Professional Documents
Culture Documents
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at
http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may use content in the JSTOR archive only for your personal, non-commercial use.
Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at
http://www.jstor.org/action/showPublisher?publisherCode=aea.
Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.
JSTOR is a not-for-profit organization founded in 1995 to build trusted digital archives for scholarship. We work with the
scholarly community to preserve their work and the materials they rely upon, and to build a common research platform that
promotes the discovery and use of these resources. For more information about JSTOR, please contact support@jstor.org.
American Economic Association is collaborating with JSTOR to digitize, preserve and extend access to Journal
of Economic Literature.
http://www.jstor.org
Journal of Economic Literature
Vol. XXXVI (September 1998), pp. 1426-1464
ALEX ANAS
RICHARD ARNOTT
and
KENNETH A. SMALL1
ates the need to produce at discrete what is the level of spatial resolution at
points in space because of increasing which such phenomena are best ana-
returns to scale in production (David lyzed?
Star'rett 1974). Because of these perva-
sive externalities and nonconvexities,
2. History and Description of Urban
economic analysis when applied to ur-
Spatial Structure
ban geography yields results that differ
in important and interesting ways from We begin with a sketch of how urban
results of other branches of economics. form has evolved in modern times, fol-
Agglomeration economies also create lowed by some observations on how to
first-mover advantages and regional measure its characteristics.
specializations that are important in in-
2.1. Recent Evolution of Urban Formn
ternational trade (Paul Krugman
1991a), and some first-mover disadvan- The spatial structure of modern cities
tages that prevent optimal dynamic was shaped, in large measure, by ad-
growth paths from being realized. We vances in transport and communication.
discuss these in Section 5. The history of urban development in
Agglomeration economies are, of North America since colonial times al-
course, not new. As eloquently expos- lows us to document aspects of this pro-
ited by Raymond Vernon (1960) and cess (Charles Glaab and Theodore
Benjamin Chinitz (1961), they are at Brown 1967).
the heart of our current understanding Prior to about 1840, most cities were
of central business districts. But recent tied to waterways such as harbors, riv-
changes in the technology of agglomera- ers, and canals. The average cost of
tion, due to advances in information processing freight fell sharply with the
processing and telecommunications, quantity processed at a particular port,
may profoundly alter the pattern of spa- creating substantial scale economies at
tial developmelnt (Jess Gaspar and Glae- harbors or river junctions with access to
ser 1998). Understanding these new the sea. Similarly, as railroads competed
forces will help us understand newly with waterways later in the 19th cen-
emerging forms of urban structure as tury, scale economies in rail terminals
well as basic determinants of industrial created accessibility advantages near
structure and interregional and interna- them as well. Meanwhile intra-urban
tional trade. freight transport took place mainly by
While our focus is on explaining ur- horse and wagon, which was time con-
ban spatial structure as a result of mar- suming and unreliable in bad weather.
ket processes, we touch on two related These conditions favored the growth of
issues as well. The first concerns the a single manufacturing district located
role of government. Government poli- near the harbor or railhead, with resi-
cies-notably land-use controls and the dences surrounding it (Leon Moses and
provision of transportation infrastruc- Harold Williamson 1967).
ture-play a major role in shaping cit- In the last quarter of the century, the
ies. What can we say about optimal pol- telegraph greatly speeded the flow of
icy? The second issue concerns the information from city to city (Alexander
importance of space in economics. Ac- Field 1992). But economies of scale
counting for location yields new insights prevented it from being used much
into economic phenomena that are nor- within a city-instead, messengers re-
mally analyzed in aspatial models. But mained the primary means of intra-city
Anas, Arnott, Smnall:Urban Spatial Structure 1429
business communication. The high cost The next big changes were the dis-
of intra-urban communication meant semination of the internal combustion
that even light manufacturing and ser- engine and the telephone in the early
vice industries tended to concentrate twentieth century. Gradually the horse
within the central manufacturing core, and wagon were replaced by the small
as shown for New York by Chinitz urban truck, and the messenger by the
(1960). But this small core area was far telephone. For example, in the single
from homogeneous; rather, it was di- decade from 1910 to 1920, truck regis-
vided into districts, each specialized in trations in Chicago increased from 800
an activity such as commercial banking, to 23,000, while horse-drawn vehicle
pawnbrokerage, or light or heavy manu- registrations dropped almost by half.
facturing. In late nineteenth-century Moses and Williamson (1967) estimate
Chicago, four-fifths of the city's jobs that variable costs and travel time for
were located within four miles of State the truck were less than half those for
and Madison streets, according to Ray- the horse and wagon. The truck and the
mond Fales and Moses (1972), who go telephone allowed businesses to spread
on to show how a pattern of specialized outward from the center, thereby taking
districts arose due to agglomerative advantage of lower land values while
forces within industries and the link- maintaining their links to the central
ages among them. port or railhead. Thus central business
Before 1850, personal transport districts expanded. In Chicago, firms
within the city was mainly by foot and that moved in 1920 located on average
horse-drawn carriage, causing the great 1.5 miles from the core, as opposed to
majority of rich and poor alike to live 0.92 miles in 1908 (Moses and William-
close to the city center. For the most son 1967).
part, the rich outbid the poor for the The automobile, at first restricted to
most central and hence most convenient richer families, rapidly increased in im-
sites, so that income declined markedly portance with assembly-line production
with distance from the CBD, as is docu- of the Model T Ford starting in 1908.
mented in studies of Milwaukee, Pitts- Cars broadened the coverage of motor-
burgh, and Toronto (Stephen LeRoy ized personal transport, causing the ar-
and Jon Sonstelie 1983). eas between the streetcar suburbs to be
Between 1850 and 1900, the advent settled and the residential apron to ex-
of horse-drawn and then electric street- pand. The automobile competed suc-
cars enabled large numbers of upper- cessfully with mass transit, despite tran-
and middle-income commuters to move sit fares remaining flat in nominal terms
further out. This migration gave rise to from the beginning of the century until
"streetcar suburbs," residential enclaves approximately World War II; it did this
organized around a station on a radial mainly by providing speed, privacy, and
streetcar line (Sam Warner 1962). To- convenience, al-though it was also facili-
ward the turn of the century, subways tated by an active program of building
further contributed to this pattern in and upgrading public roads (Paul Bar-
the largest cities. Thus developed a spa- rett 1983).
tial structure now known as the "nine- As assembly-line production became
teenth century city," consisting of a widespread, the lower capital-land ra-
compact production core surrounded by tios characterized by flat buildings in-
an apron of residences concentrated creased the attractiveness of locations
around mass transport spokes. where land was cheap. Nevertheless,
1430 Journal of Economic Literature, Vol. XXXVI (September 1998)
inside the urban boundary. The fractal terns somewhat more regular, at least in
dimension D for this situation can be western Europe. Batty and Longley sug-
measured by observing how rapidly the gest that a possible reason is the more
fraction of zones containing urban de- extensive imposition of land-use con-
velopment falls as zonal size is de- trols and other forms of urban planning.
creased, i.e., as resolution becomes Unfortunately, the estimated areal
finer. (More precisely, D is twice the fractal dimension of a city is quite sen-
elasticity of the number of zones con- sitive to just how the land-use data are
taining any development with respect to summarized (Batty and Longley, p.
the total number of zones into which 236). Another problem is that in such a
the fixed urban area is divided.) We call measurement, a city's fine structure is
this dimension the areal fractal dimen- assumed to look like a miniature of the
sion; it can vary from 0, indicating that coarse structure, whereas in fact the
nearly all the interior space is empty processes operating at the micro and
when examined at a fine enough resolu- macro scales are probably very differ-
tion, to 2, indicating that each coarsely- ent: fine structure may reflect local zon-
defined zone that contains development ing rules or developers' detailed design
is in fact fully developed. Long narrow strategies, while coarse structure may
development would have D = 1 (since as reflect regional planning, regional
we increase the total number N of transportation facilities, or land specu-
zones into which a well-defined region lation based on anticipated regional
is divided, the number of zones contain- growth. Nevertheless, the fractal ap-
ing any development would grow only proach highlights the inadequacy of a
as N). deterministic view of development,
Batty and Longley (1994, Table 7.1) adopted especially in earlier economic
report estimated areal fractal dimen- models, in accounting for the irregulari-
sions for many cities around the world, ties in urban structure. As we discuss in
with the result most often in the range Section 5, more recent advances such as
1.55 to 1.85. For Paris in 1981 the esti- random utility theory enable us to deal
mate is 1.66. For Los Angeles in the with irregularities in a way that is better
same year, it is 1.93, tied with Beijing suited to economic modeling.
for the highest among the 28 cities re- Most urban economists have used
ported. This latter estimate implies that more intuitive, if simplified, depictions
the fraction of area developed is almost of urban structure, identifying one or
constant at different scales, indicating a more employment centers and estimat-
relative absence of fine-structure ir- ing how these centers affect employ-
regularities in development patterns. mnentand population densities around
Apparently Los Angeles has grown in a them. Much of the early literature on
more homogeneous manner than Car- subcenters used criteria based on local
diff or Paris. knowledge in planning organizations or
Time series observations of London real estate firms. More recent work has
from 1820 to 1962, and of Berlin from used objective definitions based on em-
1875 to 1945, suggest that the areal ployment data for a large number of
fractal dimension has been increasing zones within a metropolitan area (John
steadily throughout these time periods. McDonald 1987). Genevieve Giuliano
This lends further support to the con- and Kenneth Small (1991) define a
clusion that urban growth during the in- "center"-either a main center (the one
dustrial era has made development pat- containing the CBD) or a subcenter-as
1434 Journal of Economic Literature, Vol. XXXVI (September 1998)
indifferent between the two locations, more than one type of household or to
lot rent must be lower at the more dis- other sectors bidding on land outside the
tant location by the same amount: that is, CBD; in such generalizations, the mar-
Ldb = - T'(x)dx. ket rent function is the upper envelope
For each household, there is a family of applicable bid-rent functions.
of residential bid-rent functions, in- The comparative statics of the model
dexed by ui. Households are treated as were first fully worked out by William
identical and costlessly mobile. Hence, Wheaton (1974). To illustrate their
they all obtain the same utility in equi- derivation, consider the case of an in-
librium, and the equilibrium rent func- crease in household population, N. This
tion r(x) coincides with one of these causes no change in the family of bid-
bid-rent functions. To determine which rent functions (1) or in the lot-size
one, two conditions are needed. First, function L[.] corresponding to any given
there is an arbitrage condition at the net income and utility. But from (3) the
city boundary (whose value x* is yet to higher population would create excess
be determined): residential rent there demand for land if the solution were
must equal the rent on land in non- unchanged. Equilibrium is reestab-
urban use, rA. (This opportunity cost of lished with higher densities, lower util-
land, often called "agricultural rent," is ity, a steeper rent function, and an ex-
assumed not to vary with location.) Sec- panded outer boundary.
