Fallah Et Al (2010)

doi:10.1111/j.1435-5957.2010.00330.
Urban sprawl and productivity: Evidence from US

metropolitan areas*
Belal N. Fallah1, Mark D. Partridge2, M. Rose Olfert3
1
Department of Administrative Sciences and Informatics, Palestine Polytechnic University, Hebron West Bank,
Palestine (e-mail: bfallah2000@yahoo.com)
2
Department of Agricultural, Environmental, and Development Economics, Ohio State University, Columbus,
Ohio,USA (e-mail: partridge.27@osu.edu)
3
Johnson-Shoyama Graduate School of Public Policy, University of Saskatchewan, Canada
(e-mail: rose.olfert@usask.ca)
Received: 4 September 2009 / Accepted: 30 July 2010
Abstract. This paper draws on urban agglomeration theories to empirically investigate the
relationship between the economic performance of US metropolitan areas and their respective
amounts of sprawl. To measure urban sprawl, we construct a distinctive measure that captures
the distribution of population density and land-use within metropolitan areas. Using both
ordinary least squares (OLS) and instrumental variables (IVs) approaches, we find that higher
levels of urban sprawl are negatively associated with average labour productivity. This pattern
holds even within given industries or within given occupational classifications.
JEL classification: R11, R12, R14
Key words: Sprawl, productivity, agglomeration
1 Introduction
The emergence of urban sprawl during the past century has ignited a growing debate regarding
its costs and benefits.1 Opponents of sprawl suggest that it generates adverse socio-economic
* We thank Infrastructure Canada for their support in funding part of this research under a grant entitled ‘Mapping
the rural-urban interface: Partnerships for sustainable infrastructure development’. We also thank the Canada Rural
Revitalization Foundation and the Federation of Canadian Municipalities for their support of this project, in particular
Robert Greenwood. An earlier version of this paper was presented at the North American Regional Science Association
Meetings in San Francisco, CA.
1
Among the underlying forces generating urban sprawl, urban economists have highlighted lower transport costs and
rising income as postulated by the monocentric model (Alonso 1964; Mills 1967). Other culprits include the advent of
the automobile (Glaeser and Khan 2004) and improved highways (Baum-Snow 2007). Others suggest that urban sprawl
is an outcome of bad government policies including over-reliance on property taxes (Brueckner and Kim 2003), political
fragmentation (Carruthers 2003), crime, and racial tension in central cities (Mills and Price 1984).
© 2010 the author(s). Papers in Regional Science © 2010 RSAI. Published by Blackwell Publishing, 9600 Garsington Road,
Oxford OX4 2DQ, UK and 350 Main Street, Malden MA 02148, USA.
Papers in Regional Science, Volume 90 Number 3 August 2011.

452 B.N. Fallah et al.
outcomes. These include increasing the concentration of poor households in low-income

central-city neighbourhoods, reducing tax revenue (Glaeser and Sacerdote 1999), which leads to
a reduced ability to provide public goods and services in central cities (Jordan et al. 1998).2
Moreover, unlike in dense settings, low concentration of individuals and firms along with the
possibility of lower frequency of face-to-face communication in sprawling cities is likely to
provide an environment with reduced opportunity for knowledge exchange activities, leading to
less innovation (McCann 2007, 2008; Partridge et al. 2008b, 2009a, 2009b).
Brueckner (2000) points to the importance of recognizing the potential negative externali-
ties associated with sprawl. For example, several observers have argued that sprawling cities
are inefficient in terms of providing infrastructure and other public services. In particular,
low density and expansion of a city’s spatial structure leads to a greater per-unit cost of
development (e.g., roadways, sewage, and electricity) (Knaap and Nelson 1992; Knaap
et al. 2001; Carruthers 2002; Carruthers and Ulfarsson 2008).3 Furthermore, urban sprawl
represents an important shift in land-use that greatly alters agricultural uses, potentially
creating environmental degradation, and changing the residence-work relationship through
longer commutes and traffic congestion. These potential negative impacts of urban sprawl
have led to exhortations to regional planners and local governments to curb sprawl and
promote compact cities, namely, ‘smart growth’.4 Advocates contend that ‘planned’ urban
development achieves better social outcomes, especially in terms of transportation/
environmental efficiencies and public infrastructure provision, compared with pure market
solutions.
Yet, the negative perception of urban sprawl has been challenged by others who argue that
it is a result of market processes fuelled by individual preferences rather than by bad urban
planning. Gordon and Richardson (1997) contend that urban sprawl is a natural product of
growing cities and that the arguments against urban sprawl do not justify promoting compact
cities.5 Urban sprawl has also been credited for creating opportunities for people to enjoy
consumption of land and housing at lower prices (Glaeser and Khan 2004), and through
self-sorting, a preferred level of local public services in sprawling areas (Ellickson 1971).
Advocates of individual choice and the free market solutions form the base of the opposition to
planning processes that would curb urban sprawl. However, bottom-up processes may also be
incorporated into urban planning (Batty 2007). Further there is evidence that sprawl and ‘smart
growth’ may co-exist (Torrens 2008), suggesting definitional and measurement problems with
these concepts.
A fundamentally important, and neglected, question regarding sprawl is how it affects
labour productivity. For example, from the spatial mismatch hypothesis, Wheeler (2001)
contends that lower commuting rates and search costs associated with dense settings may
enhance labour productivity via better matching between firms and workers. Likewise,
urban economists have long argued that a productivity premium of denser environments
2
Putnam (2000) contends that sprawling cities are associated with social isolation and deteriorating urban social
capital with concomitant adverse economic outcomes.Yet, the results in Brueckner and Largey (2006) do not support a
negative association between social interaction and urban sprawl.
3
Real Estate Research Corporation (RERC 1974) has estimated the direct cost of serving alternative development
patterns and find that sprawl-type development costs twice as much to serve as high-density development. Yet, the
relationship between urban form and cost of public service remains controversial as other researchers find less
supportive results. For a detailed discussion, see Carruthers and Ulfarsson (2003).
4
Muro and Puentes (2004) define smart growth as a set of planning goals and policies that limit outward expansion
and encourage higher density development by concentrating housing and employment.
5
Brueckner (2000) attributes the source of possible distortions to failure to account for: (i) the benefit of open space;
(ii) the social cost of congestion; and (iii) the infrastructure cost of sprawling areas.

Urban sprawl and productivity 453
originates from lower transportation costs of delivering goods and services. For example,
Ciccone and Hall (1996) show that if production technologies exhibit constant returns to
scale, but the marginal costs of transporting intermediate goods rise with distance, then the
ratio of output to input (productivity), rises with density. They find that doubling county
density translates into a 6 percent increase in labour productivity. Despite their rigorous work,
their findings cannot be generalized to the economic performance of sprawling cities. Spe-
cifically, land-use and average density can imply entirely different things – namely, cities with
the same average density can have very different land-use patterns in which one city could be
more uniformly distributed while the other city has large patches of very low density use if
offset by (say) a dense core. Moreover, land-use and sprawl are a metropolitan land-use
phenomenon and cannot be explored using data at the state level as in Ciccone and Hall
(1996).
Endogenous growth theories also emphasize the importance of dynamic externalities, par-
ticularly knowledge spillovers, in dense settings that enhance productivity. This is evident in
Lynch’s (1981) normative theory of urban form (see also Marshall 1890; Jacobs 1969; Lucas
1988; Glaeser 1998). Knowledge spillovers imply that improvements and innovations in one
firm or industry increase the productivity of other firms. Jaffe et al. (1993) use patent data to
confirm that knowledge spreads slowly and diminishes over distance, making geographical
concentration important to firms.
The above suggests that policies that encourage urban sprawl could diminish productivity
gains from dense agglomeration economies. However, this conclusion is highly contested. The
rapid advances in communication and information technologies may reduce the need for
face-to-face communication and thus, for a dense environment to generate productivity gains
(Partridge et al. 2008b, 2009a, 2009b). Lower transportation costs due to improved highways
have further diminished the cost-savings of agglomerated areas and may have been a significant
force favouring more sprawl (Glaeser and Khan 2004; Baum-Snow 2007). Glaeser and Khan
(2004) argue that well-functioning sprawling cities (e.g., Silicon Valley), may be conducive to
greater productivity.
In sum, the relationship between urban sprawl and labour productivity is ambiguous. We
might expect that more compact cities enhance productivity through greater social capital,
increased knowledge spillovers, better labour matching, and lower transportation costs. But
other factors, such as the benefits of Tiebout self-sorting, advancement in communication
technologies and modern transportation call into question this prediction. Thus, urban sprawl’s
effect on productivity is an empirical question.
The aim of this paper is to provide an empirical examination of this issue. Specifically, it
investigates the effect of urban sprawl on US metropolitan labour productivity. Our empirical
analysis draws upon panel and cross sectional data from US metropolitan statistical areas
(MSAs). To measure urban sprawl, we use disaggregate-level data to construct a distinctive
index that, unlike many existing measures, addresses the distribution of population density
within a given metropolitan area. Unlike most standard measures of sprawl such as overall
metropolitan area population density, our measure captures variations in intra-metropolitan area
density, and hence, is more of measure of land-use not just density. We then find robust evidence
that metropolitan sprawl is positively associated with lower average labour productivity. More-
over, the productivity-sprawl association holds even within industries and within occupations,
suggesting it is not a simple matter of firm or worker self-sorting that affects occupational or
industry composition.
The rest of the paper is organized as follows. Section 2 outlines the theoretical framework.
Section 3 discusses the empirical models and the measure of urban sprawl. Section 4 presents
the empirical results with sensitivity analysis following in Section 5. Section 6 presents our
concluding thoughts.

