Employment Growth in the American Urban Hierarchy: Long Live Distance

Abstract

New information technologies and reductions in transportation costs have led pundits to pronounce the "death of distance." These claims would suggest that distance is no longer a barrier to growth for remote areas and small urban centers. Despite extensive research on the localized effects of agglomeration, very few studies have empirically investigated the broader spillover effects of proximity and location in the urban system. This study attempts to fill this void using U.S. county level employment data. A primary innovation is that urban centers, from which distance is measured, are differentiated by their position within six tiers (rural plus 5 urban) of the American urban hierarchy. Net agglomeration economies can thus originate from multiple sources throughout the entire 360° span. Our findings indicate that proximity to higher-tiered urban centers continues to be an important positive determinant of local job growth, all the way from the smallest to largest urban centers.

Full text

Employment Growth in the American Urban Hierarchy: Long Live Distance*
by
Mark D. Partridge1
Dan S. Rickman2
Kamar Ali3
M. Rose Olfert4
August 16, 2006
Abstract: The emergence of new information technologies and reductions in the cost of transporting
goods, combined with numerous other developments led many pundits to pronounce the “death of
distance.” Yet, long-standing settlement patterns have seemingly remained unchanged. Large urban
centers continue to decentralize, placing growth pressures on sprawling suburbs, whereas remote
communities struggle. Despite the extensive research on urban agglomeration, it is remarkable how little
research has empirically investigated the relationship between distance from core urban areas and job
growth. This study fills this void using U.S. county level data linked to an immense geographic
information system database. A primary innovation is that distance effects are differentiated for six tiers
in the American urban hierarchy. We estimate ceteris paribus relationships between county employment
growth and distance to the nearest urban center in each tier. The results indicate that closer proximity to
larger urban centers is an important positive determinant of job growth, becoming stronger over time. We
conclude that not only is distance not dead, it appears stronger than ever.
*We thank Jordan Rappaport for generously providing Stata code for the estimation of the GMM models
and his related advice.
1. Department of Agricultural, Environmental, and Development Economics, The Ohio State University, 2120 Fyffe
Road, Columbus, OH 43210, USA. Phone: 306-966-4037; Fax: 614-688-3622; Email: partridge.27@osu.edu,
webpage: http://aede.osu.edu/programs/Swank/. (contact author)
2. Department of Economics and Legal Studies, Oklahoma State University, Stillwater, OK 74078, USA. Phone:
405-744-1434, Fax: 405-744-5180. Email: dan.rickman@okstate.edu.
3. Department of Agricultural Economics, University of Saskatchewan. 51 Campus Drive, Saskatoon, SK, S7N 5A8,
CANADA, E-mail: kamar.ali@usask.ca
4. Department of Agricultural Economics, University of Saskatchewan. Email: rose.olfert@usask.ca.
http://www.crerl.usask.ca.

Employment Growth in the American Urban Hierarchy: Long Live Distance*
Abstract: The emergence of new information technologies and reductions in the cost of transporting
goods, combined with numerous other developments led many pundits to pronounce the “death of
distance.” Yet, long-standing settlement patterns have seemingly remained unchanged. Large urban
centers continue to decentralize, placing growth pressures on sprawling suburbs, whereas remote
communities struggle. Despite the extensive research on urban agglomeration, it is remarkable how little
research has empirically investigated the relationship between distance from core urban areas and job
growth. This study fills this void using U.S. county level data linked to an immense geographic
information system database. A primary innovation is that distance effects are differentiated for six tiers
in the American urban hierarchy. We estimate ceteris paribus relationships between county employment
growth and distance to the nearest urban center in each tier. The results indicate that closer proximity to
larger urban centers is an important positive determinant of job growth, becoming stronger over time. We
conclude that not only is distance not dead, it appears stronger than ever.

1
1. Introduction
The economic role of distance is the raison d’être for regional economics, the foundation of central
place theory and the core of the epiphany that more recently propelled the New Economic Geography
(NEG) into prominence. Advantages afforded by minimizing distance costs, in the form of agglomeration
economies, have often been used to rationalize a century-long urbanization trend. Yet suburbanization
and episodic U.S. rural rebound during the 1970s and early 1990s have generated pronouncements of the
diminishing effect of distance, in which rural stagnation and decline during the 1980s has been
characterized as an aberration (Johnson and Beale, 1995; 1998; Gordon et al., 1998). And despite slower
economic growth during the late 1990s and the recent recession, growth in nonmetropolitan areas once
again exceeds metropolitan growth in the United States (USDA, 2006). Several recent developments have
been highlighted as underlying the perceived diminished role of distance in local economic growth.
Advances in communications technology have been argued to greatly reduce the necessity of close
physical proximity for personal communications, leading to the coining of the phrase “the death of
distance” by Frances Cairncross (1995; 1997). Concurrently, the declining role of goods (vs. services) in
the economy, the emphasis on quality rather than quantity, and improved transportation technologies have
contributed to a declining role of transport costs (Glaeser and Kohlhase, 2004). Positive agglomerating
externalities also may diminish as traditional sectors mature and their products become standardized,
leading to a spatial dispersal of industries (Rossi-Hansberg, 2005). Numerous conditions in the core areas
of large cities such as higher crime, taxes, land prices, traffic congestion, and environmental pollution,
may cause firms and households to disperse outward (Glaeser, 1997), giving distance a positive, rather
than negative, economic value.
It is not clear, however, that these developments are tantamount to the death or diminishment of
distance. Advances in information technology may complement face-to-face interaction rather than
substitute for it. For example, there may be an urban bias in the provision of telecommunication
infrastructure (Gillespie et al., 2001). Likewise, the internet may be a complement to cities rather than a
substitute as distance may affect access, and location may affect the dissemination of information in
cyberspace (Wang et al., 2003; Sinai and Waldfogel, 2004). And although transport costs have fallen,
transactions costs related to distance may have increased (Rietveld and Vickerman, 2004).
Deconcentration has occurred primarily in mature industries where innovation has waned, while in new

2
industries such as information technology, positive externalities lead to concentration (Le Bas and
Miribel, 2005). Finally, distance may matter greatly for people who commute to work and for household
access to urban amenities and higher-order services.
Assessment of whether distance continues to play a role in the economy or has diminished in
importance requires empirical analysis beyond examining aggregate nonmetro growth rates. Although
there have been several studies of distance effects on the spatial interaction among cities (e.g., Hanson,
1998b; Dobkins and Ioannides, 2001, Ioannides and Overman, 2004), there are surprisingly few studies
on how distance from higher-tiered urban centers affects employment growth for lower tiers of the urban
hierarchy.1 Moreover, to our knowledge there have not been any studies on how employment growth in
the hinterlands is influenced by distance to levels of the urban hierarchy.2 An empirical investigation of
the relationships among these cities with their hinterlands using different time periods would reveal
whether distance continues to play a critical role in the determination of regional growth patterns.
Therefore, in this study we examine how distance from successively larger urban centers affects
U.S. county employment growth in the 1990s. All else equal, if economic activity continues to
concentrate in and near urban areas, then a county’s growth should be inversely related to its distance
from successively higher-level urban tiers. The analysis is performed for both aggregate and sector-level
employment. To isolate the effect of distance, the econometric specification controls for numerous other
growth factors related to initial conditions and changes in exogenous factors during the period of analysis.
The next section presents the theoretical framework for considering how distance can affect regional
employment growth through influencing both firm and household location decisions. Section 3 develops
the empirical model, in which county growth depends on incremental distances to larger employment
centers, other initial conditions, and changes in exogenous conditions. Empirical implementation and a
discussion of the results follow in section 4. For aggregate employment, we find no evidence supporting
the notion of a “death of distance,” or even for a greatly diminished role for distance. For hinterland areas,
there are growth penalties for incremental distances from each higher-tier urban center. There is also less
1Partial exceptions for the U.S. are Desmet and Fafchamps (2005; 2006), which examined county employment
growth. Although they incorporated distance effects for all counties, there were no separate analyses for urban and
non-urban counties, delineation of the urban hierarchy, or consideration for the 1990s, all necessary elements for
assessing whether nonmetropolitan growth was associated with a changing role of distance. Other related studies
have considered Canada (Partridge et al., forthcoming a) and France (Combes, 2000).
2Other studies have simply focused on whether the concentration of production has changed, while ignoring
considerations of distance (e.g., Ellison and Glaeser, 1997; Dumais et al., 2002).

3
evidence of growth penalties among smaller urban areas for their distances from larger urban centers. By
contrast, incremental changes in the population size of nearby higher-tiered urban centers generally has
no significant impacts on a community’s job growth, indicating that distance to urban tiers matter more
than the urban centers at any tier being marginally larger. Generally, we find that increasing the size of a
larger city in the county’s urban hierarchy (say 500,000 to 800,000) has a much smaller effect than
changing the county’s distance to the higher-tiered city. At the sectoral level, growth penalties found in
the hinterlands were mostly for retail trade and services. A summary and conclusions follow in section 5.
2. Regional Employment Growth and Urban Proximity
Our conceptual framework characterizes regional employment change as the result of location
decisions by firms and households at the margin. Adjustments by profit maximizing firms and utility-
maximizing households occur in response to spatially varying factors such as agglomeration economies
and natural amenities.
2.1 Firm Location Decisions
In the urban hierarchy, firms choose their location (i) to maximize profits (Πi ), the difference
between total revenues (TRi) and costs (TCi) associated with production in i:
(1) Πi = TRi - TCi
Revenues reflect fob prices and quantities, both of which will vary with location as they are sensitive to
costs related to the distribution of the product to the market and to market size. TCi is determined by the
wage rate (wi), the cost of acquiring material inputs (mi), and the cost of commercial land (rc
i). Costs are
also affected by an exogenous firm “amenity” component (Xi) that relates to location-specific attributes
such as public capital, natural resources, and labor-force quality. A region’s population density (Ni) will
influence the size of the local market, as well as firm productivity and wages through agglomeration
economies (Rosenthal and Strange, 2001; Black and Henderson, 1999; Stelder, 2005).
Of particular interest to this study is the relationship between profits and distance from the urban
core. Core urban centers are typically associated with agglomeration economies or production
externalities, the spatial benefits of which decay with distance (Hanson, 1997; Lucas and Rossi-Hansberg,
2002). Agglomeration economies may occur within industries possessing similar input structures
(localization economies), or between industries of dissimilar input structure (urbanization economies)

4
(Rosenthal and Strange, 2001).3 For externalities depending on personal contacts, their positive effects on
profits likely attenuate quickly with distance (Wallsten, 2001). Yet congestion costs associated with urban
size may stimulate growth in nearby areas as firms attempt to avoid these costs.
From New Economic Geography (NEG), close proximity to suppliers and customers lowers
transportation costs (Venables, 1996), in which scale economies may exist in the production of non-traded
intermediate inputs (Fujita, 1988). This reduces relative profitability of producing in the hinterlands and
draws resources into the urban core. Yet, increased competition associated in close proximity to urban
agglomerations acts as a dispersal force, in what has become referred to as “Krugman’s agglomeration
shadow” (Dobkins and Ioannides, 2001, Ioannides and Overman, 2004).
Thus, representative firm k’s indirect profit function in region i can be denoted as:
(2) Πk
i = TRk
i (DISTij, ·) - TCk
i(wi, mi, rC
i, Xi , Ni, DISTij, ·),
where DISTij is the distance of region i from urban area j. Distance effects can be either positive or
negative for either, or both, revenues and costs, and likely depend on the size of the urban core from
which the distance is measured. Long-run equilibrium dictates equalization of regional profits, in which
higher profits in region i relative to the national average induce a subsequent net movement of firms into i
and expansions or contractions by existing firms (Partridge and Rickman, 2003).
Several recent developments suggest the role of distance and agglomeration economies may be
changing over time, with some impacts being offsetting, making the net impact a priori ambiguous.
Transportation costs of goods have declined (Glaeser and Kohlhase, 2004), communication technology
improved (Cairncross, 1995; 1997), international trade patterns shifted, industries matured, and
congestion thresholds may have been reached in the largest urban areas. Falling transportation costs make
agricultural goods cheaper to ship into core urban areas, increasing the tendency towards agglomeration
(Fujita and Thisse, 1996). Yet, falling transportation costs also can disperse manufacturing activity to less
congested areas due to weakened agglomerative upstream and downstream linkages. Improved
communication technology reduces the tendency toward agglomeration if it is a substitute for face-to-face
contact, but increases agglomeration to the extent it is a complement (Glaeser, 1997). The agglomeration
effect of the increasing prevalence of the Internet depends upon whether it is a complement to cities or a
3Intra-metropolitan urban agglomeration effects suggest that productivity levels and wage rates are higher in
urban areas (Glaeser and Maré, 2001), which supports faster job growth. Likewise, wages are also higher in regions
with greater population and income, with spillovers that may extend for hundreds of miles (Hanson, 1997, 1998b).

