Papers Reg. Sei. 83, 197-228 (2004) DOI: 10.1007/sl0110-003-0183-x -Papers in- Regional Science © RSAI 2004 Cities, regions and the decline of transport costs* Edward L. Glaeser1, Janet E. Kohlhase2 1 Department of Economics, Littauer Center 315A, Harvard University, Cambridge, MA 02138, USA (e-mail: eglaeser@kuznets.fas.harvard.edu) 2 Department of Economics, University of Houston, Houston, TX 77204-5019, USA (e-mail: jkohlhase@uh.edu) Abstract. The theoretical framework of urban and regional economics is built on transportation costs for manufactured goods. But over the twentieth century, the costs of moving these goods have declined by over 90% in real terms, and there is little reason to doubt that this decline will continue. Moreover, technological change has eliminated the importance of fixed infrastructure transport (rail and water) that played a critical role in creating natural urban centres. In this article, we document this decline and explore several simple implications of a world where it is essentially free to move goods, but expensive to move people. We find empirical support for these implications. JEL classification: R12, R14, R23, R41, J30 Key words: transport costs, congestion, spatial distribution of economic activity, concentration and decentralisation, productivity, growth of cities and regions, density 1 Introduction The new regional economics has been built on transport costs. The models of Krug-man (1991a,b), Fujita et al. (1999), Fujita and Thisse (2002) follow the classic firm location literature of Weber (1909, 1929), the central place theories of Christaller (1933, 1966), Lösch (1944, 1954), and the spatial economy approach of Isard (1956), and develop models where economic decisions are both created - and lim- * Glaeser thanks the National Science Foundation, the Taubman Center, and the Rappaport Institute for support and Jesse Shapiro for providing tremendous assistance. Kohlhase thanks Jia-Huey Ling and Anuja Krishnan for helpful research assistance. 198 E.L. Glaeser, J.E. Kohlhase ited - by the costs of moving goods over space. What is the benefit of being close to one another? Being able to buy goods that are produced locally. What is the cost of agglomerating? Having to ship basic commodities farther over space. The foundational models of urban economics are also, if less clearly, founded on transportation technologies. The classic monocentric urban models developed by Alonso (1960, 1964), Muth (1969) and Mills (1967, 1972), hereafter referred to as the Alonso-Mills-Muth model, are consistent with a world where people walk and take public transportation.2 The central business district (CBD) can be viewed as the hub for transportation technologies. Public transportation brings people to the hub and people walk from that point to their work places and use their feet to interact during the workday. The workers' physical output then gets shipped from the hub to consumers using rail and water transport. As such, a monocentric model, where firms are extremely close to one another in the central business district (CBD), is natural when thinking about cities built around feet and trains. Historically, these models capture the essence of urban economies. In the United States transport costs before 1900 were enormously high. People moved by foot and goods were carried by water. Both the structure and location of cities reflect high transport costs. Because roads and rail were rare and costly, every large city in 1900 was located on a waterway. Of the 20 largest cities in America in 1900, seven were ocean ports where rivers meet the sea (Boston, Providence, New York, Jersey City, Newark, Baltimore, and San Francisco); five were ports where rivers meet the Great Lakes (Milwaukee, Chicago, Detroit, Cleveland, and Buffalo); three were on the Mississippi river (Minneapolis, St. Louis and New Orleans); three were on the Ohio river (Louisville, Cincinnati and Pittsburgh); and the remaining two were on east coast rivers, close to the Atlantic (Philadelphia and Washington). America's largest city, New York, was clearly its best port. Moreover, high transport costs created a series of cities scattered throughout the country, each oriented towards exploiting the rich American hinterland. Cronon (1991) details how Chicago was built as a transport hub enabling the movement of commodities such as lumber, wheat and cattle from America's vast hinterland to the east coast and to Europe.3 Moreover, transport costs dictated the existence of a hierarchy of cities (Christaller 1933, 1966; Losch 1944, 1954; Henderson 1974). Small cities dotted the interior of the country and were specialised in providing basic services for men making their living from the earth. Larger cities served as depots for goods coming from and headed to the east coast and Europe. They also increasingly became centres of manufactured goods, which were moved naturally, exploiting their comparative advantage in transportation. Several texts develop the classic regional economics models and location theory models. See Beckmann (1968), Greenhut (1970), Nourse (1968) or Hoover and Giarratani (1985). The last is available online at the web book of regional science http://www.rri.wvu.edu/regscweb.htm. 2 The classic monocentric model is discussed in detail in the text by Fujita (1989). Several summary articles, including Anas et al. (1998) and Brueckner (1987), discuss the basic monocentric model and comparative statics. The agricultural land use model developed by von Thiinen (1826, 1966) can be considered one of the antecedents to the monocentric urban land use models. 