going o the rails: the e ect of railroad abandonments on...
TRANSCRIPT
Going Off the Rails: The Effect of Railroad Abandonments on
Population and Industrial Decline ∗
Dustin Frye†
Vassar College
February 19, 2018
Abstract
The paper exploits the abandonment of American railroads over the twentieth century to studythe impact of reductions in transportation infrastructure on population and industrial growth. Usingnewly digitized spatial data on the location and timing of railroad abandonments, I find that com-munities with more severe reductions in railroads experienced declines in population, urbanization,and manufacturing. There is also suggestive evidence of declines in wholesale trade, retail trade,and agricultural property values. These declines persist despite investments in highway infrastruc-ture. These findings reaffirm the importance of transportation networks for economic development.Despite their long-run presence, the removal of railroads was able to upset the spatial distributionof economic activity, suggesting that continued investment and improvements in infrastructure areimportant for maintaining the growth benefits of transit infrastructure.
Keywords: Railroad Abandonment,
JEL Classification:
∗I would like to thank Jacob Sowder for excellent research assistance. All errors are my own.†Department of Economics, Vassar College, 124 Raymond Ave, Box 76, Poughkeepsie, NY 12604 (e-mail:
[email protected]; website: http://pages.vassar.edu/dustinfrye/ )
1 Introduction
The expansion of railroads transformed the landscape of trade and travel in the nineteenth and
early twentieth centuries, resulting in over 250,000 miles of track by 1916. Following this high-water
mark in railroad mileage, competition, changing economic conditions, and deregulation contributed
to the steady abandonment of railroad mileage. Despite being the leading method of shipping high
volume goods within the US, over 40 percent of railroad miles were abandoned during the 20th
century. These abandonments introduced a potentially substantial shock for rail reliant locations
and industries. This article exploits this substantial reduction in infrastructure and examines the
impact of railroad abandonment on population and the economic conditions in five industries:
agriculture, manufacturing, retail trade, services, and wholesale trade.
Railroads were a key contributor to the spatial arrangement of both population and industry
throughout the nineteenth and early twentieth centuries (Atack, et al., 2010; Atack, Haines, &
Margo, 2008; Atack & Margo, 2012). The abandonment of rail lines provide a setting to understand
the fragility of these spatial equilibria to negative shocks in infrastructure and market access.
Additionally, abandonments were more severe in less populated, rural counties, potentially removing
a valuable channel of access to larger outside markets. This reduction in market access directly
affects traded sectors and a decline in those sectors could have negative spillover effects to the rest
of the local economy. Taken together, these effects would likely exacerbate regional differences in
economic growth between urban and rural areas.
Using county-level data from the Censuses of Agriculture and Population from 1910 to 1990 and
the County Business Patterns from 1947 to 1994, the empirical specifications use newly digitized
locations and dates of railroad abandonments to estimate the effect of increased abandonments
on county-level outcomes. To address the endogeneity regarding selection into abandonment, this
paper exploits techniques in graph theory to measure railway network importance. This technique
establishes the most unnecessary segments for travel and therefore the most likely segments of
railroad for abandonment, which I use to instrument for actual abandoned segments.
Preliminary estimates indicate railroad abandonment led to a consistent pattern of decline
in population and industry after accounting for differences in population trends, geography, and
access to alternative forms of transit infrastructure. A county experiencing the average level of
1
abandonment saw population fall by 4.71 percent. This same county also experienced industrial
decline. The declines were most severe and consistent in manufacturing, which experienced a 4.67
percent reduction in the number of manufacturing firms and an 8.37 percent decline in the value
added of manufactured goods. Abandonments led to slightly weaker declines in retail trade, services,
wholesale trade, and agricultural property values. These declines are more severe in places with
fewer miles of substitutable highway infrastructure, suggesting that a reduction in market access is
partially responsible for the declines.
Despite the large literature devoted to understanding the benefits of expanding transportation
infrastructure, very little is known about the potential effects of removing infrastructure. In Ghana,
the neglect of colonial railroads did not alter the spatial equilibrium that developed following
construction (Jedwab & Moradi, 2016). The authors propose that the lack of developed cities
created a stronger environment for railroads to create persistent local returns to scale, thereby
creating a suitable environment for a persistent spatial equilibrium to develop around railroads. The
relative level of development, fluidity of capital and labor markets, and availability of alternative
locations in the United States likely explains the difference in persistence across settings. During a
similar period to the US, Britain instituted a series of cutbacks to rail financing, an event commonly
referred to as the ‘Beeching Axe’. These cuts reduced mileage by over 41 percent and there is still
uncertainty to how these reductions impacted rural Britain (Loft, 2013).
