The Effect of Permanent Infrastructure Loss Evidence from the UK Beeching Report Rail Cuts

Stephen Gibbons, LSE Stephan Heblich, University of Bristol Scott Swisher, University of Cambridge

October 2015 Introduction

I Question: Effect of permanent infrastructure closure

I Time period: 1950-80

I Actor: British Rail (BR)

I Event: Beeching Report

I Published in 1963 I Proposed large cuts to UK rail network I 75% of proposed cuts implemented by 1980

I What was the effect of the Beeching cuts on the UK?

I Look at effect on outcomes in UK census data

I We find that cuts lowered population growth, employment rate I Effects of cuts persist to present day Motivation

I Permanent unanticipated loss of infrastructure

I Rare historical situation I General cuts expected, but not specific distribution

I Cuts proposed by an outsider technocrat

I Cuts based on local traffic levels

I 25% of cuts averted due to political interference

I Related literature

I Adding infrastructure: Donaldson, Hornbeck (2015) on railroads; Duranton, Morrow, Turner (2014) on roads I Temporary closure: Feyrer (2009) on Suez Crisis

I We are not aware of another paper on effect of permanent infrastructure loss on economic outcomes Research Design

I Planned rail cuts announced in 1963

I Difference-in-difference (DiD) design

I Pre-cuts: 1951 census I Post-cuts: 1981 census

I Unit of analysis: 1951 Local Government Districts (LGDs)

I Treatment: km of rail lines lost, continuous

I Controls

I LGD characteristics up to and including 1951 I Change in road network 1951-81

I Alternate control group: Averted cuts

I Definition: proposed in 1963, line remains open in 1981 I 25% of planned cuts Preview of Main Results

I Beeching cuts had long-lasting negative effect

I Baseline DiD rail cut response

I Length: employment rate, relative size of service sector fall; relative size of professional and unskilled workers, manufacturing sector increase

I “All Measures” DiD

I Using rail length

I Length: same as baseline, but with no effect on service sector I Market access: lower pop. growth, fewer professionals, more agriculture, less manufacturing I Traffic: more professionals Brief History of UK Rail

I Initial private build-out: 1830-WWI

I Interim “Big Four” period: 1923-47

I British Rail: 1948-95

I Nationalized public rail company I Declining service scope and quality

I Privatization: 1995-present

I Complete breakup of industry I Persistent growth in ridership

I Current issues

I Inability to reopen lines that were closed in the past I Price inflation of rail tickets I Accidents due to lack of coordination between operators I Re-nationalization debate Beeching Biography

I Richard Beeching, Baron, 1913-85

I Physicist, mechanical engineer

I First Chairman of British Railways Board in 1961

I Installed by Conservative government to cut losses of BR I No prior experience in rail industry I Wrote Beeching Report, launched Beeching cuts

I Left BR in 1965 when later plan rejected Beeching Report

I British Rail suffering financial losses in early 1960s

I Decreased ridership and revenue

I Impetus to cut unprofitable rail lines

I 1963: Beeching I

I Proposed closure of 9700 km of lines, 33% of the current system, 2400 stations

I 1965: Beeching II

I Outlined development of a new minimal trunk line system I Never implemented directly as a package

I Implementation

I 75% of Beeching I closures complete by 1980 I Proposed cuts averted in 25% of cases I Closures intentionally irreversible Illustrative Rail Cut: Varsity Line

I Oxford-Cambridge rail line built in 1846

I Failed attempt to close in 1959

I Not listed for closure in Beeching I

I Closed in 1967 by BR - faster to go through London

I Reopening plans

I Started in 2006, now 50% open I Plans for 75% open I Barrier to last 25%: Mullard Radio Astronomy Observatory UK Rail Passengers, 1830-2014

Source: Association of Train Operating Companies (ATOC), Office of Rail Regulation (ORR) British Rail Route Mileage, 1948-82

Source: Fig. 1, Dodgson (1984) Journal of Transport Economics and Policy Treatment: Overlay of Rail Network in 1980 vs. 1950 Alternate Control Group: Averted Rail Cuts in 1980 Alternate Control Group: Averted vs. Actual Cuts Groups (1)

