Understanding Divergence in the Performance of Central Business District Economies Among U.S. Metropolitan Regions,1995

Abstract

Although references to an urban resurgence are pervasive, on average, the share of regional jobs found in central business district (CBD) areas has fallen since the 1990s. Yet, this fact masks great variation in the performance of CBD economies. In several metropolitan regions, the share of jobs found in CBD areas increased over this period. This paper seeks to understand the factors that explain such variation. Relying on a panel of metropolitan regions for the period 1995 to 2019, the paper finds that, as a regional economy increasingly comprises tasks that are more cognitive in nature, the share of its jobs found in CBD areas increases. Although the “great divergence” has been linked to an increasing dispersion of incomes, health care outcomes, and cultural identities among regions, the paper's findings suggest it may also be associated with divergent economic performance among CBD areas.

Keywords

urban resurgence central business districts urban form

Since the turn of the century, references to an “urban resurgence” have been pervasive (Cheshire, 2006; Glaeser & Gottlieb, 2006; Manville & Storper, 2006). In the United States, the term “urban resurgence” is typically used to refer to two distinct geographic scales. The first refers to a reversal in the fortunes of certain principal cities in the Northeast and Midwest, which after decades of decline, experienced improved performance in the 1990s (Glaeser & Gottlieb, 2006; Manville & Storper, 2006). The second refers to the resurgence of central city areas (Manville & Storper). The evidence reveals that resurgence is far from widespread in either case (Frey, 2012; Glaeser & Gottlieb, 2006; Osman, 2022). Although New York City and Boston are prominent examples of the revival of deindustrialized cities, since the 1990s, jobs and population levels have continued to decline in cities such as Buffalo, Cleveland, Detroit, and Pittsburgh (Frey, 2012; U.S Census Bureau, 2020a). With respect to central city areas, consider the performance of central business district (CBD) economies. Although there are notable success stories, in most regions, CBD population and employment growth have greatly lagged growth in non-CBD areas since the 1990s (Osman, 2022). This paper focuses on the urban resurgence of central city areas, seeking to understand the factors that account for the divergent performance of CBD economies across the largest metropolitan regions in the United States between 1995 and 2019.

Although central city areas were once the hub of regional economies in the United States, today they account for the great minority of residents and jobs in most regions (Baum-Snow, 2014; Boustan & Shertzer, 2012; Gardner & Marlay, 2013; Hartley et al., 2016; Kneebone, 2013). In response to decentralization within metropolitan regions, the demise of central city economies, and the impact of decentralization on inner-city residents, policy makers at all levels of government have targeted the revitalization of central city areas for decades (Euchner & McGovern, 2003; Teaford, 1990). Federal efforts at urban renewal began in the late 1940s. They have been well analyzed and documented (e.g., Euchner & McGovern; Teaford). Three decades of programs targeting inner-city housing and poverty mostly became a mechanism for the redevelopment of downtown business districts (Euchner & McGovern). As the federal government began withdrawing from urban policy in the 1970s, local leaders started targeting the development of CBDs. Waves of local policies to encourage CBD renewal included fiscal incentives for firms, as well as a variety of measures to entice out-of-town visitors, such as convention centers, sports stadia, festival marketplaces, and waterfront development (Euchner & McGovern).

Despite such policy attention, since the 1990s much of the population and employment gains in metropolitan regions have occurred outside of CBDs (Hartley et al., 2016; Osman, 2022). According to data analysis in this paper, in the 100 largest U.S. metropolitan regions in 2019, CBD areas accounted for 9.6% of jobs and 3% of residents, on average. Analyses of central cities––defined as the principal cities of metropolitan regions that cover geographies larger than CBD areas––shed light on the extent of decentralization over time. In 1950, 58% of metropolitan area residents lived in central cities, compared to 36% in 2000 (Boustan & Shertzer, 2012), while Baum-Snow (2014) estimated that the share of regional employment in central cities fell from 61% in 1960 to 34% in 2000. Across the 100 largest metropolitan regions in the United States, CBD employment has grown since the 1990s, on average, but at a slower rate than in the non-CBD areas of these regions. For example, Osman (2022) found that, for the 100 largest metropolitan regions over the period 1994 to 2019, CBD employment grew by 24%, on average, compared to 46% in non-CBD areas, and that CBD areas accounted for only 6.2% of the net job growth in these regions over the period. In around a third of these CBD areas, job losses occurred. However, in some CBD areas, the rate of job growth has exceeded growth in non-CBD areas.

Although urban resurgence can be defined in several different ways (see Teaford, 2000), population and employment growth in some central city areas is the most used metric of resurgence. This paper focuses on the economies of CBD areas, the historic cores of regional economies, and considers changes in the share of jobs located in the CBD economies of the 100 largest U.S. metropolitan regions¹ over the period 1995 to 2019. This measure of resurgence gauges the relative performance of CBD areas and the extent to which they have added jobs at a quicker rate than the rest of their regional economies. Across the 100 largest regions, the share of jobs in CBD areas fell from 11% in 1995, to 9.6% in 2019, on average. However, this aggregate decline masks great variation in CBD performance. The share of regional jobs found in CBD areas increased in 14 of these regions, where the share of jobs in CBD areas increased from 13.8% to 15.6% over this time, on average. In fact, just 10 metropolitan regions account for the equivalent of 81% of the net CBD job gains across the 100 largest regions since 1994. The primary contribution of this paper is to explore why the performance of CBD economies, measured as the share of regional employment found in CBD areas, differs among regions, over the period 1995 to 2019. Although there is research into the factors that contribute to employment decentralization within regions (Baum-Snow, 2014; Glaeser & Kahn, 2001), there is little work considering the factors that might account for employment growth in CBD areas.

Variation in CBD performance has occurred against the backdrop of a renewed divergence in the performance of U.S. regional economies (Moretti, 2012; Storper et al., 2015). Although during much of the 20th century incomes among U.S. regions converged, this trend began to unravel in the 1980s. Since then, a handful of large, “superstar” metropolitan regions have outperformed smaller regions, many deindustrialized regions and rural locations. It is widely held that regional specialization is at the root of this divergence. The shift from a manufacturing-based economy to an information-based “new economy” has created “winner” and “loser” regions (Glaeser, 2011; Moretti, 2012; Storper et al., 2015). The key feature of the new economy is the extent to which its core activities rely on tasks performed by highly skilled workers who typically have higher levels of formal education (Moretti).

