, Trade, and Income - The to Closing of the 15557 Suez Canal as a Natural Experiment.” NBER Working Papers , National Bureau of Economic Research, Inc. 2009 Feyrer, James, b. “Trade and Income – Exploiting Time Series in Geography.” NBER 14910 Working Papers , National Bureau of Economic Research, Inc. 1995 14 95 Helpman, E., . “The Size of Regions.” Papers - , Tel Aviv. 2017 Holmes, Thomas J. and Singer, Ethan, . “Indivisibilities in Distribution.” Tech. Rep. 739 Working Paper , Federal Reserve Bank of Minneapolis. 2010 Horn,Kevin,Wilson,Erin,Knight,Kevin,andDurden,Susan, . “NationalEconomic Development (NED) Manual For Deep Draft Navigation.” Tech. rep., U. S. Army Engineer Institute for Water Resources. 2007 Hummels, David, . “Transportation Costs and International Trade in the Second Era 21 3 131 154 of Globalization.” Journal of Economic Perspectives ( ): – . 2013 Hummels, David and Schaur, Georg, . “Time as a Trade Barrier.” American 103 7 2935 2959 Economic Review ( ): – . 1986 Kendall, L.C., . The Business of Shipping. Cornell Maritime Press, fifth edn. 2013 Kovak, Brian K., . “Regional Effects of Trade Reform: What Is the Correct Measure 103 5 1960 1976 of Liberalization?” American Economic Review ( ): – . 1992 Kuby, Michael and Reid, Neil, . “Technological Change and the Concentration of 1970 88 68 3 272 289 the U.S. General Cargo Port System: - .” Economic Geography ( ): – . 2008 Levinson, Mark, . The Box: How the Shipping Container Made the World Smaller and the World Economy Bigger. Princeton, NJ: Princeton University Press. 1950 1949 1950 Mendershausen, Horst, . “Dollar Shortage and Oil Surplus in - .” Tech. 11 Rep. No. , International Finance Section, Department of Economics and Social Institutions, Princetion Univeristy. 2008 Michaels, Guy, . “The effect of trade on the demand for skill: Evidence from the 09 683 701 Interstate Highway System.” Review of Economics and Statistics : – . 2004 Moretti, Enrico, . “Human Capital Externalities in Cities.” In Vernon Henderson and Jacques Thisse, (Eds.) “Handbook of Regional and Urban Economics,” Elsevier. 2011 Moretti, Enrico, . “Local Labor Markets.” In David Card and Orley Ashenfelter, 4 1237 1313 (Eds.) “Handbook of Labor Economics,” vol. , pages – . Elsevier. 38

1952 Morgan, F. W., . Ports and Harbours. London, UK: Hutchison’s University Library. 1953 1953 National Geospatial-Intelligence Agency, . “World Port Index, .” H.O. Pub. No. 950 , First Edition. 2015 2015 NationalGeospatial-IntelligenceAgency, . “WorldPortIndex, .” Twenty-Fourth Edition. 2017 Pascali, Luigi, . “The Wind of Change: Maritime Technology, Trade, and Economic 107 9 2821 54 Development.” American Economic Review ( ): – . 2003 Rappaport, Jordan and Sachs, Jeffrey D, . “The United States as a Coastal Nation.” 8 1 5 46 Journal of Economic Growth ( ): – . 2008 Redding,StephenandSturm,DanielM., . “TheCostsofRemoteness: Evidencefrom 98 5 1766 1797 German Division and Reunification.” American Economic Review ( ): – . 2017 Redding, Stephen J. and Rossi-Hansberg, Esteban, . “Quantitative Spatial Eco- 9 1 21 58 nomics.” Annual Review of Economics ( ): – . 2015 Redding, StephenJ.andTurner, MatthewA., . “TransportationCostsandtheSpatial Organization of Economic Activity.” In Gilles Duranton and William Strange, (Eds.) “Handbook of Regional and Urban Economics,” Elsevier B. V. 2017 Rodrigue, Jean-Paul, . The Geography of Transport Systems. Hofstra University, Department of Global Studies and Geography. 1999 Romer, David H. and Frankel, Jeffrey A., . “Does Trade Cause Growth?” American 89 3 379 399 Economic Review ( ): – . 2014 Rua, Gisela, . “Diffusion of Containerization.” Finance and Economics Discussion 88 Series , Board of Governors of the Federal Reserve System (U.S.). 2016 Sanderson, Eleanor and Windmeijer, Frank, . “A weak instrument F-test in linear IV 190 2 212 221 modelswithmultipleendogenousvariables.” JournalofEconometrics ( ): – . 1938 Sargent, A. J., . Seaports and Hinterlands. London, UK: Adam and Charles Black. 2010 Topalova, Petia, . “Factor Immobility and Regional Impacts of Trade Liberalization: Evidence on Poverty from India.” American Economic Journal: Applied Economics 2 4 1 41 ( ): – . 2013 United Nations Conference on Trade and Development, . Review of Maritime Transport. New York, NY: United Nations. 1952 U.S. Department of the Navy, . “Catalog of Naval Shore Activities.” Tech. Rep. 213 105 OPNAV P - . 39