More broadly this paper contributes to our understanding of the conditions under which per-
sistent historic spatial equilibriums can change. There is a large literature documenting the im-
portance of increasing returns to scale in promoting persistent effects brought about by historical
shocks (Ahlfeldt et al., 2015; Bleakley & Lin, 2012; Bosker et al., 2007, 2008; Kline & Moretti, 2014;
Michaels & Rauch, 2016; Redding, Sturm, & Wolff, 2011). My preliminary findings suggest that
transit infrastructure and market access are important components for maintaining agglomeration
due to increasing returns to scale and after removing infrastructure or decreasing access to mar-
kets, the channels promoting agglomeration can be undermined. My setting is similar to the case
of German reunification, where Redding & Sturm (2008) find evidence that reunifying Germany
altered market access and reversed the persistent differences in population growth across regions in
former West Germany. My paper highlights a similar reversal brought about by the abandonment
of infrastructure.
2
This paper also relates to the broader literature on the employment effects of transportation
infrastructure.1 The American railroad literature has primarily focused on the era of railroad
expansion or the more recent period post-deregulation. Early American railways promoted changes
in agriculture, property values, manufacturing, banking, trade, and structural change (Atack &
Margo, 2012; Atack et al., 2010; Atack, Jaremski & Rousseau, 2014; Donaldson & Hornbeck, 2016;
Fogel 1964; Fajgelbaum & Redding, 2014). More recent work has focused on price discrimination
and pricing behavior following deregulation (Busse & Keohane, 2007; Ellig, 2002; Hughes, 2011).
This paper bridges the gap between those two eras. It also overlaps with the period of interstate
highway expansion. Interstate expansion led to increases and changes in the locations of employ-
ment, both overall and by sector (Chandra & Thompson, 2000; Duranton, Morrow, & Turner 2014;
Duranton & Turner 2012; Frye, 2017; Michaels, 2008). I find that the abandonment of railway net-
works has a similar, although inverse, effect. Additionally, I find that the negative affects of railroad
abandonment are more severe in the absence of highways, suggesting that railroad abandonment
and highway expansion offset through a similar mechanism.
Railways were of similar importance to historic development outside of the American setting.
Jedwab & Moradi (2016) document persistent benefits for colonial railways in Africa. Indian railway
expansion produced substantial benefits to trade, income, and welfare (Donaldson, 2010). In Ger-
many, railroad development promoted manufacturing, industrialization, and economic integration
(Gutberlet, 2014; Walker, 2015). My paper speaks to how these industrial developments continue
to evolve as the railway ages and alternative forms of transportation develop.
The paper proceeds as follows. Section 2 provides a brief overview of railroad history and the
process of railway abandonments. Section 3 discusses the data. Section 4 introduces the empirical
strategy and addresses the endogeneity concerns. Section 5 presents the major findings, and Section
7 concludes.
1For a recent survey of this literature see Redding & Turner (2015).
3
2 Historical Background
2.1 Railroad Expansion and Decline
Railroad expansion dates back to the early 1800s, but the first major wave followed the federal gov-
ernment’s opening of land grant subsidies in 1850. This era of construction was largely decentralized
and the network was mostly constructed by entrepreneurial investors (White, 2011). Following the
Civil War, the federal government became increasingly involved in developing a transcontinental
railroad. Through the Pacific Railway Act of 1862 it reinstated land grant subsidies and instituted
a system of loans that varied based on construction costs. In the land grant process, railroad
companies were given blocks of public land along the proposed routes. This policy incentivized
speculation and building ahead of demand as railway companies competed to secure land rights.
This led to massive expansion in the railroad network. Between 1860 and 1880, over 100,000 miles
of railroad track were constructed (Fishlow, 1965; Fishlow, 2000). Growth slowed in the 1890s and
shifted to the Midwest where agricultural traffic boosted profits. By the turn of the century total
railroad mileage exceeded 190,000 miles and railroads continued as the primary mode of trans-
portation. Railroad mileage peaked in 1916 with over 250,000 miles of track. Despite the sustained
expansion in mileage, the number of railroad companies actually fell over time as regulation and
financial panics contributed to the consolidation of the industry.
In 1887, Congress passed the Interstate Commerce Act and created the Interstate Commerce
Commission (ICC), which established early railroad regulations targeting rates and discrimination
(Stone & Landry, 2012). The Transportation Act of 1920 expanded the regulatory arm of the ICC
to cover entry, exit, rates, and service. This act formalized the process of railroad abandonment.
