Lecture 6: Transport Costs and Congestion Forceserossi/Urban/Lecture_6_538.pdf · Baum-Snow (2007)...
Transcript of Lecture 6: Transport Costs and Congestion Forceserossi/Urban/Lecture_6_538.pdf · Baum-Snow (2007)...
Lecture 6: Transport Costs and Congestion ForcesWWS 538
Esteban Rossi-Hansberg
Princeton University
ERH (Princeton University ) Lecture 6: Transport Costs and Congestion Forces 1 / 43
Baum-Snow (2007)
Between 1950 and 1990 the population of central cities declined by 17%despite population growth of 72% in MSAs
Why?I Paper argues that construction of limited access highways is importantI Lower commuting cost lead to suburbanization as demand for space increasesI Explains about one third of total suburbanizationI Central city population would have grown by 8% without the interstatehighway system
Other explanations?I Amenity value of suburbsI Tiebout sorting and racial preferencesI Crime and blightI Desegregation of central city schools
ERH (Princeton University ) Lecture 6: Transport Costs and Congestion Forces 2 / 43
Empirical Strategy
Highway system designed to connect far away places not for local commuting
So number of highway links varies independently from city characteristics
Still, endogeneity is a possibilityI Instrument using 1947 highway planI The legislation stipulated that highways in the planned system should be “. . .so located as to connect by routes as direct as practicable, the principalmetropolitan areas, cities, and industrial centers, to serve the national defense,and to connect at suitable border points with routes of continental importancein the Dominion of Canada and the Republic of Mexico.”
I Growth in rays between 1950 and 1990 is correlated with urban populationgrowth in 1940-1950, but planned rays are not correlated
I Planned rays are correlated with 1940 population, so essential to control forpopulation
Count rays emanating from central cities
ERH (Princeton University ) Lecture 6: Transport Costs and Congestion Forces 3 / 43
Aggregate Trends in Suburbanization
determining the provision of local public goods. The fact thatconstant geography central city population fell significantlyfaster than central city population within political boundariesimplies that space is an important factor needed to explain fallingurban population density. The fact that large inland MSAs sawsharper declines in central city population than all large MSAsdespite their faster total MSA population growth further sup-ports this view. Tiebout sorting models imply that people movingfurther from the core would endeavor to form new communitiesthat provide different levels of public services, rather than relo-cate within the same political jurisdiction. The aggregate datathus show that there is ample opportunity for a spatial mecha-nism to be an important driver of urban population decentraliza-tion, though it may partially interact with a Tiebout sortingmechanism.2
2. Mieszkowski and Mills [1993] make a similar point, noting that populationdispersal is not just a post-WWII phenomenon in the United States and thatsuburbanization has been occurring around the world independent of the geogra-phy of local political jurisdictions. Nevertheless, there are clearly some publicgoods, like crime, that are more neighborhood-based than city-based. This argu-ment is only relevant for services provided at the city level.
TABLE IAGGREGATE TRENDS IN SUBURBANIZATION, 1950–1990
1950 1960 1970 1980 1990
Percentchange
1950–1990
Panel A: Large MSAsMSA population 92.9 115.8 134.0 144.8 159.8 72Total CC population 44.7 48.5 51.3 49.2 51.0 14Constant geography CC population 44.7 44.2 42.6 37.9 37.1 �17N for constant geog. CC population 139 132 139 139 139
Panel B: Large Inland MSAsMSA population 39.2 48.9 57.0 65.0 73.5 88Total CC population 16.8 19.7 22.1 22.1 23.2 38Constant geography CC population 16.8 16.5 15.4 13.3 12.5 �26N for constant geog. CC population 100 94 100 100 100
Total U. S. population 150.7 178.5 202.1 225.2 247.1 64
Notes: All populations are in millions. CC stands for central city. The sample includes all metropolitanareas (MSAs) of at least 100,000 people with central cities of at least 50,000 people in 1950. The sample inPanel B excludes MSAs with central cities located within 20 miles of a coast, major lake shore, or interna-tional border. MSA populations are for geography as of year 2000. Constant geography central city populationuses 1950 central city geography. Census tract data are not available to build constant geography central citypopulations for some small cities in 1960. These cities are assigned a population of 0 for constructing theaggregates. Reported total U. S. population excludes Alaska and Hawaii.
777DID HIGHWAYS CAUSE SUBURBANIZATION?
at Princeton University on M
arch 14, 2012http://qje.oxfordjournals.org/
Dow
nloaded from
ERH (Princeton University ) Lecture 6: Transport Costs and Congestion Forces 4 / 43
Counting Rays
FIGURE IThe Projected System of Interstate Highways in 1947
780Q
UA
RT
ER
LY
JO
UR
NA
LO
FE
CO
NO
MIC
S
at Princeton University on March 14, 2012 http://qje.oxfordjournals.org/ Downloaded from
ERH (Princeton University ) Lecture 6: Transport Costs and Congestion Forces 5 / 43
First Stage
importance of holding population levels constant in an evaluationof the effect of rays on suburbanization.
First stage results reported in Table II show that the numberof planned rays is a strong predictor of actual highway construc-
TABLE IIFIRST STAGE RESULTS
LARGE MSAS IN 1950
Panel A: Long difference 1950–1990
Change in Rays
1 2 3
Rays in the plan .677 .526 .510(.074)** (.076)** (.074)**
1950 Central city radius .325 .306(.071)** (.072)**
Change in simulated log income �.939(1.819)
Change in log of MSA population.856
(.279)**Constant .866 .218 .463
(.218)** (.247) (1.231)
Observations 139 139 139R-squared .38 .47 .50
Panel B: Full panel 1950–1990
Rays Smoothed Rays
1 2 4 5
Smoothed rays in plan .826 .448 .833 .712(.029)** (.040)** (.023)** (.030)**
Log simulated income 1.563 .198(.234)** (.179)
Log MSA population .170 .591(.200) (.154)**
MSA Fixed effects Yes Yes Yes Yes
Groups 132 132 132 132R-squared .55 .84 .67 .88
Notes: Panel A shows the first stage results for the regressions in Table IV. Panel B shows the first stageresults for the regressions in Table VI. Listed specification numbers match those in the corresponding secondstage tables. “Smoothed rays” is calculated by multiplying the stock of rays in 1990 in each MSA by thefraction of these rays’ mileage that is completed at each point in time. “Smoothed rays in the plan” iscalculated by multiplying the number of rays in the 1947 plan by the fraction of federally funded highwaymileage in the 1956 Federal Aid Highway Act completed at each point in time. All coefficients remainsignificant when estimated using the more selected sample in Table I Panel B.
