subways and urban growth: evidence from...

Subways and urban growth: Evidence from Earth

Marco Gonzalez-Navarro

University of Toronto

Matthew A. Turner

Brown University

July 13 2015

Motivation

I There are 138 cities worldwide with subway systems.Construction costs ∼ 3 trillion 2005usd

I Construction costs average 250 million per kilometer

I The Societe du Grand Paris will invest 26 billion Euros to buildmore than 200 kilometres of new rapid transit lines by 2031.

I The Big Move Ontario - a 45 billion rail project

Motivation

I There are 138 cities worldwide with subway systems.Construction costs ∼ 3 trillion 2005usd

I Construction costs average 250 million per kilometer

I The Societe du Grand Paris will invest 26 billion Euros to buildmore than 200 kilometres of new rapid transit lines by 2031.

I The Big Move Ontario - a 45 billion rail project

Motivation

I There are 138 cities worldwide with subway systems.Construction costs ∼ 3 trillion 2005usd

I Construction costs average 250 million per kilometer

I The Societe du Grand Paris will invest 26 billion Euros to buildmore than 200 kilometres of new rapid transit lines by 2031.

I The Big Move Ontario - a 45 billion rail project

Motivation

I There are 138 cities worldwide with subway systems.Construction costs ∼ 3 trillion 2005usd

I Construction costs average 250 million per kilometer

I The Societe du Grand Paris will invest 26 billion Euros to buildmore than 200 kilometres of new rapid transit lines by 2031.

I The Big Move Ontario - a 45 billion rail project

Motivation

I There are 138 cities worldwide with subway systems.Construction costs ∼ 3 trillion 2005usd

I Construction costs average 250 million per kilometer

I The Societe du Grand Paris will invest 26 billion Euros to buildmore than 200 kilometres of new rapid transit lines by 2031.

I The Big Move Ontario - a 45 billion rail project

Policy motivation

I Subways are simply another form of transport andconstruction decisions should be based on transport demand(i.e. McFadden, 1974)

I In reality capital and operating costs receive large subsidiesI Example: Mexico City subway requires USD 500 million per

year from general revenue representing 60% of budgetI Example: Boston system requires a subsidy of around 80% of

its operating budget

I And these subsidies are justified on three main grounds(externalities):

I Job creation (city size)I Ontario: “Over the next 25 years we can create 430,000 jobs

and generate approximately 70 billion in business revenue.”I Paris: “[Allow us to] Create one million jobs over the next 25

years”

I pollution reduction (Chen and Whalley, 2012)I congestion reduction (Anderson, 2014)

I In this paper we assess the validity of the first claim

Policy motivation

I Subways are simply another form of transport andconstruction decisions should be based on transport demand(i.e. McFadden, 1974)

I In reality capital and operating costs receive large subsidies

I Example: Mexico City subway requires USD 500 million peryear from general revenue representing 60% of budget

I Example: Boston system requires a subsidy of around 80% ofits operating budget

I And these subsidies are justified on three main grounds(externalities):

I Job creation (city size)I Ontario: “Over the next 25 years we can create 430,000 jobs

and generate approximately 70 billion in business revenue.”I Paris: “[Allow us to] Create one million jobs over the next 25

years”

I pollution reduction (Chen and Whalley, 2012)I congestion reduction (Anderson, 2014)

I In this paper we assess the validity of the first claim

Policy motivation

I Subways are simply another form of transport andconstruction decisions should be based on transport demand(i.e. McFadden, 1974)

I In reality capital and operating costs receive large subsidiesI Example: Mexico City subway requires USD 500 million per

year from general revenue representing 60% of budgetI Example: Boston system requires a subsidy of around 80% of

its operating budget

I And these subsidies are justified on three main grounds(externalities):

I Job creation (city size)I Ontario: “Over the next 25 years we can create 430,000 jobs

and generate approximately 70 billion in business revenue.”I Paris: “[Allow us to] Create one million jobs over the next 25

years”

I pollution reduction (Chen and Whalley, 2012)I congestion reduction (Anderson, 2014)

I In this paper we assess the validity of the first claim

Policy motivation

I Subways are simply another form of transport andconstruction decisions should be based on transport demand(i.e. McFadden, 1974)

I In reality capital and operating costs receive large subsidiesI Example: Mexico City subway requires USD 500 million per

year from general revenue representing 60% of budgetI Example: Boston system requires a subsidy of around 80% of

its operating budget

I And these subsidies are justified on three main grounds(externalities):

I Job creation (city size)

I Ontario: “Over the next 25 years we can create 430,000 jobsand generate approximately 70 billion in business revenue.”

I Paris: “[Allow us to] Create one million jobs over the next 25years”

I pollution reduction (Chen and Whalley, 2012)I congestion reduction (Anderson, 2014)

I In this paper we assess the validity of the first claim

Policy motivation

I Subways are simply another form of transport andconstruction decisions should be based on transport demand(i.e. McFadden, 1974)

I In reality capital and operating costs receive large subsidiesI Example: Mexico City subway requires USD 500 million per

year from general revenue representing 60% of budgetI Example: Boston system requires a subsidy of around 80% of

its operating budget

I And these subsidies are justified on three main grounds(externalities):

I Job creation (city size)I Ontario: “Over the next 25 years we can create 430,000 jobs

and generate approximately 70 billion in business revenue.”

I Paris: “[Allow us to] Create one million jobs over the next 25years”

I pollution reduction (Chen and Whalley, 2012)I congestion reduction (Anderson, 2014)

I In this paper we assess the validity of the first claim

Policy motivation

I Subways are simply another form of transport andconstruction decisions should be based on transport demand(i.e. McFadden, 1974)

I In reality capital and operating costs receive large subsidiesI Example: Mexico City subway requires USD 500 million per

year from general revenue representing 60% of budgetI Example: Boston system requires a subsidy of around 80% of

its operating budget

I And these subsidies are justified on three main grounds(externalities):

I Job creation (city size)I Ontario: “Over the next 25 years we can create 430,000 jobs

and generate approximately 70 billion in business revenue.”I Paris: “[Allow us to] Create one million jobs over the next 25

years”

I pollution reduction (Chen and Whalley, 2012)I congestion reduction (Anderson, 2014)

I In this paper we assess the validity of the first claim

Policy motivation

I Subways are simply another form of transport andconstruction decisions should be based on transport demand(i.e. McFadden, 1974)