This shift in the nature of economic activities is relevant to this paper since many new economic activities are theorized to thrive in dense urban environments (Duranton & Puga, 2015; Glaeser, 2011). It should be expected, therefore, that as a region's industrial specialization increasingly comprises new economic activities, the share of employment found in its densest areas, such as CBDs, will increase. As such, the primary research question of this paper is: Is the nature of regional specialization a key contributor to variation in the performance of CBD economies among U.S. regions over the period 1995 to 2019? The statistical analysis employs two primary estimators (ordinary least squares and 2SLS) for a panel of 100 metropolitan statistical area (MSA) regions over the period 1995 through 2019. The analysis employs two measures of regional industry specialization: the extent to which regional economies comprise tasks that are cognitive and manual in nature. This study finds that as a region's economy comprises more cognitive activities, the share of its jobs found in CBD areas increases, when controlling for other factors that might account for CBD performance.

Literature Review

For more than 100 years, observers have repeatedly declared the death of American cities (Manville & Storper, 2006), but the United States is fundamentally an urban nation and has been for many decades. Since 1920, more than 50% of the nation's population has lived in urbanized areas (U.S. Census Bureau, 2020b). In 2019, 86% of the nation's population was found in one of the nation's 384 metropolitan regions (U.S. Census Bureau, 2020a). The fall of the American city more accurately referred to the demise of central city areas, and the decline of many cities in the Northeast and the Midwest, which came to prominence during the industrial revolution (Manville & Storper). The extent of urbanization in the United States suggests that, despite the negative externalities associated with urban areas, they remain the most efficient way to organize the economy and society (Glaeser, 2011). Yet within urbanized areas, there has been a pronounced decentralization of jobs and people since World War II.

Theory identifies four primary factors that could influence the location of economic activity within regional economies: transportation costs, agglomeration economies, Tiebout sorting, and the location preferences of workers. Each of these factors could influence the performance of CBDs within regional economies. In the foundational models of urban economics, the minimization of transportation costs governs firm location within a city. In these early models, firms outbid residents to locate in the center of cities, which are at the hubs of inter- and intraregional transportation networks. Central locations reduce firms’ costs of shipping to, and receiving from, nonlocal markets. It also reduces the commuting costs of workers, maximizing welfare for firms and workers alike (Alonso, 1964; Duranton & Puga, 2015). Although monocentric cities were prevalent in the country throughout the 19th century, as the 20th century unfolded, city development increasingly expanded outward and became less anchored to cities’ original cores (Duranton & Puga, 2015; Glaeser & Kohlhase, 2003).

Several modeling extensions incorporated polycentric cities into models of urban economics. Originally, these extensions allowed for more than one exogenously located employment center within cities, preserving the transportation cost framework of foundational models, where firms trade CBD access for lower land costs (White, 1976). Over time, models of urban structure incorporated the endogenous actions of firms. In some models, for example, firms in certain industries locate within cities to reduce transaction costs and enhance access to information spillovers (Duranton & Puga, 2015; Lucas & Rossi-Hansberg, 2002; Ogawa & Fujita, 1980). In such models, density facilitates information exchange and lowers transaction costs, but localized clustering generates diseconomies of agglomeration. Firms, therefore, locate within cities depending on the nature of their core activities. Firms for which information is a key input of production, and for which localized external economies of scale are key to their operations, benefit from proximity to other firms and can support the higher costs of density. Yet, these same costs lead firms that are less reliant on such factors to decentralize within cities to locations where land costs are lower; they can compensate workers less for their commutes and they can still benefit from region-wide agglomeration effects. This leads to the emergence of additional, smaller employment centers within cities. According to these models, therefore, firms locate within a city with respect to the benefits they gain from, and the spatial decay of, agglomeration economies, their intensity of land use in production, and the minimization of commuting costs (Duranton & Puga).

As has been widely documented, over the past 40 years or so, manufacturing industries have declined in relative importance, and knowledge sector activities have become the drivers of growth and wealth for regional economies in the United States (Moretti, 2012; Storper et al., 2015). One of the most striking features of this transition is the extent to which emergent sectors of the economy are highly localized, engendering renewed income divergence among regional economies (Moretti; Storper et al.). Variation in worker skills is one of the most common indicators of differences in economic activities among regions. Moretti showed that the share of residents who hold at least a college degree in Stamford, Connecticut, is five times greater than in Merced, California, which has the lowest share. In the 10 regions with the highest degree of educational attainment, at least 42% of residents hold a bachelor's degree, while in the 10 least-educated regions, less than 15% of residents hold a university degree. Furthermore, the skill gap among regions has increased over time (Diamond, 2016). Such skill sorting is the expression of the geography of knowledge sectors of the economy, which due to the nature of their core activities, are highly agglomerated (Moretti; Storper et al.). Given the uneven dispersion of knowledge economy activities and the tendency of such industries to agglomerate at a highly localized scale within regions, this paper examines the extent to which the dispersion of knowledge industries across regional economies has contributed to the divergent performance of CBD economies among regions.

There are two other factors that could influence CBD performance. The first is local government fragmentation. Local government fragmentation is a key feature of regional economies in the United States, where local governments have more autonomy, especially over land use and local taxation, than their counterparts in many other parts of the world. There were 90,126 local governments in the United States in 2017, driven by local government fragmentation within metropolitan regions (U.S. Census of Governments, 2019). Such fragmentation could influence the share of regional employment found in CBD areas in two ways. First, local fragmentation can lead to greater competition among local governments for mobile capital. Such interjurisdictional competition could lead to a lower share of jobs located in CBD areas, if local competition for business disperses economic activity within a region. Second, fragmentation can facilitate Tiebout-sorting effects whereby, absent outright competition among local governments, a diversity of public goods could emerge across jurisdictions within a region. This process could encourage firms to locate in communities that provide the best match of public goods for the level of taxes they are willing to pay (Tiebout, 1956).