2 Theoretical framework
Productivity differences are modeled using an aggregate production function in the tradition of
Carlino (1992) and Sveikauskas (1975).6 We assume that all firms, across metropolitan areas, are
perfectly competitive and that their production is characterized by constant returns to scale.
Metropolitan productivity differences are generated due to externalities related to differences in
site-specific characteristics, which are embedded in the Hicks-neutral multiplier (Ai). The
aggregate production function in a given metropolitan area is assumed to take a Cobb Douglas
form, such that:
Qi = Ai ( K iα L1i −α ) (1)
Where Qi is output in MSA i, a and 1-a are the shares of the metropolitan capital stock (Ki) and
labour (Li), respectively.
Since data on capital stock is hard to obtain at a metropolitan level, we assume that the rental
price of capital, r, is equalized across all areas.7 Differentiating output with respect to capital
stock, the marginal product of capital (MPK) can be equated with rental price of capital, such
that:
MPK i = ∂Qi ∂ K i = Aiα K iα −1L1i −α = r (2)
Dividing (2) by Qi = Ai ( K iα L1i −α ) yields:
α Qi
Ki = (3)
r
Substituting (3) into (1) and collecting Qi terms yields:

α
⎛ α ⎞1−α L
1
Qi = Ai 1−α (4)
⎝r ⎠ i
The labour productivity expression can be obtained by dividing Equation (4) by Li, such that:
α
1
α 1−α
= Ai 1−α ⎛ ⎞
Qi
(5)
Li ⎝ r⎠
We allow other metropolitan characteristics related to productivity to be included in the

productivity shifter Ai such as industry mix, labour force characteristics, and other location-
specific characteristics:
⎡ ⎤
k
Ai = exp ⎢ϕ 0 + ∑ ϕ k x ki ⎥ (6)
⎣ k =1 ⎦
Substituting Equation (6) into Equation (5) and taking the logarithm yields the labour
productivity expression that is used as the basis for the empirical analysis:
6
Moomaw (1983), in a survey of cross-sectional analyses of spatial variations in manufacturing productivity,
suggests future improvements will include a more detailed specification of the role of agglomeration economies. Our
inclusion of a sprawl measure is an example of such a refinement.
7
This assumption is realistic since there is little variation in the capital rental price across the US.

⎛Q ⎞
k
log ⎜ i ⎟ = B0 + ∑ Bk x ki (7)
⎝ Li ⎠ k =1
Where B0 = [a/1 - a]log(a/r) + 1/(1 - a)j0 and Bk = 1/(1 - a) jk. As discussed below, xki
denotes the metropolitan characteristics, including sprawl, the main variable of interest.
Before proceeding to the empirical specification, it should be noted that examining labour
productivity is not the same thing as examining welfare or average firm profitability. For
example, assume that sprawl reduces labour productivity, but allows workers to live closer to
work. It is possible then that workers would accept lower wages due to shorter commutes, which
could mean higher firm profitability even though productivity is lower. Likewise, overall welfare
increases could result if household utility increases sufficiently.
3 Empirical model
3.1 Measuring urban sprawl
Urban sprawl can take different forms. It may involve low-density development, clustering of
population and economic activities at the urban fringe (edge cities), and separation of land-use
known as leapfrog development (Nechyba and Walsh 2004). Although there have been several
attempts to develop measures of sprawl, researchers have focused mainly on population density.8
Low density is admittedly an important aspect of urban sprawl though there are many other
dimensions as well.9 A variant, density gradients, has also been used to assess the rate of decline
in density from the Central Business District to the perimeter of the city (Mieszkowski and Mills
1993). Higher-density development is seen by many sprawl critics as an antidote to many
unwanted aspects of sprawling US urban structures. Moreover, relying on the traditional popu-
lation density measure is simple and avoids the difficulty and cost of obtaining data on alter-
native measures that require technologies such as geographical information systems (GIS) or
satellite imagery (Lopez and Hynes 2003; Burchfield et al. 2006).
One shortcoming of using an aggregate measure of density alone is that large areas of
counties that are contained in metropolitan areas are exurban and even rural. This leads to an
upward bias in measuring sprawl in the immediate vicinity of urban areas (Lang 2003). Partially
addressing this, other researchers (e.g., Fulton et al. 2001), use census urbanized areas as smaller
geographic bases.10 Yet, the latter measure excludes a relatively large area of ‘developed’ land at
the urban fringe, leading to a downward bias in the sprawl measure (Cutsinger et al. 2005). For
these reasons, an improvement would be a more disaggregate measure that avoids problems of
8
Glaeser and Khan (2004) provide evidence that job and population sprawl in US metropolitan areas go hand in
hand. Measuring sprawl using distance in miles from central cities for the major 150 metropolitan areas, they find that
the correlation between the median person’s distance and median worker’s distance is 0.89. Moreover, when measuring
urban sprawl using density, they find that the correlation between metropolitan average population density and average
employment density is 0.77.
9
Galster et al. (2000) offer a definition that describes eight dimensions of sprawl: density, continuity, concentration,
compactness, centrality, nuclearity, diversity and proximity. They subsequently condense these to density, concentration,
clustering, centrality, nuclearity and proximity in the ranking of 13 urban areas. Torrens (2008) suggests that the best
way to measure sprawl may be through an examination of a multitude of sprawl symptoms to determine which are the
key drivers and where there may be codependencies. These authors focus on offering complete definitions and
descriptions of what is meant by sprawl. It is not likely that a single measure will capture all dimensions. Our measure
is constructed to include consideration of variations in density, concentration, and perhaps also an indicator of centrality.
For example, a metropolitan area with a relatively high average density will be shown by our measure to have
considerable sprawl if their land-use pattern is one where a significant share of their population lives in low-density
developments.
10
A census urbanized area is densely settled territory that consist of core census block groups or blocks that have a
population density of at least 1,000 people per square mile and surrounding census blocks that have an overall density
of at least 500 people per square mile. See below for the definition of census block group.

either focusing too narrowly on the most highly urbanized areas or being ‘contaminated’ by
including large sections of rural areas in the outer parts of metropolitan areas.
Another important concern of using aggregate population density measures is that by not
incorporating the distributional aspect of metropolitan population, it is assumed that two metro-
politan areas with the same population density are equivalent even if they possess very different
land-use patterns. Aggregate measures do not distinguish between metropolitan areas where
population is scattered, evenly distributed, or highly concentrated (Lopez and Hynes 2003). In
particular, average density could miss factors such as infill land, leapfrog development, and loss
of prime farmland and open space to low density development. In this paper, we develop a sprawl
index that incorporates the density distribution aspect of sprawl using data at a disaggregated
scale, specifically, census block groups. A block group is an area comprising a group of co-located
census blocks that contain between 600–3,000 people, with a typical size of 1,500. Using block
group data derived from the Geolytics database (www.geolytics.com), allows us to access
information about the intra-metropolitan land-use pattern including adjustments for rural spaces.
This ensures that fine differences in the distribution of population density within metropolitan
areas can be captured, information that would be lost with an aggregate measure such as
population density.
Our basic sprawl measure is then defined as:
Sprawl = (( L% − H %) + 1) ∗ 0.5 (8)
Where L% is the share of metropolitan population living in a block group with density below the
overall US metropolitan block group median and H% is the share of metropolitan population
living in a block group with density above the overall US metropolitan block group median.
The sprawl measure in Equation (8) is an index that ranges between 0 and 1; values closer
to 1 represent greater sprawl. For each metropolitan area, block groups are sorted and aggre-
gated into high density (above the US metropolitan block group median), and low density
(below the US metropolitan block group median). A metropolitan area with a high percentage of
its population living in block groups with density below the median suggests more sprawl. To
account for ‘rural clusters’ in metropolitan areas, block groups with density below 200 persons
per square mile are excluded. This cut off corresponds to one residential unit per eight acres -
about the lowest residential density of many new exurban housing developments.
Illustrating that our sprawl measure is distinct from aggregate metropolitan population
density, the correlation between the two measures is only -0.42. Further analysis suggested that
our sprawl measure is only similar to the traditional density measure in the least sprawled (most
dense) metropolitan areas. Since we are examining the role of land-use and sprawl, the fact that
our sprawl measure provides different information in the most sprawled areas strikes us as
particularly appealing. To illustrate this pattern, Table 1 reports three groupings of 12 metro-
politan areas denoted by having the most sprawl, medium sprawl, and the least sprawl based on
our sprawl measure, as well as the metropolitan area’s corresponding rank based on population
density. In the latter case, the least-dense metropolitan area receives a value of 1 and the
most-dense metropolitan area receives the highest value of 363. The table illustrates that there
is relatively little correspondence between our sprawl measure’s ranking and population den-
sity’s ranking for high and medium-sprawl metropolitan areas, but there is close association for
the least sprawled metropolitan areas.
Lopez and Hynes (2003) use a similar measure of sprawl, but employ fixed density cut offs
(high density corresponds to greater than 3,500 persons per square mile, while low density lies
between 200 and 3,500 persons per square mile). The main concern with their choice of high and
low density cut off is that it is somewhat arbitrary and is not based on a widely recognized
benchmark such as the mean or median, though we consider their measure in sensitivity analysis
as well.

