Public transit access and the changing spatial distribution of poverty

Author’s Accepted Manuscript Public Transit Access and the Changing Spatial Distribution of Poverty Rahul Pathak, Christopher K. Wyczalkowski, Xi Huang www.elsevier.com

PII: DOI: Reference:

S0166-0462(17)30240-5 http://dx.doi.org/10.1016/j.regsciurbeco.2017.07.002 REGEC3277

To appear in: Regional Science and Urban Economics Received date: 27 April 2016 Revised date: 22 June 2017 Accepted date: 3 July 2017 Cite this article as: Rahul Pathak, Christopher K. Wyczalkowski and Xi Huang, Public Transit Access and the Changing Spatial Distribution of Poverty, Regional Science and Urban Economics, http://dx.doi.org/10.1016/j.regsciurbeco.2017.07.002 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting galley proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Public Transit Access and the Changing Spatial Distribution of Poverty

Rahul Pathak, Christopher K. Wyczalkowski*, Xi Huang

1

Georgia State University, Andrew Young School of Policy Studies, 14 Marietta Street NW, 3rd

Floor, Atlanta, GA 30303

[email protected] [email protected] [email protected]

*

Corresponding author.

Abstract This article examines whether access to public transportation plays a significant role in determining the spatial distribution of poverty in a metropolitan area. Our empirical strategy relies on long-term changes in poverty and access to bus transit at the neighborhood level in the Atlanta metropolitan area. We estimate the effect of bus transit access on poverty using fixedeffects models to control for time-invariant unobservable characteristics. Furthermore, we undertake several robustness checks using a combination of instrumental variable regression, subsample analysis, and propensity score matching. Our results indicate that, on average, after 

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. 1

controlling for neighborhood characteristics, census tracts with better access to public bus transportation have a higher proportion of low-income households – in both the central city and the suburbs. Thus, policies that improve access to transit in underserved areas can plausibly expand residential opportunities for the poor and reduce spatial inequities in urban centers.

JEL classification: R1, R3, R4

Keywords: Urban Poverty, Concentrated Poverty, Public Transportation, Bus Transit

1. Introduction The key focus of recent academic research on urban poverty and transportation in the United States has been on the central city. However, there are currently more poor people (defined as households below the federal poverty threshold) living in the suburbs of American cities than in the urban cores (Cooke & Denton, 2015, Kneebone & Berube, 2014). Between 2000 and 2013, the population of the suburban poor increased almost twice as fast as the

2

population of the central city poor, and by 2013 nearly 56 percent of the poor in the hundred largest metro areas lived in the suburbs (Kneebone & Holmes, 2014). The number of suburban poor living in concentrated poverty rose by 188 percent during the same period (Kneebone & Holmes, 2016). Previous research has suggested that factors such as job decentralization, gentrification, subsidized housing, antipoverty programs, and immigrant preferences are among the potential drivers of this rising suburban poverty (Freedman & McGavock, 2015; Howell & Timberlake, 2013; Jargowsky, 1997; Kneebone, 2016; Lees, Slater & Wyly, 2008; McKinnish & White, 2011; Persky & Kurban, 2003; Popkin, Levy, & Buron, 2009; Raphael & Stoll, 2010). However, the role of public transportation in influencing the spatial distribution of poverty in metropolitan areas has received limited attention. Traditionally, policies and programs aiming to tackle concentrated poverty have focused on housing-based initiatives. Housing-based programs for poverty alleviation can potentially revitalize neighborhoods and increase the housing options for the poor, but they may concentrate poverty in the targeted neighborhoods. Public transit (especially bus transit), a local public good, on the other hand, may facilitate the deconcentration of poverty and reduction in spatial inequities by linking low-income populations to new housing opportunities and better public services, such as schools, outside of the urban cores. Therefore, a focus on public transportation (especially bus transit, which is substantially more expansive than rail transit in American cities) is of particular relevance in understanding the spatial distribution of low-income households. In this context, this paper examines the relationship between bus transportation access and the changing spatial distribution of poverty in the Atlanta metro area. We contribute to the current literature on the subject of public transportation access and spatial distribution of poverty in at least three ways. First, previous work has examined the

3

relationship between public transport and poverty largely in the context of the central city, and much of the work is concentrated on uniquely large cities such as New York City. However, considering the substantial rise in suburban poverty in many American cities and the expansion of the transit footprint outside central cities, it is important to examine the nature of the transitpoverty relationship over an entire metropolitan area in a representative American city. Second, previous work has examined this relationship mostly in the context of intra-urban rail transit, largely due to the paucity of data on other forms of public transportation. Although rail transit is an important public transportation mode in metropolitan areas, it captures only a small footprint of the public transit system, with limited coverage outside the urban core. To address this issue, we use a novel bus route data set to capture the long-term changes in transit availability in the entire Atlanta metropolitan area. The bus route data set is constructed from cartographic records of the Atlanta public transportation footprint from 1970 to 2010. Third, we improve on the methodology of previous studies by employing a combination of fixed-effect, instrumental variable, and matching methods. Our fixed-effect estimates are expansive in terms of time and space, providing an appreciably long study period. Furthermore, to improve confidence in our results, we undertake several additional analyses to test the robustness of our results to different subsamples and estimations techniques. The rest of the paper proceeds as follows. We begin the next section by discussing our conceptual framework on the relationship between urban poverty, its changing geography, and public transportation access. Section three summarizes the key dimensions of the poverty and public transportation in our study area – the Atlanta metropolitan area. Section four provides the details of cartographic and demographic data that we use in this study. In section five, we present the results of the fixed-effects models and several robustness checks. The last section discusses 4

the limitations of this study and considers the results in terms of their implications for social and urban policy. 2. The Conceptual Framework Income-based segregation of households in urban areas has motivated a stream of academic work in areas such as land-use theory, spatial mismatch analysis, Tiebout sorting, and more recently, studies on gentrification (Alonso, 1964; Kain, 1968; Lees, Slater, & Wyly, 2013; Meyer, Kain, & Wohl, 1965; Mills, 1972; Muth, 1969; Mieszkowsli & Mills, 1993; Vigdor, Massey, & Rivlin, 2002). Several technological, economic, geographic, political, and social determinants of residential spatial distribution and location decisions have received significant attention across disciplines (Chan, 2004; McFadden, 1978). However, less is known about the role of public transportation in shaping the intra-urban spatial distribution of the poor, especially relating to the recent suburbanization of poverty. Suburbanization has been studied through a variety of academic lenses. There is evidence that the construction of highways and the decreasing intra-urban commuting costs have contributed to population decentralization from inner cities to their suburbs (Ahlfeldt & Wendland, 2011; Baum-Snow, 2007). Better quality of amenities, preferable housing stock, better school quality, less crime and racial tension, higher environmental quality, and lower taxes in the suburbs have also contributed to this process (Cullen & Levitt 1999; Reber 2005; Tiebout, 1956). Though these accounts provide insightful explanations, they have been mostly developed to explain the decentralization phenomenon of the middle- and high-income households to the suburbs and poverty concentration in the inner city. The recent rise of suburban poverty has challenged the existing wisdom and directed research and policy interests to factors that may explain this phenomenon. While Raphael and Stoll (2010) highlight the role of employment 5