5
substitute; e.g., internet access and the dissemination of information in cyberspace has ambiguous impacts
across rural and urban areas (Forman et al., 2005; Wang et al., 2003; Sinai and Waldfogel, 2004). While
spatial dispersal of an existing mature industry often occurs (Rossi-Hansberg, 2005), in new industries
such as information technology, positive externalities may lead to concentration (Le Bas and Miribel,
2005). Core urban centers may reach thresholds where the balance of agglomeration benefits, versus
congestion costs, is significantly altered (Hansen, 2001). Finally, shifting trade patterns can cause
economic activity to move away from dense urban cores to border areas to reduce transportation costs
associated with international trade (Hanson, 1997).
2.2 Household Location Decisions
Representative household l located in region i is assumed to maximize utility derived from a traded
good T (the numeraire), nontraded housing H that costs ri
H, and nonpecuniary-amenity attributes. We
assume households provide labor for a wage rate wi, which is accessed with some probability pi as a result
of labor-market conditions (0 ≤ pi ≤ 1).
Amenities depend on both a fixed level of natural amenity stocks (Si) as well as endogenously
determined population density of region i (Ni). Greater Ni can enhance household amenities as it supports
more retail, cultural and recreational venues, and a diverse and interesting urban milieu (Glaeser, 1997;
Krugman, 1993), acting as an agglomeration force. These amenities may be particularly attractive to
higher educated and skilled workers (Adamson et al., 2004). The level of human capital itself may attract
additional households through positive wage spillovers (Rauch, 1993). Moreover, migrants may likewise
be a source of patents, enhancing area productivity (Ottaviano and Puga, 1998). However, as population
or density rises, there will be increasing household disamenities such as congestion and pollution
(Glaeser, 1997). Also, Ni positively affects the price of housing (ri
H) due to land scarcity in a given region,
such that households may move to the hinterland to save on housing costs (Ottaviano and Puga, 1998).
Proximity or distance (DISTij) of location i to the nearest urban center j affects commuting access to
jobs and the availability of urban services and related amenities. Distance is differentiated by the size of
the urban center, as higher-level cities may offer a greater range of employment opportunities, higher
wages, and higher-order urban amenities including, for example, upscale shopping, professional sporting
events, and more diverse cultural and recreational venues.

6
Thus, we can write the indirect utility function as:
(3) Vi
l(wi, ri
H, Ni, Si, pi, DISTij, ·).
We hypothesize that at least over some range, utility is inversely related to distance. Long-run equilibrium
implies household mobility equalizing expected utility across regions. Higher utility from greater
amenities, higher wages, or lower housing costs in region i relative to the national average induce a net
movement of households into i (Partridge and Rickman, 2003).
The relative important of distance in the household location decision may have changed with recent
developments. Preferences for amenities increase with income and may be related to household life-cycle
considerations (Graves and Mueser, 1993). With declining relative costs of transporting goods, the
opportunity costs of moving people become more important with rising real wages (Glaeser and
Kohlhase, 2004), affecting commuting choices. Higher crime, taxes, land prices, traffic congestion, and
environmental pollution, associated with city size thresholds may cause households to move outside
cities, particularly from the inner core areas (Glaeser, 1997).
2.3 Regional Employment Change
We specify area employment growth as a reduced-form outcome of firm and household location
decisions, making it a function of factors in equations (2) and (3). Because of the high degree of U.S.
regional factor mobility, we interpret growth differences as the result of the transition from one steady
state equilibrium to another (Dumais et al., 2002; Desmet and Fafchamps, 2005). Since distance effects
may vary across sectors, we also examine growth at the sectoral level. We focus on job growth because it
better reflects regional advantages than income growth, which is confounded by amenities and
occupational structure (Glaeser et al., 1995; Beeson et al., 2001). Further, urban centers are differentiated
by size class due to the nature of the urban hierarchy. Nevertheless, the spatial relationships within urban
centers may differ greatly from those between urban centers (and between urban and rural areas). These
relationships are depicted by distance from any region i to cities of varying size j. For example, even if
distance is “alive and well,” fringe counties within urban areas could grow faster due to sprawl.
The array of agglomeration effects produces varying predictions about how location in the urban
hierarchy and urban scale influence total and sectoral employment growth. Generally, we focus on how
employment growth is affected by local urbanization effects and by proximity to other cities in the urban

7
hierarchy. For example, if distance is dead, or dying, the local incremental effects of an increase in
population size of a (nearby) urban center would likely have a much stronger effect on job growth than
the incremental effect of being more remote in the urban hierarchy (or being more distant from larger
urban centers). Likewise, we would expect the role of distance to decline over time.
2.4 Measuring the Distance Penalty
The specific inter-region/community growth effects likely depend on the nature of the spatial
interactions, which are influenced by the density and (urban) proximity of the interacting regions. For
example, Central Place Theory (CPT) and the New Economic Geography (NEG) predict a hierarchy of
cities (Christaller, 1933; Fujita and Thisse, 1996; Krugman, 1996). Higher-tiered cities offer a greater
variety of amenities and services, where successively fewer are available in lower-tiered cities and
especially in the hinterlands. Households and firms in the hinterland and lower-tiered cities must traverse
the distance to the nearest higher-tiered cities to access higher-order services and urban amenities. With
each tier of cities offering additional amenities and services, there is a distance “penalty” for access to
each incrementally higher-order service or urban amenity. Nevertheless, if spatial competition is the
primary force, then distance protection may enable remote communities to thrive as they compete less
directly with their larger neighbors. Of course, whether there is a distance “penalty” or distance “benefit”
depends on the spatial interaction and the particular tier in the hierarchy.
The distance penalties/benefits for community c in tier j of n tiers of cities in the urban hierarchy can
be denoted as follows. Let the incremental distance to a higher-order place with greater than the jth level
of services equal d and the marginal effect of greater incremental distance from a place with greater than
the jth level of services equal φ. The sum of the accessibility penalties for a given community c in the jth
tier can be depicted as:
(4) Penaltycj = ∑t dtφt,
where the summation over t is from j+1 to n tiers. Thus, incremental distances reflect the additional
penalty (or benefit) a resident/business of a county faces because they have to travel a greater distance to
access progressively higher-tiered urban centers. The penalty associated with accessing urban amenities,
higher-order services, higher-paying jobs, and lower cost goods and services (factors associated with
agglomeration economies), consistently increases the further a region is from its higher-tiered places. Yet,

8
for some areas the “penalty” could be negative, i.e., distance could be conducive to growth. For example,
consistent with growth shadows in CPT and NEG models, close proximity to the highest-tier areas inhibit
growth because of negative effects on price and profitability from spatial competition.
3. Empirical Implementation
We use U.S. counties as our units of analysis and separate them into four groups to examine the
different transmission mechanisms across the urban hierarchy. Counties that are not part of a micropolitan
area or a metropolitan area (MA) are categorized as rural/hinterlands, the lowest tier in the ‘urban’
hierarchy. The next tier consists of micropolitan areas (MICRO) which contain counties with tight
commuting links to small urban centers of 10,000–50,000 people.4,5 We then divide MA counties into
those with less than 250,000 people (“small” MAs) and those containing more than 250,000 people
(“medium” and “large” MAs). The 250,000 division point seems reasonable as it splits the metro sample
approximately in half, though we conduct sensitivity analysis using different break points. MAs and
MICRO are referred to as “urban” areas in our description.
We expect different growth dynamics across the types of counties. Rural counties may fare better
because of distance protection from spatial competition, and the importance of this protection may vary
across sectors. Yet greater distance from urban centers may limit rural growth because of fewer
commuting opportunities, weaker trade linkages, and greater distance to urban consumption amenities.
Further up the urban hierarchy, enhanced agglomeration economies from geographical spillovers and
congestion-induced decentralization of large centers may benefit counties proximate to them (Dobkins
and Ioannides, 2001). However, NEG theory generally suggests that urban centers likely need some
distance protection from higher-order centers (Fujita et al., 1999).
In implementing the model, employment growth is examined primarily over the 1990-2000 period,
4The 10,000 minimum-population distinction for a city anchoring a micropolitan area was defined by the Census
Bureau as the smallest grouping to have developed a surrounding regional labor market. A micropolitan area is
roughly the county(ies) that contain a city of between 10,000-50,000 and other counties with tight commuting links.
A metropolitan area (MA) is roughly the county/counties that contain a city of at least 50,000 plus other counties
with tight commuting links. Generally, we use the 2003 metro/micropolitan area definitions, which allows us to use
micropolitan areas (first defined in 2003) and to include counties in the metro sample if they had emerging
commuting linkages. We desire a broad definition of MAs to isolate within MA growth due to changing commuting
patterns versus inter urban center interactions due other factors. For details, see the Census Bureau MA and
Micropolitan definitions at http://www.census.gov/population/www/estimates/metrodef.html (on Dec. 15, 2005).
5Not including the fastest growing outer MA counties in the rural sample weakens any negative rural-distance to
urban center response that may have occurred (which would actually strengthen our results). However, we conduct
sensitivity analysis using the earlier 1999 definitions that mostly use 1980s commuting patterns to establish
boundaries for the then existing (1990) MAs along with any new MAs that were defined during the 1990s.

9
along with more limited analysis for other time periods in sensitivity analysis. Counties in the lower 48
states and the District of Columbia are the units of observation.6 The most complete reduced-form
specification for a given county i, located in state s is represented as equation (5) for total employment
and (6) for sector k employment:
(5) %ΔEMPis(t-0) = α + δ EMP is0 +φ GEOGis0+θ DEMOGis0 +ψECONis0 + γAMENITYis0 + σs +εis(t-0),
(6) %ΔEMPisk(t-0) = α + δ EMP isk0 +φ GEOGis0+θ DEMOGis0 +ψECONisk0 + γAMENITYis0 + σs +εisk(t-0),
where the dependent variable is the percent change in employment between periods 0 and t, which is
measured for either total employment or employment for each of 10 sectors. GEOG, DEMOG, ECON,
and AMENITY are vectors representing: spatial attributes such as distance to different tiers in the urban
hierarchy; demographic characteristics; economic characteristics; and amenities. The regression coefficients
are α, δ, φ, θ, ψ, and γ; σs are state fixed effects that account for common factors within a state; and ε is the
residual. Given the predetermined nature of many of the explanatory variables, specifying job growth as a
function of period 0 characteristics helps mitigate possible statistical endogeneity, though we consider the
possibility that statistical endogeneity exists.7
For sector-level employment, initial conditions include initial-period level of sectoral employment to
capture dynamic localization effects or reversion to the mean effects (Dumais et al., 2002). Changes in
exogenous conditions include job growth exogenous to that sector. For total employment, the initial level
of aggregate employment is included to proxy for dynamic urbanization effects (and offsetting congestion
effects), also by total population in the urban center (when relevant). Other initial area conditions can
include climate, initial education, and region. While many “initial” conditions are fixed over time, their
importance can change. For example, preferences for amenities may increase over time with income and
wealth (Graves and Mueser, 1993; Glaeser, 1997). Since these other conditions may be correlated with
6A key benefit of using counties is that they range from very rural to highly urban. Moreover, unlike city or
metropolitan area, county analysis does not suffer from selectivity bias in that counties that never “succeeded” in
terms of becoming urban centers are still included in the sample. Counties also have the advantage that their borders
are not affected by their recent growth experience (such as MAs) (Hanson, 1998a). Following the U.S. Bureau of
Economic Analysis, there are cases where independent cities are merged with the surrounding county to form a
more functional economic area (especially in Virginia). Forty three mostly small rural counties are also omitted due
to the lack of economic data. For details of sample construction, see Partridge and Rickman (2006).
7Examining growth rates also helps difference out any fixed effects associated with levels or scale of the locality
(Hanson, 2001). A full-scale fixed effect approach is inappropriate. First, because a county’s distance to the urban
centroid is constant over time, the key distance measures could not be included as variables. Second, because local
job growth and its underlying determinants are often persistent, a fixed effects (or first difference) approach nets out
these cross-sectional effects, subsuming them into the local-area’s fixed effect. The regression coefficients for the
other explanatory variables would then not reflect the long-run effects we desire (Partridge et al., forthcoming b).