Prior to 1930, the abandonment of rail lines was overwhelmingly driven by the exhaustion of natural
resources (Weissman, 1958). By the early 1930s, the primary cause shifted to non-rail competition
and insufficient traffic (Weissman, 1958).
2.2 Railroad Legislation and Abandonments
The Transportation Act of 1920 required railroad companies to apply for abandonment. The
ICC weighed the potential losses by railroad companies against the losses to the shippers and
communities from the loss of the line (GAO, 1976). A key consideration for the Commission
4
was the availability of alternative forms of transit. Between 1920 and 1963, over 90 percent of
abandonment applications were approved, partly because railroad companies only submitted cases
with a high probability of success (Due, 1978). From 1920 to 1972, the ICC approved an average
of 1,023 miles per year of abandonments, totaling over 53,000 miles. Despite these abandonments,
railroad companies, especially those in the Northeast, were still facing financial hardship due to
increased competition from highways and air travel and ICC regulation.
Congress introduced several measures during the 1970s to alleviate the financial burdens of rail
companies, including relaxing abandonment regulations. Passenger travel had steadily declined
and in 1970 Congress created Amtrak to alleviate the burden of passenger travel. In the mid-1970s
Congress passed the 3R Act of 1973 and the 4R Act of 1976 aimed to accelerate the abandonment
process.2 Deregulation of railroads started with the Staggers Rail Act of 1980 and was expanded
with the Northeast Rail Service Act of 1981. These acts reduced the role of the ICC and allow
carriers to abandon lines more freely. The early 1980s saw a sharp increase in the number of
abandonment applications and total mileage of abandonment. By 1987, over 88,000 miles of track
had been abandoned, roughly 35 percent of the 1916 peak mileage.
3 Data Construction and Description
3.1 Data Construction
Historic county-level data are drawn from the Census of Population and the County Business
Patterns (Haines, 2010). The main variables of interest are population, population in urban areas,
and economic indicators across five industries: agriculture, manufacturing, retail trade, services,
and wholesale trade. The industrial variables include the number of establishments, employment,
payrolls, and measures of the value of output, including sales and value added in the case of
manufacturing. Table 1 provides summary statistics for the full sample of each outcome. The
number of observations vary across outcomes as the availability depends on the data collection and
suppression to maintain confidentiality. Table 2 lists the source document, year of interest, and
2The formal names for the two acts are the Regional Rail Reorganization Act of 1973 and the Railroad Revital-ization and Regulatory Reform Act fo 1976. For a survey of their legal features see Due (1978).
5
the number of observations available.3 The empirical analysis focuses on an unbalanced panel of
counties, from 1910 to 1990, for which data is available across multiple years. The final sample is
restricted to exclude counties that did not have railroads in 1916, counties in Alaska and Hawaii,
and counties that had substantive county border changes between 1920 and 1990.4
Figure 1 maps the national system of railways at near peak mileage in 1916 (Atack, 2016).
Railroad abandonment information comes from a series of maps and tables created by Waldo Nielsen
(1992). The maps and tables include information about each railroad abandonment prior to 1992
including the start and end points for each abandoned segment and a date range of abandonment. I
digitized these maps and used contemporary and historic maps to match the start and end points of
each segment. I then isolated the segments from the original Atack 1916 map that were eventually
abandoned. I cross-referenced Nielsen’s date and location information with published railroad
maps that span the period. Figure 2 maps the full railroad network split by eventual abandonment
status in 1987. Figure 3 maps only the set of abandoned railroad segments. In general, the spatial
distribution of abandonments is consistent with overbuilding in the midwest and northeast, but
every state experienced some degree of abandonment.
My primary measure of abandonment intensity is the share of 1916 railroad lines that were
abandoned in a county. I use the date information from Nielsen and additional railroad maps to
measure the abandonment intensity at different points in time. Nielsen’s date information is limited
to six date ranges between 1939 and 1987.5 I create both a discrete and smoothed abandonment
measure. The discrete measure only attributes track as abandoned in the final year of the date
range. The smoothed measure assumes abandonment occurred linearly over the date range.6 My
preferred specification uses the discrete measure as it is likely to underestimate the true effect of
railroad abandonment by understating the degree of abandonment that actually occurred.7
Table 1 shows the average level of abandonment across the entire period and the average level of
abandonment in 1987. By the end of the sample period, the average county in the sample lost about
3In future drafts I will add some analysis to understand the characteristics of these omitted counties, and how, ifat all, they might impact the main findings.
4The empirical models control for 1920 population, so counties without population measures in 1920 are alsoexcluded from the analysis.