decentralization in propelling this trend, other researchers find explanations in housing-related forces (Brueckner & Rosenthal, 2009). Based on evidence that redevelopment brings new housing to central-city neighborhoods that cater to the affluent, Brueckner and Rosenthal (2009) predict the current location patterns of different income groups using the geography of dwelling ages. Some suburban communities have become more affordable over time as their housing stock deteriorates with age and filters to lower income occupants. The rising housing costs in the urban core have also led residents to move outward in search of affordable options. LeRoy & Sonstelie (1983) were among the first researchers to directly suggest that income-stratified urban geography could be driven by the choice of transportation mode – the slow and low-cost public transportation versus the faster and costlier private vehicle. They point out that mode choice is driven by costs, and varies by income group. Automobiles are more affordable to the wealthy, giving them more autonomy in choosing residential locations. The poor are instead dependent on public transportation; therefore, they are geographically constrained in central cities where public transportation has a large presence. In this paper, we examine whether this relationship holds true when new public bus transportation expands beyond the central city, potentially providing increased residential opportunities for low-income households. In other words, does public transit moderate the spatial distribution of poverty within a metropolitan area? Only a few studies have focused on the relationship between public transportation access and poverty, and the empirical evidence remains mixed (Barton & Gibbons, 2015; Brueckner & Rosenthal, 2009; Giuliano, 2005; Glaeser, Kahn, & Rappaport, 2008; Taylor & Ong, 1995).1 Glaeser and his colleagues (2008) empirically examined the relationship between rail public 1

Several studies on spatial mismatch have focused on the relationship between transportation and employment, particularly of the minority groups (Ihlanfeldt & Sjoquist, 1998; Sanchez, 1999; Stoll, 2006). The spatial mismatch literature does not directly addresses this question, thus we do not review this literature here.

6

transportation access and poverty in New York City and argued that better availability of rail transit is one of the primary reasons for the concentration of poverty in the central city. The time and cost structure associated with public transportation is believed to be more compatible with the preferences and budget constraints of the poor. Glaeser et al. (2008) also highlight that traditional housing market explanations are not sufficient to explain the spatial sorting of the poor across urban census tracts. Brueckner & Rosenthal (2009) also use census tract data from 331 MSAs and find that the age of housing stock, as well as public transit access, are key determinants of the incomebased spatial distribution in metropolitan areas. On the other hand, Barton and Gibbons (2015) examine the relationship between the concentration of public transportation and the income profile of neighborhoods in New York City. They argue that the residential choices of the poor in response to increased public transportation coverage is not generalizable to places outside of New York City. In this study, we focus on the case of the Atlanta metropolitan area, which is more representative of other mid-size American metro areas2 and has a significant spatial correlation between transit footprint and poverty distribution (Figure 1).

2

Atlanta Metropolitan Area is the ninth largest Metropolitan Statistical Area in the United States with an estimated population of around 4.5 million in 2010, much lower than large MSAs like New York (18.3 million) or Los Angeles (12.1 million), but significantly comparable to other mid-size MSAs such as Houston (4.9 million), Philadelphia (5.4 million), Boston (4.1 million), and Washington D.C. (4.5 million).

7

Fig 1: Public Transit Access and Spatial Distribution of Poverty in Atlanta Metro, 2010 In Figure 1, the pink lines on the map show the bus routes of the Atlanta public transportation system and the blue dots show the bus stops. The map exhibits higher levels of poverty concentration (dark red shading) around transit not just in the central city (within the yellow polygon) but also along the spokes of bus routes that fan out into the suburbs from the central city. Notably, on the outskirts, census tracts with bus stops have distinctly higher levels of poverty concentration than those that have bus routes passing through them without bus stops. Several factors can theoretically explain this overlap between bus transit and the spatial distribution of poverty. First, since poor households value public transportation as an amenity, 8

they move to transit-rich neighborhoods to improve their mobility and minimize their transportation costs. Second, transit is assigned to neighborhoods that are poor to begin with, due to the higher demand for public transit in those neighborhoods. Third, it is plausible that a certain set of omitted variables related to housing or economic factors are driving this relationship. Fourth, the arrival of transit in a neighborhood is leading to out-migration of the rich, due to some inherent aversion to the existence of bus transit in their neighborhood. Lastly, it can be argued that through a certain unknown mechanism public transportation is making the existing residents of a neighborhood poorer over time. In this study, we undertake the analysis at the aggregate census tract level and thus cannot provide evidence for the fourth and fifth mechanisms, which would require a differentiation between income changes caused by inmovers versus out-movers or an examination of the potential negative effects of transit on earnings of the same individuals over time. The inability to address the last two mechanisms remains a limitation of this study and warrants further research. We will come back to limitations in the final part of the paper, but before moving into the empirics, the next section provides some background information about the transit and poverty patterns in the Atlanta metropolitan area.

3. Public Transportation and Poverty in Atlanta The Metropolitan Atlanta Rapid Transit Authority (MARTA), the public agency that operates the public transportation system in Atlanta, manages 38 rail stops and 740 bus stops covering areas in Fulton, Clayton, and DeKalb counties. Transit expansion in Atlanta has gone through its share of political and fiscal problems, but the expansion of public transportation in 9

the region is likely to gain momentum. In 2014, Clayton County rejoined MARTA by approving a one percent sales tax to fund buses and future high capacity public transportation. Furthermore, in a November 2016 referendum, 71 percent of city of Atlanta voters approved a sales tax hike to support a $2.6 billion expansion and upgradation of transit services. The public transit system in Atlanta has regularly expanded its footprint since the Authority’s creation in 1965. After a voter referendum in 1971, MARTA started its first heavy rail line construction in 1975, which went into service in 1979 and underwent expansions during the 1980s and early 1990s. However, the rail system only covers DeKalb and Fulton counties, a footprint significantly smaller than that of the bus transit system. Therefore, the bus system essentially shapes the transportation footprint in the Atlanta metropolitan area. In Figure 2, we outline the historical evolution of the bus transportation system in Metro Atlanta from 1970 to 2010. By 1970, almost all the census tracts in the central city had bus routes passing through them. As suburbanization gained momentum, the bus routes continued to expand beyond Fulton and DeKalb counties. The expansion was relatively slow in the 1990s, but there was a period of significant expansion in bus routes during the 2000s. In 2000, the agency opened two rail stations north of the city that fueled further expansion in the northern neighborhoods. In 2001, MARTA opened a Compressed Natural Gas (CNG) garage facility that enabled fleet expansion, and the agency signed an agreement with the Georgia Regional Transit Authority (GRTA) to expand commuter bus services into the Clayton County to the South. These expansions to the northern and southern suburbs are clearly visible in the Figure 2.

10

Fig 2: Expansion of Bus Transit Footprint in the Atlanta Metropolitan Area

The expansion of the bus transit footprint was concomitant with the changing demographics of the suburbs, characterized by increasing representations of minorities, immigrants, and low-income households (Pandey & Sjoquist, 2017). Atlanta’s suburbs have witnessed a steep increase in the number of distressed and high-poverty neighborhoods. During the 2000s, the number of high-poverty (census tracts with poverty rates between 20 and 40 percent) and distressed (census tracts with poverty rates of 40 percent or more) tracts in the suburbs grew from 32 to 197, registering a 500% increase (Kneebone and Berube, 2014).