783DID HIGHWAYS CAUSE SUBURBANIZATION?
at Princeton University on M
arch 14, 2012http://qje.oxfordjournals.org/
Dow
nloaded from
ERH (Princeton University ) Lecture 6: Transport Costs and Congestion Forces 6 / 43
The Effect of Lower Commuting Costs
Basic monocentric city model implies that lower commuting costs result inlower density at the center
I Careful, more general theory does not
Basic conclusion from the standard model:I population in metropolitan areas should spread out along new highwaysI central city population should increase with metropolitan area population andthe radius of the central city
I central city population should decline with the number of highway rays
ERH (Princeton University ) Lecture 6: Transport Costs and Congestion Forces 7 / 43
Austin, TX
FIGURE IIDevelopment Patterns in Austin, TX.
786Q
UA
RT
ER
LY
JO
UR
NA
LO
FE
CO
NO
MIC
S
at Princeton University on March 14, 2012 http://qje.oxfordjournals.org/ Downloaded from
ERH (Princeton University ) Lecture 6: Transport Costs and Congestion Forces 8 / 43
Population Density and Distance to Center
estimated gradient is much greater in the sample, including morecentrally located MSAs, likely because the less restricted space inthese MSAs generates equilibria with less population restrictedto live near the highways. Panel B shows that conditional on CBDand 1970 highway distance, portions of MSAs near highwaysbuilt between 1970 and 1990 had faster population growth thanother areas. Population density decreased by about 1 percent for
TABLE IIITHE SPATIAL DISTRIBUTION OF METROPOLITAN AREA POPULATIONS
Panel A: 1970 and 1990 Cross-Sections
Sample
Log populationdensity
1970 1990
Large MSAs in 1950 (36,250tracts, 139 MSAs)
Distance to CBD �.132 �.114(.001)** (.001)**
Distance to highway �.014 �.019(.002)** (.002)**
Large MSAs in 1950 withcentral cities at least 20 milesfrom a coast or border (17,336tracts, 100 MSAs)
Distance to CBD �.134 �.117(.002)** (.001)**
Distance to highway �.055 �.054(.003)** (.003)**
Panel B: Evolution between 1970 and 1990
Sample�Log population
density
Large MSAs in 1950 (36,250tracts, 139 MSAs)
Distance to CBD .021(.000)**
�Distance to highway �.015(.002)**
Large MSAs in 1950 withcentral cities at least 20miles from a coast or border(17,336 tracts, 100 MSAs)
Distance to CBD .021(.001)**
�Distance to highway �.008(.003)**
Notes: Each pair of entries lists coefficients and standard errors from a regression of log populationdensity on the listed variables at the census tract level. All regressions include MSA fixed effects. Regressionsin Panel B also include the distance to the nearest highway in 1970. Estimated coefficients on distance to thenearest highway in 1970 are between �0.002 and 0.004. Regressions using the distance to planned highwaysas an instrument for the distance to observed highways yield similar results. When standard errors areclustered by MSA, results for the larger sample in Panel B and results for the smaller sample in Panel Aremain significant at the 5 percent level. Other results are not statistically significant with clustering.Regressions are weighted by the fraction of MSA population that is represented in the tract. Analogousunweighted regressions produce highway distance coefficients that are larger in absolute value. All distancesare in miles.
788 QUARTERLY JOURNAL OF ECONOMICS
at Princeton University on M
arch 14, 2012http://qje.oxfordjournals.org/
Dow
nloaded from
ERH (Princeton University ) Lecture 6: Transport Costs and Congestion Forces 9 / 43
Main Specification
The main specification used is
∆ logNci = δ0 + δ1∆rayi + δ2rci + δ3∆w̃i + δ4∆ logNMSAi + δ5∆Gi + εi
where Nc represents 1950-definition central city population, rc denotescentral city radius, NMSA denotes MSA population, w̃i represents mean logannual personal income, and Gi represents the Gini coeffi cient of the incomedistribution for MSA i
ERH (Princeton University ) Lecture 6: Transport Costs and Congestion Forces 10 / 43
Main Results: Long Differences
suburbanization. Coefficients on rays in all specifications havethe expected negative sign and are usually statistically signifi-cant and sizable. OLS results indicate that conditional on controlvariables, each additional ray is associated with about a 6 percentdecline in central city population. In this primary specification,instrumental variables estimates imply about a 12 percent de-cline in central city population for each additional ray. The onlycontrol variable that significantly influences the coefficient onrays is the central city radius. A positive correlation between thecentral city radius and the number of rays, planned and actual,accounts for the absolute increase in the coefficient on rays whenthe radius enters as a control.10 Using central city area as acontrol instead produces similar results.
10. A land use model would imply that the appropriate functional form mayinclude an interaction between rays and central city radius. The resulting coeffi-
TABLE IVLONG-DIFFERENCE REGRESSIONS OF THE DETERMINANTS OF CONSTANT GEOGRAPHY
CENTRAL CITY POPULATION GROWTH, 1950–1990
Large MSAs in 1950
OLS3
Change in log population in constant geographycentral cities
IV1 IV2 IV3 IV4 IV5
Change in number ofrays
�.059 �.030 �.106 �.123 �.114 �.101(.014)** (.022) (.032)** (.029)** (.026)** (.046)*
1950 central city radius .080 .111 .113 .106 .125(.014)** (.023)** (.023)** (.023)** (.021)**
Change in simulated logincome
.084 .048 �6.247 �.137(.378) (.417) (6.174) (.480)
Change in log of MSApopulation
.363 .424 .374 .405(.082)** (.094)** (.079)** (.108)**
Change in Gini coeff ofsimulated income
�23.416(23.266)
Log 1950 MSApopulation
�.062(.062)
Constant �.640 �.203 �.359 �.588 4.580 �.611(.260)* (.078)* (.076)** (.281)* (5.091) (.265)*
Observations 139 139 139 139 139 139R-squared .39 .00 .01 .30 .33 .37
Notes: In columns IV1–IV5, the number of rays in the 1947 plan instruments for the change in thenumber of rays. Standard errors are clustered by state of the MSA central city. Standard errors are inparentheses. ** indicates significant at the 1 percent level, * indicates significant at 5 percent level. Summarystatistics are in the Appendix Table. First stage results are in Table II.
