SLIDE 1 1 09-30-2017 Note from the authors A more complete manuscript will be shared with the discussant and session participants on or around Monday Oct 16, 2017 (available on request from the lead author). This is a work in progress and we apologize for the delay in its
- dissemination. The material contained in this document is an extended abstract that
includes background on, and motivation for, the project, includes the presentation of some descriptive findings but at this time excludes formal model results. THE SCALE AND SPATIAL PATTERNING OF RACIAL/ETHNIC SEGREGATION
BASED ON BOTH HOME AND WORKPLACE ENVIRONMENTS:
THE CASE OF THE ATLANTA MSA Robert J. Zuchowski (rxz155@psu.edu) Stephen A. Matthews (sxm27@psu.edu) both Department of Sociology & Population Research Institute Pennsylvania State University, University Park, PA 16802, USA September 30, 2017 Paper prepared for the International Population Conference XXVIII International Union for the Scientific Study of Population (IUSSP), Cape Town, South Africa October 29-November 4, 2017 Acknowledgements: This work was supported by the National Science Foundation under IGERT Award #DGE-1144860, Big Data Social Science. Words: 4000 approx.
SLIDE 2 2 Abstract Motivated by Ellis, Wright and Parks (2004) emphasis on measuring both residential and workplace segregation we utilize a relatively new data product to provide a micro-geographic lens
- n the study of segregation and spatial mismatch by race/ethnic groups in the Atlanta-Sandy
Springs-Marietta Metropolitan Statistical Area (MSA). We examine the spatial clustering of both residence and workplace segregation using exploratory spatial data analysis (ESDA) approaches such as Local Indicators of Spatial Association (LISA) measures. We examine segregation at different census hierarchical scales within Atlanta, and utilize the race/ethnic diversity of Atlanta to compare and contrast the segregation and spatial mismatch experiences of African American, Hispanic and Asian residents in different counties within the MSA. We will close our paper with a discussion of housing and labor market policy implications based off of findings. Our intent is to showcase LODES and to demonstrate their utility to demographers. Key words: spatial mismatch theory; segregation, exploratory spatial data analysis, local analysis, spatial heterogeneity and nonstationarity.
SLIDE 3 3 Introduction In recent years profound changes have occurred in the political, economic, social and demographic structures of U.S. metropolitan areas. These changes include increased functional specialization, spatial economic differentiation and inequality within metropolitan areas including edge cities, gentrified neighborhoods, ethnic enclaves, new immigrant destinations, Black ghettos, etc. The changing urban racial/ethnic landscape and patterns of segregation are seemingly driven not by a single process but potentially many. Our work falls under the spatial segregation research, which over the past decade has promoted attention to ‘spatial’ issues and has introduced both more nuanced and more sophisticated methods of analysis (Reardon et al, 2008). Some of this literature has been case- study description and informed by exploratory spatial data analysis (ESDA) methods (see for example Brown and Chung, 2006). Following Brown and Chung (2006) but motivated by Ellis, Wright and Parks (2004) emphasis on measuring both residential and workplace segregation we utilize a new data product to provide a micro-geographic lens on the study of segregation based
- n where people live and where they work—and spatial mismatch by race/ethnic groups—in the
Atlanta-Sandy Springs-Marietta Metropolitan Statistical Area (MSA), henceforth Atlanta. Contribution This paper offers three improvements on conventional practice used in studies of segregation and can potentially contribute to empirical research on the spatial mismatch hypothesis (Kain, 1968; Wilson, 1987, Ihlanfeldt and Sjoquist, 1991; 1998), and at the very least provide a different lens on an examination of non-residential segregation.