5The dates are before 1939, 1940 - 1944, 1945 - 1959, 1960 - 1971, 1972 - 1984, and 1984 - 1987. I am currentlyin the process of collecting more refined segment abandonment information from ICC annual reports.
6For the period prior to 1939, I assume abandonments started in 1920.7When I compare the smoothed model to the discrete model the OLS estimates are consistent with this assumption.
6
33 percent of it’s railroad mileage, with some counties not losing any and others abandoning the
railroad completely. Figure 4 maps these final abandonment shares. Abandonments are spatially
distributed with some clustering in western New York, the upper midwest, and western Colorado.
The baseline Atack 1916 railroad file contains information about the operation dates of each rail
line. Figure 5 plots the distribution of the first known operation dates for the set of abandoned
lines and the full sample of rail lines. The distribution of abandoned lines is consistent with the
distribution of overall rail lines.
Table 3 reports summary statistics for each outcome in the first period I observe the outcome,
split by the mean fraction of abandonment. Counties with higher eventual shares of abandonment
were more economic developed in the early periods. High abandonment counties were smaller in
population, less urbanized, and were less developed across all industry measures.
3.2 Trends Relative to Abandonment
Figures 6 through 11 plot each outcome of interest relative to the date of each counties first
abandonment. Each plot shows the mean residual of the outcome variable after de-trending by
state and year. Across the outcomes, the figures show a striking relationship relative to the date
of the first abandonment. Importantly, prior to abandonment the outcomes are either steady or
increasing. From Figure 6, both population and the percent of the population living in urban areas
show a sharp decline following abandonment. A similar trend develops across the manufacturing
outcomes in Figure 7, the services outcomes in Figure 8, the wholesale trade outcomes in Figure
9, and the retail trade outcomes in Figure 10. The agricultural outcomes in Figure 11 appear
to be the exception, with the downward trend in the amount of acreage under cultivation and
average property value per acre starting several years in advance of railroad abandonment. These
provide some suggestive evidence that railroad abandonments negatively impacted communities in
the decades after abandonment.
7
4 Empirical Framework
4.1 Baseline Specification
To estimate the effect of railroad abandonment on population and industry outcomes, I exploit
variation in the severity of railroad abandonment across locations and over time. Outcome Y in
county c and year t is regressed on the fraction of railroad that was abandoned at time t, a vector
of geographic and demographic controls, X, state-by-year effects, and county fixed-effects:
Yct = βRRAbandShrct + δst + γc +X ′ρct + εct (1)
The specifications control for several time invariant geographic characteristics and baseline pre-
abandonment measures. These time invariant controls are interacted with year-dummy variables to
flexibly control for these covariates over time. The identifying assumption is that, if not for railroad
abandonment, counties with abandoned rail would have changed similarly to non-abandoned coun-
ties in the same state and year and with similar geographic and pre-abandonment characteristics.
The bottom panel of Table 1 lists the key covariates of interest. Controlling for initial railroad
mileage flexibly accounts for the degree of initial overbuilding and speculation that occurred prior
to abandonment. Controlling for distance from county centroid to the closest MSA helps account
for potential differences in the probability of abandonment at different distances from major cities.
County areas differ substantially by region, with the southwest having larger counties on average.
Controlling for county area helps account for differential trends in population and industrial move-
ment to larger counties, which often had less abandonment. County centroid latitude and longitude
controls allow for spatial patterns in economic changes that may be correlated with abandonment.
The model controls for three types of highway infrastructure, two time-invariant measures of major
highway mileage and time varying county level measures of interstate highway mileage. These con-
trols are potentially endogenous as highways are often used as a form of stimulus for lagging areas
(Frye, 2017; Jaworski & Kitchens 2017). Particularly in the case of interstate highways, regions
that were more negatively affected by abandoned lines might be more likely to receive or be near
an interstate in the future. The substitution between railroads and highways was a fundamental
part of the abandonment of railroads that it is a useful control despite the potential endogeneity
8
concerns. In Section 6, I show the results without the interstate highway control. Finally, control-
ling for population in 1920 allows for the impact of pre-treatment population levels to vary over
time.8
4.2 Endogeneity Concerns and Network Centrality
Measuring differences between counties with abandoned and non-abandoned segments of rail will
likely result in biased estimates because segments of track selected for abandonment are likely
correlated with unobserved time varying characteristics that are correlated with both demographic
changes and industry growth. Railroad companies were selecting their segments based on both
current and expected future profitability. The motives for their selection and the willingness of
the ICC to grant an abandonment permit create an endogeneity problem that the controls cannot
account for.