Table 1. Sprawl index compared to population density ranking
Top 12 sprawling metropolitan areas (1) (2) (3)

Sprawl index Sprawl rank Inverse density rank
(density in reverse order)
Barnstable Town, MA 0.9497 1 13

Sebastian-Vero Beach, FL 0.9232 2 113
Punta Gorda, FL 0.9041 3 80
Panama City-Lynn Haven, FL 0.8868 4 171
Spartanburg, SC 0.8767 5 12
Pensacola-Ferry Pass-Brent, FL 0.8722 6 149
Burlington, NC 0.8657 7 154
Fayetteville, NC 0.8560 8 127
Lakeland, FL 0.8538 9 82
Chattanooga, TN-GA 0.8478 10 50
Hickory-Lenoir-Morganton, NC 0.8404 11 1
Fort Walton Beach-Crestview-Destin, FL 0.8402 12 237
Middle 12 sprawling metropolitan areas
Boise City-Nampa, ID 0.6534 176 254

Rapid City, SD 0.6528 177 88
Terre Haute, IN 0.6504 178 133
Owensboro, KY 0.6496 179 188
Chico, CA 0.6473 180 202
Cumberland, MD-WV 0.6437 181 26
Flint, MI 0.6434 182 166
Lebanon, PA 0.6432 183 69
Kansas City, MO-KS 0.6429 184 278
Atlantic City, NJ 0.6415 185 148
Waterloo-Cedar Falls, IA 0.6411 186 108
Pittsfield, MA 0.6405 187 128
12 least sprawling metropolitan areas
Miami-Fort Lauderdale-Pompano-Beach, FL 0.3405 352 358

Stockton, CA 0.3394 353 349
Chicago-Naperville-Joliet, IL-IN-WI 0.3329 354 353
El Paso, TX 0.3315 355 354
San Diego-Carlsbad-San Marcos, CA 0.3176 356 356
Honolulu, HI 0.3170 357 357
New Orleans-Metairie-Kenner, LA 0.3104 358 347
Laredo, TX 0.2650 359 348
New York-Northern New Jersey-Long-Island, NY-NJ-PA 0.2479 360 361
San Francisco-Oakland-Fremont, CA 0.2313 361 359
San Jose-Sunnyvale-Santa Clara, CA 0.2058 362 360
Los Angeles-Long Beach-Santa Ana, CA 0.1630 363 355
Notes: For the 363 metropolitan areas defined for the 2000 census, column (1) shows the value for our sprawl index;
column (2) shows the sprawl ranking using our sprawl index; and column (3) shows the inverse ranking of population
density to run from most to least sprawl, namely, the MSA with the lowest population density receives a value of 1 and
the one with the highest population density receives a value of 363.
3.2 Empirical specification
To examine the effect of urban sprawl on labour productivity, we first employ cross-sectional
reduced-form models. The unit of observation is metropolitan statistical areas (MSAs),

producing a total sample of 357 MSAs.11,12 The following empirical estimation issues are
addressed. First, all the explanatory variables are measured at the beginning of the period
(1990), to mitigate the direct simultaneity between the dependent and independent variables. In
the sensitivity analysis section, we further investigate whether the results are affected by
statistical endogeneity using a variety of instrumental variable approaches. Previewing those
results, endogeneity does not appear to affect the results.
Second, the base model adds a number of control variables, including city size, exogenous
amenities, and other economic variables. This reduces omitted-variable bias and also ensures
that we are not confounding other agglomeration effects with the influence of urban sprawl. The
stochastic version of Equation (7) can then be written as:
⎛Q ⎞
Ln ⎜ i ⎟ = β0 + β1metropolitan _ sprawli ,1990 + β2 popi ,1990 + β3 popi2,1990 +
⎝ Li ⎠ i ,2001 (9)
β4 educi,1990 + β5 Natural _ amenityi + β6indust _ sharei1990 + vi + ei
The dependent variable is a measure of labour productivity that is derived by taking the log of
metropolitan gross domestic product (GDP) divided by total metropolitan employment, mea-
sured in 2001. Metropolitan GDP data is collected from the US Bureau of Economic Analysis
(www.bea.gov). Total metropolitan employment includes both full and part-time employment,
which are collected from the US Bureau of Economic Analysis regional economic information
system (REIS).13
It is regularly argued that the use of GDP to measure labour productivity is superior to other
measures of output that use the concept of value added or total value of production (Bernard and
Jones 1996; Rigby and Essletzbichler 2002). Ciccone and Hall (1996) argue that the findings of
such studies “are seriously flawed by their reliance on unsatisfactory measures of output from
the Census of Manufactures”. Nonetheless, labour productivity has also been measured using
average wages, on the basis that higher labour productivity should be reflected in higher wages
(Glaeser and Maré 2001; Wheaton and Lewis 2002). In the sensitivity analysis section, the base
model is re-estimated using average metropolitan wages to assess the robustness of our findings,
for which we find that the results are robust to different measures of productivity.
Metropolitan_sprawli is specified in expression (8). To disentangle sprawl (and population
distribution) effects from more standard agglomeration and urbanization effects, total metro-
politan population (popi) is included as an additional variable. Larger cities are expected to have
higher labour productivity through advantages such as labour pooling and the availability of
intermediate goods and services (Glaeser 1998; Quigley 1998). Yet, crossing a specific metro-
politan threshold size might raise congestion costs, which at some point could overwhelm
favourable agglomeration effects. This is particularly true when increases in metropolitan size
cannot be supported by existing levels of public goods, including infrastructure and accessible
transportation networks. The squared-term of total metropolitan population (pop2i) is added to
allow for the effects of congestion costs or diseconomies.
A larger stock of human capital raises labour productivity by exposing workers to more
information and allowing them to learn and exchange ideas more quickly. Human capital
accumulation is accounted for by including the share of metropolitan population above 25 years
11
MSAs are defined by the US Office of Management and Budget (OMB). An MSA includes one or more counties,
which consist of a core urban area of at least 50,000 people, plus any adjacent counties that have a high degree of
integration with the urban core. We employ the June 2003 MSA definitions based on 2000 census data. The listing of
MSAs and their county components can be accessed from: URL: http://www.census.gov/population/estimates/metro-
city/03msa.txt.
12
In other reported regressions below, the number of observation varies across regressions due to missing data. The
number of observations is reported in the tables.
13
Metropolitan GDP estimates are not available prior to 2001.

old that have: (i) high school degree; (ii) some college with no degree: (iii) associate degree; and
(iv) at least a bachelor degree. The estimates of the impact of these categories are relative to the
omitted category, those who did not complete high school. Education attainment data is derived
from the Geolytics data base (www.geolytics.com).
Industry composition could also influence average labour productivity. This effect is cap-
tured by including the employment shares of 15 major industries (indust_sharei) using data from
the US Bureau of Economic Analysis REIS database. The capital stock is an influential factor
for labour productivity. We do not include it explicitly because of our assumption that capital
returns are equalized across MSAs (due to the lack of data). However, our industry employment-
share variables reflect capital differences across industries. An index of 1 to 7 constructed by the
US Department of Agriculture (USDA), is used to proxy for the availability of natural_ameni-
tiesi.14 Natural amenities enhance labour productivity in a given MSA if they are associated with
attracting high-skilled labour or high productivity firms (Partridge et al. 2008a). Finally, state
dummies (vi) are included to pick up effects that are common across all metropolitan areas
within the same state, including cross industry differences in the capital intensity.15 Because
some metropolitan areas extend across state boundaries, each metropolitan area is assigned to
the state where the majority of the metropolitan population resides.
One potential concern is that the error terms could be spatially correlated. To correct for this
possible bias, the base model is estimated assuming that the residuals are correlated within
geographical clusters (Bureau of Economic Analysis [BEA] economic areas), but uncorrelated
across clusters.16 The advantage of using the clustering approach is that it does not impose
restrictions on the spatial correlation of the residuals within clusters. This is unlike other spatial
econometric models that use more restrictive assumptions, such as distance or adjacency weight
matrices.
4 Empirical results
4.1 Base model
Our discussion of results will emphasize the relationship between metropolitan sprawl and
average labour productivity. This section first discusses the base model results from Equation
(9). Section 5 presents sensitivity analyses to test the robustness of the findings. The base results
reported in column 1, Table 2, reveal that, ceteris paribus, a higher level of urban sprawl is
associated with lower average metropolitan labour productivity. The coefficient of metropoli-
tan_sprawli is negative and highly significant at the 1 percent level. This finding suggests that a
one standard deviation decrease in the level of metropolitan sprawl is associated with a 2.5
percentage point increase in labour productivity, suggesting that denser urban settlement pat-
terns are on average associated with higher productivity.
Some other results are worth noting. First, the coefficient of popi is positive and highly
significant at the 1 percent level, illustrating how population scale and the sprawl measure are
14
The amenity index combines six measures of natural amenities: warm winter, winter sun, temperate summer, low
summer humidity, topographic variation, and water area. A higher index value reflects more amenities. Since USDA
produces this data at the county level, the index is averaged over all counties belonging to the same MSA. For more
discussion of the role of natural amenities in US growth, see Partridge (2010).
15
When controlling for state fixed effects, the coefficients of the other explanatory variables reflect the within-state
variation of those explanatory variables on the dependent variable.
16
The BEA economic areas include one or more economic nodes that are regional centres of economic activities
along with the surrounding counties with economic links to the regional centre(s); in the sample, there are 161 BEA
economic areas. For more information on the definition of BEA economic areas, see URL: http://www.bea.gov/regional/
docs/econlist.cfm. The stata cluster command is used for the estimation.