11

In Figure 3, the changing poverty footprint (red shading) in Atlanta is quite clear.3 In 1970, high-poverty tracts were noticeably located in the city core and the outer-ring counties, which until this time had a rural character. However, as the first wave of suburbanization gained momentum in the 1970s and 1980s, the median incomes in suburbanizing counties rose and poverty rates declined, leading to poverty concentration in the central city as captured by the 1990 census. However, as has been noted by several other commentators (Burns, 2013; Semuels, 2015), the proportion of people living in poverty started to increase in the suburbs in the 1990s, often along the transit corridors (grid lines representing the bus transportation footprint), and the trend continued into the 2000s. In Figure 3, we can observe several such overlaps between transit coverage and poverty. The census tracts just north of the Fayette County border gained access to transportation around 1990 when the poverty rates at that time were around 5.0 to 7.5 percent. Over the course of the ensuing decades, these census tracts have experienced a substantial increase in poverty rates that reached the levels of 20.1 to 40.0 percent. A similar pattern exists in the census tracts to the southwest of Gwinnett County. To a large extent, we see a spatial overlap in the changing proportions of the low-income population, and the changing share of minority populations in these neighborhoods (Figure 4). The northern suburbs were predominantly white until the 1980s. The pattern started to change in the 1990s, with the increases in the minority share of the population, again, mostly occurring along the transit corridors. Figures 1, 3, and 4 suggest a spatial correlation between transit expansion and the changing geography of poverty in the Atlanta metropolitan area. These patterns lead us to ask whether the observed correlation can be attributed to a causal effect of bus public transportation 3

Figures 2 and 3 only show the poverty data for only 8-County metro area since consistent data from 1970-2010 is available only for these countries. U.S. Census Bureau started collecting tract-level data for the remaining 21 counties only from 1990. Section 4 provides further details of the census tract and public transportation data.

12

on the spatial distribution of poverty in cities, after accounting for other neighborhood characteristics. If so, is the relationship true only in the central city like previous studies (Glaeser et al., 2008) have suggested, or is it also true for the suburban neighborhoods?

13

Fig 3: Transit Expansion and Poverty in the Atlanta Metro, Eight County Metro Area 14

Fig 4: Transit Expansion and Minorities (% Non-Whites) in the Atlanta Metro

15

4. Data Sources There are several challenges to studying neighborhood-level changes in access to public transportation and poverty over time. First, the neighborhood level data on socioeconomic outcomes is available from the U.S. census bureau at the census-tract level, but the geographical boundaries of census tracts can change during each intercensal period. In this article, we use the census-tract data standardized to 2010 geographies, from the Longitudinal Tract Database (LTDB) to address the problem of changing boundaries (Logan, Xu & Stults, 2014). Second, even though the Census Bureau and other agencies collect information about public transportation use, there is a lack of reliable and consistent data on access to public transportation at the neighborhood level. Also, it is infeasible to collect historical transportation access data that are comparable over time for all the metropolitan areas in the country. Thus, we focus on the Atlanta metropolitan region and construct cartographic data on access to public transportation using historical bus route maps. We obtain our measures of population, poverty, housing and demographic characteristics from the Longitudinal Tract Database (LTDB) for the 1970-2010 study period. Given the challenges in compiling a panel dataset of spatial units with changing geography across decades, LTDB employs a combination of areal interpolation techniques to “re-shape” census tract information from 1970 to 2000 into a consistent 2010 tract-level geography.4 Census tracts were created for the entire country in 1990, but in prior decades only a limited number of counties available for analysis. We obtain consistent estimates from 1970 to 2010 only for the eight county area that constitutes the core of the Atlanta metropolitan region. These eight counties are Clayton, Cobb, DeKalb, Douglas, Fulton, Gwinnett, Henry, and Rockdale (green areas in Figure

4

For a detailed methodological note on Longitudinal Tract Database, see Logan et al. (2014)

16

5). However, the present Atlanta-Sandy Springs-Roswell Metropolitan Statistical Area (MSA) is composed of 29 counties, and we obtain consistent estimates at the census-tract level of aggregation for the entire MSA from 1990 onwards. Thus, we construct two sets of panels, first a 29-county panel that covers the period of 1990-2010, and second an 8-county panel for the period of 1970-2010. After dropping tracts with missing information, we obtain 688 census tracts in the 8-county panel and 947 census tracts in the 29-county panel. Figure 5 provides a visual representation of the county boundaries and the coverage of the two panels.

Fig 5: Counties covered under Atlanta-Sandy Springs-Roswell MSA, Georgia

17

The data on public transportation access come from historical bus maps. We obtained paper bus route maps corresponding to each decennial census period and used these maps to construct our measures of public transportation access. In an ideal scenario, we would like to use one-year lagged maps, but given their unavailability – the best set we have is 1970, 1979, 1988, 2000, 2009, i.e. lags for 1970 and 2000 are not available. Given the slow pace of change of bus routes and population, we do not expect a one-year difference significantly affect our results (Figure 6 illustrates a one-year change in transit footprint during 2000-2001). For 2009, we were able to obtain a Geographic Information Systems (GIS) shapefile containing transit route information in the Atlanta metropolitan area. The 2009 shapefile enables us to account for each bus route in a census tract, and we use this detailed information in our instrumental variable estimations. Using GIS, we georeference the scanned paper maps and overlay them on the 2010 census tract boundaries. We create two measures of transit access using the information from the maps, first a binary variable indicating whether the census tract had a bus route passing through it and second, for only 2009, a continuous variable measuring the number of bus routes passing through each census tract. We employ the former binary measure in the fixed-effects regression estimations and propensity score matching. The latter continuous measure is used in the two-stage least squares estimation. For the instrumental variable regressions, we use a streetcar network map from 1943 to construct a continuous measure of streetcar density, which we will discuss in greater detail in the next section.

18

Fig 6: Short-term Changes in Transit Footprint in Atlanta between 2000 and 2001

19

5.

Estimation and Results The identification of causal effects is a challenge in urban studies, as an urban ecosystem

develops over time through a complex interaction of social, economic, cultural, technological, political, and geographic factors. A range of unobserved characteristics may influence the probability of a city or neighborhood obtaining access to the treatment condition and the outcome of interest, thereby rendering cross-sectional design susceptible to selection bias. The examination of changes in the same neighborhoods over time (‘within’ neighborhood change) may help reduce bias to the extent that some unobserved characteristics of a place are timeinvariant. Thus, we use fixed-effect panel estimates at the census tract level to estimate the changes in a neighborhood’s poverty rate as a function of access to public transit. Our primary specification in the fixed effects model assumes the form of equation 1, (1) where

is the percentage of the population in census tract i and year t living below the federal

poverty threshold,

is a dummy indicating whether census tract i receives access to bus transit

during year t. The parameter

measures the impact of transit access on the poverty percentage of

a census tract. All other covariates are included in vector associated with the covariates. In addition,

is the vector of coefficients

denotes a full set of census tract fixed effects, and

is the error term with the assumption that ( In vector

, and

)

for all i and t.