3 Other classics which emphasise the link between transport technologies and urban form include the multiple nuclei model of Harris and Ullman (1945) and Blumenfeld (1955). Cities, regions and the decline of transport costs 199 We know this urban world because it still surrounds us. The durability of physical infrastructure ensures us that cities built around Great Lakes still remain even though the lakes' importance as a transport mechanism has declined. Baltimore, Buffalo, Cleveland, Detroit, Pittsburgh and St. Louis (in the Midwest) and Baltimore and Philadelphia (on the coast) excelled in moving the products of the American country by water or rail. These were seven of the 15 largest cities in the country in 1950, and every one of them has lost population in every decade since 1950. They continue to exist, but their gradual weakening should remind us that the world has changed, and the factors that made these places centres of productivity in 1900 seem unimportant 100 years later. The great force that has reshaped the city in the twentieth century is the engine; it has become both more powerful and noticeably lighter over the last 100 years.4 As a result, people have increasingly been able to propel themselves and their goods over long distances with better and better machines. The average cost of moving a ton a mile in 1890 was 18.5 cents (in 2001 dollars). Today, this cost is 2.3 cents. At their height, the transportation industries represented 9% of GDP. Today, if we exclude air travel, they represent 2% of national product. Two factors have acted to decrease the importance of transportation costs for goods. First, the technologies designed for moving goods have improved. Second, the value of goods lies increasingly in quality, rather than quantity, so that we are shipping far fewer tons of goods relative to GDP than we have in the past. These reduced costs, and the declining importance of the good-producing sector of the economy, means that in our view, it is better to assume that moving goods is essentially costless than to assume that moving goods is an important component of the production process. The implication of this claim is that the prevailing urban models are poorly suited for the twenty-first century-city, although they do help us to understand the 19th century cities that still surround us. As we contemplate the future of urban and regional economics, it becomes crucial to build a new theoretical paradigm built on forces other than the costs of moving physical products across space. In this article we do three things. First, we review the evidence on the decline of transportation costs. The evidence indicates a large change that appears to be continuing. Certainly it is an exaggeration to claim that moving goods is free, but it is becoming an increasingly apt assumption. We also note that moving people is not free and that the twentieth century has seen a switch from infrastructure-heavy transport (railroads, ports) to infrastructure-light transport (cars, trucks). This also has implications for urban form. Second, we discuss the implications of this change to the analysis of urban and regional economics. The new regional economics is built around fixed cost technologies with substantial transport costs. Population is anchored by a desire to be in close proximity to natural resources. In Krugman (1991) it is the fixed agricultural sector that ties workers to both regions of the economy. These assumptions fit nineteenth century America perfectly, but are inappropriate for the twenty-first century. 4 Mumford (1934, 1963) emphasised the powerful role of technology in transforming social organisation. 200 E.L. Glaeser, J.E. Kohlhase Instead, models should continue to emphasise agglomeration effects, which come ultimately from the benefits of easy access to other people, not from saving transport costs for goods. Natural resources in production are becoming increasingly irrelevant, and the items that provide features to the landscape are consumption-related natural amenities (e.g., warmth) and state or city specific government policies. Third, we empirically test, in Sects. 4 and 5, a number of implications stemming from our approach. People should leave areas that were once centres for natural resource extraction and instead should increasingly live in a small handful of metropolitan areas. The main factors that stop convergence to a single area are heterogeneity of tastes for weather, government policies and historically developed infrastructure. Manufacturing should be located in lower density areas; services should be in areas with greater density. Service firms should locate near their suppliers and customers; manufacturing firms should not. Extreme congestion on urban roads may have negative productivity impacts. In the last section of the article, we synthesize our discussion and offer suggestions for future modelling efforts. 2 Transportation costs over time A standard theoretical element in the new regional economics-new economic geography literature follows Samuelson (1954) and assumes iceberg transportation costs.5 The iceberg specification maintains that the cost of moving goods involves the loss of some proportion of the product during shipment. Obviously, this assumption is a simplification and misses much that is interesting in the economics of moving goods, but nonetheless it provides us with a convenient starting point for thinking about the magnitude of transportation costs. If these costs were truly iceberg, then we could estimate the size of the cost parameter just by examining the share of GDP spent on moving goods. As such, one starting point for thinking about the magnitude of transportation costs is the share of GDP involved in the transportation industry. At a more detailed level, a measure of transport costs is the price of moving a commodity as a fraction of the total value of the commodity. Figure 1 shows the overall share of transport in GDP from the late nineteenth century until today. This sector of the economy includes rail, water and pipeline transportation, trucking and warehousing, air transport, transportation services, and local and interurban passenger transit. Prior to 1929, we use the Martin Series (Historical Statistics of the United States, F-250-261), which includes both transportation and public utilities. Since this series includes components unrelated to transportation, we multiplied the series by 0.67 - the ratio of transportation spending to transportation plus public utilities in 1929. If this ratio was not constant, but rather decreasing between 1870 and 1930 (which seems probable), then transportation spending would be declining more sharply before 1929. 