Download Impact of Urban Highways on Commuting and Growth in Cities and Suburbs and more Study notes Economics in PDF only on Docsity!
Changes in Transportation Infrastructure and Commuting Patterns
in U.S. Metropolitan Areas, 1960-
Nathaniel Baum-Snow
Department of Economics
Box B
Brown University
Providence, RI 02912
Nathaniel_Baum-Snow@brown.edu
401-863-2697 (phone)
401-863-1970 (FAX)
Presented at the 2010 American Economic Association Meetings in Atlanta, GA
in the published session
“Housing and Labor Markets”
chaired by Enrico Moretti, University of California, Berkeley
Changes in Transportation Infrastructure and Commuting Patterns
in U.S. Metropolitan Areas, 1960-
Nathaniel Baum-Snow *
Population decentralization has been a salient feature of the landscape of most U.S. urban
areas since 1950. Nathaniel Baum-Snow (2007) documents that the aggregate population of
central cities of the 139 largest metropolitan areas (henceforth, MSAs) declined by 17 percent
between 1950 and 1990 while aggregate MSA population growth was 72 percent during this
period. Expansion of the highway network in urban areas accounts for about one-third of the gap
in central city and MSA population growth rates. While transport network expansions clearly
generated urban population decentralization, there is little evidence to date on how this
decentralization manifested itself as changes in employment locations and commuting patterns.
In this paper, I present evidence indicating that employment decentralization occurred
apace with residential decentralization between 1960 and 2000 such that their relative spatial
concentrations remained remarkably unchanged. A byproduct has been that most commutes of
MSA residents no longer involve central cities at all. Central cities as defined by their
geographies in 1960 were the origin and/or destination of only 38 percent of commutes made by
MSA residents in 2000, down from 66 percent in 1960. Using planned portions of the interstate
highway system as a source of exogenous variation, estimates reported in Section III indicate
that urban highway construction played a pivotal role in generating this shift. New highways
primarily increased the number and fraction of commuting flows within suburban areas at the
expense of commutes within central cities. Because within suburb commutes are longer than
other types of commutes on average, results are consistent with Gilles Duranton & Matthew
- (^) Department of Economics, Box B, Brown University, Providence, RI 02912. Nathaniel_Baum-Snow@brown.edu
I thank David Card for inviting me to participate in this session. Kailin Clarke provided excellent research assistance.
Using these data, Table 1 presents some facts about the evolution of the spatial distribution of
employment and residences for the 99 MSAs with at least 250 thousand residents in 1960. This
table only includes counts of workers. Table 1 indicates that employment and working
population decentralized at remarkably similar rates between 1960 and 2000. While in 1960 50
percent of MSA residents and 60 percent of MSA jobs were in central cities, by 2000 these
numbers had declined to 23 and 32 percent respectively for the same geographic regions.^3 The
higher spatial concentration of jobs means that while central cities lost residents, they gained
jobs over time. The number of central city jobs increased by 14 percent between 1960 and 2000
while working population declined by 7 percent. Since labor force participation rates increased
rapidly over this time period, total population in central cities declined at a faster rate than did
working population. In addition, note the importance of holding the central city geography
(^3) Due to data limitations, in 2000 I measure metropolitan areas as county agglomerations. This does not affect
results much because MSAs have expanded spatially with their populations.
Data Cen tral City Liv e in W o rk in Liv e in W o rk in Year Geo g rap h y Cen tral City Cen tral City (S)M SA (S)M SA 1960 1960 17.8 21.1 35.5 35. (Fractio n o f M SA ) (0.50) (0.60) 2000 2000 21.4 29.1 73.0 75. (Fractio n o f M SA ) (0.29) (0.39) 2000 1960 16.6 24.1 73.0 75. (Fractio n o f M SA ) (0.23) (0.32) Percen t 2000 0.21 0.38 1.06 1. Ch an g e 1960 -0.07 0.14 1.06 1. Ch an g e in 2000 -0.21 -0. Fractio n 1960 -0.27 -0.
Table 1 : Chang es in Res idential and W ork Locations , 1 9 6 0 -2 0 0 0 9 9 Metropolitan Areas of O ver 2 5 0 ,0 0 0 in 1 9 6 0
No te: Co u n ts are in millio n s o f wo rkers. Co u n ts are calcu lated u s in g 1960 an d 2000 cen s u s jo u rn ey to wo rk d ata u s in g co n temp o ran eo u s SM SA /M SA d efin itio n s. T h o s e co n trib u tin g to co u n ts in Co lu mn s 1 an d 3 may wo rk an y wh ere. T h o s e co n trib u tin g to co u n ts in co lu mn s 2 an d 4 may liv e an y wh ere. Data fro m 1960 in co rp o rate th e au th o r's imp u tatio n s fo r n o n rep o rted wo rk lo catio n s wh ile th e 2000 d ata in co rp o rates s u ch imp u tatio n s d o n e b y th e Cen s u s Bu reau an d th e au th o r.
constant over time for understanding trends in aggregate counts. Absent this adjustment, the
number of workers and jobs in central cities appear to grow by 21 percent and 38 percent
respectively. However, the fractions living and working in contemporaneously defined central
cities both declined by 0.21, exhibiting a similar pattern as seen in data built using constant
geographic units.