I 1951 Local Government Districts (1843)

I Proposed (1367) I Not proposed (476)

I Proposed cuts

I Implemented (1285) I Averted (343) Groups (2)

Averted YN Y 261 1024 1285 Implemented N 82 476 558

343 1500 1843 Pre-Trends: Total Population, Treated vs. Untreated

1951 LGDs; blue = untreated or “Implemented = N” group (0), red = treated or “Implemented = Y” group (1); 95% CIs Pre-Trends: Tot. Pop., Treated vs. Untreated, Ex. Urban

1951 LGDs; blue = untreated or “Implemented = N” group (0), red = treated or “Implemented = Y” group (1); 95% CIs; exclude top 5 urban areas Pre-Trends: Total Population, Implemented vs. Averted

1951 LGDs; blue = averted cuts or “Averted = Y” group (0), red = implemented cuts or “Implemented = Y” group (1); 95% CIs Pre-Trends: Tot. Pop., Imp. vs. Averted, Ex. Urban

1951 LGDs; blue = averted cuts or “Averted = Y” group (0), red = implemented cuts or “Implemented = Y” group (1); 95% CIs; exclude top 5 urban areas Outcome Data

I UK Census years

I Pre-Beeching: 1911, 21, 31, 51 I Interim: 1961, 71 I Post-Beeching: 1981, 91, 2001, 11

I Geographic units

I Local Government Districts (LGDs): 1951 I Districts: 1951-81 I Enumeration Districts (EDs): 1981 I Output Areas (OAs): 1991-2011

I Reconciling boundary changes in geographic units

I Merge into 1951 LGDs I Weighted by proportional area overlap I Assumes uniform population distribution within LGD Outcomes: Population Growth, %, 1951-81

1951 LGDs Treatment: Local

I Local: depends only on within-LGD rail network

I Local Measure 1: Rail length

I Defined as km of rail passing through LGD I Does not account for variation in LGD size

I Local Measure 2: Rail density 2 I Defined as km of rail per km of LGD area I May be distorted by large dispersion in LGD size

I Local Measure 3: Rail coverage

I Defined as fraction of LGD within 5 km of rail line I Sensitive to buffer size Local Treatment: ∆ in Rail Length, 1963-84

Rail Lines (km); 1951 LGDs; darker red = more negative, white = unchanged Local Treatment: Rail Density, 1963/1984

Rail Lines (km) per Area (km2); 1951 LGDs; red = high, blue = low Local Treatment: ∆ in Rail Density, 1963-84

Rail Lines (km) per Area (km2); 1951 LGDs; darker red = more negative, white = unchanged Local Treatment: ∆ in Rail Coverage, 1963-84

Rail Lines with 5km buffer; 1951 LGDs; darker red = more negative, white = unchanged Treatment: Global

I Global: depends on structure of entire rail network

I Global Measure 1: Closeness centrality

I Population-weighted closeness centrality on rail network I Interpret as market access, average distance to other units

I Global Measure 2: Betweenness centrality

I Population-weighted betweenness centrality on rail network I Interpret as volume of traffic that passes through unit Treatment: Change in Quantile of Closeness Centrality Distribution, Population Weighted, 1963-84

1961 Districts Treatment: Change in Quantile of Betweenness Centrality Distribution, Population Weighted, 1963-84

1961 Districts Identification

I Stated goal of Beeching: remove unprofitable rail lines

I Rail lines with low revenue have low local traffic

I Local traffic is correlated with local economic conditions

I Main concern for identification of effect

I “By cutting according to local traffic, Beeching removed rail lines from places that were already struggling economically and would have fallen behind regardless of the cuts”

I Alternate control group: averted cuts

I Instruments for cuts

I Initial betweenness centrality in 1963 I On shortest path between top 5/10 cities

I Show robustness to including pre-treatment controls

I Other alternative: RDD in local traffic at “rail line” level Beeching’s Criterion: Measured Local Traffic per km of Rail Line, Beeching Report, Passenger/Freight

1961 Districts Beeching’s Criterion: Measured Local Traffic per km of Rail Line, Beeching Report, Total vs. ∆ Rail Coverage