10
distance to core urban centers (Hanson, 1997), our empirical approach consists of examining alternative
specifications of equations (5) and (6) to more fully sort out how distance affects growth.
We begin our analysis with more parsimonious models than reflected by equations (5) and (6) to assess
whether potential multicollinearity or endogeneity are affecting the key results. The most parsimonious
models should have the fewest multicollinearity and endogeneity concerns, but more likely suffer from
omitted variable bias, whereas the opposite applies to the most complete model. Robustness across
specifications would suggest that these concerns are minor. The county residual is assumed to be spatially
correlated with neighboring counties in which the strength of the correlation is inversely related to the
distance between the two counties. We use a generalized method of moments (GMM) procedure to produce
t-statistics that are robust to cross-sectional spillovers (Conley, 1999; Rappaport, 2004).8 Appendix Table 1
presents the detailed variable definitions, sources, and descriptive statistics.
The GEOG vector contains several spatial measures that reflect proximity to urban areas differentiated
by their status in the hierarchy. Thus, the first measure is distance to the nearest urban center of any size.
This is the distance from the center of the county to the center of the urban area if it is part of a multi-county
MA or MICRO, which reflects any offsetting effects of concentration or sprawl. If the county is not part of a
MA or MICRO, it is the distance from the county center to the center of the nearest urban center.9
For counties where the nearest urban center is not at the highest level in the urban hierarchy, we
include the incremental distances to more populous higher-tiered urban centers to reflect additional
“penalties” or “protection.” First, we include the incremental distance in kilometers from the county to reach
an MA.10 We also include variables that measure the incremental distance to reach an urban center of at
least 250,000, at least 500,000, and at least 1.5 million people.11 The population thresholds we use, to reflect
8The bandwidth extends 200kms, after which we assume no correlation in county residuals. We also calculated
robust t-statistics using the Stata Cluster command, in which we assumed that the residuals are correlated within a
given region, but uncorrelated across regions. The clustered t-statistics produced estimates that were quite similar to
the GMM t-statistics.
9The population-weighted county centroids are from the U.S. Census Bureau, which are also used to calculate the
population-weighted centroids of micro/metropolitan areas. The population category for MAs is based on their
initial year population—i.e., 1970, 1980, or 1990, since an MA area’s distance and population categories can differ
depending on the initial period.
10For example, if rural county A is 50kms from a micropolitan area and 110kms from the nearest MA, the
incremental distance to the nearest MA would be 60kms. Conversely assume county B is located in a MICRO, being
15kms from the center of its micropolitan area and 60kms from the nearest MA. Then the corresponding incremental
value to the nearest MA would be 45kms (60-15). For a county already located in a MA, the incremental value is
zero because it already is a MA.
11Incremental distance is calculated as before. If the county is already nearest to a MA that is either larger than or
equal to its own size classification, then the incremental value is zero. For example, if the county’s nearest urban
center of any size (or MA of any size) is already over 500,000, then the incremental values for the at least 250,000

11
agglomeration/congestion effects on both the firm and household side, are roughly consistent with those of
Overman and Ioannides (2001). The largest category generally corresponds to national and top-tier regional
centers, with the 500,000-1.5 million category reflecting sub-regional tiers. Other smaller-size centers reflect
different-size labor markets (for commuting) and varying access to personal and business services.
Other variables in the GEOG vector include population of the nearest or actual urban center (MICRO
or MA) to account for the competing effects of urbanization economies and congestion. Analogous to the
distance variables, we include variables for incremental population sizes of the nearest MA, a MA of at least
250,000, 500,000, and 1.5 million.12 These variables measure any added “cost” or “benefit” for job growth
attributable to being farther from incrementally more populous urban centers. That is, beyond the effects of
the nearest urban center, a county may benefit from additional agglomeration economies if its second closest
urban center is larger (and so on). To be sure, because we already include incremental distance to urban
centers within various tiers, these population terms account for any marginal population impact.
The incremental population variables account for within urban tier effects of urban size, while the
incremental distance terms account for the penalties of reaching an urban center of at least the specified
size.13 Indeed, if urbanization/congestion scale effects are the driving force in spatial employment patterns,
then these marginal population variables will dominate the influence of the incremental distance variables,
while if distance matters (a lot), then the incremental distance variables will dominate. Other specifications
(below) that used the actual population rather than the incremental population led to no change in the key
results.
Other control variables account for potential causes of growth in total and sectoral employment aside
from geographic location. First, we account for natural AMENITIES measured by climate, topography,
and at least 500,000 categories are both equal to zero. To take another example, suppose rural county A is 20kms
from a micropolitan area (its nearest urban center), 80kms from its nearest MA of any size (say 150,000 population),
140kms from a MA >250,000 people (say 400,000 population), 220kms from a MA >500,000 (which happens to be
2 million). Then the incremental distances are 20kms to the nearest urban center, 60 incremental kms to the nearest
MA (80-20), 60 incremental kms to a MA >250,000 (140-80), 80 incremental kms to a MA >500,000 (220-140),
and 0 incremental kms to a MA >1.5million (220-220).
12For example, if the nearest/actual urban center is 25,000 people (MICRO), the next closest urban center is
670,000 population, the third closest urban center is 2.5million people, then the incremental population of nearest
MA is 645,000, the incremental population of a MA that is >250,000 is 0, the incremental population of a MA
>500,000 is 0, and the incremental population of a MA that is at least 1.5 million is 1.83million (i.e., 2.5million
minus 670,000).
13For example, for the 500,000 cutoff, the incremental distance to an urban center variable accounts for penalties
to reach an urban center of at least 500,000 people. The corresponding incremental population variable accounts for
any marginal spillovers due to the relevant urban center having a population in excess of 500,000.

12
percent water area, and a related amenity scale constructed by U.S. Department of Agriculture (see
Appendix Table 1). The AMENITIES vector is included in all models as it reflects exogenous location
effects that are independent of the urban hierarchal effects that are of interest.
State fixed effects are also included in some models to account for state-specific factors including
policy differences, geographic location with respect to coasts, settlement period, or differing geographic size
of counties (they tend to be larger in the west).14 When state fixed effects are included, the other regression
coefficients are interpreted as the average response for a within-state change in the explanatory variable.
To account for human capital migration effects and initial human capital spillovers, in some models we
include initial-period DEMOG measures of racial composition, age distribution, and education (e.g., see
Glaeser et al., 1995). To control for disequilibrium economic migration, some models incorporate the
following ECON measures: initial 1989 median household income and 1990 agriculture employment share.
We also include the 1990-2000 industry mix employment growth in the sectoral employment models.15 To
account for nearby county spillovers, some models use BEA-region values of median income, excluding the
county of interest from the regional calculation (see the Appendix Table 1). We also control for log
employment in surrounding counties within the county’s BEA economic region to account for many factors
including agglomeration spillovers that attenuate more slowly over space and market potential effects
(Hanson, 1998a; Black and Henderson, 1999; Head and Mayer, 2003).
4. Empirical Results
Appendix Table 1 reports descriptive statistics for the full sample, while Appendix Table 2 reports
descriptive statistics for selected sub-samples. Table 1 contains the regression results for counties located in
core rural areas, MICRO areas, MAs with less than 250,000 population, and large MAs with more than
250,000, all using 2003 definitions. Sensitivity analysis using earlier metro definitions is described below.
4.1 Base 1990-2000 Total Employment Growth Models
With the exception of the over 250,000 MA sample, we start with a parsimonious model including
14We also experimented with including three indicators for close proximity (within 50kms) to the Atlantic Ocean,
Pacific Ocean, and the Great Lakes. However, our results were essentially unaffected and we omitted these measures
from our analysis assuming that state fixed effects adequately capture these initial/natural advantage effects.
15Industry mix growth is a common exogenous measure of demand shifts. Industry mix employment growth is the
sum of the county’s initial industry employment shares multiplied by the corresponding national industry growth
rates over the subsequent period. Because national industry growth should be exogenous to industry growth in a
given county, it is routinely used as an instrument for local job growth and local demand shifts (e.g., Blanchard and
Katz, 1992).

13
only the distance and amenity measures, followed by a second model that adds the state fixed effects, and a
third model that adds other urban population, economic, and demographic variables. This staged approach
permits an assessment of the robustness of the results to alternative specifications and addresses econometric
concerns. As shown in Table 1, the key distance results across columns (1)-(3) for rural counties, columns
(4)-(6) for MICRO counties, and columns (7)-(9) for small MA counties, are very similar.16 Thus, it appears
that the results are not artifacts of endogeneity, omitted variables, or multicollinearity.
Given the robustness of the results, we focus our discussion on the more fully specified models in
columns (3), (6), (9), and (10). We first examine the core rural area results. This is followed by examination
of the smaller urban center results and then the largest MA results. The incremental distance results are
stressed due to their first-order importance in the paper.
4.1.1 Rural Total Employment Results
The distance to the nearest urban center coefficient indicates that for every additional kilometer a rural
county is from its nearest urban center (of any size), its total employment grew 0.1% less during the 1990s,
ceteris paribus. A one standard deviation increase in the distance to the nearest urban center reduces
expected employment growth by 3.2%. If the nearest urban center is only a MICRO area, the rural county
loses an additional 0.036% of employment growth potential per incremental km to reach a MA of any size.
If the second closest urban center is a MA of less than 250,000 people, there is another penalty of 0.021%
per incremental km to reach an urban center of at least 250,000. Finally, there is a corresponding growth
penalty of about 0.027% per incremental km to reach MAs of at least 500,000 and 0.012% per incremental
km to reach a MA of at least 1.5 million (though the latter is only marginally significant). In total, all else
equal, being a core rural community that was one standard deviation further away from all urban center tiers
(in terms of incremental distances) would have experienced a distance penalty of about 10.5% less
employment growth. Without the marginally significant highest tier, the penalty would be 9.5%.
These results convincingly show that distance is not dead for core rural communities. Nor is there any
evidence of a growth shadow in terms of aggregate employment. Indeed, we find evidence of a much larger
distance penalty for rural areas than that found by Desmet and Fafchamps (2005) (though their approach
significantly differs from ours).17 Proximity to urban centers for input-output linkages and other information
16The greater than 250,000 MA sample results are also robust across these specifications, but we do not report
them for brevity.
17Defmet and Fafchamps (2005) consider proximity to total (or sectoral) employment clusters without

14
spillovers overwhelm the additional competition from more proximate urban businesses. Likewise, closer
proximity for rural commuters attracts/retains residents, who in turn purchase products from rural
businesses. These results suggest that remote rural communities that do not possess offsetting positive
factors such as natural amenities will likely struggle economically.
Unlike the distance results, there is no evidence of a penalty for the nearest MA being smaller, as none
of the urban center population terms are significant, a finding that generally applies for other urban groups.
For a rural county, having larger cities in its regional urban hierarchy appears to be much less consequential
than being geographically closer to higher-tiered urban centers. There is also no evidence of (localized)
urbanization effects for rural communities, as job growth is inversely related to the initial level of county
employment. More employment in the surrounding region is also inversely related to rural job growth. Thus,
access to a nearby urban center can increase rural employment, but more dispersed employment clusters
surrounding a rural community may produce congestion or spatial competition without the offsetting
agglomeration spillovers that occur with greater concentration.
4.1.2 Small Urban Center Employment Results
The MICRO results reveal few spatial employment interactions with neighboring urban centers in
terms of the distance or the population variables, or the employment level in the neighboring BEA region.
Thus, for micropolitan areas, proximity to nearby urban centers does not appear to significantly affect its
employment growth. Yet, there are significant spatial interactions for small MAs in which the magnitudes of
the coefficients are even larger than those in the rural models. For example, one standard deviation increases
in incremental distances to reach a MA of at least 250,000 and an MA of at least 500,000 is associated with
17% less employment growth during the 1990s (with the incremental distance to reach a MA of at least 1.5
million being insignificant). Thus, for smaller MAs, closer proximity to larger urban centers appears to be
paramount in terms of job growth. One likely reason is that these MAs have firms for which closer access to
highly-specialized legal, financial, information, and consulting services is essential.
Within micropolitan areas and small MAs, job growth is negatively associated with being farther away
from the core, though this effect is only statistically significant in the MICRO model. However, this
concentration effect is offset in both models by county employment growth being inversely related to its
distinguishing whether the employment is concentrated in one city, or its relation to other cities in the urban
hierarchy. Conversely, we more directly consider distance to cities in the urban hierarchy, based on CPT.