To address this issue I use an application from graph theory to predict a set of underutilized seg-
ments of the railroad that would be ideal candidates for abandonment. Once I have this pseudo-set
of abandonment locations, I can calculate a predicted share of abandonment to use as an instru-
mental variable. To identify underutilized segments, I calculate the edge betweenness centrality for
the 1916 railroad network similar to Frye (2017). The process converts the 1916 railway network
into a mathematical network of nodes and edges and weights each edge by distance. Nodes occur at
the intersection of each two edges or at the endpoint of an edge. I use Brandes’s (2008) algorithm
to calculate the betweenness centrality of each edge. The betweenness centrality value indicates
the number of times that particular edge was used when traveling using the shortest path between
every two pairs of nodes in the network. Higher betweenness centrality values are travel more
frequently. I identify the lowest 25 percent of betweenness centrality values as the set of segments
that will be abandoned.9 Figure 10 plots the set of predicted abandonments.
Once I have a set of candidates for abandonment I assume that segments were abandoned
according to their betweenness centrality value, so the lowest centrality value would be abandoned
first. I rank the centrality values, and “abandon” an equal percentage of mileage each year between
8I have altered the year and number of population controls and the results are extremely stable.9The 25 percent cutoff is arbitrary, however it matches closely in mileage to the actual eventual mileage that was
abandoned. In future drafts I plan on testing the sensitivity to this threshold.
9
1939 and 1990.10 This gives me a predicted amount of mileage each year that varies by county. I
then use this to calculate a predicted share of abandonment, which I use to instrument for actual
abandonment shares.
Figure 12 plots the flow of abandonments and predicted flow of abandonments over time. The
predicted flow of abandonments is consistently higher because it is a mechanically smoothed process
from 1939 to 1990, whereas actual abandonments move variably with the legislative eras. The figure
illustrates the sharp increase in abandonments after deregulation in 1980. Table 3 illustrates the
fraction of the instances where the predicted shares are below the actual shares, are equal to the
actual shares, and are less than the actual shares in each year. Consistent with Figure 12, the
predicted shares are consistently higher in early periods, but the IV generates predicted values
that are both above and below the actual values in each period. Figure 13 shows the actual and
predicted distributions of both the distance abandoned and the share abandoned. The distributions
of both measures are very similar.
4.3 Defending Network Centrality
The validity of the instrument hinges on the degree that edge betweenness centrality of the railroad
network is orthogonal to changes in population and industry over the twentieth century. One source
of variation in railroad centrality is driven by the connectedness of the graph. By controlling for the
number of railroad intersections in the county, I account for the number of available shortest paths
originating in the county, junctions where shortest paths could change direction in the county, and
potential growth that could occur around the intersections.
Low betweenness values are primarily driven by redundancies and under-traveled spurs in the
network. Redundancies are more likely in places with excessive overbuilding. One concern is that
these places experienced an unusual boom in construction due to some persistent process and
this process is correlated with changes after railroad abandonment. Given the history of railroad
expansion and subsidies, railroad expansion was largely motivated by land speculation and railroad
entrepreneurs were focused on short-term gains. The speculative motives of railroad entrepreneurs
in the late-nineteenth century is unlikely to directly affect population changes, except through the
channel of potentially redundant lines, which is a key source of the variation in the instrument.
10I have also adjusted the date range to begin in 1920 and the results are similar.
10
Additionally, the nature and structure of the US economy changed dramatically from the late-
nineteenth century to the mid-twentieth century. Late arriving rail line developers were interested
in income from shipping agricultural products. One concern is that rail line network betweenness
is simply measuring areas developed for the quality of their agricultural land. To address this
concern, I control for agricultural suitability.
5 Main Results
5.1 Population and Urbanization
Table 4 reports estimated declines in population and the percent of population living in urban
areas when estimating equation 1 and including the full set of controls. The table also reports first-
stage results and the mean and standard deviation of the treatment variable. The population and
urbanization results indicate that the abandonment of railroads led to a sharp decline in population
and urbanization. Results from the instrumental variable specification indicate that a one standard
deviation increase in railroad abandonment led to an 8.16 percent decline in population. These
population declines appear to target urban areas, with populations over 2500 people. Following a
similar one standard deviation increase in abandonment, the percent of the population living in
urban areas fell by 8.94 percentage points. Railroad abandonment is correlated with lower initial
levels of population, indicating that these population declines are most likely affecting small and
medium sized urban centers.
5.2 Industrial Growth
Table 5 reports changes to the manufacturing and wholesale traded sectors following abandonment.