791DID HIGHWAYS CAUSE SUBURBANIZATION?
at Princeton University on M
arch 14, 2012http://qje.oxfordjournals.org/
Dow
nloaded from
ERH (Princeton University ) Lecture 6: Transport Costs and Congestion Forces 11 / 43
Main Results: City Fixed Effects
in the long-difference regressions because 1950-definition centralcity population is not available for all cities in all years.
Panel OLS regressions using discrete rays as an explanatoryvariable produce coefficients of less than �0.02. These smallcoefficients reflect adjustment costs and measurement problems.To get at estimating a long-run effect, I also estimate the panelregressions using a measure of smoothed rays. I define smoothedrays as the stock of rays in 1990 multiplied by the fraction of therays’ mileage completed in each MSA at each point in time.Smoothing the rays increases the coefficients in Specification 3 to�0.04, about 0.02 smaller in magnitude than the OLS coefficientsin Tables IV and V.
Table VI presents instrumental variables results with MSAfixed effects. Specifications 1 to 3 use discrete rays as the explan-atory variable while Specifications 4–6 use smoothed rays. Thepanel IV results are consistent with results from the long differenceswhether endogenous rays are smoothed or not. The coefficient on
TABLE VIPANEL IV REGRESSIONS OF THE DETERMINANTS OF CONSTANT GEOGRAPHY CENTRAL
CITY POPULATION, 1950–1990
Large MSAs in 1950
Log central city population
1 2 3 4 5 6
Number of rays �0.111 �0.142 �0.140(0.016)** (0.026)** (0.028)**
(1990 Rays) �(Fraction of Ray
�0.097 �0.089 �0.086(0.016)** (0.012)** (0.013)**
miles completed at t)Log simulated income �0.083 �0.061 �0.288 �0.229
(0.117) (0.109) (0.075)** (0.077)**Log MSA population 0.266 0.263 0.294 0.286
(0.104)* (0.105)* (0.100)** (0.098)**Simulated Gini
coefficient�0.623 �1.415(1.106) (0.847)
MSA Fixed Effects Yes Yes Yes Yes Yes Yes
R-Squared 0.20 0.22 0.22 0.14 0.56 0.57
Notes: The instrument used is (rays in the plan) � (MSA mileage of highways running through thecentral city at time t)/(MSA mileage of highways running through the central city in 1990). Standard errorsare clustered by the state of the central city. Standard errors are in parentheses. ** indicates significant atthe 1 percent level, * indicates significant at the 5 percent level. First stage results are in Table II. Eachregression includes 132 MSAs with five observations each, one for each year 1950–1990. There are fewerMSAs in this sample than that in Table IV because of lack of census tract data for seven MSAs in 1960.
797DID HIGHWAYS CAUSE SUBURBANIZATION?
at Princeton University on M
arch 14, 2012http://qje.oxfordjournals.org/
Dow
nloaded from
ERH (Princeton University ) Lecture 6: Transport Costs and Congestion Forces 12 / 43
The Fundamental Law of Road Congestion, Duranton andTurner (2011)
What is the effect of building more roads on the vehicle-km traveled (VKT)?
Fundamental Law of Highway Congestion:I VKT increases one to one with highwaysI So building more roads does not relieve congestion
Costs of congestion and transportation are largeI in 2001 an average American household spent 161 person-minutes per day in apassenger vehicle
I These minutes allowed 134 person-kilometer of auto travel at an averagespeed of 44 km/h
Cost of transport infrastructure are largeI Current policy based on the idea that infrastructure relieves congestion
Linked to environmental problem of carbon emissions
ERH (Princeton University ) Lecture 6: Transport Costs and Congestion Forces 13 / 43
Conceptual Framework
2619dURAnTOn And TURnER: ThE fUndAmEnTAL LAW Of ROAd cOngESTIOnVOL. 101 nO. 6
That is, willingness to pay equals average cost.Increasing the supply of road lane kilometers from R to R′ reduces the average
cost of driving for any level of VKT.4 It thus shifts the average cost curve to the right. With R lane kilometers of roads in the city, the demand curve intersects with the supply curve at Q * , the equilibrium VKT. With R′ lane kilometers of road, the corresponding equilibrium implies a VKT of Q′ * .
We would like to learn the effect of an increase in the stock of roads on driving in cities. That is, we would like to learn about the function Q * (R) defined implicitly by equation (1). Indexing cities by i and years by t, our problem may be stated as one of estimating,
(2) ln( Q it ) = A 0 + ρ R Q ln ( R it ) + A 1 X it + ϵ it ,
where X denotes a vector of observed city characteristics and ϵ describes unobserved contributors to driving. We are interested in the coefficient of R, the road elasticity of VKT, ρ R Q ≡ ∂ ln Q/∂ ln R.
With data describing driving and the stock of roads in a set of cities, we can estimate equation (2) with OLS to obtain consistent estimates of ρ R Q , provided that cov (R, ϵ | X) = 0. In practice, we hope that roads will be assigned to growing cit-ies and fear that they are assigned to prop-up declining cities. In either case, the required orthogonality condition fails. Thus, we are concerned that estimating equa-tion (2) will not lead to the true value of ρ R Q .
4 There are pathological examples where increases in the extent of a road network can reduce its capacity, in particular the “Braess paradox” described in Small and Verhoef (2007). We ignore such pathological examples here.
Figure 1. Supply and Demand for Road Traffic
0
P
VKTQ* Q*′
P(Q)
AC(R)
AC(R′)
ERH (Princeton University ) Lecture 6: Transport Costs and Congestion Forces 14 / 43
Empirical Framework
Goal is to learn about Q∗ (R)
Or estimate ρQR in
ln (Qit ) = A0 + ρQR ln (Rit ) + A1Xit + εit
where Xit denotes city characteristics and εit unobserved contributors todriving
If εit = δi + ηιt then using fixed effects one can estimate
ln (Qit ) = A0 + ρQR ln (Rit ) + A1Xit + δi + ηιt
Or in first differences (∆)
∆ ln (Qit ) = ρQR ∆ ln (Rit ) + A1∆Xit + ∆ηιt
ERH (Princeton University ) Lecture 6: Transport Costs and Congestion Forces 15 / 43
Empirical Framework
Problem is still that roads could be assigned to cities in response to acontemporaneous shock to the city’s traffi c
So use
ln (Rit ) = B0 + B1Xit + B2Zit + µit
ln (Qit ) = A0 + ρQR̂ln (Rit ) + A1Xit + δi + ηιt
where the instrument Z satisfies cov (Z ,R) 6= 0 and cov(Z , ε) = 0.