ond, all households must be accommo- Land use in the simple monocentric
dated, which means the integral of model is efficient-that is, the equilib-
household density (i/L) over the resi- rium density pattern is Pareto optimal
dential area must equal the number of (Fujita 1989). This is basically because
households: there are no externalities; land-use de-
cisions are based entirely on tradeoffs
~()- dx =N, (3) between desire for space and recogni-
XcL[y - T(x),u] tion of commuting costs, both of which
are purely private. The need for com-
where p(x)dxis the residential land area muting is exogenous in the model, so no
between x and x + dx.8 These two con- agglomerative effects are present. Of
ditions provide two equations in the un- course, these nice properties disappear
knowns x* and ui;we denote the solution in more realistic models with conges-
for u by ue. tion, air pollution, neighborhood quality
The land rent at any location is the effects, and economies of agglomera-
maximum of the bid rents there: tion-the last being of prime interest in
this essay.
r(x) = max [b(x,ue),rA] Several comments are in order about
{ b(x,ue)forx<x* the limitations of the monocentric
IrA for x > x*. model. The model implicitly assumes
that businesses have steeper bid-rent
This expresses the principle that, in the functions than residents, so that all jobs
land market, each piece of land goes to are centrally located. But most of its re-
the highest-bidding use. This principle is sults can follow from the weaker as-
the basis for generalizing the model to sumption that employment is dispersed
in a circularly symmetric manner, so
8 If all urban land is used for residential pur- long as it is less dispersed than resi-
poses and the city is circular, then ?(x) = 2irx. dences-that is, within any circle there
1436 Journal of Economic Literature, Vol. XXXVI (September 1998)
are more jobs than resident workers. In changes in urban structure over the last
this case the wage varies over location century and a half? It obviously throws
so as to offset differences in commuting no light on the trend toward polycen-
costs (Robert Solow 1973; Michelle tricity. If it applies to anything, it
White 1988), and commuters still should help explain the broad popula-
choose to travel radially inward to tion decentralization that has occurred
work. in most cities of the world (Mills and
The model is also easily extended to Jee Peng Tan 1980). To see how the
incorporate different groups of resi- model performs, we need to quantify
dents. For example, it can predict the the empirically observed trends and
pattern of residential location by in- provide some plausible parameters for
come. In order to do this, marginal the model.
transport cost T'(x) has to be reinter- Pioneered by Colin Clark (1951), re-
preted to include the shadow value of searchers have estimated urban popula-
time, which turns out to be its dominant tion density functions for an enormous
component in modern developed na- range of places and times.9 In most
tions. (Deriving this shadow value en- of this work, a negative exponential
dogenously would require adding lei- function is assumed: D(x) = Doe-yx
sure and a time budget to the model.) where D(x) is population density at
Because this shadow value rises with in- distance x from the CBD and Do and y
come, so does marginal transport cost. are positive constants. The negative
If T'(x) is less elastic with respect to in- exponential function is convenient
come than is lot size L[.], equation (2) because it is easy to estimate after
predicts that rich households will have taking logarithms. The constant
flatter bid-rent functions than poor y=--D'/D is the proportional rate at
households and hence will locate more which population density falls with
peripherally. Whether this condition distance and is known as the density
holds for a typical U.S. city is under gradient. It is a useful index of popula-
some dispute (Wheaton 1977). tion centralization.
A more fundamental limitation is that Two of the strongest empirical regu-
the model is static. Two interpretations larities relating to urban spatial struc-
are possible, both unrealistic. One is ture can be concisely stated using the
that the model describes a stationary gradient as defined earlier. First, den-
state with durable housing, which a real sity declines with distance from the
city would approach asymptotically. The center-the density gradient is positive.
other is that the model describes short- Second, virtually all cities in the devel-
term equilibrium at a point in time, oped world and most others elsewhere
with perishable housing being continu- have decentralized over the last century
ously replaced. The trouble with both or more-the density gradient has de-
interpretations is that the typical life- clined over time. Table 1 provides just a
times of buildings greatly exceed the tiny sampling of empirical support for
time over which the model's parameters these assertions; corroborating evidence
can be expected to remain unchanged. is provided for Japan by Mills and Kat-
sutoshi Ohta (1976), for Latin America
3.2 Explanations of Post-war
by Gregory Ingram and Alan Carroll
Suburbanization
9 McDonald (1989) and Mills and Tan (1980)
What has the monocentric model en- provide good surveys of methodology and results,
abled us to say about the dramatic respectively.
Anas, Arnott, Small: Urban Spatial Structure 1437
gradient of y = 0.234 per mile.11 By over that period. Again, the simple
way of comparison, Edmonston (1975, model appears to be in the right ball
Table 5.5) and Mills and Ohta (1976) park.13
report average values of 0.38 and 0.12, However, there are some unsatisfac-
respectively, for various samples of U.S. tory aspects to the attempt to explain
cities in 1970. So our "guesstimate" of (5) density gradients in this way. Peter
is near the average of their estimates. Mieszkowski and Mills (1993) give a co-
How does (5) do in explaining decen- gent account. First, attempts to explain
tralization in U.S. cities? Comparisons differences in gradients across cities
across decades are tenuous, but we can and across times have not been very
very roughly ask whether changes in in- successful at isolating transport costs as
comes and transport costs could ac- an explanatory factor; this may be be-
count for the changes in y observed be- cause such costs are inaccurately meas-
tween 1950 and 1970. According to our ured and are strongly correlated with
model, from 1950 to 1970 the gradient income. Second, many of the density
should have fallen from 0.318 to 0.234 gradient estimates are based on just two
or by 26 percent.12 By comparison, Ed- observations, population in the central
monston reports a 41 percent decline in city and in the suburbs, along with the
the density gradient for a sample of cities area covered by the central city; but
this method appears to be highly inac-
curate in certain cases, particularly in
11 Housing costs were probably around 20% of
smaller cities. Third, because of lack of
after-tax income net of commuting cost, and land
costs about 20% of housing costs (Small 1981, land-use data at a fine scale, most of the
p.320), giving I-ct = 0.04. We assume that each empirical work uses gross density
commuter had nine hours daily for commuting (population divided by total land area)
plus work, and that income was taxed at a constant
rate T. We assume that the average one-way com- although the theory would be better
mute was 10 miles and took place at a speed of represented by net density (population
25 miles per hour, requiring 48 minutes of round divided by residential land area); unfor-
trip per day. Suppose that the only cost of
travel is time, valued at the after-tax wage rate. tunately, using gross density may drasti-
So total daily commuting cost averaged over x is cally overstate the size of density gradi-
(48/60)w, while marginal daily commuting cost ents because the outer reaches of an
(per mile of one way trip) is one-tenth as large. Fig-
uring in taxes: y = (1-t)9w, T = (1-t)(48/60)w, and urban area contain much higher propor-
T= (0.10)T. Hence T/y = 0.0889 and T'/y = 0.00889. tions of undeveloped land (Mieszkowski
(This implies that commuting time is, on average, and Barton Smith 1991). Finally, a
about 9% of the consumer's time endowment,
which is quite plausible.) Hence, y = (0.96/0.04) strong negative correlation is observed
(0.00889)/(1-0.0889) = 0.234. To be better aligned between the density gradient and total
with the empirical evidence (see Small 1992, pp. population, with larger cities more de-
44, 84), we would have to recognize that travel
time is valued at w/2, or somew at less than the centralized; whereas the monocentric
after-tax wage rate; and also that there is a vari- model predicts either no correlation or,
able money cost of automobile commuting equal in our version, a mild positive correla-
to about half the time cost. These corrections hap-
pen to approximatelycancel, so do not change the
gradient estimate by much.
12We assume that I-oc remained at 0.04 figures for the 25th and 75th percentiles.) Then,
throughout the period. LeRoy and Sonstelie the 1950 value of y predicted by equation (5) is
(1983, Table 4) estimate that real income rose ap- found by replacing the 1970 value of (T'/y) by
proximately 88% over those two decades while [(T'/1.43)/(y/1.88)]= 1.315 (T'/y), and similarly for
real marginal transport costs (including the value T/y. The result is y = (0.96/0.04) (1.315)(0.00889)/
of time) rose only 43%. (They give nominal fig- [1 - (1.315 times 0.0889)] = 0.318.
ures which we deflated by the CPI. We have esti- 13More refined predictions could be made us-
mated the mean by interpolating between their ing available extensions of the simple monocentric
Anas, Arnott. Small: Urban Svatial Structure 1439
tion.14 Mills and Tan (1980) suggest and less reliable than those for popula-
that the observed negative correlation, tion. The general conclusion from the
"though not a consequence of the empirical literature is that the density
model, is strongly suggested by com- gradient is larger for jobs than for
mon sense" because larger cities sup- households, but has been falling faster
port outlying employment subcenters (Mieszkowski and Mills 1993). This evi-
(p. 315). This of course is an appeal to dence weakly supports the hypothesis
forces outside the monocentric model. that jobs have been following people;
Probably the most serious deficiency but there are many other reasons for
of the monocentric model as an expla- jobs to have decentralized, as described
nation of urban decentralization is its in Section 2.
failure to account for the durability of Other possible explanations of popu-
housing. David Harrison and John Kain lation decentralization, especially in the
(1974) observed that cities tend to grow U.S., include variants of a "flight from
outward by adding rings of housing at a blight" hypothesis. First is deteriorating
density which reflects contemporaneous central housing quality, due to style or
economic conditions, with the density technological obsolescence combined
of earlier rings remaining unchanged with rational decisions by owners to run
due to housing durability. The same down housing quality. Second is the ex-
phenomenon is demonstrated by istence of racial preferences combined
Mieszkowski and Smith (1991), who with the tendency of poorer African-
show that the density of developed resi- Americans to live in central cities.
dential land (i.e. net density) in Hous- Third is negative neighborhood exter-
ton is approximately constant all the nalities associated with many poor
way to the outer edge of the metropoli- neighborhoods. Fourth is the working
tan area. A variety of dynlamic versions out of Tiebout mechanisms for provid-
of the monocentric model with durable ing local public goods (Charles Tiebout
housing has been constructed. In such 1956), resulting in better-off residents
models, the density gradient depends with high demands for local public
not only on the past time path of in- goods abandoning the central city and
come and transport costs, but also on excluding the poor from the suburbs
developers' expectations and the pros- through minimum lot-size zoning. All
pects for redevelopment. Explanations these explanations imply that the
for observed density gradients are cor- poor live near downtown and the rich
respondingly complex. are pushed or pulled out to the suburbs.