, we include a number of variables to capture the variation in housing stock

and economic characteristics across neighborhoods. If we assume that the housing markets work efficiently, then we can expect the variation in unobserved and unobserved housing characteristics is capitalized in real estate values. The LTDB data provide standardized 20

information on median home value and median rent in each census tract for each time-period. We use the consumer price index from the Bureau of Labor Statistics to adjust for inflation, converting to 2010 dollars. We denote the log of these variables as HRENT and HVALUE. Furthermore, Brueckner & Rosenthal (2009) highlight that the age of the housing stock is a key determinant of income-based sorting in urban areas; thus, we include the percentage of housing that is more than thirty years old (H30OLD) as a measure of housing quality. We control for the percentage of manufacturing jobs in a census tract (MANUFPER) to account for changes in semi-skilled and low-skilled jobs and economic changes in a neighborhood. Lastly, we include population density at the tract level (POPDENSITY) in all specifications. We primarily focus on the poverty rate as an outcome variable, but run some additional models with the share of minorities (% non-whites) and immigrants as dependent variables to gain a deeper understanding of the demographic changes that accompany the changes in poverty. Table 1 reports the descriptive statistics for all variables in the 29-county sample in the year 2010 to provide an overview of the variables. We employ the same fixed effects specifications for our two panels: the full 29-county Atlanta MSA panel from 1990-2010 and the 8-county Atlanta MSA panel from 1970-2010. We conduct an analysis for two subsamples to examine the variation in effects within the metropolitan area: the tracts that are within the 15-mile radius of the city center and the tracts that are outside the 15-mile radius of the city center. We chose the 15-mile ring as the city boundary because the ring broadly corresponds to the ‘perimeter’, a colloquial term for the major interstate corridor (I-285) that encircles the city of Atlanta and effectively divides the city’s inner core from its outer ring suburbs. The I-285 corridor forms a continuous highway around the city creating a geographic barrier that is much more tangible than a city or county political boundary.

21

Table 1: Descriptive Statistics for 29-County Atlanta MSA

Variable

Min

Max

POVERTYRATE MINORITY FOREIGN HRENT ($) HVALUE ($000s) H30OLDPER MANUFPER POPDENSE

Twenty-Nine County Atlanta MSA (2010) 947 13.78 11.40 0.00 947 32.79 30.69 0.00 947 13.32 12.17 0.00 930 798.49 236.02 182.00 942 212.56 117.72 10.00 947 36.84 24.80 0.00 947 8.90 4.76 0.00 947 2335.68 2296.54 21.26

95.93 100.00 73.06 2001.00 1000.00 100.00 29.37 27000.00

POVERTYRATE MINORITY FOREIGN HRENT ($) HVALUE ($000s) H30OLDPER MANUFPER POPDENSE

Census Tracts with Access to Public Transit (2010) 478 10.58 7.98 0.00 478 21.70 21.06 0.00 478 10.99 9.91 0.00 467 803.99 256.96 283.00 478 195.89 80.47 67.70 478 28.00 18.43 0.00 478 10.73 4.81 1.23 478 1452.11 1387.68 21.26

47.92 99.24 56.87 1960.00 658.20 87.39 27.79 9624.49

POVERTYRATE MINORITY FOREIGN HRENT ($) HVALUE ($000s) H30OLDPER MANUFPER POPDENSE

Observations

Mean

Std. Dev.

Census Tracts without Access to Public Transit (2010) 469 17.05 13.30 0.00 95.93 469 44.08 34.62 0.00 100.00 469 15.70 13.72 0.00 73.06 463 792.95 212.95 182.00 2001.00 464 229.73 144.61 10.00 1000.00 469 45.85 27.13 0.00 100.00 469 7.04 3.91 0.00 29.37 469 3236.20 2662.36 96.92 27000.00

22

Table 2 presents the results from the fixed effects regression for the 29-county MSA from 1990 to 2010. Columns 1–3 show the results for all census tracts in the Atlanta MSA, and columns 4–6 and columns 7–9 show the results for the census tracts that are within and outside the 15-mile radius, respectively. The first specification in each set (columns 1, 4, and 7) shows the naïve relationship between transit access and poverty rate with population density as the only control variable. The second specification (columns 2, 5 and 8) adds four variables that capture the housing characteristics of a neighborhood. The last specification controls for the percentage of the labor force in the manufacturing sector to capture the economic structure of a place. Following the same structure, Table 3 reports the results of the second 8-County panel from 1970-2010. In Tables 2 and 3, the key independent variable is a dummy variable for whether the census tract has access to a public bus transit route. Access is defined as the bus route going through a given census tract or running along the border of the census tract. The Metropolitan Area Rapid Transit Authority (MARTA) operates a heavy rail subway system in addition to the bus system. The footprint of the rail system, however, is very small relative to the bus system, and every rail station has a bus connection. Therefore, given our data, the effects of the two systems cannot be separated in areas where they co-exist without making strong assumptions, and we effectively include rail lines in our analysis when studying the bus system. 5 We find that census tracts with access to public transit have a significantly higher percentage of people living under the federal poverty threshold than those people not living in proximity to public transit. In Table 2, we observe a significantly higher poverty rate in census tracts with transit access in the naïve estimates (Column 1). This relationship is also observed

5

We ran these models for a subsample of census tracts that excluded the tracts with access to rail stations. The results did not significantly differ from the full sample.

23

after the inclusion of the housing and economic variables, albeit with smaller effect size (Columns 2 and 3). The full specification, presented in column 3, shows that in the 29-county panel, on average, census tracts with access to public transit have around three percentage point higher poverty than tracts without access to public transportation. This relationship between bus transit availability and poverty is true not just for the central city but also for the suburbs, as shown in Columns 4 to 9 of Table 2. In the suburbs (more than 15 miles from city center), census tracts that have access to public transit, on average, have about 2.32 percentage point higher poverty rate than those that do not have bus transit access. In Table 3, we report the results for the 8-county panel that spans forty years from 1970 to 2010. The results across this sample are substantially similar to those reported in Table 2 with minor differences in the magnitude of the coefficients. Within the 15-mile radius of the city center, neighborhoods with access to transit tend to have higher levels of poverty after controlling for other neighborhood characteristics (Columns 4–6). This relationship between transit access and poverty is also true for the suburbs (Columns 7–8). The results for the central city are consistent with previous research on the relationship between poverty and transit (Glaeser et al., 2008), but the transit effects in the suburbs are new findings that can add to the urban studies literature. To gain a deeper understanding of the demographic changes, Table 4 estimates the full fixed effects specification for two demographic characteristics, % minority and % immigrants, which are highly correlated with poverty status in the Atlanta metro. On average, census tracts that are proximate to bus routes have higher percentages of minorities and foreign-born residents. The patterns in the suburbs are particularly interesting since they suggest that suburban neighborhoods with bus access have a significantly higher immigrant share, in contrast to

24

insignficant results for the central city. This is not surprising given that during the last decade, immigrants have accounted for almost 30 percent of the population growth and 17 percent of the growth in the number of people in poverty in the suburbs of American cities (Suro et al., 2011). Table 2: Public Transit Access and Poverty in 29-County Atlanta MSA (1990-2010)

(1) (2) (3) All Census Tracts

(4) (5) (6) Less than 15 miles

(7) (8) (9) More than 15 miles

TRANSIT

5.00** 3.22*** 2.92*** 6.76*** 3.15*** 2.79*** 3.40** * * (0.47) (0.42) (0.41) (1.01) (0.91) (0.88) (0.45) POPDENS 0.00 0.00*** 0.00*** 0.00 0.00 0.00 0.00** E * (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) HRENT 4.81*** 5.73*** 9.90*** 9.92*** (0.90) (0.92) (2.13) (2.15) HVALUE -1.82** -1.04 -1.87 2.82*** (0.78) (0.87) (1.36) (1.47) H30OLD 0.15*** 0.14*** 0.14*** 0.12*** (0.01) (0.01) (0.01) (0.01) MANUFP -0.28** ER 0.18*** (0.04) (0.11) Constant 8.60** 45.21** 59.26** 10.72** 77.06** 85.07** 5.50** * * * * * * * (0.64) (5.54) (6.73) (1.20) (12.15) (12.70) (0.29) Observatio 2,841 2,817 2,817 ns R-squared 0.07 0.29 0.30 No. of 947 946 946 Tracts Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.10