1959 U.S. Department of the Navy, . “Catalog of Naval Shore Activities.” Tech. Rep. 09 23 105 OPNAV P B - . 1982 Wilson, R. P., . “The Containerization of Ports.” UC Santa Cruz Working Paper. 40

1 1956 2008 Figure : Adoption of Containerization: – (a) United States (b) Worldwide ← Int'l standard ← Int'l diffusion plateau: 90% of countries have >= 1 container port ediwdlrow strop reniatnoc fo rebmuN 006 004 002 0 1960 1970 1980 1990 2000 2010 Year Note: TheupperpanelshowsthediffusionofcontainerizationacrossUSports;thebottompanelrepeats thisexerciseforworldports. Source: ContainerisationInternationalYearbook,volumes1968and1970–2010. 41

2 Figure : Graphical Intuition (a) First Stage: Depth and Likelihood of Containerization (b) Reduced Form: Depth and Population Changes Notes: Inbothfigures,thicklinesdenotedepthsthatwelabel“verydeep”inourestimation. Figure2a showsthelikelihoodthatacountywillhaveacontainerizedportwithin300kminyear t bythedepthof thedeepestportwithin300kmin1953. Onaverage,deeperportsaremorelikelytoevercontainerize, andmorelikelytocontainerizeearly. Figure2bplotsthelogarithmofpopulationovertimebythedepth ofthedeepestportwithin300kmin1953. 42

3 Figure : Geographic Variation in Treatment and Instrument 2010 (a) Counties Near a Containerized Port in 2010 1953 (b) Counties Near a Containerized Port in and Near a Very Deep Port in Notes: Figure3ashowsthedistancetothenearestcontainerizedportin2010. Bluepolygonsarecounties d to d kmfromthenearestcontainerizedport. Distancebins {d ,d } are{0to100,100to200,200to 1 2 1 2 300}. Figure3bshowsthedistancetothenearestcontainerizedportin2010aswellasthedistancetothe nearest“verydeep”portin1953. Greencolorsrepresentcountiesthatare d to d kmfromthenearest 1 2 containerizedportand d to d kmfromthenearest“verydeep”portin1953. 1 2 43

4 Figure : Port Depth Unrelated to Pre-Containerization Growth Notes: Thispictureshowsthedistributionofcountypopulationchange1910to1950,conditionalon regionalfixedeffectsanddistancetotheocean. Countiesnearverydeepportsareinredandthosenot nearverydeepportsareinblue. Regressionsresultsshownosignificantdifferencebetweenthesetwo means. 44

1 Table : County Characteristics by Distance to Nearest Containerized Port Distance to Containerized Port, km 100 200 to to Ever Never 0 100 to 200 300 Cont. Cont. 1 2 3 4 5 ( ) ( ) ( ) ( ) ( ) Log Population 1910 1031 1003 1002 1011 947 . . . . . 122 082 080 095 096 [ . ] [ . ] [ . ] [ . ] [ . ] 1950 1081 1023 1014 1036 958 . . . . . 147 097 097 116 096 [ . ] [ . ] [ . ] [ . ] [ . ] 2010 1170 1075 1052 1094 979 . . . . . 150 116 115 135 132 [ . ] [ . ] [ . ] [ . ] [ . ] Log Employment 1956 902 819 804 837 718 . . . . . 194 144 145 165 143 [ . ] [ . ] [ . ] [ . ] [ . ] 2011 1037 931 908 953 835 . . . . . 183 145 147 166 155 [ . ] [ . ] [ . ] [ . ] [ . ] Log Payroll Per Employee 1956 027 037 040 035 050 - . - . - . - . - . 033 029 031 031 032 [ . ] [ . ] [ . ] [ . ] [ . ] 2011 219 204 202 208 197 . . . . . 029 020 019 024 022 [ . ] [ . ] [ . ] [ . ] [ . ] Region 019 017 012 016 000 Northeast . . . . . 019 028 038 029 039 Midwest . . . . . 049 048 045 047 043 South . . . . . 013 007 005 008 017 West . . . . . Share Employment, Manufacturing 1956 042 041 042 042 026 . . . . . 019 019 020 019 022 [ . ] [ . ] [ . ] [ . ] [ . ] 2011 010 015 014 013 010 . . . . . 009 012 012 011 012 [ . ] [ . ] [ . ] [ . ] [ . ] 370 523 442 1335 1688 Observations Note: Thistablereportsmeansandstandarddeviationsinbrackets. Thenumberofobservationsatthe bottomofthetableappliestoallvariablesexceptthe1910populationandthepayrollandemployment variables;eachhasslightlyfewerobservations. 45

