The spatial structure of US metropolitan employment: New insights from administrative data

Abstract

Urban researchers have long debated the extent to which metropolitan employment is monocentric, polycentric, or diffuse. In this paper I use high-resolution data based on unemployment insurance records to show that employment in US metropolitan areas is not centralized but is spatially concentrated. Unlike residents, who form a continuous surface covering most parts of each metropolitan area, jobs have a bimodal spatial distribution, with most blocks containing no jobs whatsoever and a small number having extremely high employment densities. Across the 100 largest Metropolitan Statistical Areas, about 75% of jobs are located on the 6.5% of built land in Census blocks with at least twice as many jobs as people. These relative proportions are extremely consistent across cities, even though they vary greatly in the physical density at which they are constructed. Motivated by these empirical regularities, I introduce an algorithm to identify contiguous business districts and classify them into four major types. Based solely on the relative densities of employment and population, this algorithm is both simpler to implement and more flexible than current approaches, requiring no metro-specific tuning parameters and no assumptions about urban spatial layout.

Keywords

Big data metropolitan areas spatial analysis employment subcenters

Scholars of cities have long sought to uncover regularities in the spatial structure of urban areas. In recent decades, a key portion of this effort has been directed at understanding the decentralization of employment. Following the suburbanization of the late 20th century, it is now clear that the monocentric city model formalized by Alonso (1964) is no longer adequate to explain the geographic profiles of American cities. But there is debate about whether the most accurate alternative is a polycentric city featuring many subcenters or a truly “edgeless” city (Lang, 2003) where employment is scattered almost at random throughout the metropolitan area.

This paper seeks to answer the question: how is employment distributed within US cities? Specifically, it asks whether employment in the 100 largest US Metropolitan Statistical Areas (MSAs) is more polycentric or more scattered. In answering this question, I make two major contributions to the literature on the spatial distribution of employment. First, I use Census Longitudinal Employer-Household Dynamics Origin-Destination Employment Statistics (LODES) data based on unemployment insurance records to study employment at the Census block level, a spatial resolution approximately 100 times finer than that typically used in studies of employment subcenters. These high-resolution data allow me to show definitively that while employment in US cities is not centralized, it is highly spatially concentrated. Jobs are found in all corners of most metros, but even in the exurbs they are tightly clustered into a relatively small number of blocks occupying only a minute percentage of land area. These employment blocks form contiguous districts that define the employment geography of their cities. The bimodal spatial distribution of jobs, where a few centers have extremely high employment density and most blocks have no jobs whatsoever, is fundamentally different from the more even distribution of residents.

The second contribution of this paper is to develop a new algorithm for business district identification and classification. Most attempts at subcenter identification have used employment density alone, but here I reintroduce the employment–population ratio as an alternative statistic (Forstall and Greene, 1997; McDonald, 1987; Shearmur et al., 2007). This results in an algorithm that is both conceptually simpler and analytically more flexible than current approaches. It does not require local tuning parameters or any prior knowledge of a city’s built form, not even the location of that city’s central business district (CBD). In fact, it is able to identify the CBD with remarkable accuracy. Because it uses no metro area-specific parameters, it allows for the direct comparison of business districts across US MSAs, which previous approaches do not.

Such a comparison reveals that while metros are remarkably consistent in the relative amounts of land used for employment and residences, they differ a great deal in the form that each takes. I divide business districts into four categories based on their density, employment, and physical size. Traditional urban cores—the best approximation of the CBD from the monocentric city model—today contain about 16% of all metropolitan jobs, continuing to influence the employment landscapes of many MSAs. But large, continuously developed, yet low-density districts are much more prominent, accounting for more than 40% of all jobs.

In the following section I revisit previous research on the spatial distribution of employment and subcenter identification methods. I then introduce the LODES data and show that urban employment in the United States today is decentralized but still highly concentrated. I motivate the use of the employment–population ratio by showing that the absolute densities of jobs and residences vary dramatically across the 100 largest MSAs, but the relative densities are remarkably constant. Further, metros are consistent in the proportions of their built land allocated to employment and residences, and most blocks contain exactly zero jobs. Given this, it is analytically straightforward to identify employment blocks as those few places with high employment densities and to group contiguous employment blocks into business districts. I conclude by outlining my typology of business district categories.

Theorizing metropolitan spatial structure

The spatial location of economic activity has been of interest to scholars since at least Von Thünen (1966). He theorized an isolated city on a featureless plain, and described how in such a state land use would be determined by rent and transportation costs, resulting in a series of concentric zones of decreasing productivity and rents. Elaborations on Von Thünen’s theory include the ecological model of the city (Burgess, 1925) and central place theory (Christaller, 1933; Losch, 1954). For the most part, these early theoretical models described idealized cities with a single focal point or strict hierarchy of centers, an approach that was formalized in the monocentric city model (Alonso, 1964).