ERH (Princeton University ) Lecture 6: Transport Costs and Congestion Forces 16 / 43
Data2623dURAnTOn And TURnER: ThE fUndAmEnTAL LAW Of ROAd cOngESTIOnVOL. 101 nO. 6
Table 1—Summary Statistics for Our Main HPMS and Public Transportation Variables
Year: 1983 1993 2003
Mean daily VKT (IH, ’000 km) 7,777 11,905 15,961(16,624) (24,251) (31,579)
Mean AADT (IH) 4,832 7,174 9,361(2,726) (3,413) (4,092)
Mean lane km (IH) 1,140 1,208 1,280(1,650) (1,729) (1,858)
Mean lane km (IH, per 10,000 population) 26.7 24.3 22.1(26.9) (20.9) (16.4)
Mean daily VKT (MRU, ’000 km) 14,553 22,450 31,242(36,303) (49,132) (70,692)
Mean AADT (MRU) 3,146 3,646 3,934(847) (947) (1,059)
Mean lane km (MRU) 3,885 5,071 6,471(7,926) (9,119) (12,426)
Mean VKT share urbanized (IHU/IH) 0.38 0.44 0.48Mean lane km share urbanized (IHU/IH) 0.29 0.36 0.40 Mean share truck AADT (IH) 0.11 0.12 0.13 Peak service large buses per 10,000 population 1.20 1.09 1.34
(1.02) (0.98) (0.98) Peak service large buses 169 165 217
(563) (562) (742) Number MSAs 228 228 228Mean MSA population 753,726 834,290 950,054
notes: Cross MSA means and standard deviations in parentheses. IH denotes interstate high-ways for the entire MSA. IHU denotes interstate highways for the urbanized areas within an MSA. MRU denotes major roads for the urbanized areas within an MSA.
Table 2—VKT as a Function of Lane Kilometers, Univariate OLS by Decade
1983 1993 2003Year: (1) (2) (3)Panel A. dep. var.: ln VKT for interstate highways, entire mSAs
ln (IH lane km) 1.24*** 1.25*** 1.23***(0.04) (0.02) (0.02)
R 2 0.86 0.87 0.88
Panel B. dep. var.: ln VKT for interstate highways, urbanized areas within mSAs
ln (IHU lane km) 1.26*** 1.23*** 1.20***(0.02) (0.02) (0.02)
Panel c. dep. var.: ln VKT for major roads, urbanized areas within mSAs
ln (MRU lane km) 1.08*** 1.13*** 1.14***(0.02) (0.01) (0.01)
Panel d. dep. var.: ln VKT for interstate highways, outside urbanized areas within mSAs
ln (IHNU lane km) 1.06*** 1.03*** 1.00***(0.03) (0.04) (0.04)
notes: The same regressions for different types of roads are performed in all four panels. All regressions include a constant. Robust standard errors in parentheses; 228 observations for each regression in panel A and 192 in panels B–D.
*** Significant at the 1 percent level. ** Significant at the 5 percent level. * Significant at the 10 percent level.
ERH (Princeton University ) Lecture 6: Transport Costs and Congestion Forces 17 / 43
Basic OLS Results2624 ThE AmERIcAn EcOnOmIc REVIEW OcTOBER 2011
panel C, and nonurban MSA interstates in panel D. Columns 1 to 3 consider the 1983, 1993, and 2003 cross sections in turn.
Depending on the decade, the elasticity of MSA interstate highway VKT with respect to lane kilometers is between 1.23 and 1.25. Focusing only on interstate highways in the urbanized part of MSAs yields similar results. For major urban roads and nonurban MSA interstates, we obtain slightly lower estimates between 1.00 and 1.14.
In Table 3, we consider richer specifications. In panel A of this table, the dependent variable is again MSA interstate VKT. Columns 1 to 3 consider the 1983 cross sec-tion. In the first column we include our variable of interest, the log of lane kilometers of road, MSA population, and a constant. In the second we add nine census division dummy variables along with five measures of physical geography: elevation range within the MSA, the ruggedness of terrain in the MSA, two measures of climate, and a measure of how dispersed is development in the MSA. Details about these variables
Table 3—Vkt as a Function of Lane Kilometers, Ols by Decade
1983 1983 1983 1993 1993 1993 2003 2003 2003Year: (1) (2) (3) (4) (5) (6) (7) (8) (9)
Panel A. dependent variable: ln VKT for interstate highways, entire mSAs ln (IH lane km) 0.92*** 0.94*** 0.92*** 0.73*** 0.76*** 0.77*** 0.71*** 0.75*** 0.76***
(0.06) (0.06) (0.05) (0.05) (0.04) (0.04) (0.05) (0.04) (0.04)ln (population) 0.43*** 0.42*** 1.01*** 0.54*** 0.51*** 0.46* 0.53*** 0.49*** 0.39
(0.04) (0.05) (0.37) (0.04) (0.04) (0.25) (0.04) (0.04) (0.35)Elevation range −0.057 −0.076 −0.027 −0.038 −0.026 −0.030
(0.060) (0.054) (0.056) (0.054) (0.053) (0.048)Ruggedness 6.81* 5.29 5.86* 3.90 5.72* 3.46
(3.46) (3.24) (3.00) (3.00) (3.06) (3.11)Heating degree days −0.014*** −0.015*** −0.012*** −0.013*** −0.011*** −0.013***
(0.004) (0.01) (0.003) (0.004) (0.003) (0.004)Cooling degree days −0.019* −0.027** −0.019*** −0.022** −0.019** −0.020**
(0.010) (0.012) (0.007) (0.009) (0.007) (0.009)Sprawl 0.0059* 0.0061* 0.0033 0.0019 0.0021 0.0016
(0.0031) (0.0036) (0.0028) (0.0029) (0.0027) (0.0027)Census divisions Y Y Y Y Y YPast populations Y Y YSocioeconomic characteristics
Y Y Y
R 2 0.93 0.94 0.95 0.94 0.95 0.96 0.94 0.96 0.96
Panel B. dependent variable: ln VKT for interstate highways, urbanized areas within mSAs
ln (IHU lane km) 1.04*** 1.05*** 1.06*** 0.95*** 0.97*** 1.00*** 0.92*** 0.94*** 0.97***(0.03) (0.03) (0.03) (0.03) (0.03) (0.04) (0.03) (0.03) (0.04)
Panel c. dependent variable: ln VKT for major roads, urbanized areas within mSAs
ln (MRU lane km) 0.90*** 0.89*** 0.88*** 0.72*** 0.78*** 0.80*** 0.66*** 0.67*** 0.70***(0.03) (0.03) (0.03) (0.04) (0.04) (0.04) (0.04) (0.04) (0.04)
Panel d. dependent variable: ln VKT for interstate highways, outside urbanized areas within mSAs
ln (IHNU lane km) 0.83*** 0.85*** 0.84*** 0.81*** 0.83*** 0.82*** 0.82*** 0.84*** 0.83***(0.05) (0.04) (0.03) (0.04) (0.03) (0.03) (0.03) (0.03) (0.03)
notes: The same regressions for different types of roads are performed in all four panels. All regressions include a constant. Robust standard errors in parentheses; 228 observations for each regression in panel A and 192 in pan-els B–D.