2 Table : Containerization Associated with Increased Population, Particularly Near the Port OLS IV 1 2 3 4 5 6 7 8 ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) Closest container port is 0 100 0684 0604 0464 0453 0685 0642 0410 0529 to km . *** . *** . *** . *** . *** . *** . ** . *** 0064 0063 0095 0094 0082 0087 0187 018 ( . ) ( . ) ( . ) ( . ) ( . ) ( . ) ( . ) ( . ) 100 200 0348 0351 0256 0249 0237 0371 0219 0285 to km . *** . *** . *** . *** . *** . *** . . * 0054 0053 0078 0076 0086 0087 0154 0147 ( . ) ( . ) ( . ) ( . ) ( . ) ( . ) ( . ) ( . ) 200 300 0235 0202 0156 0132 0215 0267 0175 0204 to km . *** . *** . ** . * . ** . *** . . 0057 0056 007 007 0097 0102 014 0139 ( . ) ( . ) ( . ) ( . ) ( . ) ( . ) ( . ) ( . ) Covariates Regional fixed effects x x x x x x x x Demographics x x x x x x 1920 1940 Log of population, - x x x x x x Distance to the ocean x x x x 1953 Number of ports x x x x 1955 Total int’l trade at ports, x x x x 1950 s-era transportation x x Share manufacturing employment, 1956 x x 0186 0328 0356 0372 0183 0327 0355 0371 R-squared . . . . . . . . 991 957 21 211 Kleinberg-Paap F Stat . . . Notes: Starsdenotesignificancelevels: *0.10,**0.05,and***0.01. Allregressionsuse3,023observationsandclusterstandarderrorsatthe 2010commutingzone. Thedependentvariableisthechangeinlogpopulation,1950-2010. Themeanofthedependentvariableis0.373. DemographicsisshareofpeoplewithacollegedegreeormoreandshareAfricanAmerica,bothmeasuredasof1950. Numberof1953 portsandtotalinternationaltradeatportsin1955arebothvectorswithtotalsby100kmbins. 1950s-eratransportationisavectorwhich measuresthekilometersofhighwaysc. 1960,kilometersofnavigablewaterways,andkilometersofrailroadsc. 1957ineachcounty,allper squarekilometeroflandarea. Seedataappendixforcompletedetailsonyearsandsources. 46

3 Table : Impact of Containerization Robust to Alternative Specifications 400 Main spec., Within Omit Additional With R & S Squares of 2 1953 Table , col. km of a Midwest demographic coast control population 8 port region covariates 1 2 3 4 5 6 ( ) ( ) ( ) ( ) ( ) ( ) Closest Container Port is within 0 100 0529 0423 0510 0448 0443 0449 to km . *** . ** . ** . ** . ** . ** 018 0201 0202 0182 0201 0183 ( . ) ( . ) ( . ) ( . ) ( . ) ( . ) 100 200 0285 0260 0237 0231 0174 0236 to km . * . * . . . . 0147 0146 0176 0147 0166 0145 ( . ) ( . ) ( . ) ( . ) ( . ) ( . ) 200 300 0204 0205 0228 0171 0109 0164 to km . . . . . . 0139 0139 0173 0139 0146 0136 ( . ) ( . ) ( . ) ( . ) ( . ) ( . ) 0357 0363 0311 0371 0293 0371 R-squared . . . . . . 213 177 214 216 246 213 Kleinberg-Paap F Stat . . . . . . 3023 3023 1767 3023 1975 3023 Observations 0373 0373 0514 0373 0508 0373 Mean, dependent variable . . . . . . Notes: Starsdenotesignificancelevels: *0.10,**0.05,and***0.01. Allspecificationsinstrumentalvariableregressionswithclustered standarderrorsatthe2010commutingzone. Logofpopulationisthedependentvariableandallregressionsincludethemostcomplete covariatelistfromTable2. Column1repeatsthemostsaturatedestimationfromTable2Column8. Column2controlsfortheRappaport andSachs(2003)measureofcoastalproximity. Column3restrictsthesampletocountieswithin400kmofa1953port. Column4includes squaresof1920,1930and1940population. Column5omitstheMidwestcensusregion,whichhasnoverydeepports. Column6includes additionaldemographiccovariatesmeasuredin1950: shareofpeople25orolderwithlessthanahighschooldegree,shareforeignborn, governmentworkerspercapita,andshareage65andolder. 47

4 Table : Containerization Impacts Growth in World Cities OLS IV 1 2 3 4 5 6 7 8 ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) Closest container port is 0 100 0010 0069 0007 0090 0047 0134 0216 0310 to km - . . . . . . * . * . *** 0056 0056 0066 0065 0071 0070 0122 0120 ( . ) ( . ) ( . ) ( . ) ( . ) ( . ) ( . ) ( . ) 100 200 0040 0035 0037 0015 0188 0165 0310 0307 to km - . - . - . - . . ** . * . ** . *** 0060 0058 0067 0066 0091 0086 0124 0118 ( . ) ( . ) ( . ) ( . ) ( . ) ( . ) ( . ) ( . ) 200 300 0038 0027 0035 0012 0040 0027 0114 0113 to km - . - . - . - . . . . . 0064 0060 0067 0064 0112 0105 0127 0120 ( . ) ( . ) ( . ) ( . ) ( . ) ( . ) ( . ) ( . ) Covariates Country fixed effects x x x x x x x x 1950 Log of population, x x x x Distance to the ocean x x x x 1953 Number of ports x x x x 0655 0684 0663 0690 0648 0678 0652 0680 R-squared . . . . . . . . 437 439 425 436 Kleinberg-Paap F Stat . . . . Notes: Starsdenotesignificancelevels: *0.10,**0.05,and***0.01. Allregressionsuse1,051observations,andtheunitofobservationisa citywithasleast50,000inhabitantsin1950. Thedependentvariableisthechangeinlogpopulation,1950to2010. Themeanofthe dependentvariableis1.54. 48