5 See Fujita et al. (1999) (FKV) for a discussion of iceberg transport costs and its use in regional science, urban economics, economic geography, and international trade models. FKV point out (p. 59) that the agricultural land-use model of von Thiinen (1826,1966) contains the predecessor to Samuelson's iceberg transport costs; in that model one of the assumptions was that the oxen pulling the loaded carts ate some of the grain being shipped to the market. Cities, regions and the decline of transport costs 201 Share of GDP in transport 0.10- ■Q- Transportation share of GDP a Without air 0.04- 0.08 0.06- A A 0.02 1850 1900 1950 2000 Year Fig. 1. The share of GDP in transportation industries. Source: Department of Commerce (since 1929), and Historical Statistics of the U.S. (Martin Series) before then As late as 1929 (the first year we have Department of Commerce data available), transportation represented 8% of gross domestic product. By 1990, only 3% of GDP is being spent on transportation. This figure understates the true decline of transportation because air travel, which is overwhelmingly involved in transporting people, not goods, is a major component of transportation expenditures during the later time period. The triangles in the figure represent the transport cost series without air transportation. This figure is, unsurprisingly, almost the same as total transportation expenditures in 1949, but by the 1990s, more than one-quarter of total spending in this category was on air transport. Without that category, transportation represents only 2.3% of GDP in the 1990s. Of course this figure does not truly represent an estimate of iceberg costs, even in the best of circumstances, because a significant fraction of GDP is not shipped. Services tend to involve little freight shipment. Other more physical goods only involve small amounts of shipping (e.g., construction). Moreover, many physical goods are actually consumed at home and not shipped. Since only a fraction of GDP (perhaps one-half) is in physical goods that are traded, the share of GDP spent on transportation is something of an underestimate of the hypothetical iceberg costs, perhaps by as much as one-half. Another reason that these numbers may tend to underestimate the overall importance of shipping costs in the economy is that they exclude shipping that is done in-house. When a manufacturing firm hires an external shipper, that payment is included in the share of GDP in the transportation industries. When a firm uses its own trucks, the salaries of the trucks will not be attributed to the transportation industry. Furthermore, to the extent that the government subsidises the trucking industry through the construction and maintenance of roads, those costs will not be 202 E.L. Glaeser, J.E. Kohlhase Freight bill divided by GDP 0.09- 0.08- 0.07- 0.06- 1960 1970 1980 1990 2000 Year Fig. 2. Transportation bill (freight only) divided by GDP. Source: Bureau of Transportation Statistics Annual Reports included in these figures. The Eno Foundation has attempted to calculate a national transportation bill to include all of the different components of transportation expenditures. This bill is not exactly comparable to GDP, because if this procedure were done for all of the different functions of the economy, the sum of the results would add to more than GDP. Figure 2 uses the Eno Foundation data to show the shift in the ratio of the nation's freight bill to GDP between 1960 and today. This graph shows a decline from 0.09 in 1960 to 0.06 today. Most of the decrease took place between 1960 and 1990, and little change in the share has occurred in the 1990s. While this is certainly a larger number than the 2.3% of GDP cited earlier, it nevertheless displays a significant downward trend over time. These aggregate movements combine together several different changes: (1) movements in the real cost of moving goods within modes of transport, (2) overall changes in the ratio of goods shipped to GDP, and (3) changes in modes of transport. Indeed, the overall decline in transportation costs has been quite muted because people have moved from a cheaper technology (rail) to a more expensive technology (trucking). We now turn to these different components of the movement in transport costs. Figure 3 shows our longest series on transport costs over time: costs per ton-mile on railroads between 1890 and today. Figures are in 2001 dollars and show a decline from more than 18 cents per ton-mile in 1890 to 2.3 cents today. The decline has been essentially continuous, except for an increase during the 1920s and 1930s. Of course, this graph itself does not control for the average length of the haul and other factors, which will tend to influence the cost of rail shipping. Nonetheless, Cities, regions and the decline of transport costs 203 Dollars per ton-mile (real) 0.185 0.023 1890 2000 Year Fig. 3. The costs of railroad transportation over time. Source: Historical Statistics of the US (until 1970), 1994, Bureau of Transportation Statistics Annual Reports 1994 and 2002 Revenue per ton-mile 0.406 0.011 aaaaa-a-a- 1947 2000 Year -e- Rail —□- Truck -A-Pipe Fig. 4. Revenue per ton-mile, all modes. Source: Bureau of Transportation Statistics Annual Reports the data does suggest a remarkable reduction in the real cost of shipping goods over the twentieth century. Figure 4 shows the trends in costs for other industries. We have included data since 1947 for trucks and pipeline (water is the missing major mode). These figures illustrate nicely the huge gap in shipping costs between trucks and the other modes 204 E.L. Glaeser, J.E. Kohlhase Dollars per ton-mile 0.16 0.11 1960 1992 Year Fig. 5. Revenue per ton-mile, all modes together. Source: Bureau