Hot city, cool city: explaining neighbourhood-level losses in low-cost rental housing in southern US cities

Abstract

We examine losses in affordable rental units in central cities in a key region of the United States, the South. We examine changes in low-cost rented units across eight large central cities, and then identify neighbourhood characteristics associated with such changes. Finally, we estimate a similar model in two of the cities with the greatest amount of loss in low-cost units and that have different overall housing market conditions – one a ‘hot market’ (Nashville, Tennessee) and one a ‘cool market’ (Memphis, Tennessee). We find that, generally, a number of neighbourhood conditions are associated with greater losses of low-cost units, including more young (25–34) adults, fewer public housing units, more initial low-cost units, and a larger percentage of newer units. However, these relationships do not show up consistently in the Memphis and Nashville models. The presence of younger adults (especially those aged 25–34) is a strong predictor of losing low-cost rental units in Nashville but is not a significant predictor in Memphis. Second, the presence of public housing units appears to serve as a buffer against the loss in low-cost units in a neighbourhood, but not in Memphis or Nashville, perhaps because they have fewer public housing units altogether.

Keywords:

• gentrification
• housing
• affordable housing
• rental housing
• abandonment

Acknowledgements

The authors acknowledge the support of the Federal Reserve Bank of Atlanta. They would like to thank Karen Leone De Nie and Carl Hudson for comments on earlier versions of this article. The views expressed here are the authors' and are not necessarily those of the Federal Reserve Bank of Atlanta or the Federal Reserve System. Any remaining errors are the authors' responsibility. Finally, the authors thank the editor and referees at the International Journal of Housing Policy for helpful comments.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1. The Census's South region includes Alabama, Arkansas, Delaware, the District of Columbia, Florida, Georgia, Kentucky, Louisiana, Maryland, Mississippi, North Carolina, Oklahoma, South Carolina, Tennessee, Texas, Virginia, and West Virginia.

2. It is important to note that these medians are not constant-quality figures. That is, some of the change in median rent is due to the change in the housing stock that is being rented. Constant-quality indices show smaller rent increases, about 3 per cent annually. But changes in medians are important because they indicate what the median renter is actually paying for rent. One of the reasons the median is rising substantially is because lower cost stock is essentially exiting the market, while newer stock is generally quite expensive.

3. We initially considered, including New Orleans, but subsequently omitted it because of its unique housing market history since Hurricane Katrina. Over the period focused on here, 2006–2010 to 2010–2014, New Orleans actually saw an increase in low-cost rented units. However, the number of such units in 2014 was still very low compared to pre-Katrina levels. In 2010, the number of rented units with gross rents under $750 per month was only just over 17,000, down from over 82,000 in 2000. While there was an increase in such units from 2000 to 2014, the number only grew to 23,000, still more than 70 per cent below the 2000 level.

4. The ACS is conducted annually, but for census tracts, the Census Bureau pools five years of data together to achieve adequate sample size. Therefore, we compare the five-year ACS sample data, including the years 2006 to 2010, to the five-year ACS sample, including the years 2010 to 2014. The data, then, measure change over a four-year span – the period ending in 2010 to the period ending in 2014.

5. Rather than adjusting the rent level defined here as ‘low cost’ for differences in median incomes across the eight cities, we used a constant $750 threshold for three reasons: (1) simplicity; (2) because the ACS data do not allow for highly specific thresholds; and (3) because we employ a tract level regression across all eight cities, which accounts for differences in variables closely related to housing costs (such as median home value and poverty rate). HUD's estimated area median incomes (AMI) for 2010 ranged from just over $40,000 in Memphis to just over $50,000 in Atlanta for a one-person household. A $750 per month gross rent is significantly higher relative to AMI in Memphis than in Atlanta. The average AMI for the eight cities was approximately $45,000 for one-person households and $64,200 for four-person households. At these average levels, a $750 per month rent is affordable at 47 per cent of the four-person AMI and 67 per cent of the one-person AMI.

6. Tracts do not align perfectly with city boundaries in most cities. Therefore, only tracts whose centroids lie within the cities are included. This means that the totals of the rows in this table will not match the citywide totals in Table 1.

7. These ratios are not closely related to the magnitude of total low-cost unit losses in the city.

8. The small number of cities suggests that multilevel modelling is inappropriate for this task, despite the fact that the tracts are nested in cities (O'Dwyer and Parker, 2014). In a subsequent section, we compare the results in two cities with substantial numbers of census tracts to examine inter-city heterogeneity issues.

9. The estimation results were checked for concerns over multicollinearity, non-normal error terms, and heteroskedasticity. Residuals were normally distributed. All variance inflation factors were under five, with most under three. Some heteroskedasticity was evident in the residuals, so robust standard errors were used in all estimations.

10. All percentages are specified in a form from 0 to 100. Some additional housing stock variables (that is, percent of units that are in multifamily buildings) were also entered in various permutations of the models, but these variables had little effect on the results and were not statistically significant. To conserve statistical power and prevent problems of severe multicollinearity, these variables were not retained in the model.

11. Multilevel modelling at the city-tract levels is likely to be unreliable here due to the small number of cities (O'Dwyer and Parker, 2014).