I In reality capital and operating costs receive large subsidiesI Example: Mexico City subway requires USD 500 million per

year from general revenue representing 60% of budgetI Example: Boston system requires a subsidy of around 80% of

its operating budget

I And these subsidies are justified on three main grounds(externalities):

I Job creation (city size)I Ontario: “Over the next 25 years we can create 430,000 jobs

and generate approximately 70 billion in business revenue.”I Paris: “[Allow us to] Create one million jobs over the next 25

years”

I pollution reduction (Chen and Whalley, 2012)

I congestion reduction (Anderson, 2014)

I In this paper we assess the validity of the first claim

Policy motivation

I Subways are simply another form of transport andconstruction decisions should be based on transport demand(i.e. McFadden, 1974)

I In reality capital and operating costs receive large subsidiesI Example: Mexico City subway requires USD 500 million per

year from general revenue representing 60% of budgetI Example: Boston system requires a subsidy of around 80% of

its operating budget

I And these subsidies are justified on three main grounds(externalities):

I Job creation (city size)I Ontario: “Over the next 25 years we can create 430,000 jobs

and generate approximately 70 billion in business revenue.”I Paris: “[Allow us to] Create one million jobs over the next 25

years”

I pollution reduction (Chen and Whalley, 2012)I congestion reduction (Anderson, 2014)

I In this paper we assess the validity of the first claim

Policy motivation

I Subways are simply another form of transport andconstruction decisions should be based on transport demand(i.e. McFadden, 1974)

I In reality capital and operating costs receive large subsidiesI Example: Mexico City subway requires USD 500 million per

year from general revenue representing 60% of budgetI Example: Boston system requires a subsidy of around 80% of

its operating budget

I And these subsidies are justified on three main grounds(externalities):

I Job creation (city size)I Ontario: “Over the next 25 years we can create 430,000 jobs

and generate approximately 70 billion in business revenue.”I Paris: “[Allow us to] Create one million jobs over the next 25

years”

I pollution reduction (Chen and Whalley, 2012)I congestion reduction (Anderson, 2014)

I In this paper we assess the validity of the first claim

Theoretical motivation

I All open city models (utility is equal across locations) predictthat a reduction in transportation costs in a given city willmake it larger

I This fact has been shown for roads, but remains an openquestion for subways

Objective

I Estimate the effect of subway system extent onI city population size

I lights at nightI spatial configuration

I Track cities at 5 year intervals from 1950 to 2010 for thelargest 632 cities in the world.

I Identification of causal effects from panel time series variationin subway extent, population and lights using OLS and IV.

Objective

I Estimate the effect of subway system extent onI city population sizeI lights at night

I spatial configuration

I Track cities at 5 year intervals from 1950 to 2010 for thelargest 632 cities in the world.

I Identification of causal effects from panel time series variationin subway extent, population and lights using OLS and IV.

Objective

I Estimate the effect of subway system extent onI city population sizeI lights at nightI spatial configuration

I Track cities at 5 year intervals from 1950 to 2010 for thelargest 632 cities in the world.

I Identification of causal effects from panel time series variationin subway extent, population and lights using OLS and IV.

Objective

I Estimate the effect of subway system extent onI city population sizeI lights at nightI spatial configuration

I Track cities at 5 year intervals from 1950 to 2010 for thelargest 632 cities in the world.

I Identification of causal effects from panel time series variationin subway extent, population and lights using OLS and IV.

Objective

I Estimate the effect of subway system extent onI city population sizeI lights at nightI spatial configuration

I Track cities at 5 year intervals from 1950 to 2010 for thelargest 632 cities in the world.

I Identification of causal effects from panel time series variationin subway extent, population and lights using OLS and IV.

Related Literature

I Recent literature has studied how roads and railroads affectthe growth and organization of economic activity

I Baum-Snow (QJE 2007), Duranton and Turner (RES 2012),Garcia-Lopez et al.(JUE 2013), Faber (2014)

I Despite their policy relevance, subways are mostlyunexamined. This reflects a data problem that this paperovercomes for the first time

Related Literature

I Existing papers on subways perform a D-D within a singlecity:

I Billings (RSUE 2011) - property valuesI Gibbons and Machin (JUE 2005) - property values

I By construction, they can’t estimate city level effects(Ahlfeldt, 2013)

I This is the only investigation of the relationship between citylevel variation in subway extent and city growth.

I Other papers to consider city level measures of subways(Baum-Snow and Kahn 2005) examine ridership (USA).

DataPopulation Data

UN World Cities data

I annual population of every metropolitan area with pop>750,000 any time between 1950-2010

I standardized definition of city extent

I 632 large cities in dataset

I we mainly rely on t > 1970

These are metropolitan area definitions and are comparableacross time and countries.

DataSubways

I Subway defined as:I rapid transit rail system (aka Metro); high platform electric

vehicle, not necessarily underground, typically 2-8 carsI must have an exclusive right-of-way: No at grade crossings

with cars or peopleI this excludes: trams; trolleys; bus lines; bus-rapid transit; long

distance rail and suburban railI common sense ‘subways’, e.g., NYC subway, Paris Metro,

BART.

I Our data consists of latitude, longitude and opening year forevery subway station in the world up to 2010.

I We also create system maps by connecting stations alongeach subway line.

DataLights at night

I Night light intensity for 1km squares on a regular grid for thewhole world, for ca. 1995, 2000, 2005, 2010.

I We use ‘radiance calibrated lights at night’. They are nottopcoded and show internal structure of big cities.

I Henderson, Storeygard and Weil (2012) show that lights arecorrelated with country GDP. They are also highly correlatedwith city population.

This gives a consistent panel describing all large cities in the world.It allows an investigation of the internal structure of cities.

DataOther control variables

Not time varying:

I climate and topography,

I distance to coast and international border,

I capital city status.

Time varying:

I country GDP per person from Penn World Tables.

Subways over time

London lights and subways1995 - Ellipses are projected 5km and 25km radius circles, blue are bodies of water.

Shanghai lights and subways1995

Shanghai lights and subways2000

Shanghai lights and subways2005

Shanghai lights and subways2010

Descriptive statistics for 2010 - All cities

World Africa Asia Aus. Europe N. America S. America

All cities

N 632 73 341 6 57 99 56

Population (1000’s) 2,427 2,104 2,511 2,429 1,921 2,441 2,825Total Stations 7,886 51 2,977 0 2,782 1,598 478Total route km 10,686 56 4,224 0 3,558 2,219 628Mean light in 25km disk 62 24 52 65 84 116 58Corr. lights & pop. 0.57 0.56 0.58 0.85 0.69 0.71 0.92

Lights data are based on 2010 lights at night imagery.