of Transportation Statistics Annual Reports of transportation. It also illustrates that trucking costs remained essentially constant over much of the time period. Rising fuel prices and a regulated industry kept trucking prices at essentially their 1947 levels through 1985. Since 1985, deregulation has enabled technological change and trucking costs have fallen from 38 cents a ton-mile (in 2001 dollars) to 28 cents a ton-mile in 1999. Since the Motor Carrier Act of 1980, which effectively decontrolled the industry, trucking costs have been falling by 2% per year, which is similar in magnitude to the 2.5% per year decline that rail experienced over the entire time period. Although the low costs of pipe transport make the graph difficult to understand, between 1978 and 1999, the real costs of pipeline transport fell 25% from 2 cents per ton-mile to 1.5 cents per ton-mile. Both before 1975 and after 1978, real pipeline costs fell by about 2% per year. Only during the mid-1970s, when pipeline costs shot up by one-third, did this trend reverse. Overall, across all modes there have been declining costs, and in the absence of outside factors (the oil crisis, government regulation) costs per ton-mile, within each mode, appear to be declining by about 2% per year. Figure 5 combines all of the modes and shows a steady downward trend, with the exception of the remarkable year of 1978. Between 1960 and 1992, costs per ton-mile fell from 16 cents to 11 cents, or an average of 0.15 cents per year, or 1.1% per year. This average is declining by somewhat less than the within-mode numbers - in part because of the increasing importance of trucking in the overall share of transportation. The rise of trucking has been a major factor in the postwar transportation industry. As late as 1947, more than 50% of total transportation spending was on rail. Today trucking represents 77.4% of the nation's freight bill (Bureau of Transporta- Cities, regions and the decline of transport costs 205 Ton-miles of freight 1,500,000- 306,000 - 1960 1970 1980 1990 2000 Year Truck Rail Water Pipe Fig. 6. Ton-miles of freight over time. Source: Bureau of Transportation Statistics Annual Reports tion Statistics 1994). Of course, as Fig. 6 shows, rail is still the dominant technology measured in terms of ton-miles and ton-miles by rail are still rising. However, since trucking is more than ten times more expensive on average than rail, it accounts for the lion's share of overall spending on transportation. These numbers tell us the costs of moving a ton of goods one mile (on average), but to understand how big a cost this actually represents, we need to connect this with average length of hauls and with the value of goods transported. Using the 1997 Commodity Flow Survey (Table 1-52, National Transportation Statistics 2002), we have been able to calculate for selected industries the relationship between average transport costs and average value. The Commodity Flow Survey tells us both the average length of haul, by industry group, and the average value per ton in this industry grouping. In Table 1, we then multiply that average haul by 2.4 cents (for rail transport) and 26 cents (for truck transport) to give two different estimates of the costs of transporting the goods. The first column of Table 1 describes the industry; fuller descriptions are available in the commodity flow survey. The second column gives the total value of shipments of these industries in 1997. The third column shows the total ton-miles travelled by this industry and the fourth column gives the value per ton. This is calculated by dividing total value by total tons. Column five shows the average length of haul. In Columns six and seven, we multiplied column five by 2.4 cents and 26 cents, respectively, and then divided by the average value per ton. This calculation is meant to give us the transport cost, relative to value, if the good is shipped by rail and truck, respectively. Naturally, the length of haul is itself endogenous. Commodities with lots of bulk tend not to be shipped far. Indeed, the relationship between value per ton and 206 E.L. Glaeser, J.E. Kohlhase Table 1. Transportation costs and commodity value, selected industries Commodity Value Ton-miles Value per Average Shipping Shipping description ($ billion) (billion) ton ($) miles per costs/value costs/value shipment (Rail) (Truck) Meat, fish, 183.8 36.4 2,312 137 0.001 0.015 seafood, and their preparations Milled grain products, 109.9 48.5 1,069 122 0.003 0.029 preparations, and bakery products Alcoholic beverages 87.9 27.8 1,085 58 0.001 0.013 Tobacco products 56.4 1.0 13,661 296 0.0005 0.006 Gasoline and 217.1 136.6 225 45 0.005 0.052 aviation turbine fuel Basic chemicals 159.6 136.8 539 332 0.014 0.160 Pharmaceutical 224.4 5.6 22,678 692 0.0007 0.008 products Chemical 209.5 45.0 2,276 333 0.004 0.038 products and preparations (NEC) Plastics 278.8 69.1 2,138 451 0.005 0.054 and rubber Wood products 126.4 96.9 384 287 0.018 0.194 Printed products 260.3 22.8 3,335 431 0.003 0.033 Textiles, leather, 379.2 24.7 8,266 912 0.003 0.028 and articles of textiles or leather Base metal in 285.7 117.5 851 276 0.008 0.084 primary or semi- finished forms and in finished basic shapes Articles of 227.2 48.7 2,133 403 0.005 0.049 base metal Machinery 417.1 27.0 8,356 356 0.001 0.010 Electronic and 869.7 27.1 21,955 640 0.0007 0.008 electrical equipment, components and office equipment Motorised and 571.0 45.9 5,822 278 0.001 0.012 other vehicles (including parts) Source: National Transportation Statistics 2002 and authors' calculations assuming that the cost per ton-mile is 26 cents by truck and 2.4 cents by rail. Cities, regions and the decline of transport costs 207 Log of average haul length 7 " 6 - 5 - 4 - 3 - n i i 0 5 10 Log of value per ton Fig. 7. Distance and value per ton. Source: National Transportation Statistics, 2001, Table 1-52 average length of haul is comfortingly tight (shown in Fig. 7). The regression line is: Log(Miles per Average Haul)= 3.22 + 