*** Significant at the 1 percent level. ** Significant at the 5 percent level. * Significant at the 10 percent level.
ERH (Princeton University ) Lecture 6: Transport Costs and Congestion Forces 18 / 43
Fixed Effects Pooled
2626 ThE AmERIcAn EcOnOmIc REVIEW OcTOBER 2011
Columns 5–10 of Table 4 estimate equation (3) by including MSA fixed effects in our cross-sectional regression. Because they condition out permanent determinants of VKT for each city that are potentially correlated with roadway, we prefer the specifications with MSA fixed effects to those without. In column 5 we replicate column 1 of the same table but include MSA fixed effects. In column 6, we augment the specification of column 2 with MSA fixed effects. In column 7, we repeat this for column 4. In column 8 we replicate column 6 using only the 192 MSAs that have urban interstate highways in all years instead of the 228 MSAs that report interstate highways in all three of our sample years. Columns 9 and 10 run the same regression again on MSAs with below- and above-median 1990 population size, respectively. All the fixed-effect estimates of the interstate VKT elasticity of interstate lane kilo-meters are slightly above one, except for column 8 where the estimate is slightly below one. This is obtained for the more restricted sample of MSAs with interstate highways in their urbanized area. Given the similarity between the results, however, we do not concern ourselves further with sample selection. While it is estimated precisely in all specifications, ρ R Q is not statistically different from one at standard levels of confidence in columns 5 through 10. Overall, we note that including MSA fixed effects leads to slightly higher estimates of ρ R Q .
We now estimate the interstate VKT elasticity of interstate lane kilometers using our first difference estimating equation (4). Unlike the fixed-effects estimations of Table 4, in the first difference regressions of Table 5, we allow the levels of MSA ini-tial characteristics to affect the growth of traffic. Using our three cross sections we compute two cross sections of first differences. In panel A of Table 5 we pool these two cross sections of first differences to estimate equation (4). Our dependent vari-able is the ten-year change in interstate VKT. In column 1, we include only a constant and year dummies as controls. In column 2, we add changes in MSA population.
Table 4—VKT as a Function of Lane Kilometers, Pooled OLS
All All All All All All All w. IHU Big Small MSA sample (1) (2) (3) (4) (5) (6) (7) (8) (9) (10)dependent variable: ln VKT for interstate highways, entire mSAs
ln (IH lane km) 1.24*** 0.82*** 0.86*** 0.85*** 1.05*** 1.06*** 1.05*** 0.95*** 1.05*** 1.12***(0.02) (0.05) (0.05) (0.04) (0.04) (0.04) (0.04) (0.03) (0.04) (0.08)
ln (population) 0.48*** 0.44*** 0.32*** 0.34*** 0.39*** 0.32*** 0.44*** 0.31**(0.04) (0.04) (0.12) (0.09) (0.09) (0.09) (0.11) (0.12)
Geography Y Y Census divisions Y Y Socioeconomic characteristics
Y Y
Past populations Y MSA fixed effects Y Y Y Y Y Y
R 2 0.88 0.94 0.95 0.96 0.94 0.94 0.95 0.94 0.96 0.93
notes: All regressions include year effects. Robust standard errors in parentheses (clustered by MSA in columns 1–4). Complete sample of 228 MSAs (684 observations) with interstate highways in columns 1–7; 192 MSAs (576 observations) with urban interstate highways in column 8; 114 MSAs (342 observations) above the median popu-lation size in 1990 in column 9; 114 MSAs (342 observations) below the median population size in 1990 in col-umn 10.
*** Significant at the 1 percent level. ** Significant at the 5 percent level. * Significant at the 10 percent level.
ERH (Princeton University ) Lecture 6: Transport Costs and Congestion Forces 19 / 43
First Differences2627dURAnTOn And TURnER: ThE fUndAmEnTAL LAW Of ROAd cOngESTIOnVOL. 101 nO. 6
In column 3, we also control for initial VKT. In column 4, we add physical geog-raphy and census division dummies. Column 5 adds decennial MSA population levels from 1920 to 1980 and initial socioeconomic characteristics of cities. In each case, our point estimate of ρ R Q is very close to one and is precisely estimated.
Columns 6–8 consider more restricted samples of observations. Column 6 repli-cates column 2 using only observations with increases in lane kilometers greater than 5 percent. Column 7 uses the same selection rule to replicate column 5. Column 8 replicates column 5 again but this time using only observations with declines in lane kilometers greater than 5 percent. The results for large increases in lane kilometers are the same as for the whole sample of MSAs. The elasticity we estimate in column 8 is 0.8. These estimations do not allow us to determine whether the response of traffic to roads is nonlinear in the amount of change to the road network, or if metropolitan areas experiencing large changes are different from those experiencing small changes.8
Finally, column 9 of Table 5 estimates equation (4) including MSA fixed effects and year fixed effects as controls, while column 10 adds MSA population. These
8 Apart from measurement error, decreases in lane kilometers are likely to reflect temporary closures while increases reflect new and permanent construction.