5 Table : More Employment and Higher Earnings Near Containerized Ports IV, Dependent Variable is All industries Employment Share Log pth percentile income, where p is Log Transporta- Log em- Manufac- 10 50 90 payroll/ tion ployment turing employee Services 1 2 3 4 5 6 7 ( ) ( ) ( ) ( ) ( ) ( ) ( ) Closest container port is 0 100 0347 0049 0019 0003 0175 0276 0152 to km . * . - . . ** . * . *** . ** 0201 0068 0021 0001 0083 0069 0055 ( . ) ( . ) ( . ) ( . ) ( . ) ( . ) ( . ) 100 200 0124 0043 0 0001 0082 0109 0084 to km . . . . . * . + 0156 0044 0017 0001 0067 0055 0045 ( . ) ( . ) ( . ) ( . ) ( . ) ( . ) ( . ) 200 300 002 0018 0008 0 0058 005 001 to km . . . . . - . 0147 0039 0017 0001 0067 0056 0046 ( . ) ( . ) ( . ) ( . ) ( . ) ( . ) ( . ) 0178 0155 076 0075 0298 0436 0324 R-squared . . . . . . . 211 22 211 211 214 214 214 Kleinberg-Paap F Stat . . . . . . 2985 2981 2985 2985 3022 3022 3022 Observations 1135 2448 0215 0001 3547 3147 3176 Mean, Dependent Variable . . - . . . . . Notes: Starsdenotesignificancelevels: *0.10,**0.05,and***0.01. AllspecificationsareinstrumentalvariableregressionswithCounty BusinessPatternsandCensusincomedata,andincludethemostcompletecovariatelistfromTable2. Weclusterthestandarderrorsatthe 2010commutingzone. Thesecondpairofcolumnsreportfewerobservationsbecausesomecountiesaresufficientlysmalltosuppressall payrollinformation. 49

6 Table : Greater Containerization-Induced Growth in Initially Lagging Places 1950 Interaction Variable is Highway 1950 1956 Manuf. Rail km / km / share of Population Assessed county sq county sq Employmt Density Land Value km km 1 2 3 4 5 ( ) ( ) ( ) ( ) ( ) Closest container port is within 0 100 0293 0293 0559 0094 0375 to km . . . ** . . 0224 0192 0255 0218 0231 ( . ) ( . ) ( . ) ( . ) ( . ) 100 200 0072 016 0521 001 0293 to km - . . . ** - . . 0168 0159 0235 0186 0193 ( . ) ( . ) ( . ) ( . ) ( . ) 200 300 0105 0056 0231 0038 0127 to km - . . . - . . 0144 0141 0196 0182 0167 ( . ) ( . ) ( . ) ( . ) ( . ) Container port distance * 1 {County ≤ median(column header variable)} 0 100 0267 0389 0038 0523 0166 to km . * . *** . . *** . 0145 0138 0159 0123 012 ( . ) ( . ) ( . ) ( . ) ( . ) 100 200 0531 0218 0238 0314 0032 to km . *** . ** - . . ** - . 0141 01 0145 0131 0123 ( . ) ( . ) ( . ) ( . ) ( . ) 200 300 0467 0282 0004 0267 0104 to km . *** . ** . . * . 0147 0113 0147 0152 0136 ( . ) ( . ) ( . ) ( . ) ( . ) 0368 0368 0355 0377 0358 R-squared . . . . . 109 106 76 127 69 Kleinberg-Paap F Stat . . . . . 044 171 001 0 007 Median, interaction variable . . . . Share of observations ≤ median 0 100 048 037 035 066 044 to km . . . . . 100 200 049 054 052 081 054 to km . . . . . 200 300 053 057 06 081 050 to km . . . . . Note: Starsdenotesignificancelevels: *0.10,**0.05,and***0.01. AllspecificationsareinstrumentalvariableestimatesofEquation(3)with populationasthedependentvariable. Allregressionshave3,023observationsandclusterstandarderrorsatthe2010commutingzone. The firstpanelofcoefficientsreportstheaverageimpactofcontainerizationbydistancefromtheport;thesecondpanelofcoefficientsreports whetherthereisanyadditionalpopulationgrowthifthecounty’svalueofvariable h in1950isbelowthemedianamongtreated i observations. 50

FOR ONLINE PUBLICATION A Data Appendix A.1 Data Sources We use data from a variety of sources. This appendix provides source information. 1 . County Business Patterns These data include total employment, total number of establishments (with some variation in this definition over time), and total payroll. • 1956 : Courtesy of Gilles Duranton and Matthew Turner. See Duranton et al. 2014 ( ) for source details. We collected a small number of additional counties that were missing from the Duranton and Turner data. – In these data, payroll is defined as the “amount of taxable wages paid for covered employment [covered by OASI, or almost all “nonfarm indus- 30 trial and commercial wage and salary employment” (page VII) ] during 1956 the quarter. Under the law in effect in , taxable wages for covered 4200 employment were all payments up to the first $ , paid to any one employee by any one employer during the year, including the cash value of payments in kind. In general, all payments for covered employment in the first quarter were taxable unless the employee was paid at the rate of 16800 1956 more than $ , per year. For the first quarter of , it is estimated 970 that . percent of total non-agricultural wages and salaries in covered employment was taxable. The taxable proportion of total wages becomes smaller in the later quarter of the year. Data are presented for the first quarter because wages for this quarter are least affected by the provisions 4200 of the law limiting taxable wages to $ , per year.” (page VI, Section III, 1956 Definitions in County Business Patterns report.) • 1967 to 1985 : U.S. National Archives, identifier 313576 . • 1986 to 2011 : U.S. Census Bureau. Downloaded from https://www.census. gov/econ/cbp/download/ – For comparability, we also use total first quarter payroll from these data. 2 . Decennial Census: Population and demographics data by county • 1910 : ICPSR 02896 , Historical, Demographic, Economic and Social Data: The 1790 2002 38 1950 United States, - , Dataset : Census I (County and State) 30 Dataalsoexcluderailroademployment. 51