Finally, the location preferences of workers could influence job location within regions. Following World War II, the suburbanization of populations was one of the defining features of metropolitan regions in the United States (Euchner & McGovern, 2003). At the same time, a great decentralization of work occurred within metropolitan regions (Baum-Snow, 2014). Many jobs are dispersed within regions to provide services to suburbanized residents. Other firms moved to suburban locations to benefit from cheaper land and lower urban costs. However, in the 1990s, wealthier, more educated residents returned to central city locations for consumption-oriented lifestyle considerations (Florida, 2002; Glaeser et al., 2001). The return of skilled workers might have led certain firms to locate there for better access to their primary workforce (Glaeser, 2011).

There is not a large body of empirical work devoted to understanding the growth of CBD employment within cities, but some studies seek to understand the reasons for employment decentralization. Baum-Snow (2014) found that over the period 1960 to 2000, the supply of highways had a statistically significant effect on the decentralization of employment within the largest metropolitan regions in the United States. The addition of each new highway reduced job centralization by 6%. In a cross-section analysis, Glaeser and Kahn (2001) considered factors that affect the level of employment decentralization within U.S. metropolitan regions. They provided evidence that, as the share of a region's manufacturing labor force increases, employment decentralization increases. By contrast, as the regional share of service workers increases, there is less employment decentralization within regions. They also found that the number of jurisdictions within a region and a region's land area are both associated with employment decentralization, while the demographic factors they examined, notably race and education, do not have a statistically significant impact on employment decentralization. The remainder of this paper will consider how the factors identified here have influenced the performance of CBD economies in the United States since the 1990s.

Data and Methods

There is not a formal, universal definition of CBDs since they are not administrative units, but rather neighborhoods within cities. As such, their definition varies across studies (Glaeser & Kahn, 2001; Hartley et al., 2016; Landis, 2009). In the following analysis, I define CBDs based on the definitions of the 1982 Census of Retail Trade, which asked local business leaders to identify the CBDs within their respective cities. The U.S. Census Bureau assigned census tracts to delineate the areas described by respondents for 427 cities. I use these census tracts to identify the CBD centroid in each region.While dated, these CBD definitions have fared well over time and are used as a basis for defining CBDs in several studies (Glaeser & Kahn, 2001; Hartley et al., 2016; Kneebone, 2013). For each region, I define CBDs as those zip codes that have a centroid within a 2-mile radius of the CBD centroid.

Zip code boundaries (the basic geographic unit in this study) have significantly different shapes and sizes from census tracts, which render it infeasible to closely match census tract areas with zip code areas. Furthermore, the census tracts that defined CBDs in 1982 will not precisely define CBD boundaries today due to economic expansion and contraction within these parts of regional economies. In this analysis, zip codes are preferred to census tracts because the time series for zip code employment data (the Zip Business Patterns) begins 8 years earlier than the time series for census tract employment data (the Longitudinal Employer-Household Dynamics). Including zip codes witihin a 2-mile radius of the CBD centroid, therefore, is not intended to capture perfectly the shape of each CBD area, but rather approximate the historic, economic core of each region's economy. The data used in this study reveal that, on average, job density (jobs divided by land area) for the zip codes with a centroid within a 2-mile ring of CBD centroids is eight times denser than zip codes with a centroid within the 3- to 4-mile ring, and 27 times denser than in the ring that is 10 to 11 miles from CBD centroids. This relationship holds for cities across the size distribution. This provides further evidence that the 1982 definition of CBD areas provides a good basis for understanding CBD location today.

Based on the CBD definition employed here, if the land area of each CBD in this analysis were formed into the shape of a circle, the average radius of a CBD area would be 1.51 miles and the median radius would be 1.55 miles.² At the 10th percentile of the distribution, this figure would be 1.03 miles, and at the 90th percentile of the distribution, it would be 2.08 miles. Across the sample, there is variation in CBD land areas. While some of this is due to scale differences––some CBDs are found in larger regions, while some account for a larger share of jobs in a given region––variation is also due to the size of zip code areas, which can vary by region. This could lead to inflated CBD land areas, and potentially job counts, in regions that are home to large zip code boundaries. The average CBD in this analysis accounts for 0.27% of a region's total land area. At the 10th percentile of the distribution, the CBD accounts for 0.05% of the region's total land area, while at the 90th percentile, it accounts for 0.7% of the region's land area. To account for the potential that such variation could affect the subsequent statistical analysis (in other words, the CBD boundaries employed here could both underestimate and overestimate actual CBD sizes due to the uneven sizes of zip code boundaries among regions), I will test to what extent the paper's analysis is sensitive to the tails of the distribution of CBD land areas.

The economy of each CBD is situated within its respective MSA. According to the Office of Management and Budget, MSAs comprise a core county that has an urbanized area with a population of at least 50,000 residents, along with the surrounding counties with which it has strong commuting ties (to be considered part of the same region, at least 25% of the workers living in surrounding counties must work in the core county). This study considers the 100 largest MSAs based on their population in 2019, according to estimates from the U.S. Census Bureau's American Community Survey (ACS). Large MSAs are the focus of this study because there are too few zip codes in smaller regions to detect meaningful differences in employment location within regions (which is necessary to discern CBD areas from the rest of a regional economy). For example, for the 100 largest MSAs, there is an average of 124 zip codes per region, compared to 21 zip codes, on average, in regions that rank 100 to 200 by population. The average land area for an MSA in this analysis is 4,853 square miles, and the average CBD area is 7.69 square miles.

In the following analysis, I estimate which factors are associated with changes in the share of MSA jobs found in CBD areas for the 100 largest metropolitan regions in the United States over the period 1995 to 2019, using the following equation:
$Y_{c, t} = β_{0} + β_{1} l n (X_{c, t}) + μ_{c} + η_{t} + e_{c, t}$
(1)
where $Y_{c, t}$ is the share of MSA employment found in the CBD of MSA c in year t (total CBD employment is divided by total MSA employment for each year of analysis). $X_{c, t}$ is a vector of relevant time-varying MSA characteristics. μ is an MSA-specific fixed effect, the purpose of which is to capture unobserved but relatively stationary features of regions, while η is a year-fixed effect, included to capture time-varying but economy-wide shocks. Across the years of analysis, error terms are likely to be correlated within, but independent across MSA areas, so standard errors are clustered by MSA. The standard random error term is represented by $e_{c t}$ . The key parameter to be estimated is β1, which estimates the impact of different variables on the share of MSA jobs found in CBD areas.