0.32 x Log(Average Dollars per Ton) (0.318) (0.045) (1) where R2 = 0.56, the standard errors in parentheses, and the number of observations is 42. Dollars per ton is the inverse of tons per dollar or the average weight of a fixed value of goods. If the costs of shipment are roughly proportional to weight, then this suggests that as transport costs rise by 10%, the average length of distance between supplier and consumer falls by —3.2%. Despite the endogeneity, these numbers can inform us about the importance of transport costs across a number of industries. Transport costs for some industries still appear to be quite important. For example, if wood products were shipped their average haul of 287 miles by truck, this would cost approximate one-fifth of the value of the shipment. If base metal was shipped its average haul of 276 miles by truck, transport would eat up 8.4% of the value of the commodity. Other commodities, such as basic chemicals or plastics and rubber, also feature significant transport costs, at least if shipped by truck. However, many bigger industries all face trivial transportation costs. For machinery, electrical equipment and transportation equipment costs are always less than 1.2% of total product if shipped by truck and one-tenth of 1% of total product if shipped by rail. These three industries together account for one-quarter of the value of all shipments within the US, and 36% of all shipments (measured by value) fall in this very low cost category. Indeed, these calculations suggest that only 18% of all shipments occur in industries where transport costs are more than 6% of total value - even if all transport was by truck. If we assume that all industries with an 208 E.L. Glaeser, J.E. Kohlhase Ton-miles per dollar of GDP 0.498 □ □ □ □ □ □ □ □ □ □ □ □ n 0.387 □ □ 1965 1999 Year Fig. 8. Ton-miles divided by GDP (in 2001 dollars). Source: Bureau of Transportation Statistics Annual Reports average haul above 400 miles uses rail, then 80% of all shipments (again by value) occur in industries where transport costs are less than 4% of total value. The previous discussion clarifies that effective transportation costs depend as much on value per ton as on the real costs of transport. In general, just as there is a secular trend towards lower real transport costs, there is a secular trend in the value of dollars per shipment. While Fig. 5 showed that shipments were rising, this figure was not normed relative to GDP. In Fig. 8, we show total ton-miles in the economy divided by GDP. As the figure shows, a downward trend reflects increasing value of commodities shipped (and an increasing reliance on the service sector). Not only is transport becoming cheaper, it is also becoming less important relative to GDP Another aspect of the cost of moving goods is the time cost; and other authors have documented the rapid decline of time costs over a 300-year period, 1658-1966, in moving goods between cities. (For an example of the time costs in shipping between Edinburgh and London, see Janelle 1968.) These two effects reinforce each other so that the full-cost of transporting goods - money costs plus time costs - is also rapidly declining over time. Transport costs of course still remain expensive in one area - the movement of human beings. While the car has simplified the movement of people within cities, and the airplane has greatly facilitated movement between cities, these technologies still represent large amounts of our national resources. Indeed, according to the 2001 Consumer Expenditure Survey, 18% of total expenditures for the average household is spent on vehicle purchases, gasoline and other vehicular expenses (e.g., insurance). This cash cost fails to include the far more important time costs of moving people, and these time costs are not withering away with technological Cities, regions and the decline of transport costs 209 progress. Instead, as wages continue to rise, these time costs should rise roughly in proportion to wages. Table 2 reports trends in the movement of people within cities6 over the three-decade period 1980-2000. Data are shown for the 10 largest urbanised areas in the United States and for size-class averages for 75 small, medium, large and very large cities. Various measures of congestion are taken from the Texas Transportation Institutes's (TTI) 2002 Urban Mobility Report (Schrank and Lomax 2002) and include the travel time index, the percentage of daily travel in congestion, and the average annual hours of delay. The travel time index shows the amount of additional time (in percentage terms) that would be required to make a trip because of congested conditions of the roadways. For example, the largest travel time index is 1.90 for Los Angeles in 2000, which implies that it would take a traveller about 90% longer to make a given trip during peak periods than if the person could move at freeflow speeds. Also reported are average travel times for work trips. An overwhelming pattern emerges from examining the table: over the past three decades, congestion and delay have been increasing in all size classes of cities, not just the very large urban areas. For all 75 urbanised areas taken together, on average the annual delay increased by over 280%, with the most dramatic increase occurring for large cities - those between 1 and 3 million - about 450% percent. For small cities, ranging from 100,000-500,000, the delay has increased by over 300%. Real incomes per capita have, at the same time, been rising, about 25% for the combined 75 metropolitan areas. Both trends reinforce our contention that people-moving costs are not declining within US cities. The pattern with respect to commute times is similar, but not as dramatic. Between 1980 and 2000, commute times rose by about 13% for the combined 75 areas. On average commute times are greater in larger cities for each of the decades, but the time pattern differs. Average commute times rose in all 10 of the very large urbanised areas and in all size classes between 