1961 Districts Instrument: Dummy =1 if on Shortest Path Over 1963 Rail Network Between Top 5 Cities by Pop.

1961 Districts Instrument: Dummy =1 if on Shortest Path Over 1963 Rail Network Between Top 10 Cities by Pop.

1961 Districts Instrument: Quantile of Betweenness Centrality Distribution, Population Weighted, 1963

1961 Districts Controls: Road Network in 1956/2015

1951 LGDs; Source: Ordnance Survey, 10 mi. - in. planning map, May 1956 black = M-road, red = A-road, green = B-road; Source: Ordnance Survey Open Roads, March 2015 Controls: Road Network, Density in 1956/2015

Roads (km) per Area (km2); 1951 LGDs Data: Outcome Variables

d pop Population growth, % LGD

d ur Change in unemployment rate, percentage points Dist

d emp Change in employment rate, percentage points LGD

d class1 frac Change in fraction social class I, professionals LGD

d class5 frac Change in fraction social class V, unskilled LGD

d ag frac Change in fraction employment in agriculture Dist

d man frac Change in fraction employment in manufacturing Dist

d con frac Change in fraction employment in construction Dist

d util frac Change in fraction employment in utilities Dist

d serv frac Change in fraction employment in services Dist

d ag emp Change in ag. employment, % Dist

d man emp Change in manufacturing employment, % Dist

d con emp Change in construction employment, % Dist

d util emp Change in utilities employment, % Dist

d serv emp Change in services employment, % Dist Data: Treatment Variables

Local

rr cuts length Cut in railroad lines (km) LGD

rr cuts density Cut in railroad lines (km) per area (km2) LGD

rr cuts cover Cut in railroad coverage, 5 km buffer LGD

Global

d cent close qtile Change in quantile (100) of closeness centrality dist. Dist

d cent btwn qtile Change in quantile (100) of betweenness centrality dist. Dist

Placebo

rr cuts averted length Averted cut in railroad lines (km) LGD

rr cuts averted density Averted cut in railroad lines (km) per area (km2) LGD Data: Control Variables

Roads

d road density Change in road density (km per km2), 1956-2015 LGD

Geography

dist near cities Distance (km) to nearest top 5 city LGD

dist near coast Distance (km) to nearest coastline LGD

slope Average slope LGD Data: Instrumental Variables

Local

rr line slope Average slope within 100m buffer around rail lines LGD

rr line curve Abs. value of normalized diff. between actual and straight-line dist. for rail lines LGD

Global

cities top5 Dummy =1 if on shortest path over 1963 rail network between top 5 cities Dist

cities top10 Dummy =1 if on shortest path over 1963 rail network between top 10 cities Dist/LGD

cent btwn 1963 qtile Betweenness centrality, 1963, quantile (100) Dist Baseline DiD: Rail Cuts, Length Treatment

(1) - Full Sample (2) - Ex. Top 5 Cities (3) - (2), Ex. “Not proposed” group

d pop -0.227 (1.19) -0.195 (1.01) -0.094 (0.45) d emp -0.082 (5.58) -0.085 (5.66) -0.060 (3.90) d class1 frac 0.019 (2.86) 0.018 (2.70) 0.021 (2.93) d class5 frac 0.058 (5.36) 0.062 (5.75) 0.055 (4.74)

d ag frac -0.028 (1.22) -0.029 (1.26) -0.028 (1.14) d man frac 0.065 (2.77) 0.067 (2.83) 0.062 (2.37) d con frac 0.004 (1.24) 0.004 (1.31) 0.006 (1.66) d util frac 0.006 (0.40) 0.008 (0.53) 0.014 (0.84) d serv frac -0.029 (2.13) -0.031 (2.31) -0.029 (2.05) “All Measures”DiD: Rail Cuts, Length Treatment

rr cuts length d cent close qtile d cent btwn qtile

d pop -0.111 (0.53) -0.715 (3.11) 0.077 (1.08) d emp -0.060 (3.95) -0.033 (1.93) -0.003 (0.61) d class1 frac 0.020 (2.80) -0.024 (2.70) 0.007 (2.40) d class5 frac 0.055 (4.73) -0.005 (0.40) -0.001 (0.36)

d ag frac -0.024 (0.94) 0.151 (2.65) -0.020 (1.09) d man frac 0.056 (2.26) -0.147 (3.56) 0.027 (1.37) d con frac 0.006 (1.48) 0.012 (1.44) 0.005 (1.53) d util frac 0.030 (1.46) -0.001 (0.06) -0.005 (0.45) d serv frac -0.031 (1.79) -0.012 (0.28) -0.014 (0.74)