15
1990 initial level, which would disperse job growth. Yet, there are favorable urbanization effects as reflected
by the statistically significant coefficient on total own population in the MICRO and small MA models. In
their totality, these results illustrate the complexity of spatial interactions of small urban centers.
4.1.3 Large Urban Center Employment Results
The spatial interactions among larger urban centers are similar to their smaller counterparts. For MAs
over 250,000 people, there is a penalty for distance from larger urban centers, especially urban centers of
between 250,000-500,000 people. Residents are also penalized for distance from urban centers between
500,000-1.5 million. One standard deviation increases in incremental distance to urban centers of at least
500,000 and at least 1.5 million are associated with 12.1% (8.0 and 4.1% respectively) less job growth for
MAs of more than 250 thousand residents. Using a similar formulation when considering population
growth, Partridge et al. (2006) detect few spatial interactions for larger MAs, which suggest that job growth
is more influenced by spatial location within the urban hierarchy than is population growth.
Generally, the evidence suggests employment is concentrating nearer larger urban centers, but this is
not the same thing as saying that the largest MAs are growing faster. In fact, Desmet and Fafchamps(2006)
find that very large metro areas are losing jobs to intermediate sized urban areas. Our results suggest these
intermediate sized urban areas are more likely to be closer to the very large metro areas. We believe these
results reflect the importance of access in terms of proximity or distance to a broad range of input-output
linkages, diverse markets, and labor pooling effects (as well as access to urban amenities on the household
side). Conversely, we believe these results have less to do with incremental population or market potential
effects because the incremental population terms are consistently insignificant.
Within large MAs, all things equal, there is no simple pattern in regards to whether employment is
concentrating or deconcentrating. The negative within urban center coefficient suggests that more remote
counties are experiencing less employment growth, but this is offset by the negative coefficient on the 1990
employment level which suggests more concentrated counties experienced less job growth. Thus, the pattern
is consistent with Chatterjee and Carlino (2001) in the sense that the counties with the highest employment
densities are experiencing less rapid employment growth. Job growth is more likely in less congested
counties that are relatively close to the urban center. Partridge et al. (2006) found that the fringe areas of a
metropolitan area experienced the fastest population growth—consistent with sprawl. Together, the

16
population and employment results suggest that jobs do not necessarily follow people within a given
metropolitan area—i.e., employment dynamics appear to differ from population dynamics.
4.2 Results from Different Periods
In sensitivity analyses, we also estimated the models for periods 1990-2004, 1970-1980, and 1980-
1990.18 First, the 1990-2004 results (not shown) are nearly identical to those reported for 1990-2000,
suggesting that there were no significant changes in the underlying patterns in the first half of the
subsequent decade. The only difference is the coefficients are of slightly larger magnitude when using the
longer period, most likely because there are larger employment changes.
Results for the periods 1970-1980 and 1980-1990 are reported in Table 2. Generally, the 1980-1990
results show that, if anything, distance became more important in the 1990s. For example, compared to the
corresponding 1990-2000 rural results in Table 1, the incremental distance terms have a much smaller
impact in the 1980s model. For the three urban models, there is no clear pattern; the 1980-1990 distance
terms are more consequential in the MICRO sample, while the 1990-2000 distance terms are generally more
important in the small MA model. The results for the 1970s are generally consistent with the results for the
1980s, which again suggests that distance effects are, if anything, intensified in the 1990s versus earlier
decades. Thus, there is no indication of the death or diminishment of distance since the 1990-2000 distance
results are at least as strong (if not stronger) as they are for the 1970s and the 1980s.
4.3 Robustness Tests
Several additional models were estimated to assess the robustness of the findings. Column (1) of
Appendix Table 3 reports the results of combining the core-rural and MICRO samples in a non-MA sample
(using 2003 definitions). These results most resemble the core-rural results. Thus, it appears to be important
to divide nonmetro counties into separate MICRO and rural sub-samples because they appear to have
different growth dynamics that are masked in a combined nonmetro sample.
Appendix Table 3 also reports the results of using 1999 MA boundaries to define the sub-samples
(MICRO areas were not defined in 1999).19 The key difference between the 1999 MA boundary samples
and the previously defined samples is that generally the fastest growing outlying counties in an MA as
18For brevity, only the base model corresponding to columns (3), (6), (9), and (10) are reported in Table 2. The
1970-1980 and 1980-1990 models use initial period 1970 or 1980 explanatory variables. Yet, the initial
demographic and economic variables are not included due to electronic data availability constraints.
19In defining the sample of small and large MAs, 1999 boundaries are used in determining whether the MA fit
into its particular population category (e.g., less than or greater than 250,000).

17
defined in 2003 are moved to the non-MA models reported in column (2) of Appendix Table 3. As
expected, when comparing columns (1) and (2) in Appendix Table 3, there is an even steeper distance to
urban-center gradient for non-MAs because there is now a sharper decline in growth when moving further
into the fringe. Regarding the urban center results, removing the outlying counties from MAs weakens some
of the spatial interactions detected using the more inclusive 2003 urban definitions.
In the next set of robustness tests, instead of dividing the MA sample at a population of 250,000, we
re-estimated the model after dividing the MA sample at 500,000. These results were similar to those already
reported though the distance penalty was a little less severe in the newly defined “small” MA group and a
little larger in the “large” MA group (not shown). Yet, given the similarities in the results, we did not further
pursue these different sample divisions.
We next examined whether employment growth in the urban hierarchy affects lower-tiered centers. In
some sense, employment growth “spread” effects would not be expected because the U.S. Census Bureau
defines MICROs and MAs on the basis of tight commuting linkages among the member counties. By
design, counties that do not belong to the same urban center should have weak labor-market linkages.
Nevertheless, to assess this issue, we include measures of urban center employment growth for the same
urban-tier categories used above. Because commuting effects likely die out after 160kms, we set the
corresponding urban center job growth equal to zero if it is farther than 160kms from the county. Even
nearby, commuting effects likely decline with distance. Hence, we also included interactions of the nearest
urban center’s job growth with the county’s distance to the corresponding urban center. Finally, we
substitute the relevant 1990-2000 industry mix employment growth as an exogenous proxy because county
job growth could be simultaneously determined with job growth in the urban hierarchy.
With few exceptions, the industry mix employment growth variables and the associated distance
interactions were not significant (not shown). This finding is generally not surprising for urban centers given
the definition of MICROs and MAs. Yet, we are somewhat surprised that employment growth in the nearest
urban center does not have a statistically significant impact on rural county job growth. This finding likely
means that general accessibility to an urban center (as measured by distance to urban center) is of first-order
importance for rural employment growth, but not whether that urban center is experiencing job growth. In
sum, we conclude that the spatial interactions revealed in Table 1 are mostly due to spillovers and

18
accessibility issues such as input-output externalities, urban amenities, and labor pooling that relate to the
community’s location or proximity in the urban hierarchy.
4.4 Employment Results by Sector
The previous results demonstrate that distance still matters for aggregate employment growth. Yet,
research in economic geography traditionally suggests sectoral differences in how the spatial dimension
affects industry location. For example, CPT and subsequent NEG suggest that agriculture locates in the
hinterlands while manufactures concentrate in larger urban areas. On the other hand, recent trends of start-
ups of manufactures in rural green fields suggest they are deconcentrating to take advantage of lower land
and labor costs (Quigley, 2002). Most hypotheses on spatial location of industries though have focused on
traded goods sectors, with relatively little emphasis on the spatial location of the much larger service sector
(broadly defined) (Hanson, 2001). In particular, economic base theory suggests that services generally play
a passive role, locating where there is sufficient population and economic base.
Thus, Table 3 reports the results from re-estimating the 1990-2000 model by replacing total
employment growth with employment growth from 10 one-digit (SIC) industries. In general, these results
reveal that the traded goods sectors are not the reason for the spatial patterns uncovered in Table 1. While
total employment is concentrating closer to larger cities, farming and to a smaller extent mining (at least
away from very large MAs), is moving away from urban centers. Yet, there is some evidence that
manufactures are favoring rural locations near urban centers, and there is evidence that for MAs of 250,000
to 1.5 million, manufacturing growth was greater when the MA was more proximate to a higher-tiered MA.
Nevertheless, it would be a stretch to argue that the traded goods sectors are behind the spatial concentration
of employment growth nearer to higher-tiered urban centers.
Regarding the non-traded sectors, the rural service and retail sectors experience distinctly lower
employment growth in more remote counties, ceteris paribus. One explanation for this pattern relates to the
importance of input-output linkages with urban centers, rather than urban counterparts representing a source
of spatial competition—i.e., stronger overall rural growth from close proximity to urban centers helped
promote its retail and service sector growth. Likewise, there is evidence that these two sectors also gained
more employment in MAs more proximate to higher ordered urban centers. Though not as strongly evident,
wholesale employment also tended to grow faster in MAs that are closer to higher-tiered urban centers.

19
Overall, non-traded goods sectors tend to be concentrating near higher-tiered urban centers. This is not
the same thing as stating that these sectors are concentrating within large cities. To some extent, this is
consistent with previous findings that services are concentrating in and near employment agglomerations
(Desmet and Fafchamps 2005; 2006). Nevertheless, when considering all 10 sectors, the overall sectoral
results do not reveal as strong a pattern of spatial concentration as do the aggregate employment results. Yet,
they suggest that services and (and perhaps retail) may be an underlying driver of spatial location that is not
predicted by traditional models. More research is needed to assess whether the observed concentration of
service and trade sector employment near large urban centers is stronger in more innovative sectors versus
mature sectors—i.e., consistent with product cycle analysis.
Across the individual sectors, the evidence also does not support localization economies as being a
strong force in describing the spatial location of employment growth. Specifically, there was generally a
strong inverse association between a sector’s initial level of employment and its subsequent job growth,
which is consistent with Combes’ (2000) findings for France. However, sectoral job growth was greater in
counties that had higher initial levels of total employment, consistent with urbanization and diversity effects.
To some extent, such a pattern would support regional concentration of aggregate employment, but not
regional concentrations of employment in particular sectors (consistent with Dumais et al., 2002).
5. Conclusion
In this study we examined the effects of distance from core urban areas on U.S. county employment
growth during the 1990s. We conclude that distance remains a strong force behind U.S. regional growth
dynamics. Although our methodology did not address whether U.S. employment became more concentrated
in certain regions or within the largest cities, we found that within the existing urban hierarchy employment
became more concentrated near the largest urban centers during the 1990s, even as congestion costs may
have limited the growth within some of the largest centers. Further analysis suggested that this pattern was,
if anything, stronger in the 1990s than in the 1970s and 1980s.
Rural counties were penalized in terms of lower employment growth for greater distance from each
higher-tiered urban area, with a particularly large penalty for remoteness from the nearest urban center,
regardless of its size. Yet, the initial level of employment in the county and the level of employment in non-
urbanized surrounding counties were negatively related to subsequent employment growth. There was little