Employment density functions can be The implied effect on the value of den-
estimated in the same way as population sity gradients is, however, ambiguous.
density functions, although data on the
location of jobs are less readily available
4. The Polycentric City:
Empirical Descriptions
model. For example, accounting for income differ- We now turn to one of the most in-
ences would increase the predicted density gradi- teresting features of modern urban
ent if parameters are such that higher income peo-
ple live more peripherally, since they also choose landscapes-the tendency of economic
more land per dwelling for a given rent. activity to cluster in several interacting
14Looking at the outer boundary, rising popula- centers of activity. This section de-
tion does not change marginal transport cost but it
does increase total transport cost, hence lowering scribes empirical findings. The next re-
the denominator in (5) and causing y to rise. views theoretical models of polycentric-
1440 Journal of Economic Literature, Vol. XXXVI (September 1998)
ity. Throughout, we use "center" to mean (ii) The number of subcenters and their
either the main center or a subcenter. boundaries are quite sensitive to defini-
It is not hard to discover subcenters tion. Both the Los Angeles and the Chi-
lurking in spatial employment or popu- cago studies mentioned above find that
lation data for most large cities. Giu- with changes in density cutoffs, certain
liano and Small (1991) provide a review employment clusters could be viewed
of studies, and new ones are steadily ap- either as several large subcenters or as
pearing. Here we consider some tenta- one mega-center. In the Chicago data,
tive generalizations about the nature for example, the criteria just listed pro-
and role of subcenters in U.S. cities, for duce a huge subcenter near O'Hare Air-
which polycentricity has been examined port, with 420,000 employees,16 whereas
in greater detail than anywhere else. doubling the density cutoff breaks this
(i) Subcenters are prominent in both new subcenter into five smaller ones. The
and old cities. Evidence is emerging that Los Angeles case, discussed in the next
in each of the largest metropolitan areas subsection, shows even more sensitivity
in the United States, twenty or so sub- to subcenter definition.
centers can be identified using the crite- Such sensitivity is not surprising con-
ria described in Section 2 with minimum sidering the observations made in Sec-
gross density (D) of 10 employees per tion 2. The urban landscape is highly ir-
acre and minimum total employment (E) regular when viewed at a fine scale, and
of 10,000. Giuliano and Small (1991) how one averages these local irregulari-
find 29 such centers in Los Angeles in ties determines the look of the resulting
1980, and add three smaller outlying pattern. It may be that the patterns that
centers with prominent density peaks. occur at different distance scales are in-
Daniel McMillen and McDonald (1998b) fluenced by different types of agglom-
find 15 subcenters outside the city limits eration economies, each based on inter-
of Chicago meeting an identical crite- action mechanisms with particular
rion, using a combination of 1980 and requirements for spatial proximity. This
1990 data. Cervero and Wu (1997) find observation applies also to clustering at
22 such centers in the San Francisco Bay a regional scale such as the U.S. eastern
Area for 1990.
Each of these studies covers a Con- CMSA typically combines several adjacent areas
solidated Metropolitan Statistical Area formerly classified as SMSAs, most of which are
(CMSA), a census concept that is the now called Primary Metropolitan Statistical Areas
(PMSAs). For example, the New York-Northern
most inclusive of the various types of New Jersey-Long Island CMSA consists of 11
metropolitan areas defined in official PMSAs including New York (New York City plus
U.S. statistics. For example, San Fran- three adjacent counties), Nassau-Suffolk (two
counties constituting Long Island), and Newark
cisco's CMSA includes nine counties, (five counties in New Jersey). The Los Angeles-
from the Napa Valley wine country in Anaheim-Riverside CMSA consists of four
the north to San Jose and Silicon Valley PMSAs: Los Angeles County, Riverside and San
Bernardino Counties, Orange County, and Ven-
in the south.15 tura County. See U.S. Bureau of the Census
(1996, pp. 937-945). Because we are not inter-
15 Smaller urban regions, and a few large ones ested in municipal boundaries, in this essay we
like that surrounding Washington, D.C., are not generally designate a CMSA just by the name of
classified as CMSAs but rather as Metropolitan its largest city.
Statistical Areas (MSAs). Both CMSAs and MSAs 16 O'Hare airport is annexed to the City of Chi-
are collections of whole counties (except in New cago, despite its being surrounded entirely by sub-
England) that are highly integrated; the MSA is urbs. For this reason employment at the airport
closest to what before 1983 was defined as a Stan- itself is missing in these data, which cover only the
dard Metropolitan Statistical Area (SMSA). The suburbs.
Anas, Arnott, Small: Urban Spatial Structure 1441
(7) that replaces -7nXmn by 7n/Xmn over- (v) Subcenters have not eliminated the
comes this difficulty, and seems to fit importance of the main center. When-
even better.17 ever a downtown center and one or more
An intermediate case is the additive subcenters have been defined using the
form (8), used by Gordon, Richardson, same criteria, downtown has more total
and H.L. Wong (1986) and by Small and employment, higher employment den-
Shunfeng Song (1994). It is based on sity, and usually a larger statistical effect
the idea that the accessibility of a loca- on surrounding densities and land prices
tion is determined by the sum of expo- than does any subcenter. Because so
nentially declining influences from vari- many people believe that big-city down-
ous centers. Here every center has an towns are passe, it is worth reviewing
influence as in (7), but unlike in (7) a this evidence in some detail.
center's influence becomes negligible at Let us begin with Chicago. In ex-
large distances. Unfortunately, estima- plaining 1980 employment density pat-
tion of (8) requires nonlinear estimation terns outside the city limits of Chicago,
and often produces convergence prob- three large subcenters are found by
lemns. McDonald and Prather (1994) to have
Considerable success has been at- exerted an important influence, but
tained using these models to explain none has a t-statistic even one-fourth as
density and land-value patterns in Los large as does the CBD. In a remarkable
Angeles, Chicago, San Francisco, and a study of land values over a century and
few other places. The pioneering stud- a half, McMillen (1996) finds a clear
ies were Daniel Griffith (1981) and and marked land-value peak at the CBD
Gordon, Richardson and Wong (1986). for each of 10 different years from 1836
Small and Song (1994) are able to ex- to 1990, despite the steady rise in im-
plain roughly 50 to 75 percent of the portance of centers several miles to the
variance in employment or population northwest.
density across the entire Los Angeles In their study of San Francisco,
region using equation (8) with five cen- Cervero and Wu (1997) list the sizes
ters for 1970 and eight centers for 1980. of the 22 centers emerging from the
In all cases the special case of mono- Giuliano-Small criterion described
centricity is soundly rejected. The earlier. The largest and densest by far is
population density patterns fit well the one containing downtown San Fran-
even though population data were not cisco. This center accounts for 15 per-
used to determine the locations of the cent of the region's employment. Sili-
centers used in the specification. Small con Valley is the second largest center,
and Song also show that monocentric and the third (despite Gertrude Stein)
density estimates fit poorly, especially is centered in downtown Oakland.
in the later year, reinforcing the belief Now consider Los Angeles, famous
that polycentricity is an increasingly for its sprawl. Garreau (1991) names
prominent feature of the landscape. more actual plus emerging "edge cities"
there than in any other metropolitan
17 McDonald and Paul Prather (1994), McMil-
area in the United States.18 Yet of the
len and McDonald (1998a, b). A different modifi-
cation replaces the distances xmn to specific cen-
18 Garreau's definition of an edge city includes
ters n in (7) with distance to the nearest center,
the second nearest center, and so forth. Rena Sivi- five criteria: 5,000,000 square feet of office space;
tanidou (1996) uses this form successfully to ex- 600,000 square feet of retail space; a daily inflow
plain Los Angeles office and commercial land val- of commuters; a "local perception as a single end
ues. destination for mixed use"; and a location that was
Anas, Arnott, Small: Urban Spatial Structure 1443
terns discussed in the previous sec- be clothed and fed; depending on the
tions? To fix ideas, imagine first a structure of transport costs, some stages
"backyardeconomyy"with no patterns- of the production or processing of
just a uniform distribution of economic clothing and food are performed locally.
activity over space. This would be the If the cost of shipping unprocessed ore
equilibrium under certain restrictive as- is high, ore processing also occurs lo-
sumptions: land is homogeneous, pro- cally. A similar example is a town form-
duction of each good exhibits constant ing at a river rapids, since transship-
returns to scale, goods and people are ment activity creates a demand for
costly to transport, and there is no in- other goods causing local production-
teraction over space. To understand ag- early Montreal is one such case.
glomeration, we can ask, following Pa- Spatial inhomogeneities can create
pageorgiou and T. Smith (1983): What subcenters as well as central business
are some alternative assumptions that districts. For example, a CBD may form
would make this uniform distribution of on a harbor and a secondary employ-
activity unstable? The classical answers ment center may form at the site of a
are spatial inhomogeneities and internal river landing. The early model of White
scale economies in production. More (1976) stressed such causes of subcen-
recent answers involve scale economies ter formation.
external to firms, including those aris-
5.2. Internal Scale Economnies
ing from spatial contacts and imperfect
competition. When any of these alterna- The second classical explanation for
tive assumptions holds in an environ- agglomeration is economies of scale in
ment where transport and communica- some production process. An important
tion costs are not too high, spatial example is scale economies in the load-
agglomeration can occur. ing and unloading of goods. Even in the
In this section we explore each of absence of a natural advantage such as a
these alternative assumptions in turn. protected harbor, port activities would
We then consider dynamics, and finally tend to concentrate for this reason, a
examine some approaches to agglomera- tendency which helped produce the
tion from outside economics. port or railhead orientation of the nine-
teenth century city (Moses and William-
5.1 Spatial Inhomogeneities
son 1967; Mills 1972). The advent of
Locations differ in factors such as containerization has, if anything, inten-
soil, climate, mineral deposits, and ac- sified the economies of scale in port op-
cess to waterways. Given such sources erations; trucking, by contrast, appears
of Ricardian comparative advantage, to require only small-scale loading and
trade arises and production specializes unloading equipment, so its terminal
by location, unless transport costs are operations are widely dispersed along
prohibitively high in which case the major highways.
backyard economy persists but with Another source of scale economies is
backyards that differ from one another. the production of local public goods
Thus even with constant returns to (Joseph Stiglitz 1977), as suggested by
scale in production, spatial inhomo- many of the classic explanations for the
geneities can give rise to towns (Marcus historical origin of cities the city as
Berliant and Hideo Konishi 1996). Ani temple, citadel, capitol, marketplace,
example is a mineral deposit which at- granary, or theater. Their counterparts
tracts workers to a mine. Miners have to in modern cities include civic buildings,
1446 Journal of Economic Literature, Vol. XXXVI (September 1998)
vacancy is more likely to find a match their job sites-the backyard economy
with an unemployed worker when the again; if commuting costs are very low
labor market is larger; likewise, special- and the decay rate is small, agglomera-
ized capital that is unemployed due to a tion benefits dominate and firms cluster
firm's closing is more likely to be suc- around one location giving rise to a
cessfully redeployed the larger the monocentric city; and if commuting
number of other firms using similar costs are moderate and the decay rate is
types of capital (Robert Helsley and high (so that a firm benefits a lot from
William Strange 1991). Another type is nearby firms but not much from more
information exchange within or between distant firms) then equilibrium is poly-
industries, for example, learning about centric. This model produces multiple
the efficacy of new techniques by ob- equilibria-for example with one, three,
serving the successes and failures of or five centers-under the same set of
competitors. Yet another type derives parameter values, suggesting that a city's
from education: because labor special- structure at a point in time may be path-
ization encourages investment in human dependent even when the durability of
capital, larger cities have more edu- structuresis ignored. Also, the comparative
cated work forces which may in turn re- statics of this model are catastrophic-
sult in more experimentation, more in- i.e., the solution changes discontinu-
novation, greater adaptability, and ously as parameter values are varied.