For a larger sample of 152 metropolitan areas, Table 2 breaks out the working population
of central city and suburban metropolitan regions into commuting flows to central cities and
other destinations.^4 Evidence in Table 2 indicates that the nature of commutes changed
dramatically from 1960 to 2000. While 45 percent of workers living in MSAs lived and worked
in central cities in 1960, just 15 percent did in 2000. To compensate, the fraction living and
working in the suburbs almost doubled from 0.34 to 0.62, representing a near tripling in number.
(^4) Data limitations preclude separate identification of suburban and ex-metropolitan area destinations.
Data Cen tral City Liv e in CC Liv e in CC Liv e in Su b u rb s Liv e in Su b u rb s Year Geo g rap h y W o rk in CC W o rk Els ewh ere W o rk in CC W o rk Els ewh ere 1960 1960 17.3 2.4 6.1 13. (Fractio n o f M SA ) (0.45) (0.06) (0.16) (0.34) 2000 2000 18.3 6.8 15.8 42. (Fractio n o f M SA ) (0.22) (0.08) (0.19) (0.51) 2000 1960 12.4 6.1 12.7 51. (Fractio n o f M SA ) (0.15) (0.07) (0.15) (0.62) Percen t 2000 0.05 1.89 1.61 2. Ch an g e 1960 -0.29 1.61 1.08 2. Ch an g e in 2000 -0.23 0.02 0.03 0. Fractio n 1960 -0.30 0.01 0.00 0. No tes : Reg io n s o u ts id e o f cen tral cities acco rd in g to co n temp o ran eo u s M SA d efin itio n s are as s ig n ed as s u b u rb s. T h e 15.8 millio n wh o are in d icated to liv e in th e s u b u rb s an d wo rk in 2000 d efin itio n cen tral cities in y ear 2000 may in fact liv e in s id e o r o u ts id e o f an M SA. T h e 42.2 millio n liv in g in th e s u b u rb s o f 2000 d efin itio n cen tral cities in y ear 2000 an d wo rkin g els ewh ere is calcu lated as a res id u al o f th e o th er th ree co mmu tin g flo ws fo r th is g eo g rap h y an d time p erio d.
Table 2 : Chang es in Commuting P atterns : 1 9 6 0 -2 0 0 0 1 5 2 Metropolitan Areas of O ver 1 0 0 ,0 0 0 in 1 9 6 0
M SA W o rkin g Res id en ts
133 in 2000. However, construction progress was not sufficient such that much of this 1960
infrastructure constituted full radial highways connecting cities to suburbs. Therefore, to be
conservative, the key explanatory variable used in this analysis is the change in the number of
rays constructed between 1950, when only 7 MSAs had highways, and 2000. Experimentation
with various alternative measures using different cutoff levels of mileage open to traffic in metro
areas as of 1960 to constitute rays always generates point estimates that are larger in magnitude
than those reported below. This occurs because the first stage coefficient on planned rays is
smaller the more actual rays that were open as of 1960.
The first stage coefficient on rays received by each city as planned in 1947 is 0.47 with a
standard error of 0.07. This coefficient changes little as a function of included control variables
except that inclusion of square root of 1960 central city area reduces it from 0.65.^5 Analogous
regressions that count a ray as being open in 1960 if at least 4 miles of the highway was
operational consistently yield statistically significant first stage coefficients of 0.20. This
estimate is remarkably robust, even to the exclusion of central city radius. First stage regressions
also indicate that central cities with more area and MSAs with greater employment growth also
received more rays between 1950 and 2000, all else equal.
III. Results
Table 3 presents IV estimates of the effects of highways on the four types of commuting
flows discussed above as measured using 1960 central city geography. Because it is arguably
exogenous and it is the only potential control variable correlated with endogenous rays
conditional on planned rays, the primary specification controls for the square root of central city
(^5) Included control variables are listed in Table 3. In Baum-Snow (2007) I argue that a measure of central city size is
a crucial control since spatially larger cities had higher populations and thus received more planned highways. Furthermore, a broad class of land use models predict that magnitudes of responses of commuting flows to transport infrastructure depend crucially upon the spatial size of origin and destination regions.
area. Consistent with standard theories of land use, a robustness specification also controls for
simulated MSA mean income and MSA employment growth with no meaningful effects on
coefficients of interest. The simulated income measure is built using 1940 MSA employment
shares by industry and national skill prices from 1960 and 2000 excluding states that include the
MSA of interest. Panel A presents results for central city residents while Panel B shows results
for suburban residents.