20
evidence of spatial interactions between micropolitan areas and higher-tiered urban areas. Nevertheless,
distance from higher-tiered urban areas was even more important for small metropolitan areas than for rural
counties. A possible reason for the spatial linkage of small metropolitan areas with larger areas is the need
for specialized services which are only available in larger metropolitan areas. Consistent with this
interpretation, growth in medium-sized metro areas also was dependent on proximity to larger metro areas.
In addition, we generally find that increasing the size of a larger city in the nearby urban hierarchy (say
500,000 to 800,000) has a much smaller effect than changing the lower-tier city’s distance to the larger
city, further suggesting that it is the access to any urban services that is especially beneficial.
Sector-level analysis suggested that the aggregate employment distance effects were mostly driven
by trends in the service and retail sectors. Yet, manufacturing continued to show some tendency towards
higher growth in rural areas nearer an urban area and in medium-sized MA’s near larger MAs. There were
generally inverse relationships between a sector’s initial level of employment and its subsequent growth,
which is inconsistent with localization externalities. However, consistent with the existence of urbanization
and diversity externalities, employment growth at the sector level was generally greater in counties that had
higher initial levels of total employment.
Distance-related employment growth dynamics found in this study differ in some important ways from
population growth dynamics found by Partridge et al. (2006). Further investigation appears warranted to
explain the differences and better understand the causal forces. Additional research at a finer industrial level
or the firm level also may prove to be fruitful regarding whether the observed concentration of service and
trade sector employment near large urban centers is related to the product cycle. In summary, it appears that
“distance” is alive and well, and like a fine wine it has become more complex and nuanced with time.
References
Adamson, Dwight, David Clark and Mark Partridge. 2004. “Do Urban Agglomeration Economies and
Amenities Have a Skill Bias?” Journal of Regional Science 44: 201-224.
Beeson, Patricia E., David N. DeJong, and Werner Troesken, 2001. “Population growth in U.S. counties,
1840-1990,” Regional Science and Urban Economics 31(6) 669-99.
Black, Duncan and Vernon Henderson. 1999. “Spatial Evolution of Population and Industry in the United
States.” American Economic Review 89: 321-327.
Blanchard, O.J. and L.F. Katz. 1992. “Regional Evolutions.” Brookings Papers on Economic Activity, (1):
1-75.

21
Cairncross, Frances, 1995. “The Death of Distance,” The Economist 336, No. 7934, September 30, 5-6.
_____, 1997. The Death of Distance: How the Communications Revolution Will Change Our Lives.
Harvard Business School Press: Cambridge: MA.
Chatterjee, Satyajit and Gerald A. Carlino. 2001. “Aggregate metropolitan employment growth and the
deconcentration of metropolitan employment.” Journal of Monetary Economics. 48: 549-583.
Christaller, W. 1933. “Central Places in Southern Germany.” Translated from Die Zentralen Orte in
Süddeutschland by Charles W. Baskin. Englewood Cliffs, N.J. Prentice Hall.
Combes, Pierre-Philippe, 2000. “Economic Structure and Local Growth: France, 1984-1993,” Journal of
Urban Economics, 47: 329-355.
Conley, T.G. (1999): “GMM estimation with cross sectional dependence,” Journal of Econometrics 92:
1-45.
Desmet, Klaus, and Marcel Fafchamps, 2005. “Changes in the Spatial Concentration of Employment
across US Counties: A Sectoral Analysis 1972-2000,” Journal of Economic Geography 5(3), 261-84.
_____, 2006. “Employment Concentration across U.S. Counties,” Regional Science and Urban
Economics 36, 482-509.
Dobkins Linda H. and Yannis M. Ioannides. 2001. “Spatial Interactions Among Cities,” Regional Science
and Urban Economics 31: 701-731.
Dumais, Guy, Glenn Ellison, and Edward L. Glaeser. 2002 “Geographic Concentration as a Dynamic
Process,” Review of Economics and Statistics 84: 193-204.
Ellison, Glenn, and Edward L. Glaeser, 1997. “Geographic Concentration in U.S. Manufacturing
Industries: A Dartboard Approach,” Journal of Political Economy 105(5), 889-927.
Forman, Chris, Avi Goldfarb, and Shane Greenstein. 2005. “How Did Location Affect Adoption of the
Commercial Internet? Global Village, Urban Density, and Industry Composition,” Journal of Urban
Economics, 58(3), 389-420.
Fujita, Masahisa, 1988. “A Monopolistic Competition Model of Spatial Agglomeration,” Regional
Science and Urban Economics 18(February), 87-124.
Fujita, Masahisa and Jacques-Francois Thisse. 1996. “Economics of Agglomeration,” Journal of the
Japanese and International Economies 10(4), 339-78.
Fujita, Masahisa, Paul Krugman, and Tomoya Mori. 1999. “On the Evolution of Hieracrchical Urban
Systems,” European Economic Review 43: 209-251.
Gillespie, A., Richardson, R., Cornford, J. 2001. “Regional Development and the New Economy,” EIB
Papers 6(1), 109-131.
Glaeser, Edward L., 1997. Are Cities Dying? Journal of Economic Perspectives 12(2), 139-60.
Glaeser, Edward L. and David C. Maré. 2001. “Cities and Skills,” Journal of Labor Economics 19:316-
342.
Glaeser, Edward L., Jose A. Scheinkman, and Andrei Schleifer, 1995. “Economic Growth in a Cross-
Section of Cities,” Journal of Monetary Economics 36(1), pp. 117-43.
Glaeser, Edward L. and Janet E. Kohlhase, 2004. “Cities, Regions and the Decline of Transport Costs,”
Papers in Regional Science 83(1), 197-228.

22
Gordon, Peter, Harry W. Richardson, and Gang Yu, 1998. “Metropolitan and Non-metropolitan
Employment Trends in the US: Recent Evidence and Implications,” Urban Studies 35(7), 1037-1057.
Graves, Philip E., and Peter R. Mueser, 1993. “The Role of Equilibrium and Disequilibrium in Modeling
Regional Growth and Decline: A Critical Reassessment,” Journal of Regional Science 33, 69-84.
Hansen, Niles, 2001. “Are Very Large Cities Successful? Searching for Dynamic Externalities versus
Evidence from Population Movements,” International Regional Science Review 24, 344-359.
Hanson, Gordon H., 1997. “Increasing Returns, Trade and the Regional Structure of Wages,” Economic
Journal 107, 113-33.
_____, 1998a. “Market Potential, Increasing Returns, and Geographic Concentration,” NBER Working
Paper # 6429.
_____, 1998b. “Regional Adjustment to Trade Liberalization,” Regional Science and Urban Economics
28, 419-444.
_____. 2001. “Scale Economies and Geographic Concentration of Industry,” Journal of Economic
Geography 1, 255-276.
Head, Keith and Thierry Mayer. 2003. “The Empirics of Agglomeration and Trade.” Handbook of
Regional and Urban Economics Vol. 4., accessed at http://strategy.sauder.ubc.ca/head/Papers/neat.pdf.
Ioannides, Yannis M. and Henry G. Overman, 2004. “Spatial Evolution of the US Urban System,”
Journal of Economic Geography 4, 131-56.
Johnson, Kenneth M. and Calvin L. Beale, 1995. “The Rural Rebound Revisited,” in American
Demographics 17(7), 46-51.
_____, 1998. “The Rural Rebound: Recent Nonmetropolitan Demographic Trends in the United States,”
The Wilson Quarterly Spring 16. Accessed on May 26, 2006 at
http://www.luc.edu/depts/sociology/johnson/p99webn.html.
Krugman, Paul, 1993. "On the Number and Location of Cities," European Economic Review 37, 293-98.
_____, 1996. “Confronting the Mystery of Urban Hierarchy,” Journal of the Japanese and International
Economies 10(4), 399-418.
Le Bas, Christian and Frederic Miribel, 2005. “The Agglomeration Economies Associated with
Information Technology Activities: An Empirical Study of the US Economy,” Industrial and Corporate
Change 14(2), 343-363.
Lucas, Robert E. and Esteban Rossi-Hansberg, 2002. “On the Internal Structure of Cities,” Econometrica
70(4), 1445-1476.
Ottaviano, Gianmarco I. P. and Diego Puga, 1998. “Agglomeration in the Global Economy: A Survey of
the 'New Economic Geography’,” World Economy 21(6), 707-31.
Overman, Henry G. and Yannis M. Ioannides. 2001. “Cross-Sectional Evolution of the U.S. City Size
Distribution,” Journal of Urban Economics 49: 543-566.
Partridge, M.D. and D.S. Rickman, 2003. “Do We Know Economic Development When We See It?” The
Review of Regional Studies 33(1), 17-39.
Partridge, Mark D. and Dan S. Rickman. 2006. The Geography of American Poverty: Is There a Role for
Place Based Policy? Kalamazoo, MI: W.E. Upjohn Employment Institute.

23
Partridge, M.D., M. R. Olfert, and A. Alasia. forthcoming a. “Agglomeration or Amenities: Canadian
Cities as Regional Engines of Growth,” Canadian Journal of Economics.
Partridge, M.D., R. Bollman, M.R. Olfert, and A. Alasia. forthcoming b. “Riding the Wave of Urban
Growth in the Countryside: Spread, Backwash, or Stagnation.” Land Economics.
Partridge, M.D., D.S. Rickman, K. Ali, and M.R. Olfert, 2006. “Does the New Economic Geography
Explain U.S. Core-Periphery Population Dynamics?” unpublished manuscript available at
http://www.bus.okstate.edu/ecls/rickman/neg_paper.pdf.
Quigley, John M. 2002. “Rural Policy and the New Regional Economics: Implications for Rural
America.” Conference proceedings of The New Power of Regions: A Policy Focus for Rural America
held at the Federal Reserve Bank of Kansas City, pp. 7-27. Accessed at:
http://www.frbkc.org/PUBLICAT/PowerofRegions/RC02_Quigley.pdf.
Rappaport, Jordan, 2004. “Moving to nice weather,” RWP 03-07. Research Division. Federal Reserve
Bank of Kansas City.
Rauch, James E., 1993. “Productivity Gains from Geographic Concentration of Human Capital: Evidence
from the Cities,” Journal of Urban Economics 34(3), 380-400.
Rietveld, Piet, and Roger Vickerman 2004. “Transport in Regional Science: The ‘Death of Distance’ Is
Premature,” Papers in Regional Science 83, 229-248.
Rosenthal, Stuart S. and William C. Strange. 2001. “The Determinants of Agglomeration,” Journal of
Urban Economics 50: 191-229.
Rossi-Hansberg, E., 2005. “A Spatial Theory of Trade,” American Economic Review 95, 1464-1491.
Sinai, Todd and Joel Waldfogel, 2004. “Geography and the Internet: Is the Internet a Substitute or a
Complement for Cities?” Journal of Urban Economics 56, 1-24.
Stelder, Dirk. 2005. “Where Do Cities Form? A Geographical Agglomeration Model for Europe,”
Journal of Regional Science 45, 657-679.
USDA, 2006. ERS/USDA Briefing Room: Rural Labor and Education: Rural Employment and
Unemployment, http://www.ers.usda.gov/Briefing/LaborAndEducation/employunemploy/, accessed May
26, 2006.
Venables, A.J., 1996. “Equilibrium Locations of Vertically Linked Industries,” International Economic
Review 37, 341-59.
Wallsten, Scott J. 2001. “An empirical test of geographical information systems and firm-level data,”
Regional Science and Urban Economics 31, 571-599.
Wang, Yong, Phillip Lai, and Daniel Sui, 2003. “Mapping the Internet using GIS: The Death of Distance
Hypothesis Revisited,” Journal of Geographical Systems 5: 381-405.