improved management skills. What might lie behind a location po-
How do inter-firm externalities affect tential function? One possibility is sim-
spatial structure? We can learn a lot ply spatial contact. Consider, for exam-
just by specifying how their strength ple, a very basic fixed interaction model
varies with spatial proximity, even with- in the spirit of Robert Solow and Wil-
out describing their source. Using such liam Vickrey (1971) or E. Borukhov and
a "pure externality" approach, Fujita Oded Hochman (1977). The city's geog-
and Hideaki Ogawa (1982) consider a raphy is described by a finite space such
closed market economy on a line seg- as a line segment or a disc, with a geo-
ment with a fixed number of workers, metric center but no predetermined
each of whom consumes a single pro- economic center. The city is populated
duced good and a residential lot of fixed by homogeneous agents (either firms or
size. They assume an equal number of households but not both), each of whom
firms, each employing one worker and occupies a lot of unit size and interacts
occupying an industrial lot of fixed size. by traveling the same fixed number of
Workers commute to their jobs at a times to visit every other agent. These
constant cost per unit distance. Firms abstract interactions can be interpreted
benefit from proximity to other firms, as social contact, information acquisi-
as described by a location potential tion, search, or exchange.22
function in which the external produc-
tivity benefit conferred by one firm on 22 In actual cities, many such interactions are
another falls off with the distance be- face to face. Because formal contracting is costly,
tween them according to a negative ex- much contracting takes place informally; this re-
quires honest dealing, and honesty is communi-
ponential function with a fixed decay cated by body language and eye contact. The fact
rate. All agents are price takers. If com- that humans have developed unconscious signals
muting costs are very high, equilibrium of their intentions, as well as the ability to deci-
pher those signals, can be explained by theories of
entails a completely mixed land use pat- evolutionarily stable strategies as postulated by
tern with all workers living adjacent to John Maynard-Smith (1976). See also Robert
1448 Journal of Economic Literature, Vol. XXXVI (September 1998)
ties on one another. In aspatial contexts the benefits from the increased product
this can create critical-mass effects as in variety to be found there (which in their
some "big push" models of industrializa- model lowers search costs). Retailers, in
tion (Kevin Murphy, Andrei Schleifer turn, trade off the larger potential vol-
and Robert Vishny 1989). In spatial ume of customers at a center offering
contexts, imperfect competition can the advantages of product variety
cause agglomeration in an analogous against the lower degree of monopoly
way. Indeed, from Harold Hotelling power achieved there. This type of
(1929) on, one of the central issues ad- model leads one to expect more homo-
dressed by spatial competition theory is geneous products to be sold in smaller
the circumstances under which firms centers, and more differentiated prod-
have incentives to co-locate. Jean Gab- ucts, as well as big ticket items, to be
szewicz and Jacques-Francois Thisse sold in larger centers. The result is a
(1986) provide a review. hierarchy of centers analogous to the hi-
If economies of scale internal to the erarchy of cities in the central place
firm are large, the number of firms in theories of Christaller (1933) and Au-
the industry will be small. Given the re- gust Losch (1940). The pattern is fur-
sulting market power, determining ther complicated by complementarities
equilibrium location patterns entails that arise if consumers purchase multi-
game-theoretic considerations. In such ple goods on a single trip, giving retail-
spatial oligopoly models, firms may ers of different goods an added incen-
compete in price, product quality, prod- tive to locate in the same place (Robert
uct mix, and location, conferring market Bacon 1984).
advantages and disadvantages on each When economies of scale are less im-
other. Such firms are typically con- portant but product variety is still val-
ceived to be retailers or, more recently, ued, firms are more numerous and so
developers (J. Vernon Henderson and may engage in monopolistic competi-
Eric Slade 1993). Typically, product va- tion, in which strategic considerations
riety is assumed to be valued because of are absent. One particular model of
convex preferences, idiosyncratic pref- such a situation, by Avinash Dixit and
erences, or specialized intermediate Stiglitz (1977), has been used by others
goods. Such models easily produce ex- to derive results on agglomeration
ternalities: suppose, for example, that which can be interpreted as applying
expansion of the market occurs, causing either at an intraurban or regional scale
one more firm to enter and the accessi- (e.g. HeshamnAbdel-Rahman and Fujita
bility or variety of products to be 1990; Fujita and Tomoya Mori 1997). In
thereby enlarged; this creates additional two models by Krugman (1991b, 1993),
consumer surplus that is not fully cap- co-location of all of the monopolistically
tured by the entrant. competitive firms at a single point is a
Agglomeration may arise in situations stable outcome when transport costs are
of spatial oligopoly, depending on the low. Fujita (1988) has shown that intro-
balance of advantages and disadvan- ducing a land market into such models
tages of clustering. In the model of causes the agglomeration of firms to
Norbert Schulz and Konrad Stahl spread out as firms economize on rent,
(1996), shoppers trade off the higher and generates a variety of possible equi-
transport costs from traveling to a libria in which residential and commer-
larger activity center (which on average cial land uses can be either mixed or
is farther away from consumers) against segregated, monocentric or polycentric,
1450 Journal of Economic Literature, Vol. XXXVI (September 1998)
ing CBD can become too large because tals, dissipative structures, and self-or-
of coordination failures among potential ganization. All involve some form of
outmigrants. positive feedback (Brian Arthur 1990),
The process of "edge city" formation which in the urban growth context takes
envisioned by Henderson and Arindam the form of development at one location
Mitra (1996) is one way in which sub- somehow enhancing the development
centers can be sized and timed more ef- potential of nearby locations. This, of
ficiently. In their model firms decide course, is just another description of ag-
whether to relocate from the monocen- glomeration economies; the difference
ter to a new edge city. The essential in- is that this strain of literature has em-
novation is the introduction of a devel- phasized the dynamic analytics of such
oper who helps the migration process feedback mechanisms rather than their
along by internalizing some of the ex- economic underpinnings. In this sense
ternal benefits that migrants to the it resembles many macroeconomic mod-
edge city confer on each other. The de- els.
veloper is engaged in a game with the These models typically explore sys-
city government, which exercises influ- tems that are out of equilibrium, an ap-
ence over conditions in the original cen- proach now also established in evolu-
ter. Henderson and Mitra examine the tionary economics (Richard Nelson
strategic considerations facing the de- 1995) and one that is amply justified by
veloper, finding a rich set of possible the durability of urban structures. Un-
decisions concerning the location and fortunately, the models often lack
size for an edge city. The developer in- prices and so may neglect forces tend-
ternalizes some of the externalities, but ing toward the restoration of equilib-
introduces new ones due to strategic ef- rium. But are spatial interactions medi-
fects. The role of developers is only just ated through prices more important
beginning to receive attention in the than unpriced spatial influences and ex-
economic literature, but clearly it is ternalities? Since unpriced externalities
quite important in practice.26 probably play a dominant role in shap-
ing urban spatial structure, the chal-
5.6 Noneconomic Dynamic Models
lenge posed by the noneconomic mod-
The existence of multiple centers, the els cannot be easily dismissed. What
irregularity of spatial forms, and the un- follows is a sampler of these none-
predictability of how they evolve are conomic models from a quite eclectic
important features of the modern urban literature centered mostly in geography
landscape. Similar properties are also and regional science. We attempt to ex-
known to arise in a variety of nonlinear tract some basic insights which are use-
dynamic processes in chemistry, phys- ful to economic models.
ics, and biology. As a result, some of the Markovian models explain the transi-
more interesting infusions of ideas into tions of micro units from one state to
urban economics and urban geography another: development or redevelop-
are coming from those fields. In par- ment of a parcel of land, household mi-
ticular, urban structure is proving to be grations, and the birth or death of
a fertile application of generalized con- firms. Agglomeration effects imply that
cepts such as chaos, complexity, frac- individual transition probabilities de-
26 It is also important for equilibrium in Tiebout
pend on the number of actors in each
models of local public goods, as demonstrated by state, as in interactive Markov chain
Henderson (1985). models (John Conlisk 1992). A model
1452 Journal of Economic Literature, Vol. XXXVI (September 1998)
tion, in which the development prob- rators have put some of the same ideas
ability for a given site increases with the into dynamic models intended to de-
proximity of other occupied sites and scribe urban or regional growth pro-
decreases with distance from an exoge- cesses that may be far from equilib-
nous monocenter. Simulations yield rium. This work is part of a more
growth patterns that resemble, at least general movement, inspired by Ilya
impressionistically, the historical devel- Prigogine, to describe systems that
opment of Berlin from 1875 to 1944, maintain organized structure against
which especially in the later years the ravishes of entropy. Such systems
showed a high degree of irregularity. are called dissipative structures (G.
