Race and Assisted Housing

Abstract

This article explores racial disparities between assisted housing outcomes of black and white and white households with children. We compare the assisted housing occupied by black and white households with children, and examine whether young adult education, employment, and earnings outcomes in 2011 differ between blacks and whites who spent part of their childhood in assisted housing in the 2000s. We use a special version of the Panel Study of Income Dynamics (PSID) that has been address-matched to federally assisted housing, and the PSID’s Transition to Adulthood supplement, along with geocode-matched data from the U.S. Census Bureau, American Community Survey (ACS), CoreLogic real estate data, and U.S. Department of Housing and Urban Development (HUD). Statistical methods include difference in means, logit and general linear models. We find no evidence of racial disparities in the type of assisted housing program, the physical quality of project-based developments, or the management of public housing developments in the 2000 decade. But black households with children are more likely to live in assisted housing that is located in poorer quality neighborhoods. Multivariate tests reveal that the worse outcomes of black young adults compared with whites disappear once socioeconomic differences are taken into account. The discrepancy in assisted housing neighborhood quality experienced by black and white children makes no additional contribution to predicting young adult outcomes. Nonetheless, black children living in relatively better assisted housing neighborhoods tend to have better outcomes in young adulthood than those who live in poorer quality assisted housing neighborhoods. We discuss sources of racial disparity in neighborhood quality, and the policies enacted and proposed to address it.

Keywords:

• Race
• children
• assisted housing

Acknowledgments

We gratefully acknowledge support for this research from the W. K. Kellogg Foundation. We are also grateful to A. Polikoff, F. Roisman, P. Tegeler, and S. Popkin for their assistance; K. McGonagle, R. Schoeni, and other PSID colleagues for collaboration in creating the PSID-AHD; HUD colleagues for assistance with administrative data; and A. Robie and L. Staub for excellent research assistance.

Notes

1. Assisted housing refers to housing that is subsidized by the federal government to reduce the monthly rent and that is targeted on income-eligible households. Subsidies take two forms: demand side, through a rent voucher provided to income-eligible households to use in the private rental market; and supply side, through public housing developed and managed by local Public Housing Authorities, concessionary financing to private entities to develop or rehabilitate housing, or tax credits under the Treasury Department’s Low-Income Housing Tax Credit program.

2. Authors’ calculations using the 2011 AHS national data. The 2011 AHS provides a larger national sample than the typical national AHS, and also oversampled assisted housing units identified through an address match to HUD administrative data.

3. During the 2000s, there were roughly 80 field offices across the United States.

4. We are not the first to find the more powerful effects of family background relative to childhood neighborhood (e.g., Duncan, Boisjoly, & Harris, 2001).

5. Documentation of analyses not included in the tables in this article is available in a technical appendix from the authors.

6. The 1989 AHS Follow-On Study attempted to categorize whether each renter-occupied address in the AHS sample was part of one of the three assisted housing program types (public, multifamily, voucher).

7. Unit quality was measured with the AHS indices for severe and moderate defects. Neighborhood quality was measured by a 10-point neighborhood rating and an assessment of crime as a problem in the neighborhood. All are self-reports.

8. We exclude research that focuses on pre- and post neighborhood quality for blacks versus whites because this does not directly address our central question of differential effects.

9. The only exceptions are studies of the effects of court orders emerging from class action lawsuits (see, e.g., Patterson & Yoo, 2012).

10. Although WtWV also studied educational attainment outcomes, they did not address race differences in education (Gubits, email correspondence, October 30, 2015).

11. Chetty et al. construct an earnings-based index of college quality based on earnings at age 31 and on the college that was attended at age 20.

12. This is not entirely true because the comparison group lived in inner-city public housing prior to voucher receipt and volunteered to participate in the MTO experiment.

13. MTO research found no effects of the intervention on education, employment, and earnings but did not publish results by race in the final evaluation report (Sanbonmatsu et al., 2011; Sanbonmatsu, email correspondence, October 30, 2015).