2.37*** 2.32*** (0.45) (0.45) 0.00*** 0.00*** (0.00) (0.00) 2.34*** 2.85*** (0.68) (0.69) 2.80*** 3.54*** (0.67) (0.86) 0.17*** 0.16*** (0.01) (0.01) -0.08** (0.03) 32.23** 40.70** * * (4.61) (6.06)

1,146

1,137

1,137

1,692

1,677

1,677

0.05 382

0.29 381

0.30 381

0.16 564

0.35 564

0.36 564

25

Table 3: Public Transit Access and Poverty in 8-County Atlanta MSA (1970-2010)

(1) (2) (3) All Census Tracts

(4) (5) (6) Less than 15 miles

(7) (8) (9) More than 15 miles

TRANSIT

4.40** 2.69*** 2.23*** 5.76*** 2.69*** 1.92*** 3.14** * * (0.49) (0.40) (0.38) (0.69) (0.56) (0.57) (0.61) POPDENS 0.00 0.00*** 0.00*** 0.00 0.00*** 0.00** -0.00 E (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) HRENT 4.30*** 5.78*** 7.32*** 8.04*** (0.50) (0.60) (1.26) (1.24) HVALUE -2.04** 2.71*** 3.76*** 3.22*** (0.53) (0.58) (0.82) (0.85) H30OLD 0.17*** 0.16*** 0.18*** 0.15*** (0.01) (0.01) (0.01) (0.01) MANUFP ER 0.18*** 0.25*** (0.03) (0.04) Constant 9.21** 45.42** 63.78** 10.03** 62.30** 77.59** 7.06** * * * * * * * (0.40) (2.43) (4.33) (0.73) (6.52) (7.00) (0.25) Observatio 3,401 3,342 3,341 ns R-squared 0.04 0.40 0.42 No. of 688 687 687 Tracts Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.10

2.61*** 2.61*** (0.54) (0.54) 0.00*** 0.00*** (0.00) (0.00) 3.49*** 3.57*** (0.41) (0.40) 3.64*** 3.78*** (0.58) (0.73) 0.16*** 0.16*** (0.02) (0.02) -0.01 (0.03) 44.29** 45.78** * * (2.62) (4.37)

1,905

1,877

1,877

1,491

1,460

1,459

0.04 382

0.37 381

0.39 381

0.03 305

0.56 305

0.56 305

Table 4: Public Transit Access and Demographic Differences in 29-County Atlanta MSA (1970-2010)

26

Minority (%) (1) All Tracts

Foreign Born (%)

(2) Less than 15 miles

(3) More than 15 miles

(4) All Tracts

(5) Less than 15 miles

(6) More than 15 miles

15.63*** (1.30) POPDENSE 0.01*** (0.00) HRENT -3.37* (1.91) HVALUE -2.63 (1.69) H30OLD 0.34*** (0.02) MANUFPER -0.84***

13.12*** (2.93) 0.00*** (0.00) -4.03 (3.31) -3.91* (2.28) 0.31*** (0.03) -1.11***

11.93*** (1.36) 0.02*** (0.00) -2.54 (1.91) -2.76 (1.87) 0.42*** (0.04) -0.49***

1.80 (1.36) 0.00*** (0.00) -0.49 (1.50) 4.32*** (0.91) 0.15*** (0.02) -0.27***

5.25*** (0.69) 0.01*** (0.00) -1.19* (0.71) -0.16 (1.48) 0.12*** (0.02) -0.22***

(0.08) 52.85***

(0.18) 75.69***

(0.08) 34.96**

(0.10) -21.82***

(0.04) 8.49

(13.63)

(21.47)

(13.63)

5.54*** (0.70) 0.00*** (0.00) -0.79 (0.78) 3.21*** (0.95) 0.13*** (0.01) 0.26*** (0.04) 12.64** (6.39)

(8.33)

(8.90)

1,677 0.64 564

2,817 0.46 946

1,137 0.36 381

1,677 0.64 564

TRANSIT

Constant

Observations 2,817 1,137 R-squared 0.50 0.42 Number of 946 381 trtid10 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.10

Robustness Test 1: Matching Census Tracts The fixed effects estimates discussed above adequately address concerns related to unobservable time-invariant characteristics of neighborhoods and control for the main neighborhood characteristics. However, differences between treated and untreated tracts could be time-varying and systematically correlated with the error term, potentially biasing our fixed

27

effects estimates. For example, local government programs and amenities for low-income households may vary substantially across jurisdictions potentially introducing omitted variable bias. Although the median rent and median home values could capture the variation in amenities and programs, we may still want to compare tracts that are similar in housing and economic characteristics, but different in transit treatment to further isolate the effect of transit access. To do this, we match census tracts using two approaches. First, we assign as controls the tracts located within a 1-mile radius of the bus transit’s 1-mile outer ring in each decade as shown in Figure 7. Census tracts within these rings should arguably share similar characteristics, given their proximity, but still, possess the most important difference for our estimation – absence of the transit treatment. Table 4 reports the difference in poverty rates in the treated and untreated rings using Ordinary Least Squares (OLS) regression, after controlling for the same set of variables as in the fixed effect models. We find that across all five time periods the treated tracts, on average, have poverty rates that are about 2 percentage points higher than those of the untreated tracts. This finding suggests robustness of the fixed effects model results. Though proximity enables us to compare similar treated and untreated neighborhoods, we further improve our matches by employing a propensity score matching technique – comparing census tracts that have similar propensities for treatment. Informed by previous research regarding the advantages and disadvantages of each matching algorithm (Caliendo & Kopeinig, 2008; Heckman et al., 1997; Smith & Todd, 2005), we use a combination of three matching techniques to ensure the robustness of our estimates: nearest neighbor matching with replacement, nearest neighbor matching with calipers (with 0.10 caliper), and kernel matching (0.10 bandwidth). In addition to the distance from the city center, we include housing and economic variables in the matching vector to increase the resemblance of the treated tracts and

28

the matched untreated tracts.6 The matching variables are population density, median rent, median home value, the age of housing units, and manufacturing employment, the same set of variables used in the fixed effects models above. As reported in Table 5, we obtain significant and positive results for all of the time periods and across the different matching approaches (except kernel matching in 1970), though the magnitude of the estimates varies. In 2010, the treatment effects across the matching specifications are fairly comparable: the treated tracts have around 6 to 8 percentage point higher poverty rates than the untreated tracts. The results stay relatively robust to the inclusions of alternative calipers and bandwidths for nearest-neighbor and kernel matching approaches (results not shown in the table), providing further support for the results observed in the fixed effects approach and outer-ring comparisons.

6

The treated tracts and matched untreated tracts are able to achieve substantial covariate balance along all dimensions specified in the models in all matching methods. Tables are available upon request.