1990 and 2000, but a few (New York, Chicago, and Houston) experienced shorter average commutes between 1980 and 1990. We end this section by concluding that transport costs for goods are not negligible, but they are low and getting lower. Across all modes, transport costs are declining and the economy is moving away from producing high bulk products and toward more expensive products where transport costs are less relevant. As such, transportation costs, at least for goods, should play an increasingly irrelevant role in the urban economy. Conversely, as the costs of moving people are certainly not disappearing, these should continue to be a dominating presence in the structure of urban form. An open question remains whether improvements in the Internet and other forms of information technology will reduce the demand for face-to-face contact altogether.7 6 In this table we use the concept of "urbanised area" as defined by the Census where the geographic extent of the city is defined by minimum population density. 7 Gaspar and Glaeser (1998) argue that the implications of information technology for the city depend on whether face-to-face contact is a substitute or complement to electronic interactions (see also Kolko 2000a). 210 E.L. Glaeser, J.E. Kohlhase Table 2. Trends in commuting and congestion for US metropolitan areas0 Metropolitan Areaa Yearb Popu- Travel time Daily Delay per Mean travel Real income lation index travel in person time to per capita (X 1,000) congestion (hrs/yr) work (min) ($1982-1984) (%) Very large ( > 3 million in 2000) New York 1980 15,500 1.13 14 6 32.3 11,088 NY-North- 1990 15,925 1.31 27 18 31.3 14,887 eastern NJ 2000 17,090 1.41 35 23 35.1 15,616 Los Angeles CA 1980 9,900 1.34 31 22 24.1 11,738 1990 11,420 1.91 42 63 26.2 13,587 2000 12,680 1.90 45 62 28.7 12,615 Chicago 1980 7,080 1.19 23 7 31.6 11,716 IL-Northwestern 1990 7,510 1.37 35 18 28.5 13,298 IN 2000 8,090 1.47 40 27 31.2 14,812 Philadelphia 1980 4,090 1.11 16 5 26.0 10,168 PA-NJ 1990 4,370 1.18 23 9 24.9 12,959 2000 4,590 1.28 30 15 28.2 14,286 San Francisco- 1980 3,290 1.21 27 12 25.2 13,219 Oakland CA 1990 3,675 1.50 41 37 26.5 16,293 2000 4,030 1.59 41 41 29.9 18,151 Detroit MI 1980 3,810 1.12 17 7 22.9 11,767 1990 4,000 1.28 30 20 23.1 12,723 2000 4,025 1.34 35 25 26.0 14,702 Dallas-Fort 1980 2,450 1.07 9 6 22.4 11,660 Worth TX 1990 3,150 1.18 19 18 23.5 13,135 2000 3,800 1.33 29 37 26.8 14,262 Washington 1980 2,700 1.18 25 10 28.1 14,355 DC-MD-VA 1990 3,100 1.34 37 22 29.0 17,578 2000 3,560 1.46 40 35 32.2 18,948 Houston TX 1980 2,400 1.28 26 19 26.2 12,540 1990 2,880 1.31 28 18 25.9 12,404 2000 3,375 1.38 33 36 28.4 13,115 Boston MA 1980 2,850 1.14 16 9 23.1 11,113 1990 2,955 1.27 30 18 24.1 15,456 2000 3,025 1.45 38 28 28.3 16,760 Average by size class in 2000 Very large 1980 5,407 1.20 22 10 27.3 11,937 (> 3 million) 1990 5,899 1.47 23 28 27.5 14,232 2000 6,427 1.53 38 35 30.6 15,327 Large 1980 1,214 1.08 13 4 21.8 10,952 (1-3 million) 1990 1,398 1.18 23 12 22.4 12,393 2000 1,664 1.30 32 22 25.5 13,629 Medium 1980 542 1.05 8 2 19.8 10,148 (0.5-1 million) 1990 615 1.09 14 6 20.3 11,644 2000 710 1.18 23 14 22.8 13,231 Small 1980 196 1.03 6 2 18.1 9,732 (0.1-0.5 million) 1990 224 1.06 9 4 18.8 10,374 2000 274 1.11 15 7 21.3 11,513 All sizes 1980 1,395 1.14 17 7 24.3 10,626 (75 metro areas) 1990 1,560 1.31 26 19 24.6 12,052 2000 1,772 1.39 33 27 27.5 13,350 Notes: a We use the concept "urbanised areas" as defined by the Census. b For rows labelled "1980" data in columns 2-5 are for 1982, the earliest year reported by TTI. Sources: Columns 2-5 from The 2002 Urban Mobility Study, Texas Transportation Institute (TTI), Texas A&M University System, June 2002, http://mobility.tamu.edu/ums/. Columns 6-7 from various US Censuses: Table 118,124 from Vol. 1, Ch. C 1980 Census of Population; 1990 STF3C and 2000 SF3C available from http://factfinder.census.gov/servlet/BasicFactsServlet Cities, regions and the decline of transport costs 211 3 Implications for cities and regions of the decline of transport costs for goods We documented in the previous section that transport costs in 1900 were high and have been falling throughout the past century. Other facts also support the relevance of the new economic geography as a tool for understanding nineteenth century American geography. Natural resources were a large share of the economy. In 1900, 40% of the labour force worked on farms (Historical Statistics of the United States, Dl-10). Another 600,000 worked in mines. In the non-agricultural sector, manufacturing dominated services. There were 5.4 million manufacturing workers and only 1.7 million workers in services in 1900. Manufacturing, unlike services, displayed then (as now) the large fixed costs and returns to scale that are a critical element in the Krugman model. Furthermore, there are two other important aspects of transportation technologies not highlighted in the new economic geography, but which are also important for the structure of cities. Water-borne transport remained important. As late as 1924, water borne domestic tonnage was about one-fifth that of rail tonnage. All relevant forms of transportation technologies in 1900 also displayed significant scale economies that created centralisation. Having dozens of railroad stations spread throughout a city did not pay, and major stations were centred in a particularly large area. Rail is optimally organised in a hub and spokes network that naturally led to large rail centres (such as Chicago). The transportation technologies without these fixed costs in 1900 were walking and, to a much less significant degree, riding. Individuals primarily had to reach their homes on foot. Together these facts help us better understand American economic geography at the end of the last century. People are situated across the hinterland where the proximity to natural resources and saving on transport costs by crowding manufacturing activities together, and are close to ports or rail depots. Moreover, the rail depots and