Descriptive statistics for 2010 - All cities

World Africa Asia Aus. Europe N. America S. America

All cities

N 632 73 341 6 57 99 56Population (1000’s) 2,427 2,104 2,511 2,429 1,921 2,441 2,825

Total Stations 7,886 51 2,977 0 2,782 1,598 478Total route km 10,686 56 4,224 0 3,558 2,219 628Mean light in 25km disk 62 24 52 65 84 116 58Corr. lights & pop. 0.57 0.56 0.58 0.85 0.69 0.71 0.92

Lights data are based on 2010 lights at night imagery.

Descriptive statistics for 2010 - All cities

World Africa Asia Aus. Europe N. America S. America

All cities

N 632 73 341 6 57 99 56Population (1000’s) 2,427 2,104 2,511 2,429 1,921 2,441 2,825Total Stations 7,886 51 2,977 0 2,782 1,598 478Total route km 10,686 56 4,224 0 3,558 2,219 628

Mean light in 25km disk 62 24 52 65 84 116 58Corr. lights & pop. 0.57 0.56 0.58 0.85 0.69 0.71 0.92

Lights data are based on 2010 lights at night imagery.

Descriptive statistics for 2010 - All cities

World Africa Asia Aus. Europe N. America S. America

All cities

N 632 73 341 6 57 99 56Population (1000’s) 2,427 2,104 2,511 2,429 1,921 2,441 2,825Total Stations 7,886 51 2,977 0 2,782 1,598 478Total route km 10,686 56 4,224 0 3,558 2,219 628Mean light in 25km disk 62 24 52 65 84 116 58Corr. lights & pop. 0.57 0.56 0.58 0.85 0.69 0.71 0.92

Lights data are based on 2010 lights at night imagery.

Descriptive statistics for 2010 - Subway cities

World Africa Asia Aus. Europe N. America S. America

N 138 1 53 0 40 30 14Pop. 4,706 11,031 5,951 2,260 4,814 6,300

stations 57 51 56 70 53 34Route km 77 56 80 89 74 45Pop. per route km 61 196 75 25 65 141Pop. per station 82 216 106 32 90 185log(Pop.) 14.9 16.2 15.1 14.4 15.1 15.3log(Stations) 3.55 3.93 3.51 3.87 3.33 3.25∆t log(Stations) 0.204 0.101 0.305 0.158 0.140 0.117∆2

t log(Stations) -0.047 0.000 -0.054 -0.018 -0.066 -0.088∆ log(Pop.) 0.113 0.124 0.144 0.045 0.123 0.170∆2

t log(Pop.) -0.011 -0.014 -0.012 -0.005 -0.013 -0.017Mean light in 25km disk 122 212 117 95 171 109Corr. lights & pop. 0.67 0.67 0.69 0.78 0.91

Population levels reported in thousands. Lights data are based on 2010 lights at night imagery. Allpopulation and subway growth rates are averages over the period 1950 to 2010.

Descriptive statistics for 2010 - Subway cities

World Africa Asia Aus. Europe N. America S. America

N 138 1 53 0 40 30 14Pop. 4,706 11,031 5,951 2,260 4,814 6,300stations 57 51 56 70 53 34Route km 77 56 80 89 74 45Pop. per route km 61 196 75 25 65 141Pop. per station 82 216 106 32 90 185

log(Pop.) 14.9 16.2 15.1 14.4 15.1 15.3log(Stations) 3.55 3.93 3.51 3.87 3.33 3.25∆t log(Stations) 0.204 0.101 0.305 0.158 0.140 0.117∆2

t log(Stations) -0.047 0.000 -0.054 -0.018 -0.066 -0.088∆ log(Pop.) 0.113 0.124 0.144 0.045 0.123 0.170∆2

t log(Pop.) -0.011 -0.014 -0.012 -0.005 -0.013 -0.017Mean light in 25km disk 122 212 117 95 171 109Corr. lights & pop. 0.67 0.67 0.69 0.78 0.91

Population levels reported in thousands. Lights data are based on 2010 lights at night imagery. Allpopulation and subway growth rates are averages over the period 1950 to 2010.

Descriptive statistics for 2010 - Subway cities

World Africa Asia Aus. Europe N. America S. America

N 138 1 53 0 40 30 14Pop. 4,706 11,031 5,951 2,260 4,814 6,300stations 57 51 56 70 53 34Route km 77 56 80 89 74 45Pop. per route km 61 196 75 25 65 141Pop. per station 82 216 106 32 90 185log(Pop.) 14.9 16.2 15.1 14.4 15.1 15.3log(Stations) 3.55 3.93 3.51 3.87 3.33 3.25∆t log(Stations) 0.204 0.101 0.305 0.158 0.140 0.117∆2

t log(Stations) -0.047 0.000 -0.054 -0.018 -0.066 -0.088

∆ log(Pop.) 0.113 0.124 0.144 0.045 0.123 0.170∆2

t log(Pop.) -0.011 -0.014 -0.012 -0.005 -0.013 -0.017Mean light in 25km disk 122 212 117 95 171 109Corr. lights & pop. 0.67 0.67 0.69 0.78 0.91

Population levels reported in thousands. Lights data are based on 2010 lights at night imagery. Allpopulation and subway growth rates are averages over the period 1950 to 2010.

Descriptive statistics for 2010 - Subway cities

World Africa Asia Aus. Europe N. America S. America

N 138 1 53 0 40 30 14Pop. 4,706 11,031 5,951 2,260 4,814 6,300stations 57 51 56 70 53 34Route km 77 56 80 89 74 45Pop. per route km 61 196 75 25 65 141Pop. per station 82 216 106 32 90 185log(Pop.) 14.9 16.2 15.1 14.4 15.1 15.3log(Stations) 3.55 3.93 3.51 3.87 3.33 3.25∆t log(Stations) 0.204 0.101 0.305 0.158 0.140 0.117∆2

t log(Stations) -0.047 0.000 -0.054 -0.018 -0.066 -0.088∆ log(Pop.) 0.113 0.124 0.144 0.045 0.123 0.170∆2

t log(Pop.) -0.011 -0.014 -0.012 -0.005 -0.013 -0.017Mean light in 25km disk 122 212 117 95 171 109Corr. lights & pop. 0.67 0.67 0.69 0.78 0.91

Population levels reported in thousands. Lights data are based on 2010 lights at night imagery. Allpopulation and subway growth rates are averages over the period 1950 to 2010.