Perhaps the main advantage of such Nicholis and Prigogine 1977; John Fos-
models is the tools they offer for analyz- ter 1993).
ing irregularity-for example, the fit- Allen's models are based upon inter-
ting of power laws to the size distri- dependent growth equations for popula-
butions of local spatial fluctuations. tion and employment which incorporate
Per Bak and Kan Chen (1991) have both agglomeration economies and con-
shown that many dramatic physical phe- gestion diseconomies. For example, in
nomena, including avalanches and the model of Allen and M. Sanglier
earthquakes, occur when the dynamics (1981), employment S in a given region
of a system push it to an ordered state and sector obeys a dynamic equation in
that is just on the edge of breakdown. which dS/dt is proportional to S.(E-S),
Given such a state of self-organized where E is a measure of "potential em-
criticality, small fluctuations cause ployment demand." This potential de-
chain reactions whose sizes typically mand is in turn determined by other
obey a power-law distribution. Krugman equations in the system that account for
(1996) hints that the interactions among the location's relative attractiveness,
economic agents may produce similar crowding, and a rather arbitrary "natu-
states in cities, as well as in other eco- ral carrying capacity." Thus existing em-
nomic situations, and that this may ex- ployment attracts new employment, but
plain the prevalence of sudden transi- eventually the location becomes satu-
tions such as the extremely rapid rated. The authors create simulations in
growth of new edge cities. Extensions which random fluctuations cause the
of economic models that produce sud- spontaneous creation of centers, which
den growth, such as those of Krugman subsequently grow along a path resem-
(1996) and Anas (1992), could perhaps bling a logistic curve. Most simulations
produce temporary states of self-orga- lead to a stable but not necessarily
nized criticality with testable statistical unique steady state. Constraints such as
properties. zoning regulations, if added early in the
Regional scientists have long been in- simulation, can affect which of the pos-
terested in models in which the attrac- sible steady states occurs. This model
tiveness of a location, for example a and related ones have been calibrated
shopping center, is enhanced by large for a number of cities and regions in
size. As already discussed, such models Belgium, France, Senegal, and the
are capable of generating bifurcations, United States (Allen 1997).
in which small shifts of parameter val- Most of the noneconomic models de-
ues produce qualitatively different scribed here lack a price system and any
equilibrium configurations, some stable explicit description of rational economic
and some not. Peter Allen and collabo- decision-making. Furthermore, their
1454 Journal of Economic Literature, Vol. XXXVI (September 1998)
dynamic behavior is backward- rather tle faith in either the efficiency or the
than forward-looking. Thus, for all their equity of market-determined outcomes,
tantalizing success in portraying the and advocate detailed land use plan-
complexity in the dynamics of urban ning. To evaluate these conflicting
structure, they fail to incorporate eco- points of view we need to explore the
nomic explanations. Fortunately, they welfare economics of urban land use. In
tend to be based on the behavior of this section we attempt to show how
individual units and so are not funda- some of the prominent policy questions
mentally incompatible with economic can be illuminated, if not answered, by
reasoning. This suggests that advances building on the theoretical models and
might be achieved by some merging of empirical observations of the previous
modeling techniques. Either economic sections.
behavior might be inserted rigorously
6.1 Can Agglomeration Economies
into existing noneconomic models, or
Be Internalized?
attractive analytical features from
those models might be blended into ex- We have seen that although agglom-
isting models in urban economics. eration economies are the raison d'etre
An example of the first approach is by of most cities, their exact nature is in
Hsin-Ping Chen (1996), who shows that flux and only partially understood. Our
a rigorous microeconomic model can current understanding of them is based
generate macro-level equations like on a variety of factors including
those of Allen and Sanglier. Chen's Smithian specialization, idiosyncratic
model contains land and labor prices, matching, interaction, and innovation.
development and abandonment deci- Because these notions are broad ones,
sions, and other recognizable micro- no one has really succeeded in coming
economic constructs, all within a frame- to grips with how they affect the indus-
work of agglomeration economies and trial organization of the modern city.
congestion. She produces abstract simu- Why, if there are economies of scale, is
lations much like those of Allen and production not undertaken by a single
Sanglier, and in other work (Chen large firm? Why do some forms of inter-
1993) makes a plausible case for repli- action occur within firms, while others
cating the 1970-80 growth of the Los operate through the market and yet oth-
Angeles region with a calibrated version ers take place informally? And why do
of the model. some interactions appear to require
face-to-face contact while others can be
effected via telecommunication? The
6. The Welfare Economics of
answers given to these questions often
Urban Structure
refer to transactions costs, incomplete
In defense of the low-density devel- contracts, trust, and flexibility.
opment that increasingly characterizes Does the market-broadly speak-
modern cities, Gordon and Richardson ing-deal efficiently with agglomeration
(1997) have argued that the urban spa- economies? The standard answer is
tial structure generated by market negative. If scale economies are inter-
forces reflects the will of the people- nal to firms, then efficient pricing can-
or more precisely, that it is a successful not be supported by competition. If
and largely desirable adaptation to the they are external, firms will under-em-
forces of urban growth and congestion. ploy those business practices that con-
Planners, in contrast, typically have lit- tribute social value to their neighbors.
Anas, Arnott, Small: Urban Spatial Structure 1455
The standard argument, however, ne- selves, are formed from the tension be-
glects that efficiency could be achieved tween agglomerative and dispersive
by competition among private city-de- forces. Both sets of forces entail strong
velopers who would set up efficient cit- externalities-external economies pro-
ies, thereby internalizing the agglom- ducing the agglomerative tendencies,
eration economies. Each city would and congestion or nuisance externalities
operate at minimum average cost-a limiting the size and density of the ag-
point of locally constant returns to glomeration that is achieved. The first
scale-with increasing returns in the set of externalities is largely positive,
production of goods being balanced by suggesting an inadequate private incen-
decreasing returns in the production of tive to join an agglomeration and hence
accessible land, due to the higher costs excessive dispersion. The second set
of transport and communications in consists of negative externalities, so
larger cities. Under marginal-cost pric- may cause too many activities to locate
ing, the losses from production of goods close together. Since different exter-
would be just offset by profits on the nalities operate at different scales, it is
production of accessible land, which are quite possible for the spatial pattern of
manifested as land rents-a variant of economic activity to be too centralized
the Henry George Theorem (Richard at one scale (e.g. cities that are too big)
Arnott and Stiglitz 1979). When devel- and too dispersed at another (e.g. sub-
opers make decisions concerning the in- centers that are too small). To further
ternal structure of edge cities, they complicate matters, the externalities are
are to a limited extent playing this role. linked. For example, downtown conges-
We do not, however, observe developers tion, along with the excessive residen-
trading cities in a competitive market; tial decentralization caused by under-
so it is doubtful that agglomeration priced transport, may give rise to
economies can be fully internalized in excessive employment decentralization
this way. Government intervention can (because jobs follow households), which
help in principle, but until the sources may in turn spawn excessively large sec-
of market failure are better understood ondary agglomerations.
it risks making things worse instead of The two-location model of Anas
better-as has also been argued in the (1992), reviewed in the previous sec-
international trade context (Krugman tion, illustrates these problems in a dy-
1987). namic setting. As the population of the
first center grows, there comes a time
6.2 How Efficient Is Subcenter
when it is optimal for a mass of popula-
Formation?
tion to move to the second location.
We have seen how agglomeration Since, however, the social gains from
economies tend to create clusters of relocation exceed the private gains, un-
economic activity within a city and how der atomistic migration the second cen-
these clusters influence surrounding ter will not be established until prob-
residential densities. Given the rich na- ably much later. According to this
ture of interactions within urban areas, reasoning, some collective action is
such clusters play a variety of roles. needed not only to establish the second
What can we say about the optimality of center at the right time but also to pro-
the resulting pattern? tect it until it becomes stable and self-
Our theoretical review suggests that sustaining. In principle, a private devel-
urban subcenters, like cities them- oper has a profit incentive to form the
1456 Journal of Economic Literature, Vol. XXXVI (September 1998)
second center at the right time;27 but urban spatial structure is efficient. It is
in a more realistic model with multiple reassuring that the Invisible Hand can
locations, the strategic rivalry among work with respect to the location of
potential developers, each trying to cre- economic activities. Unfortunately, this
ate a subcenter, results in other ineffi- efficiency property is not very robust
ciencies (Henderson and Slade 1993). theoretically, and is of questionable
There may therefore be a role for gov- practical relevance because of the per-
ernment in assisting subcenter forma- vasiveness of externalities in actual
tion-for example by providing infra- cities. One of the most serious is traffic
structure, regulating or subsidizing congestion.
developers, or subsidizing firm location. The congestion externality arises be-
On a regional or national scale, analo- cause the user of a motor vehicle does
gous issues have been raised in the de- not pay for its marginal contribution to
bates over France's "poles de crois- congestion. Consequently, the private
sance," Britain's New Towns policy, and cost of travel during peak periods falls
policies of less developed nations to di- short of the social cost. Travel is misal-
vert growth away from their "primate" located across transport modes, routes,
cities which contain large percentages and times of the day, and overall travel
of the national urban populations. may be excessive too. As is well known,
The comparison between optimal and this externality can be internalized by
market-determined spatial structure is means of a congestion toll equal to the
further complicated by history depen- marginal congestion externality evalu-
dence. The most obvious source is the ated at the optimum. However, optimal
durability of structures and infrastruc- congestion tolls are charged nowhere
ture. But as we have seen, even in the and congested travel is underpriced al-
absence of durability one can have mul- most everywhere. Uncongested travel,
tiple stable equilibria, with some more by contrast, may be considerably over-
efficient than others and with history priced, especially in nations with high
determining which obtains. On balance, fuel taxes.
therefore, it appears formidably diffi- What does this imply about urban
cult to ascertain how the actual size dis- form? Even in today's complex urban
tribution and composition of subcenters structures, the most severe congestion
differs from the optimum under realis- continues to occur on radial travel to
tic situations. While there is certainly and from the central business district
scope for ameliorative government ac- (CBD), and it is here that underpricing
tion, a precise prescription of good is most severe. If urban structure is fun-
planning in this arena remains elusive. damentally shaped by commuting costs
to the CBD, as postulated by the mono-
6.3 Does Traffic Congestion Cause
centric model, then such underpricing
Excessive Decentralization?
causes the city to be more spread out
In the basic monocentric-city model, than is optimal. This excessive residen-
tial decentralization is compounded by
27 On a smaller scale, James Rauch (1993) shows
a less obvious effect: underpricing
how the developer of an industrial park, in which
there are pro uction complementaries between travel distorts land values in a way that
firms, can achieve efficiency by subsidizing the encourages planners to allocate too
first firms moving into the park in order to attract much downtown land to roads (Arnott
additional tenants. Shopping centers employ a
similar strategy by giving rental discounts to an- 1979). To see why, suppose the only
chor stores. cost associated with a road is the oppor-
Anas, Arnott, Small: Urban Spatial Structure 1457
tunity cost of the land it uses. Now let derpriced and hence is developed at in-
the planner employ the following "na- efficiently low density. So the resulting
ive" cost-benefit rule: at each location, land use pattern is likely to be ineffi-
expand the road until the incremental ciently dispersed (not clustered
travel-cost saving from further expan- enough). It is more difficult to say if the
sion equals the residential market value pattern is also inefficiently decentral-
of the incremental land required. How- ized (too spread out from the center)
ever, the market value of residential because the timing of polycentric devel-
land reflects only the private transport- opment depends on how the land devel-
cost savings from a more central loca- opment industry is organized. If the in-
tion, not the social savings which-be- dustry is dominated by a few large
cause of underpriced congestion-are developers, then timing is affected by
greater. The market therefore underval- strategic interdependence; whereas if
ues downtown residential land, so that' there are instead many small developers,
application of the naive rule results in timing is influenced by coordination fail-
too much land there being devoted to ure and the dynamics of herd behavior.
roads. Another way of viewing it is that Possible second-best policies to cor-
the naive rule ignores the contribution rect excessive decentralization, if such
to congestion of "induced traffic," i.e., is the case, include more sophisticated
traffic caused by land-use changes in- cost-benefit analysis of transport proj-
duced by the highway investment. ects, minimum-density controls, and
Wheaton (1978) has argued that such a greenbelts. In fact, policies in the
mechanism resulted in massive over- United States have worked in exactly
building of urban highways in the U.S. the opposite direction, as emphasized
during the 1950's and 1960's. by Anthony Downs (1992) and others.
