Suburban gentrification? Examining the geographies of New Urbanism in Atlanta’s inner suburbs

ABSTRACT

Since the 1990s, Atlanta’s inner suburbs have experienced rising rates of poverty alongside growing racial/ethnic diversity. ​ Meanwhile, on the built landscape, high-end (re)development projects incorporating New Urbanist features such as pedestrian accessibility, compact densities, and mixed land uses have become increasingly common. Although practitioners suggest that these developments promote social inclusivity and enhanced “livability,” individual cases suggest that some New Urbanist projects may be gentrifying underserved communities of color. This paper tests the extent of this concern by examining Atlanta's inner-suburban geographies of New Urbanism using a logit model. Results indicate that neighborhoods with older housing stock, more renters, lower incomes, lower rents relative to home prices, and higher Latino populations in close proximity to inner-suburban downtowns are most likely to be targeted for New Urbanist (re)development. Thus, despite the optimistic language, New Urbanism in this region appears to be serving as the inner-suburban face of gentrification.​

KEYWORDS:

• New Urbanism
• suburbs
• gentrification
• housing
• Atlanta

Acknowledgments

This study is part of the master’s thesis research of the author, titled “Pre- and post-crisis geographies of New Urbanism in Atlanta’s inner suburbs,” defended at the University of Tennessee. An earlier version of this manuscript was presented at the 2016 meeting of the American Association of Geographers in San Francisco, California, winning the 2016 Student Paper Competition of the Ethnic Geography Specialty Group. The author would like to thank Madhuri Sharma, Nicholas Nagle, Stephanie Bohon, Steve Holloway, and three anonymous reviewers for their insightful comments and suggestions that greatly helped shape and refine this work. All errors and omissions are my own.

Disclosure statement

No potential conflict of interest was reported by the author.​

Notes

1. Although this method is vulnerable to irregular population distributions within block groups, this issue is not as prevalent in inner suburbs as in more sparsely populated regions.

2. If projects broke ground before December 2007 and completed the construction of housing units without any evidence from satellite imagery suggesting that the project had stalled, then it is classified as a Time Period 1 project. If ground was cleared before December 2007 but instead remained without any signs of progress beyond the next time period (April 2008) and then was eventually resumed, the project is classified as a Time Period 2 project. Projects could be classified into both time periods if at least 20 units were constructed in separate phases that straddled the housing crash. This is the case for only one development.

3. One block group fits this criteria but is excluded because the portion that lies outside the City of Atlanta contains no housing units.

4. Block groups that did not meet the 95 percent threshold due to a river or tributary running through them are not excluded.

5. Sources include realtor and local government websites, Google Earth Street View, Zillow.com, Apartments.com, and local news reports.

6. Although assigning tract-level variables to block groups subjects the model to spatial misalignment, the inclusion of this variable did not significantly impact the signs of any other variables in the model.

7. Activity centers primarily serving non-commercial functions (e.g. Dobbins Air Reserve Base) are excluded. Block groups (and tracts) are considered to overlap activity centers if they share at least 500,000 square feet, unless that space is not usable (e.g., the overlap spans a highway where no NU development would be possible).

8. For the 2006–2010 ACS data, 170 block groups are missing median contract rent data, while 54 are missing median home value data. By using tract values and existing values in the block groups within the same tract, it was possible to impute estimates. When this was not possible or when this calculation yielded unlikely results, the tract value was assigned to the block group. Weighted means from neighboring tracts were used to impute values in the rare cases where tract-level data is missing.

9. Other variables run in the analysis but eventually excluded due to non-significance or collinearity include the share of housing built during the previous decade, share of single-family detached housing, occupancy rate, and interactions terms with the Latino population and the other two types of activity center. Interactions were also tested between activity-center types and the Black population, as well as the median household income.

10. All variables that are positive or negative in the block-group model remain positive or negative in the tract-level model. But some variables – housing age, regional center, and station community – are removed because their effect on the model is no longer significant. Besides this, the percent Black variable is no longer significant. The interaction variable weakens, but is still significant at 10 percent. Upon closer inspection, it seems more likely that this change reflects issues with using geographic units larger than block groups rather than a problem with the block-group data.