29

Fig 7: Bus Transit’s Outer Ring and Adjoining 1-mile Controls: 1970-2010 30

Table 4: Poverty Difference in Transit’s Outer Ring and adjoining 1-mile tracts

(1)

(2)

(3)

(4)

(5)

1970

1980

1990

2000

2010

1.48* (0.81) 0.00*** (0.00) -11.18*** (1.64) -2.61*** (0.96) 0.13*** (0.02) 0.02 (0.13) 92.80*** (9.81) 199 0.61

1.90*** (0.69) 0.00*** (0.00) -14.36*** (1.70) -4.73*** (0.94) 0.09*** (0.02) 0.12 (0.10) 125.69*** (10.68) 357 0.51

TRANSIT

2.20** 2.26** 1.59* (0.86) (1.13) (0.84) POPDENSE 0.00 0.00 0.00 (0.00) (0.00) (0.00) HRENT -7.85*** -15.00*** -19.22*** (1.48) (2.07) (1.94) HVALUE -4.41*** -1.45 0.20 (1.54) (1.55) (1.24) H30OLD 0.02 0.08 0.15*** (0.08) (0.08) (0.04) MANUFPER 0.01 -0.05 -0.14 (0.07) (0.13) (0.16) Constant 77.75*** 110.06*** 134.31*** (9.00) (10.94) (10.97) Observations 168 141 171 R-squared 0.44 0.52 0.66 Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.10

Table 5: Average Treatment Effect on the Treated (ATT), 8-County Matched Samples (1)

(2)

(3)

(4)

(5)

1970

1980

1990

2000

2010

8.55***

11.46***

13.02***

10.33***

8.76***

(1.93)

(2.50)

(3.13)

(3.11)

(3.19)

Nearest Neighbor with Calipers

2.33*

11.45***

13.01***

10.33***

8.51***

Kernel

(1.20) 2.48

(2.50) 11.53***

(3.12) 12.99***

(3.10) 9.87***

(2.92) 6.46***

(1.68)

(3.07)

(2.51)

(2.40)

(1.76)

687

609

687

686

672

Nearest Neighbor

No. of Observations

Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.10

31

Robustness Test 2: Instrumental Variable Approach As highlighted earlier, the fixed effects method accounts for all time-invariant factors that influence the outcome of interest. However, some unobserved or uncontrolled neighborhood or housing characteristics may vary over time, leading to biased estimates. Matching provides an additional layer of robustness checks, but still reverse causality can pose a threat if public transit assignment is a function of certain income-related neighborhood characteristics. To address this issue, we utilize the instrumental variable approach. A valid instrument should predict access to public transportation in a neighborhood but should not be correlated with current levels of neighborhood poverty through any mechanism but public transit. Several recent studies have employed historical cartographic instruments to overcome the simultaneity between urban the spatial configuration and urban socioeconomic outcomes (Ananat, 2011; Baum-Snow, 2007; Rosenthal & Strange, 2008). Following this approach, we construct our instrument using the 20th-century streetcar network in Atlanta, which was built between the late 1800s and early 1900s and completely dismantled in the late 1940s. Over the years, the neighborhoods and roads with access to the early streetcars had developed a transit-friendly urban environment that was welcoming to later public transportation technologies, first trackless trolleys (rubber wheeled buses powered by overhead electric lines) and ultimately internal combustion engine buses. Studies in the context of other cities, such as Los Angeles, show persisting effects of streetcar systems, which went extinct in the 1960s, on the current land use intensity, density, and urban form (Brooks and Lutz, 2016). In Atlanta, we also observe such a continued existence of transit corridors along the extinct streetcar lines – the density and location of streetcar routes is a strong predictor for the density and location of current bus routes. 32

Early 20th century streetcar access as an instrument The streetcar system in Atlanta was created in the late 19th and early 20th centuries in anticipation of future residential growth in the outskirts of the city, and in response to prevailing political factors. Many cities in the United States experienced a speculative building boom around the streetcar lines from late 19th to early 20th century forming the first set of suburbs (Ward, 1964). The Atlanta Street Railway Company started in 1871 and began its operation as an animal-powered street railway line. With the onset of electrification, electric streetcars began their operation in Atlanta in 1891. In anticipation of residential growth along streetcar lines, Atlanta Consolidated Railway Company and Georgia Railways and Electric Company competed to gain a foothold in the expanded housing market (Carson, 1981), leading to a significant expansion of streetcar lines (Carson, 1981). The resulting residential developments became Atlanta’s first suburbs – termed streetcar suburbs (Warner, 1978). The electric streetcar lines continued to operate until 1949 and covered a significantly large portion of the urban core of the city, with several streetcar lines stretching into the current suburban settlements. When the streetcars were discontinued in 1949, most of the neighborhoods and streets that had better access to streetcars had a built environment that facilitated the creation of bus transit – that makes streetcars a useful instrumental variable for the bus transportation system. We use a 1943 map of the streetcar network from the Georgia Power Company and georeference it to the 2010 census tract boundaries to create the instrument. The instrument is measured as the number of streetcar lines passing through a given census tract’s boundary. The variation we exploit here is the portion of the current bus transit system that is created by a range of historical factors that plausibly had no direct effect on the potential poverty outcomes in the 21st century. The instrumental variable approach uses a two-stage least square (2SLS) estimator, 33

which regresses the endogenous regressor on exogenous variables in the first stage and uses the predicted value of the regressor in the second stage. The 2SLS specification assumes the following form: (2) ̂

(3)

Z is the instrumental variable, which is the number of streetcar routes in a given census tract, X is a vector of control variables, and D is the treatment variable for public transit access, measured by the number of bus lines passing through the census tract. In equation (3), ̂ is the predicted value from the first stage in equation (2) that parcels out the exogenous variation in public transit configuration that is uncorrelated with neighborhood poverty. The control variables in the IV specification are the same as the ones used in the fixed effects specification. We also run our specification with an additional variable of distance from the city center to account for geographically defined unobserved factors and processes.7 Instrument Validity A valid instrumental variable strategy must satisfy three key requirements: valid first stage, monotonicity, and the exclusion restriction (Imbens & Angrist, 1994). For the first requirement, we observe a significant positive relationship between the number of streetcar lines and the number of bus routes in a neighborhood (census tract) with an F-statistic of greater than 30, which confirms the first-stage validity (Table 7). The monotonicity requirement is also credible in our case since those census tracts that had streetcar access were also more likely to

7

We estimate the centroid of each census tract using GIS and calculate its distance from the city center, identified by the census tract that contain Five Points which is historically considered to be the center of downtown Atlanta, to obtain the distance between the census tract and the city center.

34

receive bus lines after the streetcar system was dismantled, as discussed above. The exclusion restriction assumption stipulates that better access to a streetcar network is a predictor of bus transit access in a neighborhood but has no relationship to the current poverty level of the neighborhood, except through bus transit access. The streetcar history in Atlanta indicates that this requirement is likely to be met since the streetcar network was a product of inter-company competition and speculative investment in real estate (Carson, 1981; Ward, 1964; Warner, 1978). As highlighted above, many streetcar neighborhoods were developed after or in conjunction with the construction of streetcar lines (Carson, 1981). Therefore, we argue that the provision of public transit to the low-income population was not a driver of the streetcar network configuration at that time and that the 1943 streetcar network is unlikely to affect the current distribution of poverty within Metro Atlanta. 2SLSRegression Results The 2SLS estimates of the relationship of public transit with poverty are shown in Table 6. The IV specification uses a continuous measure of the bus transit rather than the binary measure used in the fixed effects estimations. Thus, coefficients on the treatment variable in the IV specification should be interpreted as the effect of each additional bus route on poverty holding the other factors constant. Following a specification structure similar to the one in the previous section, Column 1 reports the results with only population density, Column 2 adds the median rent and median housing values, Column 3 shows the results with the full set of control variables, and Column 4, the final specification, further includes a control of the distance from the city center. We obtain broadly similar results across all the specifications. The results of the IV specifications confirm the positive and significant relationship between the transit and poverty that we observed in the fixed effects and matching estimations. 35