SLIDE 4
4 First, we take advantage of an innovative, high-resolution geospatial data set; the Longitudinal Employer-Household Dynamics (LEHD) Origin-Destination Employment Statistics; known as LODES data (see LEHD program: http://lehd.ces.census.gov; Graham, Kutzbach & McKenzie, 2014). LODES includes information on the characteristics of the labor force population for every home and workplace census block (for the Atlanta MSA there are data on 88,527 blocks); LODES also includes a limited dataset on origin-destination (not used in this paper). LODES are available annually 2002-2014, which permits an analysis of change (note, race/ethnicity only available post 2009). Using LODES allows us to move beyond the conventional residential focus within the segregation literature. Our research on non-residential segregation (workplace segregation), is not unique in its framing. Indeed, our work in this area stems from a broader interest in non-residential places across the social and health sciences (e.g., Matthews, 2011, Siordia and Matthews, 2016), but we also acknowledge a debt to the earlier work of Ellis, Wright and Parks (2004) and as noted above the spatial mismatch literature. It is important to note, that while the focus is on race/ethnicity we are also examining ‘segregation’ using other dimensions of stratification of workers based on gender, age, income and education. In the full paper we will be expanding on these other dimensions. Second, the high spatial resolution data in LODES provides the building blocks of our analysis of the effects of scale and larger spatial contexts (tracts and counties). Specifically, we examine the spatial clustering of both home and workplace segregation using exploratory spatial data analysis (ESDA) approaches such as Local Indicators of Spatial Association (LISA) measures focusing on the block level data. In addition, we will (a) examine segregation at different census hierarchical scales (census blockgroup, census tract) within Atlanta, and (b) utilize the diversity of Atlanta to compare and contrast the segregation and spatial mismatch experiences of the
SLIDE 5
5 residents of different counties within the MSA. A key theme of our ESDA is the description of the variability of results across scales and across different subdivisions of the MSA (e.g., by county). We will leverage this heterogeneity in patterns by scale and in the full paper/presentation will include vignettes that focus on specific counties (rather than the entire Atlanta MSA – see below). These county vignettes provide an illustration of the complexity of urban systems and also hint at the potential for the relationships between variables in any formal model to vary across contexts (i.e., for their to be non-stationarity in the processes generating the patterns we observe). Ongoing work includes examining for the presence of non-stationarity via the use of geographically weighted regressions and we will include examples of this work in the full paper/presentation. Third, as just alluded, Atlanta is especially suited for a study of racial/ethnic segregation. The Atlanta MSA is a 30-county region that contains almost 6 million residents (2015); making it the ninth largest MSA in the U.S. and second fastest growing MSA in the past decade (a locator map is provided in Figure 1). More importantly Atlanta has a rich Black history, a growing Black middle class, an increasingly mixed-race/ethnic population (49% minority), is one of the largest of the new immigrant gateway regions of the country and yet is still an area beset with extreme racial and economic inequalities (Ingwerson 1987, Sjoquist 2000, Baylor 2000, Liu 2012, Aka 2012). More recently, Atlanta was hit particularly badly by the Great Recession (2007-2012), with elevated rates of foreclosure (Hall et al, 2015) that impacted the housing markets, especially for minority homeowners. (Dis)Aggregating Atlanta The question “why Atlanta?” will become evident as we present our work and especially as we zoom in-and zoom-out to compare results for the MSA by different scales but also in our focus
SLIDE 6 6
- n specific counties. The county vignettes will focus on sub-areas within the MSA that are of
substantive interest in the context of race/ethnic and socioeconomic diversity. A few descriptive facts reveal why Atlanta is an excellent site to examine variations in race/ethnic and socioeconomic groups by home and workplace. First, Atlanta is frequently identified as the MSA with one of the longest average commutes
- f any major urban area in the United States; with a typical commute 12.8 miles (Bookings Report
2015). One quick way of presenting information on the spatial extent and volume of commuting is to present employment/residence (E/R) ratios; the E/R ratio measures the total number of workers working in an area relative to the total number of workers living in the area. Table 1 provides county-specific E/R ratios for the 30 counties in the Atlanta MSA. Fulton County, the core county of the MSA, has 440,000 workers who live in the county, of which 320,000 (72%) work In the county itself, and each day it received a net gain of 357,000 workers coming in from other counties generating an E/R ratio of 1.81. It is worth noting that the 72% of Fulton residents who also work in the county represents the highest percent of ‘local’ (i.e., within county) commuters across all counties in the MSA. As implied it is important to also note that the flows of workers are not just into Fulton County; twenty-eight percent of Fulton’s working residents cross county boundaries to work in other counties. When we examine all Atlanta counties there is a lot of commuting that crosses a county boundary. The daily net losses exceed 10,000 workers in thirteen counties; six of which have a net loss of over 20,000. We have generated county-to-county flow maps and will include selections of these in the full paper/presentation. Second, and has already been mentioned, Atlanta is both seen to be a highly segregated MSA but also one that has seen considerable change in its race/ethnic composition over the last few decades. In total the Atlanta MSA was in 2010 just over 50% white, 32% black, 10% Hispanic