Table 2. Metropolitan sprawl-average labour productivity, base model*
Variable -1- -2- -3-
OLS 2SLS 2SLS (IV: 1950 density

(IV: 1940 density) and non-white share)
1990 metropolitan sprawl -0.203 -0.897 -0.997

(-3.09) (-2.16) (-2.26)
1990 total metropolitan population 4.31E-08 1.22E-08 7.76E-09
(3.78) (0.48) (0.30)
1990 total population_squared -2.49E-15 -1.38E-15 -1.22E-15
(-3.42) (-1.13) (-0.96)
Natural amenity scale -0.017 -0.025 -0.027
(-1.58) (-2.03) (-1.93)
1990 share of at least bachelor degree 1.996 2.343 2.394
(7.25) (6.93) (6.87)
1990 share of associate degree -0.428 0.134 0.214
(-0.81) (0.17) (0.24)
1990 share of some degree 0.15 0.48 0.53
(0.60) (1.26) (1.28)
1990 share of high school 0.593 0.920 0.960
(2.88) (2.75) (2.70)
1990 shares of Industry employment Y Y Y
State dummiesa Y Y Y
Constant -3.44 -2.76 -2.66
(-12.35) (-5.03) (-4.73)
R-square 0.84 0.76 0.74

N 357 357 357
First stage: F-statb 11.02 10.95
Weak Inst. Wald Fc 64.42 140.78
Hausman test: p-vald 0.47 0.20
Notes: *Robust (spatially clustered) t-statistics are in parentheses. In calculating the robust t-statistics, the clusters are
formed based on BEA economic areas, which are defined as the relevant regional markets surrounding metropolitan or
micropolitan statistical areas. See: URL: http://www.bea.doc.gov/bea/regional/docs/econlist.cfm.
a
Metropolitan areas that cross state boundaries are assigned to the state in which the majority of metropolitan
population resides.
b
The F-test statistic on the identifying instrument (either 1940 population density or 1950 population density and 1950
share of non-white population) in the first-stage model. Values above 10 suggest that the instrument is strong.
c
Exceeds Stock and Yogo (2005) critical values for 10% maximal IV size distortion indicating that the IVs are strong.
d
Hausman test for the null hypothesis that the OLS results does not suffer from (statistical) endogeneity bias. Higher
p-values suggest that the null hypothesis cannot be rejected, suggesting that OLS is the preferred estimator.
producing separate effects. Yet, the coefficient of pop2i is negative and highly significant at the
1 percent level, suggesting adverse congestion effects on productivity. Using the linear and
quadratic population terms, the estimated population size that maximizes metropolitan labour
productivity is 8.7 million, which is near the upper range of the size of US metropolitan areas.
Finally, as expected, average labour productivity is positively related to higher population shares
having completed high school and to having at least a bachelor degree. The coefficients of the
these variables are highly significant at the 1 percent level.
4.2 Productivity: Linkages across metropolitan size
To investigate whether metropolitan population is an intervening factor that shapes the impact
of sprawl on labour productivity, the sample is divided into small MSAs (those below the

Table 3. Metropolitan sprawl-labour productivity across different metropolitan size*
Variables -1- -2-
Smaller MSAsa Larger MSAs
1990 metropolitan sprawl -0.31 -0.197

(-2.94) (-1.95)
Natural amenity scale -0.042 0.002
(-2.41) (0.09)
1990 share of at least bachelor degree 1.73 2.31
(3.82) (6.15)
1990 share of associate degree 0.014 -0.002
(1.25) (-0.23)
1990 share of some degree 0.004 -0.004
(0.72) (-0.9)
1990 share of high school 0.007 0.005
(2.04) (1.2)
1990 shares of Industry Employment Y Y
State dummiesb Y Y
Constant -3.44 -3.69
(-8.29) (-7.35)
R-square 0.72 0.87

N 179 180
micropolitan statistical areas. See: URL: http://www.bea.doc.gov/bea/regional/docs/econlist.cfm.
a
Smaller MSAs are those below the median metropolitan size of 190,352 population; larger MSAs are those above the
median.
b
population resides.
1990 median metropolitan size of 190,352 and large MSAs above the 1990 median). The
results reported in Table 3 reveal that the negative relationship between sprawl and labour
productivity holds in both samples. Yet, it turns out that metropolitan sprawl induces greater
labour productivity penalties in smaller MSAs.17 The estimates suggest that increasing sprawl
by one standard deviation is associated with 3.2 and 2.7 percentage points lower labour
productivity in smaller and larger MSAs respectively (recall this controls for population). This
finding weakly suggests that lower levels of urbanization/agglomeration economies in smaller
MSAs magnify the negative consequences of sprawl (or dispersed intra-MSA population
distribution).18
We next explore how the influence of our sprawl measure varies across regions because
sprawl may have different effects in (say) the ‘densely’ populated Northeast versus the
growing Sunbelt regions. Specifically, we re-estimate the base model by interacting our sprawl
variable with a dummy variable representing one of the eight major BEA region dummy
variables. The results modestly suggest that the negative effect of sprawl varies spatially with
the negative sprawl/productivity association being the smallest in the Southeast and Far West
(not shown).
17
The sprawl coefficients statistically differ between the small and large MSA samples at the 1 percent level.
18
To assess whether the residential location of workers in terms of commuting affects our results, we added median
commute time to our base model (not shown). However, the commute time was insignificant and the other results were
essentially not affected.

4.3 Sprawl/productivity: Industry effects
The above results suggest that sprawl and productivity are inversely related on average - even
when controlling for population and industry composition. We now ask an even stronger
question of whether the sprawl-productivity relationship holds within given industries in each
metropolitan area by estimating a model that pools data for 20 main industries:19
Ln ⎛ ⎞
Q
= β0 + β1metropolitan _ sprawli ,1990 + β2 popi ,1990 + β3 popi2,1990 +
⎝ L ⎠ij ,2001 (10)
β4 educi,1990 + β5 Natural amenityi + μ j + vi + ei
Where (Q/L)ij is GDP per worker in industry j located in MSA i, mj is the industry dummy and
vi is the state dummy. The model estimated from Equation 9 includes all industries aggregated
in the output measure, Equation 10 includes a Q/L ratio for each industry pooled into one
sample. That is, the number of observations is greater by a factor of 20. In this model, with the
industry fixed effects, we are examining whether sprawl affects productivity within each major
industry in the MSA.20 The other explanatory variables are specified as in the base model and the
error terms are assumed to cluster across industries (Moulton 1986). The results reported in
column 1, Table 4 reveal that the negative relationship between metropolitan sprawl and labour
productivity remains robust, suggesting that the sprawl-productivity association holds even
within industries, a stronger statistical result.
We next investigate whether the negative productivity-sprawl association varies across major
industries by estimating a separate regression for each of the major 20 industries, using GDP per
worker in industry j located in MSAi (Q/L)ij as the dependent variable (not shown). These results
reveal some expected differences. A greater level of metropolitan sprawl is especially negatively
related to productivity in industries that use considerable information and face-to-face interac-
tion, such as the finance and insurance industry. On the other hand, sprawl did not have a
statistically significant influence for industries in which productivity is less reliant on immediate
proximity of firms such as the wholesale and retail industries. In total, metropolitan sprawl is
statistically associated with lower metropolitan labour productivity in the following industries:
manufacturing; construction, finance and insurance, professional and technical services, gov-
ernment; and health care and social assistance.
4.4 Metropolitan sprawl-productivity linkages
Given our results, we now ask why is metropolitan sprawl associated with lower metropolitan
labour productivity? Rather than focusing only on the demand side as an explanation of the
agglomeration effects (as mentioned in the Introduction), we now consider consumption effects
on the labour supply side. A growing literature suggests that household location decisions are
driven not only by economic opportunities, but also by the level of urban amenities, including:
entertainment, diversity, lifestyle and better access to goods and services (Costa and Khan 2000;
Lloyd and Clark 2001; Adamson et al. 2004). These characteristics, strongly associated with
dense urban areas, are particularly applicable to more skilled and talented workers who are more
19
The list of 20 industries is shown in the descriptive statistics table in the Appendix.
20
Including industry dummies controls for industry specific characteristics that affect labour productivity, such as
R&D (innovation), intensity or skilled labour intensity (Klenow 1996; Ngai and Samaniego 2009). Also, controlling for
industry fixed effects would account for cross industry differences in the capital intensity.

