![Sample our Urban Studies journals, sign in here to start your FREE access for 14 days]({"type":"image" , "src" : "/sda/110826/TOC-Subject-Sample-2021-Urban-Studies.jpg"})
ABSTRACT
Drawing on the nationally representative Census Bureau Household Pulse Survey (N = 443,375), this study examined how pre-pandemic and pandemic-induced vulnerabilities shaped racial inequalities in housing insecurity from August to December 2020. Theorizing the pandemic as a “systemic shock,” this article helps explain how racial inequalities are reproduced both as legacies of long-standing disparities and through widespread, societal moments of disruption. Our results indicated that Black, Hispanic, and Asian households were all significantly more likely to fall behind on housing payments than white households. We found that preexisting racial inequities, including in educational attainment, household income, and homeownership rates, largely explained racial disparities in housing insecurity during COVID-19. However, the pandemic also induced new pathways to racial inequalities. Black and Hispanic households were more likely to lose income and when they lost income, they faced a greater likelihood of experiencing housing insecurity. We conclude by highlighting the importance of theorizing disruptive systemic shocks to understand racial inequalities and discuss findings of both disparities faced by Black and Hispanic households, as well as underrecognized housing insecurity among Asian Americans.
Acknowledgments
We thank Whitney Airgood-Obrycki, Christopher Herbert, Hyojung Lee, Daniel McCue, Giselle Routhier, Jonathan Spader, Lillian Leung, and Kristin Perkins for their thoughtful suggestions on earlier versions. We also thank the reviewers and the editor for their helpful feedback. We are grateful for the opportunity to present this work at the 2021 Housing Research Seminar Series and the 2022 Urban Affair Association meeting and thank the audience for their questions and ideas.
Disclosure statement
No potential conflict of interest was reported by the author(s).
Notes
1. States differential speeds and methods of distribution of federal Emergency Rental Assistance funds fall outside of the period examined in this paper: the second half of 2020. For a timeline of the distribution of various types of federal aid to renters throughout the pandemic, see Harvard Joint Center for Housing Studies (Citation2022b).
2. Survey respondents were sampled from the Census Bureau’s Master Address File (Census Bureau, Citation2021b) and recruited by e-mail or text messaging. The survey was completed online. The HPS’s response rate was lower than traditional in-person or mail-in surveys—during Phase 2 and 3 it ranged between 5.7% to 10.5%. A Census Bureau analysis of HPS revealed that the use of survey-provided weights mitigated some but not all of the potential non-response bias (Peterson et al., Citation2021). Due to the large sampling frame, weekly sample sizes during Phase 2 and 3 ranged from 58,729 to 110,019 respondents.
3. Phase 1 included a panel component, with repeat-respondents over several weeks—a design that was later dropped. Moreover, survey questions on housing insecurity were changed after Phase 1: making survey responses before and after this change incomparable. Additionally, we only included Phase 3 respondents through December, as some survey questions were changed after this date.
4. For example, the Panel Study of Income Dynamics includes questions for homeowners on late mortgage payments, forbearance, and foreclosure experiences, while it lacks equivalent questions for renters (The Institute for Social Research, Citation2019).
5. HPS weights for Phase 2 and early Phase 3 were based on the distribution of households in 2018 American Community Survey 5-year Estimates (Census Bureau, Citation2021b).
6. Most omitted respondents with housing payments did not complete the survey. To illustrate, just 2% of omitted respondents with housing payments did not answer the question on lost employment income, the 9th question in Phase 3, while 96% did not answer the 50th and last question on household income.
7. Even with sample attrition, the households in our sample were almost identical to respondents with housing payments overall in the HPS by age, race/ethnicity, educational attainment, and gender. Indeed, 27% of households in our sample were under age 35, the exact share of households under age 35 with a housing payment in the HPS. Likewise, 65% of respondents were white and 53% were female in both our sample and among all households with housing payments in the HPS. Our sample had roughly the same share of households with a bachelor’s degree (35%) compared with all households with mortgage or rent payments (34%).
8. Respondents answering “no” are considered housing insecure.
9. The models did not account for differences in the eviction moratorium process across states, nor for differential interpretation and application of the CDC federal eviction moratorium.
10. Index values for each state by day were averaged to correspond with the dates for each HPS week. Higher index values indicated greater stringency. Across the 459 state-weeks in our sample, the index averaged 57, with a low of 37 in Florida in week 19 and a high of 83 in Maine in week 13. Scores were generally highest in August, declined during the fall, and then climbed again in November and December.
11. We decided to use linear probability models over alternatives, such as logit or probit models. Our main rationale for LPM was its greater ease of interpreting coefficients, including interaction terms. One weakness of LPM models is that they assume that a unit increase in the predictor always changes the probability of the outcome by the same amount, regardless of the initial value of x—i.e., the assumption of linearity (Wooldridge, Citation2002). This makes these models potentially unreliable at extreme values of x (Wooldridge, Citation2002, p. 455). However, this was less of a concern for our purposes, as most of our predictors were ordinal or binary. We also used non-linear transformations (such as interactions and square terms; Mood, Citation2010, p. 79). As a robustness check, we ran logit models, and this did not substantively change our findings.
12. Like some other demographic controls, this referred to the educational attainment of household heads.
13. Additional models not shown break out our estimates by tenure. These models produce similar results and are available upon request.
14. In models stratified by tenure that are available upon request, state eviction moratoria had a positive, statistically significant association with housing insecurity for renter households but not for homeowner households.
Additional information
Notes on contributors
Sharon Cornelissen
Sharon Cornelissen is a Postdoctoral Fellow at the Harvard Joint Center for Housing Studies, who received her PhD in Sociology from Princeton University. She is a community-based ethnographer who has studied racial inequalities in housing amidst both urban decline and urban growth. Her book on life after depopulation in Detroit is forthcoming with the University of Chicago Press and her work has also appeared in Theory and Society, Sociological Forum, and Urban Affairs Review.
Alexander Hermann
Alexander Hermann is a Research Associate at the Harvard Joint Center for Housing Studies. His research focuses on housing markets, affordability, and policy, and he has contributed to or co-led the Center’s annual The State of the Nation’s Housing report since 2017. Alexander’s work has been featured in the New York Times, Wall Street Journal, Associated Press, and NPR, among others. He received a Master’s in Public Policy and a Master’s in Urban Planning from the University of Michigan in 2016.