Manufacturing experienced the sharpest declines in the number of establishments and manufactur-
ing value added. A one standard deviation increase in abandonment is associated with a 14.5
percent reduction in the value added of manufactured goods. Counties that lost all of their rail-
roads would be expected to lose over 65 percent of the value added of their manufactured goods
compared to a county that retained all of it’s rail lines. Changes in wholesale trade tell a similar
story in the magnitude of declines for both the number of establishments and the value of total
sales, although the later is not statistically distinguishable from zero. Taken together these results
11
suggest traded sectors faced sharp declines following the abandonment of railroads.
Estimated changes in the agricultural sector, given in Table 6, suggest that the number of
farms and acreage under cultivation were unaffected by railroad abandonment. However, the value
of farm property per acre fell by over 9 percent following a one standard deviation increase in
abandonments. The estimated effects of abandonment on the service sector suggest similar declines
to those in wholesale trade. Table 7 reports the estimated impact of abandonment for retail trade.
The results suggest potentially small declines in the number of firms, employment, and total payroll.
However, there are much larger declines in the value of total sales, where a one standard deviation
increase in abandonments led to a 13.6 percent decline.
6 Robustness
6.1 Excluding Interstate Highway Controls
Table 8 reproduces the OLS and IV results with the full set of covariates from Table 4. Columns 3
and 4 exclude interstate highway mileage from the set of covariates. As expected, the OLS estimates
become larger in magnitude, consistent with interstate mileage being positively associated with
growth. However, the IV estimates also become larger. Indicating that interstate highway mileage
is correlated with railroad centrality. Given that railroad locations have been used to instrument
for interstate highway connections, it is likely that the centrality of highways is correlated with the
centrality of railroads (Duranton, Morrow, & Turner, 2014; Duranton & Turner, 2012). Given this
correlation, controlling for interstate highway mileage seems important for the IV estimation.
7 Conclusion
The process of structural transformation, characterized by the transition of workers into increas-
ingly productive and skilled occupations, requires modern technology that can support the changes.
As the US transitioned from a largely agrarian and small manufacturing economy to a more ad-
vanced manufacturing and service based economy it required transportation infrastructure that
could support changing industrial development. This modernization led to a transition in the over-
land transportation networks, shifting away from railroads towards highways and air travel. These
12
changes resulted in the abandonment of over 88,000 miles of rail lines during the mid-twentieth
century. These abandonments changed the economic conditions in formerly connected areas, de-
creasing population, urbanization, and industrial growth across several industries.
These findings reaffirm the importance of transportation networks for economic development.
Despite their long-run presence, the removal of railroads was able to upset the spatial distribution
of economic activity, suggesting that continued investment and improvements in infrastructure are
important for maintaining the growth benefits of transit infrastructure.
13
References
Ahlfeldt, G. M., Redding, S. J., Sturm, D. M., and Wolf, N. (2015). The economics of density:Evidence from the berlin wall. Econometrica, 83(6):2127–2189.
Atack, J. (2016). Historical geographic information systems (gis) database of u.s. railroads for 1916.Technical report.
Atack, J., Bateman, F., Haines, M., and Margo, R. A. (2010). Did Railroads Induce or FollowEconomic Growth?: Urbanization and Population Growth in the American Midwest, 1850-1860.Social Science History, 34(2):171–197.
Atack, J., Haines, M. R., and Margo, R. A. (2008). Railroads and the rise of the factory: Evidencefor the united states, 1850-70. Technical report, National Bureau of Economic Research.
Atack, J., Jaremski, M., and Rousseau, P. L. (2014). American banking and the transportationrevolution before the civil war. The Journal of Economic History, 74(4):943–986.
Atack, J. and Margo, R. A. (2012). The Impact of Access to Rail Transportation on AgriculturalImprovement: The American Midwest as a Test Case, 1850-1860. Journal of Transport and LandUse, 4(2).
Bleakley, H. and Lin, J. (2012). Portage and path dependence. The quarterly journal of economics,127(2):587–644.
Bosker, M., Brakman, S., Garretsen, H., and Schramm, M. (2007). Looking for multiple equilibriawhen geography matters: German city growth and the wwii shock. Journal of Urban Economics,61(1):152–169.
Bosker, M., Brakman, S., Garretsen, H., and Schramm, M. (2008). A century of shocks: theevolution of the german city size distribution 1925–1999. Regional Science and Urban Economics,38(4):330–347.
Brandes, U. (2008). On variants of shortest-path betweenness centrality and their generic compu-tation. Social Networks, 30(2):136–145.
Busse, M. R. and Keohane, N. O. (2007). Market effects of environmental regulation: coal, railroads,and the 1990 clean air act. The RAND Journal of Economics, 38(4):1159–1179.