SLIDE 7 7 and almost 5% Asian. As one might expect, race/ethnic data for different counties within the MSA reveals considerable heterogeneity in composition. Just focusing on the four central counties (see Table 2) we see both very high entropy (diversity) scores but variability in the composition of the populations in these counties. Cobb County, north-west of Fulton, is the only one of these four counties that is majority white. Dekalb County, to the east of Fulton, is majority black. In Fulton County both the white and black populations exceed 40 percent. In Gwinnett, a county north east
- f Dekalb and east of Fulton, all four primary race/ethnic groups of interest exceeds 10 percent of
the population with Hispanic and Asian groups representation more than double that of their average for the MSA as a whole. While it is common to focus on black-white segregation, especially in the context of spatial mismatch research, these county aggregate data justify the need to pay particular attention to the internal heterogeneity of Atlanta and motivate not just the use of local spatial analytical tools but also the use of county vignettes. Exploratory Spatial Data Analysis using LODES In our ESDA we focus on the use of both Moran’s I and Local Indicators of Spatial Association (LISA) to examine spatial structure (spatial autocorrelation) and spatial clustering. Both Moran’s I and LISA have been used in segregation research (see for example, Frank, 2003; Logan and Zhang, 2004; Brown and Chung, 2006; Martin, Matthews and Lee, 2017; and also for a recent example using data on Atlanta see Ambinakudige et al., 2017). In our full paper we describe Moran’s I and LISA. Using the LODES data we have calculated Moran’s I for both home (residential) and workplace areas on the primary race/ethnic variables (and also on other social stratification variables) for the entire Atlanta MSA at the block, blockgroup and tract level. We have repeated this analysis within each of the counties for which
SLIDE 8 8 the N units of analysis permitted this (e.g., N > 30 units). In all of our analysis, and for the purposes
- f consistency across scales and MSA and County study areas we use the Queen’s first order
spatial weights matrix. We will summarize the general findings from this Moran’s I analysis in the full paper/presentation. Foreshadowing these results it is worth noting that while we of course find differences in Moran’s I associated with a change in scale from block to blockgroup to tract (due to MAUP related issues – Openshaw 1983) we were particularly interested in the variability in the Moran’s I values for each county in comparison to each other and to the overall MSA averages and if there were any potential association between the Moran’s I values within a county based on LODES home data vs LODES workplace data. We are preparing graphics that summarize some of the variability in Moran’s I by county LODES source. Just to illustrate the variability we are finding, consider that the Moran’s I for percent black based on workplace area characteristics varies from a low of 0.036 (not significant) to a high of 0.644 (Dekalb) with a median Moran’s I of around 0.298. The workplace Moran’s I for Asians varies between -0.005 and 0.198 (Fulton County) and for Hispanics from a low of -0.01 to 0.426 (Rockdale County). The range in the strength of the Moran’s I across counties and by race/ethnic groups is not necessarily surprising but further motivates our interest in local spatial analysis and specific issues of spatial
- nonstationarity. Note, that as we have an N of 30 counties we have examined scatterplots of the
Moran’s I values derived for each race/ethnic group based on both the residential vs workplace measures and examined these associations at the block, blockgroup and tract levels. We will summarize these data in the full paper/presentation. Figure 2 is a Local Indicators of Spatial Association (LISA) cluster map, in this instance a bivariate LISA cluster map, that overlays block data on (a) the clustering of White and African Americans by residential location (the blue areas) and (b) the clustering of White and African
SLIDE 9 9 Americans by their workplace locations (the red areas). Each map represents each race group, and the pattern shows a clear inverse between clusters of White and Black persons in both home residence and workplace location. In other words, Whites and Blacks tends to be separated (isolated) from each other in terms of their home and work locations. Besides showing this inverse spatial pattern, the maps also show some overlap, especially with African Americans working in areas of Fulton that Whites live, suggesting an exposure process. The LISA measure compares the standardized (z-score) of a variable (e.g., percent African American) in a block and the weighted standardized value of the same variable in all connected blocks, where connected is defined as a spatial weights matrix (in our case using a Queen’s 1st order contiguity matrix).1 In the color figure, parts of north-central Atlanta have a low percent of African American (a lack of blue clusters) but there are clusters of African Americans at workplace blocks (presence of red clusters). This map and others like it (on different race-ethnic groups) provide a visual tool but
- ur descriptive analysis also includes other representations of the data.