CBD and MSA employment data are drawn from the U.S. Census Bureau's annually released County Business Patterns. The data are extracted from the Business Register, a database of all known single and multi-establishment employers in the country. In addition to county files, since 1994 the Census Bureau has released zip code level data (ZBP). For the zip codes that form each CBD, CBD employment counts are generated from the ZBP “totals” files, which have provided an annual count of total employment for each zip code since 1994. MSA employment counts are drawn from county files, according to the Office of Management and Budget boundaries.

Independent Variables

Regional Industrial Specialization

My primary motivation for this paper is to consider the extent to which variation in the nature of regional economies shapes the share of regional employment found in CBD areas. As described above, dense urban environments are theorized to facilitate information spillovers among workers, thus one might expect the extent to which a regional economy is comprised of knowledge intensive industries, for example, to influence the share of jobs located in CBD areas, which are typically the densest locations within a given MSA. To measure variation among regional economies, I exploit differences across three variables. First, among regional economies, the share of workers employed across the different sectors of the economy varies. Industries are defined according to the North American Industrial Classification System (NAICS) and, for each region, I draw industry data from the Quarterly Census of Employment and Wages. Second, each industry is comprised of a different mix of occupations. For the national economy, the Bureau of Labor Statistics' Occupational Employment and Wage Statistics (OEWS) measures the occupational composition of each industry (for example, the share of managers employed in each sector of the economy). Third, the tasks workers perform vary by occupation. The Department of Labor's ONET program identifies the tasks workers perform in each occupation (OEWS), such as the complex problem-solving skills required to perform a given occupation. From these variables, I create two different measures of differences in the nature of regional economies.

Acemoglu and Autor (2011) created a framework in which certain ONET variables are combined to identify the extent to which the activities of a given occupation are cognitive, meaning that their tasks require complex problem solving, intuition, persuasion, and creativity. They also identifiedthe extent to which occupations comprise manual tasks, which are more physical in nature, require less problem solving, and can be filled by workers with less formal education. I use these two measures to proxy for the extent to which regional economies comprise activities involving relatively complex tasks and the extent to which they comprise relatively lower-skilled activities (manual economy activities). Exploiting the NAICS-OEWS-O*NET nexus, for each region and for each year I calculate a cognitive and manual score, based on how workers are divided across industries (four-digt NAICS sectors) and occupations within industries. The cognitive and manual score for a given region-year is the weighted average based on the variables Acemoglu and Autor identified, according to the number of workers in each occupation within each industry and the regional share of workers in each industry.

CBD Population Characteristics

As described above, it's possible that, as workers move to CBD areas, employers will locate there to have better access to them. To account for this possibility, I include a variable that tracks the share of a MSA's population that lives in CBD areas. For the years 2011 through 2019, I use zip code level population estimates from the 5-year sample of the ACS, to count the number of residents who live in CBD areas. Annual population counts do not exist for the other years of analysis. Therefore, I use zip code level population estimates from the decennial censuses of 1990, 2000, and 2010, and use a linear interpolation function to estimate values for the missing years, providing an estimate for each year, 1995 through 2019. MSA population counts are also drawn from the ACS for the same period. To calculate the share of MSA residents who live in CBD areas, for each region, I divide the CBD popualtion by the MSA population, for each year.

It is possible that the characteristics of CBD populations influence employment growth in CBD areas. Specifically, I test whether the shares of CBD residents who hold a bachelor's degree or higher or are either Black or Hispanic, shape relative employment location in CBD areas. For each region, data for these varaibles are drawn from the ACS, relying on the same years and using the same interpolation methods as described above.

Urban Form

The nature of urban form in some regions might also shape relative CBD job growth. For example, the features of some CBD areas might be conducive to job growth in certain sectors of the economy. For example, firms in industries that seek a dense urban environment might be drawn to CBD areas that have high levels of job density. To account for differences in urban form among regions, I employ two measures. The first measure of urban form is the job density of CBD areas in 1994. The measure from 1994 is used to estimate to what extent a job-dense CBD at a prior point in time influences the trajectory of future relative growth in CBD areas. The second proxy is a dummy variable for the census division within which an MSA is located. There are nine census divisions (groups of neighboring states). The division in which an MSA is located is a proxy for urban form since, for neighboring states, major cities were developed during similar transportation eras. The cities of the New England division, for example, began to be developed prior to the widespread adoption of the automobile, making them denser and more compact relative to cities in the Mountain Division, which are more oriented around the automobile and, as such, are less dense.

Other Control Variables

It is possible that as the number of jurisdictions in a metropolitan region grows, greater job decentralization occurs. This could be due to Tiebout-sorting effects. A higher number of jurisdictions within a region might mean that firms have more scope to better match the local taxes they pay with the bundle of public goods they receive. It is also possible that, as the number of jurisdictions within a region increases, there is greater interjurisdictional competition for business activity within the region that causes activities to disperse within the region. The U.S. Census of Governments (2019) provides data on the number of jurisdictions within regional economies. For each region, the number of counties and cities is summed to account for the degree of adminstrative fragmentation within each region. The level of MSA employment for each year is included to proxy for economic performance in each region. The natural logarithm is taken of some variables, as marked in Table 2, to account for the skewed nature of the distribution for these variables. The independent variables are lagged by a year.

Table 1 presents summary statistics for the key variables of analysis in the 100 largest MSAs for the year 2019. On average, CBD areas were home to 9.6% of regional employment. The MSA areas in the sample were home to around 950,000 jobs, on average. On average, CBD areas accounted for 3% of regional population, while 44% of CBD residents were Black or Hispanic, and 39% held a bachelor's degree or higher. In 1994, the average CBD area had a job density of 10,360 jobs per square mile. The average MSA was home to 97 local jurisdictions.

Table 2.
The Relationship Between theShare of Regional Jobs in CBD Areas and Key Predictors, 1995–2019.