24
Table 1. Dependent Variable: Percentage Change in U.S. County Employment 1990-2000
Core Rural Area 2003 Boundaries Micropolitan Area 2003 Boundaries Inside MA with pop ≤250,000 MA>250,000
Variables/var groups Dist Add state FE Add other X Dist Add state FE Add other XDist Add state FE Add other XFull model
Intercept
Distance to nearest or actual
urban center
Inc Dist to MA
Inc Dist to MA>250k
Inc Dist to MA>500k
Inc Dist to MA>1500k
Pop of nearest or actual
urban center
Inc pop of nearest MA
Inc pop of MA>250k
Inc pop of MA>500k
Inc pop of MA>1500k
Log(employment 1990)
Log(emp 1990 in
surrounding counties)
20.936**
(3.60)
-0.073**
(-3.21)
-0.050**
(-4.41)
-0.018**
(-2.96)
-0.020**
(-2.10)
-0.008
(-1.50)
N
N
N
N
N
N
N
31.008**
(3.06)
-0.086**
(-3.78)
-0.043**
(-3.32)
-0.028**
(-2.76)
-0.031**
(-2.60)
-0.015*
(-1.95)
N
N
N
N
N
N
N
37.465
(1.44)
-0.105**
(-5.11)
-0.036**
(-2.95)
-0.021**
(-2.34)
-0.027**
(-2.50)
-0.012
(-1.58)
9.1E-06*
(1.65)
-9.6E-07
(-0.63)
-4.6E-07
(-0.45)
-4.6E-07
(-0.69)
-9.7E-08
(-0.25)
-1.558*
(-1.72)
-1.791**
(-2.08)
25.855**
(3.85)
-0.051
(-0.77)
-0.005
(-0.33)
-0.023**
(-2.41)
0.005
(0.37)
-0.001
(-0.18)
N
N
N
N
N
N
N
49.230**
(3.62)
-0.072
(-1.08)
-0.003
(-0.19)
-0.019
(-1.61)
0.002
(0.16)
0.007
(0.65)
N
N
N
N
N
N
N
98.514**
(2.37)
-0.391**
(-2.31)
-0.0007
(-0.04)
-0.015
(-1.01)
0.006
(0.35)
0.009
(0.83)
0.0001**
(2.22)
1.9E-06
(1.46)
1.4E-06
(1.26)
3.1E-07
(0.61)
2.4E-08
(0.08)
-5.584**
(-2.76)
0.905
(0.69)
-10.477
(-0.47)
0.137
(1.07)
n.a.
-0.068*
(-1.65)
-0.004
(-0.15)
0.005
(0.29)
N
N
N
N
N
N
N
-27.145
(-0.86)
0.078
(0.48)
n.a.
-0.098**
(-3.44)
-0.045
(-1.43)
-0.028
(-1.22)
N
N
N
N
N
N
N
497.906*
(1.66)
-0.203
(-1.44)
n.a.
-0.118**
(-2.95)
-0.101*
(-1.79)
-0.022
(-0.89)
0.0001**
(2.08)
1.2E-06
(0.98)
n.a.
5.1E-07
(0.38)
-4.9E-07
(-0.58)
-19.416*
(-1.87)
1.582
(0.70)
148.794*
(1.91)
-0.248**
(-2.78)
n.a.
n.a.
-0.109**
(-3.15)
-0.029*
(-1.69)
8.9E-07**
(2.44)
7.5E-07
(1.24)
n.a.
n.a.
-2.3E-07
(-0.95)
-7.100**
(-3.15)
-0.414
(-0.47)
Weather/Amenitya
Economic/Demographicb
Surrounding Econc
State fixed effects (FE)
Y
N
N
N
Y
N
N
Y
Y
Y
Y
Y
Y
N
N
N
Y
N
N
Y
Y
Y
Y
Y
Y
N
N
N
Y
N
N
Y
Y
Y
Y
Y
Y
Y
Y
Y
R2 0.15 0.26 0.34 0.10 0.26 0.35 0.10 0.27 0.51 0.45
No. of observations 1300 1300 1300 672 672 672 416 416 416 641
F-statistic
All MA pop = 0
Inc MA pop = 0
Inc distance to MA = 0
N
N
11.23**
N
N
7.46**
1.32
0.19
3.77**
N
N
2.24*
N
N
1.03
1.77
0.53
0.80
N
N
2.49*
N
N
2.20*
1.37
0.20
3.39**
1.03
0.75
5.36**
Notes: Robust t-statistics following Conley (1999) and Rappaport (2004) are in the parentheses. A ** and * indicate significant at ≤ 5% and ≤10% level
respectively. N=not included, Y=included. a = sunshine hours, January temperature, July humidity, typography, amenity ranking, and percent water area. b =
1989 median household income, 1990 share ag. emp., 6 age-distribution variables for 1990, 4 education categories for 1990, and 5 race/ethnicity variables for
1990. c = weighted average 1989 median household income in surrounding counties within a BEA region.

25
Table 2. Dependent Variable: Percentage Change in U.S. County Employment 1970-1980 and 1980-1990
1970-1980 change 1980-1990 change
Variables/var groups Core
Rural
Micro-
politan
MA ≤
250,000
MA >
250,000
Core
Rural
Micro-
politan
MA ≤
250,000
MA >
250,000
Intercept
Distance to nearest or
actual urban center
Inc Dist to MA
Inc Dist to MA>250k
Inc Dist to MA>500k
Inc Dist to MA>1500k
Pop of nearest or actual
urban center
Inc pop of nearest MA
Inc pop of MA>250k
Inc pop of MA>500k
Inc pop of MA>1500k
Log(total initial emp)
Log(total initial emp in
surrounding counties)
42.604*
(1.95)
-0.003
(-0.08)
-0.020
(-1.06)
-0.016
(-1.01)
-0.010
(-0.61)
-0.025**
(-2.23)
7.3E-06
(1.14)
-5.4E-07
(-0.21)
-6.9E-07
(-0.45)
5.5E-07
(0.47)
-1.3E-07
(-0.29)
0.437
(0.28)
1.619
(1.62)
391.112**
(3.13)
-1.441*
(-1.83)
0.014
(0.45)
-0.063
(-1.62)
-0.013
(-0.51)
-0.0002
(-0.01)
0.0003
(1.17)
-4.3E-06
(-0.97)
3.0E-06
(0.55)
-1.5E-06
(-1.45)
-5.2E-07
(-1.23)
-24.663**
(-2.02)
1.840
(-1.07)
72.036**
(2.45)
-0.359**
(-3.97)
n.a.
-0.048*
(-1.77)
-0.010
(-0.35)
-0.017
(-0.86)
-4.9E-05*
(-1.90)
n.a.
1.1E-06
(1.38)
1.8E-06
(1.13)
4.5E-07
(0.60)
-3.083**
(-2.58)
-0.371
(-0.31)
78.315*
(1.79)
-0.699**
(-4.25)
n.a.
n.a.
-0.131**
(-2.92)
-0.019
(-0.90)
1.1E-07
(0.13)
n.a.
n.a.
7.7E-07
(0.93)
-2.7E-07
(-0.75)
-9.270**
(-5.20)
4.466**
(1.97)
20.791
(1.19)
-0.029
(-1.34)
-0.035**
(-2.96)
-0.028**
(-2.94)
-0.016
(-1.10)
-0.006
(-0.77)
-1.3E-06
(-0.20)
-1.2E-06
(-0.71)
2.1E-07
(0.20)
2.3E-07
(0.25)
6.6E-07
(1.39)
-1.373
(-1.35)
0.139
(0.21)
89.886**
(2.44)
-0.448*
(-1.87)
-0.034*
(-1.70)
-0.028**
(-2.21)
-0.027*
(-1.67)
0.009
(0.75)
0.0008
(1.20)
1.7E-06
(0.72)
-6.3E-07
(-0.41)
2.5E-07
(0.42)
3.9E-07
(1.11)
-3.480
(-1.30)
-0.665
(-0.79)
-34.262
(-1.45)
-0.277**
(-2.97)
n.a.
-0.077**
(-3.58)
-0.008
(-0.30)
-0.034**
(-2.30)
-8.2E-06
(-0.27)
n.a.
2.2E-07
(0.36)
-1.2E-06
(-1.45)
1.9E-06**
(2.50)
-0.786
(-0.65)
0.455
(0.46)
84.367**
(2.43)
-0.442**
(-3.66)
n.a.
n.a.
-0.142**
(-4.96)
-0.038**
(-2.53)
-5.4E-07
(-0.84)
n.a.
n.a.
1.3E-07
(0.20)
-4.5E-07
(-1.22)
-4.287**
(-3.36)
3.393*
(1.89)
R2 0.30 0.33 0.45 0.41 0.33 0.31 0.34 0.31
No. of observations 1299 671 501 556 1300 672 441 616
F-statistic
All MA pop = 0
Inc MA pop = 0
Inc dist to MA = 0
0.37
0.23
1.85
1.64
0.98
1.05
1.60
0.61
1.80
0.44
0.66
14.67**
0.79
0.98
4.43**
1.04
0.60
2.42**
1.67
2.19*
6.37**
0.44
0.47
12.24**
Notes: The categories are determined using 2003 boundary definitions of metropolitan areas (MA).
Robust t-statistics following Conley (1999) and Rappaport (2004) are in the parentheses. A ** and * indicate significant at
≤ 5% and ≤10% level respectively.
All models further include 6 weather/amenity variables: sunshine hours, January temperature, July humidity, typography,
amenity rank, and percent water area; and state fixed effect variables.