Table 3 shows the same data used in Figure 2 based on LISA classifications of the home and workplace results simultaneously in a cross-tabulation. LISA clusters allow us to identify blocks that have significantly High (H) or Low (L) percent African American relative to the distribution in their neighboring blocks (for example, the designation of a block as HL means that the percent African American is High in the block while the weighted average of the percent of African Americans in surrounding blocks is by comparison Low). It is important to note that of the 88,527 blocks, the vast majority do not contain significant clusters of African Americans in either their home or workplace blocks. The table shows that there are 1,666 blocks that are classified as High- High on both home and work locations (i.e., home and jobs cluster together); this represents
1 We will examine different specifications of the spatial weights matrix.
SLIDE 10 10 roughly 30 percent of all workplace locations (N=5,559) where there are high concentrations of African American workers. We don’t have space to go into detail but the LISA table (as with Figure 2) reveals spatial mismatch between home and work for African Americans. Work on Hispanics and Asians will be included in the paper/presentation. In our work to date we have utilize other variables available in the LODES data, specifically the characteristics of home and workplace blocks based on age (3 classes), earnings (3 classes) education (4 classes), and gender. We will present some of this work (time permitting in the presentation) and they inform some of our county vignettes. The focus on heterogeneity within the MSA has led us towards a focus on potential non-
- stationarity. We are currently using geographically weighted regression models to focus on
segregation in the workplace, drawing on the LODES characteristics data. We will present preliminary results in the full paper/presentation. Conclusions & Discussion Returning to the three contributions we hope to make in this paper/presentation we combine a new data set, LODES, with established spatial analytical tools (ESDA) and in doing so are able to present descriptive findings related to home-based and workplace-based race-ethnic patterns within Atlanta. The LISA transition matrices reveal the spatial overlap—and non-overlap— between home and workplace race/ethnic structures (as well as applied to other dimensions of social economic structure). This work and other ESDA tools reveals the internal heterogeneity and complexity of this sprawling MSA. It necessitates the need to think clearly about study area boundaries and internal divisions and how these dovetail with housing and labor market policies.
SLIDE 11
11 One of our main intents as we embarked on this descriptive and exploratory study was to showcase LODES and to demonstrate potential utility to demographers. The focus on workplace patterns remains somewhat rare within the literature and here LODES data are clearly an asset. The applications of these data and methods to residential patterns within Atlanta does not necessarily reveal new surprises (this is reassuring). The LODES data have value but the analytical possibilities are also limited. We will summarize some of the strengths and weaknesses of these data as well as what we hoped to do and were unable to do (data limitations and our own time constraints).
SLIDE 12 12 References Aka, E. (2012). Foreclosure disparities in metropolitan Atlanta counties housing market, 2000- 2010: Implications for policies and planning. International Journal of Interdisciplinary Social Sciences 6(6):89-126. Ambinakudige S., Parisi, D, Cappello, G.C. and Lotfata A. (2017) Diversity or Segregation? A multi- decade spatial analysis of demograohics of Atlanta neighborhoods? Spatial Demography 5:123-144. Bayor, R.H. (2000). Atlanta: The historical paradox. In Sjoquist, D.L. (Eds.) The Atlanta Paradox. New York, NY: Russell Sage Foundation. Brown, L.A. and Chung, S-Y. (2006) Spatial segregation, segregation indices and the geographical
- perspective. Population, Space and Place 12:125-143.