Population and subways in a few large cities

City Name Pop. Stations Stations pp.

Tokyo 36,933 255 0.69Delhi 21,935 128 0.58Mexico City 20,142 147 0.73New York 20,104 489 2.43Sao Paulo 19,649 62 0.32Shanghai 19,554 239 1.22Mumbai 19,422 . .Beijing 15,000 124 0.83Dhaka 14,930 . .Kolkata 14,283 23 0.16Karachi 13,500 . .Buenos Aires 13,370 76 0.57Los Angeles 13,223 30 0.23Rio de Janeiro 11,867 35 0.29Manila 11,654 43 0.37Moscow 11,472 168 1.46

Population levels reported in thousands. Stations per person (pp) is per 100,000 residents.

Shanghai population and subways

Brasilia population and subways

Event study - City growth and subway opening

City growth rates before and after subway opening. t = 0 isopening year of subway system, growth rates are averages over fiveyear periods. Constant sample, population > 1 million in 1970.

Event study - City growth and subway opening detrended

City growth rate residuals (after controlling for year and continent F.E.) before

and after subway opening. t = 0 is opening year of subway system, growth

rates are averages over five year periods. Constant sample, population > 1

million in 1970.

Event study - City growth and subway opening - long run

City growth rates before and after subway opening. t = 0 isopening year of subway system, growth rates are averages over fiveyear periods. Long run effects.

Subway expansion events and mean city year populationgrowth rates

t − 2 t − 1 t t + 1 t + 2 N

Panel a

0.063 0.054*** 1380.078 0.067** 0.064** 60

0.090*** 0.073 2040.120** 0.107*** 0.083 141

0.075*** 0.061 0.052** 64

Notes: Each row in panel (a) shows growth rates of cities in consecutive time periods. t is a period ofsubway expansion. j 6= t is a period with no subway expansion. Stars indicate a significant difference ofgrowth rate compared to period t. *** 1%, ** 5%, * 10% significance respectively.

Subway expansion events and mean city year populationgrowth rates

t − 2 t − 1 t t + 1 t + 2 N

Panel b

−0.001 138−0.001 −0.009 60

0.006* 2040.013* 0.012* 141

0.009* −0.007 64

Notes: Each row shows the difference in growth rates of cities (relative to period t) in consecutive timeperiods from a regression controlling for year×continent dummies. t is a period of subway expansion.j 6= t is a period with no subway expansion. Stars indicate a significant difference of growth ratecompared to period t. *** 1%, ** 5%, * 10% significance respectively.

City size thresholds?

City size thresholds? Before and after subway starts

City size thresholds? Overlapping sizes

Discussion of descriptive results

The data show that

I larger cities have more subways

I population growth rates do not increase following subwayopenings and expansions

Does this reflect the casual effect of subways or the equilibriumassignment of subways to cities? To check, examine the roles of:

I omitted variables in cross-section (or time series), e.g.,capital/coastal/flat cities are big and build subways

I endogenous subway construction

I dynamics effects

Discussion of descriptive results

The data show that

I larger cities have more subways

I population growth rates do not increase following subwayopenings and expansions

Does this reflect the casual effect of subways or the equilibriumassignment of subways to cities? To check, examine the roles of:

I omitted variables in cross-section (or time series), e.g.,capital/coastal/flat cities are big and build subways

I endogenous subway construction

I dynamics effects

Pooled ols

All cities Subway cities

(1) (2) (3) (4) (5) (6) (7) (8)ln(popt) ln(popt) ln(popt) ln(popt) ln(popt) ln(popt) ln(popt) ln(Lightst)

ln(st) 0.48∗∗∗ 0.31∗∗∗ 0.30∗∗∗ 0.23∗∗∗ 0.54∗∗∗ 0.22∗∗∗

(0.02) (0.03) (0.03) (0.04) (0.07) (0.04)

ln(route kmt) 0.20∗∗∗

(0.03)

ln(subway linest) 0.47∗∗∗

(0.06)

ln(GDPpct) 0.18∗∗∗ -0.18∗∗ -0.21∗∗ -0.20∗∗ -0.22∗∗ -0.29∗∗ 0.13(0.04) (0.09) (0.07) (0.07) (0.07) (0.13) (0.09)

Geographic controls No Yes Yes Yes Yes Yes Yes Yes

YearXContinent dummies No Yes Yes Yes Yes Yes Yes Yes

Mean of Dep Variable 13.35 13.44 14.48 14.99 14.99 14.99 15.18 4.80Mean of subways regressor 0.38 0.40 1.88 2.63 2.79 1.10 4.12 3.43SD subways regressor 1.15 1.17 1.92 1.86 1.99 0.93 0.88 1.31R-squared 0.18 0.48 0.46 0.49 0.47 0.52 0.67 0.51Number of cities 632 629 137 99 99 99 75 99Number of subway cities 138 137 137 99 99 99 75 99Number of periods 13 13 13 9 9 9 9 4p-value 1st order autocorr 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00p-value 2nd order autocorr 0.00 0.00 0.00 0.00 0.00 0.00 0.00p-value 3rd order autocorr 0.00 0.00 0.00 0.00 0.00 0.00 0.00Observations 8216 7396 1565 844 844 844 355 396

Dependent variable: Log population of metropolitan area in period t (except last column see (8) below).

City-level clustered standard errors in parentheses. Stars denote significance levels: * 0.10, ** 0.05, *** 0.01.

Geographic controls are continent dummy, capital city dummy, log km to ocean, log km to land border, and log km to major navigable river.

(1)-Pooled cross section. (2)-Add geographic controls, gdp pc control, and yearXcontinent dummies.

(3)-Only cities with subway by 2010. (4)-Cities with subway by 2010 and at least 1 million population in 1970, t >=1970.

(5)-Log route km as main regressor, t >=1970. (6)-Log subway lines as main regressor, t >=1970. (7)-Add log(st−4 + 1) > 0 restriction.

(8)-Dep. var. is log mean radiance calibrated lights in a 25km circle around the centroid of the city.