Panel 1995–2019 Pooled 1995–2019

OLS 2SLS OLS

Model I Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8

Log of MSA employment −0.128** −0. l09* −0.117* −0.084* −0.065* −0.128 −0.128 −0.112*

(0.056) (0.057) (0.062) (0.052) (0.042) (0.056) (0.056) (0.062)

Cognitive activity 0.281•• 0.692* 0.281•• 0.281•• 0.666

(0.134) (0.195) (0.134) (0.134) (0.198)

Manual activity −0.417

(0.204)

CBD population share 0.273* 0.274* 0.252* 0.283* 0.262* 0.273* 0.273* 0.259*

(0.032) (0.032) (0.037) (0.034) (0.040) (0.032) (0.032) (0.037)

CBD share of residents Black or Hispanic −0.098* −0.100••· −0.131••• −0.131* −0.175* −0.098* −0.098* −0.130***

(0.022) (0.022) (0.027) (0.027) (0.033) (0.022) (0.022) (0.027)

CBD share of residents with bachelor's degree 0.058* 0.059 0.009 0.067 0.016 0.058* 0.058* 0.011

(0.030) (0.030) (0.035) (0.032) (0.037) (0.030) (0.030) (0.035)

CBD job density I994 0.212•• 0.146*

(0.090) (0.077)

Log of number of jurisdictions −0.076 −0.072

(0.030) (0.079)

Bartik Instrument, first-stage results 0.165* 0.526***

(0.932) (0.207)

Constant 0.659 0.303 0959 0.130 0.333 0.605 2.531 2.280

(0.709) (0.719) (0.791) (0.766) (0.863) (0.683) (1.206) (1.312)

Kleibergen-Paap rk Wald F

R-squared 0.304 0.313 0.271 0.331 0.296 0.937 0.937 0.923

Sample Full FuU No tails Full No tails Full FuU No tails

Number of observations 2114 2114 1724 2022 1627 2114 2114 1724

Note. p < .05, p < .01, p < .001. Heteroscedasticity-robust standard errors in parentheses. Year effects included in all models. MSAs are the units of observation.

Results

Table 2 displays the output for eight model specifications that explore different factors associated with the share of regional jobs found in CBD areas over the period 1995 through 2019. The models employ either OLS or robust two-stage least squares (2SLS) estimators. Model 1 represents the base model for the panel of regions, which, as with all models, includes fixed effects for each MSA and year. Model 1 reveals that MSA employment growth is negatively associated with the share of regional jobs found in CBD areas, when all other variables are held constant. As regional employment increases by 1%, the share of employment in CBD areas falls by 0.13%. This finding is statistically significant with a 95% level of confidence. Regional job growth, therefore, doesn’t translate into a higher proportion of a region's jobs located in CBD areas. Perhaps this is because the fastest-growing regions in the United States, over the past 20 years or so, are found in the sprawling sunbelt cities that have had historically weak CBD areas. The findings also suggest that qualitative dimensions of regional job growth might be more important to CBD performance than the extent of quantitative job growth.

Table 1.
Summary Statistics for Model Variables in 2019.

Mean Standard deviation

CBD job share 10% 5%

MSA employment 946,634 1,192,343

Cognitive task content −0.46 0.9

Manual task content 0.43 0.41

CBD population share 3% 2%

CBD share of residents Black or Hispanic 44% 22%

CBD share of residents with bachelor's degree 39% 17%

CBD job density 1994 jobs per sq. mile) 10,360 15,540

Number of jurisdictions 99 121

There is a positive and statistically significant relationship between the share of a region's population who live in CBD areas and the share of jobs found in CBD areas, with a 99% level of confidence. As a CBD's share of a region's population increases by 1%, the share of jobs in CBD areas increases by 0.27%. As the share of CBD residents who are either Black or Hispanic increases, the CBD share of regional employment falls, a finding that is statistically significant with a 99% level of confidence. The share of the CBD population with a bachelor's degree has a positive and statistically significant relationship with CBD employment share, with a 90% level of confidence.

The relationship between the share of regional workers who live in CBD areas and the share of jobs located in CBD areas is not the primary focus of this paper. CBD population share has been employed as a control for the primary relationship of interest. Yet there is a long-standing debate about the extent to which "people follow jobs" or "jobs follow people" (e.g., see Storper & Scott, 2009). In the present context, for example, one might ask whether an increase in the share of regional workers found in CBD areas leads to an increase in the share of jobs in CBD areas as employers seek out workers. To be clear, the positive relationship between these variables represents an association rather than a causal relationship. Different specifications of the model that are not presented here reveal that the positive relationship between the variables holds whether CBD population share is employed as a lagged or a leading variable. Further work is required to tease out a causal relationship between the share of jobs and residents located in CBD areas. Likewise, it is not possible to make causal claims about the relationship between CBD population characteristics and CBD job share. For example, it is difficult to determine the nature of the relationship between the share of Black and Hispanic residents in CBD areas and CBD job share. Although it is possible that jobs are drawn to CBD areas with lower shares of Black and Hispanic residents, it is also possible that the growth of high-paying sectors of the economy in CBD areas displaces members of these communities. This relationship is not the primary focus of this paper and further research is required to better understand it.

My primary variable of interest is the level of cognitive activity in a region's economy. As cognitive activity in a region's economy increases, the share of jobs in CBD areas increases, holding all other variables constant. This relationship is statistically significant with a 95% level of confidence. In Model 2, the cognitive measure of regional economic activity is replaced by the manual measure. All other variables in Model 2 are the same as in Model 1. As the level of manual activity within a region's economy increases, the share of employment found in its CBD area decreases. The findings from the two models suggest that there is a relationship between the nature of a region's economic activities and the share of its workers found in CBD areas. The remaining models, in different ways, seek to explore how robust the relationship is between the nature of a region's economy and the share of regional employment found in CBD areas. For reasons of simplicity and focus, and because the cognitive and manual measures are correlated, the remaining models focus on the cognitive measure of a region's economy, rather than the manual measure, although the findings that are presented are consistent across the two measures.

Since zip codes can be imperfect units for measuring the geography of CBD areas, I remove the regions with the 10 smallest and 10 largest CBDs, by land area, from the analysis in Model 3. I do this to reduce the possibility that my findings are shaped by incosistent zip code sizes among regions. When these outliers are removed, the results from Model 3 are remarkably similar to the results in Model 1. One key difference is that the relationship between the cognitive measure of a region's economy and the share of workers found in CBD areas is stronger in Model 3––the coefficient is higher and the relationship is statistically significant with a 99% level of confidence, holding all other variables constant.