ports create the natural city centres, which then produce the monocentric cities modelled by Alonso, Muth and Mills. Fixed infrastructure, which facilitates the shipping of manufactured goods and the commuting of workers, created a natural centre for the city. Workers then spread out around that centre in patterns suggested by the classics of urban economic theory. Much of this has changed over the twentieth century. We have already discussed the declining importance of transport costs and the switch from fixed cost intensive means of shipping (rail) to more flexible shipping (truck). Likewise for individual consumers there has been a decreased importance of public transportation, which also created natural agglomerations - and a rise in the car, which had no need for hub or spokes. Natural resources have become far less important; in the 2000 census 1.9% of workers laboured in agriculture, fishing, forestry, and mining. Manufacturing has declined as well and now represents only 14.1% of the economy. More than 50% of the economy now works in various service industries. These facts are well known, but their implications for economic geography are surely understudied. To these facts we would add several other points. While natural resources are an increasingly irrelevant part of production, they are an important part of consumption. Space differs significantly in its natural endowment of warm weather and rainfall. These factors are the exogenous forces that make the US 212 E.L. Glaeser, J.E. Kohlhase Median income 60,000- -1-84 11.08 Log of density Fig. 9. Income in 1989 and density in 1990. Note: The regression line has a slope of 2055.6 (standard error of 60), and an R2 of 27.4 percent something other than a featureless plane. Differences in government policy will also differentiate space, although they are certainly less exogenous. Holmes (1998) uses discontinuities at borders to show how state policy impacts local employment. In the housing market as well, differences across areas in the degree of regulation are extremely important. Since the 1970s, some states, such as California and Oregon, have sharply limited new housing production; Texas has not. In a service economy where transport costs are small and natural productive resources nearly irrelevant, weather and government stand as the features, which should increasingly determine the location of people. At the top of these layers of "innate" factors are production and transportation technologies. Agglomeration economies appear to be as important as always. People in dense areas are more productive and earn more (Ciccone and Hall 1996). In a competitive labour context income should reflect the marginal product of labour. Figure 9 shows the relationship between and density across counties and suggests that a 10% increase in county density corresponds to a 206 dollar increase in median household income. Some of this relationship is due to higher skilled people living in dense areas, but if we control for schooling we find: Log(Income) = 8.76 + 0.06 x Log(Density) + 0.016 (0.02) (0.002) (0.0005) xShare w. BA + 0.015 x Share w. HS (2) (0.0004) where R2 - 0.64, standard errors are in parentheses, and there are 3,109 counties in the regression. Income refers to median household income in 1989, density is people per square mile, share w. BA refers to the percentage of the population over age 25 with college degrees, share w. HS, refers to the percentage of the population Cities, regions and the decline of transport costs 213 over age 25 with high school degrees only. There are many reasons to be sceptical about evidence of this kind. Controlling for these obvious schooling measures does not guarantee that we have controlled adequately for all forms of omitted ability (Glaeser and Mare 2001, for an extensive discussion of these issues in the urban context).9 Nonetheless, it is impossible not to start at least with the view that there remain massive economies from agglomeration. Where do these agglomeration economies come from? If there really were no transport costs, it is certainly true that agglomeration economies could not exist. There would be little possible reason for there to exist any differences across space, because without transport costs, anyone could costlessly access any other person or firm on the planet. As such, transport costs are still important, but the relevant transport costs are likely to be for moving people, not goods. The advantages from proximity to other people appear to come from saving the costs of providing and acquiring services and from improving the flow of knowledge. There has been much useful work on this in the past, and hopefully further work will help us better understand why these agglomeration effects are so critical.10 Regardless of the cause, it seems certain that proximity will remain important and likely continue to be true well into the future. However, while proximity matters, the form of proximity certainly has changed. The dominant nineteenth century mode of individual transportation - walking -has allowed people to travel one to two miles in a 30-minute walk. The automobile allows people to move between 15 and 30 miles during the same time. While the reduction of transport costs for goods has freed people from living close to natural resources and has facilitated concentration within a region, the reduction of transport costs for people has meant that individuals can live at much lower densities and still enjoy the advantages of proximity. The second significant transport technology change is the reduced importance of public transportation, ports, rail hubs, and other massive infrastructure centres.11 These centres provided cities with natural cores, and their absence means that cities can sprawl without limit. As such, we have moved from monocentric cities to polycentric regions, which are polycentric or rather uncentred entirely.12 This change is a major reason why increasingly relevant economic units are regions, rather than urban cores. 