26
Table 3. Dependent Variable: Percentage Change in U.S. County Employment by Sector 1990-2000
Sector Farm Mining
Variables/var groups Core
Rural
Micro-
politan
Inside MA
≤ 250,000
Inside MA
> 250,000
Core
Rural
Micro-
politan
Inside MA
≤ 250,000
Inside MA
> 250,000
Intercept
Distance to nearest or
actual urban center
Inc Dist to MA
Inc Dist to MA>250k
Inc Dist to MA>500k
Inc Dist to MA>1500k
Pop of nearest or actual
urban center
Log(total emp 1990)
Log(sector emp 1990)
-6.236
(-0.35)
0.016
(0.76)
0.023**
(2.43)
0.018**
(2.35)
0.010
(1.17)
0.023**
(3.71)
-4.8E-06*
(-1.76)
-0.731
(-0.86)
-0.402
(-0.54)
26.985
(0.91)
-0.046
(-0.41)
0.025*
(1.94)
0.023**
(1.96)
0.011
(0.97)
0.010
(1.14)
-5.1E-07
(-0.02)
0.058
(0.03)
-1.308
(-1.35)
10.671
(0.29)
0.541
(0.89)
n.a.
0.028*
(1.86)
0.001
(0.06)
0.015
(1.55)
-2.7E-05*
(-1.88)
1.901
(1.47)
-4.518**
(-3.38)
-77.731**
(-2.08)
0.102
(0.96)
n.a.
n.a.
0.063**
(3.22)
0.019*
(1.75)
-1.3E-06
(-1.13)
2.357
(0.76)
-5.399
(-1.50)
663.185
(1.58)
-0.803
(-0.95)
-0.202
(-0.41)
0.397
(1.13)
-0.732
(-1.33)
0.289**
(2.02)
0.0001
(1.11)
-12.713
(-0.42)
-69.054**
(-3.40)
379.307
(1.32)
-0.077
(-0.07)
0.019
(0.18)
-0.146
(-1.47)
0.063
(0.58)
-0.085
(-1.56)
2.9E-05
(0.12)
12.584
(1.10)
-30.797**
(-7.11)
319.669
(1.12)
0.108
(0.17)
n.a.
-0.034
(-0.36)
0.190
(1.09)
-0.138
(-1.07)
-0.0002
(-1.38)
14.156
(1.36)
-34.107**
(-4.91)
-345.548
(-1.46)
-0.102
(-0.28)
n.a.
n.a.
-0.059
(-0.77)
0.022
(0.43)
-3.2E-07
(-0.22)
10.460
(1.24)
-24.477**
(-3.97)
R2 0.40 0.48 0.58 0.28 0.08 0.29 0.31 0.21
F-statistic
All MA pop = 0
Inc MA pop = 0
Inc dist to MA = 0
0.67
0.50
4.60**
1.56
1.95*
1.72
1.26
0.95
1.78
1.28
0.17
1.97
0.74
0.80
2.79**
0.32
0.40
1.12
0.56
0.51
1.49
0.37
0.55
0.20
Manufacturing Transport and Public Utilities
Intercept
Distance to nearest or
actual urban center
Inc Dist to MA
Inc Dist to MA>250k
Inc Dist to MA>500k
Inc Dist to MA>1500k
Pop of nearest or actual
urban center
Log(total emp 1990)
Log(sector emp 1990)
-370.893
(-1.42)
-0.595**
(-3.22)
-0.139
(-1.17)
-0.051
(-0.59)
-0.078
(-0.96)
-0.055
(-1.06)
0.00003
(0.84)
56.269**
(4.35)
-56.132**
(-4.55)
-13.326
(-0.06)
-1.909**
(-2.01)
0.055
(0.36)
0.046
(0.39)
0.058
(0.91)
-0.002
(-0.03)
0.0004
(1.55)
11.463
(0.89)
-39.212**
(-4.98)
-322.235
(-1.23)
0.121
(0.39)
n.a.
-0.030
(-0.60)
0.075
(1.11)
0.095*
(1.75)
-5.3E-05
(-1.03)
44.261**
(2.28)
-44.660**
(-2.37)
220.137**
(2.06)
-0.339*
(-1.76)
n.a.
n.a.
-0.152**
(-2.79)
-0.041**
(-1.97)
-2.5E-09
(-0.01)
17.323*
(1.74)
-26.997**
(-2.95)
-196.558
(-0.77)
-0.077
(-0.91)
-0.055
(-0.67)
-0.097*
(-1.70)
-0.070
(-1.07)
-0.021
(-0.62)
4.3E-06
(0.23)
70.325**
(3.66)
-82.037**
(-3.57)
224.596
(1.21)
-0.136
(-0.20)
0.037
(0.73)
0.009
(0.16)
-0.108**
(-2.13)
0.012
(0.29)
0.0002
(1.00)
21.498
(1.27)
-36.325**
(-2.42)
-40.503
(-0.20)
0.235
(0.83)
n.a.
0.084
(1.20)
0.018
(0.30)
0.055
(1.22)
0.0002**
(2.39)
46.869**
(4.24)
-59.185**
(-4.43)
96.887
(0.92)
-0.276*
(-1.66)
n.a.
n.a.
-0.065*
(-1.89)
-0.033
(-1.32)
9.1E-07
(1.16)
26.801**
(4.45)
-30.048**
(-7.55)
R2 0.18 0.36 0.40 0.32 0.22 0.25 0.33 0.39
F-statistic
All MA pop = 0
Inc MA pop = 0
Inc dist to MA = 0
0.63
0.64
0.52
0.82
0.38
0.24
0.52
0.58
1.01
0.14
0.21
2.65**
0.42
0.50
0.69
0.63
0.52
1.17
0.99
0.09
0.64
0.26
0.02
1.03
No. of observations 1300 672 416 641 1300 672 416 641
See notes at the end of the table.

27
Table 3—cont. Dependent Variable: Percentage Change in U.S. County Employment by Sector 1990-2000
Sectors Wholesale Trade Retail Trade
Variables/var groups Core
Rural
Micro-
politan
Inside MA
≤ 250,000
Inside MA
> 250,000
Core
Rural
Micro-
politan
Inside MA
≤ 250,000
Inside MA
> 250,000
Intercept
Distance to nearest or
actual urban center
Inc Dist to MA
Inc Dist to MA>250k
Inc Dist to MA>500k
Inc Dist to MA>1500k
Pop of nearest or actual
urban center
Log(total emp 1990)
Log(sector emp 1990)
-433.46**
(-2.37)
-0.071
(-0.84)
0.008
(0.18)
-0.009
(-0.28)
0.007
(0.15)
-0.026
(-0.58)
-5.0E-06
(-0.31)
65.970**
(4.32)
-62.628**
(-3.76)
-761.422
(-1.55)
-0.195
(-0.41)
0.267*
(1.69)
-0.075
(-1.37)
-0.039
(-0.93)
-0.074
(-1.33)
0.00008
(0.48)
70.272**
(2.46)
-65.033**
(-2.65)
-173.060
(-0.49)
0.404
(0.77)
n.a.
-0.0006
(-0.01)
-0.047
(-0.70)
-0.185**
(-3.35)
0.0001
(1.63)
89.487**
(2.93)
-84.746**
(-4.01)
-257.130
(-1.29)
-0.175
(-0.81)
n.a.
n.a.
-0.175**
(-2.42)
0.030
(0.61)
1.8E-06*
(1.67)
77.211**
(4.18)
-80.909**
(-5.07)
-93.840**
(-2.46)
-0.107**
(-3.93)
-0.036**
(-2.32)
-0.040**
(-3.02)
-0.029*
(-1.67)
-0.025**
(-2.03)
0.00001*
(1.76)
22.144**
(5.61)
-21.691**
(-5.66)
81.737
(1.30)
-0.540*
(-1.77)
-0.023
(-1.30)
-0.036**
(-2.06)
-0.006
(-0.27)
0.020
(1.54)
0.0001
(1.34)
-0.595
(-0.08)
-8.148
(-1.50)
1.277
(0.01)
-0.278
(-1.23)
n.a.
-0.082**
(-2.90)
-0.033
(-0.97)
-0.021
(-0.88)
5.4E-06
(0.17)
15.871
(1.20)
-21.567**
(-2.00)
257.852**
(2.09)
-0.137*
(-1.87)
n.a.
n.a.
-0.049*
(-1.82)
0.011
(0.75)
9.3E-07*
(1.71)
3.690
(0.53)
-13.919**
(-2.05)
R2 0.26 0.33 0.41 0.47 0.25 0.26 0.35 0.48
F-statistic
All MA pop = 0
Inc MA pop = 0
Inc dist to MA = 0
0.43
0.53
0.13
0.46
0.54
4.27**
0.47
0.15
1.92
1.17
0.61
2.85**
0.82
0.11
2.96**
2.24**
1.83
1.91
0.35
0.46
1.41
1.06
0.72
1.63
Finance, insurance and real estate (FIRE) Services
Intercept
Distance to nearest or
actual urban center
Inc Dist to MA
Inc Dist to MA>250k
Inc Dist to MA>500k
Inc Dist to MA>1500k
Pop of nearest or actual
urban center
Log(total emp 1990)
Log(sector emp 1990)
-1054.0**
(-16.49)
-0.025
(-0.21)
-0.111
(-1.56)
-0.124*
(-1.84)
-0.024
(-0.42)
-0.068
(-1.04)
5.0E-07
(0.02)
234.502**
(6.17)
-230.73**
(-6.60)
298.600**
(2.18)
-0.731
(-1.38)
-0.096**
(-2.34)
-0.040
(-1.16)
0.0001
(0.01)
0.027
(1.02)
0.0003*
(1.86)
-0.953
(-0.08)
-31.053**
(-3.70)
125.540
(0.80)
0.220
(1.05)
n.a.
-0.076*
(-1.95)
-0.043
(-0.91)
0.015
(0.39)
0.00007*
(1.73)
-0.387
(-0.03)
-16.350
(-1.48)
586.549**
(2.34)
-0.231
(-1.35)
n.a.
n.a.
-0.061
(-1.12)
0.020
(0.53)
1.7E-06**
(1.99)
-5.902
(-0.49)
-21.394**
(-2.29)
92.710*
(1.65)
-0.104**
(-2.44)
-0.083**
(-3.12)
-0.031*
(-1.69)
-0.029
(-1.38)
-0.027*
(-1.81)
0.00001
(1.28)
28.043**
(5.28)
-30.866**
(-5.98)
127.839**
(2.25)
-0.298
(-1.28)
-0.016
(-0.62)
-0.033
(-1.35)
-0.016
(-0.78)
0.012
(0.79)
0.0001*
(1.79)
-4.858
(-0.82)
0.168
(0.03)
1510.477
(1.43)
-0.644
(-1.37)
n.a.
-0.299**
(-2.11)
-0.296
(-1.53)
-0.088
(-1.27)
0.0003**
(1.96)
-43.974
(-0.95)
-9.169
(-0.49)
72.941
(0.25)
-0.676*
(-1.81)
n.a.
n.a.
-0.346**
(-2.04)
-0.111
(-1.43)
1.8E-06
(1.13)
-32.015
(-1.32)
30.629
(0.97)
R2 0.33 0.42 0.40 0.43 0.28 0.25 0.42 0.25
F-statistic
All MA pop = 0
Inc MA pop = 0
Inc dist to MA = 0
0.29
0.36
0.70
1.92*
0.42
1.96*
0.35
0.08
0.92
0.87
0.69
0.85
0.83
0.67
3.07**
1.23
0.65
0.97
1.39
0.16
2.32*
0.18
0.23
2.60**
No. of observations 1300 672 416 641 1300 672 416 641
See notes at the end of the table.

28
Table 3—cont. Dependent Variable: Percentage Change in U.S. County Employment by Sector 1990-2000
Sectors Government Construction
Variables/var groups Core
Rural
Micro-
politan
Inside MA
≤ 250,000
Inside MA
> 250,000
Core
Rural
Micro-
politan
Inside MA
≤ 250,000
Inside MA
> 250,000
Intercept
Distance to nearest or
actual urban center
Inc Dist to MA
Inc Dist to MA>250k
Inc Dist to MA>500k
Inc Dist to MA>1500k
Pop of nearest or actual
urban center
Log(total emp 1990)
Log(sector emp 1990)
11.383
(0.47)
-0.037**
(-2.06)
0.002
(0.13)
-0.013
(-1.40)
-0.017
(-1.51)
-0.012*
(-1.81)
9.5E-06**
(2.14)
15.058**
(6.77)
-17.569**
(-7.18)
84.609**
(2.54)
-0.366**
(-2.64)
-0.009
(-0.65)
-0.017
(-1.42)
0.012
(0.73)
0.017*
(1.65)
0.00008**
(2.09)
11.625**
(5.19)
-17.829**
(-8.60)
147.991**
(2.84)
-0.047
(-0.67)
n.a.
-0.050**
(-3.01)
-0.037**
(-2.04)
-0.014
(-0.99)
0.00003**
(2.07)
9.753**
(3.64)
-19.484**
(-8.32)
69.590**
(2.11)
0.141**
(2.22)
n.a.
n.a.
-0.012
(-0.78)
-0.0001
(-0.01)
-3.6E-07
(-1.38)
18.122**
(5.60)
-26.769**
(-8.98)
125.809
(1.35)
-0.387**
(-6.52)
-0.056
(-1.22)
-0.073**
(-2.41)
-0.013
(-0.31)
0.0002
(0.08)
0.00002
(1.10)
34.632**
(5.77)
-45.719**
(-8.61)
80.667
(0.55)
0.194
(0.28)
0.028
(0.56)
0.026
(0.61)
-0.009
(-0.20)
0.008
(0.28)
0.0002
(1.23)
23.008**
(2.23)
-36.984**
(-6.22)
8.721
(0.08)
0.200
(1.18)
n.a.
-0.140**
(-4.36)
-0.056
(-1.52)
-6.8E-05
(-0.01)
0.0001**
(2.40)
30.455**
(3.74)
-44.211**
(-4.87)
85.498
(0.89)
-0.506**
(-3.26)
n.a.
n.a.
-0.111**
(-2.88)
-0.017
(-0.64)
2.3E-06**
(2.27)
30.233**
(3.44)
-47.847**
(-4.67)
R2 0.32 0.42 0.48 0.56 0.26 0.36 0.57 0.58
F-statistic
All MA pop = 0
Inc MA pop = 0
Inc dist to MA = 0
1.48
0.59
1.25
1.24
0.45
1.94
3.29**
4.08**
2.25*
0.41
0.55
2.48*
0.58
0.24
1.52
0.77
0.19
0.21
2.12*
1.45
3.50**
2.73**
1.82
3.55**
No. of observations 1300 672 416 641 1300 672 416 641
Notes: The categories are determined using 2003 boundary definitions of metropolitan areas (MA).
Robust t-statistics following Conley (1999) and Rappaport (2004) are in the parentheses. A ** and * indicate significant at
≤ 5% and ≤10% level respectively.
All models further include 6 weather/amenity variables: sunshine hours, January temperature, July humidity, typography,
amenity rank, and percent water area; 4 economic variables: 1989 median household income, 1990 share ag. emp., 1990
share goods emp., and weighted average 1989 median household income in surrounding counties within a BEA region; 6
age-distribution variables for 1990; 4 education categories for 1990; 5 race/ethnicity variables for 1990; percentage of
population immigrated during 1985-90; 4 incremental population variables; log of total emp. 1990 in surrounding counties;
and state fixed effect variables.

