Abstract
The Housing Choice Voucher (HCV) program is designed in part to expand the neighborhood choices of assisted households, thereby enabling assisted households to find a living environment that simultaneously meets their housing and neighborhood preferences. While several studies have examined the impact of rental subsidies on neighborhood satisfaction, few have examined whether access to adequate transportation enables HCV recipients to locate housing in more desirable locations. This article relies on data from the Moving to Opportunity experiment to examine the impact of transportation access, rental housing vouchers, and geographic constraints on neighborhood satisfaction. We find that access to both vehicles and public transit positively influences neighborhood satisfaction, and the influence of vehicle access varies with transit proximity. These findings point to the importance of transportation in helping low-income assisted renter households locate housing in more desirable neighborhoods.
Acknowledgements
The authors would like to thank several colleagues who provided valuable feedback on earlier versions of this paper, including Regina Gray, Salin Geevarghese, Christopher Hayes, Zach McDade, Arthur (Taz) George, Eli Knaap, Evelyn Blumenberg, Gregory Pierce, and Michael Smart. The authors would also like to thank the editors and anonymous reviewers who provided several suggestions which greatly improved the paper.
Notes
1. Beginning in 2011, the AHS began to rely on HUD administrative data to define the type of housing assistance received by AHS respondents.
2. In a previous version of this article, we defined vehicle access as having access to a vehicle or a license, based on the rationale that licensed households could gain access to a car through car-sharing services. In response to the reviewers' comments, we decided to omit licensing from our definition of vehicle access, given that licenses are often acquired for personal identification purposes unrelated to mobility. Moreover, licensing likely captures unobserved personal characteristics not associated with vehicle access. We examined the sensitivity of our results to different specifications and found that while the signs and magnitudes of the coefficients were comparable across different specifications, in models that treated licensing separately from vehicle access, access to a license alone was not statistically significant. Therefore, we have chosen to omit licensing from the definition of vehicle access. The results of these sensitivity analyses are available from the authors.
3. We chose to rely on data from several sources, including the Census, HUD administrative data, and HUD's Fair Housing Equity Assessment database. We did not have recent data for all HUD data sources, so we chose to rely on 2000 Census data in order to be consistent across data sources.
4. The sample attrition between the baseline and final sample is due to three sources: (1) households dropping out of the survey by the final sample, (2) households moving out of an MTO metropolitan area by the final survey, and (3) missing values on one or more of the covariates. In the baseline sample, about 28% of the original 4,594 observations were lost due to missing values. The original final sample was smaller (N = 3,273) due to attrition. Of the 1,335 cases lost from the final sample after accounting for attrition, 745 were lost due to missing values, while 590 were lost due to households moving out of a MTO metro by the final sample.
5. If we are to assume that the variation in the “neighborhood evaluation” coefficients across treatment groups reflects treatment group differences in the causal effects of neighborhood characteristics, we must assume that the differences are the result of, on average, random variation in those neighborhood characteristics along with random variation in the importance of a given condition as it influences household satisfaction. If unobservable factors are correlated with either of these effects, then the estimated group-level differences will be biased. While it is reasonable to assume that this issue influences the differences between MTO participants and nonparticipants (MTO participants may have chosen to enroll in the MTO program to escape undesirable neighborhood conditions), since households were randomly assigned to treatment groups following enrollment, it does not seem likely that the unobservable factors influencing neighborhood satisfaction would vary systematically across groups.
6. With the original model specification, the New York model failed to achieve convergence, due to low variation in the “friends in the same neighborhood” variable in the New York sample. Since this variable was not crucial to our analysis, we decided to drop this variable from the New York model so that the remaining coefficients could be estimated.
7. Formally, for the five-alternative ordered probit model, we define:
Additional information
Funding
Notes on contributors
Casey Dawkins
Casey Dawkins is an Associate Professor in the Urban Studies and Planning Program and Research Associate with the National Center for Smart Growth Research and Education at the University of Maryland, College Park.
Jae Sik Jeon
Jae Sik Jeon is a Doctoral student in the School of Architecture, Planning, and Preservation at the University of Maryland, College Park.
Rolf Pendall
Rolf Pendall is Director of the Urban Institute's Metropolitan Housing and Communities Policy Center.