8 The connection between average income and density is also strong but the estimated elasticity is smaller. 9 The interpretation of the schooling coefficient is difficult also because it will include any direct effect of schooling and human capital spillovers (as in Rauch 1993). 10 Many citations to the vast literature on agglomeration economies can be found in Fujita and Thisse (2002), several chapters in Henderson and Thisse (forthcoming), and in the review article by Anas et al. (1998). 11 Some ports, particularly ocean-oriented ports, may continue to flourish. See Chapters 8 and 13 of Fujita et al. (1999) or Fujita and Mori (1996). 12 Fujita and Ogawa (1982) develop the first theoretical model with simultaneous location of firms and households in a linear city. Lucas and Rossi-Hansberg (2002) generalise the Fujita-Ogawa paper to a symmetric circular city. Both papers develop various equilibriums including monocentric, polycentric or completely mixed cities. 214 E.L. Glaeser, J.E. Kohlhase Together these changes suggest a natural structure for urban models of the twenty-first century. The only innate advantages that matter are consumer amenities, and perhaps government policies (which presumably should be treated as endogenous). Individual productivity is a function of the number of people within reasonable driving distances. There are no natural centres or cores of urban areas. While it remains for theorists to develop these models more thoroughly, the implications of any model fitting the above description seems clear: we should expect to see much less population in the natural resource related sectors of the country and much more in pleasant inhabitable areas. We should see continuing agglomerations, but these agglomerations should be built around the automobile and should have few natural centres. There are no natural limits to the sprawl of these agglomerations and there is no reason why the population cannot be centred in a few big 13 areas. Indeed, the only thing to limit growth of the areas in the long run is continuing demand by some consumers for various forms of natural amenities or government policies. Without these factors we could possibly all end up in southern California. However, because some people like winters and because California regulations limit both building and certain forms of economic activity, there are likely to be a number of large, sprawling urban areas that dominate the American landscape in the future. Moreover, because housing and other infrastructure are durable, these changes are happening only slowly, and we still see the remnants of cities built around different transport technologies. 4 Testing the implications of declining transport costs for goods We now turn to our evidence on the current state of America's regions and on the twentieth century transformation of economic geography. We review some natural implications of the decline in transport costs and assess the evidence that relates to them. Implication 1: People are no longer tied to natural resources This is perhaps the most natural implication of a decline in the importance of transport costs. If in 1900, it was advantageous to be nearby natural resources, in 2000 this is no longer relevant. Areas once populated because of their natural resources should have lost large numbers of inhabitants over the twentieth century. We have no perfect measure of these resources; instead we use two proxies. First, we use the share of employment in a county that works in agriculture, fishing, forestry, or mining. Admittedly, we would prefer to have this measure at the beginning of our time period, instead of the end, but we think that the inter-temporal correlation between these measures is high enough to represent a reasonable measure of the importance of innate natural resources in the area. Figure 10 shows 13 We will define sprawl simply as decentralised population and employment. Empirically, high levels of sprawl would be captured by a large share of a metropolitan area's population and employment that is more than five miles from the central business district. Cities, regions and the decline of transport costs 215 Population growth 1920-2000 6 -I . 0 - 1-1-1-1- Agriculture, forestry 0 0.2 0.4 0.6 and mining Fig. 10. Population decline and natural resources. Source: US Census, 1920, 1990 and 2000 the relationship between this employment share and the logarithm of population in 2000 in the county divided by population of the county in 1920. The estimated regression is: / Population in 2000 \ Natural Resource Employment Log —-- = 0.95 - 4.52 x -—- \ Population in 1920/ (0.02) (0.15) Total Employment (3) where R2 - 0.22, standard errors are in parentheses, and the number of observations is 3,056. The coefficient implies that as the share of employment in natural resources rises by 10%, the growth of the county between 1920 and 2000 should be expected to fall by 45.2%. This coefficient is strongly robust to other controls. A second method of showing this change is to examine the relationship between population growth and longitude. In 1990, and we believe in 1900, the centre of the US specialised in the production of natural resource based commodities. Indeed, the peopling of America was based largely on the demand for agricultural land and the desire to exploit America's rich natural wealth. However, as transport costs fell, we should expect to see America hollow out. People should ostensibly leave the middle states, which have always had harsh environments, and move to the coasts, which are more temperate and provide easier access to Europe and Asia. To test this implication Fig. 11 indicates the relationship between population growth and longitude. We have estimated a spline with a break at -100 degrees longitude. This number was chosen fairly arbitrarily - it is the longitude of central Nebraska. The graph shows that the population increased on both coasts and 216 E.L. Glaeser, J.E. Kohlhase Log change in population 1920-2000 2 - -2 - ■8,= 0%°^-,,; o ° o »