• 1920 : ICPSR 02896 , Historical, Demographic, Economic and Social Data: The 1790 2002 38 1950 United States, - , Dataset : Census I (County and State) • 1930 : ICPSR 02896 , Historical, Demographic, Economic and Social Data: The 1790 2002 38 1950 United States, - , Dataset : Census I (County and State) • 1940 : ICPSR 02896 , Historical, Demographic, Economic and Social Data: The 1790 2002 38 1950 United States, - , Dataset : Census I (County and State) • 1950 02896 – ICPSR , Historical, Demographic, Economic and Social Data: The 1790 2002 38 1950 United States, - , Dataset : Census I (County and State) 1950 32 – Census of Population, Volume II, Part I, Table . • 1960 : ICPSR 02896 , Historical, Demographic, Economic and Social Data: The 1790 2002 38 1960 United States, - , Dataset : Census I (County and State) • 1970 : ICPSR 8107 , Census of Population and Housing, 1970 : Summary Statis- 4 tic File C – Population [Fourth Count] • 1980 : ICPSR 8071 , Census of Population and Housing, 1980 : Summary Tape 3 File A • 1990 : ICPSR 9782 , Census of Population and Housing, 1990 : Summary Tape 3 File A • 2000 : ICPSR 13342 , Census of Population and Housing, 2000 : Summary File 3 • 2010 : U.S. Census Bureau, 2010 Decennial Census Summary File 1 , Downloaded from http://www2.census.gov/census_2010/04-Summary_File_1/ • 2010 ( 2008 - 2012 ): U.S. Census Bureau, American Community Survey, 5 -Year SummaryFile,downloadedfromhttp://www2.census.gov/acs2012_5yr/summaryfile/ 2008-2012_ACSSF_All_In_2_Giant_Files%28Experienced-Users-Only%29/ 3 . Port Universe and Depth • We use these documents to establish the population of ports in any given year. • 1953 : World Port Index, National Geospatial-Intelligence Agency ( 1953 ) • 2015 : World Port Index, National Geospatial-Intelligence Agency ( 2015 ) 4 . Port Containerization Adoption Year • 1956 – 2010 : Containerisation International Yearbook for 1968 and 1970 – 2010 5 . Port Volume: Total imports and exports by port • 1948 : UnitedStatesForeignTrade,January-December 1949 : Water-borneTrade 1949 byUnitedStatesPort, , Washington, D.C.: U.S.DepartmentofCommerce, 972 Bureau of the Census. FT . 52

• 1955 : United States Waterborne Foreign Trade, 1955 , Washington, D.C. : U.S. 985 Dept. of Commerce, Bureau of the Census. FT . • 2008 : Containerisation International yearbook 2010 , pp. 8 – 11 . 6 . Highways • 2014 : 2014 National Transportation Atlas, Office of the Assistant Secretary for Research and Technology, Bureau of Transportation Statistics, United States Department of Transportation. http://www.rita.dot.gov/bts/sites/rita. dot.gov.bts/files/publications/national_transportation_atlas_database/ 2014/index.html. • c. 1960 : Office of Planning, Bureau of Public Roads, US Department of Commerce, “The National System of Interstate and Defense Highways.” Library of 3701 21 1960 5 Congress Call number G .P .U . Map reports improvement status 31 1960 as of December , . 7 . Railways • 2014 : 2014 National Transportation Atlas, Office of the Assistant Secretary for Research and Technology, Bureau of Transportation Statistics, United States Department of Transportation. http://www.rita.dot.gov/bts/sites/rita. dot.gov.bts/files/publications/national_transportation_atlas_database/ 2014/index.html. • c. 1957 : Army Map Service, Corps of Engineers, US Army, “Railroad Map of 1935 1947 8204 the United States,” prepared , revised April by AMS. Edition 5 3701 3 1957 48 -AMS. Library of Congress call number G .P .U . 8 . Waterways • 2014 : 2014 National Transportation Atlas, Office of the Assistant Secretary for Research and Technology, Bureau of Transportation Statistics, United States Department of Transportation. http://www.rita.dot.gov/bts/sites/rita. dot.gov.bts/files/publications/national_transportation_atlas_database/ 2014/index.html. 9 2014 . World Population Data: World Urbanization Prospects, Revision • Population counts for all urban agglomerations whose populations exceed 300000 1950 2010 , at any time between and . • Produced by the United Nations, Department of Economic and Social Affairs, Population Division. • Downloaded from http://esa.un.org/unpd/wup/CD-ROM/WUP2014_XLS_CD_ FILES/WUP2014-F22-Cities_Over_300K_Annual.xls 53