Table 4. Pooled industry/occupation regression: Metropolitan sprawl-labour productivity*
Variable -1- -2- -3- -4-
Pooled industry regressions Pooled occupation:

Wage dependent variable
OLSa 2SLS OLSb 2SLS

(IV: 1940 density)a (IV: 1940 density)b
1990 metropolitan sprawl -0.159 -0.968 -0.029 -0.26

(-4.08) (-2.72) (-3.21) (-4.82)
1990 total metropolitan population 1.06E-07 4.27E-08 6.23E-08 3.53E-08
(-7.57) (-1.57) (-6.69) (-3.09)
1990 total population squared -6.47E-15 -3.54E-15 -5.34E-15 -3.12E-15
(-7.64) (-2.71) (-5.62) (-2.74)
Natural amenity scale 0.0138 0.0137 0.002 -0.003
(1.15) (1.08) (0.87) (-1.12)
1990 share of at least bachelor degree 0.0146 0.0154 0.005 0.005
(5.46) (5.38) (8.95) (9.06)
1990 share of associate degree 0.023 0.03 0.004 0.007
(4.26) (4.36) (2.28) (3.64)
1990 share of some degree -0.012 -0.011 0.002 0.002
(-6.48) (-6.29) (2.61) (3.16)
1990 share of high school 0.005 0.011 0.002 0.005
(3.85) (3.82) (2.45) (3.81)
Industry dummies Y Y
Occupation dummies Y Y
State dummies Y Y Y Y
Constant -2.927 -2.695 9.798242 10.42758
(-50.63) (-25.52) (189.66) (214.15)
R-square 0.79 0.81 0.94 0.93

N 5,050 5,003 6,656 6,635
First stage: F-statc 154.43 237.58
Weak Inst. Wald Fd 1,058.65 17.23
Hausman test: p-vale 0.0098 0.104
Notes: *Robust (spatially clustered) t-statistics are in parentheses. In calculating the robust t-statistics, the error terms
are assumed to cluster by industry (columns 1–3) and occupation (column 4–6).
a
The dependent variable is log GDP per worker by industry.
b
The dependent variable is log average metropolitan wages by occupation.
c
d
e
Hausman test for the null hypothesis that the OLS results do not suffer from (statistical) endogeneity bias. P-values
greater than 0.05 suggest that the null hypothesis cannot be rejected, suggesting that OLS is the preferred estimator.
spatially mobile. Accordingly, higher levels of urban amenities in compact MSAs may enhance
labour productivity through attracting concentrations of skilled workers.
To explore whether more compact MSAs attract more skilled workers, we estimate an
auxiliary cross-section model in which the percentage change of 1990–2000 share of MSA
population above age 25 that have at least a bachelor degree (as a proxy for skilled workers), is
regressed on 1990 metropolitan sprawl and other control variables. The control variables
include: 1990 total metropolitan population; 1980–1990 employment growth; 1980–1990 per
capita income growth; squared population; the natural amenity index; and state dummies. This
exercise investigates whether low-density communities are inversely associated with skilled

Table 5. Metropolitan sprawl effects on higher education*
1990–2000 educational changea
1990 metropolitan sprawl -0.044

(-3.6)
1990 total metropolitan population 1.59E-09
-0.89
1990 total population squared -1.27E-16
(-1.15)
Natural amenity scale -0.001
(-0.38)
1980–1990 employment growth 0.045
-3.22
1980–1990 per capita income growth 0.057
-4.24
State dummiesb Y
Constant 0.019
-1.19
R-square 0.49
N 356
Notes: *Robust (spatially clustered) t-statistics are in parentheses. In calculat-

ing the robust t-statistics, the clusters are formed based on BEA economic areas,
which are defined as the relevant regional markets surrounding metropolitan or
micropolitan statistical areas. See: URL: http://www.bea.doc.gov/bea/regional/
docs/econlist.cfm.
a
The dependent variable is the 1990–2000 percentage change in the ratio of the
2000 share of population with at least a bachelor degree.
b
Metropolitan areas that cross state boundaries are assigned to the state in
which the majority of metropolitan population reside.
labour growth.21 The regression estimates shown in Table 5 are supportive of our expectations.
The coefficient of metropolitan_sprawl1990,i is negative and significant at the 1% level. Therefore,
over the long-run, one way that sprawl may reduce productivity is through an inverse link with
the growth in the college graduate share, though we leave a definitive answer to future
research.22
5 Sensitivity analysis
The previous section assessed the impact of urban sprawl on metropolitan labour productivity,
the main finding being that metropolitan areas with higher levels of sprawl tend to experience
lower average labour productivity, even within industries. We now further assess the robustness
of this finding to: (i) simultaneity between metropolitan sprawl and labour productivity in the
21
Using the 1990–2000 change in college graduate share also has the advantage of differencing out any fixed effects
in levels - that is any fixed factor associated with both sprawl and the share of college graduates. Like the 2000
university level results, the Hausman test suggested that the OLS results were not endogenous.
22
To assess whether there is any endogeneity with the sprawl variable, we employ two sets of strong instruments:
1940 population density and 1950 metropolitan population density in conjunction with the 1950 metropolitan non-white
population share (more discussion about the strength of instruments is found in Section 5). The Hausman test results
could not reject the null hypothesis that the OLS results were not biased, suggesting that OLS is the preferred method.
Even so, the same inverse sprawl-college graduate growth relationship was still obtained in the IV models when using
the 1950 population density and the 1950 share of non-white population as instruments (the results are not shown),
though this was not case when using 1940 population density.

base model and pooled industry model; (ii) the functional form of the empirical base model; (iii)
spatial correlation of the error terms; (iv) an alternative productivity measure; and (v) alternative
productivity measure.
5.1 Simultaneity between urban sprawl and labour productivity
To mitigate possible direct simultaneity between metropolitan sprawl and labour productivity,
sprawl is measured in the initial period (1990). We further address potential endogeneity by
using IV for 1990 metropolitan sprawl in the base model (Equation 9) and pooled industry
model (Equation 10) with two alternative instruments: (i) the 1940 population density; and (ii)
1950 population density and 1950 share of non-white metropolitan population. The key advan-
tage of the first is that 1940 population density predates the US entry into the Second World War,
the planning of the Interstate highway system, and modern suburban development associated
with sprawling modern cities. The second instrument set considers the role of past population
settlement patterns and early social problems such as white flight in explaining 1990 metro-
politan sprawl (Mills and Price 1984).
As shown by the F-tests at the bottom of Table 2, these instruments are strongly related to
1990 sprawl in the first stage. To further assure the validity of our instruments, we perform a
weak instrument test based on critical values provided by Stock and Yogo (2005). Reported at
the bottom of the table, the Cragg-Donald Wald F-statistics exceeds the critical value for 10%
maximal IV size distortion, indicating that the IVs are strongly robust. Yet, the reported
Hausman test results suggest that the null hypothesis that the OLS estimates do not suffer from
(statistical) endogeneity bias, cannot be rejected at even the 10 percent level, implying that
endogeneity is not a primary concern. However, given the strong concern in the land-use
literature on endogeneity, we report the IV results to show that our results are robust. Indeed, the
ensuing 2SLS results confirm that metropolitan sprawl is negatively related to productivity in
both the base and pooled models (cols. 2 and 3 Tables 2 and col. 2 in Table 4). Thus, the results
are robust when we treat metropolitan sprawl as endogenous.
5.2 The functional form of the empirical base model
To test if using different functional forms alters the estimates, the base model is re-estimated
using both linear and log-log forms. The results, not reported, show that the negative relationship
between metropolitan sprawl and average labour productivity is not an outcome of a specific
functional form.
5.3 The spatial correlation of the error terms
The base model corrects for the spatial correlation of the error terms assuming that they are
clustered in BEA areas. As a robustness check, general method of moments (GMM), is
employed, which assumes a general form of spatial correlation that declines with distance from
the metropolitan area of interest (Conley 1999).23 Not reported, the GMM results are similar to
those of the base model.
23
We use a cut-off distance of 400 km, after which, it is assumed that the error terms are not correlated. In other
regressions, the GMM model is re-estimated using different cut-off distances of 200 km and 600 km. Still, the results
are similar.

5.4 Measurement of urban sprawl
To ensure that the findings are not peculiar to the specification of the sprawl index (Equation 8),
we change the cut-off density that distinguishes between high and low-density block groups.
Instead of the median, the 25th percentile of the overall metropolitan block group density is
used. This implies that a metropolitan area with a high percentage of its population living in
block groups with density below the 25th percentile, would have more low-density sprawl
compared to those with density above the latter threshold. Reported in column 1, Table 6, the
results demonstrate that the negative impact of metropolitan sprawl on labour productivity is
robust. Likewise, we re-estimated the model in which metropolitan sprawl is reconstructed using
the 75th as the cutoff density (column 2), where again, the results are robust. We then adopt the
Lopez and Hynes (2003) fixed cut-off of 3,500 per square mile as the cut-off. These sensitivity
results also indicate robust findings (not shown).
Table 6. Metropolitan sprawl-labour productivity: Different sprawl measures*
Variables -1- -2- -3- -4-
25th percentile 75th percentile Population Base model with

cut-off cut-off density population density
Metropolitan sprawl (25th percentile cut-off ) -0.128

(-2.64)
Metropolitan sprawl (75th percentile cut-off ) -0.285
(-3.48)
Metropolitan sprawl (median cut-off ) -0.22
(-2.48)
Population density 2.90E-05 -4.02E-06
(2.07) (-0.22)
1990 total metropolitan population 4.80E-08 3.95E-08 4.59E-08 4.33E-08
(4.33) (3.28) (3.83) (-3.75)
1990 total population_squared -2.60E-15 -2.47E-15 -2.73E-15 -2.47E-15
(-3.68) (-3.2) (-3.59) (-3.45)
Natural amenity scale -0.014 -0.02 -0.015 -0.017
(-1.27) (-1.76) (-1.35) (-1.59)
1990 share of at least bachelor degree 1.934 1.926 1.924 2.00
(6.5) (7.41) (6.83) (7.23)
1990 share of associate degree -0.005 -0.003 -0.004 -0.43
(-0.96) (-0.58) (-0.84) (-0.81)
1990 share of some degree 0.0003 0.002 0.001 0.15
(0.14) (0.66) (0.43) (0.61)
1990 share of high school 0.005 0.006 0.005 0.59
(2.22) (3.18) (2.49) (2.87)
1990 shares of industry employment Y Y Y Y
State dummiesa Y Y Y Y
Constant -3.54 -3.31 -3.58 -3.32
(-13.31) (-11.55) (-13.36) (-11.43)
R-square 0.83 0.83 0.83 0.83

N 357 357 357 357
micropolitan statistical areas. See:URL: http://www.bea.doc.gov/bea/regional/docs/econlist.cfm.
a
population resides.