Table 5—Change in VKT as a Function of Change in Lane Kilometers
All All All All All Lane ↑ Lane ↑ Lane ↓ All All MSA sample (1) (2) (3) (4) (5) (6) (7) (8) (9) (10)Panel A. dependent variable: Δln VKT for interstate highways, entire mSAs, OLS
Δln (IH lane km) 1.04*** 1.05*** 1.02*** 1.00*** 0.93*** 1.09*** 0.90*** 0.82*** 1.03*** 1.03***(0.05) (0.05) (0.04) (0.04) (0.04) (0.06) (0.06) (0.09) (0.05) (0.05)
Δln (population) 0.34*** 0.40*** 0.44*** 0.39*** 0.31* 0.45** 0.16 0.51**(0.10) (0.10) (0.11) (0.13) (0.17) (0.21) (0.22) (0.20)
ln (initial VKT) −0.047*** −0.057*** −0.12*** −0.15*** −0.13***(0.006) (0.007) (0.02) (0.03) (0.04)
Geography Y Y Y YCensus divisions Y Y Y YSocioeconomic characteristics
Y Y Y
Past populations Y Y YMSA fixed effects Y Y
R 2 0.87 0.87 0.89 0.90 0.91 0.91 0.94 0.69 0.91 0.94
Panel B. dependent variable: Δln VKT for interstate highways, entire mSAs, TSLS
Δln (IH lane km) 1.05*** 1.02*** 1.00*** 0.92*** 1.07*** 0.90*** 0.82*** 1.03***(0.05) (0.04) (0.04) (0.04) (0.06) (0.05) (0.09) (0.03)
Δln (population) 0.093 0.34** 0.45 1.02** −0.16 1.14 1.50 0.62*(0.18) (0.16) (0.32) (0.45) (0.29) (0.72) (1.45) (0.37)
First stage statistic 63.3 54.3 29.2 23.9 45.7 12.3 4.05 20.1
notes: All regressions include a constant and decade effects. Robust standard errors clustered by MSA in paren-theses. 456 observations for each regression in columns 1–5 and 9–10, 205 in columns 6–7 which consider only increases in lane kilometers of more than 5 percent, and 115 in column 8 which considers declines in lane kilome-ters greater than 5 percent. Instrument for Δln (population) is expected population growth based on initial compo-sition of economic activity.
*** Significant at the 1 percent level. ** Significant at the 5 percent level. * Significant at the 10 percent level.
ERH (Princeton University ) Lecture 6: Transport Costs and Congestion Forces 20 / 43
Instrument 1: 1947 Interstate Highway Plan
FIGURE IThe Projected System of Interstate Highways in 1947
780Q
UA
RT
ER
LY
JO
UR
NA
LO
FE
CO
NO
MIC
S
at Princeton University on March 14, 2012 http://qje.oxfordjournals.org/ Downloaded from
ERH (Princeton University ) Lecture 6: Transport Costs and Congestion Forces 21 / 43
Instrument 2: 1898 Railroads
2630 THE AMERICAN ECONOMIC REVIEW OCTObER 2011
For this reason, we conduct iv estimations only for interstate highways and urban interstate highways.
We now turn to the conditional exogeneity of our two instruments. The 1947 highway plan was first drawn to “connect by routes as direct as practicable the principal metropolitan areas, cities and industrial centers, to serve the national
Figure 2. 1947 US interstate Highway Plan
Source: image based on US House of Representatives (1947).
Figure 3. 1898 US Railroads
Source: image based on Gray (c. 1898).
ERH (Princeton University ) Lecture 6: Transport Costs and Congestion Forces 22 / 43
Instrument 3: Routes of US Major Expeditions ofExploration: 1835 to 1850 2631duRANTON ANd TuRNER: THE fuNdAMENTAl lAW Of ROAd CONgESTIONVOl. 101 NO. 6
defense and to connect suitable border points with routes of continental impor-tance in the Dominion of Canada and the Republic of Mexico” (US Federal Works Agency, Public Roads Administration 1947, cited in Michaels 2008). That the 1947 highway plan was, in fact, drawn to this mandate is confirmed by both econometric and historical evidence reviewed in Duranton and Turner (2008). in particular, in a regression of log 1947 kilometers of planned interstate highways on log 1950 population, the coefficient on log 1950 population is almost exactly one, a result that is robust to the addition of various controls. On the other hand, population growth around 1947 is uncorrelated with planned highway kilometers. Thus, the 1947 plan was drawn to fulfill its mandate and connect major population centers of the mid-1940s, not to anticipate future population or traffic demand.
Note that the exclusion restriction associated with equation (5) requires the orthogonality of the dependent variable and the instruments conditional on control variables. This observation is important. Cities that receive more roads in the 1947 plan tend to be larger than cities that receive fewer. Since we observe that large cit-ies have higher levels of vKT, 1947 planned interstate highway kilometers predicts vKT by directly predicting population and indirectly by predicting 1980 road kilo-meters. Thus the exogeneity of this instrument hinges on having an appropriate set of controls, population in particular.
Next consider the case for the exogeneity of the 1898 railroad network. This net-work was built, for the most part, during and immediately after the civil war, and during the industrial revolution. At this time, the US economy was much smaller and more agricultural than during our study period. in addition, the rail network was developed by private companies with the intention to make a profit from railroad operations in the not too distant future. See Robert Fogel (1964) and Albert Fishlow
Figure 4. Routes of US Major Expeditions of Exploration, 1835 to 1850
Source: image based on US Geological Survey (1970, p. 138).