Research since the 1960s has increasingly called into question the applicability of the monocentric model to US cities (Anas et al., 1998; Ladd and Wheaton, 1991). But scholars are divided as to what alternative is superior. The most common approach is to model today’s cities as polycentric rather than monocentric (e.g. Fujita and Ogawa, 1982; Henderson and Mitra, 1996). Popularized in the book Edge City (Garreau, 2011), this approach argues that jobs and commerce are still concentrated in certain parts of the metro area, but in multiple employment centers rather than just one CBD.

Another view argues that polycentricity is just a stopping point on the way to truly decentralized development (Gordon and Richardson, 1996). Rather than edge cities, this “scatteration” view argues that employment is increasingly concentrated in “edgeless cities” (Lang, 2003), amorphous office parks, and employment districts sprinkled seemingly at random through the suburbs of most metropolitan areas.

The contrast between these two viewpoints highlights the distinction between employment centralization and employment concentration. Centralization measures the extent to which jobs and people are found near the center of a region or metro area, while concentration measures the extent to which they are found on a relatively small portion of its landmass (Galster et al., 2001; Lee, 2007). The polycentric view argues that jobs have been decentralized but remain concentrated, while the scatteration hypothesis argues that both decentralization and deconcentration have occurred.

Empirical investigations have found support for both polycentric and scattered employment. Some scholars have argued that a polycentric urban form predominates in cities from Phoenix (Leslie and Ó hUallacháin, 2006) to Barcelona (Garcia-López and Muñiz, 2010). Others have argued for the scatteration hypothesis (Lang et al., 2009), with some going so far as to claim that jobs have become almost as dispersed as residents (Wheaton, 2004). The extent to which employment is concentrated in one CBD, distributed across several subcenters, or widely dispersed may vary by city (Lee, 2007).

Methods of identifying employment centers

Accepting that some portion of employment is concentrated in certain districts, a robust literature has sought to create algorithms to identify employment centers. Four main approaches have been developed, each motivated by a slightly different theoretical idea of what an employment center is. The most straightforward and widely used approach is that of Giuliano and Small (1991), who conceptualize subcenters as large centers of employment at high absolute densities. In their study of Los Angeles, they identified subcenters as groups of contiguous transportation analysis zones with an employment density of at least 10 employees per acre and total employment of at least 10,000 jobs. This employment density threshold approach, with varying parameter values, has been adopted by other researchers in studies of Chicago (McMillen and Lester, 2003), San Francisco (Cervero and Wu, 1997), and Cleveland (Bogart and Ferry, 1999), among others (Lee, 2007; Matsuo, 2011). Ban et al. (2017) expand the Giuliano and Small’s algorithm to allow it to better identify subcenters near the periphery of metropolitan areas.

A second approach is motivated by the idea that subcenters are local maxima in an employment surface (McDonald, 1987). Several variations on this approach have been proposed (e.g. Craig and Ng, 2001; Redfearn, 2007). Perhaps the most prominent version is that of McMillen (2001a), which first fits an employment density surface to the city, then identifies subcenters as areas with large positive residuals. Unlike the threshold-based approach, this method does not require that subcenters meet an absolute density cutoff but asks instead whether they are denser than surrounding areas. The density residual approach has been widely used and applied to cities including Houston (Craig et al., 2016), Milwaukee (McMillen, 2001b), and Dublin (Vega and Reynolds-Feighan, 2008), as well as multi-city comparisons (Lee, 2007).

The third approach also identifies subcenters as areas with higher than expected employment densities for their location, but using techniques from exploratory spatial data analysis. This is most commonly done by using Local Indicators of Spatial Association (Anselin, 1995) to identify statistically significant clusters of high density employment. This approach has been used in studies of Dijon (Baumont et al., 2004), Paris (Guillain et al., 2006), and Hermosillo (Rodríguez-Gámez and Dallerba, 2012), and in comparative studies of metros in Belgium (Riguelle et al., 2007) and the US (Arribas-Bel et al., 2015). Hajrasouliha and Hamidi (2017) used an alternative form of exploratory spatial data analysis based on the Getis–Ord statistic in their study of US metros.

Finally, and most directly related to this article, previous work using data at high spatial resolution has identified employment centers based on the relative proportions of jobs and residents. This was proposed as an alternative to employment density by McDonald (1987) and developed more fully by Forstall and Greene (1997). Their study, focused on the Los Angeles metro area, noted that the employment–population ratio varied dramatically across the region, and that it was able to identify zones with low absolute employment densities that nonetheless had large commuter inflows. Since then it has been used most prominently in studies of Canadian metropolitan areas (e.g. Shearmur et al., 2007), which find that large areas of the studied metros have very few jobs, and that most jobs are found in a small number of employment zones with extremely high employment densities and few residents. Those empirical facts make the employment–population ratio a straightforward and clean way to identify employment centers, one that can easily be applied to many metropolitan areas simultaneously. Despite these appealing features, this approach has not been widely used for subcenter identification in recent years. I reintroduce it to the US context in this paper.