To assess how our results compare to the previous sprawl literature, we re-estimate our
model by replacing our sprawl measure with aggregate population density, which, despite its
weaknesses, has been widely used in the sprawl literature. Note, that like our sprawl index, rural
clusters within metropolitan areas are excluded in our calculation of average MSA density. Since
density is inversely related to sprawl, labour productivity is expected to be greater in denser
metropolitan areas. The results reported in column 3, Table 6 confirm this expectation. A one
standard deviation increase in density is associated with a 1.8 percentage point higher level of
productivity, which is less than two-thirds the size associated with our sprawl measure.
However, as explained earlier, the innovation of our sprawl index is that it captures both main
aspects of sprawl: density and population distribution (land-use pattern). To assess whether our
sprawl index independently captures differences in land-use patterns, using the base model
shown in Table 2, we add population density to account for ‘sprawl’ in addition to our sprawl
measure. The results reported in column 4 in Table 6 show that the sprawl coefficient remains
negative and highly significant, suggesting the sprawl index captures factors independent of
aggregate MSA population density.24 Conversely, the population density variable is statistically
insignificant, which together with the sprawl variable coefficient suggests that land-use has
additional effects on productivity besides just the density. More research is then needed to
determine why land-use patterns appear to affect productivity beyond the simple density effects
stressed in the literature.
Our sprawl index identifies MSA sprawl based on the national median density of all
metropolitan block groups. One possible concern is that this approach could mask cross-
metropolitan differences in land-use patterns that vary due to geography or region. To assess this
possibility, we try to two alternative density cut-offs between high and low density in Equation
(8). First, we construct a sprawl index by grouping metropolitan areas into either being on the
Atlantic or Pacific Coast and those that are not on an ocean coast. We then use the respective
coastal and non-coastal median densities as cut-offs in constructing our sprawl measure. Yet, as
shown in columns 1–3 of Table 7, the results are robust to this definitional change. Next, we use
median density cut-offs relative to the eight major BEA regions where the metropolitan area is
located.25 Specifically, for all MSAs that belong to the same BEA region, the cut-off is the
region’s median density of its MSA block groups.26 The pooled industry regression (Equation
10) uses this sprawl index to evaluate the sprawl productivity relationship. Column 4 of Table 7
reports the OLS results, which shows that the negative effect of sprawl remains robust. The same
finding is maintained with the IV estimates (columns 5 and 6).
5.5 Alternative productivity measure
Some of the empirical agglomeration literature has used average wages to study labour produc-
tivity differences across regions (e.g., Wheaton and Lewis 2002). For comparison we also
investigate the relationship between average labour productivity and metropolitan sprawl using
2001 average metropolitan wages by occupation to measure productivity. The wage data are
24
The nature of our sprawl index is that it identifies the level of sprawl of a given metropolitan area based on the share
of block groups that is below (above), the national median. Yet, it gives less consideration of how far a block group is
above or below the national threshold. To investigate how this may affect our results, we interact the sprawl index with
the aggregate metropolitan density (after removing rural clusters), to assess whether our sprawl index has an effect that
varies by average density. The results, unreported, show that the sprawl coefficient is still highly significant and the
coefficient magnitude is similar to the base model. Yet, the estimate of the interaction variable is insignificant, suggesting
that the influence of our sprawl measure does not vary by overall density.
25
Metropolitan areas that cross BEA region boundaries are assigned to the region in which the majority of
metropolitan population reside.
26
The median block group density in the eight BEA regions are: Far West (6,135); Rocky mountains (4,239); Plains
(3,785); Great Lakes (4,600); New England (3,472); Mid East (6,410); South East (2,554); and South West (3,791).

Table 7. Pooled industry regression: Metropolitan sprawl-labour productivity using regional sprawl indexes*
Variable -1- -2- -3- -4- -5- -6-
OLSa 2SLS 2SLS OLSa 2SLS 2SLS

IV: 1940 IV: 1950 IV: 1940 IV: 1950
densitya density & densitya density and
non-white non-white
sharea sharea
1990 sprawl (coastal/ -0.114 -0.867 -0.940

non-coastal MSAs)b (-2.96) (-2.66) (-3.52)
1990 sprawl (BEA -0.181 -1.03 -0.769
regions)c (-5.52) (-2.77) (-2.51)
1990 total metropolitan 1.10E-07 5.63E-08 5.12E-08 1.00E-07 1.70E-08 4.25E-08
population (6.41) (2.11) (2.24) (7.5) (0.49) (1.43)
1990 total population -6.69E-15 -4.34E-15 -4.12E-15 -6.17E-15 -2.13E-15 -3.36E-15
squared -(6.47) (-3.29) (-3.49) (-7.59) (-1.26) (-2.28)
Natural amenity scale 0.015 0.023 0.024 0.016 0.022 0.020
(0.98) (1.45) (1.48) (1.29) (1.68) (1.6)
1990 share of at least 1.457 1.521 1.527 1.404 1.197 1.261
bachelor degree (4.62) (4.56) (4.64) (5.32) (4.6) (4.9)
1990 share of associate 2.299 2.903 2.960 2.455 3.601 3.252
degree (4.06) (4.33) (4.35) (4.49) (4.39) (4.42)
1990 share of some -1.243 -1.068 -1.052 -1.270 -1.284 -1.279
degree (-6.11) (-5.49) (-5.34) (-6.69) (-7.0) (-6.92)
1990 share of high 0.479 1.081 1.139 0.542 1.258 1.039
school (3.5) (3.61) (4.4) (4.34) (3.81) (3.7)
Industry dummies. Y Y Y Y Y Y
State dummies Y Y Y Y Y Y
Constant -2.716 -2.926 -2.908 -2.663 -2.748 -2.845
(-64.29) (-34.95) (-38.77) (-55.66) (-20.11) (-24.34)
R-square 0.79 0.78 0.78 0.79 0.77 0.78

N 5,050 5,003 5,003 5,050 5,003 5,003
First stage: F-statd 155 156 101 100
Weak Inst. Wald Fe
Hausman test: p-valf 0.0336 0.0187 0.0903 0.1189
Notes: *Robust (spatially clustered) t-statistics are in parentheses. In calculating the robust t-statistics, the error terms
are assumed to cluster by industry columns 1–3) and occupation (columns 4–6).
a
The dependent variable is log GDP per worker by industry.
b
Metropolitan sprawl is constructed using the median cut-off density of the coastal metropolitan areas and the national
median cut-off density for the non-coastal metropolitan areas.
c
Metropolitan sprawl is constructed using the median cut-off density of the BEA region, where the metropolitan area
is located.
d
e
f
Hausman test for the null hypothesis that the OLS results do not suffer from (statistical) endogeneity bias. P-values
greater than 0.05 suggest that the null hypothesis cannot be rejected, suggesting that OLS is the preferred estimator.
obtained from the US Department of Labour’s Occupational Employment Statistics program

(URL: http://www.bls.gov/oes/oes_dl.htm).27 We pool the occupation data across 22 major
occupations so that occupational fixed effects can be accounted for. The explanatory variables
27
The units of observation are MSAs and PMSAs using 2000 census definition, see www.census.gov. Data avail-
ability affects the number of observations across occupations. Descriptive statistics on the 22 occupations are available
on request.