This reasoning, of course, must be Subsidies for home ownership, subsi-
modified when one takes into account dized highway construction and mainte-
non-central employment. As congestion nance, and minimum-lot-size residential
builds near the city center, some cen- zoning are just some of the measures
trally located employers respond by which have increased decentralization,
moving out of the CBD and closer to even while keeping the poor excessively
their workers and customers, with ag- concentrated in the central cities. In re-
glomerative forces causing some of this sponse to the ongoing transformation in
employment to become clustered in urban form, the planning community
subcenters. As the metropolitan area has tended to advocate policies aimed
evolves from a monocentric to a dis- at reversing decentralization, reducing
persed or polycentric structure, average automobile use, and revitalizing the
travel times and congestion levels are downtown core-for example building
reduced. This phenomenon is empiri- mass transit facilities or downtown con-
cally documented by Gordon, Ajay Ku- vention centers. But because the pric-
mar, and Richardson (1989) and Gor- ing errors of the past have been cast in
don and Richardson (1994), and occurs brick and asphalt, such policies are very
in simulations based on the theoretical expensive and have limited effectiveness.
model of Anas and Kim (1996).
6.4 When Are Land- Use Controls
Clearly, however, the process of de-
Justified?
centralization does not occur efficiently
because the congestion externality re- Given the many externalities revealed
mains. Highly accessible land is still un- by our theoretical review, it is tempting
1458 Journal of Economic Literature, Vol. XXXVI (September 1998)
to conclude that only very comprehen- ations, greenbelts are likely to spawn
sive and detailed planning can over- exurban development further out,
come the resulting inefficiencies. Be- which raises another set of issues for
cause the externalities are so poorly growth management.
understood, however, attempted cures Third, consider urban sprawl, a pejo-
may well do more harm than the dis- rative term often used for leapfrogging
ease. The brief discussion below illus- in development. This appears ineffi-
trates the complexity of determining cient at first glance. But what some
one aspect of optimal policy: land-use planners see as haphazard development
planning. may well be the seeds of future agglom-
First, consider incompatible land erations, and the land left vacant can be
uses. Cities are awash in very localized developed later at higher density than is
externalities, from the smells from a justified today.
fish shop to the blockage of ocean views Another argument for greenbelts or
by neighbors' houses. Mills and Hamil- growth boundaries is maintenance of vi-
ton (1994, pp. 252-54) argue that they able central cities. Critics of current de-
are not significant, but that may be be- velopment patterns argue, with some
cause the worst have been eliminated justification, that misguided policies
by zoning. Because pricing solutions in have produced excessively decentral-
this context would be extremely cum- ized cities at great cost in duplicative
bersome, zoning is a potentially valu- infrastructure and with disastrous re-
able tool for dealing with incompatible sults for the poor who live in concentra-
land uses. However, it can easily be tions of blight (David Rusk 1993;
overdone; for example, the complete Downs 1994; Myron Orfield 1997).
separation of retail and residential land Some of these authors argue for growth
uses results in visual monotony and un- boundaries to force new development
necessary auto travel. into the central cities in hopes of revi-
Second, consider preservation of talizing them. But such gross restric-
open space. Greenbelts and urban parks tions may well have perverse distri-
are potentially valuable public goods, butional consequences: the prior
and government intervention is prob- owners of land within the boundary en-
ably the only viable way to ensure their joy windfall gains at the expense of nas-
provision. It is important to recognize, cent businesses and new home buyers,
however, that someone is implicitly while inner-city renters-who are dis-
bearing the cost of designating areas proportionately poor-must pay more
off-limits to development. The in- for housing.
creased scarcity of residential land Finally, consider exclusionary zoning.
induced by greenbelts drives up land Many suburban municipalities enforce
rents and hence housing rents. So the minimum-lot-size restrictions, largely in
bucolic landscapes surrounding London order to exclude lower-income resi-
and Paris arguably come at the cost of dents who would pay less in property
miserable and badly overcrowded taxes while receiving the full benefits of
neighborhoods for the poor. Where the local public goods. Such restrictions
such controls divert growth from the may also be designed to exclude unde-
entire metropolitan area, they may im- sired socioeconomic, racial, or ethnic
prove the local environment but against groups. Exclusionary zoning probably
this must be weighed the environmental adds considerably to decentralization as
cost of growth elsewhere. In other situ- well as fostering social stratification,
Anas, Arnott, Small: Urban Spatial Structure 1459
segregation in education, and racial di- pecially external scale economies. Cities
vision. By forcing the poor to live in teem with positive and negative exter-
central cities, it also limits their access nalities, all acting with different
to suburban blue-collar jobs, a phe- strengths, among different agents, at
nomenon known as spatial mismatch different distances. Some people need
(Kain 1968). These are all reasons why to interact frequently face-to-face;
higher levels of government might want others carry out routine actions
to encourage suburban municipalities to remotely via telecommunications but
be more receptive to high-density hous- must meet periodically to create and
ing targeted to lower-income residents. renew trust; still others learn crucial
information by overhearing conversa-
6.5 Summary:The Role of Government
tions at restaurants, bars, parties, or
Policy
meetings. Consumers want to purchase
As in so much of economic policy some goods often, other infrequently;
analysis, it is hard to make overall rec- some want to see and touch goods,
ommendations about the scope of gov- others to hear about them from a
ernment intervention. Theory provides friend; for some any variety will do, for
clear instances of market failure, others a specific variety is required.
against which must be balanced the The pedestrian and car traffic gener-
likelihood and severity of government ated by one firm as a side effect can
failure. An interesting object lesson is make or break another firm's business,
Paris, whose urban form has been as window shoppers stop at an intrigu-
strongly influenced by government in- ing display or as disreputable patrons
tervention to limit central building scare away a neighbor's potential work-
heights and to channel exurban devel- ers, residents, or customers. Together
opment towards planned satellite towns. these many interactions, helped by
The result is a city regarded by many as history and a good deal of chance,
extremely attractive and vital. Others produce the spatial structure that we
prefer the convenience, lower cost, and see. Is it any wonder that spatial
ease of interaction of Los Angeles, patterns are complex, that they
which Paris would probably come to re- occasionally display sudden change, or
semble absent government policy. that tractable models can capture only a
What seems clear to us is that cities portion of their rich variegation?
are complex entities in which market Agglomeration economies have re-
forces are both powerful and beneficial sisted attempts to fully understand their
in many ways, obvious and subtle. microfoundations. This is illustrated by
These market forces sometimes need to urban economists' lack of confidence in
be controlled or channeled, yet they forecasting the effects of the communi-
tend to find their own way of thwarting cations revolution on urban spatial
such restrictions. Whether a particular structure. On the theoretical side, we
government policy is enlightened inter- do not know the scale at which the vari-
vention or misguided meddling will in- ous forces work or what kinds of equi-
evitably be debated case by case. libria the simultaneous interaction of
many forces will produce; nor do we
have reliable models of dynamic growth
7. Conclusion
paths with random shocks. We also do
And so we see that cities are strongly not know which external economies will
shaped by agglomeration economies, es- be internalized through private initia-
1460 Journal of Economic Literature, Vol. XXXVI (September 1998)
tive. On the empirical side, despite the belts and create satellite cities. Our re-
increasing sophistication of studies re- view suggests that such policies can in
lating a firm's productivity to the size principle elicit more efficient growth
and industrial composition of the city in paths; but that serious undesirable side
which it is located, we do not really effects are likely. As for the city cen-
know the specific forces that produce ters, whether the desire to maintain
these relationships, nor just how they their special character can stave off the
depend on industry mix, industrial pol- forces of economic change depends
icy, local public goods, or zoning. both on politics and on the ultimate
Complicating matters even more are preferences of the citizenry.
the longevity of urban structures, in-
REFERENCES
cluding public infrastructure, and the
Abdel-Rahman, Hesham and Masahisa Fujita.
stability of certain equilibria even when 1990. "Product Variety, Marshallian Externali-
other equilibria exist that would make ties, and City Sizes," J. Reg. Sci., 30:2, pp. 165-
everyone happier. Urban structure locks 83.
in past forces that may have little bear- Allen, Peter M. 1997. Cities and Regions as Self-
Organizing Systems: Models of Complexity.
ing today. Precious little traffic now Amsterdam: Gordon and Breach Science Pub.
uses the locks on the Erie Canal that and M. Sanglier. 1981. "A Dynamic Model
are the namesake of Lockport, New of a Central Place System-II," Geographical
Analysis, 13:2, pp. 149-64.
York; yet that is where its downtown re- Alonso, William. 1964. Location and Land Use.
mains. Other downtowns may be over- Cambridge, MA: Harvard U. Press.
crowded because no developer has man- Anas, Alex. 1990. "Taste Heterogeneity and Urban
Spatial Structure: The Logit Model and Mono-
aged to assemble land or obtain zoning centric Theory Reconciled," J. Urban Econ.,
variances needed to establish a much- 28:3, pp. 318-35.
needed satellite center. . 1992. "On the Birth and Growth of Cities:
Laissez-Faire and Planning Compared," Reg.
We have seen that forces that are as- Sci. Urban Econ., 22:2, pp.243-58.
cendant throughout the world are pro- Anas, Alex and Ikki Kim. 1992. "Income Distri-
bution and the Residential Density Gradient,"
ducing decentralization and dispersion
J. Urban Econ., 31:2, pp. 164-80.
at a citywide scale, and agglomeration - . 1996. "GeneraI Equilibrium Models of
at a local scale. Will Paris and Tokyo, Polycentric Urban Land Use with Endogenous
Congestion and Job Agglomeration," J. Urban
then, go the way of Los Angeles? To a Econ., 40:2, pp. 232-56.
large extent they already have; in both, Arnott, Richard J. 1979. "Unpriced Transport
as in cities throughout the developed Congestion," J. Econ. Theory, 21:2, pp. 294-
world, automobile-age development has 316.
and Joseph E. Stiglitz. 1979. "Aggregate
created a vast periphery of residential Land Rents, Expenditure on Public Goods, and
suburbs with outlying commercial, of- Optimal City Size," Quart. J. Econ., 93:4, pp.
fice, and industrial centers. Apparently 471-500.