Data

I study the spatial distribution of employment in US metros using Census LODES Workplace Area Characteristics data for the year 2014 (US Census Bureau, 2017). The LODES data are publicly available for free download from the US Census Bureau website. They are produced through a collaboration between the Census and the labor market information offices of each state, and use unemployment insurance records to provide information on the number of jobs in each Census block. Approximately 96% of wage and salary civilian jobs in the US are covered by unemployment insurance (US Bureau of Labor Statistics, 1997). All such private sector jobs appear in the LODES data, as do most state and local government jobs and a portion of federal jobs. To avoid double-counting, I limit the sample to primary jobs, the job from which each person earned the most money in 2014. In total, the 2014 LODES data contain records of 125,697,883 primary jobs, accounting for roughly 86% of the 2014 Civilian Employment Level in the Current Population Survey (US Bureau of Labor Statistics, 2017).

Besides coverage, the signal advantage of the LODES data is spatial resolution. Most previous studies of metropolitan employment patterns in the US have been conducted using transportation analysis zones (e.g. Giuliano and Small, 1991; McMillen, 2001a), Census tracts (e.g. Arribas-Bel et al., 2015), or Census block groups (e.g. Hajrasouliha and Hamidi, 2017). The LODES data exist at the Census block level, a much higher spatial resolution. Census blocks approximate city blocks in size and are defined as any land area bordered by roads, streams, or other linear features (Rossiter, 2011).¹ There are roughly 50 Census blocks in a block group and 150 blocks in a Census tract (US Census Bureau, 2019). This means the data used in this study are at least 50 times more detailed than those in previous studies. For instance, in their pioneering study of subcenters in Los Angeles, Giuliano and Small (1991) analyzed 1146 transportation analysis zones. The current analysis of the same region is built from 146,462 Census blocks, offering more than 100 times the spatial resolution.

Although the LODES data are of extremely high quality, they present a few systematic challenges. Certain large, multi-sited employers do not disaggregate their employees by establishment, meaning that all employees are recorded as working at the headquarters. This is most notable in the cases of large public agencies such as school districts and public transportation agencies. Because there is no scalable way to identify these non-disaggregated employers, I account for them by top-coding all blocks above an employment density of 500,000 jobs per square kilometer, roughly one and a half times the employment density of the block containing the Empire State Building.² In addition, there are a small number of blocks that have zero land area but are coded as having jobs, which I exclude from the analysis.

Data on the resident population of each Census block are from the 2010 Decennial Census, as compiled by the National Historical Geographic Information System (Manson et al., 2017). I focus this study on the 100 largest MSAs as defined in 2013. These include all metro areas with more than 500,000 people and collectively contain almost 204 million residents and 86 million jobs, roughly 66% of the US population in the 2010 Census and 69% of US primary jobs in the 2014 LODES data.

Empirical regularities in the spatial distribution of employment and population

Metropolitan employment is concentrated but not centralized

The dispersal of employment beyond the city center has been well documented (Giuliano and Small, 1991; Glaeser and Kahn, 2001; Lang, 2003). It appears in the LODES employment data as well. The top row of Figure 1 plots one dot per job in the Boston (left column) and Phoenix (right column) MSAs. Even in Boston, famous as a dense and centralized metro, there are plenty of jobs outside of the CBD, including many in the suburbs. From the map of Phoenix it is difficult to tell where downtown is.

Figure 1.

Dot density maps of jobs (top row) and residents (bottom row) in Boston (left) and Phoenix (right).

While the jobs shown in Figure 1 are not centralized, they do appear to be clustered. In Boston, the CBD is obvious, connecting downtown and Back Bay, as are clusters of jobs in the Kendall Square area of Cambridge and the Longwood Medical Area. Corridors of jobs are also visible following Massachusetts Avenue and the Massachusetts Turnpike from downtown out to the suburbs. In Phoenix, the downtown cluster is complemented by others in the Central Avenue Corridor, the Biltmore District, and downtown Scottsdale.

The spatial distribution of residents is completely different. As shown in the bottom row of Figure 1, in both Boston and Phoenix residents are spread throughout the metro area, and they are much more evenly distributed than jobs are. Phoenix is noticeably less densely settled than Boston, and Boston shows something of a density gradient while Phoenix does not. But the differences in settlement patterns between the two cities are much smaller than the differences between the spatial patterns of residences and jobs within each metro.

The spatial concentration of employment compared to population is present across the 100 largest MSAs. In all but five metro areas, the median employment density is higher than the median residential density, despite the fact that all MSAs have far more residents than jobs. This is the first major contribution of the paper: while employment is no longer centralized in downtowns (if it ever was), jobs remain far more spatially concentrated than residents. This finding contrasts with previous studies arguing that jobs and residents are similarly dispersed (Glaeser and Kahn, 2001; Wheaton, 2004).