ERH (Princeton University ) Lecture 6: Transport Costs and Congestion Forces 23 / 43
IV Results2633dURAnTOn And TURnER: ThE fUndAmEnTAL LAW Of ROAd cOngESTIOnVOL. 101 nO. 6
Table 6—VKT as a Function of Lane Kilometers, IV
(1) (2) (3) (4) (5)
Panel A (TSLS). dependent variable: ln VKT for interstate highways, entire mSAs Instruments: ln 1835 exploration routes, ln 1898 railroads, and ln 1947 planned interstates
ln (IH lane km) 1.32*** 0.92*** 1.03*** 1.01*** 1.04***(0.04) (0.10) (0.11) (0.12) (0.13)
ln (population) 0.40*** 0.30*** 0.34*** 0.23*(0.07) (0.09) (0.10) (0.12)
Geography Y Y Y Census divisions Y Y Y Socioeconomic characteristics Y Y Past populations Y Overidentification p-value 0.60 0.11 0.26 0.24 0.29 First-stage statistic 42.8 16.5 11.8 11.5 8.84
Panel B (LImL). dependent variable: ln VKT for interstate highways, entire mSAs Instruments: ln 1835 exploration routes, ln 1898 railroads, and ln 1947 planned interstates
ln (IH lane km) 1.32*** 0.94*** 1.05*** 1.02*** 1.06***(0.04) (0.11) (0.12) (0.13) (0.15)
Overidentification p-value 0.60 0.11 0.26 0.25 0.30
Panel c (TSLS). dependent variable: ln VKT for interstate highways, entire mSAs Instruments: ln 1947 planned interstates
ln (IH lane km) 1.33*** 1.00*** 1.10*** 1.08*** 1.12***(0.05) (0.11) (0.13) (0.13) (0.15)
First-stage statistic 99.7 41.5 29.8 29.5 26.7
Panel d (TSLS). dependent variable: ln VKT for interstate highways, entire mSAs Instruments: ln 1898 railroads
ln (IH lane km) 1.31*** 0.83*** 1.03*** 1.00*** 1.02***(0.06) (0.15) (0.18) (0.18) (0.22)
First-stage statistic 23.7 25.8 19.0 21.1 11.9
Panel E (TSLS). dependent variable: ln VKT for interstate highways, entire mSAs Instruments: ln 1835 exploration routes
ln (IH lane km) 1.25*** 0.63*** 0.75*** 0.68*** 0.72***(0.08) (0.17) (0.18) (0.21) (0.22)
First-stage statistic 53.6 13.8 9.91 7.15 6.32
Panel f (LImL). dependent variable: ln VKT for interstate highways, entire mSAs Instruments: ln 1898 railroads, and ln 1947 planned interstates
ln (IH lane km) 1.39*** 1.09*** 1.18*** 1.15*** 1.20***(0.04) (0.10) (0.11) (0.13) (0.16)
Overidentification p-value 0.69 0.10 0.31 0.25 0.29 First-stage statistic 37.9 17.7 12.1 14.4 9.51
Panel g (LImL). dependent variable: ln VKT for interstate highways, entire mSAs Instruments: ln 1898 railroads, and ln 1947 planned interstates
ln (IH lane km) 1.33*** 0.98*** 1.13*** 1.08*** 1.13***(0.05) (0.13) (0.16) (0.15) (0.17)
Overidentification p-value 0.91 0.53 0.97 0.88 0.81 First-stage statistic 53.1 22.7 14.4 15.8 11.7
Panel h (LImL). dependent variable: ln VKT for interstate highways, entire mSAs Instruments: ln 1898 railroads, and ln 1947 planned interstates
ln (IH lane km) 1.26*** 0.82*** 0.93*** 0.92*** 0.97***(0.05) (0.11) (0.13) (0.13) (0.16)
Overidentification p-value 0.77 0.55 0.96 0.98 0.93 First-stage statistic 52.2 21.0 14.2 14.4 9.76
notes: All regressions include a constant (and year effects for panels A–E). Robust standard errors in parenthe-ses (clustered by MSA in panels A–E); 684 observations corresponding to 228 MSAs for each regression for pan-els A–E and 228 observations for panels F–H.
*** Significant at the 1 percent level. ** Significant at the 5 percent level. * Significant at the 10 percent level.
ERH (Princeton University ) Lecture 6: Transport Costs and Congestion Forces 24 / 43
Commuting Today (MRRH 2015)Counties become more open over time
0.0
5.1
Den
sity
0 .2 .4 .6 .8 1Share of Residents that Work in the County Where They Live
1960 1970 1980 1990 2000
Commuting links are sizeable and heterogeneous
Min p5 p10 p25 p50 p75 p90 p95 Max MeanCommuters from Residence County 0.00 0.03 0.06 0.14 0.27 0.42 0.53 0.59 0.82 0.29Commuters to Workplace County 0.00 0.03 0.07 0.14 0.20 0.28 0.37 0.43 0.81 0.22County Employment/Residents 0.26 0.60 0.67 0.79 0.92 1.02 1.11 1.18 3.88 0.91
Commuters from Residence CZ 0.00 0.00 0.01 0.03 0.07 0.12 0.18 0.22 0.49 0.08Commuters to Workplace CZ 0.00 0.00 0.01 0.03 0.07 0.10 0.13 0.15 0.25 0.07CZ Employment/Residents 0.63 0.87 0.91 0.97 1.00 1.01 1.03 1.04 1.12 0.98Tabulations on 3,111 counties and 709 CZ after eliminating business trips (trips longer than 120km).
ERH (Princeton University ) Lecture 6: Transport Costs and Congestion Forces 25 / 43
Gravity in Goods Trade Across CFS RegionsSlope: -1.29 (after removing origin and destination fixed-effects)
-50
510
Log
Trad
e Fl
ows
(Res
idua
ls)
-8 -6 -4 -2 0 2Log Distance (Residuals)
Dashed line: linear fit; slope: -1.29
ERH (Princeton University ) Lecture 6: Transport Costs and Congestion Forces 26 / 43
Gravity in Commuting FlowsSlope: -4.43 (after removing origin and destination fixed-effects)
-50
510
Log
Com
mut
ing
Flow
s (R
esid
uals
)
-2 -1 0 1Log Distance (Residuals)
Dashed line: linear fit; slope: -4.43
ERH (Princeton University ) Lecture 6: Transport Costs and Congestion Forces 27 / 43
Elasticity of Local Employment to Productivity5% productivity shocks
S. Diego (CA)
dlnLM/dA: 0.63
New Haven (CT)
dlnLM/dA: 1.47
Arlington (VA)
dlnLM/dA: 2.35
0.2
.4.6
.81
Den
sity
0 .5 1 1.5 2 2.5Elasticity of Employment to Productivity
Eliminating bottom and top 0.5%; gray area: 95% boostrapped CI
ERH (Princeton University ) Lecture 6: Transport Costs and Congestion Forces 28 / 43
Local Employment vs. Resident Elasticity to Productivity5% productivity shocks
S. Diego (CA) dlnLM/dA: 0.63 λnn|n: .996
New Haven (CT)
dlnLM/dA: 1.47
λnn|n: .746
Arlington (VA)
dlnLM/dA: 2.35
λnn|n: .310
01
23
Den
sity
0 .5 1 1.5 2 2.5Elasticity of Employment and Residents to Productivity
Employment Residents
Eliminating bottom and top 0.5%; gray area: 95% boostrapped CI
ERH (Princeton University ) Lecture 6: Transport Costs and Congestion Forces 29 / 43
The Role of Commuting in Local Labor Demand Shocks
Announcements of Million Dollar Plants (MDP)I Compare winning county where new firm locates to runner-up counties
82 MDP announcements from Greenstone and Moretti (2004)I GHM(2010) use subset of 47 MDP openings in (confidential) Census data
We generalize GHM(2010) with commuting interactions
ln Lit = κIjτ + θ(Ijτ ·Wi
)+β
(Ijτ · λRii |i
)+ γ
(Ijτ ·Wi · λRii |i
)+
+αi+ηj+µt+εit
I i : counties; j : cases; t: calendar year; τ: treatment year index;I Lit : employment in county i , t years after announcement;I Ijτ : 1 for case j starting in treatment year;I Wi : indicator for winner county;I λRii |i : residence own-commuting share in 1990 (experiment with more)I αi , ηj , µt : counties, cases, calendar years fixed effects.