Pooled ols

All cities Subway cities

(1) (2) (3) (4) (5) (6) (7) (8)ln(popt) ln(popt) ln(popt) ln(popt) ln(popt) ln(popt) ln(popt) ln(Lightst)

ln(st) 0.48∗∗∗ 0.31∗∗∗ 0.30∗∗∗ 0.23∗∗∗ 0.54∗∗∗ 0.22∗∗∗

(0.02) (0.03) (0.03) (0.04) (0.07) (0.04)

ln(route kmt) 0.20∗∗∗

(0.03)

ln(subway linest) 0.47∗∗∗

(0.06)

ln(GDPpct) 0.18∗∗∗ -0.18∗∗ -0.21∗∗ -0.20∗∗ -0.22∗∗ -0.29∗∗ 0.13(0.04) (0.09) (0.07) (0.07) (0.07) (0.13) (0.09)

Geographic controls No Yes Yes Yes Yes Yes Yes Yes

YearXContinent dummies No Yes Yes Yes Yes Yes Yes Yes

Mean of Dep Variable 13.35 13.44 14.48 14.99 14.99 14.99 15.18 4.80Mean of subways regressor 0.38 0.40 1.88 2.63 2.79 1.10 4.12 3.43SD subways regressor 1.15 1.17 1.92 1.86 1.99 0.93 0.88 1.31R-squared 0.18 0.48 0.46 0.49 0.47 0.52 0.67 0.51Number of cities 632 629 137 99 99 99 75 99Number of subway cities 138 137 137 99 99 99 75 99Number of periods 13 13 13 9 9 9 9 4p-value 1st order autocorr 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00p-value 2nd order autocorr 0.00 0.00 0.00 0.00 0.00 0.00 0.00p-value 3rd order autocorr 0.00 0.00 0.00 0.00 0.00 0.00 0.00Observations 8216 7396 1565 844 844 844 355 396

Dependent variable: Log population of metropolitan area in period t (except last column see (8) below).

City-level clustered standard errors in parentheses. Stars denote significance levels: * 0.10, ** 0.05, *** 0.01.

Geographic controls are continent dummy, capital city dummy, log km to ocean, log km to land border, and log km to major navigable river.

(1)-Pooled cross section. (2)-Add geographic controls, gdp pc control, and yearXcontinent dummies.

(3)-Only cities with subway by 2010. (4)-Cities with subway by 2010 and at least 1 million population in 1970, t >=1970.

(5)-Log route km as main regressor, t >=1970. (6)-Log subway lines as main regressor, t >=1970. (7)-Add log(st−4 + 1) > 0 restriction.

(8)-Dep. var. is log mean radiance calibrated lights in a 25km circle around the centroid of the city.

First differences

(1) (2) (3) (4) (5) (6) (7) (8) (9)∆ ln (popt) ∆ ln (popt) ∆ ln (popt) ∆ ln (popt) ∆ ln (pop70−10) ∆ ln(Lightst) ∆ ln (popt) ∆ ln (popt) ∆ ln (popt)

∆ ln(st) 0.008∗∗ 0.007∗∗ 0.027 0.056∗∗ 0.074∗∗∗ 0.023(0.004) (0.003) (0.022) (0.023) (0.019) (0.145)

∆ ln(route kmt) 0.007∗∗

(0.003)

∆ ln(subway linest) 0.019∗∗

(0.009)

∆ln(s70−10) 0.095∗∗∗

(0.025)

∆ ln(GDPpct) 0.165∗∗∗ 0.164∗∗∗ 0.164∗∗∗ 0.806∗∗ 0.715∗∗∗ 0.098∗∗ 0.155∗∗∗ 0.107∗∗

(0.035) (0.035) (0.035) (0.330) (0.128) (0.029) (0.033) (0.045)

YearXContinent dummies No Yes Yes Yes No Yes Yes Yes Yes

Geographic controls No No No No Yes No No No No

Continent dummies No Yes Yes Yes Yes Yes Yes Yes Yes

Mean of Dep Variable 0.077 0.078 0.078 0.078 0.557 0.017 0.040 0.078 0.040Mean of subways regressor 0.31 0.30 0.32 0.12 0.11 0.28 0.09 0.30 0.09SD subways regressor 0.74 0.75 0.80 0.29 0.21 0.68 0.16 0.75 0.16R-squared 0.01 0.40 0.40 0.40 0.56 0.52 0.39 0.08 0.38Number of cities 99 99 99 99 99 99 75 99 75Number of subway cities 99 99 99 99 99 99 75 99 75Number of periods 9 9 9 9 1 3 9 9 9p-value 2nd order autocorr 0.00 0.00 0.00 0.00 0.00 0.00 0.00p-value 3rd order autocorr 0.00 0.00 0.00 0.00 0.05 0.01 0.10F-stat excluded instrument 153.49 12.88Observations 891 828 828 828 99 297 347 828 347

Dependent variable: Change in log population of metropolitan area in a 5 year period. Sample is large subway cities (over 1 million in 1970).

City-level clustered standard errors in parentheses. Stars denote significance levels: * 0.10, ** 0.05, *** 0.01. All regressions t >=1970.

(1)- No controls. (2)-Add year dummies and change in log gdp controls.

(3)- Use change in log route km as regressor. (4)-Use change in log subway lines as regressor.

(5)- Long difference regression 1970-2010. (6)-Dep. var. is change in log mean radiance calibrated lights in a 25km circle around the centroid of the city.

(7)- Add log(st−4 + 1) > 0 restriction. (8) Instrument ∆ ln (st + 1) with log(st−4 + 1). (9) Add log(st−4 + 1) > 0 restriction.

First differences

(1) (2) (3) (4) (5) (6) (7) (8) (9)∆ ln (popt) ∆ ln (popt) ∆ ln (popt) ∆ ln (popt) ∆ ln (pop70−10) ∆ ln(Lightst) ∆ ln (popt) ∆ ln (popt) ∆ ln (popt)

∆ ln(st) 0.008∗∗ 0.007∗∗ 0.027 0.056∗∗ 0.074∗∗∗ 0.023(0.004) (0.003) (0.022) (0.023) (0.019) (0.145)

∆ ln(route kmt) 0.007∗∗

(0.003)

∆ ln(subway linest) 0.019∗∗

(0.009)

∆ln(s70−10) 0.095∗∗∗

(0.025)

∆ ln(GDPpct) 0.165∗∗∗ 0.164∗∗∗ 0.164∗∗∗ 0.806∗∗ 0.715∗∗∗ 0.098∗∗ 0.155∗∗∗ 0.107∗∗

(0.035) (0.035) (0.035) (0.330) (0.128) (0.029) (0.033) (0.045)