Quantifying the concentration of employment

To quantify the phenomenon of concentration without centralization I adapt the methods used by Wheaton (2004). I first replicate his analysis of centralization, plotting the cumulative distribution of jobs and residents by distance from the CBD (as identified in Supplemental Table S4). The left panel of Figure 2 plots these distributions for the Boston MSA. In line with Wheaton’s results, employment and population show similar patterns: jobs are more centralized than residents, but not much more.

Figure 2.

Cumulative distribution of jobs and residents by distance from the CBD (left) and density (right), Boston MSA.Source: Author's analysis of LODES and Decennial Census data.

Yet the similarity in this measure of centralization clearly fails to capture the substantial difference between the patterns in the top and bottom rows of Figure 1. It is clear that the spatial distributions of employment and population are very distinctive, and if that is not captured by the centralization measure used by Wheaton an additional measure may be necessary. As an alternative, I use a density-based concentration measure, where instead of ordering blocks by their distance from the CBD I order them by their density of residents or jobs. The right panel of Figure 2 plots this density concentration curve for the Boston MSA. It shows a clear distinction between the spatial profiles of employment and population, matching that shown in Figure 1. Roughly 92% of Boston’s jobs are on the densest 10% of built land, compared to only about 66% of its residents. As shown in Supplemental Figure S1, this pattern is consistent across the 100 largest MSAs.

Distinguishing employment, residential, and mixed-use blocks

Most past investigations of intra-metropolitan employment distributions have conceptualized the spatial distribution of jobs as a continuous density surface, with subcenters defined as local maxima. The maps in Figure 1 call this approach into question. While residents do appear to form something of a density surface, with areas of high and low density but a continuous presence across most of the metro, employment has a clear bimodal distribution. A few areas have extremely high employment densities, while most parts of the metro have hardly any jobs at all. In fact, in the median MSA, 65% of built blocks representing 51% of the built land area have no jobs whatsoever.³ In contrast, no MSA lacks residents on more than 19% of its built land area. Residents form an almost continuous presence across the built portion of all 100 metro areas; jobs do not.

These two facts—that jobs are tightly concentrated, and that the majority of blocks in every major city have no jobs at all—suggest that the most productive way to characterize the spatial distribution of employment may be not as a continuous surface but as a set of discrete categories. As Shearmur et al. (2007) note in their study of Canadian metros, the majority of jobs in every city are found in a number of discrete “employment zones” scattered throughout a much larger residential area. At least at the block level, it makes sense to think of cities as composed of islands of employment in a sea of housing. Although a great deal of employment today is found in large-scale suburban office parks and shopping centers, even these occupy only a relatively small proportion of the built-up land in most cities, and they tend to be clustered together.

To formalize this observation I follow Forstall and Greene (1997) and Shearmur et al. (2007) in using the employment–population ratio within each block to classify it into one of three categories. Employment blocks are defined as those with more than twice as many jobs as residents. Residential blocks, conversely, have over twice as many residents as jobs. The remaining mixed-use blocks have comparable numbers of jobs and residents.

Figure 3 plots the proportion of jobs and built land area contained in each category of blocks for the 100 largest MSAs. In each MSA, about 75% of jobs are located in employment blocks occupying 5–10% of the built land. A further 10–20% of jobs are located in mixed-use blocks, which on average occupy about 5% of the land area. And only about 10% of jobs are found on the residential blocks that comprise the vast majority of each metro area. These proportions are remarkably consistent across cities as diverse as New York, Fresno, Pittsburgh, and Des Moines. As shown in Supplemental Figure S3, this consistency reflects a strong correlation between median employment and median residential densities across MSAs.

Figure 3.

Percentage of land and jobs in employment, residential, and mixed-use blocks, 100 largest MSAs.Source: Author's analysis of LODES and Decennial Census data.

Identifying business districts

Having shown that most jobs across the 100 largest MSAs are found in a relatively small number of employment blocks, there is still the question of whether these blocks are bunched together in large subcenters or spread out in “edgeless cities.” My second major contribution addresses this question by introducing an algorithm to identify business districts as groups of contiguous employment blocks. I group contiguous blocks together into business districts using the GeoPandas and PySAL packages in Python (Rey and Anselin, 2007). Because Census blocks can be very small and are often separated by nothing more than a street, I allow for a buffer, counting blocks as contiguous if they come within 15 meters of each other at any point. The choice of 15 meters appears to strike the best balance between grouping close-together blocks separated by small gaps such as street medians while avoiding sprawling districts with high internal variation. Further details of the specification choices are discussed in the Supplemental material.

Employment blocks are heavily clustered. More than 90% of all employment block jobs are found in business districts comprised of more than one contiguous block, and more than 77% of such jobs are found in districts of more than five. This overall pattern holds across all 100 MSAs: between 59 and 89% of each metro’s employment area jobs are found in clusters of at least five employment blocks. The probability of this concentration occurring by random chance is miniscule, as shown in Supplemental Figure S4.