10 . Property value data • 1956 : 1957 Census of Governments: Volume 5 , Taxable Property Values in the United States • 1991 : 1992 Census of Governments, Volume 2 Taxable Property Values, Number 1 Assessed Valuations for Local General Property Taxation • In both 1957 and 1992 , the Census reports a total figure for the New York City, which consists of five separate counties (equivalent to the boroughs). We attributethetotalassessedvaluefromthecensusofgovernmentstoeachcounty 1956 (borough) by using each borough’s share of total assessed value. For , we rely upon the Annual Report of the Tax Commission and the Tax Department to 30 1956 23 the Mayor of the City of New York as of June , , page which reports 1956 1957 1991 “Assessed Value of All Real Estate in New York City for - .” For , 19 24 we rely upon Department of Finance Annual Report, 1991-1992, pages - . • The District of Columbia is missing an assessed value for 1956 in the Census of Government publication listed above. However, the amount is available in Trends in Assessed Valuations and Sales Ratios, 1956-1966, US Department of 1970 Commerce, Bureau of the Census, March . We use this value. • For 2010 value, we use the sum of the value of aggregate owner occupied stock (American Community Survey) and the aggregate value of the rental occupied stock. As the Census only reports aggregate gross rent, we convert aggregate gross rent to aggregate value of the rental stock by multiplying the 12 aggregate value of the rental stock (by to generate a monthly figure) by 2008 2012 the average rent-price ratio for years - (corresponding with the ACS 31 years) from Lincoln Institute Rent-price ratio data . A.2 Data Choices 1 . U.S. County Sample 1950 2010 Our unit of analysis is a consistent-border county from to . We generate 1950 these counties by aggregating counties. Please see the final Appendix Table 32 for the specific details of aggregation. 1956 100 The County Business Patterns allowed for reporting of only jurisdictions per state, leading to the reporting of aggregate values for agglomerations of coun- 2014 tiesinstateswithmanycounties. SeeDurantonetal.( )fortheinitialcollection 1956 of these data, and additional details. To resolve the problem of making these 31http://datatoolkits.lincolninst.edu/subcenters/land-values/rent-price-ratio.asp 32 These groupings relied heavily on the very helpful work of the Applied Population Laboratory group at the University of Wisconsin. See their documentation at http://www.netmigration.wisc.edu/ datadictionary.pdf. 54

1950 1956 units consistent with the census units, we disaggregate the CBP data in the agglomerated reporting into individual counties, attributing economic activity by population weights. 1950 Alaska and Hawaii were not states in . We omit Alaska from our sample, 1950 because in it has only judicial districts, which do not correspond to modern 1950 counties. We keep Hawaii, where the borders are relatively equivalent to modern counties. We also keep Washington, DC, in all years. We also make a few additional deletions • Two counties that only appear in the data ( 1910 - 1930 ) before our major period 13 041 13 203 of analysis: Campbell, GA ( / ) and Milton, GA ( / ). • Two problematic counties. Menominee, WI ( 55 / 078 ) created in 1959 out of an 08 014 Indian reservation; it has very few people. Broomfield, CO ( / ), created 2001 in from parts of four other counties. • Two counties where land area changes are greater than 40 percent. These are 08 031 56 039 Denver County, CO ( / ) and Teton County, WY ( / ). 2 . County Business Patterns data • For some county/industry groupings, there is a disclosure risk in reporting either the total number of employees or the total payroll. In such cases, we 1974 0 convert the disclosure code (“D” in the years before ) to . • “Payroll” is first quarter payroll. 3 . Income distribution calculations • We use binned income data. In 1950 , the number in each bin is the total 2010 number of families and unrelated individuals. To be consistent, in we also use the total number of families plus unrelated individuals. • Tocalculatepercentiles,weassumethatincomeisuniformlydistributedwithin bins, with the exception of the top bin, which has no top code. • For the top bin, we assume that income is distributed following a Pareto distribution, with a parameter α. We assume that α = max(αˆ,1). Let N be the B number of people in the top income bin, and N be the number of people B−1 in the second highest bin. Similarly, L be the lower bound of the top income B bin and L be the lower bound of the second highest income bin. Then B−1 log(N + N )−log(N ) αˆ = B B−1 B log(L )−log(L ) B B−1 . 55

1 Appendix Figure : Evolution of Ship Sizes WWII technology 134x17x9 1956 1970 First container ships, to s Today, Post-Panamax Source: WWII,authors;remainingships,(Rodrigue,2017). 56

2 Appendix Figure : Instrument Variation vs. Pre-Treatment Covariates: All Instruments (a) (b) (c) (d) (e) (f) Notes: “Identifyingvariation”istheresidualfromaregressionoftheinstrument(countyiswithin d to 1 d kmofa“verydeep”portin1953)onthefullsetofcovariates. AppendixFigures2a,2c,and2eplot 2 theidentifyingvariationversustheresidualofareg5r7essionof1910logpopulationonthefullsetof covariatesfromTable2. AppendixFigures2b,2d,and2fplottheidentifyingvariationversusthe residualfromaregressionoftotaldollarsof1948internationaltradeatportsbetween d to d kmofa 1 2 county,conditionalonthefullsetofcovariates.