1
Appendix Table 1. Variable Definitions and Descriptive Statistics (full sample)
Variable Description Source Mean St. dev.
Dependent Variable
Employment change
Percentage change in total or sectoral employment
over 1990-2000 (and other years)
U.S. BEA, REIS 21.64 26.11
Dist to nearest/actual urban
center (micropolitan or
metropolitan area, CBSA)
Distance (in km) between centroid of a county and
population weighted centroid of the nearest urban
center, if the county is not in an urban center. It is
the distance to the centroid of its own urban center if
the county is a member of an urban center (in kms).
1990 Census, C-
RERL
34.61 32.44
Inc dist to metro Incremental distance to the nearest/actual
metropolitan area in kms (see text for details)
Authors’ est. 36.68 49.06
Inc dist to metro>250k Incremental distance to the nearest/actual
metropolitan area with at least 250,000 population in
1990 in kms (see text for details)
Authors’ est. 56.29 97.27
Inc dist to metro>500k Incremental distance to the nearest/actual
metropolitan area with at least 500,000 population in
1990 in kms (see text for details)
Authors’ est. 40.67 66.83
Inc dist to metro>1500k Incremental distance to the nearest/actual
metropolitan area with at least 1,500,000 population
in 1990 in kms (see text for details)
Authors’ est. 89.77 111.47
Nearest/Actual Urban Center
pop
1990 Population of the nearest/actual urban center
measured as a micropolitan or metropolitan area (see
text for details).
Authors’ est. 374,271.3 1,377,909.3
Inc pop of nearest metro Incremental population of the nearest/actual
metropolitan area, 1990 (see text for details)
Authors’ est. 186,155.0 457,600.8
Inc pop of metro>250k Incremental population of the nearest/actual
metropolitan area with at least 250,000 population in
1990 (see text for details)
Authors’ est. 475,334.1 908,103.9
Inc pop of metro>500k Incremental population of the nearest/actual
metropolitan area with at least 500,000 population in
1990 (see text for details)
Authors’ est. 613,128.6 1,263,115.2
Inc pop of metro>1500k Incremental population of the nearest/actual
metropolitan area with at least 1,500,000 population
in 1990 (see text for details)
Authors’ est. 1,235,251.3 2,159,221.8
Weather/Amenity
Sun hours Mean January sun hours ERS, USDA 151.41 33.21
January temp Mean January temperature (degree F) ERS, USDA 32.95 12.07
July humidity Mean July relative humidity (%) ERS, USDA 56.15 14.49
Typography Typography score 1 to 24, in which 24 represents
the most mountainous terrain
ERS, USDA 8.83 6.59
Amenity rank Natural amenity rank 1 to 7, with 7 being the highest ERS, USDA 3.49 1.04
Percent water Percent of county area covered by water ERS, USDA 4.61 11.29
Economic/Demographic
Median HH inc Median household income 1989 1990 Census 23,842.7 6,388.8
Industry mix growth Industry mix employment growth, calculated by
multiplying each industry's national employment
growth (between 1990 and 2000) by the initial
period (1990) county industry employ. shares in
each one-digit sector and summing across all
sectors.
1990, 2000 BEA,
Regional Econ.
Info. System,
Authors’ est.
0.16 0.04
Agriculture share 1990 Percent employed in agriculture sector 1990 Census 8.45 8.20
Percent pop under 6 years Percent of 1990 population under 6 years 1990 Census 10.08 1.45
Percent pop 7-17 years Percent of 1990 population 7-17 years 1990 Census 16.78 2.34
Percent pop 18-24 years Percent of 1990 population 18-24 years 1990 Census 9.18 3.43
Percent pop 55-59 years Percent of 1990 population 55-59 years 1990 Census 4.56 0.73
Percent pop 60-64 years Percent of 1990 population 60-64 years 1990 Census 4.70 0.98
Percent pop 65+ years Percent of 1990 population over 65 years 1990 Census 14.97 4.33
Percent HS graduate Percent of 1990 population 25 years and over that
are high school graduates
1990 Census 34.36 6.12
Percent some college Percent of 1990 population 25 years and over that
have some college
1990 Census 16.39 4.50
Percent associate degree Percent of 1990 population 25 years and over that
have an associate degree
1990 Census 5.34 2.10
Percent college graduate Percent of 1990 population 25 years and over that
are 4-year college graduates
1990 Census 13.43 6.45
Percent Hispanic Percent of 1990 population Hispanic 1990 Census 4.37 10.96

2
Percent African American Percent of 1990 population African-American 1990 Census 8.60 14.32
Percent Asian-Pacific Percent of 1990 population Asian and Pacific
islands origin
1990 Census 0.59 1.26
Percent Native American Percent of 1990 population that are Native American 1990 Census 1.44 5.59
Percent other race Percent of 1990 pop. with other race background 1990 Census 1.80 4.57
Percent immig 1985-90 Percent of 1990 pop. immigrated over 1985-90 1990 Census 0.48 0.96
Surrounding Variables
Employment density_surr Weighted average total/sectoral emploment in
surrounding counties within a BEA regiona 1990 Census,
Authors’ est.
663.44 1,553.27
Median HH inc_surr Weighted average median household income in
surrounding counties within a BEA regiona
1990 Census,
Authors’ est.
26,753.7 4795.7
State fixed effects (FE) Dummy variables n.a. n.a.
N 3029
Notes: Centroids are population weighted. The metropolitan/micropolitan definitions follow from the 2003 definitions. BEA = Bureau of
Economic Analysis, REIS; ERS, USDA = Economic Research Services, U.S. Department of Agriculture; C-RERL = Canada Rural
Economy Research Lab, University of Saskatchewan. See Partridge and Rickman (2006) for more details of the variable sources and
sample selection.
a. The surrounding BEA region variables are calculated as the average of the region net of the county in question. The BEA economic
regions are 177 functional economic areas constructed by the BEA.

3
Appendix Table 2. Mean and Standard Deviations (in parentheses) of Major Variables by Population Group
Variables Core Rural Micropolitan MA≤ 250,000 MA>250,000
Employment change 1990-2000 (%)
Distance to nearest or actual urban
center
Inc dist to MA
Inc dist to MA>250k
Inc dist to MA>500k
Inc dist to MA>1500k
Population density
Pop of nearest or actual urban center
Inc pop of nearest MA
Inc pop of MA>250k
Inc pop of MA>500k
Inc pop of MA>1500k
Log(Emp 1990)
Log(Emp in surrounding BEA
region)
16.59
(17.27)
59.91
(30.56)
43.47
(49.93)
76.02
(115.19)
45.32
(68.95)
83.45
(106.24)
23.01
(20.06)
73,743.63
(112,327.21)
183,099.40
(368,656.31)
533,758.31
(872,740.78)
648,040.08
(992,727.71)
1,097,303.21
(1,611,653.76)
8.84
(0.76)
12.96
(1.13)
19.62
(15.85)
4.63
(9.63)
78.46
(46.97)
48.96
(83.41)
38.17
(59.87)
99.81
(119.29)
61.14
(45.95)
48,875.11
(26,764.12)
274,366.21
(480,896.37)
460,874.87
(776,502.89)
619,122.45
(1,159,943.19)
1,385,394.92
(2,167,150.57)
9.63
(0.81)
13.23
(1.13)
26.16
(41.61)
17.76
(18.60)
n.a.
93.23
(93.26)
36.87
(59.07)
78.54
(115.44)
119.96
(123.82)
138,285.46
(45,144.03)
n.a.
1,048,541.50
(1,403,056.67)
619,832.98
(1,231,175.82)
1,048,521.20
(1,708,339.55)
9.96
(1.25)
13.22
(1.34)
31.09
(32.79)
28.60
(19.52)
n.a.
n.a.
36.29
(73.34)
99.37
(139.88)
793.46
(3,399.82)
1,698,737.27
(2,685,450.02)
n.a.
n.a.
531,690.68
(1,773,562.46)
1,478,848.73
(3,134,548.98)
10.77
(1.62)
13.83
(1.25)
Sample size 1300 672 416 641
Notes: The categories are determined using 2003 micropolitan and metropolitan area definitions. See the text for more
details.

4
Appendix Table 3. Dependent Variable: % Change in U.S. County Employment 1990-2000 Using Alternative
Boundaries & Definitions
1999 Boundaries
Variables/var groups
2003 Boundaries
Non-MA Non-MA MA with pop ≤ 250,000 MA with pop >250,000
Intercept
Distance to nearest or actual
urban center
Inc Dist to MA
Inc Dist to MA>250k
Inc Dist to MA>500k
Inc Dist to MA>1500k
Pop of nearest or actual urban
center
Inc pop of nearest MA
Inc pop of MA>250k
Inc pop of MA>500k
Inc pop of MA>1500k
Log(Employment 1990)
Log(Emp 1990 in surr counties)
52.057**
(2.58)
-0.098**
(-6.45)
-0.028**
(-2.86)
-0.022**
(-2.67)
-0.022**
(-1.98)
-0.005
(-0.83)
9.9E-06*
(1.71)
4.1E-07
(0.35)
3.3E-07
(0.38)
2.0E-08
(0.04)
8.8E-09
(0.04)
-2.136**
(-2.89)
-0.953
(-1.22)
122.773**
(2.41)
-0.163**
(-4.31)
-0.079**
(-3.56)
-0.044**
(-3.60)
-0.042**
(-2.83)
-0.017**
(-2.15)
7.3E-06*
(1.83)
7.0E-07
(0.45)
-1.1E-07
(-0.14)
-6.1E-08
(-0.11)
-4.1E-07
(-1.52)
-5.821**
(-2.13)
-0.524
(-0.62)
24.243
(0.32)
0.021
(0.30)
n.a.
-0.041**
(-3.14)
-0.003
(-0.13)
0.023
(1.06)
2.8E-05
(1.18)
n.a.
1.7E-07
(0.25)
4.6E-07
(0.63)
2.5E-07
(0.36)
-1.212
(-0.65)
2.793**
(3.20)
261.095**
(3.55)
-0.086
(-1.48)
n.a.
n.a.
-0.060**
(-2.96)
-0.002
(-0.19)
4.2E-07
(1.64)
n.a.
n.a.
3.5E-07
(0.73)
-4.3E-07**
(-2.18)
-9.794**
(-6.03)
-0.122
(-0.24)
R2 0.31 0.23 0.61 0.61
N 1972 2205 265 559
F-statistic
All metro pop = 0
Inc metro pop = 0
Inc distance to metro = 0
1.23
0.11
4.67**
0.93
0.33*
10.42**
0.33
0.06
1.79
1.61
1.57
3.66**
Notes: Robust t-statistics following Conley (1999) and Rappaport (2004) are in the parentheses. A ** and * indicate
significant at ≤ 5% and ≤10% level respectively.
All models further include 6 weather/amenity variables: sunshine hours, January temperature, July humidity, typography,
amenity rank, and percent water area; 4 economic variables: 1989 median household income, 1990 share ag. emp., 1990
share goods emp., and weighted average 1989 median household income in surrounding counties within a BEA region; 6
age-distribution variables for 1990; 4 education categories for 1990; 5 race/ethnicity variables for 1990; percentage of
population immigrated during 1985-90; and state fixed effect variables.