The algorithm I use to identify business districts as contiguous groups of employment blocks has several analytical advantages over subcenter identification algorithms based on employment density alone, in addition to being conceptually simple and straightforward to implement. First, this algorithm can easily and simultaneously identify business districts in suburbs and central cities alike, a task previous algorithms have struggled with. This flexibility is demonstrated in Figure 4, which maps employment areas in the New York MSA. Identified business districts are outlined in black and shaded by type (see below). The algorithm clearly identifies Midtown and Lower Manhattan, downtown Brooklyn, Long Island City, and downtown Newark as business districts, while avoiding places like the Lower East Side and the Upper East and West Sides, which have employment densities higher than many suburban business districts but are predominantly residential neighborhoods.

Figure 4.

Business districts in the New York MSA.

Particularly striking is that this algorithm is able to identify both Midtown Manhattan and the industrial areas in the Meadowlands of New Jersey as business districts while avoiding the dense residential area between them. Midtown has an overall employment density of roughly 160,000 jobs per square kilometer, while the Harmon Cove industrial area in Secaucus has a density of just 3300 jobs per square kilometer, barely 2% as high. Yet both areas are clearly dominated by employment—each has more than 15 times as many jobs as residents within its boundaries—and both are major employment centers for the metropolitan region (Midtown is the single largest business district in the country, while Harmon Cove, with more than 30,000 jobs, is the 13th largest business district in the New York MSA). As shown in Supplemental Figures S5 and S9, an algorithm employing a fixed density threshold would be unable to identify both areas while avoiding the heavily developed but primarily residential northern portion of Manhattan. An algorithm based on density residuals would similarly be unlikely to identify both areas as subcenters given their close proximity but vastly different densities.

A second advantage of this algorithm is that it returns areal units rather than single points. This allows the characteristics of each business district to be exactly quantified and studied. The size and location of business districts can be mapped, allowing for quick visual assessment of the distribution of employment within a city. Supplemental Figures S6 and S7 show such maps for Denver and Atlanta. As shown in Supplemental Table S1, most cities have 10–15 large business districts that contain at least 1% of total metro employment. In the median MSA, the largest business district accounts for 10.5% of total employment, though that fraction ranges from 3% in Detroit to 30% in San Jose.

A third advantage of this algorithm is that because it applies exactly the same algorithm to every MSA, business districts can be compared across cities. Previous approaches have used local tuning parameters, so results from one metro are not comparable to those from another. Supplemental Table S2 provides summary statistics on the 20 largest business districts in the country. In terms of pure size, Midtown Manhattan, at just over 1.1 million jobs, is the largest business district in the country. It is followed by the Chicago Loop (413,000 jobs), downtown Washington DC (359,000 jobs), Lower Manhattan (330,000 jobs), and downtown San Francisco (314,000 jobs).

Business district categorization

While cities are remarkably consistent in the percentage of jobs found in business districts, they vary a great deal in the form these business districts take. Here I identify four main types of business district based on size and density. I first distinguish between large and small clusters of employment blocks. As shown above, most jobs are found in clusters of at least five contiguous employment blocks. These large clusters can be properly thought of as true “districts:” distinct areas or neighborhoods that have some cohesive identity. In contrast, isolated clusters smaller than five blocks tend to be either the locations of a single dominant employer or small commercial areas unknown outside their immediate area.

I further divide clustered and isolated districts into two types each. Because density is so central to the experience of urban environments, I use it to differentiate among clustered business districts. I adopt a density threshold of 10,000 jobs per square kilometer as a demarcation between urban and suburban business districts. This is equivalent to approximately 150 jobs in a city block, a typical density for low-rise but walkable commercial streets, such as the main street of a small town or a neighborhood shopping area (Puget Sound Regional Council, 2015; UrbanFootprint, 2016). If at least 75% of a business district’s jobs are at this density, I label that district an “urban core.” When people imagine downtowns of skyscrapers, or even local, walkable shopping streets, this is what they have in mind. In total, urban cores make up just 0.1% of the built land area of the 100 largest MSAs but contain 16% of the jobs. However, they vary in importance across metros, containing around 30% of all jobs in some MSAs while not even existing in others. As shown in Supplemental Table S4, an MSA’s historic CBD is almost always contained within its largest urban core business district. An important concept in urban analysis, the CBD has been challenging to operationalize—especially since the largest concentration of employment today may be in the suburbs, far from the historic downtown. The algorithm introduced here offers a systematic and reproducible way of determining the CBD location.

Second are large but sparse business districts, which I label “suburban strips.” With at least five blocks grouped together but employment densities below 10,000 jobs per square kilometer, these are the commercial and industrial developments that line the interstates leading out of many cities. They include office parks, warehouses, factories, and big box stores that typically have their own parking lots but form continuous stretches of economic activity that can go on for miles.

Suburban business districts are by far the dominant employment form in the US, containing a full 44% of all metropolitan jobs on about 3% of built land area. Notably, even though they cover 27 times as much land area as urban cores, they still occupy only a tiny portion of the built area of most cities. While urban cores vary greatly in importance across the 100 largest MSAs, suburban strips are an important presence in all 100 metros, always containing at least 23% of all employment.