Ellis, M. Wright, R. and Parks V. (2004) Work together, live apart? Geographies of racial and ethnic segregation at home and at work. Annals of the Association of American Geographers 94(3):620-637. Frank, A. (2003) Using measures of spatial autocorrelation to describe socioeconomic nd racial residential patterns in US urban areas. Pp. 147-162, In Kidner D, Higgs G, and White (Eds) Socio-economic Applications of Geographic information Science. London: Taylor and Francis. Graham, M.R., Kutzbach, M.J. and McKenzie, B. (2014) Design comparison between LODES and ACS commuting data products, Working Paper CES-WP-14-38, Center for Economic Studies, U.S. Census Bureau. Hall, M., Crowder, K., and Spring, A. (2015) Variations in housing foreclosure by race and place, 2005-2012. The Annals of the American Academy of Political and Social Science 660(1):217- 237.
SLIDE 13
13 Ihlanfeldt, K.R. and Sjoquist, D.L. (1991) The effect of job access on black and whiote youth employment: A cross-sectional analysis. Urban Studies 28(2):255-265. Ihlanfeldt, K.R. and 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. Ingwerson, M. (1987). Atlanta has become a Mecca of the Black middle class in America. Christian Science Monitor 29:1-13. Kain, J.F. (1968). "Housing Segregation, Negro Employment, and Metropolitan Decentralization". Quarterly Journal of Economics. 82 (2):175–197. Logan, J.R. and Zhang W. (2004) Identifying ethnic neighborhoods with census data: Group concentraton and spatial clustering. Pp. 113-126. In Goodchild M.F. and Janelle D.G. (Eds.) Spatially Integrated Social Science. New York: Oxford University Press. Liu, C.Y. (2012). Intrametropolitan opportunity structure and the self-employment of Asian and Latino immigrants. Economic Development Quarterly 26(2):178-192. Martin, M. J.R., Matthews S.A. and Lee B.A. (forthcoming) The spatial diffusion of racial and ethnic diversity across U.S. counties. Spatial Demography https://link.springer.com/article/10.1007/s40980-016-0030-8 Matthews, S.A. (2011) Spatial polygamy and the heterogeneity of place: studying people and place via egocentric methods. Pp. 35-55, Chapter 3 in L Burton, S Kemp, M Leung, SA Matthews, & D Takeuchi (editors) Communities, Neighborhoods, and Health: Expanding the Boundaries of Place Springer, New York, NY. Openshaw, S. 1983. The Modifiable Areal Unit Problem. CATMOG #38 Norwich: GeoBooks. Reardon SF, Matthews SA, O’Sullivan D, Lee BA, Firebaugh G, Farrell CR, Bischoff K (2008) The geographic scale of metropolitan racial segregation. Demography 45(3):489-514.
SLIDE 14
14 Sharma, M. and Brown, L.A. (2012) Racial/ethnic intermixing in intra-urban space and socioeconomic context: Columbus, Ohio and Milwaukee, Wisconsin. Urban Geography 33(3):317-347. Siordia C, Matthews S.A. (2016) Extending the boundaries of place. Pp 37-56, Chapter 3 in Howell FM, Porter, JR, & Matthews SA (editors). Recapturing Space. Springer, New York, NY. Sjoquist, D.L. (2000). The Atlanta Paradox. New York, NY: Russell Sage Foundation. Wilson, W.J. (1987) The Truly Disadvantaged: The Inner-City, the Underclass, and Public Policy. Chicago, IL: The University of Chicago Press.