Table 6: Poverty and Public Transportation- Instrumental Variable Regressions (2010)

Bus transit density (2010) Instrumented: Streetcar Density, 1943

POPDENSE

(1) 0.32*** (0.12) 0.00* (0.00)

(2) 0.29*** (0.11) 0.00*** (0.00) -14.12*** (1.42) -8.20*** (0.96)

(3) 0.24** (0.11) 0.00*** (0.00) -11.36*** (1.40) -7.89*** (0.86) 0.10*** (0.02) 0.07 (0.09)

9.15*** (0.83) 947 0.11

146.08*** (8.66) 926 0.47

122.57*** (9.92) 926 0.53

HRENT HVALUE H30OLD MANUFPER DISTANCE Constant Observations R-squared

(4) 0.21** (0.11) 0.00** (0.00) -12.12*** (1.53) -8.00*** (0.87) 0.08*** (0.01) 0.17** (0.08) -0.11** (0.06) 130.95*** (11.60) 926 0.55

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

Table 7: First Stage Results for IV (Dependent Variable – Bus Routes Density, 2009)

Streetcar Density (1943) POPDENSE

(1) 7.47*** (2.04) 0.00*** (0.00)

HRENT HVALUE H30OLD MANUFPER DISTANCE

36

(2) 7.26*** (2.19) 0.00*** (0.00) -1.28 (1.99) -1.45 (1.40)

(3) 6.43*** (2.21) 0.00** (0.00) -0.18 (2.02) -1.56 (1.39) 0.09*** (0.02) -0.43*** (0.08)

(4) 5.96*** (2.20) 0.00 (0.00) -2.61 (2.16) -1.80 (1.36) 0.03 (0.02) -0.09 (0.09) -0.36*** (0.05)

Constant Observations R-squared

3.38*** (0.70) 947 0.24

19.03* (10.54) 926 0.25

14.15 (12.25) 926 0.29

40.12*** (13.51) 926 0.32

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.10

6.

Summary and Conclusion The importance of public transportation in determining the residential location of

households has long been recognized in urban studies literature (LeRoy & Sonstelie, 1983). However, there have been limited empirical studies that have examined the relationship between public transportation and the spatial distribution of poverty in metropolitan areas, especially in the context of poverty suburbanization that American metros have been witnessing in recent decades. This paper examines whether the changes in access to bus transportation have an effect on the residential location of the poor, particularly in the suburbs. Our results indicate that, on average, after controlling for neighborhood characteristics, the census tracts with access to public bus transit have a higher proportion of low-income households than tracts without bus access. This relationship is true not just in the central city, as previous research has suggested (Glaeser et al., 2008), but also in the suburbs. Our spatial and empirical analysis suggests that improved access to public transportation in the suburbs may have been an important factor contributing to poverty decentralization in the Atlanta metro in recent years. Our findings are also in concurrence with the findings of previous studies such as Glaeser et al. (2008) and Brueckner & Rosenthal (2009) and underscore the importance of public transit in the lives of the poor. In addition to the fixed-effects models, we undertake a series of robustness checks – all supporting the validity of the primary results. However, there are several caveats in these results that need to be highlighted. First, residential location is complex and likely determined by a multitude of city-level factors. This study has provided insights into the case of Atlanta 37

metropolitan area, but given the unique nature of every urban ecosystem, these results may not necessarily be generalizable to other cities. However, it can be argued that Atlanta is a more representative case study than cities used in previous studies (e.g., New York City). Second, our unit of analysis and data does not allow us to examine individual or household movements in and out of the census tracts. Studies that employ individual-level longitudinal microdata can shed more light on such individual-level patterns. Third, the standardized census tract data is created based on certain assumptions of areal projection, discussed in detail in Logan (2014) that could introduce error to our measurement. Furthermore, in an ideal setting we would have sought to include more robust measures of transit availability such as the transit density and frequency of buses at a location, but creating a historical series of such measures is hardly possible. Therefore, the main variables used in the study may suffer from measurement error. The negative externalities of families living in pockets of concentrated poverty have long been known (Cutler & Glaeser, 1997; Sampson et.al, 2002), but the traditional policy response to address the challenge of concentrated poverty has been mostly based on housing-related interventions. With growing evidence on positive intra- and inter-generational effects of households moving to mixed-income neighborhoods (Chetty et al., 2016; Ludwig et al., 2013), it is important to examine what kind of policies can nudge us towards such outcomes with minimum distortionary effects. Our findings indicate that the local public goods such as better access to public transportation, can influence the spatial distribution of low-income households in the metro areas and assist in the decentralization of poverty. However, if public transportation policies fail to expand access – either because of fiscal problems or due to discriminatory planning – then we run the risk of reconcentrating poverty in certain suburban neighborhoods.

38

Future research should explore whether the patterns that are discussed in this paper hold in other cities, so that transportation policies and programs can assist in creating more equitable cities.

References Ahlfeldt, G. M., &Wendland, N. (2011). Fifty years of urban accessibiltiy: The impact of the urban railway network on the land gradient in Berlin 1890-1936. Regional Science and Urban Economics, 41(2), 77-88. Alonso, W. (1964). Location and land use: Toward a general theory of land rent. Harvard University Press. Ananat, E. O. (2011). The wrong side(s) of the tracks: The causal effects of racial segregation on urban poverty and inequality. American Economic Journal: Applied Economics, 3(2), 34– 66. Barton, M., & Gibbons, J. (2015). A stop too far: How does public transportation concentration influence neighbourhood median household income? Urban Studies. 54(2) 1-17 Baum-Snow, N. (2007). Did Highways Cause Suburbanization? The Quarterly Journal of Economics, 122 (2), 775–805. Brooks, L., & Lutz, B. (2016) Vestiges of Transit: Urban Persistence at a Micro Scale, A working paper. Retreived from http://byron.marginalq.com/2016-0725_brooks_lutz_streetcars.submitted.pdf Brueckner, J.K., & Rosenthal, S.S. (2009). Gentrificaiton and neighbourhood housing cycles: Will America’s future downtowns be rich? Review of Economics and Statistics, 91(4), 725– 743. Burns, R. (2013) Atlanta No. 4 for suburban poverty growth. Atlanta Magazine. May 21. Retreived from http://www.atlantamagazine.com/news-culture-articles/atlanta-no-4-forsuburban-poverty-growth/ Caliendo, M., & Kopeinig, S. (2008). Some practical guidance for the implementation of propensity score matching. Journal of Economic Surveys, 22(1), 31-72. Carson, O. E. (1981). The Trolley Titans: A Mobile History of Atlanta. Glendale, California: Interurban Press.