Urban core and suburban business districts tend to be pedestrian- and auto-oriented, respectively, but a classification based purely on density will not always map perfectly to land use or design characteristics. There can be car-oriented developments that are nonetheless quite dense, and downtown-style developments with low employment densities—particularly in places with high vacancy rates. My density-based classifications are thus best thought of as a heuristic rather than a definitive description of built form.

Because isolated districts are composed of so few blocks, density is less useful than sheer size in capturing their character. Some of these isolated districts are truly massive, even though they occupy just a few blocks. For example, the headquarters of 3M in the suburbs of St. Paul, Minnesota, forms an isolated one-block business district. But it still is home to more than 17,000 jobs, making it the 11th largest business district in the Minneapolis MSA. I label isolated business districts with more than 500 jobs as “independent” districts. Though they are not always well integrated into the urban fabric, independent districts still heavily influence the employment landscape of their cities. In many cases they contain hospitals or universities that employ huge numbers of people and may serve as anchor institutions for surrounding neighborhoods. In total, these independent districts account for 10% of metropolitan employment on 0.5% of the built land area.

Finally, many isolated business districts are simply a few blocks with a small number of jobs. These “scattered” districts of fewer than five blocks and fewer than 500 jobs account for the numerical majority of business districts, and about half of the land area contained in employment blocks (3% of built land). However, they contain only 7% of jobs.

The shading in Figure 4 corresponds to the categories of business district. The black business districts are the traditional urban cores: places like Midtown Manhattan, Lower Manhattan, downtown Brooklyn, and downtown Newark that were the original centers of economic activity in the region. Suburban districts are shaded in light gray, in this case largely encompassing the industrial portions of New Jersey and New York City. Dark gray areas are independent centers, including La Guardia Airport and Columbia University. And white areas are scattered business districts with limited employment.

In sum, all jobs in the 100 largest MSAs fall into one of six types of area. About 10% of jobs are in residential blocks, where they are outnumbered more than two to one by residents. Another 13% are in mixed areas with similar numbers of jobs and residents. The employment areas that contain the remaining 77% of jobs can be divided into four types of business district with very different characters: traditional urban cores (16% of employment), suburban strips (44%), independent centers (10%), and scattered blocks (7%). Supplemental Table S3 shows the exact breakdown of jobs, residents, and built land for all 100 MSAs together.

Although the national numbers are informative, there is a huge amount of variation among MSAs in how jobs are distributed across the various types of employment district, as shown in Supplemental Figure S8. For instance, 3 of the 100 largest MSAs have no business districts that qualify as a traditional urban core, while New York has 104, which collectively account for over quarter of its employment.

Discussion

In this paper I have sought to determine how employment is spatially distributed within US metro areas. In investigating this research question, I have made two major contributions to knowledge of metropolitan spatial structure. First, I have shown definitively that US metropolitan jobs are neither centralized nor dispersed. Rather, they are heavily concentrated into comparatively small business districts that nonetheless exist throughout metropolitan areas. Across the 100 largest MSAs, about 75% of jobs are contained in the 6.5% of built land area with more than twice as many jobs as people. This is a sharp contrast to the spatial patterns of residents, who are spread more evenly across a much larger area. This pattern of concentration without centralization confirms the inadequacy of the monocentric city model for today’s cities, and also highlights the limitations of models that only consider a location’s distance from downtown.

Because the majority of built land contains no jobs at all, and because the proportions of land devoted to employment and residences are remarkably consistent across the 100 largest MSAs, I have argued that modeling employment as a continuous density surface can be misleading. Instead, I have reintroduced the employment–population ratio as a method of classifying blocks according to their land use. My algorithm groups contiguous employment blocks together into business districts, which I further categorize based on their size and density.

This algorithm, which is my second major contribution, is both more consistent and more flexible than most previous methods of subcenter identification. Because it identifies districts based on predominant land use rather than pure employment density, it is able to perform well in urban and suburban areas simultaneously. Further, since the exact same algorithm is used in all metros without any local tuning parameters, the identified business districts are comparable across cities, allowing for clear statements about the relative size of business districts in places as diverse as New York and Boise.

The business district identification algorithm introduced here has limitations. Most notably, it relies on two global parameters, the ratio of employment to residents above which blocks are deemed employment blocks, and the buffer distance below which two blocks are considered contiguous. There is no theoretical basis for choosing either parameter, and while my overall results are generally robust, individual districts can be sensitive to the buffer distance in particular.

It is also worth stating that my concept of a business district defined by commercial rather than residential land use is slightly distinct from that of an employment subcenter defined by job density. However, both approaches capture similar numbers of jobs, trading off jobs in high-density residential areas against those in low-density employment areas. It is thus debatable which best captures the spatial structure of employment. I choose land use for its greater analytical simplicity and comparability across metro areas.