are the same as those in the pooled-industry model (Equation 10).28 Thus, in this model with
occupational dummies, we are also asking the stronger question of whether the sprawl-
productivity relation holds within given occupations in a MSA. The OLS results reported
in column 3 of Table 4 are consistent with the previous findings. The coefficient of
metropolitan_sprawli remains negative and highly significant at the 1% level - namely, even
within a given MSA occupational structure, sprawl is inversely correlated with productivity
(wages).
As in the case for pooled industry model, column 4 of Table 4 present the instrumental
variable (IV) results using the 1940 population density instrument used above. Like the 2SLS
pooled industry model, the instrument passes the IV weak test. The Hausman test shows that
OLS results suffer from (statistical) endogeneity bias, suggesting that 2SLS is the preferred
estimator. The 2SLS results continue to indicate that metropolitan sprawl is negatively related to
average wages (productivity). The same results follow when using the 1950 instruments (not
shown).
6 Conclusion
The consequences of urban sprawl have been the focus of a growing economic literature.
Proponents of compact city policies suggest that urban sprawl is associated with deteriorating
socio-economic outcomes, inefficient provision of public goods, and environment degradation.
On the other hand, a growing literature contends that the negative consequences of sprawl are
exaggerated.
While productivity consequences of sprawl would seem to be of central importance in the
evaluation of sprawl, literature addressing this relationship is scant. Using a novel measure of
sprawl we seek to address this deficiency. Our sprawl measure includes consideration of the
distribution of population density within metropolitan areas, namely, not just overall density.
Using metropolitan GDP per worker to measure labour productivity, we find that sprawl is
associated with lower average labour productivity even when allowing metropolitan sprawl to be
endogenous. The finding holds when conducting the analysis within a given industry or within
a given occupation, further suggesting that the results are robust and are not an artifact of
occupational or industry sorting by firms or workers.
Our results imply that anti-sprawl policies could enhance the economic performance of
metropolitan areas. The significance of the higher levels of urban amenities in compact cities is
also linked to attracting more skilled and educated people, which may be one of the underlying
causes of the results. Yet, we caution that even though sprawl is found to be negatively linked to
labour productivity, sprawling cities are not necessarily associated with lower welfare. For
example, through residential preferences, sprawling cities may allow workers to live closer to
work, which suggests that they would accept lower wages. If so, firms may be willing to
trade-off lower productivity for lower wages. Thus, if anti-sprawl policies are considered for the
purpose of increasing productivity, we caution that residential preferences for workers need to
be considered as well.
Our results also point to the importance of exploring the metropolitan land-use character-
istics that shape the relationship between labour productivity and sprawl. Therefore, further
research is needed to assess why the distribution of population (i.e., land-use), within a given
metropolitan area has independent effects from standard scale agglomeration effects.
28
Other variables that might affect wage differences are taxes, unionization and public infrastructure, which are
largely determined at the state level, thus captured by the state dummies.

References
Adamson DW, Clark DE, Partridge MD (2004) Do urban agglomeration effects and household amenities have a skill
bias? Journal of Regional Science 44: 201–224
Alonso W (1964) Location and land-use. Harvard University Press, Cambridge
Batty M (2007) Cities and complexity: Understanding cities with cellular automata, agent-based models, and fractals.
The MIT Press, Cambridge, MA
Baum-Snow N (2007) Did highways cause suburbanization? Quarterly Journal of Economics 122: 775–805
Bernard AB, Jones CI (1996) Comparing apples to oranges: Productivity convergence and measurement across
industries and countries. American Economic Review 86: 1216–1238
Brueckner JK (2000) Urban sprawl: Diagnosis and remedies. International Regional Science Review 23: 160–171
Brueckner JK, Kim HA (2003) Urban sprawl and the property tax. International Tax and Public Finance 10: 5–23
Brueckner JK, Largey AG (2006) Social interaction and urban sprawl. CESifo Working paper Series No. 1843
Burchfield M, Overman H, Puga D, Turner M (2006) Causes of sprawl: A portrait from space. Quarterly Journal of
Economics 121: 587–633
Carlino G (1992) Accounting for differences in aggregate state productivity. Regional Science and Urban Economics
22: 597–617
Carruthers J (2002) Evaluating the effectiveness of regulatory growth management programs: An analytical framework.
Journal of Planning Education and Research 21: 406–420
Carruthers J (2003) Growth at the fringe: The influence of political fragmentation in United States metropolitan areas.
Economics of Governance 82: 475–499
Carruthers J, Ulfarsson G (2003) Urban sprawl and the cost of public services. Environment and Planning B: Planning
and Design 30: 503–522
Carruthers J, Úlfarsson G (2008) Does ‘smart growth’ matter to public finance? Urban Studies 45: 1791–1823
Ciccone A, Hall RE (1996) Productivity and the density of economic activity. American Economic Review 86: 54–70
Conley TG (1999) GMM estimation with cross-sectional dependence. Journal of Econometrics 92: 1–45
Costa D, Khan M (2000) Power couples: Changes in the location choice of college educated: 1940–1990. Quarterly
Journal of Economics 115: 1287–1315
Cutsinger J, Galster G, Wolman H, Hanson R, Towns D (2005), Verifying the multi-dimensional nature of metropolitan
land-use: Advancing the understanding and measurement of sprawl. Journal of Urban affairs 27: 235–259
Ellickson B (1971) Jurisdictional fragmentation and residential choice. American Economic Review 61: 334–339
Fulton W, Pendall R, Nguyen M, Harrison A (2001) Who sprawls most? How growth patterns differ across the U.S.
Brookings Institution, Washington, DC
Galster G, Hanson R, Woman W, Coleman S, Friebage J (2000) Wrestling sprawl to the ground: Defining and measuring
an elusive concept. Fannie Mae, Washington, DC
Glaeser EL (1998) Are cities dying? Journal of Economic Perspectives 12: 139–160
Glaeser EL, Khan ME (2004) Sprawl and urban growth. Handbook of Regional and Urban Economics 4: 2481–2527
Glaeser EL, Maré DC (2001) Cities and skills. Journal of Labour Economics 19: 316–342
Glaeser EL, Sacerdote B (1999) Why is there more crime in cities? Journal of Political Economy 107: 225–258
Gordon P, Richardson HW (1997) Are compact cities a desirable planning goal? Journal of the American Planning
Association 63: 95–105
Jacobs J (1969) The economy of cities. Random House, New York
Jaffe A, Trajtenberg M, Henderson R (1993) Geographic localization of knowledge spillovers as evidenced by patent
citations. Quarterly Journal of Economics 108: 577–598
Jordan S, Ross JP, Usowski KG (1998) U.S. suburbanization in the 1980s. Regional Science and Urban Economics 28:
611–627
Klenow PJ (1996) Industry innovation: Where and why. Carnegie-Rochester Conference Series on Public Policy 44:
125–150
Knaap GJ, Ding C, Hopkins LD (2001) Managing urban growth for the efficient use of public infrastructure: Toward a
theory of concurrency. International Regional Science Review 24: 328–343
Knaap G, Nelson A (1992) The regulated landscape: Lessons on state land-use planning from Oregon. Lincoln Institute
of Land Policy, Cambridge, MA
Lang R (2003) Open spaces, bounded places: Does the American west’s arid landscape yield dense metropolitan
growth? Housing Policy Debate 13: 755–778
Lloyd R, Clark T (2001) The city as an entertainment machine. Research in Urban Sociology: Critical Perspectives on
Urban Redevelopment 6: 357–378
Lopez R, Hynes H (2003) Sprawl in the 1990s: Measurement, distribution, and trends. Urban Affairs Review 38:
325–355
Lucas R (1988) On the mechanics of economic development. Journal of Monetary Economics 22: 3–42

Lynch, K (1981) A theory of good city form. MIT Press, Cambridge, MA

Marshall A (1890) Principles of economics. Macmillan, London.
McCann P (2007) Sketching out a model of innovation, face-to-face interaction and economic geography. Spatial
Economic Analysis 2: 117–134
McCann P (2008) Globalization and economic geography: The world is curved, not flat. Cambridge Journal of Regions,
Economy, and Society 1: 351–370
Mieszkowski P, Mills ES (1993) The causes of metropolitan suburbanization. Journal of Economic Perspectives 7:
135–147
Mills ES (1967) An aggregative model of resource allocation in metropolitan area. American Economic Review 197:
197–210
Mills ES, Price R (1984) Metropolitan suburbanization and central city problems. Journal of Urban Economics 15: 1–17
Moomaw, RL (1983) Spatial productivity variations in manufacturing: A critical survey of cross-sectional Analyses.
International Regional Science Review 8: 1–22
Moulton B (1986) Random group effects and the precision of regression estimates. Journal of Econometrics 32:
385–397
Muro M, Puentes R (2004) Investing in a better future: A review of the fiscal and competitive advantages of smarter
growth development patterns. The Brookings Institution Centre on Urban and Metropolitan Policy, Washington,
DC
Ngai LR, Samaniego RM (2009) Accounting for research and productivity growth across industries. Review of
Economic Dynamics. doi:10.1016/j.red.2009.12.002
Nechyba T, Walsh R (2004) Urban sprawl. Journal of Economic Perspectives 18: 177–200
Partridge MD (2010) The dueling models: NEG vs amenity migration in explaining U.S. engines of growth. Papers in
Regional Science 89: 513–536
Partridge MD, Rickman DS, Kamar A, Olfert MR (2008a) The geographic diversity of U.S. nonmetropolitan growth
dynamics: A geographically weighted regression approach. Land Economics 84: 241–266
Partridge MD, Rickman DS, Kamar A, Olfert MR (2008b) Lost in space: Population dynamics in the American
hinterlands and small cities. Journal of Economic Geography 8: 727–757
Partridge MD, Rickman DS, Kamar A, Olfert MR (2009a) Agglomeration spillovers and wage and housing cost
gradients across the urban hierarchy. Journal of International Economics 78: 126–140
Partridge MD, Rickman DS, Kamar A, Olfert MR (2009b) Do new economic geography agglomeration shadows
underlie current population dynamics across the urban hierarchy? Papers in Regional Science 88: 445–466
Putnam RD (2000) Bowling alone. Simon and Schuster, New York
Quigley JM (1998) Urban diversity and economic growth. Journal of Economic Perspectives 12: 127–138
Real Estate Research Corporation (RERC) (1974) The costs of sprawl: Environmental and economic costs of alternative
residential development patterns at the urban fringe. Vol. 1, Executive Summary; Vol. 2, Detailed Cost Analysis;
Vol. 3, Literature Review and Bibliography. U.S. Government Printing Office, Washington, D.C
Rigby DL, Essletzbichler J (2002) Agglomeration economies and productivity differences in U.S. cities. Journal of
Economic Geography 2: 407–432
Stock JH, Yogo M (2005) Testing for weak instruments in linear IV regression. In Andrews DW, Stock JH (eds)
Identification and inference for econometric models: Essays in honor of Thomas Rothenberg. Cambridge Univer-
sity Press, Cambridge
Sveikauskas LA (1975) The productivity of cities. Quarterly Journal of Economics 89: 392–413
Torrens PM (2008) A toolkit for measuring sprawl. Applied Spatial Analysis 1: 5–36
Wheeler CH (2001) Search, sorting, and urban agglomeration. Journal of Labour Economics 19: 879–899
Wheaton W, Lewis M (2002) Urban wages and labour market agglomeration. Journal of Urban Economics 51: 542–562