ERH (Princeton University ) Lecture 6: Transport Costs and Congestion Forces 30 / 43
The Role of Commuting in Local Labor Demand Shocks
Variable Coeffi cient (1) (2) (3) (4) (5) (6) (7) (8)Ijτ ×Wi θ 0.057∗∗ 0.250∗∗∗ 0.191∗∗∗ 0.244∗∗∗ 0.260∗∗∗ 0.223∗∗∗ 0.160∗∗ 0.159∗∗
(0.018) (0.078) (0.065) (0.068) (0.078) (0.078) (0.060) (0.066)Ijτ ×Wi × λRii |i γ -0.242∗∗ -0.219∗∗ -0.190∗∗ -0.195∗∗
(0.096) (0.096) (0.077) (0.066)Ijτ ×Wi × λLii |i γ -0.177∗∗
(0.087)Ijτ ×Wi × λARLii |i γ -0.241∗∗∗
(0.088)Ijτ ×Wi × λMRLii |i γ -0.281∗∗
(0.110)Ijτ × λRii |i β 0.012 -0.048 -0.203∗∗∗ -0.213∗∗
(0.135) (0.108) (0.075) (0.082)Ijτ × λLii |i β 0.243∗
(0.129)Ijτ × λARLii |i β 0.124
(0.160)Ijτ × λMRLii |i β 0.133
(0.145)Ijτ κ -0.015∗ -0.024 -0.200∗∗ -0.113 -0.113 0.021 0.160∗∗ 0.159∗∗
(0.008) (0.109) (0.096) (0.125) (0.106) (0.086) (0.060) (0.066)County Fixed Effects Yes Yes Yes Yes Yes Yes Yes YesCase Fixed Effects Yes Yes Yes Yes Yes Yes Yes YesYear Fixed Effects Yes Yes Yes Yes YesIndustry-year Fixed Effects YesCensus-region-year Fixed Effects YesState-year Fixed Effects YesObservations 4,431 4,431 4,431 4,431 4,431 4,431 4,431 4,431R-squared 0.991 0.0991 0.991 0.991 0.991 0.992 0.994 0.996
County observations are weighted by population at the beginning of the sample period. Standard errors are clustered by state. ∗ p-value ≤ 0.1; ∗∗
p-value ≤ 0.05l; ∗∗∗ p-value ≤ 0.01.
ERH (Princeton University ) Lecture 6: Transport Costs and Congestion Forces 31 / 43
NY Area Employment/Residents in the Data
N E W J E R S E Y
N E W Y O R K
N E W Y O R K
C O N N E C T I C U TP E N N S Y LVA N I A
P E N N S Y LVA N I A
0.263 - 0.6620.663 - 0.8090.810 - 0.9490.950 - 1.0501.051 - 1.2721.273 - 2.1802.181 - 3.878
ERH (Princeton University ) Lecture 6: Transport Costs and Congestion Forces 32 / 43
NY Area Employment ChangeShutting down commuting
N E W J E R S E Y
N E W Y O R K
N E W Y O R K
C O N N E C T I C U TP E N N S Y LVA N I A
P E N N S Y LVA N I A
-0.725 - -0.450-0.449 - -0.250-0.249 - -0.021-0.020 - 0.0200.021 - 0.2000.201 - 0.5000.501 - 1.298
ERH (Princeton University ) Lecture 6: Transport Costs and Congestion Forces 33 / 43
NY Area Residents ChangeShutting down commuting
N E W J E R S E Y
N E W Y O R K
N E W Y O R K
C O N N E C T I C U TP E N N S Y LVA N I A
P E N N S Y LVA N I A
-0.494 - -0.450-0.449 - -0.250-0.249 - -0.021-0.020 - 0.0200.021 - 0.2000.201 - 0.5000.501 - 0.575
ERH (Princeton University ) Lecture 6: Transport Costs and Congestion Forces 34 / 43
NY Area Real Income ChangeShutting down commuting
N E W J E R S E Y
N E W Y O R K
N E W Y O R K
C O N N E C T I C U TP E N N S Y LVA N I A
P E N N S Y LVA N I A
-0.456 - -0.203-0.202 - -0.109-0.108 - -0.021-0.020 - 0.0200.021 - 0.0250.026 - 0.1010.102 - 0.410
ERH (Princeton University ) Lecture 6: Transport Costs and Congestion Forces 35 / 43
Berlin Subway Extension
Public Transport Network
10 / 81
ERH (Princeton University ) Lecture 6: Transport Costs and Congestion Forces 36 / 43
Berlin Subway Extension
Public Transport Network Zoom
11 / 81
ERH (Princeton University ) Lecture 6: Transport Costs and Congestion Forces 37 / 43
Berlin Subway Extension
Mean Relative Travel Time Reduction
36 / 81
ERH (Princeton University ) Lecture 6: Transport Costs and Congestion Forces 38 / 43
Berlin Subway Extension
Relative Increase Floor Prices
37 / 81
ERH (Princeton University ) Lecture 6: Transport Costs and Congestion Forces 39 / 43
Berlin Subway Extension
Relative Increase Residence Employment
38 / 81
ERH (Princeton University ) Lecture 6: Transport Costs and Congestion Forces 40 / 43
Berlin Subway Extension
Relative Increase Real Expected Residential Income
39 / 81
ERH (Princeton University ) Lecture 6: Transport Costs and Congestion Forces 41 / 43
Berlin Subway Extension
Relative Increase Workplace Employment
40 / 81
ERH (Princeton University ) Lecture 6: Transport Costs and Congestion Forces 42 / 43
Berlin Subway Extension
Relative Reduction Wages
41 / 81
ERH (Princeton University ) Lecture 6: Transport Costs and Congestion Forces 43 / 43