The approach described in this paper invites several lines of future research. I have limited my analysis to employment patterns in a single year, 2014. Further analysis could explore changes in the composition or spatial distribution of employment over time. I have also considered only the number of jobs, not the type. Job counts in the LODES data are provided by two-digit NAICS industry code, allowing future researchers to determine whether hospitality jobs, for instance, are more likely to be found in suburban or urban core business districts. Finally, differences in commuting patterns by business district category or size might be explored using the LODES origin–destination data.

The employment geography of the United States combines incredible local intricacy with broad consistency at the metropolitan level. High-resolution administrative data are able to document these patterns in unprecedented detail. As shown in this paper, these data offer a new perspective on longstanding debates and suggest updates to how metropolitan spatial structure is theorized and modeled.

Supplemental Material

sj-pdf-1-epb-10.1177_2399808320934821 - Supplemental material for The spatial structure of US metropolitan employment: New insights from administrative data

Supplemental material, sj-pdf-1-epb-10.1177_2399808320934821 for The spatial structure of US metropolitan employment: New insights from administrative data by Robert Manduca in Environment and Planning B: Urban Analytics and City Science

Footnotes

Acknowledgements

I am grateful to Luc Anselin, Daniel Arribas-Bel, Philip Chodrow, Julia Koschinsky, David Plane, James Saxon, and participants in the Spatial Structures in the Social Sciences workshop at Brown University, the Study Group in Spatial Analysis at the University of Chicago, and the Western Regional Science Association 2018 Annual Meeting for their helpful comments and feedback, and to Matthew Graham and the Longitudinal Employer-Household Dynamics Program at the US Census Bureau for answering my questions about the LODES dataset.

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was supported by the Harvard Multidisciplinary Program in Inequality and Social Policy.

ORCID iD

Robert Manduca

Supplemental material

Supplemental material for this article is available online.

Notes

Robert Manduca is an Assistant Professor of Sociology at the University of Michigan. His research focuses on income inequality and urban and regional economic development.

References

Alonso

(1964) Location and Land Use: Toward a General Theory of Land Rent. Cambridge, MA: Harvard University Press.

Anas

Arnott

Small

(1998) Urban spatial structure. Journal of Economic Literature 36(3): 1426–1464.

Anselin

(1995) Local indicators of spatial association – LISA. Geographical Analysis 27(2): 93–115.

Arribas-Bel

Ramos

Sanz-Gracia

(2015) The size distribution of employment centers within the US metropolitan areas. Environment and Planning B: Planning and Design 42(1): 23–39.

Ban

Arnott

Macdonald

(2017) Identifying employment subcenters: The method of exponentially declining cutoffs. Land 6(1): 17.

Baumont

Ertur

Gallo

(2004) Spatial analysis of employment and population density: The case of the agglomeration of Dijon 1999. Geographical Analysis 36(2): 146–176.

Bogart

Ferry

(1999) Employment centres in greater Cleveland: Evidence of evolution in a formerly monocentric city. Urban Studies 36(12): 2099–2110.

Burgess

(1925) The growth of the city: An introduction to a research project. In: Park RE and Burgess EW (eds) The City. Chicago, IL: University of Chicago Press.

Cervero

K-L

(1997) Polycentrism, commuting, and residential location in the San Francisco Bay area. Environment & Planning A 29(5): 865–886.

10.

Christaller

(1933) Central Places in Southern Germany. Englewood Cliffs, NJ: Prentice-Hall.

11.

Craig

Kohlhase

Perdue

(2016) Empirical polycentricity: The complex relationship between employment centers. Journal of Regional Science 56(1): 25–52.

12.

Craig

(2001) Using quantile smoothing splines to identify employment subcenters in a multicentric urban area. Journal of Urban Economics 49(1): 100–120.

13.

Forstall

Greene

(1997) Defining job concentrations: The Los Angeles case. Urban Geography 18(8): 705–739.

14.

Fujita

Ogawa

(1982) Multiple equilibria and structural transition of non-monocentric urban configurations. Regional Science and Urban Economics 12(2): 161–196.

15.

Galster

Hanson

Ratcliffe

, et al. (2001) Wrestling sprawl to the ground: Defining and measuring an elusive concept. Housing Policy Debate 12(4): 681–717.

16.

Garcia-López

M-À

Muñiz

(2010) Employment decentralisation: Polycentricity or scatteration? The case of Barcelona. Urban Studies 47(14): 3035–3056.

17.

Garreau

(2011) Edge City: Life on the New Frontier. New York: Anchor Books.

18.

Giuliano

Small

(1991) Subcenters in the Los Angeles region. Regional Science and Urban Economics 21(2): 163–182.

19.

Glaeser

Kahn

(2001) Decentralized employment and the transformation of the American city. Brookings-Wharton Papers on Urban Affairs 2: 1–63.

20.

Gordon

Richardson

(1996) Beyond polycentricity: The dispersed metropolis, Los Angeles, 1970–1990. Journal of the American Planning Association 62(3): 289–295.

21.