Arthur, W. Brian. 1989. "Competing Technologies,
these patterns are not just the product Increasing Returns, and Lock-In by Historical
of crazy Americans in love with their Events," Econ. J., 99:394, pp. 116-31.
cars. Yet Paris and Tokyo have each 1990. "Positive Feedbacks in the Econ-
omy," Sci. Amer., 262:2, pp. 92-99.
preserved a distinctive city center, in Bacon, Robert W. 1984. Consumer Spatial Behav-
part through strict zoning and by valu- ior. Oxford, UK: Clarendon Press.
ing historic preservation. In addition Bailey, Jeff and Calmetta Y. Coleman. 1996. "De-
spite Tough Years, Chicago Has Become a Nice
Paris, like other cities including London Place to Live," Wall Street Journal, Aug. 21,
and Seoul, has seen the shape of decen- 135:37, pp. 1, 6.
tralization and dispersion altered by Bak, Per and Kan Chen. 1991. "Self-Organized
Criticality," Sci. Amer., 264:1, pp. 46-53.
central-government policies tha*t zone Barrett, Paul. 1983. Th-e Automobile and Urban
large tracts of outlying land as green- Transit: The Formation of Public Policy in Chi-
Anas, Arnott, Small: Urban Spatial Structure 1461
cago, 1900-1930, Philadelphia: Temple U. Dixit, Avinash K. and Joseph E. Stiglitz. 1976.
Press. "Monopolistic Competition and Optimum Prod-
Batty, Michael and Paul Longley. 1994. Fractal uct Diversity," Amer. Econ. Rev., 67:3 , pp.
Cities: A Geometry of Form and Function. Lon- 297-308.
don: Academic Press. Downs, Anthony. 1992. Stuck in Traffic: Coping
Berliant, Marcus and Hideo Konishi. 1996. "The with Peak-Hour Traffic Congestion. Washing-
Endogenous Foundations of a City: Population ton, D.C.: Brookings Institution.
Agglomeration and Marketplaces in a Location- - . 1994. New Visions for Metropolitan Amer-
Specific Production Economy," working paper, ica. Washington, D.C. and Cambridge, MA:
U. Rochester. Brookings Institution and Lincoln Institute of
Borukhov, E. and Oded Hochman. 1977. "Opti- Land Policy.
mum and Market Equilibrium in a Model of a Eberts, Randall W. and Daniel P. McMillen.
City without a Precetermined Center," Envi- Forthcoming. "Agglomeration Economies and
ronment & Planning A, 9:8, pp. 849-56. Urban Public Infrastructure,"in: Handbook of
Brueckner, Jan. 1987. "The Structure of Urban Regional and Urban Economics, Volume 3: Ap-
Equilibria: A Unified Treatment of the Muth- plied Urban Economics, Paul Cheshire and Ed-
Mills Model," in Handbook of Regional and Ur- win S. Mills, eds. Amsterdam:North-Holland.
ban Economics, Vol. II: Urban Economics. Ed- Edmonston, Barry. 1975. Population Distribution
win S. Mills, ed. North-Holland, Amsterdam, in American Cities. Lexington, MA: D.C.
pp. 821-45. Heath.
Bunde, Armin and Shlomo Havlin. 1996. "Percola- Eldredge, Niles and Stephen Jay Gould. 1972.
tion I," in Fractals and Disordered Systems. "Punctuated Equilibria: An Alternative to
Armin Bunde and Shlomo Havlin, eds. Berlin: Phyletic Gradualism, " in Models in Paleobiol-
Springer-Verlag, pp. 59-113. ogy. T.J.M. Schopf, ed. San Francisco: Free-
Cervero, Robert and Kang-Li Wu. 1996. "Subcen- man, Cooper & Co., pp. 82-115.
tering and Commuting: Evidence from the San Ellison, Glenn and Glaeser, Edward L. 1997.
Francisco Bay Area, 1980-1990," Working "GeographicConcentration in U.S. Manufactur-
Paper, Inst. of Urban & Regional Development, ing Industries: A Dartboard Approach,"J. Polit.
U. C. Berkeley. Econ., 105:5, pp. 889-927.
. 1997. "Polycentrism, Commuting, and Fales, Raymond and Leon N. Moses. 1972. "Land
Residential Location in the San Francisco Bay Use Theory and the Spatial Structure of the
Area," Environment and Planning A, 29:5, pp. Nineteenth Century City," Papers & Proceed-
865-86. ings Reg. Sci. Association, 28, pp.49-80.
Chen, Hsin-Ping. 1993. Theoretical Derivation Field, Alexander J. 1992. "The Magnetic Tele-
and Simulation of a Nonlinear Dynamic Urban graph, Price and Quantity Data and the New
Growth Model. Ph.D. Dissertation, Dept. of Management of Capital,"J. Econ. History, 52:2,
Econ., U. California Irvine. pp.401-13.
. 1996. "The Simulation of a Proposed Non- Foster, John. 1993. "Economics and the Self-Orga-
linear Dynamic Urban Growth Mo el," Annals nisation Approach: Alfred Marshall Revisited,"
Reg. Sci., 30:3, pp. 305-19. Econ. J., 103:419, pp. 975-91.
Chinitz, Benjamin. 1960. Freight and the Me- Fotheringham, A. Stewart, Michael Batty, and
tropolis, Cambridge, MA: Harvard U. Press. Paul A. Longley. 1989. "Diffusion-Limited Ag-
--. 1961. "Contrasts in Agglomeration: New gregation and the Fractal Nature of Urban
York and Pittsburgh," Amer. Econ. Rev., Papers Growth," Papers & Proceedings Reg. Sci. Asso-
& Proceedings, 51:2, pp. 279-89. ciation, 67, pp. 55-69.
Christaller, Walter. 1933. Central Places in South- Frank, Robert H. 1988. Passions Within Reason:
ern Germany, C.W.Baskin. Trans. London: The Strategic Role of the Emotions. NY: Norton.
Prentice-Hall [1966]. Fujita, Masahisa. 1988. "A Monopolistic Competi-
Clark, Colin. 1951. "Urban Population Densities." tion Model of Spatial Agglomeration : Differen-
J. Royal Statistical Society: Series A, 114:4, pp. tiated Products Approach," Reg. Sci. & Urban
490-96. Econ., 18:1, pp.87-124.
-. 1967. Population Growth and Land Use, . 1989. Urban Economic Theory: Land Use
London: Macmillan. and City Size, Cam-bridge,UK: Cambridge U.
Conlisk, John. 1992. "Stability and Monotonicity Press.
for Interactive Markov Chains," J. Math. Sociol- - and Tomoya Mori. 1997. "Structural Sta-
ogy, 17:2-3, pp. 127-43. bility and Evolution of Urban Systems," Reg.
Craig, Steven G., Janet E. Kohlhase and Steven C. Sci. & Urban Econ., 27:4-5, pp.399-442.
Pitts. 1996. "The Impact of Land Use Restric- Fujita, Masahisa and Hideaki Ogawa. 1982. "Mul-
tions in a Multicentric City," working paper, U. tiple Equilibria and Structural Transition of
Houston. Non-monocentric Urban Configurations," Reg.
Cronon, William. 1991. Nature's Metropolis: Chi- Sci. & Urban Econ., 12:2, pp.161-96.
cago and the Great West, NY: Norton. Gabszewicz, Jean Jaskold and Jacques-Fran9ois
David, Paul A. 1985. "Clio and the Economics of Thisse. 1986. "Spatial Competition and the Lo-
QWERTY," Amer. Econ. Rev., 75:2, pp. 332-37. cation of Firms," in Location Theory, by Jean
1462 Journal of Economic Literature, Vol. XXXVI (September 1998)
Jaskold Gabszewicz, Jacques-Francois Thisse, Policentric City," Environ. & Planning A, 21:2,
Masahisa Fujita and Urs Schweizer. Vol. 5 of 221-32.
Fundamentals of Pure and Applied Economics Hepep.
sley, Robert W. and William C. Strange. 1991.
series. Chur, Switzerland: Harwood Academic "Agglomeration Economies and Urban Capital
Publishers, pp. 1-71. Markets,"J. Urban Econ., 29:1, pp. 96-112.
Gaspar, Jess and Edward L. Glaeser. 1998. "Infor- Henderson, J. Vernon. 1985. "The Tiebout Model:
mation Technology and the Future of Cities,"J. Bring Back the Entrepreneurs,"J. Polit. Econ.,
Urban Econ., 43:1, pp. 136-56. 93:2, pp. 248-64.
Garreau, Joel. 1991. Edge City: Life on the New and Arindam Mitra. 1996. "The New Ur-
Frontier. NY: Doubleday. ban Landscape: Developers and Edge Cities,"
Getis, Arthur. 1983. "Second-Order Analysis of Reg. Sci. & Urban Econ., 26:6, pp. 613-43.
Point Patterns: The Case of Chicago as a Multi- Henderson, J. Vernon and Eric Slade. 1993. "De-
Center Urban Region," Professional Geogra- velopment Games in Non-monocentric Cities,"
pher, 35:1, pp. 73-80. J. Urban Econ., 34:2, pp.207-29.
Giuliano, Genevieve and Kenneth A. Small. 1991. Hoover, Edgar M. 1948. The Location of Eco-
"Subcenters in the Los Angeles Region," Reg. nomic Activity. NY: McGraw-Hill.
Sci. & Urban Econ., 21:2, pp. 163-82. Hotelling, Harold. 1929. "Stability in Competi-
1993. "Is the journey to Work Explained tion," Econ. J., 39:1, pp. 41-57.
by Urban Structure?" Urban Studies, 30:9, pp. Ingram, Gregory K. and Alan Carroll. 1981. "The
1485-500. Spatial Structure of Latin American Cities," J.
Glaab, Charles N. and Theodore Brown. 1967. A Urban Econ., 9:2, pp. 257-73.
History of Urban America, London: Macmillan Jacobs, Jane. 1969. The Economy of Cities. NY:
Press. Random House.
Glaeser, Edward L. et al. 1992. "Growth in Cit- . 1984. Cities and the Wealth of Nations:
ies," J. Polit. Econ., 100:6, pp. 1126-52. Principles of Economic Life. NY: Random
Gordon, Peter, Ajay Kumar, and Harry W. House.
Richardson. 1989. "The Influence of Metropoli- Koopmans, Tjalling C., and Martin Beckmann.
tan Spatial Structure on Communting Time," J. 1957. "Assignment Problems and the Location
Urban Econ., 26:2, pp. 138- 151. of Economic Activities," Econometrica, 25:1,
Gordon, Peter and Harry W. Richardson. 1994. pp. 53-76.
"Congestion Trends in Metropolitan Areas," in Krugman, Paul. 1987. "Is Free Trade Passe?" J.
Curbing Gridlock: Peak-Period Fees to Relieve Econ. Perspectives, 1:2, pp. 131-44.
Traffic Congestion, Volume 2: Commissioned . 1991a. Geography and Trade. Cambridge,
Papers, Transportation Research Board Special MA: M.I.T. Press.
Report 242. Committee for Study on Urban . 1991b. "IncreasingReturns and Economic
Transportation Congestion Pricing, National Geography,"J.Polit. Econ., 99:3, pp. 483-99.
Research Council. Washington, D.C.: National . 1993. "First Nature, Second Nature and
Academy Press, pp. 1-31. Metropolitan Location," J. Reg. Sci., 33:2,
. 1996. "Beyond Polycentricity: The Dis- pp. 129-44
persed Metropolis, Los Angeles, 1970-1990," J. 1996. The Self-Organizing Economy.