SLIDE 15
15 Figures and Tables Figure 1: Map of Atlanta-Sandy Springs-Roswell, Georgia Metropolitan Statistical Area by County Note: Ambinakudige et al. (2017) list Fulton and DeKalb as the “city-proper;” and Cherokee, Gwinnett, Rockdale, Henry, Clayton, Fayette, Douglas, and Cobb as “Suburb-I, and all other counties as “Suburb-II”
SLIDE 16
16 Figure 2: LISA Clusters for White and African American Home and Workplace areas based on LODES 2010, in core Atlanta
SLIDE 17 17
Table 1: Atlanta MSA Employment-Residence Ratios By County (ACS, 5-year estimates 2011-2015) County Total Pop. Workers Working in County Workers Living in County Daytime Pop Day Pop Change due to Commute Percent Day Pop Change due to Commute Workers Live & Work in County Percent Workers Live & Work in County Employ- ment/ Residence Ratio Fulton 948,554 797,538 440,476 1,305,616 357,062 41.40 320,111 72.70 1.81 Gwinnett 825,911 339,014 380,147 784,778
221,918 58.40 0.89 Cobb 699,235 328,753 346,467 681,521
198,498 57.30 0.95 DeKalb 700,308 300,189 322,473 678,024
147,863 45.90 0.93 Clayton 262,455 95,636 107,102 250,989
37,411 34.90 0.89 Cherokee 218,277 60,283 102,499 176,061
43,818 42.80 0.59 Henry 206,349 55,312 89,559 172,102
33,225 37.10 0.62 Forsyth 182,916 67,298 83,000 167,214
38,105 45.90 0.81 Douglas 133,486 41,638 58,108 117,016
21,878 37.60 0.72 Coweta 129,397 36,583 58,486 107,494
26,659 45.60 0.63 Carroll 111,160 39,006 45,316 104,850
26,819 59.20 0.86 Paulding 143,845 26,094 65,775 104,164
18,268 27.80 0.40 Fayette 107,105 42,985 47,971 102,119
20,957 43.70 0.90 Bartow 100,382 35,525 41,725 94,182
24,226 58.10 0.85 Newton 100,808 28,020 41,809 87,019
17,290 41.40 0.67 Rockdale 85,650 35,776 36,780 84,646
15,155 41.20 0.97 Walton 84,397 21,156 36,490 69,063
13,840 37.90 0.58 Spalding 64,011 22,576 24,130 62,457
13,172 54.60 0.94 Barrow 69,933 16,704 30,691 55,946
9,887 32.20 0.54 Haralson 28,594 9,406 10,580 27,420
4,950 46.80 0.89 Pickens 29,486 9,976 12,440 27,022
6,680 53.70 0.80 Butts 23,563 7,079 8,886 21,756
3,512 39.50 0.80 Dawson 22,387 7,707 9,656 20,438
3,115 32.30 0.80 Meriwether 21,695 5,283 7,793 19,185
2,690 34.50 0.68 Morgan 17,870 8,853 7,667 19,056 1,186 7.30 3,938 51.40 1.15 Lamar 18,139 4,502 6,978 15,663
2,684 38.50 0.65 Pike 17,807 2,907 7,443 13,271
1,811 24.30 0.39 Jasper 13,748 2,654 5,904 10,498
1,912 32.40 0.45 Heard 11,708 2,371 4,488 9,591
1,364 30.40 0.53 Clay 3,133 580 901 2,812
346 38.40 0.64
Note: The four counties in yellow are the largest four counties (each with 700,000 or more resident population). The other shaded counties are all neighbors of Fulton County.
SLIDE 18
18 Table 2: Selected County Data: Population, Change in Entropy, and Race/Ethnic Composition.
Total Population Entropy Racial/Ethnic Populations (2014) % County Census 2000 ACS 2010- 14 E 2000 (A) E ACS 10-14 (B) Absolut e Δ in E (B-A) White Black Hispan ic Asian /PI Cobb 607751 708920 58.74 72.60 13.86 55.0 25.4 12.5 4.9 DeKalb 665865 707185 68.28 72.38 4.09 29.7 53.2 9.3 5.5 Fulton 816006 967100 64.20 73.13 8.92 40.6 43.5 7.7 6.1 Gwinnett 588448 842091 60.68 84.82 24.13 42.1 24.2 20.3 10.9
Notes: The shaded cells under Racial/Ethnic Population indicate two components; (a) the dark shaded cells and bold text indicate the largest race/ethnic group within a county, and (b) the light shaded cells indicate all groups that exceed 10 percent of the population within a specific county.
SLIDE 19
19 Table 3: Cross-tabulation of LISA Cluster Class for Percent African American in Home Block by Percent African American in Workplace Block, LODES 2010 Workplace block High- High High- Low Not Sign. Low- High Low-Low Total High- High 1,665 122 7,413 1,887 152 11,239 High- Low 47 15 446 15 93 616 Not Sign 3,088 455 55,538 2,652 1,300 63,033 Low- High 279 11 1,640 481 279 2,426 Low-Low 480 183 8,818 184 1,548 11,213 Total 5,559 786 73,855 5,219 3,108 88,527 Home block