Ex. Top 5 Cities & “Not proposed” group Baseline DiD: Rail Cuts, Density Treatment

(1) - Full Sample (2) - Ex. Top 5 Cities (3) - (2), Ex. “Not proposed” group

d pop -0.330 (3.68) -0.329 (3.64) -0.282 (3.20) d emp -0.005 (0.63) -0.005 (0.60) 0.007 (0.73) d class1 frac -0.010 (2.70) -0.010 (2.85) -0.009 (2.38) d class5 frac -0.011 (1.79) -0.010 (1.60) -0.021 (2.99)

d ag frac 0.025 (2.22) 0.027 (2.34) 0.026 (2.08) d man frac 0.005 (0.37) 0.004 (0.30) -0.001 (0.07) d con frac 0.005 (2.01) 0.005 (2.01) 0.006 (2.35) d util frac -0.004 (0.39) -0.005 (0.45) -0.001 (0.17) d serv frac 0.029 (2.58) 0.030 (2.67) 0.017 (1.53) “All Measures”DiD: Rail Cuts, Density Treatment

rr cuts density d cent close qtile d cent btwn qtile

d pop -0.290 (3.33) -0.723 (3.16) 0.072 (1.03) d emp 0.006 (0.66) -0.031 (1.78) -0.004 (0.75) d class1 frac -0.009 (2.45) -0.025 (2.80) 0.007 (2.45) d class5 frac -0.021 (2.98) -0.007 (0.57) -0.001 (0.21)

d ag frac 0.025 (2.05) 0.152 (2.66) -0.020 (1.10) d man frac 0.000 (0.01) -0.149 (3.56) 0.028 (1.40) d con frac 0.006 (2.46) 0.012 (1.43) 0.005 (1.59) d util frac -0.001 (0.16) -0.002 (0.09) -0.004 (0.40) d serv frac 0.017 (1.49) -0.011 (0.26) -0.014 (0.75)

Ex. Top 5 Cities & “Not proposed” group Conclusion

I Beeching cuts had long-lasting negative effect I Baseline DiD rail cut response

I Length: employment rate, relative size of service sector fall; relative size of professional and unskilled workers, manufacturing sector increase I Density: fall in pop. growth, professionals; increase in agriculture, construction

I “All Measures” DiD

I Using rail length

I Length: same as baseline, but with no effect on service sector I Market access: lower pop. growth, fewer professionals, more agriculture, less manufacturing I Traffic: more professionals

I Using rail density:

I Density: baseline, plus fall in unskilled I Market access: same as using rail length I Traffic: same as using rail length Outcomes: Change in Unemployment Rate, 1951-81

1961 Districts Outcomes: ∆ in Frac. Employed in Ag., 1951-81

1961 Districts Outcomes: ∆ in Frac. Emp. in Manufacturing, 1951-81

1961 Districts Outcomes: ∆ in Frac. Emp. in Construction, 1951-81

1961 Districts Outcomes: ∆ in Frac. Emp. in Utilities, 1951-81

1961 Districts Outcomes: ∆ in Frac. Emp. in Services, 1951-81

1961 Districts Outcomes: ∆ in Ag. Emp., %, 1951-81

1961 Districts Outcomes: ∆ in Manufacturing Emp., %, 1951-81

1961 Districts Outcomes: ∆ in Construction Emp., %, 1951-81

1961 Districts Outcomes: ∆ in Utilities Emp., %, 1951-81

1961 Districts Outcomes: ∆ in Services Emp., %, 1951-81

1961 Districts