YearXContinent dummies No Yes Yes Yes No Yes Yes Yes Yes

Geographic controls No No No No Yes No No No No

Continent dummies No Yes Yes Yes Yes Yes Yes Yes Yes

Mean of Dep Variable 0.077 0.078 0.078 0.078 0.557 0.017 0.040 0.078 0.040Mean of subways regressor 0.31 0.30 0.32 0.12 0.11 0.28 0.09 0.30 0.09SD subways regressor 0.74 0.75 0.80 0.29 0.21 0.68 0.16 0.75 0.16R-squared 0.01 0.40 0.40 0.40 0.56 0.52 0.39 0.08 0.38Number of cities 99 99 99 99 99 99 75 99 75Number of subway cities 99 99 99 99 99 99 75 99 75Number of periods 9 9 9 9 1 3 9 9 9p-value 2nd order autocorr 0.00 0.00 0.00 0.00 0.00 0.00 0.00p-value 3rd order autocorr 0.00 0.00 0.00 0.00 0.05 0.01 0.10F-stat excluded instrument 153.49 12.88Observations 891 828 828 828 99 297 347 828 347

Dependent variable: Change in log population of metropolitan area in a 5 year period. Sample is large subway cities (over 1 million in 1970).

City-level clustered standard errors in parentheses. Stars denote significance levels: * 0.10, ** 0.05, *** 0.01. All regressions t >=1970.

(1)- No controls. (2)-Add year dummies and change in log gdp controls.

(3)- Use change in log route km as regressor. (4)-Use change in log subway lines as regressor.

(5)- Long difference regression 1970-2010. (6)-Dep. var. is change in log mean radiance calibrated lights in a 25km circle around the centroid of the city.

(7)- Add log(st−4 + 1) > 0 restriction. (8) Instrument ∆ ln (st + 1) with log(st−4 + 1). (9) Add log(st−4 + 1) > 0 restriction.

First differences

(1) (2) (3) (4) (5) (6) (7) (8) (9)∆ ln (popt) ∆ ln (popt) ∆ ln (popt) ∆ ln (popt) ∆ ln (pop70−10) ∆ ln(Lightst) ∆ ln (popt) ∆ ln (popt) ∆ ln (popt)

∆ ln(st) 0.008∗∗ 0.007∗∗ 0.027 0.056∗∗ 0.074∗∗∗ 0.023(0.004) (0.003) (0.022) (0.023) (0.019) (0.145)

∆ ln(route kmt) 0.007∗∗

(0.003)

∆ ln(subway linest) 0.019∗∗

(0.009)

∆ln(s70−10) 0.095∗∗∗

(0.025)

∆ ln(GDPpct) 0.165∗∗∗ 0.164∗∗∗ 0.164∗∗∗ 0.806∗∗ 0.715∗∗∗ 0.098∗∗ 0.155∗∗∗ 0.107∗∗

(0.035) (0.035) (0.035) (0.330) (0.128) (0.029) (0.033) (0.045)

YearXContinent dummies No Yes Yes Yes No Yes Yes Yes Yes

Geographic controls No No No No Yes No No No No

Continent dummies No Yes Yes Yes Yes Yes Yes Yes Yes

Mean of Dep Variable 0.077 0.078 0.078 0.078 0.557 0.017 0.040 0.078 0.040Mean of subways regressor 0.31 0.30 0.32 0.12 0.11 0.28 0.09 0.30 0.09SD subways regressor 0.74 0.75 0.80 0.29 0.21 0.68 0.16 0.75 0.16R-squared 0.01 0.40 0.40 0.40 0.56 0.52 0.39 0.08 0.38Number of cities 99 99 99 99 99 99 75 99 75Number of subway cities 99 99 99 99 99 99 75 99 75Number of periods 9 9 9 9 1 3 9 9 9p-value 2nd order autocorr 0.00 0.00 0.00 0.00 0.00 0.00 0.00p-value 3rd order autocorr 0.00 0.00 0.00 0.00 0.05 0.01 0.10F-stat excluded instrument 153.49 12.88Observations 891 828 828 828 99 297 347 828 347

Dependent variable: Change in log population of metropolitan area in a 5 year period. Sample is large subway cities (over 1 million in 1970).

City-level clustered standard errors in parentheses. Stars denote significance levels: * 0.10, ** 0.05, *** 0.01. All regressions t >=1970.

(1)- No controls. (2)-Add year dummies and change in log gdp controls.

(3)- Use change in log route km as regressor. (4)-Use change in log subway lines as regressor.

(5)- Long difference regression 1970-2010. (6)-Dep. var. is change in log mean radiance calibrated lights in a 25km circle around the centroid of the city.

(7)- Add log(st−4 + 1) > 0 restriction. (8) Instrument ∆ ln (st + 1) with log(st−4 + 1). (9) Add log(st−4 + 1) > 0 restriction.

First differences

(1) (2) (3) (4) (5) (6) (7) (8) (9)∆ ln (popt) ∆ ln (popt) ∆ ln (popt) ∆ ln (popt) ∆ ln (pop70−10) ∆ ln(Lightst) ∆ ln (popt) ∆ ln (popt) ∆ ln (popt)

∆ ln(st) 0.008∗∗ 0.007∗∗ 0.027 0.056∗∗ 0.074∗∗∗ 0.023(0.004) (0.003) (0.022) (0.023) (0.019) (0.145)

∆ ln(route kmt) 0.007∗∗

(0.003)

∆ ln(subway linest) 0.019∗∗

(0.009)

∆ln(s70−10) 0.095∗∗∗

(0.025)

∆ ln(GDPpct) 0.165∗∗∗ 0.164∗∗∗ 0.164∗∗∗ 0.806∗∗ 0.715∗∗∗ 0.098∗∗ 0.155∗∗∗ 0.107∗∗

(0.035) (0.035) (0.035) (0.330) (0.128) (0.029) (0.033) (0.045)

YearXContinent dummies No Yes Yes Yes No Yes Yes Yes Yes

Geographic controls No No No No Yes No No No No

Continent dummies No Yes Yes Yes Yes Yes Yes Yes Yes

Mean of Dep Variable 0.077 0.078 0.078 0.078 0.557 0.017 0.040 0.078 0.040Mean of subways regressor 0.31 0.30 0.32 0.12 0.11 0.28 0.09 0.30 0.09SD subways regressor 0.74 0.75 0.80 0.29 0.21 0.68 0.16 0.75 0.16R-squared 0.01 0.40 0.40 0.40 0.56 0.52 0.39 0.08 0.38Number of cities 99 99 99 99 99 99 75 99 75Number of subway cities 99 99 99 99 99 99 75 99 75Number of periods 9 9 9 9 1 3 9 9 9p-value 2nd order autocorr 0.00 0.00 0.00 0.00 0.00 0.00 0.00p-value 3rd order autocorr 0.00 0.00 0.00 0.00 0.05 0.01 0.10F-stat excluded instrument 153.49 12.88Observations 891 828 828 828 99 297 347 828 347

Dependent variable: Change in log population of metropolitan area in a 5 year period. Sample is large subway cities (over 1 million in 1970).