Guillain

Le Gallo

Boiteux-Orain

(2006) Changes in spatial and sectoral patterns of employment in Ile-De-France, 1978–97. Urban Studies 43(11): 2075–2098.

22.

Hajrasouliha

Hamidi

(2017) The typology of the American metropolis: Monocentricity, polycentricity, or generalized dispersion? Urban Geography 38(3): 420–444.

23.

Henderson

Mitra

(1996) The new urban landscape: Developers and edge cities. Regional Science and Urban Economics 26(6): 613–643.

24.

Ladd

Wheaton

(1991) Causes and consequences of the changing urban form: Introduction. Regional Science and Urban Economics 21(2): 157–162.

25.

Lang

(2003) Edgeless Cities: Exploring the Elusive Metropolis. Washington, DC: Brookings Institution Press.

26.

Lang

Sanchez

Oner

(2009) Beyond edge city: Office geography in the new metropolis. Urban Geography 30(7): 726–755.

27.

Lee

(2007) “Edge” or “edgeless” cities? Urban spatial structure in US metropolitan areas, 1980 to 2000. Journal of Regional Science 47(3): 479–515.

28.

Leslie

Ó hUallacháin

(2006) Polycentric phoenix. Economic Geography 82(2): 167–192.

29.

Losch

(1954) Economics of Location. New Haven, CT: Yale University Press.

30.

McDonald

(1987) The identification of urban employment subcenters. Journal of Urban Economics 21(2): 242–258.

31.

McMillen

(2001a) Nonparametric employment subcenter identification. Journal of Urban Economics 50(3): 448–473.

32.

McMillen

(2001b) Polycentric Urban Structure: The Case of Milwaukee. Chicago, IL: Federal Reserve Bank of Chicago.

33.

McMillen

Lester

(2003) Evolving subcenters: Employment and population densities in Chicago, 1970–2020. Journal of Housing Economics 12(1): 60–81.

34.

Manson

Schroeder

Van Riper

, et al. (2017) IPUMS National Historical Geographic Information System: Version 12.0. Minneapolis: University of Minnesota.

35.

Matsuo

(2011) US metropolitan spatial structure and labour accessibility. Urban Studies 48(11): 2283–2302.

36.

Puget Sound Regional Council (2015) Transit-supportive densities and land uses. Available at: https://www.psrc.org/sites/default/files/tsdluguidancepaper.pdf (accessed 17 January 2020).

37.

Redfearn

(2007) The topography of metropolitan employment: Identifying centers of employment in a polycentric urban area. Journal of Urban Economics 61(3): 519–541.

38.

Rey

Anselin

(2007) PySAL: A python library of spatial analytical methods. The Review of Regional Studies 37(1): 5–27.

39.

Riguelle

Thomas

Verhetsel

(2007) Measuring urban polycentrism: A European case study and its implications. Journal of Economic Geography 7(2): 193–215.

40.

Rodríguez-Gámez

Dallerba

(2012) Spatial distribution of employment in Hermosillo, 1999–2004. Urban Studies 49(16): 3663–3678.

41.

Rossiter

(2011) What are census blocks? Random Samplings. Available at: https://www.census.gov/newsroom/blogs/random-samplings/2011/07/what-are-census-blocks.html (accessed 16 January 2020).

42.

Shearmur

Coffey

Dube

, et al. (2007) Intrametropolitan employment structure: Polycentricity, scatteration, dispersal and chaos in Toronto, Montreal and Vancouver, 1996–2001. Urban Studies 44(9): 1713–1738.

43.

UrbanFootprint (2016) UrbanFootprint Place Types. Los Angeles. Available at: https://planning.lacity.org/eir/SCEAs/documents/WeingartProjects/appendices/C_Urban.pdf (accessed 10 January 2020).

44.

US Bureau of Labor Statistics (1997) Ch. 5, Employment and wages covered by unemployment insurance. In: BLS Handbook of Methods. Washington, DC: US Department of Labor, pp.42–47.

45.

US Bureau of Labor Statistics (2017) Civilian Employment Level [CE16OV]. 23 September. FRED, Federal Reserve Bank of St Louis. Available at: https://fred.stlouisfed.org/series/CE16OV.

46.

US Census Bureau (2017) LEHD Origin-Destination Employment Statistics Data (2002–2015). Washington, DC: US Census Bureau, Longitudinal Employer-Household Dynamics Program.

47.

US Census Bureau (2019) 2010 census tallies. Available at: https://www.census.gov/geographies/reference-files/2010/geo/tallies.html#par_textimage_1 (accessed 16 January 2020).

48.

Vega

Reynolds-Feighan

(2008) Employment sub-centres and travel-to-work mode choice in the Dublin region. Urban Studies 45(9): 1747–1768.

49.

Von Thünen

(1966) Isolated State. New York: Pergamon Press.

50.

Wheaton

(2004) Commuting, congestion, and employment dispersal in cities with mixed land use. Journal of Urban Economics 55(3): 417–438.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

1.16 MB