According to the findings from Models 1, 2, and 3, the nature of activities in a regional economy is associated with the share of jobs found in CBD areas, when controlling for other factors that might influence the relative growth of CBD employment. The nature of the relationship between the two variables, however, is not clear. Although the nature of regional economic activities might influence the share of regional jobs found in CBD areas, the share of jobs in CBD areas could be correlated with some types of urban form, which might be attracitve to certain types of industries and therefore shape regional specialization. To tease out the direction of causality between the two variables, I employ a dynamic shift-share as an instrument variable (Bartik, 1991; Faggio & Overman, 2014; Ottaviano & Peri, 2006). The purpose of the measure is to apply economy-wide exogenous changes to predict changes in specialization across regions.

For each four-digit NAICS sector in each region, I replace observed employment change in each year with the exogenous, national rate of change for that sector, meaning that changes in regional specialization are not determined by local conditions, such as the influence of urban form. In theory, national change in a sector should be a good predictor of regional employment change in that sector. Consider two regions that are differently specialized in 1995—one with an economy requiring higher levels of skilled workers, for example. As the nature of the national economy becomes more cognitive in nature, the MSA with an initial specialization grounded in a highly skilled workforce will experience proportionately more growth in such employment.

I calculate the shift-share measure as follows:
$\hat{E_{c, t}^{T}} = E_{c, t - 1}^{T} [1 + \frac{(E_{U S, t}^{T} - E_{c, t}^{T}) - (E_{U S, t - 1}^{T} - E_{c, t - 1}^{T})}{(E_{U S, t - 1}^{T} - E_{c, t - 1}^{T})}]$
(2)
where E_US denotes national employment levels, and E_C represents employment in region, c. In Equation 2, the level of employment for each four-digit NAICS industry, T, in each metropolitan region in period t-1 is multiplied by the national growth rate in industry T between t-1 and t. Since a region with a high concentration of one industry could influence the national growth rate of that industry, the national growth rate applied to each sector in each region excludes the performance of that region.

In Model 4, the cognitive measure employed in Model 1 is replaced by a cognitive measure derived from the shift-share instrument. A 2SLS estimator is employed with heteroscedastic-robust standard errors. The instrument passes tests of under and weak identification. The second-stage results closely resemble the findings from the baseline model (Model 1), although the cofficient of the IV-derived estimate is smaller. In Model 5, I replicate Model 4, but again remove from the analysis the 10 regions at either end of the CBD land area size distribution. As in Model 4, removing these outliers increases the coefficient of the instrument variable and increases its statistical significance. Overall, the consistency across the 2LSL and OLS results provides some evidence to support the idea that regional specialization (as a regional economy becomes more cognitive in nature) has an independent, positive effect on the share of jobs found in CBD areas.

Above, I identify other factors that could influence the share of jobs found in CBD areas, such as differences in urban form as well as the level of government fragmentation within regions. Neither of these variables vary with time––or at least, vary little with time––and as such, cannot be incorporated into the panel structure of Models 1 through 5. In a panel structure, these variables would simply drop out of the model. Model 6, therefore, replicates Model 1, but observations are pooled into cross-sections across years. Dummy variables are employed for each year and MSA area. Functionally, Model 6 is equivalent to Model 1, so the findings are identical across the two models.

In Model 7, I add the measures of urban form and fragmention to Model 6. Model 7 reveals that CBD job density in 1994 is a positive and significant predictor of the share of jobs in CBD areas, with a 99% level of confidence. In other words, the nature of CBD areas in 1994 is a predictor of subsequent growth in the share of jobs found in CBD areas. A dummy variable is employed to account for the census division within which a region is located, the output from which is excluded here. Again, a census division is a proxy for urban form in so much as metropolitan regions in different parts of the country share similar urban structures based on the approximate date of their early development. New England is the omitted category. If a CBD is in the Middle Atlantic, which includes New York City, this has a positive impact on the concentration of jobs in CBD areas compared to New England. If a CBD is in the East North Central (the Midwest), South Atlantic (the Southeast), or Mountain regions, this has a negative impact on the share of jobs located in CBD areas, compared to New England, with a 95% level of confidence. In other words, when I control for underlying differences in urban form among regions, regional specialization still has a positive and statistically significant relationship with the share of jobs found in CBD areas. Model 7 also reveals that the number of jurisdictions in a regional economy does not have a statistically significant relationship with the share of jobs found in CBD areas. These associations hold in Model 8, when the regions containing the CBDs with largest and smallest land areas are removed from the analysis. In summary, when controlling for a variety of factors that could influence the share of regional jobs found in CBD areas, the results presented here suggest that the nature of economic activities within regions influences the relative performance of CBD economies.

Conclusion

Since the turn of the century, references to an urban resurgence have been pervasive. With respect to employment, the evidence reveals that the urban resurgence is limited in scope. Across the 100 largest MSAs in the United States, CBD job growth, on average, has lagged job growth in the rest of regional economies. Yet, CBD economies in some regions have countered this trend, where CBD job growth has exceeded job growth in non-CBD areas. My results find that, as a regional economy increasingly comprises economic activities that are cognitive in nature, the share of jobs found in CBD areas increases, when controlling for other factors that might influence CBD performance. In other words, the performance of CBD areas seems to be associated with the nature of a region's economic activities. This finding is robust across several model specifications.

My findings should be considered within the context of structural changes that have occurred in the economy over the past 40 years or so. Specifically, changes from a manufacturing-based to an information-oriented economy, and the tendency of new economic activities to be clustered in a handful of ‘superstar regions.' The concentration of these activities has renewed income divergence among regions and has contributed to divergence in cultural identity and health care outcomes (Moretti, 2012). My findings suggest that the geography of knowledge economy activities might also influence differences in urban form among regions. This paper does not consider why knowledge-intensive activities find a home in some regions to a greater extent than in others, but it does provide evidence that CBD performance seems to be the outcome of a region's industrial base, rather than its cause.