City-level clustered standard errors in parentheses. Stars denote significance levels: * 0.10, ** 0.05, *** 0.01. All regressions t >=1970.

(1)- No controls. (2)-Add year dummies and change in log gdp controls.

(3)- Use change in log route km as regressor. (4)-Use change in log subway lines as regressor.

(5)- Long difference regression 1970-2010. (6)-Dep. var. is change in log mean radiance calibrated lights in a 25km circle around the centroid of the city.

(7)- Add log(st−4 + 1) > 0 restriction. (8) Instrument ∆ ln (st + 1) with log(st−4 + 1). (9) Add log(st−4 + 1) > 0 restriction.

Long differences (1970-2000)

Long differences (1970-2000 and 1970-2010)

First differences

(1) (2) (3) (4) (5) (6) (7) (8) (9)∆ ln (popt) ∆ ln (popt) ∆ ln (popt) ∆ ln (popt) ∆ ln (pop70−10) ∆ ln(Lightst) ∆ ln (popt) ∆ ln (popt) ∆ ln (popt)

∆ ln(st) 0.008∗∗ 0.007∗∗ 0.027 0.056∗∗ 0.074∗∗∗ 0.023(0.004) (0.003) (0.022) (0.023) (0.019) (0.145)

∆ ln(route kmt) 0.007∗∗

(0.003)

∆ ln(subway linest) 0.019∗∗

(0.009)

∆ln(s70−10) 0.095∗∗∗

(0.025)

∆ ln(GDPpct) 0.165∗∗∗ 0.164∗∗∗ 0.164∗∗∗ 0.806∗∗ 0.715∗∗∗ 0.098∗∗ 0.155∗∗∗ 0.107∗∗

(0.035) (0.035) (0.035) (0.330) (0.128) (0.029) (0.033) (0.045)

YearXContinent dummies No Yes Yes Yes No Yes Yes Yes Yes

Geographic controls No No No No Yes No No No No

Continent dummies No Yes Yes Yes Yes Yes Yes Yes Yes

Mean of Dep Variable 0.077 0.078 0.078 0.078 0.557 0.017 0.040 0.078 0.040Mean of subways regressor 0.31 0.30 0.32 0.12 0.11 0.28 0.09 0.30 0.09SD subways regressor 0.74 0.75 0.80 0.29 0.21 0.68 0.16 0.75 0.16R-squared 0.01 0.40 0.40 0.40 0.56 0.52 0.39 0.08 0.38Number of cities 99 99 99 99 99 99 75 99 75Number of subway cities 99 99 99 99 99 99 75 99 75Number of periods 9 9 9 9 1 3 9 9 9p-value 2nd order autocorr 0.00 0.00 0.00 0.00 0.00 0.00 0.00p-value 3rd order autocorr 0.00 0.00 0.00 0.00 0.05 0.01 0.10F-stat excluded instrument 153.49 12.88Observations 891 828 828 828 99 297 347 828 347

Dependent variable: Change in log population of metropolitan area in a 5 year period. Sample is large subway cities (over 1 million in 1970).

City-level clustered standard errors in parentheses. Stars denote significance levels: * 0.10, ** 0.05, *** 0.01. All regressions t >=1970.

(1)- No controls. (2)-Add year dummies and change in log gdp controls.

(3)- Use change in log route km as regressor. (4)-Use change in log subway lines as regressor.

(5)- Long difference regression 1970-2010. (6)-Dep. var. is change in log mean radiance calibrated lights in a 25km circle around the centroid of the city.

(7)- Add log(st−4 + 1) > 0 restriction. (8) Instrument ∆ ln (st + 1) with log(st−4 + 1). (9) Add log(st−4 + 1) > 0 restriction.

System age and growth

System age and growth

System growth and (20 year) lagged size

Note: Restricted to L4lop>0.

First differences

(1) (2) (3) (4) (5) (6) (7) (8) (9)∆ ln (popt) ∆ ln (popt) ∆ ln (popt) ∆ ln (popt) ∆ ln (pop70−10) ∆ ln(Lightst) ∆ ln (popt) ∆ ln (popt) ∆ ln (popt)

∆ ln(st) 0.008∗∗ 0.007∗∗ 0.027 0.056∗∗ 0.074∗∗∗ 0.023(0.004) (0.003) (0.022) (0.023) (0.019) (0.145)

∆ ln(route kmt) 0.007∗∗

(0.003)

∆ ln(subway linest) 0.019∗∗

(0.009)

∆ln(s70−10) 0.095∗∗∗

(0.025)

∆ ln(GDPpct) 0.165∗∗∗ 0.164∗∗∗ 0.164∗∗∗ 0.806∗∗ 0.715∗∗∗ 0.098∗∗ 0.155∗∗∗ 0.107∗∗

(0.035) (0.035) (0.035) (0.330) (0.128) (0.029) (0.033) (0.045)

YearXContinent dummies No Yes Yes Yes No Yes Yes Yes Yes

Geographic controls No No No No Yes No No No No

Continent dummies No Yes Yes Yes Yes Yes Yes Yes Yes

Mean of Dep Variable 0.077 0.078 0.078 0.078 0.557 0.017 0.040 0.078 0.040Mean of subways regressor 0.31 0.30 0.32 0.12 0.11 0.28 0.09 0.30 0.09SD subways regressor 0.74 0.75 0.80 0.29 0.21 0.68 0.16 0.75 0.16R-squared 0.01 0.40 0.40 0.40 0.56 0.52 0.39 0.08 0.38Number of cities 99 99 99 99 99 99 75 99 75Number of subway cities 99 99 99 99 99 99 75 99 75Number of periods 9 9 9 9 1 3 9 9 9p-value 2nd order autocorr 0.00 0.00 0.00 0.00 0.00 0.00 0.00p-value 3rd order autocorr 0.00 0.00 0.00 0.00 0.05 0.01 0.10F-stat excluded instrument 153.49 12.88Observations 891 828 828 828 99 297 347 828 347

Dependent variable: Change in log population of metropolitan area in a 5 year period. Sample is large subway cities (over 1 million in 1970).