39

Chan, Y. (2004). Location, Transport and Land-Use: Modelling Spatial-Temporal Information. Springer Science & Business Media B.V. Chetty, R., Hendren, N., & Katz, L. F. (2016). The effects of exposure to better neighborhoods on children: New evidence from the Moving to Opportunity Experiment. American Economic Review 106(4), 855-902 Cooke, T. J., & Denton, C. (2015). The suburbanization of poverty? An alternative perspective. Urban Geography, 36(2), 300-313. Cullen, J. B., & Levitt, S. D. (1999). Crime, urban flight, and the consequences for cities. Review of Economics and Statistics, 81(2), 159-169. Cutler, D. M., & Glaeser, E. L. (1997). Are Ghettos Good or Bad? The Quarterly Journal of Economics, 112, 827–872. Freedman, M., & McGavock, T. (2015). Low-Income Housing Development, Poverty Concentration, and Neighborhood Inequality. Journal of Policy Analysis and Management, 34(4), 805–834. Lees, L., Slater, T., & Wyly, E. (2008). Gentrification. Routledge. Giuliano, G. (2005). Low Income, Public Transit, and Mobility. Transportation Research Record. 1927 http://doi.org/10.3141/1927-08 Glaeser, E. L., Kahn, M. E., & Rappaport, J. (2008). Why do the poor live in cities? The role of public transportation. Journal of Urban Economics, 63(1), 1–24. Heckman, J. J., Ichimura, H., & Todd, P. E. (1997). Matching as an economic evaluation public estimator: Evidence from evaluating a job training programme. The Review of Economic Studies, 64(4), 605-654. Howell, A. J., & Timberlake, J. M. (2013). Racial and ethnic trends in the suburbanization of poverty in U.S. metropolitan areas, 1980-2010. Journal of Urban Affairs, 36(1), 79–98. Ihlanfeldt, K. R., & Sjoquist, D. L. (1998). The spatial mismatch hypothesis: a review of recent studies and their implications for welfare reform. Housing policy debate, 9(4), 849-892. Imbens, G. W., & Angrist, J. D. (1994). Identification and estimation of local average treatment effects. Econometrica: Journal of the Econometric Society, 62, 467–475. Jargowsky, P. (1997). Poverty and Place: Ghettos, Barrios, and the American City. Russell Sage Foundation. Kain, J. F. (1968). Housing Segregation, Negro Employment, and Metropolitan Decentralization. The Quarterly Journal of Economics, 82(2), 175–197. 40

Kneebone, E., & Berube, A. (2014). Confronting suburban poverty in America. Washington, D.C.: Brookings Institution Press. Kneebone, E., & Holmes, N. (2014). New Census Data Show Few Metro Areas Made Progress Against Poverty in 2013. Retrieved from http://www.brookings.edu/research/reports/2014/09/19-census-metros-progress-povertykneebone-holmes Kneebone, E. (2016). The changing geography of US poverty. Retrieved from https://www.brookings.edu/testimonies/the-changing-geography-of-us-poverty/ Lees, L., Slater, T., & Wyly, E. (2013). Gentrification. Routledge. Logan, J. R., Xu, Z., & Stults, B. J. (2014). Interpolating U.S. Decennial Census Tract Data from as Early as 1970 to 2010: A Longitudinal Tract Database. The Professional Geographer, 66(3), 412–420. LeRoy, S. F., & Sonstelie, J. (1983). Paradise lost and regained: Transportation innovation, income, and residential location. Journal of Urban Economics, 13(1), 67–89. Ludwig, J., Duncan, G. J., Gennetian, L. A., Katz, L. F., Kessler, R. C., Kling, J. R., & Sanbonmatsu, L. (2013). Long-term neighborhood effects on low-income families: Evidence from moving to opportunity. American Economic Review. 103(3), 226–231. McFadden, D. (1978). Modelling the choice of residential location. In Spatial Interaction Theory and Planning Models 673, 75–96 McKinnish T. & White T.K. (2011) Who moves to mixed-income neighborhoods? Regional Science and Urban Economics, 41(3), 187-195 Meyer, J. R., Kain, J. F., & Wohl, M. (1965). The Urban Transportation Problem. Harvard University Press. Mieszkowski, P., & Mills, E. (1993). The causes of metropolitan sububranization. Journal of Economic Perspectives, 7(3), 135-147. Mills, E. S. (1972). Studies in the structure of the urban economy. John Hopkins Press. Muth, R. F. (1969). Cities and Housing. University of Chicago Press. Pandey L. & Sjoquist, D.L. (2017) “The Changing Face of Atlanta”, Center for State and Local Finance, Georgia State University. Retrieved from http://cslf.gsu.edu/changing-face-atlanta/ Persky, J. & Kurban, H. (2003). Do federal spending and tax policies build cities or promote sprawl. Regional Science and Urban Economics, 33(3), 361-378.

41

Popkin, S. J., Levy, D. K., & Buron, L. (2009). Has Hope Vi Transformed Residents’ Lives? New Evidence From The Hope Vi Panel Study. Housing Studies. 24(4), 477-502 Raphael, S., & Stoll, M. (2010). Job Sprawl and the Suburbanization of Poverty. Retrieved from http://www.brookings.edu/research/reports/2010/03/30-job-sprawl-stoll-raphael Reber, S. J. (2005). Court-ordered desegregation successes and failures integrating American schools since Brown versus Board of Education. Journal of Human Resources, 40(3), 559590. Rosenthal, S. S., & Strange, W. C. (2008). The attenuation of human capital spillovers. Journal of Urban Economics, 64(2), 373–389. Sampson, R. J., Morenoff, J. D., & Gannon-Rowley, T. (2002). Assessing “neighborhood effects”: Social processes and new directions in research. Annual Review of Sociology, 28(1), 443–478. http://doi.org/10.1146/annurev.soc.28.110601.141114 Sanchez, T. W. (1999). The connection between public transit and employment: the cases of Portland and Atlanta. Journal of the American Planning Association, 65(3), 284-296. Semuels, Alana. (2015) Suburbs and the New American Poverty. The Atlantic. January 7. Retreived from https://www.theatlantic.com/business/archive/2015/01/suburbs-and-thenew-american-poverty/384259/ Smith, J. A., & Todd, P. E. (2005). Does matching overcome LaLonde’s critique of nonexperimental estimators? Journal of Econometrics, 125(1), 305-353. Sjoquist, D. L. (2000). The Atlanta Paradox. Russell Sage Foundation. Stoll, M.A. (2006) Job Sprawl, Spatial Mismatch, and Black Employment Disadvantage, Journal of Policy Analysis and Management, 25(4) pp. 827-854. Taylor, B., & Ong, P. M. (1995). Spatial Mismatch or Automobile Mismatch? An Examination of Race, Residence and Commuting in US Metropolitan Areas. Urban Studies. 32(9) 14531473 Tiebout, C. M. (1956). A pure theory of local expenditures. The journal of political economy, 64(5), 416-424. Vigdor, J. L., Massey, D. S., & Rivlin, A. M. (2002). Does Gentrification Harm the Poor? Brookings-Wharton Papers on Urban Affairs, 133–182. Warner, S. B. (1978). Streetcar suburbs (Vol. 133). Harvard University Press. Ward, D. (1964). A Comparative Historical Geography of Streetcar Suburbs in Boston Massachusetts and Leeds, England: 1850-1920. Annals of the Association of American Geographers, 54(4), 477-489. 42

Highlights    

This paper examines whether access to public transportation plays a significant role in determining the spatial distribution of poverty in a metropolitan area, and if expansion of bus public transit is a factor contributing to the suburbanization of poverty. The paper finds that in the Atlanta metropolitan area, the census tracts with access to bus public transit, on average, have a higher proportion of low-income households – in both the central city and the suburbs. The main results of the paper are robust to a variety of estimation approaches – fixed effects methods, matching techniques, and instrumental variable regression (we use historical streetcar density as an instrument for current bus transit density). The findings underscore the importance of public transportation for low-income households and suggest that improving access to bus transportation may assist in changing the spatial distribution of poverty and creating more equitable and inclusive cities.

43