3 300 Appendix Figure : IV Estimates Indistinguishable From Zero at km Notes: Thispicturereportscoefficientsfromthespecificationincolumn8ofTable2,butparamaterize ∆C assixindicatorvariables,oneforeachdistancebinof{0to50,50to100,100to150,150to200,200 i,t to250,250to300}km. Eachdotisanestimatedcoefficientfromthisregressionandgraybandsportray the90%and95%confidenceintervals. 58

4 Appendix Figure : Containerization’s Impact Increases Over Time Notes: Thispicturereportscoefficientsfromthespecificationincolumn8ofTable2,butwherethe dependentvariableisthechangeinlogpopulationfrom1950totheyearreportedonthehorizontalaxis andtheendogenousvariableisthechangeincontainerizationstatusfrom1950totheyearreportedon thehorizontalaxis. Eachdotcorrespondstoanestimatedcoefficientbydistancebin. Fullcirclesare significantatthe5percentlevel;hollowcirclesareinsignificantcoefficients. 59

5 Appendix Figure : Depth and Likelihood of Containerization, World Cities {35-40} {30-35} {25-30} {>40} {15-20} {20-25} {10-15} {0-10} t raey yb noitazireniatnoc fo doohilekiL 00.1 57.0 05.0 52.0 00.0 1955 1975 1995 2015 Notes: Linesinthefigurereportthelikelihoodthatacitywillhaveacontainerizedportwithin300kmin year t bythedepthofthedeepestportwithin300kmin1953. Weusethicklinestodrawcountiesnear portsthatweclassifyas“verydeep,”andthinlinesfortheremainderofcities. Thelikelihoodofbeing proximatetoacontainerportisgreaterthecloserthecityistoaverydeep1953port. 60

1 Appendix Table : Complete First Stage Specification 1 if Nearest Container Port is d to d km of county 1 2 0 100 100 200 200 300 to to to 1 2 3 ( ) ( ) ( ) County is d to d of a very deep port 1 2 0 100 0542 0068 0012 to km . *** . - . 0067 0066 0046 ( . ) ( . ) ( . ) 100 200 0015 0605 0013 to km . . *** - . 0034 0049 0042 ( . ) ( . ) ( . ) 200 300 0016 0017 0632 to km - . - . . *** 0027 004 0052 ( . ) ( . ) ( . ) 0584 0462 0416 R-squared . . . 224 59 568 Joint F test, instruments . . 0122 0173 0146 Mean, dependent variable . . . Notes: Allestimationsuse3,023observations. Starsdenotesignificancelevels: *0.10,**0.05,and***0.01. TheFtestvaluesinthistablearefromatestofjointsignificanceofthethreereportedcoefficients. Table2 reportstheKleinberg-PaapFstatistic,assuggestedbySandersonandWindmeijer(2016). 61

2 Appendix Table : Midwest Counties Have No First Stage and Reduced Form Impacts Are Zero First Stage Reduced Form 1 if Closest Container Port is d to d km of county 1 2 Change in Log 0 100 100 200 200 300 1950 to to to Population, 2010 to 1 2 3 4 ( ) ( ) ( ) ( ) County is d to d of a very deep port 1 2 0 100 0065 0332 0248 004 to km . - . * . * - . 0146 0188 0146 0158 ( . ) ( . ) ( . ) ( . ) 100 200 0015 0011 006 0078 to km . - . - . . 0083 0175 0143 0103 ( . ) ( . ) ( . ) ( . ) 200 300 0031 0376 0496 005 to km - . - . *** . *** . 0073 0137 0106 0081 ( . ) ( . ) ( . ) ( . ) 0512 0264 0269 0332 R-squared . . . . 04 53 135 Joint F test, instruments . . . . 016 0329 035 0397 Mean, dependent variable . . . . Notes: Starsdenotesignificancelevels: *0.10,**0.05,and***0.01. ThesampleisrestrictedtotheMidwestCensusregion,whichhasno verydeepports. Allregressionsuse702observationsandclusterstandarderrorsatthe2010commutingzone. 62

3 Appendix Table : World City Characteristics by Distance to Nearest Containerized Port Distance to Containerized Port, km 100 200 to to Ever Never 0 100 to 200 300 Cont. Cont. 1 2 3 4 5 ( ) ( ) ( ) ( ) ( ) Log Population 1950 1252 1203 1205 1232 1198 . . . . . 111 094 088 106 081 [ . ] [ . ] [ . ] [ . ] [ . ] 2010 1397 1355 1360 1381 1361 . . . . . 104 085 080 098 080 [ . ] [ . ] [ . ] [ . ] [ . ] Continent 010 009 009 010 005 Africa . . . . . 036 040 040 038 059 Asia . . . . . 002 000 000 001 000 Australia . . . . . 025 023 023 024 019 Europe . . . . . 018 019 020 019 011 North America . . . . . 009 008 009 009 008 South America . . . . . 373 159 102 634 417 Observations Note: Theunitofobservationinthistableisacitywithatleast50,000inhabitantsin1950. Wereport meansandstandarddeviationsinbrackets. Seedataappendixformoredetailsontheworldsample. 63