Although I found a positive relationship between the share of regional jobs and the share of residents found in CBD areas, future research could explore the extent to which there is a causal relationship between the variables. This would contribute to the long-standing debate about the extent to which people follow jobs or jobs follow people, albeit at an intra-metropolitan scale. Furthermore, the global pandemic has impacted many facets of the economy and society, but downtown areas have been uniquely affected since they are home to a larger share of office jobs than is the case in the rest of regional economies. The increase in the number of employees working from home, brought about by the pandemic, has been especially pronounced in offices. Although there is little evidence that companies have abandoned offices in prime, CBD locations, a greater acceptance of work-from-home arrangements could shift some consumption from places of work to places of residence within regions. Further research could determine the extent to which work-from-home arrangements have shifted the nature of localized multiplier effects within regional economies, one consequence of which could be a negative impact on local services firms in CBDs. Finally, CBD areas are not the only job-dense locations in modern, polycentric cities. Although my analysis has treated job location within regions in a binary way, it is likely that secondary employment subcenters within regions would also experience job gains, as cognitive activities increase within regions.

	Panel 1995–2019	Pooled 1995–2019
Log of MSA employment	−0.128**	−0. l09*	−0.117*	−0.084*	−0.065*	−0.128**	−0.128**	−0.112*
	(0.056)	(0.057)	(0.062)	(0.052)	(0.042)	(0.056)	(0.056)	(0.062)
Cognitive activity	0.281••		0.692***			0.281••	0.281••	0.666**
	(0.134)		(0.195)			(0.134)	(0.134)	(0.198)
Manual activity		−0.417**
		(0.204)
CBD population share	0.273***	0.274***	0.252***	0.283***	0.262***	0.273***	0.273***	0.259***
	(0.032)	(0.032)	(0.037)	(0.034)	(0.040)	(0.032)	(0.032)	(0.037)
CBD share of residents Black or Hispanic	−0.098***	−0.100••·	−0.131•••	−0.131***	−0.175***	−0.098***	−0.098***	−0.130***
	(0.022)	(0.022)	(0.027)	(0.027)	(0.033)	(0.022)	(0.022)	(0.027)
CBD share of residents with bachelor's degree	0.058*	0.059**	0.009	0.067**	0.016	0.058*	0.058*	0.011
	(0.030)	(0.030)	(0.035)	(0.032)	(0.037)	(0.030)	(0.030)	(0.035)
CBD job density I994							0.212••	0.146*
							(0.090)	(0.077)
Log of number of jurisdictions							−0.076	−0.072
							(0.030)	(0.079)
Bartik Instrument, first-stage results				0.165*	0.526***
				(0.932)	(0.207)
Constant	0.659	0.303	0959	0.130	0.333	0.605	2.531	2.280
	(0.709)	(0.719)	(0.791)	(0.766)	(0.863)	(0.683)	(1.206)	(1.312)
Kleibergen-Paap rk Wald F
R-squared	0.304	0.313	0.271	0.331	0.296	0.937	0.937	0.923
Sample	Full	FuU	No tails	Full	No tails	Full	FuU	No tails
Number of observations	2114	2114	1724	2022	1627	2114	2114	1724

	Mean	Standard deviation
CBD job share	10%	5%
MSA employment	946,634	1,192,343
Cognitive task content	−0.46	0.9
Manual task content	0.43	0.41
CBD population share	3%	2%
CBD share of residents Black or Hispanic	44%	22%
CBD share of residents with bachelor's degree	39%	17%
CBD job density 1994 jobs per sq. mile)	10,360	15,540
Number of jurisdictions	99	121

Footnotes

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) received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Taner Osman

Notes

Author Biography

Taner Osman's research focuses on themes of local economic development planning and economic and urban geography. He researches the evolving location of economic activity within metropolitan regional economies. His research and teaching interests intersect with other areas of urban planning and urban geography, such as land use and transportation planning.

References

Acemoglu

Autor

(2011). Skills, tasks, and technologies: Implications for employment and earnings. In Ashenfelter

Card

(Eds.), Handbook of labor economics (Vol. 4, pp. 1043‐1171). Elsevier.

Alonso

(1964). Location and land use. Harvard University Press.

Bartik

T. J.

(1991). Who benefits from state and local economic development policies? W.E. Upjohn Institute for Employment Research. https://doi.org/10.17848/9780585223940

Baum-Snow

(2014). Urban transport expansions, employment decentralization, and the spatial scope of agglomeration economies. Processed, Brown University. https://scholar.google.com/citations?view_op=view_citation&hl=en&user=2CslxQ4AAAAJ&citation_for_view=2CslxQ4AAAAJ:dhFuZR0502QC

Boustan

Shertzer

(2012). Population trends as a counterweight to central city decline, 1950–2000. Demography, 50(1), 125‐147. https://doi.org/10.1007/s13524-012-0137-5

Cheshire

(2006). Resurgent cities, urban myths, and policy hubris: What we need to know. Urban Studies, 43(8), 1231‐1246. https://doi.org/10.1080/00420980600775600

Diamond

(2016). The determinants and welfare implications of US workers’ diverging location choices by skill: 1980–2000. American Economic Review, 106(3), 479‐524. https://doi.org/10.1257/aer.20131706

Duranton

Puga

(2015). Urban land use. In Duranton

Henderson

J. V.

Strange

W. C.

(Eds.), Handbook of regional and urban economics (Vol. 5A, pp. 467‐560). North-Holland.

Euchner

McGovern

(2003). Urban policy reconsidered: The problems and prospects of American cities. Routledge.

10.

Faggio

Overman

(2014). The effect of public sector employment on local labour markets. Journal of Urban Economics, 79, 91‐107. https://doi.org/10.1016/j.jue.2013.05.002

11.

Florida

(2002). The rise of the creative class. Basic Books.

12.

Frey

W. H.

(2012). Population growth in metro America since 1980: Putting the volatile 2000s in perspective. Brookings Institution.

13.

Gardner

T. K.

Marlay

(2013). Whither the jobs: Identifying concentrations of employment in metropolitan areas. Urban Geography, 34(6), 795‐814. https://doi.org/10.1080/02723638.2013.790692

14.

Glaeser

(2011). Triumph of the city. Penguin.

15.

Glaeser

E. L.

Gottlieb

J. D.

(2006). Urban resurgence and the consumer city. Urban Studies, 43(8), 1275‐1299. https://doi.org/10.1080/00420980600775683

16.

Glaeser

E. L.

Kahn

M. E.

(2001). Decentralized employment and the transformation of the American city. Brookings/Wharton Papers on Urban Affairs, 2. https://doi.org/10.1353/urb.2001.0008

17.