14. Access to PSID Sensitive Data Files was obtained under special arrangements to protect the anonymity of respondents.

15. For example, if a household lived in public housing from 1995 to 2005, we use the 1995 observation. Alternatively, if a household lived in public housing from 1995 to 1999, left housing assistance, but then began a new spell of assisted housing receipt from 2001 to 2005, it contributes two observations (1995 and 2001).

16. The power analysis adjustments are based on the average cluster sample size of 1.336 and the observed intraclass correlations, r, for the various outcomes, which yield design effects ranging from 1 to 1.132. See Technical Appendix for more details. The spells sample for 2000 is based on household observations and is not affected by nonindependence.

17. A few cases were more than three standard deviations above the mean.

18. The Bonferroni correction adjusts the critical value/confidence interval when running multiple tests.  It is a highly conservative bound, not an exact correction.  It is calculated by dividing the desired significance level by the number of tests. For example, if the goal is to attain a 5% significance level and we perform five tests, results must attain p = .01 (.05/5) to be considered statistically significant.

19. Disability is defined as having any physical or nervous condition that limits the type of work or the amount of work he/she can do.

20. The CoreLogic measures were excluded from the factor analysis because of substantial missing data.

21. The economic returns to a GED is a topic of debate (see, e.g., Heckman & Rubinstein, 2001). However, tests including and excluding GEDs yield similar estimates. For simplicity, we report results including the GED.

22. We also find no relationship between program type and household disadvantage either overall or by race (see Technical Appendix).

23. Quality scores are based on the inspection of a sample of units and of public spaces in and around the property. There are no differences between cases with and without PHAS or REAC inspection or overall PHAS score data. See the Technical Appendix available from the authors for further details on missing data and matching cases to HUD administrative data.

24. In an effort to boost the number of observations for whites, we reran this analysis combining the most and the middle groups of disadvantaged. This raised the number of cases in the white subsample to 10 and 18 for the overall management and inspection scores, respectively. This, too, revealed no differences between blacks and whites.

25. The two-group t-test accounts for differences in sample sizes because the standard errors are based on the sample sizes of the two groups (i.e., it is the standard deviation divided by sample size for each group).  Although the statistical power of a two-group t-test is maximized when the sample is evenly divided, it is not necessary for the groups to be of equal size to support comparisons between them.

26. Although we also examined administrative data on HOPE VI and public housing demolition, the number of observations for each is too small for further analysis. Only three households in the sample lived in a development affected by the HOPE VI public housing revitalization program, and two white and 22 black households lived in a public housing development either before demolition or after rebuilding.

27. Identifying the causes of racial disparities in income is beyond the focus of this article. Additional analysis (not shown) demonstrates that income is highly correlated with labor-force participation measures. Black parents in the analysis sample report significantly fewer annual work hours compared with white parents. This disparity persists even after accounting for the larger fraction of black parents who have a disability that limits the type and amount of work they can do.

28. Because there may be some degree of selection into neighborhoods, the child’s family background is included in the test of neighborhood quality.

29. It will be recalled that the reported p values are adjusted using the Bonferroni correction.

30. We tested two definitions: a neighborhood poverty rate of less than 30%, and a composite neighborhood-quality measure based on neighborhood segregation, poverty, and house values. For the latter, any neighborhood that did not qualify as a poor-quality neighborhood (i.e., highest quintile for segregation and poverty, lowest quintile for house value) was included.

31. The one possible exception is that the high school graduation rates of blacks in better assisted housing neighborhoods are nearly equivalent to those of whites.

32. Tabulations of the 2011 AHS by the authors.

33. Tabulations of the PSID-AHD data enriched with urbanicity measures from the Neighborhood Change Data Base. The public housing development siting process in some cities also resulted in the location of public housing in very disadvantaged neighborhoods (Schill & Wachter, 1995).