City-level clustered standard errors in parentheses. Stars denote significance levels: * 0.10, ** 0.05, *** 0.01. All regressions t >=1970.

(1)- No controls. (2)-Add year dummies and change in log gdp controls.

(3)- Use change in log route km as regressor. (4)-Use change in log subway lines as regressor.

(5)- Long difference regression 1970-2010. (6)-Dep. var. is change in log mean radiance calibrated lights in a 25km circle around the centroid of the city.

(7)- Add log(st−4 + 1) > 0 restriction. (8) Instrument ∆ ln (st + 1) with log(st−4 + 1). (9) Add log(st−4 + 1) > 0 restriction.

Second differences

(1) (2) (3) (4)∆2 ln (popt) ∆2 ln (popt) ∆2 ln (popt) ∆2 ln(Lightst)

∆2 ln(st) -0.0021 -0.0014 -0.0077 0.0382(0.0014) (0.0014) (0.0093) (0.0250)

∆2 ln(GDPpct) 0.0605∗∗∗ 0.0406∗∗ 1.0103∗∗

(0.0161) (0.0197) (0.3448)

YearXContinent dummies No Yes Yes YesMean of Dep Variable -0.009 -0.008 -0.003 -0.167Mean of subways regressor 0.02 0.02 -0.01 -0.03SD subways regressor 1.04 1.04 0.21 0.92R-squared 0.002 0.184 0.314 0.407Number of cities 99 99 75 99Number of subway cities 99 99 75 99Number of periods 9 9 9 2p-value 3rd order autocorr 0.81 0.41 0.48Observations 891 812 338 198

Dependent variable: 2nd difference in log population of metropolitan area in a 5 year period.

Sample is large subway cities (over 1 million in 1970). All regressions t >= 1970.

City-level clustered standard errors in parentheses.

Stars denote significance levels: * 0.10, ** 0.05, *** 0.01.

(1)-No controls. (2)-Add year dummies and change in log gdp controls.

(3)-Add log(st−4) > 0 restriction.

(4)-Dep. var. is second change in log mean radiance calibrated lights in a 25km circle around the centroid of

the city.

Distributed lag

(1) (2) (3) (4) (5) (6) (7)∆ ln(popt) ∆ ln(popt) ∆ ln(popt) ∆ ln(popt) ∆ ln(popt) ∆ ln(popt) ∆ ln(popt)

∆ ln(st+3) 0.013∗∗

(0.005)

∆ ln(st+2) 0.012∗∗

(0.005)

∆ ln(st+1) 0.007(0.005)

∆ ln(st) 0.007∗∗ 0.007∗∗ 0.007∗

(0.003) (0.003) (0.004)

∆ ln(st−1) 0.004 0.004 0.005(0.003) (0.003) (0.004)

∆ ln(st−2) -0.000 0.000 -0.000 0.004(0.003) (0.003) (0.003) (0.004)

∆ ln(st−3) -0.003 -0.002 -0.003 0.003(0.003) (0.003) (0.003) (0.004)

∆ ln(GDPpct) 0.165∗∗∗ 0.165∗∗∗ 0.166∗∗∗ 0.166∗∗∗ 0.164∗∗∗ 0.165∗∗∗ 0.129∗∗∗

(0.035) (0.036) (0.036) (0.036) (0.035) (0.036) (0.031)

YearXContinent dummies Yes Yes Yes Yes Yes Yes YesMean of Dep Variable 0.08 0.08 0.08 0.08 0.08 0.08 0.09Number of cities 99 99 99 99 99 99 96Number of subway cities 99 99 99 99 99 99 96Number of periods 9 9 9 9 9 9 6Observations 828 828 828 828 828 828 531

Dependent variable: Change in log population in a 5 year period.

Sample is large cities (over 1 million in 1970) with subway in 2010. City-level clustered standard errors in parentheses.

Stars denote significance levels: * 0.10, ** 0.05, *** 0.01. All regressions t >= 1970.

Decentralization - Boston

Note: 5km and 25km circles.

Decentralization

Mean light ratio Log mean light

(1) (2) (3) (4) (5) (6)lights5km/lights25km ∆(lights5km/lights25km) log(lights[0,5]km) log(lights[5,25]km) ∆log(lights[0,5]km) ∆log(lights[5,25]km)

ln(st) -0.38∗∗∗ 0.12∗∗∗ 0.24∗∗∗

(0.091) (0.031) (0.040)

∆ ln(st) -0.26∗∗ -0.020 0.040∗

(0.089) (0.026) (0.023)

ln(GDPpct) 0.32 0.20∗∗ 0.11(0.38) (0.084) (0.10)

∆ ln(GDPpct) 0.41 0.77∗∗∗ 0.69∗∗∗

(0.29) (0.17) (0.12)

YearXContinent dummies Yes Yes Yes Yes Yes YesGeographic controls Yes No Yes Yes No NoMean of Dep Variable 4.198 -0.293 6.139 4.654 -0.053 0.032Mean of subways regressor 3.43 0.28 3.43 3.43 0.28 0.28SD subways regressor 1.31 0.68 1.31 1.31 0.68 0.68R-squared 0.34 0.30 0.41 0.50 0.52 0.50Number of cities 99 99 99 99 99 99Number of subway cities 99 99 99 99 99 99Number of periods 4 3 4 4 3 3Observations 396 297 396 396 297 297

Lights refer to mean of radiance calibrated lights in rings or circles of different diameters.

City-level robust standard errors in parentheses. Stars denote significance levels: * 0.10, ** 0.05, *** 0.01.

Geographic controls are continent dummy, capital city dummy, log km to ocean, log km to land border, and log km to major navigable river.

Summary

I B ∼ 0.2 in cross-section. Same for route km and count oflines. Big effects on lights, too.

I B < 0.01 in first-differences. Bigger in bigger cities? Zero forlights.

I B ∼ 0 in FD-IV using fourth lag of subways to predictsubways.

I B = 0 in second differences.

That is, regressions confirm what we saw in the figures.

Conclusion

I First ever city level analysis of subways and growth

I Population elasticity of subways < 0.01

I Confirmed with lights at night data

I Subsidies based on agglomeration effects should be tiny

I Subways decentralize central cities