Amer. Planning Association, 62:3, pp. 289-95. Cambridge, MA: Blackwell.
. 1997. "Are Compact Cities a Desirable LeRoy, Stephen F. and Jon Sonstelie. 1983. "Para-
Planning Goal?"J. Amer. Planning Association, dise Lost and Regained: TransportationInnova-
63:1, pp. 95-106. tion, Income, and Residential Location,"J. Ur-
and H.L. Wong. 1986. "The Distribution ban Econ., 13:1, pp. 67- 89.
of Population and Employment in a Polycentric Losch, August. 1940. The Economics of Location,
City: The Case of Los Angeles," Environ. & W.H. Woglom and W.F. Stolper. Trans. New
Planning A, 18:2, pp. 161-73. Haven: Yale U. Press [1954].
Griffith, Daniel A. 1981. "Evaluatingthe Transfor- Makse, Hernan A., Shlomo Havlin, and H. Eugene
mation from a Monocentric to a Polycentric Stanley. 1995. "Modelling Urban Growth Pat-
City," Professional Geographer, 33:2, pp. 189- terns," Nature, Oct., 377, pp. 608-12.
96. Maynard Smith, John. 1976. "Evolution and the
Hamilton, Bruce W. 1982. "WastefulCommuting," Theory of Games," Amer. Scientist, 64:1, pp.
J. Polit. Econ., 90:5, pp. 1035-53. 41-45.
Harris, B. and A.G. Wilson. 1978. "Equilibrium McDonald, John F. 1987. "The Identification of
Values and Dynamics of Attractiveness Terms Urban Employment Subcenters," J. Urban
in Production-Constrained Spatial-Interaction Econ., 21:2, pp. 242-58.
Models," Environ. & Planning A, 10:4, pp. 371- . 1989. "Econometric Studies of Urban
88. Population Density: A Survey,"J. Urban Econ.,
Harrison, David, and Kain, John F. 1974. "Cumu- 26:3, pp. 361-85.
lative Urban Growth and Urban Density Func- and Paul J. Prather. 1994. "SuburbanEm-
tions,"J. Urban Econ., 1:1, pp. 61-98. ployment Centres: The Case of Chicago," Ur-
Heikkila, E. et al. 1989. "What Happened to the ban Studies, 31:2, pp. 201-18.
CBD-Distance Gradient?: Land Values in a McMillen, Daniel P. 1996. "One Hundred Fifty
Anas, Arnott, Small: Urban Spatial Structure 1463
Years of Land Values in Chicago: A Nonpara- erarchy: Network Forms of Organization," in
metric Approach," J. Urban Econ., 40:1, pp. Res. Organizational Behavior, 12, pp. 295-336.
100-24. Rauch, James E. 1993. "Does History Matter Only
- and John F. McDonald. 1998a. "Population When It Matters Little? The Case of City-In-
Density in Suburban Chicago: A Bid-Rent Ap- dustry Location," Quart. J. Econ., 108:434, pp.
proach," Urban Studies, 55:7, pp. 1119- 30. 843-67.
. 1998b. "Suburban Subcenters and Em- Robinson, E.A.G. 1931. The Structure of Competi-
ployment Density in Metropolitan Chicago,"J. tive Industry. Cambridge, UK: Cambridge U.
Urban Econ., 43:2, pp. 157-80. Press.
Mieszkowski, Peter and Edwin S. Mills. 1993. Romer, Paul M. 1986. "Increasing Returns and
"The Causes of Metropolitan Suburbanization," Long-Run Growth," J. Polit. Econ., 94:5, pp.
J.Econ. Perspectives, 7:3, pp. 135-47. 1002-37.
and Barton Smith. 1991. "AnalyzingUrban Rosen, Kenneth T. and Mitchel Resnick. 1980.
Decentralization: The Case of Houston," Reg. "The Size Distribution of Cities: An Examina-
Sci. & Urban Econ., 21:2, pp. 183-99. tion of the Pareto Law and Primacy,"J. Urban
Mills, Edwin S. 1967. "An Aggregative Model of Econ., 8:2, pp. 165-86.
Resource Allocation in a Metropolitan Area," Rusk, David. 1993. Cities Without Suburbs. Balti-
Amer. Econ. Rev., 57, pp. 197-210. more: Johns Hopkins U. Press.
-. 1972. Studies in the Structure of the Ur- Saxenian, AnnaLee. 1994. Regional Advantage:
ban Economy, Baltimore:Johns Hopkins Press. Culture and Competition in Silicon Valley and
and Bruce W. Hamilton. 1994. Urban Eco- Route 128. Cambridge, MA: HarvardU. Press.
nomics, New York:Harper Collins. Schulz, Norbert and Konrad Stahl. 1996. "Do
Mills, Edwin S. and Katsutoshi Ohta. 1976. "Ur- Consumers Search for the Highest Price? Oli-
banization and Urban Problems," in Asia's gopoly Equilibrium and Monopoly Optimum in
New Giant: How the Japanese Economy Differentiated-Products Markets," Rand J.
Works. Hugh Patrick and Henry Rosovsky, eds. Econ., 27:3, pp. 542-62.
Washington: Brookings Institution, pp. 673- Scott, Allen J. 1988. Metropolis: From the Division
751. of Labor to Urban Form. Berkeley: U. Califor-
Mills, Edwin S. and Jee Peng Tan. 1980. "A Com- nia Press.
parison of Urban Population Density Functions . 1991. "Electronics Assembly Subcontract-
in Developed and Developing Countries," Ur- ing in Southern California: Production Pro-
ban Studies, 17:3, pp. 313-21. cesses, Employment, and Location," Growth
Mirrlees, James A. 1972. "The Optimum Town," and Change, 22:1, pp. 22- 35.
Swedish J. Econ., 74:1, pp. 114-35. Sivitanidou, Rena. 1996. "Do Office-Commercial
Moses, Leon and Harold F. Williamson, Jr. 1967. Firms Value Access to Service Employment
"The Location of Economic Activity in Cities," Centers? A Hedonic Value Analysiswithin Poly-
Amer. Econ. Rev., 57:2, pp. 211-22. centric Los Angeles," J. Urban Econ., 40:2, pp.
Murphy, Kevin M., Andrei Schleifer, and Robert 125-149.
W. Vishny. 1989. "Industrializationand the Big Small, Kenneth A. 1981. "A Comment on Gasoline
Push,"J. Polit. Econ., 97:5, pp. 1003-26. Prices and Urban Structure," J. Urban Econ.,
Muth, Richard F. 1969. Cities and Housing. Chi- 10:3, pp. 311-22.
cago: The U. of Chicago Press. - . 1992. Urban Transportation Economics,
Nelson, Richard R. 1995. "Recent Evolutionary Vol. 51 of Fundamentals of Pure and Applied
Theorizing About Economic Change," J. Econ. Economics series. Chur, Switzerland: Harwood
Lit., 33:1, pp. 48-90. Academic Publishers.
Nicholis, G., and Ilya Prigogine. 1977. Self-Orga- and Shunfeng Song. 1992. "'Wasteful'
nisation in Non-equilibrium Systems: From Dis- Commuting: A Resolution," J. Polit. Econ.,
sipative Structures to Order through Fluctua- 100:4, pp. 888-98.
tions. NY: John Wiley. . 1994. "Population and Employment Den-
6 hUallachain, Breandan. 1989. "Agglomerationof sities: Structure and Change," J. Urban Econ.,
Services in American Metropolitan Areas," 36:3, pp. 292-313.
Growth & Change, 20:3, pp. 34-49. Solow, Robert M. 1973. "On Equilibrium Models
Orfield, Myron. 1997. Metropolitics: A Regional of Urban Location," in Essays in Modern Eco-
Agenda for Communit and Stability. Washing- nomics. Michael Parkin with A.R. Nobay, eds.
ton, D.C. and Cambridge: Brookings Institution London: Longman, pp. 2-16.
and Lincoln Institute of Land Policy. and William S. Vickrey. 1971. "Land Use
Papageorgiou, Yorgos Y. and David Pines. 1989. in a Long Narrow City," J. Econ. Theory, 3:4,
"The Exponential Density Function: First Prin- pp. 430-47.
ciples, Comparative Statics, and Empirical Evi- Song, Shunfeng. 1994. "Modelling Worker Resi-
dence," J. Urban Econ., 26:2, pp. 264-68. dence Distribution in the Los Angeles Region,"
and T.R. Smith. 1983. "Agglomeration as Urban Studies, 31:9, pp. 1533-44.
Local Instability in Spatially Uniform Steady- Starrett, David A. 1974. "Principles of Optimal
States," Econometrica, 51:4, pp. 1109-19. Location in a Large Homogeneous Area," J.
Powell, Walter W. 1990. "Neither Market nor Hi- Econ. Theory, 9:4, pp. 418-48.
1464 Journal of Economic Literature, Vol. XXXVI (September 1998)
Stiglitz, Joseph E. 1977. "The Theory of Local Warner, Sam Bass Jr. 1962. Streetcar Suburbs:
Public Goods," in The Economics of Public Ser- The Process of Growth in Boston :1870-1900,
vices. Martin S. Feldstein and Robert P. Inman, Cambridge, MA: HarvardU. Press.
eds. London: Macmillan. pp. 274-333. Wheaton, William C. . 1974. "A Comparative Stat-
Thomas, R.W. 1981. "Point Pattern Analysis," in ics Analysis of Urban Spatial Structure," J.
Quantitative Geography: A British View. N. Econ. Theory, 9:2, pp. 223-37.
Wrigley and R.J. Bennett, eds. London: Rout- . 1977. "Income and Urban Residence: An
ledge & Kegan Paul, pp. 164-76. Analysis of Consumer Demand for Location,"
Tiebout, Charles M. 1956. "A Pure Theory of Local Amer. Econ. Rev., 67:4, pp. 620-31.
Expenditures,"J. Polit. Econ., 64:5, pp. 416-424. . 1978. "Price-Induced Distortions in Ur-
U.S. Bureau of the Census. 1996. Statistical Ab- ban Highway Investment," Bell J. Econ., 9:2,
stract of the United States: 1996. Washington, pp. 622-32.
D.C.: U.S. Government Printing Office. White, Michelle J. 1976. "Firm Suburbanization
Vernon, Raymond. 1960. Metropolis 1985, Cam- and Urban Subcenters," J. Urban Econ., 3:4
bridge, MA: HarvardU. Press. pp. 323-43.
von Thuinen,Johann Heinrich. 1826. Der Isolierte . 1988. "Location Choice and Commuting
Staat in Beziehung auf Landwirtschaft und Na- Behavior in Cities with Decentralized Employ-
tionalokonomie. Hamburg:F. Perthes. ment,"J. Urban Econ., 24:2, pp.129-52.