Glaeser

E. L.

Kohlhase

J. E.

(2003). Cities, regions, and the decline of transport costs. Papers in Regional Science, 83, 197‐228. https://doi.org/10.1007/s10110-003-0183-x

18.

Glaeser

E. L.

Kolko

Saiz

(2001). Consumer city. Journal of Economic Geography, 1(1), 27‐50. https://doi.org/10.1093/jeg/1.1.27

19.

Hartley

D. A.

Kaza

Lester

T. W.

(2016). Are America’s inner cities competitive? Evidence from the 2000s. Economic Development Quarterly, 30(2), 137‐158. https://doi.org/10.1177/0891242416638932

20.

Kneebone

(2013). Job sprawl stalls: The great recession and metropolitan employment location. Brookings Institution.

21.

Landis

(2009). The changing shape of metropolitan America. Annals of the American Academy of Political and Social Science, 626, 154‐191. https://doi.org/10.1177/0002716209344837

22.

Lucas

Rossi-Hansberg

(2002). On the internal structure of cities. Econometrica, 70(4), 1445‐1476. https://doi.org/10.1111/1468-0262.00338

23.

Manville

Storper

(2006). Behaviour, preferences and cities: Urban theory and urban resurgence. Urban Studies, 43(8), 1247‐1274. https://doi.org/10.1080/00420980600775642

24.

Moretti

(2012). The new geography of jobs. Houghton Mifflin Harcourt.

25.

Ogawa

Fujita

(1980). Equilibrium land use patterns in a nonmonocentric city. Journal of Regional Science, 20(4), 455‐475. https://doi.org/10.1111/j.1467-9787.1980.tb00662.x

26.

Osman

(2022). A tale of a few CBDs: The uneven resurgence of central business district economies across US Cities, 1994–2019. Journal of Planning Education and Research, https://doi.org/10.1177/0739456X22113174

27.

Ottaviano

G. I. P.

Peri

(2006). The economic value of cultural diversity: Evidence from US cities. Journal of Economic Geography, 6(1), 9‐44. https://doi.org/10.1093/jeg/lbi002

28.

Storper

Kemeny

Makarem

Osman

(2015). The rise and fall of urban economies: Lessons from San Francisco and Los Angeles. Stanford University Press.

29.

Storper

Scott

A. J.

(2009). Rethinking human capital, creativity, and urban growth. Journal of Economic Geography, 9, 147‐167. https://doi.org/10.1093/jeg/lbn052

30.

Teaford

(1990). The rough road to renaissance. Johns Hopkins Press.

31.

Teaford

J. C.

(2000). Urban renewal and its aftermath. Housing Policy Debate, 11(2), 443‐465. https://doi.org/10.1080/10511482.2000.9521373

32.

Tiebout

(1956). A pure theory of local expenditures. Journal of Political Economy, 64, 416‐424. https://doi.org/10.1086/257839

33.

U.S. Census Bureau (2020a). Annual estimates of the resident population: April 1, 2010 to July 1, 2019, metropolitan statistical area; and for Puerto Rico . https://www.census.gov/newsroom/press-kits/2020/pop-estimates-county-metro.htm

34.

U.S. Census Bureau (2020b). Population Statistics . https://www.census.gov/topics/population.html

35.

U.S. Census of Governments (2019). From municipalities to special districts, official count of every type of local government in 2017 Census of Governments . https://www.census.gov/library/publications/2019/econ/from_municipalities_to_special_districts.html

36.

White

M. L.

(1976). Firm suburbanization and urban subcenters. Journal of Urban Economics, 323‐343. https://doi.org/10.1016/0094-1190(76)90033-4

	Panel 1995–2019					Pooled 1995–2019
	OLS			2SLS		OLS
	Model I	Model 2	Model 3	Model 4	Model 5	Model 6	Model 7	Model 8
Log of MSA employment	−0.128**	−0. l09*	−0.117*	−0.084*	−0.065*	−0.128**	−0.128**	−0.112*
	(0.056)	(0.057)	(0.062)	(0.052)	(0.042)	(0.056)	(0.056)	(0.062)
Cognitive activity	0.281••		0.692***			0.281••	0.281••	0.666**
	(0.134)		(0.195)			(0.134)	(0.134)	(0.198)
Manual activity		−0.417**
		(0.204)
CBD population share	0.273***	0.274***	0.252***	0.283***	0.262***	0.273***	0.273***	0.259***
	(0.032)	(0.032)	(0.037)	(0.034)	(0.040)	(0.032)	(0.032)	(0.037)
CBD share of residents Black or Hispanic	−0.098***	−0.100••·	−0.131•••	−0.131***	−0.175***	−0.098***	−0.098***	−0.130***
	(0.022)	(0.022)	(0.027)	(0.027)	(0.033)	(0.022)	(0.022)	(0.027)
CBD share of residents with bachelor's degree	0.058*	0.059**	0.009	0.067**	0.016	0.058*	0.058*	0.011
	(0.030)	(0.030)	(0.035)	(0.032)	(0.037)	(0.030)	(0.030)	(0.035)
CBD job density I994							0.212••	0.146*
							(0.090)	(0.077)
Log of number of jurisdictions							−0.076	−0.072
							(0.030)	(0.079)
Bartik Instrument, first-stage results				0.165*	0.526***
				(0.932)	(0.207)
Constant	0.659	0.303	0959	0.130	0.333	0.605	2.531	2.280
	(0.709)	(0.719)	(0.791)	(0.766)	(0.863)	(0.683)	(1.206)	(1.312)
Kleibergen-Paap rk Wald F
R-squared	0.304	0.313	0.271	0.331	0.296	0.937	0.937	0.923
Sample	Full	FuU	No tails	Full	No tails	Full	FuU	No tails
Number of observations	2114	2114	1724	2022	1627	2114	2114	1724

Understanding Divergence in the Performance of Central Business District Economies Among U.S. Metropolitan Regions,1995–2019

Abstract

Keywords

Literature Review

Data and Methods

Independent Variables

Regional Industrial Specialization

CBD Population Characteristics

Urban Form

Other Control Variables

Results

Conclusion

Footnotes

Declaration of Conflicting Interests

Funding

ORCID iD

Notes

Author Biography

References