Appendix
Table A1. Descriptive statistics
Variable Obs Mean Std. Dev. Min Max
GDP/L 359 51,163.3 9,777 32,954.20 110,281

1990 metropolitan sprawl 360 0.639 0.128 0.163 0.950
1990 total metropolitan population 358 568,658 1,345,379 40,714 16,900,000
1990 total population_squared 358 2.13E+12 1.71E+13 1.66E+09 2.85E+14
Natural amenity scale 360 3.81 1.235 1.5 7
2000 share of at least bachelor degree 360 0.225 0.073 0.099 0.524
1990 share of at least bachelor degree 360 0.167 0.053 0.075 0.366
1990 share of associate degree 360 0.061 0.016 0.024 0.123
1990 share of some degree 360 0.219 0.046 0.105 0.399
1990 share of high school 360 0.311 0.055 0.172 0.485
2001 Labour Productivity across industries:
Administrative and waste services 241 24,659 6,039 12,797 54,865

Art, entertainment & recreation 280 20,933 11,567 8,415 135,839
Health care & social assistance 221 40,137 4,535 24,586 53,233
Educational service 226 20,287 6,989 3,846 53,595
Accommodation and food services 283 20,543 4,862 13,158 54,319
Other services 330 24,318 3,582 17,669 41,519
Wholesale 249 77,996 18,465 40,816 175,332
Construction 326 41,453 8,174 21,795 71,420
Manufacturing 334 73,754 19,843 39,967 161,788
Retail 351 34,019 5,301 21,144 58,360
Finance & insurance 323 71,703.3 29,431.6 38,572.8 346,820.3
Real estate & rental & leasing 316 170,585.6 73,976 42,049.9 485,252.6
Information 291 81,498 25,354 40,495 220,434
Professional & technical services 231 46,493 13,405 25,499 109,980
Management of cos. & enterprises 200 71,025 32,252 18,692 275,600
Mining 134 67,020.95 2,101,751 6,369.4 647,115.8
Utilities 149 209,512.9 1,470,521 93,750.34 948,114.5
Government 358 47,523.07 1,127,666 33,538.88 78,567.28
Agriculture, forestry, fishing, hunting) 114 78,813.58 1,981,315 10,928.95 388,889.5
Transportation 187 46,657.57 1,230,179 22,774.36 127,380.1
1990 shares of industry employment
Public administration 360 0.050 0.031 0.017 0.218

Other professional & related services 360 0.060 0.015 0.030 0.154
Educational service 360 0.094 0.039 0.046 0.383
Health 360 0.086 0.020 0.043 0.221
Entertainment 360 0.014 0.011 0.005 0.140
Personal service 360 0.032 0.014 0.018 0.188
Business & repair 360 0.042 0.009 0.024 0.067
FIRE 360 0.057 0.018 0.026 0.163
Retail 360 0.180 0.021 0.124 0.260
Whole sale industry 360 0.040 0.012 0.014 0.107
Communication and utility 360 0.026 0.008 0.010 0.074
Transportation 360 0.039 0.011 0.018 0.097
Manufacturing 360 0.179 0.079 0.043 0.484
Construction 360 0.064 0.016 0.036 0.129
Mining industry 360 0.007 0.017 0.000 0.175
Labour productivity at the industry level is not available for every metropolitan area due to confidentiality requirements
of the reporting agencies.

doi:10.1111/j.1435-5957.2011.00330.x
Resumen. Este artículo se basa en las teorías de aglomeración urbana para investigar empíri-
camente la relación entre el desempeño económico de las áreas metropolitanas de los EE.UU.
y la dispersión de cada una. Para medir la dispersión urbana, creamos una medida distintiva que
incorpora la distribución de la densidad de población y el uso del suelo dentro de áreas
metropolitanas. Por medio del uso de métodos de mínimos cuadrados ordinarios (MCO) y de
variables instrumentales (VI), descubrimos que los niveles elevados de dispersión urbana están
asociados con una productividad laboral media. Esta patrón se mantiene incluso para sectores o
clasificaciones laborales específicas.
© 2011 the author(s). Papers in Regional Science © 2011 RSAI. Published by Blackwell Publishing, 9600 Garsington Road,
Oxford OX4 2DQ, UK and 350 Main Street, Malden MA 02148, USA.

Fallah Et Al (2010)

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Fallah Et Al (2010)

Uploaded by

Copyright:

Available Formats

doi:10.1111/j.1435-5957.2010.00330.

Urban sprawl and productivity: Evidence from US

Received: 4 September 2009 / Accepted: 30 July 2010

JEL classification: R11, R12, R14

Key words: Sprawl, productivity, agglomeration

Papers in Regional Science, Volume 90 Number 3 August 2011.

outcomes. These include increasing the concentration of poor households in low-income

Papers in Regional Science, Volume 90 Number 3 August 2011.

Papers in Regional Science, Volume 90 Number 3 August 2011.

MPK i = ∂Qi ∂ K i = Aiα K iα −1L1i −α = r (2)

Dividing (2) by Qi = Ai ( K iα L1i −α ) yields:

Substituting (3) into (1) and collecting Qi terms yields:

We allow other metropolitan characteristics related to productivity to be included in the

Papers in Regional Science, Volume 90 Number 3 August 2011.

3.1 Measuring urban sprawl

Papers in Regional Science, Volume 90 Number 3 August 2011.

Papers in Regional Science, Volume 90 Number 3 August 2011.

Table 1. Sprawl index compared to population density ranking

Top 12 sprawling metropolitan areas (1) (2) (3)

Barnstable Town, MA 0.9497 1 13

Middle 12 sprawling metropolitan areas

Boise City-Nampa, ID 0.6534 176 254

12 least sprawling metropolitan areas

Miami-Fort Lauderdale-Pompano-Beach, FL 0.3405 352 358

3.2 Empirical specification

Papers in Regional Science, Volume 90 Number 3 August 2011.

Papers in Regional Science, Volume 90 Number 3 August 2011.

4.1 Base model

Papers in Regional Science, Volume 90 Number 3 August 2011.

Table 2. Metropolitan sprawl-average labour productivity, base model*

Variable -1- -2- -3-

OLS 2SLS 2SLS (IV: 1950 density

1990 metropolitan sprawl -0.203 -0.897 -0.997

R-square 0.84 0.76 0.74

4.2 Productivity: Linkages across metropolitan size

Papers in Regional Science, Volume 90 Number 3 August 2011.

Table 3. Metropolitan sprawl-labour productivity across different metropolitan size*

Variables -1- -2-

Smaller MSAsa Larger MSAs

1990 metropolitan sprawl -0.31 -0.197

R-square 0.72 0.87

Papers in Regional Science, Volume 90 Number 3 August 2011.

4.3 Sprawl/productivity: Industry effects

4.4 Metropolitan sprawl-productivity linkages

Papers in Regional Science, Volume 90 Number 3 August 2011.

Table 4. Pooled industry/occupation regression: Metropolitan sprawl-labour productivity*

Variable -1- -2- -3- -4-

Pooled industry regressions Pooled occupation:

OLSa 2SLS OLSb 2SLS

1990 metropolitan sprawl -0.159 -0.968 -0.029 -0.26

R-square 0.79 0.81 0.94 0.93

Papers in Regional Science, Volume 90 Number 3 August 2011.

Table 5. Metropolitan sprawl effects on higher education*

1990–2000 educational changea

1990 metropolitan sprawl -0.044

Notes: *Robust (spatially clustered) t-statistics are in parentheses. In calculat-

Papers in Regional Science, Volume 90 Number 3 August 2011.

5.1 Simultaneity between urban sprawl and labour productivity

5.2 The functional form of the empirical base model

5.3 The spatial correlation of the error terms

Papers in Regional Science, Volume 90 Number 3 August 2011.

5.4 Measurement of urban sprawl

Table 6. Metropolitan sprawl-labour productivity: Different sprawl measures*