Chandra, A. and Thompson, E. (2000). Does public infrastructure affect economic activity?: Evi-dence from the rural interstate highway system. Regional Science and Urban Economics.
Commission, I. C. (1976). Better information needed in railroad abandonment. Technical report,Working Paper.
Donaldson, D. (2010). Railroads of the Raj: Estimating the Impact of Transportation Infrastruc-ture. pages 1–51.
Donaldson, D. and Hornbeck, R. (2016). Railroads and american economic growth: A marketaccess approach. The Quarterly Journal of Economics, 131(2):799–858.
Due, J. F. (1978). Railway abandonment: What next? Technical report, Working Paper.
14
Duranton, G., Morrow, P. M., and Turner, M. A. (2014). Roads and Trade: Evidence from the US.The Review of Economic Studies, 81(2):681–724.
Duranton, G. and Turner, M. A. (2012). Urban Growth and Transportation. The Review ofEconomic Studies, 79(4):1407–1440.
Ellig, J. (2002). Railroad deregulation and consumer welfare. Journal of Regulatory Economics,21(2):143–167.
Fajgelbaum, P. and Redding, S. J. (2014). External integration, structural transformation andeconomic development: Evidence from argentina 1870-1914. Technical report, National Bureauof Economic Research.
Fishlow, A. (1965). American Railroads and the Transformation of the Ante-bellum Economy,volume 127. Harvard University Press Cambridge, MA.
Fishlow, A. (2000). Internal transportation in the nineteenth and early twentieth centuries. TheCambridge economic history of the United States, 2:543–642.
Fogel, R. W. (1964). Railroads and American economic growth. Johns Hopkins Press Baltimore.
Frye, D. (2017). Transportation networks and the geographic concentration of industry. Technicalreport, Working Paper.
Gutberlet, T. (2014). Mechanization and the spatial distribution of industries in the GermanEmpire, 1875 to 1907 - Gutberlet - 2013 - The Economic History Review - Wiley Online Library.The Economic History Review.
Haines, M., university Consortium for Political, I., and Research, S. (2010). Historical, Demo-graphic, Economic, and Social Data: The United States, 1790-2002.
Hughes, J. E. (2011). The higher price of cleaner fuels: Market power in the rail transport of fuelethanol. Journal of Environmental Economics and Management, 62(2):123–139.
Jaworski, T. and Kitchens, C. T. (2017). National policy for regional development: Evidence fromappalachian highways. Technical report, National Bureau of Economic Research.
Jedwab, R. and Moradi, A. (2016). The permanent effects of transportation revolutions in poorcountries: evidence from africa. Review of economics and statistics, 98(2):268–284.
Kline, P. and Moretti, E. (2014). Local Economic Development, Agglomeration Economies, andthe Big Push: 100 Years of Evidence from the Tennessee Valley Authority. Quarterly Journal ofEconomics, 129(1):275–331.
Loft, C. (2013). Last Trains: Dr Beeching and the death of rural England. Biteback Publishing.
Michaels, G. (2008). The effect of trade on the demand for skill: evidence from the InterstateHighway System. The Review of Economics and Statistics, 90(4):683–701.
Michaels, G. and Rauch, F. (2016). Resetting the urban network: 117–2012. The Economic Journal.
Nielsen, W. (1992). Right-of-way: A guide to abandoned railroads in the United States. Maverick.
Redding, S. J. and Sturm, D. M. (2008). The Costs of Remoteness: Evidence from German Divisionand Reunification. The American Economic Review, 98(5):1766–1797.
15
Redding, S. J., Sturm, D. M., and Wolf, N. (2011). History and industry location: Evidence fromgerman airports. The Review of Economics and Statistics, 93(3):814–831.
Redding, S. J. and Turner, M. A. (2015). Transportation costs and the spatial organization ofeconomic activity. In Handbook of regional and urban economics, volume 5, pages 1339–1398.Elsevier.
Stone, R. D. and Landry, M. (2012). The railroad capacity crisis: After cutting to the bone andmore, trains are back. Essays in Economic & Business History, 26(1).
Walker, S. (2015). On the right track? estimating the effects of transportation infrastructure oneconomic integration and welfare. Unpublished Working Paper.
Weissman, J. (1958). Railroad abandonments: The impact of competition. Iowa L. Rev., 44:492.
White, R. (2011). Railroaded: the transcontinentals and the making of modern America. WWNorton & Company.