Point estimates reported in Table 3 indicate that highways caused declines in all three types
of commutes involving central cities with precise estimates for within-central city and reverse
commutes. Highways caused commensurate increases in commutes of suburban residents to
suburban or ex-MSA areas. In particular, each radial highway caused an estimated 18 percent
decline in the number of people who both lived and worked in a central city and a 10 percent
decline in reverse commuters. Together, this amounts to a 16 percent decline in central city
working residents caused by each ray.
These estimates indicate that the average city, which received 2.5 rays, saw the number of
workers living there decline by 40 percent as a result of new highways. While this estimate
seems large, it only implies a growth in resident workers of 33 percent absent no highway
construction, well below the overall growth in metropolitan workers of over 100 percent.
Similar to my estimates of the effects of highways on total working and non-working central city
population, this calculation indicates that absent highways, about one-third of the gap between
central city and metro area working population growth rates would be closed.
Table 3 Panel B indicates that the central city residents that leave because of highway
construction do not generally keep jobs in the city. While not statistically significant, the
primary specification yields a point estimate indicating that each ray caused an 18 percent
The evidence is clear that the primary way that highways serving central cities caused
declines in central city population was by inducing those who had lived and worked in central
cities to live and work in suburban areas instead. However, it is important to recognize that part
of the way radial highways may have brought about this change is by encouraging more
commuting by car. In addition, other types of urban transport infrastructure including
Ch an g e in Ray s -0.18* * * -0.17* * * -0.10* -0.11* * 1960 to 2000 (0.05) (0.05) (0.06) (0.05) Sq u are Ro o t o f 1960 0.10* * * 0.10* * * 0.07* * * 0.06* * * Cen tral City A rea (0.02) (0.01) (0.01) (0.02) Ch an g e in Simu lated 2.07 -0. In co me (1.44) (1.57) Ch an g e in Lo g 0.35* * * 0.63* * * M SA Emp lo y men t (0.10) (0.11) Co n s tan t -0.74* * * -1.61* * * 0.73* * * 0. (0.12) (0.41) (0.15) (0.44) N 152 152 152 152 R-Sq u ared 0.13 0.34 0.09 0.
Ch an g e in Ray s -0.17 -0.18 0.25* * * 0.23* * * 1960 to 2000 (0.10) (0.11) (0.08) (0.07) Sq u are Ro o t o f 1960 0.10 0.08 -0.04 -0.06* * Cen tral City A rea (0.06) (0.05) (0.03) (0.02) Ch an g e in Simu lated 0.64 -0. In co me (2.19) (2.94) Ch an g e in Lo g 1.05* * * 1.23* * * M SA Emp lo y men t (0.20) (0.21) Co n s tan t 0.74* * * -0.07 1.18* * * 0. (0.17) (0.69) (0.17) (0.86) N 152 152 152 152 R-Sq u ared 0.08 0.40 0.05 0. No tes : Reg res s io n s are o f th e ch an g e in th e lo g o f o u tco mes lis ted in co lu mn h ead ers o n v ariab les lis ted at left. T h e s amp le in clu d es th e s ame 152 M SA s u s ed fo r T ab le 2. T h e ch an g e in th e n u mb er o f ray s is in s tru men ted with ray s in th e 1947 n atio n al p lan. Stan d ard erro rs are clu s tered b y M SA.
P anel B : S uburban Res idents Liv e in Su b u rb s Liv e in Su b u rb s W o rk in CC W o rk Els ewh ere
Table 3 : Effects of Hig hways on Commuting P atterns
W o rk in CC W o rk Els ewh ere
P anel A: Central City Res idents Liv e in CC Liv e in CC
circumferential highways and transit may have interacted with radial highways to generate a
portion of estimated effects.^8
IV. Conclusions
U.S. cities have decentralized to the point that most working residents no longer have any
contact with their MSAs’ central cities. Highway construction has played a crucial role in this
growth of the suburbs and decline of cities, not just as residential locations but also as work
locations. Evidence from this paper indicates that employment and residential decentralization
occurred in tandem and that highways primarily led within-central city commuters to become
within-suburb commuters. Estimates indicate that had the urban highway systems not been built,
the total number of within city commutes would be about double its 2000 number and the total
number of within-suburb commutes would be cut by about one-half. This reallocation would
change the fraction of total commutes in these two categories to 0.31 and 0.30 respectively, a
dramatic shift.
These results provide evidence that declining city transport costs not only allowed
commuters to spread out spatially but also allowed firms to attain the same productivity
advantages from proximity as before but at further distances. This means that firm productivity
has potentially increased because of new urban transport infrastructure even as firms have
decentralized. Welfare improvements of new highways for workers have thus potentially come
in the form of higher wages that reflect this increased productivity in addition to faster commutes
and lower housing costs.
(^8) While, Baum-Snow & Kahn (2005) find that the large expansion of rail transit lines in U.S. cities did little to stem
the secular trend of declining public transit ridership over time, Baum-Snow (2007) argues that the existence of circumferential highways may augment the negative effect of radial highways on central city population.