4 Appendix Table : Complete First Stage Estimates for World Sample 1 if Closest Container Port is d to d km of city 1 2 0 100 100 200 200 300 to to to 1 2 3 ( ) ( ) ( ) City is d to d of a very deep port 1 2 0 100 0573 0019 0033 to km . *** - . - . 0045 0033 0028 ( . ) ( . ) ( . ) 100 200 0020 0579 0033 to km . . *** - . 0048 0050 0032 ( . ) ( . ) ( . ) 200 300 0006 0099 0511 to km . . * . *** 0045 0047 0055 ( . ) ( . ) ( . ) 0653 0457 0406 R-squared . . . 757 584 377 Joint F test, instruments . . . 0355 0151 0097 Mean, dependent variable . . . Notes: Allestimationsuse1,051observations. Starsdenotesignificancelevels: *0.10,**0.05,and***0.01. TheFtestvaluesinthistablearefromatestofjointsignificanceofthethreereportedcoefficients. Table4 reportstheKleinberg-PaapFstatistic,assuggestedbySandersonandWindmeijer(2016). 64

5 Appendix Table : County Groupings for Consistent Counties Initial Counties Grouped State State County County Name County Notes FIPS FIPS FIPS Used to be part of Yuma 04 027 012 Arizona La Paz County 04 027 County ( / ) Name change, from Dade 12 086 025 Florida Miami Dade County to Miami-Dade, yielded a numbering change. 15 010 005 Hawaii Kalawao County 15 010 009 Hawaii Maui County Yellowstone County merged 30 067 113 Montana Yellowstone County 30 067 is to Park County ( / ) Becomes Carson City 32 510 025 Nevada Ormsby County 32 510 ( / ) Used to be part of Valencia 35 061 006 New Mexico Cibola County 35 061 County ( / ) Is merged into Dewey 46 041 001 South Dakota Armstrong County 46 041 County ( / ) Is merged into Jackson 46 071 131 South Dakota Washabaugh County 46 071 County ( / ) 51 900 003 Virginia Albermarle County 51 901 005 Virginia Alleghany County 51 906 013 Virginia Arlington County 51 902 015 Virginia Augusta County 51 903 019 Virginia Bedford County 51 903 031 Virginia Campbell County 51 904 035 Virginia Carroll County 51 905 041 Virginia Chesterfield County 51 915 053 Virginia Dinwiddie County 51 924 055 Virginia Elizabeth City 65

51 906 059 Virginia Fairfax County 51 907 069 Virginia Frederick Couty 51 904 077 Virginia Grayson County 51 908 081 Virginia Greensville County 51 909 083 Virginia Halifax County 51 905 087 Virginia Henrico County 51 910 089 Virginia Henry County 51 911 095 Virginia James City County 51 912 121 Virginia Montgomery County Is later folded into Suffolk 51 800 123 Virginia Nanasemond City 51 800 County ( / ) 51 913 129 Virginia Norfolk County 51 914 143 Virginia Pittsylvania County 51 915 149 Virginia Prince George County 51 913 151 Virginia Princess Anne 51 916 153 Virginia Prince William County 51 917 161 Virginia Roanoake County 51 918 163 Virginia Rockbridge County 51 919 165 Virginia Rockingham County 51 920 175 Virginia Southhampton County 51 921 177 Virginia Spotsylvania County 51 924 189 Virginia Warwick County 51 922 191 Virginia Washington County 51 923 195 Virginia Wise County 51 924 199 Virginia York County 51 906 510 Virginia Alexandria City 51 903 515 Virginia Bedford City 51 922 520 Virginia Bristol City 51 918 530 Virginia Buena Vista City 51 900 540 Virginia Charlottesville City 51 913 550 Virginia Chesapeake City 51 901 560 Virginia Clifton Forge City 51 905 570 Virginia Colonial Heights City 51 901 580 Virginia Covington City 66

51 914 590 Virginia Danville City 51 908 595 Virginia Emporia City 51 906 600 Virginia Fairfax City 51 906 610 Virginia Falls Church City 51 920 620 Virginia Franklin City 51 921 630 Virginia Fredricksburg City 51 904 640 Virginia Galax City 51 924 650 Virginia Hampton City 51 919 660 Virginia Harrisonburg City 51 915 670 Virginia Hopewell City 51 918 678 Virginia Lexington City 51 903 680 Virginia Lynchburg City 51 916 683 Virginia Manassas City 51 916 685 Virginia Manassas Park City 51 910 690 Virginia Martinsville City Appears for a few years in 51 800 695 Virginia Nanasemond County County Business Patterns data as a county. 51 924 700 Virginia Newport News City 51 913 710 Virginia Norfolk City 51 913 710 Virginia Portsmouth City 51 923 720 Virginia Norton City 51 915 730 Virginia Petersburg City 51 924 735 Virginia Poquoson City 51 912 750 Virginia Radford City 51 905 760 Virginia Richmond City 51 917 770 Virginia Roanoake City 51 917 775 Virginia Salem City 51 909 780 Virginia South Boston City 51 913 785 Virginia South Norfolk City 51 902 790 Virginia Staunton City 51 913 810 Virginia Virginia Beach City 51 902 820 Virginia Waynesboro City 51 911 830 Virginia Williamsburg City 67

51 907 840 Virginia Winchester City Yellowstone Park Is merged into Teton County 56 039 047 Wyoming 56 039 County ( / ) 68