16
8 Figures
Figure 1: National Railroad Map
17
Figure 2: National Railroad Map by Abandonment Status
Figure 3: Abandoned Railroads
18
Figure 4: Share of Abandoned Railroad
Figure 5: Distribution of Dates of Operation for Rail Lines
0.0
5.1
.15
Dens
ity
1820 1840 1860 1880 1900 1920Date of Operation
Abandoned Lines
0.0
5.1
.15
Dens
ity
1820 1840 1860 1880 1900 1920Date of Operation
All Lines
19
Figure 6: Population
-150
00-1
0000
-500
00
5000
1000
0Po
pula
tion
-75 -50 -25 0 25 50Years Relative to First Abandonment
(a) Total Population
-2-1
01
2Pc
t. Po
p in
Urb
an A
reas
-75 -50 -25 0 25 50Years Relative to First Abandonment
(b) Pct. Urban Population
20
Figure 7: Manufacturing
-40
-20
020
40M
anuf
actu
ring
Esta
blis
hmen
ts
-75 -50 -25 0 25 50Years Relative to First Abandonment
(a) Establishments
-200
0-1
000
010
0020
00M
anuf
actu
ring
Empl
oym
ent
-75 -50 -25 0 25 50Years Relative to First Abandonment
(b) Employment
-150
-100
-50
050
100
Man
ufac
turin
g Va
lue
Adde
d ('0
00s)
-75 -50 -25 0 25 50Years Relative to First Abandonment
(c) Value Added
21
Figure 8: Services
-150
-100
-50
050
Serv
ice
Esta
blis
hmen
ts
-75 -50 -25 0 25 50Years Relative to First Abandonment
(a) Establishments
-150
0-1
000
-500
050
0Se
rvic
e Em
ploy
men
t
-75 -50 -25 0 25 50Years Relative to First Abandonment
(b) Employment
-60
-40
-20
020
Serv
ice
Payr
oll (
'000
s)
-75 -50 -25 0 25 50Years Relative to First Abandonment
(c) Total Payroll
22
Figure 9: Wholesale Trade
-40
-20
020
40W
hole
sale
Est
ablis
hmen
ts
-75 -50 -25 0 25 50Years Relative to First Abandonment
(a) Establishments
-100
0-5
000
500
Who
lesa
le E
mpl
oym
ent
-75 -50 -25 0 25 50Years Relative to First Abandonment
(b) Employment
-600
-400
-200
020
040
0W
hole
sale
Sal
es ('
000s
)
-75 -50 -25 0 25 50Years Relative to First Abandonment
(c) Total Sales
23
Figure 10: Retail Trade
-100
-50
050
100
Ret
ail S
tore
s
-75 -50 -25 0 25 50Years Relative to First Abandonment
(a) Establishments
-100
0-5
000
500
Ret
ail E
mpl
oym
ent
-75 -50 -25 0 25 50Years Relative to First Abandonment
(b) Employment
-150
-100
-50
050
Ret
ail S
ales
('00
0s)
-75 -50 -25 0 25 50Years Relative to First Abandonment
(c) Total Sales
-20
-10
010
Ret
ail P
ayro
ll ('0
00s)
-75 -50 -25 0 25 50Years Relative to First Abandonment
(d) Total Payroll
24
Figure 11: Agriculture
-200
-100
010
020
0N
umbe
r of F
arm
s
-75 -50 -25 0 25 50Years Relative to First Abandonment
(a) Total Farms
-15
-10
-50
510
Acre
age
in F
arm
ing
('000
s)
-75 -50 -25 0 25 50Years Relative to First Abandonment
(b) Total Acreage
-600
-400
-200
020
0To
tal F
arm
Val
ue P
er A
cre
-75 -50 -25 0 25 50Years Relative to First Abandonment
(c) Farm Value Per Acre
25
Figure 12: Predicted Location of Abandoned Railroads
Figure 13: Railroad Abandonment vs. Predicted Railroad Abandonment Over Time
010
2030
40
1900 1920 1940 1960 1980 2000year
Pct. RR Abandoned Predicted Pct. RR Abandoned
26
Figure 14: Distribution of Railroad Abandonment and Predicted Abandonment
0.0
1.0
2.0
3.0
4D
ensi
ty
0 100 200 300 400RR Abandoned (km)
05
1015
Den
sity
0 .2 .4 .6 .8 1Pct. RR Abandoned
0.0
05.0
1.0
15.0
2.0
25D
ensi
ty
0 100 200 300 400 500Predicted RR Abandoned (km)
02
46
810
Den
sity
0 .2 .4 .6 .8 1Predicted Pct. RR Abandoned
27
9 Tables
28
29
30
31
32
33
34
35
36